krqj "0xhI'`|>d@ _ᇺ@PAH 0I ADA b) E) DGs|3dà 1Ad A*ȡ ,RXI_\Jb2j !Pؐ25?E㐘ץICADL%8wO!k~$Zx)]kp ` vwU*&P 9B!y7n3"n2&m7trZS ygHRKQ7la \@{D'a05S`F=v~pgY{ +dw!akW0aZF. DPcmAl(E hTQ~CL{y~ A.8}w 3XPxh 8]8ܤ_i yFsO6#g[^ʷ3 &u`(9ɯ sǞqaN
$/>w#PIq 9&-q"mq)"nԙu\mBTWb{w5/^Ķ 4 ѯ]=kt'
ے\KtcdHB R> `h L:R52DEY' ݄E@9>)imnzG!D cgp~!?jaX !x 4ǤCIG~NR 8({:(@j|?42^͵[;?VC~=9TqOS?iτN'bVmV4' (>y흁 dS"}gQ~R5Yj3+!-le)Pe
§mo OfH[ 4t;7?/~<"jb3Gkb+B#T~ Л mC$ 9cb
ehs:(@ɐ?V}B w X0aA-4')E5g -C !PtU_zm լbFska3p H Mʗ6kc
5$/hP`uUjژuJm!תeaңZ5
}ɼQ G@G$%M}>e
{J6AfŘXǬn>XE& SXU? g? ||cx4SY7㥼Zi=Ҫ?/^{;[ v܈? +wD p@(<*پZ!~C=ۧ艆MZY" q߄41f ; &H'|$d*ёY.fℎP5@}Wf0Q/Kpx\d6?HQEV5Y2R>[]]oym9]RRŨx.|kFʕ ^ё *Ʉ]_=Q_]T
' L4uZʱxGXV(U暷;q%dW/]
;2F9 GUf}Ŗ:9FA7
SYAk/0*B扢'C/I$^UeJX%Wك EIB 8ъѣ[ "C.p tnZ mice/man/ 0000755 0001762 0000144 00000000000 14437176012 011747 5 ustar ligges users mice/man/cc.Rd 0000644 0001762 0000144 00000001617 14330031606 012617 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/cc.R
\name{cc}
\alias{cc}
\title{Select complete cases}
\usage{
cc(x)
}
\arguments{
\item{x}{An \code{R} object. Methods are available for classes
\code{mids}, \code{data.frame} and \code{matrix}. Also, \code{x}
could be a vector.}
}
\value{
A \code{vector}, \code{matrix} or \code{data.frame} containing the data of the complete cases.
}
\description{
Extracts the complete cases, also known as \emph{listwise deletion}.
\code{cc(x)} is similar to
\code{na.omit(x)}, but returns an object of the same class
as the input data. Dimensions are not dropped. For extracting
incomplete cases, use \code{\link{ici}}.
}
\examples{
# cc(nhanes) # get the 13 complete cases
# cc(nhanes$bmi) # extract complete bmi
}
\seealso{
\code{\link{na.omit}}, \code{\link{cci}}, \code{\link{ici}}
}
\author{
Stef van Buuren, 2017.
}
\keyword{univar}
mice/man/as.mitml.result.Rd 0000644 0001762 0000144 00000001160 14330031606 015264 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/as.R
\name{as.mitml.result}
\alias{as.mitml.result}
\title{Converts into a \code{mitml.result} object}
\usage{
as.mitml.result(x)
}
\arguments{
\item{x}{An object of class \code{mira}}
}
\value{
An S3 object of class \code{mitml.result}, a list
containing $m$ fitted analysis objects.
}
\description{
The \code{as.mitml.result()} function takes the results of repeated
complete-data analysis stored as a list, and turns it
into an object of class \code{mitml.result}.
}
\seealso{
\code{\link[mitml]{with.mitml.list}}
}
\author{
Stef van Buuren
}
mice/man/mice.impute.mpmm.Rd 0000644 0001762 0000144 00000006130 14347334322 015422 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.mpmm.R
\name{mice.impute.mpmm}
\alias{mice.impute.mpmm}
\alias{mpmm}
\title{Imputation by multivariate predictive mean matching}
\usage{
mice.impute.mpmm(data, format = "imputes", ...)
}
\arguments{
\item{data}{matrix with exactly two missing data patterns}
\item{format}{A character vector specifying the type of object that should
be returned. The default is \code{format = "imputes"}.}
\item{...}{Other named arguments.}
}
\value{
A matrix with imputed data, which has \code{ncol(y)} columns and
\code{sum(wy)} rows.
}
\description{
Imputes multivariate incomplete data among which there are specific relations,
for instance, polynomials, interactions, range restrictions and sum scores.
}
\details{
This function implements the predictive mean matching and applies canonical
regression analysis to select donors fora set of missing variables. In general,
canonical regressionanalysis looks for a linear combination of covariates that
predicts a linear combination of outcomes (a set of missing variables)
optimally in a least-square sense (Israels, 1987). The predicted
value of the linear combination of the set of missing variables
would be applied to perform predictive mean matching.
}
\note{
The function requires variables in the block have the same missingness pattern.
If there are more than one missingness pattern, the function will return
a warning.
}
\examples{
# simulate data
beta2 <- beta1 <- .5
x <- rnorm(1000)
e <- rnorm(1000, 0, 1)
y <- beta1 * x + beta2 * x^2 + e
dat <- data.frame(y = y, x = x, x2 = x^2)
m <- as.logical(rbinom(1000, 1, 0.25))
dat[m, c("x", "x2")] <- NA
# impute
blk <- list("y", c("x", "x2"))
meth <- c("", "mpmm")
imp <- mice(dat, blocks = blk, method = meth, print = FALSE,
m = 2, maxit = 2)
# analyse and check
summary(pool(with(imp, lm(y ~ x + x2))))
with(dat, plot(x, x2, col = mdc(1)))
with(complete(imp), points(x[m], x2[m], col = mdc(2)))
}
\seealso{
\code{\link{mice.impute.pmm}}
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-knowledge.html#sec:quadratic}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Mingyang Cai and Gerko Vink
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/nimp.Rd 0000644 0001762 0000144 00000003541 14334522175 013205 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/nimp.R
\name{nimp}
\alias{nimp}
\title{Number of imputations per block}
\usage{
nimp(where, blocks = make.blocks(where))
}
\arguments{
\item{where}{A data frame or matrix with logicals of the same dimensions
as \code{data} indicating where in the data the imputations should be
created. The default, \code{where = is.na(data)}, specifies that the
missing data should be imputed. The \code{where} argument may be used to
overimpute observed data, or to skip imputations for selected missing values.
Note: Imputation methods that generate imptutations outside of
\code{mice}, like \code{mice.impute.panImpute()} may depend on a complete
predictor space. In that case, a custom \code{where} matrix can not be
specified.}
\item{blocks}{List of vectors with variable names per block. List elements
may be named to identify blocks. Variables within a block are
imputed by a multivariate imputation method
(see \code{method} argument). By default each variable is placed
into its own block, which is effectively
fully conditional specification (FCS) by univariate models
(variable-by-variable imputation). Only variables whose names appear in
\code{blocks} are imputed. The relevant columns in the \code{where}
matrix are set to \code{FALSE} of variables that are not block members.
A variable may appear in multiple blocks. In that case, it is
effectively re-imputed each time that it is visited.}
}
\value{
A numeric vector of length \code{length(blocks)} containing
the number of cells that need to be imputed within a block.
}
\description{
Calculates the number of cells within a block for which imputation
is requested.
}
\examples{
where <- is.na(nhanes)
# standard FCS
nimp(where)
# user-defined blocks
nimp(where, blocks = name.blocks(list(c("bmi", "hyp"), "age", "chl")))
}
\seealso{
\code{\link{mice}}
}
mice/man/md.pattern.Rd 0000644 0001762 0000144 00000003612 14330031647 014310 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/md.pattern.R
\name{md.pattern}
\alias{md.pattern}
\title{Missing data pattern}
\usage{
md.pattern(x, plot = TRUE, rotate.names = FALSE)
}
\arguments{
\item{x}{A data frame or a matrix containing the incomplete data. Missing
values are coded as NA's.}
\item{plot}{Should the missing data pattern be made into a plot. Default is
`plot = TRUE`.}
\item{rotate.names}{Whether the variable names in the plot should be placed
horizontally or vertically. Default is `rotate.names = FALSE`.}
}
\value{
A matrix with \code{ncol(x)+1} columns, in which each row corresponds
to a missing data pattern (1=observed, 0=missing). Rows and columns are
sorted in increasing amounts of missing information. The last column and row
contain row and column counts, respectively.
}
\description{
Display missing-data patterns.
}
\details{
This function is useful for investigating any structure of missing
observations in the data. In specific case, the missing data pattern could be
(nearly) monotone. Monotonicity can be used to simplify the imputation model.
See Schafer (1997) for details. Also, the missing pattern could suggest which
variables could potentially be useful for imputation of missing entries.
}
\examples{
md.pattern(nhanes)
# age hyp bmi chl
# 13 1 1 1 1 0
# 1 1 1 0 1 1
# 3 1 1 1 0 1
# 1 1 0 0 1 2
# 7 1 0 0 0 3
# 0 8 9 10 27
}
\references{
Schafer, J.L. (1997), Analysis of multivariate incomplete data.
London: Chapman&Hall.
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}: Multivariate
Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
}
\author{
Gerko Vink, 2018, based on an earlier version of the same function by
Stef van Buuren, Karin Groothuis-Oudshoorn, 2000
}
\keyword{univar}
mice/man/mice.mids.Rd 0000644 0001762 0000144 00000004332 14330031647 014104 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.mids.R
\name{mice.mids}
\alias{mice.mids}
\title{Multivariate Imputation by Chained Equations (Iteration Step)}
\usage{
mice.mids(obj, newdata = NULL, maxit = 1, printFlag = TRUE, ...)
}
\arguments{
\item{obj}{An object of class \code{mids}, typically produces by a previous
call to \code{mice()} or \code{mice.mids()}}
\item{newdata}{An optional \code{data.frame} for which multiple imputations
are generated according to the model in \code{obj}.}
\item{maxit}{The number of additional Gibbs sampling iterations.}
\item{printFlag}{A Boolean flag. If \code{TRUE}, diagnostic information
during the Gibbs sampling iterations will be written to the command window.
The default is \code{TRUE}.}
\item{...}{Named arguments that are passed down to the univariate imputation
functions.}
}
\description{
Takes a \code{mids} object, and produces a new object of class \code{mids}.
}
\details{
This function enables the user to split up the computations of the Gibbs
sampler into smaller parts. This is useful for the following reasons:
\itemize{ \item RAM memory may become easily exhausted if the number of
iterations is large. Returning to prompt/session level may alleviate these
problems. \item The user can compute customized convergence statistics at
specific points, e.g. after each iteration, for monitoring convergence. -
For computing a 'few extra iterations'. } Note: The imputation model itself
is specified in the \code{mice()} function and cannot be changed with
\code{mice.mids}. The state of the random generator is saved with the
\code{mids} object.
}
\examples{
imp1 <- mice(nhanes, maxit = 1, seed = 123)
imp2 <- mice.mids(imp1)
# yields the same result as
imp <- mice(nhanes, maxit = 2, seed = 123)
# verification
identical(imp$imp, imp2$imp)
#
}
\references{
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
}
\seealso{
\code{\link{complete}}, \code{\link{mice}}, \code{\link{set.seed}},
\code{\link[=mids-class]{mids}}
}
\author{
Stef van Buuren, Karin Groothuis-Oudshoorn, 2000
}
\keyword{iteration}
mice/man/estimice.Rd 0000644 0001762 0000144 00000003671 14330031606 014036 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.norm.R
\name{estimice}
\alias{estimice}
\title{Computes least squares parameters}
\usage{
estimice(x, y, ls.meth = "qr", ridge = 1e-05, ...)
}
\arguments{
\item{x}{Matrix (\code{n} x \code{p}) of complete covariates.}
\item{y}{Incomplete data vector of length \code{n}}
\item{ls.meth}{the method to use for obtaining the least squares estimates. By
default parameters are drawn by means of QR decomposition.}
\item{ridge}{A small numerical value specifying the size of the ridge used.
The default value \code{ridge = 1e-05} represents a compromise between stability
and unbiasedness. Decrease \code{ridge} if the data contain many junk variables.
Increase \code{ridge} for highly collinear data.}
\item{...}{Other named arguments.}
}
\value{
A \code{list} containing components \code{c} (least squares estimate),
\code{r} (residuals), \code{v} (variance/covariance matrix) and \code{df}
(degrees of freedom).
}
\description{
This function computes least squares estimates, variance/covariance matrices,
residuals and degrees of freedom according to ridge regression, QR decomposition
or Singular Value Decomposition. This function is internally called by .norm.draw(),
but can be called by any user-specified imputation function.
}
\details{
When calculating the inverse of the crossproduct of the predictor matrix,
problems may arise. For example, taking the inverse is not possible when the
predictor matrix is rank deficient, or when the estimation problem is
computationally singular. This function detects such error cases and
automatically falls back to adding a ridge penalty to the diagonal of the
crossproduct to allow for proper calculation of the inverse.
}
\note{
This functions adds a star to variable names in the mice iteration
history to signal that a ridge penalty was added. In that case, it
also adds an entry to \code{loggedEvents}.
}
\author{
Gerko Vink, 2018
}
mice/man/fix.coef.Rd 0000644 0001762 0000144 00000003571 14330031606 013734 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/fix.coef.R
\name{fix.coef}
\alias{fix.coef}
\title{Fix coefficients and update model}
\usage{
fix.coef(model, beta = NULL)
}
\arguments{
\item{model}{An R model, e.g., produced by \code{lm} or \code{glm}}
\item{beta}{A numeric vector with \code{length(coef)} model coefficients.
If the vector is not named, the coefficients should be
given in the same order as in \code{coef(model)}. If the vector is named,
the procedure attempts to match on names.}
}
\value{
An updated R model object
}
\description{
Refits a model with a specified set of coefficients.
}
\details{
The function calculates the linear predictor using the new coefficients,
and reformulates the model using the \code{offset}
argument. The linear predictor is called
\code{offset}, and its coefficient will be \code{1} by definition.
The new model only fits the intercept, which should be \code{0}
if we set \code{beta = coef(model)}.
}
\examples{
model0 <- lm(Volume ~ Girth + Height, data = trees)
formula(model0)
coef(model0)
deviance(model0)
# refit same model
model1 <- fix.coef(model0)
formula(model1)
coef(model1)
deviance(model1)
# change the beta's
model2 <- fix.coef(model0, beta = c(-50, 5, 1))
coef(model2)
deviance(model2)
# compare predictions
plot(predict(model0), predict(model1))
abline(0, 1)
plot(predict(model0), predict(model2))
abline(0, 1)
# compare proportion explained variance
cor(predict(model0), predict(model0) + residuals(model0))^2
cor(predict(model1), predict(model1) + residuals(model1))^2
cor(predict(model2), predict(model2) + residuals(model2))^2
# extract offset from constrained model
summary(model2$offset)
# it also works with factors and missing data
model0 <- lm(bmi ~ age + hyp + chl, data = nhanes2)
model1 <- fix.coef(model0)
model2 <- fix.coef(model0, beta = c(15, -8, -8, 2, 0.2))
}
\author{
Stef van Buuren, 2018
}
mice/man/mice.impute.2l.norm.Rd 0000644 0001762 0000144 00000005760 14330031647 015747 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.2l.norm.R
\name{mice.impute.2l.norm}
\alias{mice.impute.2l.norm}
\title{Imputation by a two-level normal model}
\usage{
mice.impute.2l.norm(y, ry, x, type, wy = NULL, intercept = TRUE, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{type}{Vector of length \code{ncol(x)} identifying random and class
variables. Random variables are identified by a '2'. The class variable
(only one is allowed) is coded as '-2'. Random variables also include the
fixed effect.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{intercept}{Logical determining whether the intercept is automatically
added.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes univariate missing data using a two-level normal model
}
\details{
Implements the Gibbs sampler for the linear multilevel model with
heterogeneous with-class variance (Kasim and Raudenbush, 1998). Imputations
are drawn as an extra step to the algorithm. For simulation work see Van
Buuren (2011).
The random intercept is automatically added in \code{mice.impute.2L.norm()}.
A model within a random intercept can be specified by \code{mice(...,
intercept = FALSE)}.
}
\note{
Added June 25, 2012: The currently implemented algorithm does not
handle predictors that are specified as fixed effects (type=1). When using
\code{mice.impute.2l.norm()}, the current advice is to specify all predictors
as random effects (type=2).
Warning: The assumption of heterogeneous variances requires that in every
class at least one observation has a response in \code{y}.
}
\references{
Kasim RM, Raudenbush SW. (1998). Application of Gibbs sampling to nested
variance components models with heterogeneous within-group variance. Journal
of Educational and Behavioral Statistics, 23(2), 93--116.
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}: Multivariate
Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
Van Buuren, S. (2011) Multiple imputation of multilevel data. In Hox, J.J.
and and Roberts, J.K. (Eds.), \emph{The Handbook of Advanced Multilevel
Analysis}, Chapter 10, pp. 173--196. Milton Park, UK: Routledge.
}
\seealso{
Other univariate-2l:
\code{\link{mice.impute.2l.bin}()},
\code{\link{mice.impute.2l.lmer}()},
\code{\link{mice.impute.2l.pan}()}
}
\author{
Roel de Jong, 2008
}
\concept{univariate-2l}
\keyword{datagen}
mice/man/toenail2.Rd 0000644 0001762 0000144 00000004122 14330031606 013741 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/toenail2.R
\docType{data}
\name{toenail2}
\alias{toenail2}
\title{Toenail data}
\format{
A data frame with 1908 observations on the following 5 variables:
\describe{
\item{\code{patientID}}{a numeric vector giving the ID of patient}
\item{\code{outcome}}{a factor with 2 levels giving the response}
\item{\code{treatment}}{a factor with 2 levels giving the treatment group}
\item{\code{time}}{a numeric vector giving the time of the visit
(not exactly monthly intervals hence not round numbers)}
\item{\code{visit}}{an integer giving the number of the visit}
}
}
\source{
De Backer, M., De Vroey, C., Lesaffre, E., Scheys, I., and De
Keyser, P. (1998). Twelve weeks of continuous oral therapy for
toenail onychomycosis caused by dermatophytes: A double-blind
comparative trial of terbinafine 250 mg/day versus itraconazole 200
mg/day. Journal of the American Academy of Dermatology, 38, 57-63.
}
\description{
The toenail data come from a Multicenter study comparing two oral
treatments for toenail infection. Patients were evaluated for the
degree of separation of the nail. Patients were randomized into two
treatments and were followed over seven visits - four in the first
year and yearly thereafter. The patients have not been treated
prior to the first visit so this should be regarded as the
baseline.
}
\details{
Apart from formatting, this dataset is identical to
\code{toenail}. The formatting is taken identical to
\code{data("toenail", package = "HSAUR3")}.
}
\references{
Lesaffre, E. and Spiessens, B. (2001). On the effect of the number of
quadrature points in a logistic random-effects model: An example.
Journal of the Royal Statistical Society, Series C, 50, 325-335.
G. Fitzmaurice, N. Laird and J. Ware (2004) Applied Longitudinal Analysis,
Wiley and Sons, New York, USA.
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-catoutcome.html#example}{\emph{Flexible
Imputation of Missing Data. Second Edition.}} Chapman & Hall/CRC.
Boca Raton, FL.
}
\seealso{
\code{\link{toenail}}
}
\keyword{datasets}
mice/man/appendbreak.Rd 0000644 0001762 0000144 00000002077 14330031606 014507 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/auxiliary.R
\name{appendbreak}
\alias{appendbreak}
\title{Appends specified break to the data}
\usage{
appendbreak(data, brk, warp.model = warp.model, id = NULL, typ = "pred")
}
\arguments{
\item{data}{A data frame in the long long format}
\item{brk}{A vector of break ages}
\item{warp.model}{A time warping model}
\item{id}{The subject identifier}
\item{typ}{Label to signal that this is a newly added observation}
}
\value{
A long data frame with additional rows for the break ages
}
\description{
A custom function to insert rows in long data with new pseudo-observations
that are being done on the specified break ages. There should be a
column called \code{first} in \code{data} with logical data that codes whether
the current row is the first for subject \code{id}. Furthermore,
the function assumes that columns \code{age}, \code{occ},
\code{hgt.z}, \code{wgt.z} and
\code{bmi.z} are available. This function is used on the \code{tbc}
data in FIMD chapter 9. Check that out to see it in action.
}
mice/man/make.visitSequence.Rd 0000644 0001762 0000144 00000002522 14330031606 015771 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/visitSequence.R
\name{make.visitSequence}
\alias{make.visitSequence}
\title{Creates a \code{visitSequence} argument}
\usage{
make.visitSequence(data = NULL, blocks = NULL)
}
\arguments{
\item{data}{A data frame or a matrix containing the incomplete data. Missing
values are coded as \code{NA}.}
\item{blocks}{List of vectors with variable names per block. List elements
may be named to identify blocks. Variables within a block are
imputed by a multivariate imputation method
(see \code{method} argument). By default each variable is placed
into its own block, which is effectively
fully conditional specification (FCS) by univariate models
(variable-by-variable imputation). Only variables whose names appear in
\code{blocks} are imputed. The relevant columns in the \code{where}
matrix are set to \code{FALSE} of variables that are not block members.
A variable may appear in multiple blocks. In that case, it is
effectively re-imputed each time that it is visited.}
}
\value{
Vector containing block names
}
\description{
This helper function creates a valid \code{visitSequence}. The
\code{visitSequence} is an argument to the \code{mice} function that
specifies the sequence in which blocks are imputed.
}
\examples{
make.visitSequence(nhanes)
}
\seealso{
\code{\link{mice}}
}
mice/man/print.mads.Rd 0000644 0001762 0000144 00000000652 13666252075 014327 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/print.R
\name{print.mads}
\alias{print.mads}
\title{Print a \code{mads} object}
\usage{
\method{print}{mads}(x, ...)
}
\arguments{
\item{x}{Object of class \code{mads}}
\item{...}{Other parameters passed down to \code{print.default()}}
}
\value{
\code{NULL}
}
\description{
Print a \code{mads} object
}
\seealso{
\code{\link[=mads-class]{mads}}
}
mice/man/as.mids.Rd 0000644 0001762 0000144 00000007425 14334522175 013605 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/as.R
\name{as.mids}
\alias{as.mids}
\title{Converts an imputed dataset (long format) into a \code{mids} object}
\usage{
as.mids(long, where = NULL, .imp = ".imp", .id = ".id")
}
\arguments{
\item{long}{A multiply imputed data set in long format, for example
produced by a call to \code{complete(..., action = 'long', include = TRUE)},
or by other software.}
\item{where}{A data frame or matrix with logicals of the same dimensions
as \code{data} indicating where in the data the imputations should be
created. The default, \code{where = is.na(data)}, specifies that the
missing data should be imputed. The \code{where} argument may be used to
overimpute observed data, or to skip imputations for selected missing values.
Note: Imputation methods that generate imptutations outside of
\code{mice}, like \code{mice.impute.panImpute()} may depend on a complete
predictor space. In that case, a custom \code{where} matrix can not be
specified.}
\item{.imp}{An optional column number or column name in \code{long},
indicating the imputation index. The values are assumed to be consecutive
integers between 0 and \code{m}. Values \code{1} through \code{m}
correspond to the imputation index, value \code{0} indicates
the original data (with missings).
By default, the procedure will search for a variable named \code{".imp"}.}
\item{.id}{An optional column number or column name in \code{long},
indicating the subject identification. If not specified, then the
function searches for a variable named \code{".id"}. If this variable
is found, the values in the column will define the row names in
the \code{data} element of the resulting \code{mids} object.}
}
\value{
An object of class \code{mids}
}
\description{
This function converts imputed data stored in long format into
an object of class \code{mids}. The original incomplete dataset
needs to be available so that we know where the missing data are.
The function is useful to convert back operations applied to
the imputed data back in a \code{mids} object. It may also be
used to store multiply imputed data sets from other software
into the format used by \code{mice}.
}
\note{
The function expects the input data \code{long} to be sorted by
imputation number (variable \code{".imp"} by default), and in the
same sequence within each imputation block.
}
\examples{
# impute the nhanes dataset
imp <- mice(nhanes, print = FALSE)
# extract the data in long format
X <- complete(imp, action = "long", include = TRUE)
# create dataset with .imp variable as numeric
X2 <- X
# nhanes example without .id
test1 <- as.mids(X)
is.mids(test1)
identical(complete(test1, action = "long", include = TRUE), X)
# nhanes example without .id where .imp is numeric
test2 <- as.mids(X2)
is.mids(test2)
identical(complete(test2, action = "long", include = TRUE), X)
# nhanes example, where we explicitly specify .id as column 2
test3 <- as.mids(X, .id = ".id")
is.mids(test3)
identical(complete(test3, action = "long", include = TRUE), X)
# nhanes example with .id where .imp is numeric
test4 <- as.mids(X2, .id = 2)
is.mids(test4)
identical(complete(test4, action = "long", include = TRUE), X)
# example without an .id variable
# variable .id not preserved
X3 <- X[, -2]
test5 <- as.mids(X3)
is.mids(test5)
identical(complete(test5, action = "long", include = TRUE)[, -2], X[, -2])
# as() syntax has fewer options
test7 <- as(X, "mids")
test8 <- as(X2, "mids")
test9 <- as(X2[, -2], "mids")
rev <- ncol(X):1
test10 <- as(X[, rev], "mids")
# where argument copies also observed data into $imp element
where <- matrix(TRUE, nrow = nrow(nhanes), ncol = ncol(nhanes))
colnames(where) <- colnames(nhanes)
test11 <- as.mids(X, where = where)
identical(complete(test11, action = "long", include = TRUE), X)
}
\author{
Gerko Vink
}
\keyword{mids}
mice/man/anova.Rd 0000644 0001762 0000144 00000001162 14330031606 013331 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/anova.R
\name{anova.mira}
\alias{anova.mira}
\title{Compare several nested models}
\usage{
\method{anova}{mira}(object, ..., method = "D1", use = "wald")
}
\arguments{
\item{object}{Two or more objects of class \code{mira}}
\item{...}{Other parameters passed down to \code{D1()}, \code{D2()},
\code{D3()} and \code{mitml::testModels}.}
\item{method}{Either \code{"D1"}, \code{"D2"} or \code{"D3"}}
\item{use}{An character indicating the test statistic}
}
\value{
Object of class \code{mice.anova}
}
\description{
Compare several nested models
}
mice/man/pops.Rd 0000644 0001762 0000144 00000004671 14330031606 013216 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/pops.R
\docType{data}
\name{pops}
\alias{pops}
\alias{pops.pred}
\title{Project on preterm and small for gestational age infants (POPS)}
\format{
\code{pops} is a data frame with 959 rows and 86 columns.
\code{pops.pred} is the 86 by 86 binary predictor matrix used for specifying
the multiple imputation model.
}
\source{
Hille, E. T. M., Elbertse, L., Bennebroek Gravenhorst, J., Brand, R.,
Verloove-Vanhorick, S. P. (2005). Nonresponse bias in a follow-up study of
19-year-old adolescents born as preterm infants. Pediatrics, 116(5):662666.
Hille, E. T. M., Weisglas-Kuperus, N., Van Goudoever, J. B., Jacobusse, G.
W., Ens-Dokkum, M. H., De Groot, L., Wit, J. M., Geven, W. B., Kok, J. H., De
Kleine, M. J. K., Kollee, L. A. A., Mulder, A. L. M., Van Straaten, H. L. M.,
De Vries, L. S., Van Weissenbruch, M. M., Verloove-Vanhorick, S. P. (2007).
Functional outcomes and participation in young adulthood for very preterm and
very low birth weight infants: The Dutch project on preterm and small for
gestational age infants at 19 years of age. Pediatrics, 120(3):587595.
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-selective.html#pops-study-19-years-follow-up}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\description{
Subset of data from the POPS study, a national, prospective study on preterm
children, including all liveborn infants <32 weeks gestational age and/or <1500
g from 1983 (n = 1338).
}
\details{
The data set concerns of subset of 959 children that survived up to the age
of 19 years.
Hille et al (2005) divided the 959 survivors into three groups: Full
responders (examined at an outpatient clinic and completed the
questionnaires, n = 596), postal responders (only completed the mailed
questionnaires, n = 109), non-responders (did not respond to any of the
mailed requests or telephone calls, or could not be traced, n = 254).
Compared to the postal and non-responders, the full response group consists
of more girls, contains more Dutch children, has higher educational and
social economic levels and has fewer handicaps. The responders form a highly
selective subgroup in the total cohort.
Multiple imputation of this data set has been described in Hille et al (2007)
and Van Buuren (2012), chapter 8.
}
\note{
This dataset is not part of \code{mice}.
}
\examples{
pops <- data(pops)
}
\keyword{datasets}
mice/man/make.method.Rd 0000644 0001762 0000144 00000005004 14334522175 014432 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/method.R
\name{make.method}
\alias{make.method}
\title{Creates a \code{method} argument}
\usage{
make.method(
data,
where = make.where(data),
blocks = make.blocks(data),
defaultMethod = c("pmm", "logreg", "polyreg", "polr")
)
}
\arguments{
\item{data}{A data frame or a matrix containing the incomplete data. Missing
values are coded as \code{NA}.}
\item{where}{A data frame or matrix with logicals of the same dimensions
as \code{data} indicating where in the data the imputations should be
created. The default, \code{where = is.na(data)}, specifies that the
missing data should be imputed. The \code{where} argument may be used to
overimpute observed data, or to skip imputations for selected missing values.
Note: Imputation methods that generate imptutations outside of
\code{mice}, like \code{mice.impute.panImpute()} may depend on a complete
predictor space. In that case, a custom \code{where} matrix can not be
specified.}
\item{blocks}{List of vectors with variable names per block. List elements
may be named to identify blocks. Variables within a block are
imputed by a multivariate imputation method
(see \code{method} argument). By default each variable is placed
into its own block, which is effectively
fully conditional specification (FCS) by univariate models
(variable-by-variable imputation). Only variables whose names appear in
\code{blocks} are imputed. The relevant columns in the \code{where}
matrix are set to \code{FALSE} of variables that are not block members.
A variable may appear in multiple blocks. In that case, it is
effectively re-imputed each time that it is visited.}
\item{defaultMethod}{A vector of length 4 containing the default
imputation methods for 1) numeric data, 2) factor data with 2 levels, 3)
factor data with > 2 unordered levels, and 4) factor data with > 2
ordered levels. By default, the method uses
\code{pmm}, predictive mean matching (numeric data) \code{logreg}, logistic
regression imputation (binary data, factor with 2 levels) \code{polyreg},
polytomous regression imputation for unordered categorical data (factor > 2
levels) \code{polr}, proportional odds model for (ordered, > 2 levels).}
}
\value{
Vector of \code{length(blocks)} element with method names
}
\description{
This helper function creates a valid \code{method} vector. The
\code{method} vector is an argument to the \code{mice} function that
specifies the method for each block.
}
\examples{
make.method(nhanes2)
}
\seealso{
\code{\link{mice}}
}
mice/man/make.blocks.Rd 0000644 0001762 0000144 00000005501 14330031606 014417 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/blocks.R
\name{make.blocks}
\alias{make.blocks}
\title{Creates a \code{blocks} argument}
\usage{
make.blocks(
data,
partition = c("scatter", "collect", "void"),
calltype = "type"
)
}
\arguments{
\item{data}{A \code{data.frame}, character vector with
variable names, or \code{list} with variable names.}
\item{partition}{A character vector of length 1 used to assign
variables to blocks when \code{data} is a \code{data.frame}. Value
\code{"scatter"} (default) will assign each column to it own
block. Value \code{"collect"} assigns all variables to one block,
whereas \code{"void"} produces an empty list.}
\item{calltype}{A character vector of \code{length(block)} elements
that indicates how the imputation model is specified. If
\code{calltype = "type"} (the default), the underlying imputation
model is called by means of the \code{type} argument. The
\code{type} argument for block \code{h} is equivalent to
row \code{h} in the \code{predictorMatrix}.
The alternative is \code{calltype = "formula"}. This will pass
\code{formulas[[h]]} to the underlying imputation
function for block \code{h}, together with the current data.
The \code{calltype} of a block is set automatically during
initialization. Where a choice is possible, calltype
\code{"formula"} is preferred over \code{"type"} since this is
more flexible and extendable. However, what precisely happens
depends also on the capabilities of the imputation
function that is called.}
}
\value{
A named list of character vectors with variables names.
}
\description{
This helper function generates a list of the type needed for
\code{blocks} argument in the \code{[=mice]{mice}} function.
}
\details{
Choices \code{"scatter"} and \code{"collect"} represent to two
extreme scenarios for assigning variables to imputation blocks.
Use \code{"scatter"} to create an imputation model based on
\emph{fully conditionally specification} (FCS). Use \code{"collect"} to
gather all variables to be imputed by a \emph{joint model} (JM).
Scenario's in-between these two extremes represent
\emph{hybrid} imputation models that combine FCS and JM.
Any variable not listed in will not be imputed.
Specification \code{"void"} represents the extreme scenario that
skips imputation of all variables.
A variable may be a member of multiple blocks. The variable will be
re-imputed in each block, so the final imputations for variable
will come from the last block that was executed. This scenario
may be useful where the same complete background factors appear in
multiple imputation blocks.
A variable may appear multiple times within a given block. If a univariate
imputation model is applied to such a block, then the variable is
re-imputed each time as it appears in the block.
}
\examples{
make.blocks(nhanes)
make.blocks(c("age", "sex", "edu"))
}
mice/man/mice.Rd 0000644 0001762 0000144 00000054313 14334522175 013162 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice-package.R, R/mice.R
\docType{package}
\name{mice}
\alias{mice}
\title{\pkg{mice}: Multivariate Imputation by Chained Equations}
\usage{
mice(
data,
m = 5,
method = NULL,
predictorMatrix,
ignore = NULL,
where = NULL,
blocks,
visitSequence = NULL,
formulas,
blots = NULL,
post = NULL,
defaultMethod = c("pmm", "logreg", "polyreg", "polr"),
maxit = 5,
printFlag = TRUE,
seed = NA,
data.init = NULL,
...
)
}
\arguments{
\item{data}{A data frame or a matrix containing the incomplete data. Missing
values are coded as \code{NA}.}
\item{m}{Number of multiple imputations. The default is \code{m=5}.}
\item{method}{Can be either a single string, or a vector of strings with
length \code{length(blocks)}, specifying the imputation method to be
used for each column in data. If specified as a single string, the same
method will be used for all blocks. The default imputation method (when no
argument is specified) depends on the measurement level of the target column,
as regulated by the \code{defaultMethod} argument. Columns that need
not be imputed have the empty method \code{""}. See details.}
\item{predictorMatrix}{A numeric matrix of \code{length(blocks)} rows
and \code{ncol(data)} columns, containing 0/1 data specifying
the set of predictors to be used for each target column.
Each row corresponds to a variable block, i.e., a set of variables
to be imputed. A value of \code{1} means that the column
variable is used as a predictor for the target block (in the rows).
By default, the \code{predictorMatrix} is a square matrix of \code{ncol(data)}
rows and columns with all 1's, except for the diagonal.
Note: For two-level imputation models (which have \code{"2l"} in their names)
other codes (e.g, \code{2} or \code{-2}) are also allowed.}
\item{ignore}{A logical vector of \code{nrow(data)} elements indicating
which rows are ignored when creating the imputation model. The default
\code{NULL} includes all rows that have an observed value of the variable
to imputed. Rows with \code{ignore} set to \code{TRUE} do not influence the
parameters of the imputation model, but are still imputed. We may use the
\code{ignore} argument to split \code{data} into a training set (on which the
imputation model is built) and a test set (that does not influence the
imputation model estimates).
Note: Multivariate imputation methods, like \code{mice.impute.jomoImpute()}
or \code{mice.impute.panImpute()}, do not honour the \code{ignore} argument.}
\item{where}{A data frame or matrix with logicals of the same dimensions
as \code{data} indicating where in the data the imputations should be
created. The default, \code{where = is.na(data)}, specifies that the
missing data should be imputed. The \code{where} argument may be used to
overimpute observed data, or to skip imputations for selected missing values.
Note: Imputation methods that generate imptutations outside of
\code{mice}, like \code{mice.impute.panImpute()} may depend on a complete
predictor space. In that case, a custom \code{where} matrix can not be
specified.}
\item{blocks}{List of vectors with variable names per block. List elements
may be named to identify blocks. Variables within a block are
imputed by a multivariate imputation method
(see \code{method} argument). By default each variable is placed
into its own block, which is effectively
fully conditional specification (FCS) by univariate models
(variable-by-variable imputation). Only variables whose names appear in
\code{blocks} are imputed. The relevant columns in the \code{where}
matrix are set to \code{FALSE} of variables that are not block members.
A variable may appear in multiple blocks. In that case, it is
effectively re-imputed each time that it is visited.}
\item{visitSequence}{A vector of block names of arbitrary length, specifying the
sequence of blocks that are imputed during one iteration of the Gibbs
sampler. A block is a collection of variables. All variables that are
members of the same block are imputed
when the block is visited. A variable that is a member of multiple blocks
is re-imputed within the same iteration.
The default \code{visitSequence = "roman"} visits the blocks (left to right)
in the order in which they appear in \code{blocks}.
One may also use one of the following keywords: \code{"arabic"}
(right to left), \code{"monotone"} (ordered low to high proportion
of missing data) and \code{"revmonotone"} (reverse of monotone).
\emph{Special case}: If you specify both \code{visitSequence = "monotone"} and
\code{maxit = 1}, then the procedure will edit the \code{predictorMatrix}
to conform to the monotone pattern. Realize that convergence in one
iteration is only guaranteed if the missing data pattern is actually
monotone. The procedure does not check this.}
\item{formulas}{A named list of formula's, or expressions that
can be converted into formula's by \code{as.formula}. List elements
correspond to blocks. The block to which the list element applies is
identified by its name, so list names must correspond to block names.
The \code{formulas} argument is an alternative to the
\code{predictorMatrix} argument that allows for more flexibility in
specifying imputation models, e.g., for specifying interaction terms.}
\item{blots}{A named \code{list} of \code{alist}'s that can be used
to pass down arguments to lower level imputation function. The entries
of element \code{blots[[blockname]]} are passed down to the function
called for block \code{blockname}.}
\item{post}{A vector of strings with length \code{ncol(data)} specifying
expressions as strings. Each string is parsed and
executed within the \code{sampler()} function to post-process
imputed values during the iterations.
The default is a vector of empty strings, indicating no post-processing.
Multivariate (block) imputation methods ignore the \code{post} parameter.}
\item{defaultMethod}{A vector of length 4 containing the default
imputation methods for 1) numeric data, 2) factor data with 2 levels, 3)
factor data with > 2 unordered levels, and 4) factor data with > 2
ordered levels. By default, the method uses
\code{pmm}, predictive mean matching (numeric data) \code{logreg}, logistic
regression imputation (binary data, factor with 2 levels) \code{polyreg},
polytomous regression imputation for unordered categorical data (factor > 2
levels) \code{polr}, proportional odds model for (ordered, > 2 levels).}
\item{maxit}{A scalar giving the number of iterations. The default is 5.}
\item{printFlag}{If \code{TRUE}, \code{mice} will print history on console.
Use \code{print=FALSE} for silent computation.}
\item{seed}{An integer that is used as argument by the \code{set.seed()} for
offsetting the random number generator. Default is to leave the random number
generator alone.}
\item{data.init}{A data frame of the same size and type as \code{data},
without missing data, used to initialize imputations before the start of the
iterative process. The default \code{NULL} implies that starting imputation
are created by a simple random draw from the data. Note that specification of
\code{data.init} will start all \code{m} Gibbs sampling streams from the same
imputation.}
\item{\dots}{Named arguments that are passed down to the univariate imputation
functions.}
}
\value{
Returns an S3 object of class \code{\link[=mids-class]{mids}}
(multiply imputed data set)
}
\description{
The \pkg{mice} package implements a method to deal with missing data.
The package creates multiple imputations (replacement values) for
multivariate missing data. The method is based on Fully Conditional
Specification, where each incomplete variable is imputed by a separate
model. The MICE algorithm can impute mixes of continuous, binary,
unordered categorical and ordered categorical data. In addition, MICE
can impute continuous two-level data, and maintain consistency between
imputations by means of passive imputation. Many diagnostic plots are
implemented to inspect the quality of the imputations.
Generates Multivariate Imputations by Chained Equations (MICE)
}
\details{
The \pkg{mice} package contains functions to
\itemize{
\item Inspect the missing data pattern
\item Impute the missing data \emph{m} times, resulting in \emph{m} completed data sets
\item Diagnose the quality of the imputed values
\item Analyze each completed data set
\item Pool the results of the repeated analyses
\item Store and export the imputed data in various formats
\item Generate simulated incomplete data
\item Incorporate custom imputation methods
}
Generates multiple imputations for incomplete multivariate data by Gibbs
sampling. Missing data can occur anywhere in the data. The algorithm imputes
an incomplete column (the target column) by generating 'plausible' synthetic
values given other columns in the data. Each incomplete column must act as a
target column, and has its own specific set of predictors. The default set of
predictors for a given target consists of all other columns in the data. For
predictors that are incomplete themselves, the most recently generated
imputations are used to complete the predictors prior to imputation of the
target column.
A separate univariate imputation model can be specified for each column. The
default imputation method depends on the measurement level of the target
column. In addition to these, several other methods are provided. You can
also write their own imputation functions, and call these from within the
algorithm.
The data may contain categorical variables that are used in a regressions on
other variables. The algorithm creates dummy variables for the categories of
these variables, and imputes these from the corresponding categorical
variable.
Built-in univariate imputation methods are:
\tabular{lll}{
\code{pmm} \tab any \tab Predictive mean matching\cr
\code{midastouch} \tab any \tab Weighted predictive mean matching\cr
\code{sample} \tab any \tab Random sample from observed values\cr
\code{cart} \tab any \tab Classification and regression trees\cr
\code{rf} \tab any \tab Random forest imputations\cr
\code{mean} \tab numeric \tab Unconditional mean imputation\cr
\code{norm} \tab numeric \tab Bayesian linear regression\cr
\code{norm.nob} \tab numeric \tab Linear regression ignoring model error\cr
\code{norm.boot} \tab numeric \tab Linear regression using bootstrap\cr
\code{norm.predict} \tab numeric \tab Linear regression, predicted values\cr
\code{lasso.norm} \tab numeric \tab Lasso linear regression\cr
\code{lasso.select.norm} \tab numeric \tab Lasso select + linear regression\cr
\code{quadratic} \tab numeric \tab Imputation of quadratic terms\cr
\code{ri} \tab numeric \tab Random indicator for nonignorable data\cr
\code{logreg} \tab binary \tab Logistic regression\cr
\code{logreg.boot} \tab binary \tab Logistic regression with bootstrap\cr
\code{lasso.logreg} \tab binary \tab Lasso logistic regression\cr
\code{lasso.select.logreg}\tab binary \tab Lasso select + logistic regression\cr
\code{polr} \tab ordered \tab Proportional odds model\cr
\code{polyreg} \tab unordered\tab Polytomous logistic regression\cr
\code{lda} \tab unordered\tab Linear discriminant analysis\cr
\code{2l.norm} \tab numeric \tab Level-1 normal heteroscedastic\cr
\code{2l.lmer} \tab numeric \tab Level-1 normal homoscedastic, lmer\cr
\code{2l.pan} \tab numeric \tab Level-1 normal homoscedastic, pan\cr
\code{2l.bin} \tab binary \tab Level-1 logistic, glmer\cr
\code{2lonly.mean} \tab numeric \tab Level-2 class mean\cr
\code{2lonly.norm} \tab numeric \tab Level-2 class normal\cr
\code{2lonly.pmm} \tab any \tab Level-2 class predictive mean matching
}
These corresponding functions are coded in the \code{mice} library under
names \code{mice.impute.method}, where \code{method} is a string with the
name of the univariate imputation method name, for example \code{norm}. The
\code{method} argument specifies the methods to be used. For the \code{j}'th
column, \code{mice()} calls the first occurrence of
\code{paste('mice.impute.', method[j], sep = '')} in the search path. The
mechanism allows uses to write customized imputation function,
\code{mice.impute.myfunc}. To call it for all columns specify
\code{method='myfunc'}. To call it only for, say, column 2 specify
\code{method=c('norm','myfunc','logreg',\dots{})}.
\emph{Skipping imputation:} The user may skip imputation of a column by
setting its entry to the empty method: \code{""}. For complete columns without
missing data \code{mice} will automatically set the empty method. Setting t
he empty method does not produce imputations for the column, so any missing
cells remain \code{NA}. If column A contains \code{NA}'s and is used as
predictor in the imputation model for column B, then \code{mice} produces no
imputations for the rows in B where A is missing. The imputed data
for B may thus contain \code{NA}'s. The remedy is to remove column A from
the imputation model for the other columns in the data. This can be done
by setting the entire column for variable A in the \code{predictorMatrix}
equal to zero.
\emph{Passive imputation:} \code{mice()} supports a special built-in method,
called passive imputation. This method can be used to ensure that a data
transform always depends on the most recently generated imputations. In some
cases, an imputation model may need transformed data in addition to the
original data (e.g. log, quadratic, recodes, interaction, sum scores, and so
on).
Passive imputation maintains consistency among different transformations of
the same data. Passive imputation is invoked if \code{~} is specified as the
first character of the string that specifies the univariate method.
\code{mice()} interprets the entire string, including the \code{~} character,
as the formula argument in a call to \code{model.frame(formula,
data[!r[,j],])}. This provides a simple mechanism for specifying deterministic
dependencies among the columns. For example, suppose that the missing entries
in variables \code{data$height} and \code{data$weight} are imputed. The body
mass index (BMI) can be calculated within \code{mice} by specifying the
string \code{'~I(weight/height^2)'} as the univariate imputation method for
the target column \code{data$bmi}. Note that the \code{~} mechanism works
only on those entries which have missing values in the target column. You
should make sure that the combined observed and imputed parts of the target
column make sense. An easy way to create consistency is by coding all entries
in the target as \code{NA}, but for large data sets, this could be
inefficient. Note that you may also need to adapt the default
\code{predictorMatrix} to evade linear dependencies among the predictors that
could cause errors like \code{Error in solve.default()} or \code{Error:
system is exactly singular}. Though not strictly needed, it is often useful
to specify \code{visitSequence} such that the column that is imputed by the
\code{~} mechanism is visited each time after one of its predictors was
visited. In that way, deterministic relation between columns will always be
synchronized.
#'A new argument \code{ls.meth} can be parsed to the lower level
\code{.norm.draw} to specify the method for generating the least squares
estimates and any subsequently derived estimates. Argument \code{ls.meth}
takes one of three inputs: \code{"qr"} for QR-decomposition, \code{"svd"} for
singular value decomposition and \code{"ridge"} for ridge regression.
\code{ls.meth} defaults to \code{ls.meth = "qr"}.
\emph{Auxiliary predictors in formulas specification: }
For a given block, the \code{formulas} specification takes precedence over
the corresponding row in the \code{predictMatrix} argument. This
precedence is, however, restricted to the subset of variables
specified in the terms of the block formula. Any
variables not specified by \code{formulas} are imputed
according to the \code{predictMatrix} specification. Variables with
non-zero \code{type} values in the \code{predictMatrix} will
be added as main effects to the \code{formulas}, which will
act as supplementary covariates in the imputation model. It is possible
to turn off this behavior by specifying the
argument \code{auxiliary = FALSE}.
}
\section{Functions}{
The main functions are:
\tabular{ll}{
\code{mice()} \tab Impute the missing data *m* times\cr
\code{with()} \tab Analyze completed data sets\cr
\code{pool()} \tab Combine parameter estimates\cr
\code{complete()} \tab Export imputed data\cr
\code{ampute()} \tab Generate missing data\cr}
}
\section{Vignettes}{
There is a detailed series of
six online vignettes that walk you through solving realistic inference
problems with mice.
We suggest going through these vignettes in the following order
\enumerate{
\item \href{https://www.gerkovink.com/miceVignettes/Ad_hoc_and_mice/Ad_hoc_methods.html}{Ad hoc methods and the MICE algorithm}
\item \href{https://www.gerkovink.com/miceVignettes/Convergence_pooling/Convergence_and_pooling.html}{Convergence and pooling}
\item \href{https://www.gerkovink.com/miceVignettes/Missingness_inspection/Missingness_inspection.html}{Inspecting how the observed data and missingness are related}
\item \href{https://www.gerkovink.com/miceVignettes/Passive_Post_processing/Passive_imputation_post_processing.html}{Passive imputation and post-processing}
\item \href{https://www.gerkovink.com/miceVignettes/Multi_level/Multi_level_data.html}{Imputing multilevel data}
\item \href{https://www.gerkovink.com/miceVignettes/Sensitivity_analysis/Sensitivity_analysis.html}{Sensitivity analysis with \pkg{mice}}
}
#'Van Buuren, S. (2018).
Boca Raton, FL.: Chapman & Hall/CRC Press.
The book
\href{https://stefvanbuuren.name/fimd/}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
contains a lot of \href{https://github.com/stefvanbuuren/fimdbook/tree/master/R}{example code}.
}
\section{Methodology}{
The \pkg{mice} software was published in the {Journal of Statistical Software} (Van Buuren and Groothuis-Oudshoorn, 2011). \doi{10.18637/jss.v045.i03}
The first application of the method
concerned missing blood pressure data (Van Buuren et. al., 1999).
The term \emph{Fully Conditional Specification} was introduced in 2006 to describe a general class of methods that specify imputations model for multivariate data as a set of conditional distributions (Van Buuren et. al., 2006). Further details on mixes of variables and applications can be found in the book
\href{https://stefvanbuuren.name/fimd/}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\section{Enhanced linear algebra}{
Updating the BLAS can improve speed of R, sometime considerably. The details
depend on the operating system. See the discussion in the
"R Installation and Administration" guide for further information.
}
\examples{
# do default multiple imputation on a numeric matrix
imp <- mice(nhanes)
imp
# list the actual imputations for BMI
imp$imp$bmi
# first completed data matrix
complete(imp)
# imputation on mixed data with a different method per column
mice(nhanes2, meth = c("sample", "pmm", "logreg", "norm"))
\dontrun{
# example where we fit the imputation model on the train data
# and apply the model to impute the test data
set.seed(123)
ignore <- sample(c(TRUE, FALSE), size = 25, replace = TRUE, prob = c(0.3, 0.7))
# scenario 1: train and test in the same dataset
imp <- mice(nhanes2, m = 2, ignore = ignore, print = FALSE, seed = 22112)
imp.test1 <- filter(imp, ignore)
imp.test1$data
complete(imp.test1, 1)
complete(imp.test1, 2)
# scenario 2: train and test in separate datasets
traindata <- nhanes2[!ignore, ]
testdata <- nhanes2[ignore, ]
imp.train <- mice(traindata, m = 2, print = FALSE, seed = 22112)
imp.test2 <- mice.mids(imp.train, newdata = testdata)
complete(imp.test2, 1)
complete(imp.test2, 2)
}
}
\references{
van Buuren, S., Boshuizen, H.C., Knook, D.L. (1999) Multiple
imputation of missing blood pressure covariates in survival analysis.
\emph{Statistics in Medicine}, \bold{18}, 681--694.
van Buuren, S., Brand, J.P.L., Groothuis-Oudshoorn C.G.M., Rubin, D.B. (2006)
Fully conditional specification in multivariate imputation. \emph{Journal of
Statistical Computation and Simulation}, \bold{76}, 12, 1049--1064.
van Buuren, S., Groothuis-Oudshoorn, K. (2011). {\code{mice}:
Multivariate Imputation by Chained Equations in \code{R}}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1--67. \doi{10.18637/jss.v045.i03}
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-FCS.html#sec:MICE}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
Van Buuren, S., Brand, J.P.L., Groothuis-Oudshoorn C.G.M., Rubin, D.B. (2006)
Fully conditional specification in multivariate imputation. \emph{Journal of
Statistical Computation and Simulation}, \bold{76}, 12, 1049--1064.
Van Buuren, S. (2007) Multiple imputation of discrete and continuous data by
fully conditional specification. \emph{Statistical Methods in Medical
Research}, \bold{16}, 3, 219--242.
Van Buuren, S., Boshuizen, H.C., Knook, D.L. (1999) Multiple imputation of
missing blood pressure covariates in survival analysis. \emph{Statistics in
Medicine}, \bold{18}, 681--694.
Brand, J.P.L. (1999) \emph{Development, implementation and evaluation of
multiple imputation strategies for the statistical analysis of incomplete
data sets.} Dissertation. Rotterdam: Erasmus University.
}
\seealso{
\code{\link{mice}}, \code{\link{with.mids}},
\code{\link{pool}}, \code{\link{complete}}, \code{\link{ampute}}
\code{\link[=mids-class]{mids}}, \code{\link{with.mids}},
\code{\link{set.seed}}, \code{\link{complete}}
}
\author{
Stef van Buuren \email{stef.vanbuuren@tno.nl}, Karin
Groothuis-Oudshoorn \email{c.g.m.oudshoorn@utwente.nl}, 2000-2010, with
contributions of Alexander Robitzsch, Gerko Vink, Shahab Jolani,
Roel de Jong, Jason Turner, Lisa Doove,
John Fox, Frank E. Harrell, and Peter Malewski.
}
\keyword{iteration}
mice/man/ampute.default.patterns.Rd 0000644 0001762 0000144 00000001174 14330031606 017005 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/ampute.default.R
\name{ampute.default.patterns}
\alias{ampute.default.patterns}
\title{Default \code{patterns} in \code{ampute}}
\usage{
ampute.default.patterns(n)
}
\arguments{
\item{n}{A scalar specifying the number of variables in the data.}
}
\value{
A square matrix of size \code{n} where \code{0} indicates a variable
}
\description{
This function creates a default pattern matrix for the multivariate
amputation function \code{ampute()}.
}
\seealso{
\code{\link{ampute}}, \code{\link{md.pattern}}
}
\author{
Rianne Schouten, 2016
}
\keyword{internal}
mice/man/is.mids.Rd 0000644 0001762 0000144 00000000515 13666252075 013614 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/is.R
\name{is.mids}
\alias{is.mids}
\title{Check for \code{mids} object}
\usage{
is.mids(x)
}
\arguments{
\item{x}{An object}
}
\value{
A logical indicating whether \code{x} is an object of class \code{mids}
}
\description{
Check for \code{mids} object
}
mice/man/mads-class.Rd 0000644 0001762 0000144 00000006567 14330031606 014272 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mads.R
\docType{class}
\name{mads-class}
\alias{mads-class}
\title{Multivariate amputed data set (\code{mads})}
\description{
The \code{mads} object contains an amputed data set. The \code{mads} object is
generated by the \code{ampute} function. The \code{mads} class of objects has
methods for the following generic functions: \code{print}, \code{summary},
\code{bwplot} and \code{xyplot}.
}
\note{
Many of the functions of the \code{mice} package do not use the S4 class
definitions, and instead rely on the S3 list equivalent
\code{oldClass(obj) <- "mads"}.
}
\section{Contents}{
\describe{
\item{\code{call}:}{The function call.}
\item{\code{prop}:}{Proportion of cases with missing values. Note: even when
the proportion is entered as the proportion of missing cells (when
\code{bycases == TRUE}), this object contains the proportion of missing cases.}
\item{\code{patterns}:}{A data frame of size #patterns by #variables where \code{0}
indicates a variable has missing values and \code{1} indicates a variable remains
complete.}
\item{\code{freq}:}{A vector of length #patterns containing the relative
frequency with which the patterns occur. For example, if the vector is
\code{c(0.4, 0.4, 0.2)}, this means that of all cases with missing values,
40 percent is candidate for pattern 1, 40 percent for pattern 2 and 20
percent for pattern 3. The vector sums to 1.}
\item{\code{mech}:}{A string specifying the missingness mechanism, either
\code{"MCAR"}, \code{"MAR"} or \code{"MNAR"}.}
\item{\code{weights}:}{A data frame of size #patterns by #variables. It contains
the weights that were used to calculate the weighted sum scores. The weights
may differ between patterns and between variables.}
\item{\code{cont}:}{Logical, whether probabilities are based on continuous logit
functions or on discrete odds distributions.}
\item{\code{type}:}{A vector of strings containing the type of missingness
for each pattern. Either \code{"LEFT"}, \code{"MID"}, \code{"TAIL"} or
\code{"RIGHT"}. The first type refers to the first pattern, the second type
to the second pattern, etc.}
\item{\code{odds}:}{A matrix where #patterns defines the #rows. Each row contains
the odds of being missing for the corresponding pattern. The amount of odds values
defines in how many quantiles the sum scores were divided. The values are
relative probabilities: a quantile with odds value 4 will have a probability of
being missing that is four times higher than a quantile with odds 1. The
#quantiles may differ between patterns, NA is used for cells remaining empty.}
\item{\code{amp}:}{A data frame containing the input data with NAs for the
amputed values.}
\item{\code{cand}:}{A vector that contains the pattern number for each case.
A value between 1 and #patterns is given. For example, a case with value 2 is
candidate for missing data pattern 2.}
\item{\code{scores}:}{A list containing vectors with weighted sum scores of the
candidates. The first vector refers to the candidates of the first pattern, the
second vector refers to the candidates of the second pattern, etc. The length
of the vectors differ because the number of candidates is different for each
pattern.}
\item{\code{data}:}{The complete data set that was entered in \code{ampute}.}
}
}
\seealso{
\code{\link{ampute}}, Vignette titled "Multivariate Amputation using
Ampute".
}
\author{
Rianne Schouten, 2016
}
mice/man/stripplot.mids.Rd 0000644 0001762 0000144 00000021441 14330031647 015227 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/stripplot.R
\name{stripplot.mids}
\alias{stripplot.mids}
\alias{stripplot}
\title{Stripplot of observed and imputed data}
\usage{
\method{stripplot}{mids}(
x,
data,
na.groups = NULL,
groups = NULL,
as.table = TRUE,
theme = mice.theme(),
allow.multiple = TRUE,
outer = TRUE,
drop.unused.levels = lattice::lattice.getOption("drop.unused.levels"),
panel = lattice::lattice.getOption("panel.stripplot"),
default.prepanel = lattice::lattice.getOption("prepanel.default.stripplot"),
jitter.data = TRUE,
horizontal = FALSE,
...,
subscripts = TRUE,
subset = TRUE
)
}
\arguments{
\item{x}{A \code{mids} object, typically created by \code{mice()} or
\code{mice.mids()}.}
\item{data}{Formula that selects the data to be plotted. This argument
follows the \pkg{lattice} rules for \emph{formulas}, describing the primary
variables (used for the per-panel display) and the optional conditioning
variables (which define the subsets plotted in different panels) to be used
in the plot.
The formula is evaluated on the complete data set in the \code{long} form.
Legal variable names for the formula include \code{names(x$data)} plus the
two administrative factors \code{.imp} and \code{.id}.
\bold{Extended formula interface:} The primary variable terms (both the LHS
\code{y} and RHS \code{x}) may consist of multiple terms separated by a
\sQuote{+} sign, e.g., \code{y1 + y2 ~ x | a * b}. This formula would be
taken to mean that the user wants to plot both \code{y1 ~ x | a * b} and
\code{y2 ~ x | a * b}, but with the \code{y1 ~ x} and \code{y2 ~ x} in
\emph{separate panels}. This behavior differs from standard \pkg{lattice}.
\emph{Only combine terms of the same type}, i.e. only factors or only
numerical variables. Mixing numerical and categorical data occasionally
produces odds labeling of vertical axis.
For convenience, in \code{stripplot()} and \code{bwplot} the formula
\code{y~.imp} may be abbreviated as \code{y}. This applies only to a single
\code{y}, and does not (yet) work for \code{y1+y2~.imp}.}
\item{na.groups}{An expression evaluating to a logical vector indicating
which two groups are distinguished (e.g. using different colors) in the
display. The environment in which this expression is evaluated in the
response indicator \code{is.na(x$data)}.
The default \code{na.group = NULL} contrasts the observed and missing data
in the LHS \code{y} variable of the display, i.e. groups created by
\code{is.na(y)}. The expression \code{y} creates the groups according to
\code{is.na(y)}. The expression \code{y1 & y2} creates groups by
\code{is.na(y1) & is.na(y2)}, and \code{y1 | y2} creates groups as
\code{is.na(y1) | is.na(y2)}, and so on.}
\item{groups}{This is the usual \code{groups} arguments in \pkg{lattice}. It
differs from \code{na.groups} because it evaluates in the completed data
\code{data.frame(complete(x, "long", inc=TRUE))} (as usual), whereas
\code{na.groups} evaluates in the response indicator. See
\code{\link{xyplot}} for more details. When both \code{na.groups} and
\code{groups} are specified, \code{na.groups} takes precedence, and
\code{groups} is ignored.}
\item{as.table}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{theme}{A named list containing the graphical parameters. The default
function \code{mice.theme} produces a short list of default colors, line
width, and so on. The extensive list may be obtained from
\code{trellis.par.get()}. Global graphical parameters like \code{col} or
\code{cex} in high-level calls are still honored, so first experiment with
the global parameters. Many setting consists of a pair. For example,
\code{mice.theme} defines two symbol colors. The first is for the observed
data, the second for the imputed data. The theme settings only exist during
the call, and do not affect the trellis graphical parameters.}
\item{allow.multiple}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{outer}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{drop.unused.levels}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{panel}{See \code{\link{xyplot}}.}
\item{default.prepanel}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{jitter.data}{See \code{\link[lattice:panel.xyplot]{panel.xyplot}}.}
\item{horizontal}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{\dots}{Further arguments, usually not directly processed by the
high-level functions documented here, but instead passed on to other
functions.}
\item{subscripts}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{subset}{See \code{\link[lattice:xyplot]{xyplot}}.}
}
\value{
The high-level functions documented here, as well as other high-level
Lattice functions, return an object of class \code{"trellis"}. The
\code{\link[lattice:update.trellis]{update}} method can be used to
subsequently update components of the object, and the
\code{\link[lattice:print.trellis]{print}} method (usually called by default)
will plot it on an appropriate plotting device.
}
\description{
Plotting methods for imputed data using \pkg{lattice}.
\code{stripplot} produces one-dimensional
scatterplots. The function
automatically separates the observed and imputed data. The
functions extend the usual features of \pkg{lattice}.
}
\details{
The argument \code{na.groups} may be used to specify (combinations of)
missingness in any of the variables. The argument \code{groups} can be used
to specify groups based on the variable values themselves. Only one of both
may be active at the same time. When both are specified, \code{na.groups}
takes precedence over \code{groups}.
Use the \code{subset} and \code{na.groups} together to plots parts of the
data. For example, select the first imputed data set by by
\code{subset=.imp==1}.
Graphical parameters like \code{col}, \code{pch} and \code{cex} can be
specified in the arguments list to alter the plotting symbols. If
\code{length(col)==2}, the color specification to define the observed and
missing groups. \code{col[1]} is the color of the 'observed' data,
\code{col[2]} is the color of the missing or imputed data. A convenient color
choice is \code{col=mdc(1:2)}, a transparent blue color for the observed
data, and a transparent red color for the imputed data. A good choice is
\code{col=mdc(1:2), pch=20, cex=1.5}. These choices can be set for the
duration of the session by running \code{mice.theme()}.
}
\note{
The first two arguments (\code{x} and \code{data}) are reversed
compared to the standard Trellis syntax implemented in \pkg{lattice}. This
reversal was necessary in order to benefit from automatic method dispatch.
In \pkg{mice} the argument \code{x} is always a \code{mids} object, whereas
in \pkg{lattice} the argument \code{x} is always a formula.
In \pkg{mice} the argument \code{data} is always a formula object, whereas in
\pkg{lattice} the argument \code{data} is usually a data frame.
All other arguments have identical interpretation.
}
\examples{
imp <- mice(boys, maxit = 1)
### stripplot, all numerical variables
\dontrun{
stripplot(imp)
}
### same, but with improved display
\dontrun{
stripplot(imp, col = c("grey", mdc(2)), pch = c(1, 20))
}
### distribution per imputation of height, weight and bmi
### labeled by their own missingness
\dontrun{
stripplot(imp, hgt + wgt + bmi ~ .imp,
cex = c(2, 4), pch = c(1, 20), jitter = FALSE,
layout = c(3, 1)
)
}
### same, but labeled with the missingness of wgt (just four cases)
\dontrun{
stripplot(imp, hgt + wgt + bmi ~ .imp,
na = wgt, cex = c(2, 4), pch = c(1, 20), jitter = FALSE,
layout = c(3, 1)
)
}
### distribution of age and height, labeled by missingness in height
### most height values are missing for those around
### the age of two years
### some additional missings occur in region WEST
\dontrun{
stripplot(imp, age + hgt ~ .imp | reg, hgt,
col = c(grDevices::hcl(0, 0, 40, 0.2), mdc(2)), pch = c(1, 20)
)
}
### heavily jitted relation between two categorical variables
### labeled by missingness of gen
### aggregated over all imputed data sets
\dontrun{
stripplot(imp, gen ~ phb, factor = 2, cex = c(8, 1), hor = TRUE)
}
### circle fun
stripplot(imp, gen ~ .imp,
na = wgt, factor = 2, cex = c(8.6),
hor = FALSE, outer = TRUE, scales = "free", pch = c(1, 19)
)
}
\references{
Sarkar, Deepayan (2008) \emph{Lattice: Multivariate Data
Visualization with R}, Springer.
van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}: Multivariate
Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
}
\seealso{
\code{\link{mice}}, \code{\link{xyplot}}, \code{\link{densityplot}},
\code{\link{bwplot}}, \code{\link{lattice}} for an overview of the
package, as well as \code{\link[lattice:xyplot]{stripplot}},
\code{\link[lattice:panel.stripplot]{panel.stripplot}},
\code{\link[lattice:print.trellis]{print.trellis}},
\code{\link[lattice:trellis.par.get]{trellis.par.set}}
}
\author{
Stef van Buuren
}
\keyword{hplot}
mice/man/parlmice.Rd 0000644 0001762 0000144 00000007232 14332750363 014037 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/parlmice.R
\name{parlmice}
\alias{parlmice}
\title{Wrapper function that runs MICE in parallel}
\usage{
parlmice(
data,
m = 5,
seed = NA,
cluster.seed = NA,
n.core = NULL,
n.imp.core = NULL,
cl.type = "PSOCK",
...
)
}
\arguments{
\item{data}{A data frame or matrix containing the incomplete data. Similar to
the first argument of \code{\link{mice}}.}
\item{m}{The number of desired imputated datasets. By default $m=5$ as with \code{mice}}
\item{seed}{A scalar to be used as the seed value for the mice algorithm within
each parallel stream. Please note that the imputations will be the same for all
streams and, hence, this should be used if and only if \code{n.core = 1} and
if it is desired to obtain the same output as under \code{mice}.}
\item{cluster.seed}{A scalar to be used as the seed value. It is recommended to put the
seed value here and not outside this function, as otherwise the parallel processes
will be performed with separate, random seeds.}
\item{n.core}{A scalar indicating the number of cores that should be used.}
\item{n.imp.core}{A scalar indicating the number of imputations per core.}
\item{cl.type}{The cluster type. Default value is \code{"PSOCK"}. Posix machines (linux, Mac)
generally benefit from much faster cluster computation if \code{type} is set to \code{type = "FORK"}.}
\item{...}{Named arguments that are passed down to function \code{\link{mice}} or
\code{\link{makeCluster}}.}
}
\value{
A mids object as defined by \code{\link{mids-class}}
}
\description{
This function is included for backward compatibility. The function
is superseded by \code{\link{futuremice}}.
}
\details{
This function relies on package \code{\link{parallel}}, which is a base
package for R versions 2.14.0 and later. We have chosen to use parallel function
\code{parLapply} to allow the use of \code{parlmice} on Mac, Linux and Windows
systems. For the same reason, we use the Parallel Socket Cluster (PSOCK) type by default.
On systems other than Windows, it can be hugely beneficial to change the cluster type to
\code{FORK}, as it generally results in improved memory handling. When memory issues
arise on a Windows system, we advise to store the multiply imputed datasets,
clean the memory by using \code{\link{rm}} and \code{\link{gc}} and make another
run using the same settings.
This wrapper function combines the output of \code{\link{parLapply}} with
function \code{\link{ibind}} in \code{\link{mice}}. A \code{mids} object is returned
and can be used for further analyses.
Note that if a seed value is desired, the seed should be entered to this function
with argument \code{seed}. Seed values outside the wrapper function (in an
R-script or passed to \code{\link{mice}}) will not result to reproducible results.
We refer to the manual of \code{\link{parallel}} for an explanation on this matter.
}
\examples{
# 150 imputations in dataset nhanes, performed by 3 cores
\dontrun{
imp1 <- parlmice(data = nhanes, n.core = 3, n.imp.core = 50)
# Making use of arguments in mice.
imp2 <- parlmice(data = nhanes, method = "norm.nob", m = 100)
imp2$method
fit <- with(imp2, lm(bmi ~ hyp))
pool(fit)
}
}
\references{
Schouten, R. and Vink, G. (2017). parlmice: faster, paraleller, micer.
\url{https://www.gerkovink.com/parlMICE/Vignette_parlMICE.html}
#'Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/parallel-computation.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\seealso{
\code{\link{parallel}}, \code{\link{parLapply}}, \code{\link{makeCluster}},
\code{\link{mice}}, \code{\link{mids-class}}
}
\author{
Gerko Vink, Rianne Schouten
}
mice/man/mice.impute.norm.boot.Rd 0000644 0001762 0000144 00000004546 14330031647 016376 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.norm.boot.R
\name{mice.impute.norm.boot}
\alias{mice.impute.norm.boot}
\alias{norm.boot}
\title{Imputation by linear regression, bootstrap method}
\usage{
mice.impute.norm.boot(y, ry, x, wy = NULL, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes univariate missing data using linear regression with bootstrap
}
\details{
Draws a bootstrap sample from \code{x[ry,]} and \code{y[ry]}, calculates
regression weights and imputes with normal residuals.
}
\references{
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
}
\seealso{
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Gerko Vink, Stef van Buuren, 2018
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/D1.Rd 0000644 0001762 0000144 00000003564 14436064333 012513 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/D1.R
\name{D1}
\alias{D1}
\title{Compare two nested models using D1-statistic}
\usage{
D1(fit1, fit0 = NULL, dfcom = NULL, df.com = NULL)
}
\arguments{
\item{fit1}{An object of class \code{mira}, produced by \code{with()}.}
\item{fit0}{An object of class \code{mira}, produced by \code{with()}. The
model in \code{fit0} is a nested within \code{fit1}. The default null
model \code{fit0 = NULL} compares \code{fit1} to the intercept-only model.}
\item{dfcom}{A single number denoting the
complete-data degrees of freedom of model \code{fit1}. If not specified,
it is set equal to \code{df.residual} of model \code{fit1}. If that cannot
be done, the procedure assumes (perhaps incorrectly) a large sample.}
\item{df.com}{Deprecated}
}
\description{
The D1-statistics is the multivariate Wald test.
}
\note{
Warning: `D1()` assumes that the order of the variables is the
same in different models. See
\url{https://github.com/amices/mice/issues/420} for details.
}
\examples{
# Compare two linear models:
imp <- mice(nhanes2, seed = 51009, print = FALSE)
mi1 <- with(data = imp, expr = lm(bmi ~ age + hyp + chl))
mi0 <- with(data = imp, expr = lm(bmi ~ age + hyp))
D1(mi1, mi0)
\dontrun{
# Compare two logistic regression models
imp <- mice(boys, maxit = 2, print = FALSE)
fit1 <- with(imp, glm(gen > levels(gen)[1] ~ hgt + hc + reg, family = binomial))
fit0 <- with(imp, glm(gen > levels(gen)[1] ~ hgt + hc, family = binomial))
D1(fit1, fit0)
}
}
\references{
Li, K. H., T. E. Raghunathan, and D. B. Rubin. 1991.
Large-Sample Significance Levels from Multiply Imputed Data Using
Moment-Based Statistics and an F Reference Distribution.
\emph{Journal of the American Statistical Association}, 86(416): 1065–73.
\url{https://stefvanbuuren.name/fimd/sec-multiparameter.html#sec:wald}
}
\seealso{
\code{\link[mitml]{testModels}}
}
mice/man/pool.scalar.Rd 0000644 0001762 0000144 00000006773 14334522175 014471 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/pool.scalar.R
\name{pool.scalar}
\alias{pool.scalar}
\alias{pool.scalar.syn}
\title{Multiple imputation pooling: univariate version}
\usage{
pool.scalar(Q, U, n = Inf, k = 1, rule = c("rubin1987", "reiter2003"))
pool.scalar.syn(Q, U, n = Inf, k = 1, rule = "reiter2003")
}
\arguments{
\item{Q}{A vector of univariate estimates of \code{m} repeated complete data
analyses.}
\item{U}{A vector containing the corresponding \code{m} variances of the univariate
estimates.}
\item{n}{A number providing the sample size. If nothing is specified,
an infinite sample \code{n = Inf} is assumed.}
\item{k}{A number indicating the number of parameters to be estimated.
By default, \code{k = 1} is assumed.}
\item{rule}{A string indicating the pooling rule. Currently supported are
\code{"rubin1987"} (default, for missing data) and \code{"reiter2003"}
(for synthetic data created from a complete data set).}
}
\value{
Returns a list with components.
\describe{
\item{\code{m}:}{Number of imputations.}
\item{\code{qhat}:}{The \code{m} univariate estimates of repeated complete-data analyses.}
\item{\code{u}:}{The corresponding \code{m} variances of the univariate estimates.}
\item{\code{qbar}:}{The pooled univariate estimate, formula (3.1.2) Rubin (1987).}
\item{\code{ubar}:}{The mean of the variances (i.e. the pooled within-imputation variance),
formula (3.1.3) Rubin (1987).}
\item{\code{b}:}{The between-imputation variance, formula (3.1.4) Rubin (1987).}
\item{\code{t}:}{The total variance of the pooled estimated, formula (3.1.5)
Rubin (1987).}
\item{\code{r}:}{The relative increase in variance due to nonresponse, formula
(3.1.7) Rubin (1987).}
\item{\code{df}:}{The degrees of freedom for t reference distribution by the
method of Barnard-Rubin (1999).}
\item{\code{fmi}:}{The fraction missing information due to nonresponse,
formula (3.1.10) Rubin (1987). (Not defined for synthetic data.)}
}
}
\description{
Pools univariate estimates of m repeated complete data analysis
}
\details{
The function averages the univariate estimates of the complete data model,
computes the total variance over the repeated analyses, and computes the
relative increase in variance due to missing data or data synthesisation
and the fraction of missing information.
}
\examples{
# missing data imputation with with manual pooling
imp <- mice(nhanes, maxit = 2, m = 2, print = FALSE, seed = 18210)
fit <- with(data = imp, lm(bmi ~ age))
# manual pooling
summary(fit$analyses[[1]])
summary(fit$analyses[[2]])
pool.scalar(Q = c(-1.5457, -1.428), U = c(0.9723^2, 1.041^2), n = 25, k = 2)
# check: automatic pooling using broom
pool(fit)
# manual pooling for synthetic data created from complete data
imp <- mice(cars,
maxit = 2, m = 2, print = FALSE, seed = 18210,
where = matrix(TRUE, nrow(cars), ncol(cars))
)
fit <- with(data = imp, lm(speed ~ dist))
# manual pooling: extract Q and U
summary(fit$analyses[[1]])
summary(fit$analyses[[2]])
pool.scalar.syn(Q = c(0.12182, 0.13209), U = c(0.02121^2, 0.02516^2), n = 50, k = 2)
# check: automatic pooling using broom
pool.syn(fit)
}
\references{
Rubin, D.B. (1987). Multiple Imputation for Nonresponse in
Surveys. New York: John Wiley and Sons.
Reiter, J.P. (2003). Inference for Partially Synthetic,
Public Use Microdata Sets. \emph{Survey Methodology}, \bold{29}, 181-189.
}
\seealso{
\code{\link{pool}}
}
\author{
Karin Groothuis-Oudshoorn and Stef van Buuren, 2009; Thom Volker, 2021
}
mice/man/name.blocks.Rd 0000644 0001762 0000144 00000003226 14330031606 014424 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/blocks.R
\name{name.blocks}
\alias{name.blocks}
\title{Name imputation blocks}
\usage{
name.blocks(blocks, prefix = "B")
}
\arguments{
\item{blocks}{List of vectors with variable names per block. List elements
may be named to identify blocks. Variables within a block are
imputed by a multivariate imputation method
(see \code{method} argument). By default each variable is placed
into its own block, which is effectively
fully conditional specification (FCS) by univariate models
(variable-by-variable imputation). Only variables whose names appear in
\code{blocks} are imputed. The relevant columns in the \code{where}
matrix are set to \code{FALSE} of variables that are not block members.
A variable may appear in multiple blocks. In that case, it is
effectively re-imputed each time that it is visited.}
\item{prefix}{A character vector of length 1 with the prefix to
be using for naming any unnamed blocks with two or more variables.}
}
\value{
A named list of character vectors with variables names.
}
\description{
This helper function names any unnamed elements in the \code{blocks}
specification. This is a convenience function.
}
\details{
This function will name any unnamed list elements specified in
the optional argument \code{blocks}. Unnamed blocks
consisting of just one variable will be named after this variable.
Unnamed blocks containing more than one variables will be named
by the \code{prefix} argument, padded by an integer sequence
stating at 1.
}
\examples{
blocks <- list(c("hyp", "chl"), AGE = "age", c("bmi", "hyp"), "edu")
name.blocks(blocks)
}
\seealso{
\code{\link{mice}}
}
mice/man/quickpred.Rd 0000644 0001762 0000144 00000010765 14330031647 014232 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/quickpred.R
\name{quickpred}
\alias{quickpred}
\title{Quick selection of predictors from the data}
\usage{
quickpred(
data,
mincor = 0.1,
minpuc = 0,
include = "",
exclude = "",
method = "pearson"
)
}
\arguments{
\item{data}{Matrix or data frame with incomplete data.}
\item{mincor}{A scalar, numeric vector (of size \code{ncol(data))} or numeric
matrix (square, of size \code{ncol(data)} specifying the minimum
threshold(s) against which the absolute correlation in the data is compared.}
\item{minpuc}{A scalar, vector (of size \code{ncol(data))} or matrix (square,
of size \code{ncol(data)} specifying the minimum threshold(s) for the
proportion of usable cases.}
\item{include}{A string or a vector of strings containing one or more
variable names from \code{names(data)}. Variables specified are always
included as a predictor.}
\item{exclude}{A string or a vector of strings containing one or more
variable names from \code{names(data)}. Variables specified are always
excluded as a predictor.}
\item{method}{A string specifying the type of correlation. Use
\code{'pearson'} (default), \code{'kendall'} or \code{'spearman'}. Can be
abbreviated.}
}
\value{
A square binary matrix of size \code{ncol(data)}.
}
\description{
Selects predictors according to simple statistics
}
\details{
This function creates a predictor matrix using the variable selection
procedure described in Van Buuren et al.~(1999, p.~687--688). The function is
designed to aid in setting up a good imputation model for data with many
variables.
Basic workings: The procedure calculates for each variable pair (i.e.
target-predictor pair) two correlations using all available cases per pair.
The first correlation uses the values of the target and the predictor
directly. The second correlation uses the (binary) response indicator of the
target and the values of the predictor. If the largest (in absolute value) of
these correlations exceeds \code{mincor}, the predictor will be added to the
imputation set. The default value for \code{mincor} is 0.1.
In addition, the procedure eliminates predictors whose proportion of usable
cases fails to meet the minimum specified by \code{minpuc}. The default value
is 0, so predictors are retained even if they have no usable case.
Finally, the procedure includes any predictors named in the \code{include}
argument (which is useful for background variables like age and sex) and
eliminates any predictor named in the \code{exclude} argument. If a variable
is listed in both \code{include} and \code{exclude} arguments, the
\code{include} argument takes precedence.
Advanced topic: \code{mincor} and \code{minpuc} are typically specified as
scalars, but vectors and squares matrices of appropriate size will also work.
Each element of the vector corresponds to a row of the predictor matrix, so
the procedure can effectively differentiate between different target
variables. Setting a high values for can be useful for auxiliary, less
important, variables. The set of predictor for those variables can remain
relatively small. Using a square matrix extends the idea to the columns, so
that one can also apply cellwise thresholds.
}
\note{
\code{quickpred()} uses \code{\link[base]{data.matrix}} to convert
factors to numbers through their internal codes. Especially for unordered
factors the resulting quantification may not make sense.
}
\examples{
# default: include all predictors with absolute correlation over 0.1
quickpred(nhanes)
# all predictors with absolute correlation over 0.4
quickpred(nhanes, mincor = 0.4)
# include age and bmi, exclude chl
quickpred(nhanes, mincor = 0.4, inc = c("age", "bmi"), exc = "chl")
# only include predictors with at least 30\% usable cases
quickpred(nhanes, minpuc = 0.3)
# use low threshold for bmi, and high thresholds for hyp and chl
pred <- quickpred(nhanes, mincor = c(0, 0.1, 0.5, 0.5))
pred
# use it directly from mice
imp <- mice(nhanes, pred = quickpred(nhanes, minpuc = 0.25, include = "age"))
}
\references{
van Buuren, S., Boshuizen, H.C., Knook, D.L. (1999) Multiple
imputation of missing blood pressure covariates in survival analysis.
\emph{Statistics in Medicine}, \bold{18}, 681--694.
van Buuren, S. and Groothuis-Oudshoorn, K. (2011). \code{mice}: Multivariate
Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
}
\seealso{
\code{\link{mice}}, \code{\link[=mids-class]{mids}}
}
\author{
Stef van Buuren, Aug 2009
}
\keyword{misc}
mice/man/cci.Rd 0000644 0001762 0000144 00000001610 14330031606 012761 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/cci.R
\name{cci}
\alias{cci}
\title{Complete case indicator}
\usage{
cci(x)
}
\arguments{
\item{x}{An \code{R} object. Currently supported are methods for the
following classes: \code{mids}.}
}
\value{
Logical vector indicating the complete cases.
}
\description{
The complete case indicator is useful for extracting the subset of complete cases. The function
\code{cci(x)} calls \code{complete.cases(x)}.
The companion function \code{ici()} selects the incomplete cases.
}
\examples{
cci(nhanes) # indicator for 13 complete cases
cci(mice(nhanes, maxit = 0))
f <- cci(nhanes[, c("bmi", "hyp")]) # complete data for bmi and hyp
nhanes[f, ] # obtain all data from those with complete bmi and hyp
}
\seealso{
\code{\link{complete.cases}}, \code{\link{ici}}, \code{\link{cc}}
}
\author{
Stef van Buuren, 2017.
}
\keyword{univar}
mice/man/ampute.default.weights.Rd 0000644 0001762 0000144 00000002323 14330031606 016614 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/ampute.default.R
\name{ampute.default.weights}
\alias{ampute.default.weights}
\title{Default \code{weights} in \code{ampute}}
\usage{
ampute.default.weights(patterns, mech)
}
\arguments{
\item{patterns}{A matrix of size #patterns by #variables where \code{0} indicates
a variable should have missing values and \code{1} indicates a variable should
remain complete. Could be the result of \code{\link{ampute.default.patterns}}.}
\item{mech}{A string specifying the missingness mechanism.}
}
\value{
A matrix of size #patterns by #variables containing the weights that
will be used to calculate the weighted sum scores. Equal weights are given to
all variables. When mechanism is MAR, variables that will be amputed will be
weighted with \code{0}. If it is MNAR, variables that will be observed
will be weighted with \code{0}. If mechanism is MCAR, the weights matrix will
not be used. A default MAR matrix will be returned.
}
\description{
Defines the default weights matrix for the multivariate amputation function
\code{ampute}.
}
\seealso{
\code{\link{ampute}}, \code{\link{ampute.default.patterns}}
}
\author{
Rianne Schouten, 2016
}
\keyword{internal}
mice/man/leiden85.Rd 0000644 0001762 0000144 00000003232 14330031606 013642 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/leiden85.R
\docType{data}
\name{leiden85}
\alias{leiden85}
\title{Leiden 85+ study}
\format{
\code{leiden85} is a data frame with 956 rows and 336 columns.
}
\source{
Lagaay, A. M., van der Meij, J. C., Hijmans, W. (1992). Validation of
medical history taking as part of a population based survey in subjects aged
85 and over. \emph{Brit. Med. J.}, \emph{304}(6834), 1091-1092.
Izaks, G. J., van Houwelingen, H. C., Schreuder, G. M., Ligthart, G. J.
(1997). The association between human leucocyte antigens (HLA) and mortality
in community residents aged 85 and older. \emph{Journal of the American
Geriatrics Society}, \emph{45}(1), 56-60.
Boshuizen, H. C., Izaks, G. J., van Buuren, S., Ligthart, G. J. (1998).
Blood pressure and mortality in elderly people aged 85 and older: Community
based study. \emph{Brit. Med. J.}, \emph{316}(7147), 1780-1784.
Van Buuren, S., Boshuizen, H.C., Knook, D.L. (1999) Multiple imputation of
missing blood pressure covariates in survival analysis. \emph{Statistics in
Medicine}, \bold{18}, 681--694.
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-toomany.html#sec:leiden85cohort}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\description{
Subset of data from the Leiden 85+ study
}
\details{
The data set concerns of subset of 956 members of a very old (85+) cohort in
Leiden.
Multiple imputation of this data set has been described in Boshuizen et al
(1998), Van Buuren et al (1999) and Van Buuren (2012), chapter 7.
The data set is not available as part of \code{mice}.
}
\keyword{datasets}
mice/man/is.mitml.result.Rd 0000644 0001762 0000144 00000000575 13666252075 015325 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/is.R
\name{is.mitml.result}
\alias{is.mitml.result}
\title{Check for \code{mitml.result} object}
\usage{
is.mitml.result(x)
}
\arguments{
\item{x}{An object}
}
\value{
A logical indicating whether \code{x} is an object of class \code{mitml.result}
}
\description{
Check for \code{mitml.result} object
}
mice/man/ampute.default.type.Rd 0000644 0001762 0000144 00000001513 14330031606 016123 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/ampute.default.R
\name{ampute.default.type}
\alias{ampute.default.type}
\title{Default \code{type} in \code{ampute()}}
\usage{
ampute.default.type(patterns)
}
\arguments{
\item{patterns}{A matrix of size #patterns by #variables where 0 indicates a
variable should have missing values and 1 indicates a variable should remain
complete. Could be the result of \code{\link{ampute.default.patterns}}.}
}
\value{
A string vector of length #patterns containing the missingness types.
Each pattern will be amputed with a "RIGHT" missingness.
}
\description{
Defines the default type vector for the multivariate amputation function
\code{ampute}.
}
\seealso{
\code{\link{ampute}}, \code{\link{ampute.default.patterns}}
}
\author{
Rianne Schouten, 2016
}
\keyword{internal}
mice/man/matchindex.Rd 0000644 0001762 0000144 00000005746 14330031606 014365 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/RcppExports.R
\name{matchindex}
\alias{matchindex}
\title{Find index of matched donor units}
\usage{
matchindex(d, t, k = 5L)
}
\arguments{
\item{d}{Numeric vector with values from donor cases.}
\item{t}{Numeric vector with values from target cases.}
\item{k}{Integer, number of unique donors from which a random draw is made.
For \code{k = 1} the function returns the index in \code{d} corresponding
to the closest unit. For multiple imputation, the
advice is to set values in the range of \code{k = 5} to \code{k = 10}.}
}
\value{
An integer vector with \code{length(t)} elements. Each
element is an index in the array \code{d}.
}
\description{
Find index of matched donor units
}
\details{
For each element in \code{t}, the method finds the \code{k} nearest
neighbours in \code{d}, randomly draws one of these neighbours, and
returns its position in vector \code{d}.
Fast predictive mean matching algorithm in seven steps:
1. Shuffle records to remove effects of ties
2. Obtain sorting order on shuffled data
3. Calculate index on input data and sort it
4. Pre-sample vector \code{h} with values between 1 and \code{k}
For each of the \code{n0} elements in \code{t}:
5. find the two adjacent neighbours
6. find the \code{h_i}'th nearest neighbour
7. store the index of that neighbour
Return vector of \code{n0} positions in \code{d}.
We may use the function to perform predictive mean matching under a given
predictive model. To do so, specify both \code{d} and \code{t} as
predictions from the same model. Suppose that \code{y} contains the observed
outcomes of the donor cases (in the same sequence as \code{d}), then
\code{y[matchindex(d, t)]} returns one matched outcome for every
target case.
See \url{https://github.com/amices/mice/issues/236}.
This function is a replacement for the \code{matcher()} function that has
been in default in \code{mice} since version \code{2.22} (June 2014).
}
\examples{
set.seed(1)
# Inputs need not be sorted
d <- c(-5, 5, 0, 10, 12)
t <- c(-6, -4, 0, 2, 4, -2, 6)
# Index (in vector a) of closest match
idx <- matchindex(d, t, 1)
idx
# To check: show values of closest match
# Random draw among indices of the 5 closest predictors
matchindex(d, t)
# An example
train <- mtcars[1:20, ]
test <- mtcars[21:32, ]
fit <- lm(mpg ~ disp + cyl, data = train)
d <- fitted.values(fit)
t <- predict(fit, newdata = test) # note: not using mpg
idx <- matchindex(d, t)
# Borrow values from train to produce 12 synthetic values for mpg in test.
# Synthetic values are plausible values that could have been observed if
# they had been measured.
train$mpg[idx]
# Exercise: Create a distribution of 1000 plausible values for each of the
# twelve mpg entries in test, and count how many times the true value
# (which we know here) is located within the inter-quartile range of each
# distribution. Is your count anywhere close to 500? Why? Why not?
}
\author{
Stef van Buuren, Nasinski Maciej, Alexander Robitzsch
}
mice/man/tbc.Rd 0000644 0001762 0000144 00000004677 14330031606 013013 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/tbc.R
\docType{data}
\name{tbc}
\alias{tbc}
\alias{tbc.target}
\alias{terneuzen}
\title{Terneuzen birth cohort}
\format{
\code{tbs} is a data frame with 3951 rows and 11 columns:
\describe{
\item{id}{Person number}
\item{occ}{Occasion number}
\item{nocc}{Number of occasions}
\item{first}{Is this the first record for this person? (TRUE/FALSE)}
\item{typ}{Type of data (all observed)}
\item{age}{Age (years)}
\item{sex}{Sex 1=M, 2=F}
\item{hgt.z}{Height Z-score}
\item{wgt.z}{Weight Z-score}
\item{bmi.z}{BMI Z-score}
\item{ao}{Adult overweight (0=no, 1=yes)}
}
\code{tbc.target} is a data frame with 2612 rows and 3 columns:
\describe{
\item{id}{Person number}
\item{ao}{Adult overweight (0=no, 1=yes)}
\item{bmi.z.jv}{BMI Z-score as young adult (18-29 years)}
}
}
\source{
De Kroon, M. L. A., Renders, C. M., Kuipers, E. C., van Wouwe, J. P.,
van Buuren, S., de Jonge, G. A., Hirasing, R. A. (2008). Identifying
metabolic syndrome without blood tests in young adults - The Terneuzen birth
cohort. \emph{European Journal of Public Health}, \emph{18}(6), 656-660.
De Kroon, M. L. A., Renders, C. M., Van Wouwe, J. P., Van Buuren, S.,
Hirasing, R. A. (2010). The Terneuzen birth cohort: BMI changes between 2
and 6 years correlate strongest with adult overweight. \emph{PLoS ONE},
\emph{5}(2), e9155.
De Kroon, M. L. A. (2011). \emph{The Terneuzen Birth Cohort. Detection and
Prevention of Overweight and Cardiometabolic Risk from Infancy Onward.}
Dissertation, Vrije Universiteit, Amsterdam.
\url{https://research.vu.nl/en/publications/the-terneuzen-birth-cohort-detection-and-prevention-of-overweight}
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-rastering.html#terneuzen-birth-cohort}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\description{
Data of subset of the Terneuzen Birth Cohort data on child growth.
}
\details{
This \code{tbc} data set is a random subset of persons from a much larger
collection of data from the Terneuzen Birth Cohort. The total cohort
comprises of 2604 unique persons, whereas the subset in \code{tbc} covers 306
persons. The \code{tbc.target} is an auxiliary data set containing two
outcomes at adult age. For more details, see De Kroon et al (2008, 2010,
2011). The imputation methodology is explained in Chapter 9 of Van Buuren
(2012).
}
\examples{
data <- tbc
md.pattern(data)
}
\keyword{datasets}
mice/man/toenail.Rd 0000644 0001762 0000144 00000004126 14330031606 013663 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/toenail.R
\docType{data}
\name{toenail}
\alias{toenail}
\title{Toenail data}
\format{
A data frame with 1908 observations on the following 5 variables:
\describe{
\item{\code{ID}}{a numeric vector giving the ID of patient}
\item{\code{outcome}}{a numeric vector giving the response
(0=none or mild seperation, 1=moderate or severe)}
\item{\code{treatment}}{a numeric vector giving the treatment group}
\item{\code{month}}{a numeric vector giving the time of the visit
(not exactly monthly intervals hence not round numbers)}
\item{\code{visit}}{a numeric vector giving the number of the visit}
}
}
\source{
De Backer, M., De Vroey, C., Lesaffre, E., Scheys, I., and De
Keyser, P. (1998). Twelve weeks of continuous oral therapy for
toenail onychomycosis caused by dermatophytes: A double-blind
comparative trial of terbinafine 250 mg/day versus itraconazole 200
mg/day. Journal of the American Academy of Dermatology, 38, 57-63.
}
\description{
The toenail data come from a Multicenter study comparing two oral
treatments for toenail infection. Patients were evaluated for the
degree of separation of the nail. Patients were randomized into two
treatments and were followed over seven visits - four in the first
year and yearly thereafter. The patients have not been treated
prior to the first visit so this should be regarded as the
baseline.
}
\details{
This dataset was copied from the \code{DPpackage}, which is
scheduled to be discontinued from CRAN in August 2019.
}
\references{
Lesaffre, E. and Spiessens, B. (2001). On the effect of the number of
quadrature points in a logistic random-effects model: An example.
Journal of the Royal Statistical Society, Series C, 50, 325-335.
G. Fitzmaurice, N. Laird and J. Ware (2004) Applied Longitudinal Analysis,
Wiley and Sons, New York, USA.
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-catoutcome.html#example}{\emph{Flexible
Imputation of Missing Data. Second Edition.}} Chapman & Hall/CRC.
Boca Raton, FL.
}
\seealso{
\code{\link{toenail2}}
}
\keyword{datasets}
mice/man/tidy.mipo.Rd 0000644 0001762 0000144 00000001622 13666252233 014156 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/tidiers.R
\name{tidy.mipo}
\alias{tidy.mipo}
\title{Tidy method to extract results from a `mipo` object}
\usage{
\method{tidy}{mipo}(x, conf.int = FALSE, conf.level = 0.95, ...)
}
\arguments{
\item{x}{An object of class \code{mipo}}
\item{conf.int}{Logical. Should confidence intervals be returned?}
\item{conf.level}{Confidence level for intervals. Defaults to .95}
\item{...}{extra arguments (not used)}
}
\value{
A dataframe withh these columns:
\itemize{
\item term
\item estimate
\item ubar
\item b
\item t
\item dfcom
\item df
\item riv
\item lambda
\item fmi
\item p.value
\item conf.low (if called with conf.int = TRUE)
\item conf.high (if called with conf.int = TRUE)
}
}
\description{
Tidy method to extract results from a `mipo` object
}
\keyword{internal}
mice/man/ici.Rd 0000644 0001762 0000144 00000001334 14330031606 012772 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/cci.R
\name{ici}
\alias{ici}
\alias{ici,data.frame-method}
\alias{ici,matrix-method}
\alias{ici,mids-method}
\title{Incomplete case indicator}
\usage{
ici(x)
}
\arguments{
\item{x}{An \code{R} object. Currently supported are methods for the
following classes: \code{mids}.}
}
\value{
Logical vector indicating the incomplete cases,
}
\description{
This array is useful for extracting the subset of incomplete cases.
The companion function \code{cci()} selects the complete cases.
}
\examples{
ici(nhanes) # indicator for 12 rows with incomplete cases
}
\seealso{
\code{\link{cci}}, \code{\link{ic}}
}
\author{
Stef van Buuren, 2017.
}
\keyword{univar}
mice/man/filter.mids.Rd 0000644 0001762 0000144 00000006050 14330031606 014446 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/filter.R
\name{filter.mids}
\alias{filter.mids}
\title{Subset rows of a \code{mids} object}
\usage{
\method{filter}{mids}(.data, ..., .preserve = FALSE)
}
\arguments{
\item{.data}{A \code{mids} object.}
\item{...}{Expressions that return a
logical value, and are defined in terms of the variables in \code{.data$data}.
If multiple expressions are specified, they are combined with the \code{&} operator.
Only rows for which all conditions evaluate to \code{TRUE} are kept.}
\item{.preserve}{Relevant when the \code{.data} input is grouped.
If \code{.preserve = FALSE} (the default), the grouping structure
is recalculated based on the resulting data, otherwise the grouping is kept as is.}
}
\value{
An S3 object of class \code{mids}
}
\description{
This function takes a \code{mids} object and returns a new
\code{mids} object that pertains to the subset of the data
identified by the expression in \dots. The expression may use
column values from the incomplete data in \code{.data$data}.
}
\note{
The function calculates a logical vector \code{include} of length \code{nrow(.data$data)}.
The function constructs the elements of the filtered \code{mids} object as follows:
\tabular{ll}{
\code{data} \tab Select rows in \code{.data$data} for which \code{include == TRUE}\cr
\code{imp} \tab Select rows each imputation \code{data.frame} in \code{.data$imp} for which \code{include == TRUE}\cr
\code{m} \tab Equals \code{.data$m}\cr
\code{where} \tab Select rows in \code{.data$where} for which \code{include == TRUE}\cr
\code{blocks} \tab Equals \code{.data$blocks}\cr
\code{call} \tab Equals \code{.data$call}\cr
\code{nmis} \tab Recalculate \code{nmis} based on the selected \code{data} rows\cr
\code{method} \tab Equals \code{.data$method}\cr
\code{predictorMatrix} \tab Equals \code{.data$predictorMatrix}\cr
\code{visitSequence} \tab Equals \code{.data$visitSequence}\cr
\code{formulas} \tab Equals \code{.data$formulas}\cr
\code{post} \tab Equals \code{.data$post}\cr
\code{blots} \tab Equals \code{.data$blots}\cr
\code{ignore} \tab Select positions in \code{.data$ignore} for which \code{include == TRUE}\cr
\code{seed} \tab Equals \code{.data$seed}\cr
\code{iteration} \tab Equals \code{.data$iteration}\cr
\code{lastSeedValue} \tab Equals \code{.data$lastSeedValue}\cr
\code{chainMean} \tab Set to \code{NULL}\cr
\code{chainVar} \tab Set to \code{NULL}\cr
\code{loggedEvents} \tab Equals \code{.data$loggedEvents}\cr
\code{version} \tab Replaced with current version\cr
\code{date} \tab Replaced with current date
}
}
\examples{
imp <- mice(nhanes, m = 2, maxit = 1, print = FALSE)
# example with external logical vector
imp_f <- filter(imp, c(rep(TRUE, 13), rep(FALSE, 12)))
nrow(complete(imp))
nrow(complete(imp_f))
# example with calculated include vector
imp_f2 <- filter(imp, age >= 2 & hyp == 1)
nrow(complete(imp_f2)) # should be 5
}
\seealso{
\code{\link[dplyr]{filter}}
}
\author{
Patrick Rockenschaub
}
\keyword{manip}
mice/man/mice.impute.rf.Rd 0000644 0001762 0000144 00000010101 14436640740 015057 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.rf.R
\name{mice.impute.rf}
\alias{mice.impute.rf}
\title{Imputation by random forests}
\usage{
mice.impute.rf(
y,
ry,
x,
wy = NULL,
ntree = 10,
rfPackage = c("ranger", "randomForest"),
...
)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{ntree}{The number of trees to grow. The default is 10.}
\item{rfPackage}{A single string specifying the backend for estimating the
random forest. The default backend is the \code{ranger} package. The only
alternative currently implemented is the \code{randomForest} package, which
used to be the default in mice 3.13.10 and earlier.}
\item{\dots}{Other named arguments passed down to
\code{mice:::install.on.demand()}, \code{randomForest::randomForest()} and
\code{randomForest:::randomForest.default()}.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes univariate missing data using random forests.
}
\details{
Imputation of \code{y} by random forests. The method
calls \code{randomForrest()} which implements Breiman's random forest
algorithm (based on Breiman and Cutler's original Fortran code)
for classification and regression. See Appendix A.1 of Doove et al.
(2014) for the definition of the algorithm used.
}
\note{
An alternative implementation was independently
developed by Shah et al (2014). This were available as
functions \code{CALIBERrfimpute::mice.impute.rfcat} and
\code{CALIBERrfimpute::mice.impute.rfcont} (now archived).
Simulations by Shah (Feb 13, 2014) suggested that
the quality of the imputation for 10 and 100 trees was identical,
so mice 2.22 changed the default number of trees from \code{ntree = 100} to
\code{ntree = 10}.
}
\examples{
\dontrun{
imp <- mice(nhanes2, meth = "rf", ntree = 3)
plot(imp)
}
}
\references{
Doove, L.L., van Buuren, S., Dusseldorp, E. (2014), Recursive partitioning
for missing data imputation in the presence of interaction Effects.
Computational Statistics & Data Analysis, 72, 92-104.
Shah, A.D., Bartlett, J.W., Carpenter, J., Nicholas, O., Hemingway, H. (2014),
Comparison of random forest and parametric imputation models for
imputing missing data using MICE: A CALIBER study. American Journal
of Epidemiology, \doi{10.1093/aje/kwt312}.
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-cart.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\seealso{
\code{\link{mice}}, \code{\link{mice.impute.cart}},
\code{\link[randomForest]{randomForest}}
\code{\link[ranger]{ranger}}
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.ri}()}
}
\author{
Lisa Doove, Stef van Buuren, Elise Dusseldorp, 2012; Patrick Rockenschaub, 2021
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/mice.impute.logreg.boot.Rd 0000644 0001762 0000144 00000005164 14330031647 016677 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.logreg.R
\name{mice.impute.logreg.boot}
\alias{mice.impute.logreg.boot}
\title{Imputation by logistic regression using the bootstrap}
\usage{
mice.impute.logreg.boot(y, ry, x, wy = NULL, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes univariate missing data using logistic regression
by a bootstrapped logistic regression model.
The bootstrap method draws a simple bootstrap sample with replacement
from the observed data \code{y[ry]} and \code{x[ry, ]}.
}
\references{
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-categorical.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\seealso{
\code{\link{mice}}, \code{\link{glm}}, \code{\link{glm.fit}}
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Stef van Buuren, Karin Groothuis-Oudshoorn, 2000, 2011
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/D3.Rd 0000644 0001762 0000144 00000005137 14436133175 012514 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/D3.R
\name{D3}
\alias{D3}
\title{Compare two nested models using D3-statistic}
\usage{
D3(fit1, fit0 = NULL, dfcom = NULL, df.com = NULL)
}
\arguments{
\item{fit1}{An object of class \code{mira}, produced by \code{with()}.}
\item{fit0}{An object of class \code{mira}, produced by \code{with()}. The
model in \code{fit0} is a nested within \code{fit1}. The default null
model \code{fit0 = NULL} compares \code{fit1} to the intercept-only model.}
\item{dfcom}{A single number denoting the
complete-data degrees of freedom of model \code{fit1}. If not specified,
it is set equal to \code{df.residual} of model \code{fit1}. If that cannot
be done, the procedure assumes (perhaps incorrectly) a large sample.}
\item{df.com}{Deprecated}
}
\value{
An object of class \code{mice.anova}
}
\description{
The D3-statistic is a likelihood-ratio test statistic.
}
\details{
The \code{D3()} function implement the LR-method by
Meng and Rubin (1992). The implementation of the method relies
on the \code{broom} package, the standard \code{update} mechanism
for statistical models in \code{R} and the \code{offset} function.
The function calculates \code{m} repetitions of the full
(or null) models, calculates the mean of the estimates of the
(fixed) parameter coefficients \eqn{\beta}. For each imputed
imputed dataset, it calculates the likelihood for the model with
the parameters constrained to \eqn{\beta}.
The \code{mitml::testModels()} function offers similar functionality
for a subset of statistical models. Results of \code{mice::D3()} and
\code{mitml::testModels()} differ in multilevel models because the
\code{testModels()} also constrains the variance components parameters.
For more details on
}
\examples{
# Compare two linear models:
imp <- mice(nhanes2, seed = 51009, print = FALSE)
mi1 <- with(data = imp, expr = lm(bmi ~ age + hyp + chl))
mi0 <- with(data = imp, expr = lm(bmi ~ age + hyp))
D3(mi1, mi0)
\dontrun{
# Compare two logistic regression models
imp <- mice(boys, maxit = 2, print = FALSE)
fit1 <- with(imp, glm(gen > levels(gen)[1] ~ hgt + hc + reg, family = binomial))
fit0 <- with(imp, glm(gen > levels(gen)[1] ~ hgt + hc, family = binomial))
D3(fit1, fit0)
}
}
\references{
Meng, X. L., and D. B. Rubin. 1992.
Performing Likelihood Ratio Tests with Multiply-Imputed Data Sets.
\emph{Biometrika}, 79 (1): 103–11.
\url{https://stefvanbuuren.name/fimd/sec-multiparameter.html#sec:likelihoodratio}
\url{http://bbolker.github.io/mixedmodels-misc/glmmFAQ.html#setting-residual-variances-to-a-fixed-value-zero-or-other}
}
\seealso{
\code{\link{fix.coef}}
}
mice/man/ibind.Rd 0000644 0001762 0000144 00000002143 14433400023 013307 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/ibind.R
\name{ibind}
\alias{ibind}
\title{Enlarge number of imputations by combining \code{mids} objects}
\usage{
ibind(x, y)
}
\arguments{
\item{x}{A \code{mids} object.}
\item{y}{A \code{mids} object.}
}
\value{
An S3 object of class \code{mids}
}
\description{
This function combines two \code{mids} objects \code{x} and \code{y} into a
single \code{mids} object, with the objective of increasing the number of
imputed data sets. If the number of imputations in \code{x} and \code{y} are
\code{m(x)} and \code{m(y)}, then the combined object will have
\code{m(x)+m(y)} imputations.
}
\details{
The two \code{mids} objects are required to
have the same underlying multiple imputation model and should
be fitted on the same data.
}
\examples{
data(nhanes)
imp1 <- mice(nhanes, m = 1, maxit = 2, print = FALSE)
imp1$m
imp2 <- mice(nhanes, m = 3, maxit = 3, print = FALSE)
imp2$m
imp12 <- ibind(imp1, imp2)
imp12$m
plot(imp12)
}
\seealso{
\code{\link[=mids-class]{mids}}
}
\author{
Karin Groothuis-Oudshoorn, Stef van Buuren
}
\keyword{manip}
mice/man/mice.impute.mnar.Rd 0000644 0001762 0000144 00000020232 14330031647 015404 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.mnar.logreg.R,
% R/mice.impute.mnar.norm.R
\name{mice.impute.mnar.logreg}
\alias{mice.impute.mnar.logreg}
\alias{mice.impute.mnar.norm}
\alias{mnar.norm}
\alias{mnar.logreg}
\title{Imputation under MNAR mechanism by NARFCS}
\usage{
mice.impute.mnar.logreg(y, ry, x, wy = NULL, ums = NULL, umx = NULL, ...)
mice.impute.mnar.norm(y, ry, x, wy = NULL, ums = NULL, umx = NULL, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{ums}{A string containing the specification of the
unidentifiable part of the imputation model (the *unidentifiable
model specification"), that is, the desired \eqn{\delta}-adjustment
(offset) as a function of other variables and values for the
corresponding deltas (sensitivity parameters). See details.}
\item{umx}{An auxiliary data matrix containing variables that do
not appear in the identifiable part of the imputation procedure
but that have been specified via \code{ums} as being predictors
in the unidentifiable part of the imputation model. See details.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes univariate data under a user-specified MNAR mechanism by
linear or logistic regression and NARFCS. Sensitivity analysis under
different model specifications may shed light on the impact of
different MNAR assumptions on the conclusions.
}
\details{
This function imputes data that are thought to be Missing Not at
Random (MNAR) by the NARFCS method. The NARFCS procedure
(Tompsett et al, 2018) generalises the so-called
\eqn{\delta}-adjustment sensitivity analysis method of Van Buuren,
Boshuizen & Knook (1999) to the case with multiple incomplete
variables within the FCS framework. In practical terms, the
NARFCS procedure shifts the imputations drawn at each
iteration of \code{mice} by a user-specified quantity that can
vary across subjects, to reflect systematic departures of the
missing data from the data distribution imputed under MAR.
Specification of the NARFCS model is done by the \code{blots}
argument of \code{mice()}. The \code{blots} parameter is a named
list. For each variable to be imputed by
\code{mice.impute.mnar.norm()} or \code{mice.impute.mnar.logreg()}
the corresponding element in \code{blots} is a list with
at least one argument \code{ums} and, optionally, a second
argument \code{umx}.
For example, the high-level call might like something like
\code{mice(nhanes[, c(2, 4)], method = c("pmm", "mnar.norm"),
blots = list(chl = list(ums = "-3+2*bmi")))}.
The \code{ums} parameter is required, and might look like this:
\code{"-4+1*Y"}. The \code{ums} specifcation must have the
following characteristics:
\enumerate{
\item{A single term corresponding to the intercept (constant) term,
not multiplied by any variable name, must be included in the
expression;}
\item{Each term in the expression (corresponding to the intercept
or a predictor variable) must be separated by either a \code{"+"}
or \code{"-"} sign, depending on the sign of the sensitivity
parameter;}
\item{Within each non-intercept term, the sensitivity parameter
value comes first and the predictor variable comes second, and these
must be separated by a \code{"*"} sign;}
\item{For categorical predictors, for example a variable \code{Z}
with K + 1 categories \code{("Cat0","Cat1", ...,"CatK")}, K
category-specific terms are needed, and those not in \code{umx}
(see below) must be specified by concatenating the variable name
with the name of the category (e.g. \code{ZCat1}) as this is how
they are named in the design matrix (argument \code{x}) passed
to the univariate imputation function. An example is
\code{"2+1*ZCat1-3*ZCat2"}.}
}
If given, the \code{umx} specification must have the following
characteristics:
\enumerate{
\item{It contains only complete variables, with no missing values;}
\item{It is a numeric matrix. In particular, categorical variables
must be represented as dummy indicators with names corresponding
to what is used in \code{ums} to refer to the category-specific terms
(see above);}
\item{It has the same number of rows as the \code{data} argument
passed on to the main \code{mice} function;}
\item{It does not contain variables that were already predictors
in the identifiable part of the model for the variable under
imputation.}
}
Limitation: The present implementation can only condition on variables
that appear in the identifiable part of the imputation model (\code{x}) or
in complete auxiliary variables passed on via the \code{umx} argument.
It is not possible to specify models where the offset depends on
incomplete auxiliary variables.
For an MNAR alternative see also \code{\link{mice.impute.ri}}.
}
\examples{
# 1: Example with no auxiliary data: only pass unidentifiable model specification (ums)
# Specify argument to pass on to mnar imputation functions via "blots" argument
mnar.blot <- list(X = list(ums = "-4"), Y = list(ums = "2+1*ZCat1-3*ZCat2"))
# Run NARFCS by using mnar imputation methods and passing argument via blots
impNARFCS <- mice(mnar_demo_data,
method = c("mnar.logreg", "mnar.norm", ""),
blots = mnar.blot, seed = 234235, print = FALSE
)
# Obtain MI results: Note they coincide with those from old version at
# https://github.com/moreno-betancur/NARFCS
pool(with(impNARFCS, lm(Y ~ X + Z)))$pooled$estimate
# 2: Example passing also auxiliary data to MNAR procedure (umx)
# Assumptions:
# - Auxiliary data are complete, no missing values
# - Auxiliary data are a numeric matrix
# - Auxiliary data have same number of rows as x
# - Auxiliary data have no overlapping variable names with x
# Specify argument to pass on to mnar imputation functions via "blots" argument
aux <- matrix(0:1, nrow = nrow(mnar_demo_data))
dimnames(aux) <- list(NULL, "even")
mnar.blot <- list(
X = list(ums = "-4"),
Y = list(ums = "2+1*ZCat1-3*ZCat2+0.5*even", umx = aux)
)
# Run NARFCS by using mnar imputation methods and passing argument via blots
impNARFCS <- mice(mnar_demo_data,
method = c("mnar.logreg", "mnar.norm", ""),
blots = mnar.blot, seed = 234235, print = FALSE
)
# Obtain MI results: As expected they differ (slightly) from those
# from old version at https://github.com/moreno-betancur/NARFCS
pool(with(impNARFCS, lm(Y ~ X + Z)))$pooled$estimate
}
\references{
Tompsett, D. M., Leacy, F., Moreno-Betancur, M., Heron, J., &
White, I. R. (2018). On the use of the not-at-random fully
conditional specification (NARFCS) procedure in practice.
\emph{Statistics in Medicine}, \bold{37}(15), 2338-2353.
\doi{10.1002/sim.7643}.
Van Buuren, S., Boshuizen, H.C., Knook, D.L. (1999) Multiple
imputation of missing blood pressure covariates in survival analysis.
\emph{Statistics in Medicine}, \bold{18}, 681--694.
}
\seealso{
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Margarita Moreno-Betancur, Stef van Buuren, Ian R. White, 2020.
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/ampute.default.freq.Rd 0000644 0001762 0000144 00000001574 14330031606 016106 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/ampute.default.R
\name{ampute.default.freq}
\alias{ampute.default.freq}
\title{Default \code{freq} in \code{ampute}}
\usage{
ampute.default.freq(patterns)
}
\arguments{
\item{patterns}{A matrix of size #patterns by #variables where \code{0} indicates
a variable should have missing values and \code{1} indicates a variable should
remain complete. Could be the result of \code{\link{ampute.default.patterns}}.}
}
\value{
A vector of length #patterns containing the relative frequencies with
which the patterns should occur. An equal probability is given to each pattern.
}
\description{
Defines the default relative frequency vector for the multivariate
amputation function \code{ampute}.
}
\seealso{
\code{\link{ampute}}, \code{\link{ampute.default.patterns}}
}
\author{
Rianne Schouten, 2016
}
\keyword{internal}
mice/man/nhanes2.Rd 0000644 0001762 0000144 00000001731 14330031606 013565 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/nhanes2.R
\docType{data}
\name{nhanes2}
\alias{nhanes2}
\title{NHANES example - mixed numerical and discrete variables}
\format{
A data frame with 25 observations on the following 4 variables.
\describe{
\item{age}{Age group (1=20-39, 2=40-59, 3=60+)}
\item{bmi}{Body mass index (kg/m**2)}
\item{hyp}{Hypertensive (1=no,2=yes)}
\item{chl}{Total serum cholesterol (mg/dL)} }
}
\source{
Schafer, J.L. (1997). \emph{Analysis of Incomplete Multivariate
Data.} London: Chapman & Hall. Table 6.14.
}
\description{
A small data set with non-monotone missing values.
}
\details{
A small data set with missing data and mixed numerical and discrete
variables. The data set \code{nhanes} is the same data set, but with all data
treated as numerical.
}
\examples{
# create 5 imputed data sets
imp <- mice(nhanes2)
# print the first imputed data set
complete(imp)
}
\seealso{
\code{\link{nhanes}}
}
\keyword{datasets}
mice/man/supports.transparent.Rd 0000644 0001762 0000144 00000001344 14330031606 016466 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/supports.transparent.R
\name{supports.transparent}
\alias{supports.transparent}
\alias{transparent}
\title{Supports semi-transparent foreground colors?}
\usage{
supports.transparent()
}
\value{
\code{TRUE} or \code{FALSE}
}
\description{
This function is used by \code{mdc()} to find out whether the current device
supports semi-transparent foreground colors.
}
\details{
The function calls the function \code{dev.capabilities()} from the package
\code{grDevices}. The function return \code{FALSE} if the status of the
current device is unknown.
}
\examples{
supports.transparent()
}
\seealso{
\code{\link{mdc}} \code{\link{dev.capabilities}}
}
\keyword{hplot}
mice/man/summary.Rd 0000644 0001762 0000144 00000002416 14330031606 013725 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/summary.R
\name{summary.mira}
\alias{summary.mira}
\alias{summary.mids}
\alias{summary.mads}
\alias{summary.mice.anova}
\title{Summary of a \code{mira} object}
\usage{
\method{summary}{mira}(object, type = c("tidy", "glance", "summary"), ...)
\method{summary}{mids}(object, ...)
\method{summary}{mads}(object, ...)
\method{summary}{mice.anova}(object, ...)
}
\arguments{
\item{object}{A \code{mira} object}
\item{type}{A length-1 character vector indicating the
type of summary. There are three choices: \code{type = "tidy"}
return the parameters estimates of each analyses as a data frame.
\code{type = "glance"} return the fit statistics of each analysis
as a data frame. \code{type = "summary"} returns a list of
length \code{m} with the analysis results. The default is
\code{"tidy"}.}
\item{...}{Other parameters passed down to \code{print()} and \code{summary()}}
}
\value{
\code{NULL}
\code{NULL}
\code{NULL}
\code{NULL}
}
\description{
Summary of a \code{mira} object
Summary of a \code{mids} object
Summary of a \code{mads} object
Print a \code{mice.anova} object
}
\seealso{
\code{\link[=mira-class]{mira}}
\code{\link[=mids-class]{mids}}
\code{\link[=mads-class]{mads}}
\code{\link{mipo}}
}
mice/man/pattern.Rd 0000644 0001762 0000144 00000003271 14436640317 013721 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/pattern1.R
\docType{data}
\name{pattern}
\alias{pattern}
\alias{pattern1}
\alias{pattern2}
\alias{pattern3}
\alias{pattern4}
\title{Datasets with various missing data patterns}
\format{
\describe{ \item{list("pattern1")}{Data with a univariate missing
data pattern} \item{list("pattern2")}{Data with a monotone missing data
pattern} \item{list("pattern3")}{Data with a file matching missing data
pattern} \item{list("pattern4")}{Data with a general missing data pattern} }
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/missing-data-pattern.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\description{
Four simple datasets with various missing data patterns
}
\details{
Van Buuren (2012) uses these four artificial datasets to illustrate various
missing data patterns.
}
\examples{
pattern4
data <- rbind(pattern1, pattern2, pattern3, pattern4)
mdpat <- cbind(expand.grid(rec = 8:1, pat = 1:4, var = 1:3), r = as.numeric(as.vector(is.na(data))))
types <- c("Univariate", "Monotone", "File matching", "General")
tp41 <- lattice::levelplot(r ~ var + rec | as.factor(pat),
data = mdpat,
as.table = TRUE, aspect = "iso",
shrink = c(0.9),
col.regions = mdc(1:2),
colorkey = FALSE,
scales = list(draw = FALSE),
xlab = "", ylab = "",
between = list(x = 1, y = 0),
strip = lattice::strip.custom(
bg = "grey95", style = 1,
factor.levels = types
)
)
print(tp41)
md.pattern(pattern4)
p <- md.pairs(pattern4)
p
### proportion of usable cases
p$mr / (p$mr + p$mm)
### outbound statistics
p$rm / (p$rm + p$rr)
fluxplot(pattern2)
}
\keyword{datasets}
mice/man/cbind.Rd 0000644 0001762 0000144 00000017431 14433400023 013307 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/generics.R
\name{cbind}
\alias{cbind}
\alias{rbind}
\title{Combine R objects by rows and columns}
\usage{
cbind(...)
rbind(...)
}
\arguments{
\item{...}{
Arguments passed on to \code{\link[base:cbind]{base::cbind}}
\describe{
\item{\code{deparse.level}}{integer controlling the construction of labels in
the case of non-matrix-like arguments (for the default method):\cr
\code{deparse.level = 0} constructs no labels;\cr
the default \code{deparse.level = 1} typically and
\code{deparse.level = 2} always construct labels from the argument
names, see the \sQuote{Value} section below.}
}}
}
\value{
An S3 object of class \code{mids}
}
\description{
Functions \code{cbind()} and \code{rbind()} are defined in
the \code{mice} package in order to
enable dispatch to \code{cbind.mids()} and \code{rbind.mids()}
when one of the arguments is a \code{data.frame}.
}
\details{
The standard \code{base::cbind()} and \code{base::rbind()}
always dispatch to
\code{base::cbind.data.frame()} or \code{base::rbind.data.frame()}
if one of the arguments is a
\code{data.frame}. The versions defined in the \code{mice}
package intercept the user command
and test whether the first argument has class \code{"mids"}. If so,
function calls \code{cbind.mids()}, respectively \code{rbind.mids()}. In
all other cases, the call is forwarded to standard functions in the
\code{base} package.
The \code{cbind.mids()} function combines two \code{mids} objects
columnwise into a single
object of class \code{mids}, or combines a single \code{mids} object with
a \code{vector}, \code{matrix}, \code{factor} or \code{data.frame}
columnwise into a \code{mids} object.
If both arguments of \code{cbind.mids()} are \code{mids}-objects, the
\code{data} list components should have the same number of rows. Also, the
number of imputations (\code{m}) should be identical.
If the second argument is a \code{matrix},
\code{factor} or \code{vector}, it is transformed into a
\code{data.frame}. The number of rows should match with the \code{data}
component of the first argument.
The \code{cbind.mids()} function renames any duplicated variable or block names by
appending \code{".1"}, \code{".2"} to duplicated names.
The \code{rbind.mids()} function combines two \code{mids} objects rowwise into a single
\code{mids} object, or combines a \code{mids} object with a vector, matrix,
factor or data frame rowwise into a \code{mids} object.
If both arguments of \code{rbind.mids()} are \code{mids} objects,
then \code{rbind.mids()} requires that both have the same number of multiple
imputations. In addition, their \code{data} components should match.
If the second argument of \code{rbind.mids()} is not a \code{mids} object,
the columns of the arguments should match. The \code{where} matrix for the
second argument is set to \code{FALSE}, signalling that any missing values in
that argument were not imputed. The \code{ignore} vector for the second argument is
set to \code{FALSE}. Rows inherited from the second argument will therefore
influence the parameter estimation of the imputation model in any future
iterations.
}
\note{
The \code{cbind.mids()} function constructs the elements of the new \code{mids} object as follows:
\tabular{ll}{
\code{data} \tab Columnwise combination of the data in \code{x} and \code{y}\cr
\code{imp} \tab Combines the imputed values from \code{x} and \code{y}\cr
\code{m} \tab Taken from \code{x$m}\cr
\code{where} \tab Columnwise combination of \code{x$where} and \code{y$where}\cr
\code{blocks} \tab Combines \code{x$blocks} and \code{y$blocks}\cr
\code{call} \tab Vector, \code{call[1]} creates \code{x}, \code{call[2]}
is call to \code{cbind.mids()}\cr
\code{nmis} \tab Equals \code{c(x$nmis, y$nmis)}\cr
\code{method} \tab Combines \code{x$method} and \code{y$method}\cr
\code{predictorMatrix} \tab Combination with zeroes on the off-diagonal blocks\cr
\code{visitSequence} \tab Combined as \code{c(x$visitSequence, y$visitSequence)}\cr
\code{formulas} \tab Combined as \code{c(x$formulas, y$formulas)}\cr
\code{post} \tab Combined as \code{c(x$post, y$post)}\cr
\code{blots} \tab Combined as \code{c(x$blots, y$blots)}\cr
\code{ignore} \tab Taken from \code{x$ignore}\cr
\code{seed} \tab Taken from \code{x$seed}\cr
\code{iteration} \tab Taken from \code{x$iteration}\cr
\code{lastSeedValue} \tab Taken from \code{x$lastSeedValue}\cr
\code{chainMean} \tab Combined from \code{x$chainMean} and \code{y$chainMean}\cr
\code{chainVar} \tab Combined from \code{x$chainVar} and \code{y$chainVar}\cr
\code{loggedEvents} \tab Taken from \code{x$loggedEvents}\cr
\code{version} \tab Current package version\cr
\code{date} \tab Current date\cr
}
The \code{rbind.mids()} function constructs the elements of the new \code{mids} object as follows:
\tabular{ll}{
\code{data} \tab Rowwise combination of the (incomplete) data in \code{x} and \code{y}\cr
\code{imp} \tab Equals \code{rbind(x$imp[[j]], y$imp[[j]])} if \code{y} is \code{mids} object; otherwise
the data of \code{y} will be copied\cr
\code{m} \tab Equals \code{x$m}\cr
\code{where} \tab Rowwise combination of \code{where} arguments\cr
\code{blocks} \tab Equals \code{x$blocks}\cr
\code{call} \tab Vector, \code{call[1]} creates \code{x}, \code{call[2]} is call to \code{rbind.mids}\cr
\code{nmis} \tab \code{x$nmis} + \code{y$nmis}\cr
\code{method} \tab Taken from \code{x$method}\cr
\code{predictorMatrix} \tab Taken from \code{x$predictorMatrix}\cr
\code{visitSequence} \tab Taken from \code{x$visitSequence}\cr
\code{formulas} \tab Taken from \code{x$formulas}\cr
\code{post} \tab Taken from \code{x$post}\cr
\code{blots} \tab Taken from \code{x$blots}\cr
\code{ignore} \tab Concatenate \code{x$ignore} and \code{y$ignore}\cr
\code{seed} \tab Taken from \code{x$seed}\cr
\code{iteration} \tab Taken from \code{x$iteration}\cr
\code{lastSeedValue} \tab Taken from \code{x$lastSeedValue}\cr
\code{chainMean} \tab Set to \code{NA}\cr
\code{chainVar} \tab Set to \code{NA}\cr
\code{loggedEvents} \tab Taken from \code{x$loggedEvents}\cr
\code{version} \tab Taken from \code{x$version}\cr
\code{date} \tab Taken from \code{x$date}
}
}
\examples{
# --- cbind ---
# impute four variables at once (default)
imp <- mice(nhanes, m = 1, maxit = 1, print = FALSE)
imp$predictorMatrix
# impute two by two
data1 <- nhanes[, c("age", "bmi")]
data2 <- nhanes[, c("hyp", "chl")]
imp1 <- mice(data1, m = 2, maxit = 1, print = FALSE)
imp2 <- mice(data2, m = 2, maxit = 1, print = FALSE)
# Append two solutions
imp12 <- cbind(imp1, imp2)
# This is a different imputation model
imp12$predictorMatrix
# Append the other way around
imp21 <- cbind(imp2, imp1)
imp21$predictorMatrix
# Append 'forgotten' variable chl
data3 <- nhanes[, 1:3]
imp3 <- mice(data3, maxit = 1, m = 2, print = FALSE)
imp4 <- cbind(imp3, chl = nhanes$chl)
# Of course, chl was not imputed
head(complete(imp4))
# Combine mids object with data frame
imp5 <- cbind(imp3, nhanes2)
head(complete(imp5))
# --- rbind ---
imp1 <- mice(nhanes[1:13, ], m = 2, maxit = 1, print = FALSE)
imp5 <- mice(nhanes[1:13, ], m = 2, maxit = 2, print = FALSE)
mylist <- list(age = NA, bmi = NA, hyp = NA, chl = NA)
nrow(complete(rbind(imp1, imp5)))
nrow(complete(rbind(imp1, mylist)))
nrow(complete(rbind(imp1, data.frame(mylist))))
nrow(complete(rbind(imp1, complete(imp5))))
}
\references{
van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
}
\seealso{
\code{\link[base:cbind]{cbind}}, \code{\link{ibind}},
\code{\link[=mids-class]{mids}}
}
\author{
Karin Groothuis-Oudshoorn, Stef van Buuren
}
\keyword{manip}
mice/man/make.formulas.Rd 0000644 0001762 0000144 00000002256 14330031606 014776 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/formula.R
\name{make.formulas}
\alias{make.formulas}
\title{Creates a \code{formulas} argument}
\usage{
make.formulas(data, blocks = make.blocks(data), predictorMatrix = NULL)
}
\arguments{
\item{data}{A \code{data.frame} with the source data}
\item{blocks}{An optional specification for blocks of variables in
the rows. The default assigns each variable in its own block.}
\item{predictorMatrix}{A \code{predictorMatrix} specified by the user.}
}
\value{
A list of formula's.
}
\description{
This helper function creates a valid \code{formulas} object. The
\code{formulas} object is an argument to the \code{mice} function.
It is a list of formula's that specifies the target variables and
the predictors by means of the standard \code{~} operator.
}
\examples{
f1 <- make.formulas(nhanes)
f1
f2 <- make.formulas(nhanes, blocks = make.blocks(nhanes, "collect"))
f2
# for editing, it may be easier to work with the character vector
c1 <- as.character(f1)
c1
# fold it back into a formula list
f3 <- name.formulas(lapply(c1, as.formula))
f3
}
\seealso{
\code{\link{make.blocks}}, \code{\link{make.predictorMatrix}}
}
mice/man/extend.formula.Rd 0000644 0001762 0000144 00000001477 14330031606 015171 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/formula.R
\name{extend.formula}
\alias{extend.formula}
\title{Extends a formula with predictors}
\usage{
extend.formula(
formula = ~0,
predictors = NULL,
auxiliary = TRUE,
include.intercept = FALSE,
...
)
}
\arguments{
\item{formula}{A formula. If it is
not a formula, the formula is internally reset to \code{~0}.}
\item{predictors}{A character vector of variable names.}
\item{auxiliary}{A logical that indicates whether the variables
listed in \code{predictors} should be added to the formula as main
effects. The default is \code{TRUE}.}
\item{include.intercept}{A logical that indicated whether the intercept
should be included in the result.}
}
\value{
A formula
}
\description{
Extends a formula with predictors
}
\keyword{internal}
mice/man/name.formulas.Rd 0000644 0001762 0000144 00000004626 14330031606 015004 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/formula.R
\name{name.formulas}
\alias{name.formulas}
\title{Name formula list elements}
\usage{
name.formulas(formulas, prefix = "F")
}
\arguments{
\item{formulas}{A named list of formula's, or expressions that
can be converted into formula's by \code{as.formula}. List elements
correspond to blocks. The block to which the list element applies is
identified by its name, so list names must correspond to block names.
The \code{formulas} argument is an alternative to the
\code{predictorMatrix} argument that allows for more flexibility in
specifying imputation models, e.g., for specifying interaction terms.}
\item{prefix}{A character vector of length 1 with the prefix to
be using for naming any unnamed blocks with two or more variables.}
}
\value{
Named list of formulas
}
\description{
This helper function names any unnamed elements in the \code{formula}
list. This is a convenience function.
}
\details{
This function will name any unnamed list elements specified in
the optional argument \code{formula}. Unnamed formula's
consisting with just one response variable will be named
after this variable. Unnamed formula's containing more
than one variable will be named by the \code{prefix}
argument, padded by an integer sequence stating at 1.
}
\examples{
# fully conditionally specified main effects model
form1 <- list(
bmi ~ age + chl + hyp,
hyp ~ age + bmi + chl,
chl ~ age + bmi + hyp
)
form1 <- name.formulas(form1)
imp1 <- mice(nhanes, formulas = form1, print = FALSE, m = 1, seed = 12199)
# same model using dot notation
form2 <- list(bmi ~ ., hyp ~ ., chl ~ .)
form2 <- name.formulas(form2)
imp2 <- mice(nhanes, formulas = form2, print = FALSE, m = 1, seed = 12199)
identical(complete(imp1), complete(imp2))
# same model using repeated multivariate imputation
form3 <- name.blocks(list(all = bmi + hyp + chl ~ .))
imp3 <- mice(nhanes, formulas = form3, print = FALSE, m = 1, seed = 12199)
cmp3 <- complete(imp3)
identical(complete(imp1), complete(imp3))
# same model using predictorMatrix
imp4 <- mice(nhanes, print = FALSE, m = 1, seed = 12199, auxiliary = TRUE)
identical(complete(imp1), complete(imp4))
# different model: multivariate imputation for chl and bmi
form5 <- list(chl + bmi ~ ., hyp ~ bmi + age)
form5 <- name.formulas(form5)
imp5 <- mice(nhanes, formulas = form5, print = FALSE, m = 1, seed = 71712)
}
\seealso{
\code{\link{mice}}
}
mice/man/fico.Rd 0000644 0001762 0000144 00000002054 14330031606 013146 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/flux.R
\name{fico}
\alias{fico}
\title{Fraction of incomplete cases among cases with observed}
\usage{
fico(data)
}
\arguments{
\item{data}{A data frame or a matrix containing the incomplete data. Missing
values are coded as NA's.}
}
\value{
A vector of length \code{ncol(data)} of FICO statistics.
}
\description{
FICO is an outbound statistic defined by the fraction of incomplete cases
among cases with \code{Yj} observed (White and Carlin, 2010).
}
\references{
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/missing-data-pattern.html#sec:flux}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
White, I.R., Carlin, J.B. (2010). Bias and efficiency of multiple imputation
compared with complete-case analysis for missing covariate values.
\emph{Statistics in Medicine}, \emph{29}, 2920-2931.
}
\seealso{
\code{\link{fluxplot}}, \code{\link{flux}}, \code{\link{md.pattern}}
}
\author{
Stef van Buuren, 2012
}
\keyword{misc}
mice/man/lm.mids.Rd 0000644 0001762 0000144 00000002747 14330031647 013607 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/lm.R
\name{lm.mids}
\alias{lm.mids}
\title{Linear regression for \code{mids} object}
\usage{
lm.mids(formula, data, ...)
}
\arguments{
\item{formula}{a formula object, with the response on the left of a ~
operator, and the terms, separated by + operators, on the right. See the
documentation of \code{\link{lm}} and \code{\link{formula}} for details.}
\item{data}{An object of type 'mids', which stands for 'multiply imputed data
set', typically created by a call to function \code{mice()}.}
\item{\dots}{Additional parameters passed to \code{\link{lm}}}
}
\value{
An objects of class \code{mira}, which stands for 'multiply imputed
repeated analysis'. This object contains \code{data$m} distinct
\code{lm.objects}, plus some descriptive information.
}
\description{
Applies \code{lm()} to multiply imputed data set
}
\details{
This function is included for backward compatibility with V1.0. The function
is superseded by \code{\link{with.mids}}.
}
\examples{
imp <- mice(nhanes)
fit <- lm.mids(bmi ~ hyp + chl, data = imp)
fit
}
\references{
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
}
\seealso{
\code{\link{lm}}, \code{\link[=mids-class]{mids}}, \code{\link[=mira-class]{mira}}
}
\author{
Stef van Buuren, Karin Groothuis-Oudshoorn, 2000
}
\keyword{multivariate}
mice/man/mice.impute.lasso.logreg.Rd 0000644 0001762 0000144 00000006437 14330031647 017061 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.lasso.logreg.R
\name{mice.impute.lasso.logreg}
\alias{mice.impute.lasso.logreg}
\alias{lasso.logreg}
\title{Imputation by direct use of lasso logistic regression}
\usage{
mice.impute.lasso.logreg(y, ry, x, wy = NULL, nfolds = 10, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{nfolds}{The number of folds for the cross-validation of the lasso penalty.
The default is 10.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes univariate missing binary data using lasso logistic regression with bootstrap.
}
\details{
The method consists of the following steps:
\enumerate{
\item For a given y variable under imputation, draw a bootstrap version y*
with replacement from the observed cases \code{y[ry]}, and stores in x* the
corresponding values from \code{x[ry, ]}.
\item Fit a regularised (lasso) logistic regression with y* as the outcome,
and x* as predictors.
A vector of regression coefficients bhat is obtained.
All of these coefficients are considered random draws from the imputation model
parameters posterior distribution.
Same of these coefficients will be shrunken to 0.
\item Compute predicted scores for m.d., i.e. logit-1(X bhat)
\item Compare the score to a random (0,1) deviate, and impute.
}
The method is based on the Direct Use of Regularized Regression (DURR) proposed by
Zhao & Long (2016) and Deng et al (2016).
}
\references{
Deng, Y., Chang, C., Ido, M. S., & Long, Q. (2016). Multiple imputation for
general missing data patterns in the presence of high-dimensional data.
Scientific reports, 6(1), 1-10.
Zhao, Y., & Long, Q. (2016). Multiple imputation in the presence of
high-dimensional data. Statistical Methods in Medical Research, 25(5),
2021-2035.
}
\seealso{
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Edoardo Costantini, 2021
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/with.mids.Rd 0000644 0001762 0000144 00000003302 14330031647 014136 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/with.R
\name{with.mids}
\alias{with.mids}
\title{Evaluate an expression in multiple imputed datasets}
\usage{
\method{with}{mids}(data, expr, ...)
}
\arguments{
\item{data}{An object of type \code{mids}, which stands for 'multiply imputed
data set', typically created by a call to function \code{mice()}.}
\item{expr}{An expression to evaluate for each imputed data set. Formula's
containing a dot (notation for "all other variables") do not work.}
\item{\dots}{Not used}
}
\value{
An object of S3 class \code{\link[=mira-class]{mira}}
}
\description{
Performs a computation of each of imputed datasets in data.
}
\note{
Version 3.11.10 changed to tidy evaluation on a quosure. This change
should not affect any code that worked on previous versions.
It turned out that the latter statement was not true (#292).
Version 3.12.2 reverts to the old \code{with()} function.
}
\examples{
imp <- mice(nhanes2, m = 2, print = FALSE, seed = 14221)
# descriptive statistics
getfit(with(imp, table(hyp, age)))
# model fitting and testing
fit1 <- with(imp, lm(bmi ~ age + hyp + chl))
fit2 <- with(imp, glm(hyp ~ age + chl, family = binomial))
fit3 <- with(imp, anova(lm(bmi ~ age + chl)))
}
\references{
van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
}
\seealso{
\code{\link[=mids-class]{mids}}, \code{\link[=mira-class]{mira}}, \code{\link{pool}},
\code{\link{D1}}, \code{\link{D3}}, \code{\link{pool.r.squared}}
}
\author{
Karin Oudshoorn, Stef van Buuren 2009, 2012, 2020
}
\keyword{multivariate}
mice/man/mice.impute.lasso.select.logreg.Rd 0000644 0001762 0000144 00000007442 14330031647 020334 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.lasso.select.logreg.R
\name{mice.impute.lasso.select.logreg}
\alias{mice.impute.lasso.select.logreg}
\alias{lasso.select.logreg}
\title{Imputation by indirect use of lasso logistic regression}
\usage{
mice.impute.lasso.select.logreg(y, ry, x, wy = NULL, nfolds = 10, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{nfolds}{The number of folds for the cross-validation of the lasso penalty.
The default is 10.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes univariate missing data using logistic regression following a
preprocessing lasso variable selection step.
}
\details{
The method consists of the following steps:
\enumerate{
\item For a given \code{y} variable under imputation, fit a linear regression with lasso
penalty using \code{y[ry]} as dependent variable and \code{x[ry, ]} as predictors.
The coefficients that are not shrunk to 0 define the active set of predictors
that will be used for imputation.
\item Fit a logit with the active set of predictors, and find (bhat, V(bhat))
\item Draw BETA from N(bhat, V(bhat))
\item Compute predicted scores for m.d., i.e. logit-1(X BETA)
\item Compare the score to a random (0,1) deviate, and impute.
}
The user can specify a \code{predictorMatrix} in the \code{mice} call
to define which predictors are provided to this univariate imputation method.
The lasso regularization will select, among the variables indicated by
the user, the ones that are important for imputation at any given iteration.
Therefore, users may force the exclusion of a predictor from a given
imputation model by speficing a \code{0} entry.
However, a non-zero entry does not guarantee the variable will be used,
as this decision is ultimately made by the lasso variable selection
procedure.
The method is based on the Indirect Use of Regularized Regression (IURR) proposed by
Zhao & Long (2016) and Deng et al (2016).
}
\references{
Deng, Y., Chang, C., Ido, M. S., & Long, Q. (2016). Multiple imputation for
general missing data patterns in the presence of high-dimensional data.
Scientific reports, 6(1), 1-10.
Zhao, Y., & Long, Q. (2016). Multiple imputation in the presence of
high-dimensional data. Statistical Methods in Medical Research, 25(5),
2021-2035.
}
\seealso{
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Edoardo Costantini, 2021
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/ampute.Rd 0000644 0001762 0000144 00000030023 14433400023 013513 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/ampute.R
\name{ampute}
\alias{ampute}
\title{Generate missing data for simulation purposes}
\usage{
ampute(
data,
prop = 0.5,
patterns = NULL,
freq = NULL,
mech = "MAR",
weights = NULL,
std = TRUE,
cont = TRUE,
type = NULL,
odds = NULL,
bycases = TRUE,
run = TRUE
)
}
\arguments{
\item{data}{A complete data matrix or data frame. Values should be numeric.
Categorical variables should have been transformed to dummies.}
\item{prop}{A scalar specifying the proportion of missingness. Should be a value
between 0 and 1. Default is a missingness proportion of 0.5.}
\item{patterns}{A matrix or data frame of size #patterns by #variables where
\code{0} indicates that a variable should have missing values and \code{1} indicates
that a variable should remain complete. The user may specify as many patterns as
desired. One pattern (a vector) is possible as well. Default
is a square matrix of size #variables where each pattern has missingness on one
variable only (created with \code{\link{ampute.default.patterns}}). After the
amputation procedure, \code{\link{md.pattern}} can be used to investigate the
missing data patterns in the data.}
\item{freq}{A vector of length #patterns containing the relative frequency with
which the patterns should occur. For example, for three missing data patterns,
the vector could be \code{c(0.4, 0.4, 0.2)}, meaning that of all cases with
missing values, 40 percent should have pattern 1, 40 percent pattern 2 and 20
percent pattern 3. The vector should sum to 1. Default is an equal probability
for each pattern, created with \code{\link{ampute.default.freq}}.}
\item{mech}{A string specifying the missingness mechanism, either "MCAR"
(Missing Completely At Random), "MAR" (Missing At Random) or "MNAR" (Missing Not At
Random). Default is a MAR missingness mechanism.}
\item{weights}{A matrix or data frame of size #patterns by #variables. The matrix
contains the weights that will be used to calculate the weighted sum scores. For
a MAR mechanism, the weights of the variables that will be made incomplete should be
zero. For a MNAR mechanism, these weights could have any possible value. Furthermore,
the weights may differ between patterns and between variables. They may be negative
as well. Within each pattern, the relative size of the values are of importance.
The default weights matrix is made with \code{\link{ampute.default.weights}} and
returns a matrix with equal weights for all variables. In case of MAR, variables
that will be amputed will be weighted with \code{0}. For MNAR, variables
that will be observed will be weighted with \code{0}. If the mechanism is MCAR, the
weights matrix will not be used.}
\item{std}{Logical. Whether the weighted sum scores should be calculated with
standardized data or with non-standardized data. The latter is especially advised when
making use of train and test sets in order to prevent leakage.}
\item{cont}{Logical. Whether the probabilities should be based on a continuous
or a discrete distribution. If TRUE, the probabilities of being missing are based
on a continuous logistic distribution function. \code{\link{ampute.continuous}}
will be used to calculate and assign the probabilities. These probabilities will then
be based on the argument \code{type}. If FALSE, the probabilities of being missing are
based on a discrete distribution (\code{\link{ampute.discrete}}) based on the \code{odds}
argument. Default is TRUE.}
\item{type}{A string or vector of strings containing the type of missingness for each
pattern. Either \code{"LEFT"}, \code{"MID"}, \code{"TAIL"} or '\code{"RIGHT"}.
If a single missingness type is given, all patterns will be created with the same
type. If the missingness types should differ between patterns, a vector of missingness
types should be given. Default is RIGHT for all patterns and is the result of
\code{\link{ampute.default.type}}.}
\item{odds}{A matrix where #patterns defines the #rows. Each row should contain
the odds of being missing for the corresponding pattern. The number of odds values
defines in how many quantiles the sum scores will be divided. The odds values are
relative probabilities: a quantile with odds value 4 will have a probability of
being missing that is four times higher than a quantile with odds 1. The
number of quantiles may differ between the patterns, specify NA for cells remaining empty.
Default is 4 quantiles with odds values 1, 2, 3 and 4 and is created by
\code{\link{ampute.default.odds}}.}
\item{bycases}{Logical. If TRUE, the proportion of missingness is defined in
terms of cases. If FALSE, the proportion of missingness is defined in terms of
cells. Default is TRUE.}
\item{run}{Logical. If TRUE, the amputations are implemented. If FALSE, the
return object will contain everything except for the amputed data set.}
}
\value{
Returns an S3 object of class \code{\link{mads-class}} (multivariate
amputed data set)
}
\description{
This function generates multivariate missing data under a MCAR, MAR or MNAR
missing data mechanism. Imputation of data sets containing missing values can
be performed with \code{\link{mice}}.
}
\details{
This function generates missing values in complete data sets. Amputation of complete
data sets is useful for the evaluation of imputation techniques, such as multiple
imputation (performed with function \code{\link{mice}} in this package).
The basic strategy underlying multivariate imputation was suggested by
Don Rubin during discussions in the 90's. Brand (1997) created one particular
implementation, and his method found its way into the FCS paper
(Van Buuren et al, 2006).
Until recently, univariate amputation procedures were used to generate missing
data in complete, simulated data sets. With this approach, variables are made
incomplete one variable at a time. When more than one variable needs to be amputed,
the procedure is repeated multiple times.
With the univariate approach, it is difficult to relate the missingness on one
variable to the missingness on another variable. A multivariate amputation procedure
solves this issue and moreover, it does justice to the multivariate nature of
data sets. Hence, \code{ampute} is developed to perform multivariate amputation.
The idea behind the function is the specification of several missingness
patterns. Each pattern is a combination of variables with and without missing
values (denoted by \code{0} and \code{1} respectively). For example, one might
want to create two missingness patterns on a data set with four variables. The
patterns could be something like: \code{0,0,1,1} and \code{1,0,1,0}.
Each combination of zeros and ones may occur.
Furthermore, the researcher specifies the proportion of missingness, either the
proportion of missing cases or the proportion of missing cells, and the relative
frequency each pattern occurs. Consequently, the data is split into multiple subsets,
one subset per pattern. Now, each case is candidate for a certain missingness pattern,
but whether the case will have missing values eventually depends on other specifications.
The first of these specifications is the missing mechanism. There are three possible
mechanisms: the missingness depends completely on chance (MCAR), the missingness
depends on the values of the observed variables (i.e. the variables that remain
complete) (MAR) or on the values of the variables that will be made incomplete (MNAR).
For a discussion on how missingness mechanisms are related to the observed data,
we refer to \doi{10.1177/0049124118799376}{Schouten and Vink, 2018}.
When the user specifies the missingness mechanism to be \code{"MCAR"}, the candidates
have an equal probability of becoming incomplete. For a \code{"MAR"} or \code{"MNAR"} mechanism,
weighted sum scores are calculated. These scores are a linear combination of the
variables.
In order to calculate the weighted sum scores, the data is standardized. For this reason,
the data has to be numeric. Second, for each case, the values in
the data set are multiplied with the weights, specified by argument \code{weights}.
These weighted scores will be summed, resulting in a weighted sum score for each case.
The weights may differ between patterns and they may be negative or zero as well.
Naturally, in case of a MAR mechanism, the weights corresponding to the
variables that will be made incomplete, have a 0. Note that this may be
different for each pattern. In case of MNAR missingness, especially
the weights of the variables that will be made incomplete are of importance. However,
the other variables may be weighted as well.
It is the relative difference between the weights that will result in an effect
in the sum scores. For example, for the first missing data
pattern mentioned above, the weights for the third and fourth variables could
be set to 2 and 4. However, weight values of 0.2 and 0.4 will have the exact
same effect on the weighted sum score: the fourth variable is weighted twice as
much as variable 3.
Based on the weighted sum scores, either a discrete or continuous distribution
of probabilities is used to calculate whether a candidate will have missing values.
For a discrete distribution of probabilities, the weighted sum scores are
divided into subgroups of equal size (quantiles). Thereafter, the user
specifies for each subgroup the odds of being missing. Both the number of
subgroups and the odds values are important for the generation of missing data.
For example, for a RIGHT-like mechanism, scoring in one of the
higher quantiles should have high missingness odds, whereas for a MID-like
mechanism, the central groups should have higher odds. Again, not the size of
the odds values are of importance, but the relative distance between the values.
The continuous distributions of probabilities are based on the logistic distribution function.
The user can specify the type of missingness, which, again, may differ between patterns.
For an example and more explanation about how the arguments interact with each other,
we refer to the vignette
\href{https://rianneschouten.github.io/mice_ampute/vignette/ampute.html}{Generate missing values with ampute}
The amputation methodology is published in
\doi{10.1080/00949655.2018.1491577}{Schouten, Lugtig and Vink, 2018}.
}
\examples{
# start with a complete data set
compl_boys <- cc(boys)[1:3]
# Perform amputation with default settings
mads_boys <- ampute(data = compl_boys)
mads_boys$amp
# Change default matrices as desired
my_patterns <- mads_boys$patterns
my_patterns[1:3, 2] <- 0
my_weights <- mads_boys$weights
my_weights[2, 1] <- 2
my_weights[3, 1] <- 0.5
# Rerun amputation
my_mads_boys <- ampute(
data = compl_boys, patterns = my_patterns, freq =
c(0.3, 0.3, 0.4), weights = my_weights, type = c("RIGHT", "TAIL", "LEFT")
)
my_mads_boys$amp
}
\references{
Brand, J.P.L. (1999) \emph{Development, implementation and
evaluation of multiple imputation strategies for the statistical analysis of
incomplete data sets.} pp. 110-113. Dissertation. Rotterdam: Erasmus University.
Schouten, R.M., Lugtig, P and Vink, G. (2018)
{Generating missing values for simulation purposes: A multivariate amputation procedure.}.
\emph{Journal of Statistical Computation and Simulation}, 88(15): 1909-1930.
\doi{10.1080/00949655.2018.1491577}
Schouten, R.M. and Vink, G. (2018){The Dance of the Mechanisms: How Observed Information Influences the Validity of Missingness Assumptions}.
\emph{Sociological Methods and Research}, 50(3): 1243-1258.
\doi{10.1177/0049124118799376}
Van Buuren, S., Brand, J.P.L., Groothuis-Oudshoorn, C.G.M., Rubin, D.B. (2006)
{Fully conditional specification in multivariate imputation.}
\emph{Journal of Statistical Computation and Simulation}, 76(12): 1049-1064.
\doi{10.1080/10629360600810434}
Van Buuren, S. (2018) \href{https://stefvanbuuren.name/fimd/sec-FCS.html#sec:MICE}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
Vink, G. (2016) Towards a standardized evaluation of multiple imputation routines.
}
\seealso{
\code{\link{mads-class}}, \code{\link{bwplot}}, \code{\link{xyplot}},
\code{\link{mice}}
}
\author{
Rianne Schouten [aut, cre], Gerko Vink [aut], Peter Lugtig [ctb], 2016
}
mice/man/mice.impute.2lonly.mean.Rd 0000644 0001762 0000144 00000005552 14330031606 016610 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.2lonly.mean.R
\name{mice.impute.2lonly.mean}
\alias{mice.impute.2lonly.mean}
\alias{2lonly.mean}
\title{Imputation of most likely value within the class}
\usage{
mice.impute.2lonly.mean(y, ry, x, type, wy = NULL, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{type}{Vector of length \code{ncol(x)} identifying random and class
variables. The class variable (only one is allowed) is coded as \code{-2}.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Method \code{2lonly.mean} replicates the most likely value within
a class of a second-level variable. It works for numeric and
factor data. The function is primarily useful as a quick fixup for
data in which the second-level variable is inconsistent.
}
\details{
Observed values in \code{y} are averaged within the class, and
replicated to the missing \code{y} within that class.
This function is primarily useful for repairing incomplete data
that are constant within the class, but vary over classes.
For numeric variables, \code{mice.impute.2lonly.mean()} imputes the
class mean of \code{y}. If \code{y} is a second-level variable, then
conventionally all observed \code{y} will be identical within the
class, and the function just provides a quick fix for any
missing \code{y} by filling in the class mean.
For factor variables, \code{mice.impute.2lonly.mean()} imputes the
most frequently occuring category within the class.
If there are no observed \code{y} in the class, all entries of the
class are set to \code{NA}. Note that this may produce problems
later on in \code{mice} if imputation routines are called that
expects predictor data to be complete. Methods designed for
imputing this type of second-level variables include
\code{\link{mice.impute.2lonly.norm}} and
\code{\link{mice.impute.2lonly.pmm}}.
}
\references{
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-level2pred.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Boca Raton, FL.: Chapman & Hall/CRC Press.
}
\seealso{
Other univariate-2lonly:
\code{\link{mice.impute.2lonly.norm}()},
\code{\link{mice.impute.2lonly.pmm}()}
}
\author{
Gerko Vink, Stef van Buuren, 2019
}
\concept{univariate-2lonly}
\keyword{datagen}
mice/man/densityplot.mids.Rd 0000644 0001762 0000144 00000021155 14330031647 015547 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/densityplot.R
\name{densityplot.mids}
\alias{densityplot.mids}
\alias{densityplot}
\title{Density plot of observed and imputed data}
\usage{
\method{densityplot}{mids}(
x,
data,
na.groups = NULL,
groups = NULL,
as.table = TRUE,
plot.points = FALSE,
theme = mice.theme(),
mayreplicate = TRUE,
thicker = 2.5,
allow.multiple = TRUE,
outer = TRUE,
drop.unused.levels = lattice::lattice.getOption("drop.unused.levels"),
panel = lattice::lattice.getOption("panel.densityplot"),
default.prepanel = lattice::lattice.getOption("prepanel.default.densityplot"),
...,
subscripts = TRUE,
subset = TRUE
)
}
\arguments{
\item{x}{A \code{mids} object, typically created by \code{mice()} or
\code{mice.mids()}.}
\item{data}{Formula that selects the data to be plotted. This argument
follows the \pkg{lattice} rules for \emph{formulas}, describing the primary
variables (used for the per-panel display) and the optional conditioning
variables (which define the subsets plotted in different panels) to be used
in the plot.
The formula is evaluated on the complete data set in the \code{long} form.
Legal variable names for the formula include \code{names(x$data)} plus the
two administrative factors \code{.imp} and \code{.id}.
\bold{Extended formula interface:} The primary variable terms (both the LHS
\code{y} and RHS \code{x}) may consist of multiple terms separated by a
\sQuote{+} sign, e.g., \code{y1 + y2 ~ x | a * b}. This formula would be
taken to mean that the user wants to plot both \code{y1 ~ x | a * b} and
\code{y2 ~ x | a * b}, but with the \code{y1 ~ x} and \code{y2 ~ x} in
\emph{separate panels}. This behavior differs from standard \pkg{lattice}.
\emph{Only combine terms of the same type}, i.e. only factors or only
numerical variables. Mixing numerical and categorical data occasionally
produces odds labeling of vertical axis.
The function \code{densityplot} does not use the \code{y} terms in the
formula. Density plots for \code{x1} and \code{x2} are requested as \code{~
x1 + x2}.}
\item{na.groups}{An expression evaluating to a logical vector indicating
which two groups are distinguished (e.g. using different colors) in the
display. The environment in which this expression is evaluated in the
response indicator \code{is.na(x$data)}.
The default \code{na.group = NULL} contrasts the observed and missing data
in the LHS \code{y} variable of the display, i.e. groups created by
\code{is.na(y)}. The expression \code{y} creates the groups according to
\code{is.na(y)}. The expression \code{y1 & y2} creates groups by
\code{is.na(y1) & is.na(y2)}, and \code{y1 | y2} creates groups as
\code{is.na(y1) | is.na(y2)}, and so on.}
\item{groups}{This is the usual \code{groups} arguments in \pkg{lattice}. It
differs from \code{na.groups} because it evaluates in the completed data
\code{data.frame(complete(x, "long", inc=TRUE))} (as usual), whereas
\code{na.groups} evaluates in the response indicator. See
\code{\link{xyplot}} for more details. When both \code{na.groups} and
\code{groups} are specified, \code{na.groups} takes precedence, and
\code{groups} is ignored.}
\item{as.table}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{plot.points}{A logical used in \code{densityplot} that signals whether
the points should be plotted.}
\item{theme}{A named list containing the graphical parameters. The default
function \code{mice.theme} produces a short list of default colors, line
width, and so on. The extensive list may be obtained from
\code{trellis.par.get()}. Global graphical parameters like \code{col} or
\code{cex} in high-level calls are still honored, so first experiment with
the global parameters. Many setting consists of a pair. For example,
\code{mice.theme} defines two symbol colors. The first is for the observed
data, the second for the imputed data. The theme settings only exist during
the call, and do not affect the trellis graphical parameters.}
\item{mayreplicate}{A logical indicating whether color, line widths, and so
on, may be replicated. The graphical functions attempt to choose
"intelligent" graphical parameters. For example, the same color can be
replicated for different element, e.g. use all reds for the imputed data.
Replication may be switched off by setting the flag to \code{FALSE}, in order
to allow the user to gain full control.}
\item{thicker}{Used in \code{densityplot}. Multiplication factor of the line
width of the observed density. \code{thicker=1} uses the same thickness for
the observed and imputed data.}
\item{allow.multiple}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{outer}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{drop.unused.levels}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{panel}{See \code{\link{xyplot}}.}
\item{default.prepanel}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{\dots}{Further arguments, usually not directly processed by the
high-level functions documented here, but instead passed on to other
functions.}
\item{subscripts}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{subset}{See \code{\link[lattice:xyplot]{xyplot}}.}
}
\value{
The high-level functions documented here, as well as other high-level
Lattice functions, return an object of class \code{"trellis"}. The
\code{\link[lattice:update.trellis]{update}} method can be used to
subsequently update components of the object, and the
\code{\link[lattice:print.trellis]{print}} method (usually called by default)
will plot it on an appropriate plotting device.
}
\description{
Plotting methods for imputed data using \pkg{lattice}. \code{densityplot}
produces plots of the densities. The function
automatically separates the observed and imputed data. The
functions extend the usual features of \pkg{lattice}.
}
\details{
The argument \code{na.groups} may be used to specify (combinations of)
missingness in any of the variables. The argument \code{groups} can be used
to specify groups based on the variable values themselves. Only one of both
may be active at the same time. When both are specified, \code{na.groups}
takes precedence over \code{groups}.
Use the \code{subset} and \code{na.groups} together to plots parts of the
data. For example, select the first imputed data set by by
\code{subset=.imp==1}.
Graphical parameters like \code{col}, \code{pch} and \code{cex} can be
specified in the arguments list to alter the plotting symbols. If
\code{length(col)==2}, the color specification to define the observed and
missing groups. \code{col[1]} is the color of the 'observed' data,
\code{col[2]} is the color of the missing or imputed data. A convenient color
choice is \code{col=mdc(1:2)}, a transparent blue color for the observed
data, and a transparent red color for the imputed data. A good choice is
\code{col=mdc(1:2), pch=20, cex=1.5}. These choices can be set for the
duration of the session by running \code{mice.theme()}.
}
\note{
The first two arguments (\code{x} and \code{data}) are reversed
compared to the standard Trellis syntax implemented in \pkg{lattice}. This
reversal was necessary in order to benefit from automatic method dispatch.
In \pkg{mice} the argument \code{x} is always a \code{mids} object, whereas
in \pkg{lattice} the argument \code{x} is always a formula.
In \pkg{mice} the argument \code{data} is always a formula object, whereas in
\pkg{lattice} the argument \code{data} is usually a data frame.
All other arguments have identical interpretation.
\code{densityplot} errs on empty groups, which occurs if all observations in
the subgroup contain \code{NA}. The relevant error message is: \code{Error in
density.default: ... need at least 2 points to select a bandwidth
automatically}. There is yet no workaround for this problem. Use the more
robust \code{bwplot} or \code{stripplot} as a replacement.
}
\examples{
imp <- mice(boys, maxit = 1)
### density plot of head circumference per imputation
### blue is observed, red is imputed
densityplot(imp, ~ hc | .imp)
### All combined in one panel.
densityplot(imp, ~hc)
}
\references{
Sarkar, Deepayan (2008) \emph{Lattice: Multivariate Data
Visualization with R}, Springer.
van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}: Multivariate
Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
}
\seealso{
\code{\link{mice}}, \code{\link{xyplot}}, \code{\link{stripplot}},
\code{\link{bwplot}}, \code{\link{lattice}} for an overview of the
package, as well as \code{\link[lattice:histogram]{densityplot}},
\code{\link[lattice:panel.densityplot]{panel.densityplot}},
\code{\link[lattice:print.trellis]{print.trellis}},
\code{\link[lattice:trellis.par.get]{trellis.par.set}}
}
\author{
Stef van Buuren
}
\keyword{hplot}
mice/man/mice.impute.2l.lmer.Rd 0000644 0001762 0000144 00000006105 14330031606 015720 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.2l.lmer.R
\name{mice.impute.2l.lmer}
\alias{mice.impute.2l.lmer}
\title{Imputation by a two-level normal model using \code{lmer}}
\usage{
mice.impute.2l.lmer(y, ry, x, type, wy = NULL, intercept = TRUE, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{type}{Vector of length \code{ncol(x)} identifying random and class
variables. Random variables are identified by a '2'. The class variable
(only one is allowed) is coded as '-2'. Fixed effects are indicated by
a '1'.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{intercept}{Logical determining whether the intercept is automatically
added.}
\item{\dots}{Arguments passed down to \code{lmer}}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes univariate systematically and sporadically missing data using a
two-level normal model using \code{lme4::lmer()}.
}
\details{
Data are missing systematically if they have not been measured, e.g., in the
case where we combine data from different sources. Data are missing sporadically
if they have been partially observed.
While the method is fully Bayesian, it may fix parameters of the
variance-covariance matrix or the random effects to their estimated
value in cases where creating draws from the posterior is not
possible. The procedure throws a warning when this happens.
If \code{lme4::lmer()} fails, the procedure prints the warning
\code{"lmer does not run. Simplify imputation model"} and returns the
current imputation. If that happens we see flat lines in the
trace line plots. Thus, the appearance of flat trace lines should be taken
as an additional alert to a problem with imputation model fitting.
}
\references{
Jolani S. (2017) Hierarchical imputation of systematically and
sporadically missing data: An approximate Bayesian approach using
chained equations. Forthcoming.
Jolani S., Debray T.P.A., Koffijberg H., van Buuren S., Moons K.G.M. (2015).
Imputation of systematically missing predictors in an individual
participant data meta-analysis: a generalized approach using MICE.
\emph{Statistics in Medicine}, 34:1841-1863.
Van Buuren, S. (2011) Multiple imputation of multilevel data. In Hox, J.J.
and and Roberts, J.K. (Eds.), \emph{The Handbook of Advanced Multilevel
Analysis}, Chapter 10, pp. 173--196. Milton Park, UK: Routledge.
}
\seealso{
Other univariate-2l:
\code{\link{mice.impute.2l.bin}()},
\code{\link{mice.impute.2l.norm}()},
\code{\link{mice.impute.2l.pan}()}
}
\author{
Shahab Jolani, 2017
}
\concept{univariate-2l}
\keyword{datagen}
mice/man/futuremice.Rd 0000644 0001762 0000144 00000010325 14422737141 014407 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/futuremice.R
\name{futuremice}
\alias{futuremice}
\title{Wrapper function that runs MICE in parallel}
\usage{
futuremice(
data,
m = 5,
parallelseed = NA,
n.core = NULL,
seed = NA,
use.logical = TRUE,
future.plan = "multisession",
packages = NULL,
globals = NULL,
...
)
}
\arguments{
\item{data}{A data frame or matrix containing the incomplete data. Similar to
the first argument of \code{\link{mice}}.}
\item{m}{The number of desired imputated datasets. By default $m=5$ as with
\code{mice}}
\item{parallelseed}{A scalar to be used to obtain reproducible results over
the futures. The default \code{parallelseed = NA} will result in a seed value
that is randomly drawn between -999999999 and 999999999.}
\item{n.core}{A scalar indicating the number of cores that should be used.}
\item{seed}{A scalar to be used as the seed value for the mice algorithm
within each parallel stream. Please note that the imputations will be the
same for all streams and, hence, this should be used if and only if
\code{n.core = 1} and if it is desired to obtain the same output as under
\code{mice}.}
\item{use.logical}{A logical indicating whether logical (\code{TRUE}) or
physical (\code{FALSE}) CPU's on machine should be used.}
\item{future.plan}{A character indicating how \code{future}s are resolved.
The default \code{multisession} resolves futures asynchronously (in parallel)
in separate \code{R} sessions running in the background. See
\code{\link[future]{plan}} for more information on future plans.}
\item{packages}{A character vector with additional packages to be used in
\code{mice} (e.g., for using external imputation functions).}
\item{globals}{A character string with additional functions to be exported to
each future (e.g., user-written imputation functions).}
\item{...}{Named arguments that are passed down to function \code{\link{mice}}.}
}
\value{
A mids object as defined by \code{\link{mids-class}}
}
\description{
This is a wrapper function for \code{\link{mice}}, using multiple cores to
execute \code{\link{mice}} in parallel. As a result, the imputation
procedure can be sped up, which may be useful in general. By default,
\code{\link{futuremice}} distributes the number of imputations \code{m}
about equally over the cores.
}
\details{
This function relies on package \code{\link[furrr]{furrr}}, which is a
package for R versions 3.2.0 and later. We have chosen to use furrr function
\code{future_map} to allow the use of \code{futuremice} on Mac, Linux and
Windows systems.
This wrapper function combines the output of \code{\link[furrr]{future_map}} with
function \code{\link{ibind}} from the \code{\link{mice}} package. A
\code{mids} object is returned and can be used for further analyses.
A seed value can be specified in the global environment, which will yield
reproducible results. A seed value can also be specified within the
\code{\link{futuremice}} call, through specifying the argument
\code{parallelseed}. If \code{parallelseed} is not specified, a seed value is
drawn randomly by default, and accessible through \code{$parallelseed} in the
output object. Hence, results will always be reproducible, regardless of
whether the seed is specified in the global environment, or by setting the
same seed within the function (potentially by extracting the seed from the
\code{futuremice} output object.
}
\examples{
# 150 imputations in dataset nhanes, performed by 3 cores
\dontrun{
imp1 <- futuremice(data = nhanes, m = 150, n.core = 3)
# Making use of arguments in mice.
imp2 <- futuremice(data = nhanes, m = 100, method = "norm.nob")
imp2$method
fit <- with(imp2, lm(bmi ~ hyp))
pool(fit)
}
}
\references{
Volker, T.B. and Vink, G. (2022). futuremice: The future starts today.
\url{https://www.gerkovink.com/miceVignettes/futuremice/Vignette_futuremice.html}
#'Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/parallel-computation.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\seealso{
\code{\link[future]{future}}, \code{\link[furrr]{furrr}}, \code{\link[furrr]{future_map}},
\code{\link[future]{plan}}, \code{\link{mice}}, \code{\link{mids-class}}
}
\author{
Thom Benjamin Volker, Gerko Vink
}
mice/man/mice.impute.polr.Rd 0000644 0001762 0000144 00000011042 14330031647 015422 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.polr.R
\name{mice.impute.polr}
\alias{mice.impute.polr}
\title{Imputation of ordered data by polytomous regression}
\usage{
mice.impute.polr(
y,
ry,
x,
wy = NULL,
nnet.maxit = 100,
nnet.trace = FALSE,
nnet.MaxNWts = 1500,
polr.to.loggedEvents = FALSE,
...
)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{nnet.maxit}{Tuning parameter for \code{nnet()}.}
\item{nnet.trace}{Tuning parameter for \code{nnet()}.}
\item{nnet.MaxNWts}{Tuning parameter for \code{nnet()}.}
\item{polr.to.loggedEvents}{A logical indicating whether each fallback
to the \code{multinom()} function should be written to \code{loggedEvents}.
The default is \code{FALSE}.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes missing data in a categorical variable using polytomous regression
}
\details{
The function \code{mice.impute.polr()} imputes for ordered categorical response
variables by the proportional odds logistic regression (polr) model. The
function repeatedly applies logistic regression on the successive splits. The
model is also known as the cumulative link model.
By default, ordered factors with more than two levels are imputed by
\code{mice.impute.polr}.
The algorithm of \code{mice.impute.polr} uses the function \code{polr()} from
the \code{MASS} package.
In order to avoid bias due to perfect prediction, the algorithm augment the
data according to the method of White, Daniel and Royston (2010).
The call to \code{polr} might fail, usually because the data are very sparse.
In that case, \code{multinom} is tried as a fallback.
If the local flag \code{polr.to.loggedEvents} is set to TRUE,
a record is written
to the \code{loggedEvents} component of the \code{\link{mids}} object.
Use \code{mice(data, polr.to.loggedEvents = TRUE)} to set the flag.
}
\note{
In December 2019 Simon White alerted that the
\code{polr} could always fail silently. I can confirm this behaviour for
versions \code{mice 3.0.0 - mice 3.6.6}, so any method requests
for \code{polr} in these versions were in fact handled by \code{multinom}.
See \url{https://github.com/amices/mice/issues/206} for details.
}
\references{
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}: Multivariate
Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
Brand, J.P.L. (1999) \emph{Development, implementation and evaluation of
multiple imputation strategies for the statistical analysis of incomplete
data sets.} Dissertation. Rotterdam: Erasmus University.
White, I.R., Daniel, R. Royston, P. (2010). Avoiding bias due to perfect
prediction in multiple imputation of incomplete categorical variables.
\emph{Computational Statistics and Data Analysis}, 54, 2267-2275.
Venables, W.N. & Ripley, B.D. (2002). \emph{Modern applied statistics with
S-Plus (4th ed)}. Springer, Berlin.
}
\seealso{
\code{\link{mice}}, \code{\link[nnet]{multinom}},
\code{\link[MASS]{polr}}
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Stef van Buuren, Karin Groothuis-Oudshoorn, 2000-2010
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/mammalsleep.Rd 0000644 0001762 0000144 00000003642 14330031606 014527 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mammalsleep.R
\docType{data}
\name{mammalsleep}
\alias{mammalsleep}
\alias{sleep}
\title{Mammal sleep data}
\format{
\code{mammalsleep} is a data frame with 62 rows and 11 columns:
\describe{
\item{species}{Species of animal}
\item{bw}{Body weight (kg)}
\item{brw}{Brain weight (g)}
\item{sws}{Slow wave ("nondreaming") sleep (hrs/day)}
\item{ps}{Paradoxical ("dreaming") sleep (hrs/day)}
\item{ts}{Total sleep (hrs/day) (sum of slow wave and paradoxical sleep)}
\item{mls}{Maximum life span (years)}
\item{gt}{Gestation time (days)}
\item{pi}{Predation index (1-5), 1 = least likely to be preyed upon}
\item{sei}{Sleep exposure index (1-5), 1 = least exposed (e.g. animal sleeps in a
well-protected den), 5 = most exposed}
\item{odi}{Overall danger index (1-5) based on the above two indices and other information, 1 = least
danger (from other animals), 5 = most danger (from other animals)}
}
}
\source{
Allison, T., Cicchetti, D.V. (1976). Sleep in Mammals: Ecological and
Constitutional Correlates. Science, 194(4266), 732-734.
}
\description{
Dataset from Allison and Cicchetti (1976) of 62 mammal species on the
interrelationship between sleep, ecological, and constitutional variables.
The dataset contains missing values on five variables.
}
\details{
Allison and Cicchetti (1976) investigated the interrelationship between
sleep, ecological, and constitutional variables. They assessed these
variables for 39 mammalian species. The authors concluded that slow-wave
sleep is negatively associated with a factor related to body size. This
suggests that large amounts of this sleep phase are disadvantageous in large
species. Also, paradoxical sleep (REM sleep) was associated with a factor
related to predatory danger, suggesting that large amounts of this sleep
phase are disadvantageous in prey species.
}
\examples{
sleep <- data(mammalsleep)
}
\keyword{datasets}
mice/man/D2.Rd 0000644 0001762 0000144 00000003356 14436133175 012514 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/D2.R
\name{D2}
\alias{D2}
\title{Compare two nested models using D2-statistic}
\usage{
D2(fit1, fit0 = NULL, use = "wald")
}
\arguments{
\item{fit1}{An object of class \code{mira}, produced by \code{with()}.}
\item{fit0}{An object of class \code{mira}, produced by \code{with()}. The
model in \code{fit0} is a nested within \code{fit1}. The default null
model \code{fit0 = NULL} compares \code{fit1} to the intercept-only model.}
\item{use}{A character string denoting Wald- or likelihood-based based tests. Can be either \code{"wald"} or \code{"likelihood"}. Only used if \code{method = "D2"}.}
}
\description{
The D2-statistic pools test statistics from the repeated analyses.
The method is less powerful than the D1- and D3-statistics.
}
\note{
Warning: `D2()` assumes that the order of the variables is the
same in different models. See
\url{https://github.com/amices/mice/issues/420} for details.
}
\examples{
# Compare two linear models:
imp <- mice(nhanes2, seed = 51009, print = FALSE)
mi1 <- with(data = imp, expr = lm(bmi ~ age + hyp + chl))
mi0 <- with(data = imp, expr = lm(bmi ~ age + hyp))
D2(mi1, mi0)
\dontrun{
# Compare two logistic regression models
imp <- mice(boys, maxit = 2, print = FALSE)
fit1 <- with(imp, glm(gen > levels(gen)[1] ~ hgt + hc + reg, family = binomial))
fit0 <- with(imp, glm(gen > levels(gen)[1] ~ hgt + hc, family = binomial))
D2(fit1, fit0)
}
}
\references{
Li, K. H., X. L. Meng, T. E. Raghunathan, and D. B. Rubin. 1991.
Significance Levels from Repeated p-Values with Multiply-Imputed Data.
\emph{Statistica Sinica} 1 (1): 65–92.
\url{https://stefvanbuuren.name/fimd/sec-multiparameter.html#sec:chi}
}
\seealso{
\code{\link[mitml]{testModels}}
}
mice/man/is.mipo.Rd 0000644 0001762 0000144 00000000515 13666252075 013624 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/is.R
\name{is.mipo}
\alias{is.mipo}
\title{Check for \code{mipo} object}
\usage{
is.mipo(x)
}
\arguments{
\item{x}{An object}
}
\value{
A logical indicating whether \code{x} is an object of class \code{mipo}
}
\description{
Check for \code{mipo} object
}
mice/man/xyplot.mids.Rd 0000644 0001762 0000144 00000016364 14330031647 014536 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/xyplot.R
\name{xyplot.mids}
\alias{xyplot.mids}
\alias{xyplot}
\title{Scatterplot of observed and imputed data}
\usage{
\method{xyplot}{mids}(
x,
data,
na.groups = NULL,
groups = NULL,
as.table = TRUE,
theme = mice.theme(),
allow.multiple = TRUE,
outer = TRUE,
drop.unused.levels = lattice::lattice.getOption("drop.unused.levels"),
...,
subscripts = TRUE,
subset = TRUE
)
}
\arguments{
\item{x}{A \code{mids} object, typically created by \code{mice()} or
\code{mice.mids()}.}
\item{data}{Formula that selects the data to be plotted. This argument
follows the \pkg{lattice} rules for \emph{formulas}, describing the primary
variables (used for the per-panel display) and the optional conditioning
variables (which define the subsets plotted in different panels) to be used
in the plot.
The formula is evaluated on the complete data set in the \code{long} form.
Legal variable names for the formula include \code{names(x$data)} plus the
two administrative factors \code{.imp} and \code{.id}.
\bold{Extended formula interface:} The primary variable terms (both the LHS
\code{y} and RHS \code{x}) may consist of multiple terms separated by a
\sQuote{+} sign, e.g., \code{y1 + y2 ~ x | a * b}. This formula would be
taken to mean that the user wants to plot both \code{y1 ~ x | a * b} and
\code{y2 ~ x | a * b}, but with the \code{y1 ~ x} and \code{y2 ~ x} in
\emph{separate panels}. This behavior differs from standard \pkg{lattice}.
\emph{Only combine terms of the same type}, i.e. only factors or only
numerical variables. Mixing numerical and categorical data occasionally
produces odds labeling of vertical axis.}
\item{na.groups}{An expression evaluating to a logical vector indicating
which two groups are distinguished (e.g. using different colors) in the
display. The environment in which this expression is evaluated in the
response indicator \code{is.na(x$data)}.
The default \code{na.group = NULL} contrasts the observed and missing data
in the LHS \code{y} variable of the display, i.e. groups created by
\code{is.na(y)}. The expression \code{y} creates the groups according to
\code{is.na(y)}. The expression \code{y1 & y2} creates groups by
\code{is.na(y1) & is.na(y2)}, and \code{y1 | y2} creates groups as
\code{is.na(y1) | is.na(y2)}, and so on.}
\item{groups}{This is the usual \code{groups} arguments in \pkg{lattice}. It
differs from \code{na.groups} because it evaluates in the completed data
\code{data.frame(complete(x, "long", inc=TRUE))} (as usual), whereas
\code{na.groups} evaluates in the response indicator. See
\code{\link{xyplot}} for more details. When both \code{na.groups} and
\code{groups} are specified, \code{na.groups} takes precedence, and
\code{groups} is ignored.}
\item{as.table}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{theme}{A named list containing the graphical parameters. The default
function \code{mice.theme} produces a short list of default colors, line
width, and so on. The extensive list may be obtained from
\code{trellis.par.get()}. Global graphical parameters like \code{col} or
\code{cex} in high-level calls are still honored, so first experiment with
the global parameters. Many setting consists of a pair. For example,
\code{mice.theme} defines two symbol colors. The first is for the observed
data, the second for the imputed data. The theme settings only exist during
the call, and do not affect the trellis graphical parameters.}
\item{allow.multiple}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{outer}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{drop.unused.levels}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{\dots}{Further arguments, usually not directly processed by the
high-level functions documented here, but instead passed on to other
functions.}
\item{subscripts}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{subset}{See \code{\link[lattice:xyplot]{xyplot}}.}
}
\value{
The high-level functions documented here, as well as other high-level
Lattice functions, return an object of class \code{"trellis"}. The
\code{\link[lattice:update.trellis]{update}} method can be used to
subsequently update components of the object, and the
\code{\link[lattice:print.trellis]{print}} method (usually called by default)
will plot it on an appropriate plotting device.
}
\description{
Plotting methods for imputed data using \pkg{lattice}.
\code{xyplot()} produces a conditional scatterplots. The function
automatically separates the observed (blue) and imputed (red) data. The
function extends the usual features of \pkg{lattice}.
}
\details{
The argument \code{na.groups} may be used to specify (combinations of)
missingness in any of the variables. The argument \code{groups} can be used
to specify groups based on the variable values themselves. Only one of both
may be active at the same time. When both are specified, \code{na.groups}
takes precedence over \code{groups}.
Use the \code{subset} and \code{na.groups} together to plots parts of the
data. For example, select the first imputed data set by by
\code{subset=.imp==1}.
Graphical parameters like \code{col}, \code{pch} and \code{cex} can be
specified in the arguments list to alter the plotting symbols. If
\code{length(col)==2}, the color specification to define the observed and
missing groups. \code{col[1]} is the color of the 'observed' data,
\code{col[2]} is the color of the missing or imputed data. A convenient color
choice is \code{col=mdc(1:2)}, a transparent blue color for the observed
data, and a transparent red color for the imputed data. A good choice is
\code{col=mdc(1:2), pch=20, cex=1.5}. These choices can be set for the
duration of the session by running \code{mice.theme()}.
}
\note{
The first two arguments (\code{x} and \code{data}) are reversed
compared to the standard Trellis syntax implemented in \pkg{lattice}. This
reversal was necessary in order to benefit from automatic method dispatch.
In \pkg{mice} the argument \code{x} is always a \code{mids} object, whereas
in \pkg{lattice} the argument \code{x} is always a formula.
In \pkg{mice} the argument \code{data} is always a formula object, whereas in
\pkg{lattice} the argument \code{data} is usually a data frame.
All other arguments have identical interpretation.
}
\examples{
imp <- mice(boys, maxit = 1)
# xyplot: scatterplot by imputation number
# observe the erroneous outlying imputed values
# (caused by imputing hgt from bmi)
xyplot(imp, hgt ~ age | .imp, pch = c(1, 20), cex = c(1, 1.5))
# same, but label with missingness of wgt (four cases)
xyplot(imp, hgt ~ age | .imp, na.group = wgt, pch = c(1, 20), cex = c(1, 1.5))
}
\references{
Sarkar, Deepayan (2008) \emph{Lattice: Multivariate Data
Visualization with R}, Springer.
van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}: Multivariate
Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
}
\seealso{
\code{\link{mice}}, \code{\link{stripplot}}, \code{\link{densityplot}},
\code{\link{bwplot}}, \code{\link{lattice}} for an overview of the
package, as well as \code{\link[lattice:xyplot]{xyplot}},
\code{\link[lattice:panel.xyplot]{panel.xyplot}},
\code{\link[lattice:print.trellis]{print.trellis}},
\code{\link[lattice:trellis.par.get]{trellis.par.set}}
}
\author{
Stef van Buuren
}
\keyword{hplot}
mice/man/complete.mids.Rd 0000644 0001762 0000144 00000006540 14330031606 014775 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/complete.R
\name{complete.mids}
\alias{complete.mids}
\alias{complete}
\title{Extracts the completed data from a \code{mids} object}
\usage{
\method{complete}{mids}(data, action = 1L, include = FALSE, mild = FALSE, ...)
}
\arguments{
\item{data}{An object of class \code{mids} as created by the function
\code{mice()}.}
\item{action}{A numeric vector or a keyword. Numeric
values between 1 and \code{data$m} return the data with
imputation number \code{action} filled in. The value of \code{action = 0}
return the original data, with missing values. \code{action} can
also be one of the following keywords: \code{"all"}, \code{"long"},
\code{"broad"} and \code{"repeated"}. See the Details section
for the interpretation.
The default is \code{action = 1L} returns the first imputed data set.}
\item{include}{A logical to indicate whether the original data with the missing
values should be included.}
\item{mild}{A logical indicating whether the return value should
always be an object of class \code{mild}. Setting \code{mild = TRUE}
overrides \code{action} keywords \code{"long"}, \code{"broad"}
and \code{"repeated"}. The default is \code{FALSE}.}
\item{\dots}{Additional arguments. Not used.}
}
\value{
Complete data set with missing values replaced by imputations.
A \code{data.frame}, or a list of data frames of class \code{mild}.
}
\description{
Takes an object of class \code{mids}, fills in the missing data, and returns
the completed data in a specified format.
}
\details{
The argument \code{action} can be length-1 character, which is
matched to one of the following keywords:
\describe{
\item{\code{"all"}}{produces a \code{mild} object of imputed data sets. When
\code{include = TRUE}, then the original data are appended as the first list
element;}
\item{\code{"long"}}{ produces a data set where imputed data sets
are stacked vertically. The columns are added: 1) \code{.imp}, integer,
referring the imputation number, and 2) \code{.id}, character, the row
names of \code{data$data};}
\item{\code{"stacked"}}{ same as \code{"long"} but without the two
additional columns;}
\item{\code{"broad"}}{ produces a data set with where imputed data sets
are stacked horizontally. Columns are ordered as in the original data.
The imputation number is appended to each column name;}
\item{\code{"repeated"}}{ same as \code{"broad"}, but with
columns in a different order.}
}
}
\note{
Technical note: \code{mice 3.7.5} renamed the \code{complete()} function
to \code{complete.mids()} and exported it as an S3 method of the
generic \code{tidyr::complete()}. Name clashes between
\code{mice::complete()} and \code{tidyr::complete()} should no
longer occur.
}
\examples{
# obtain first imputed data set
sum(is.na(nhanes2))
imp <- mice(nhanes2, print = FALSE, maxit = 1)
dat <- complete(imp)
sum(is.na(dat))
# obtain stacked third and fifth imputation
dat <- complete(imp, c(3, 5))
# obtain all datasets, with additional identifiers
head(complete(imp, "long"))
# same, but now as list, mild object
dslist <- complete(imp, "all")
length(dslist)
# same, but also include the original data
dslist <- complete(imp, "all", include = TRUE)
length(dslist)
# select original + 3 + 5, store as mild
dslist <- complete(imp, c(0, 3, 5), mild = TRUE)
names(dslist)
}
\seealso{
\code{\link{mice}}, \code{\link[=mids-class]{mids}}
}
\keyword{manip}
mice/man/construct.blocks.Rd 0000644 0001762 0000144 00000004375 14330031606 015536 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/blocks.R
\name{construct.blocks}
\alias{construct.blocks}
\title{Construct blocks from \code{formulas} and \code{predictorMatrix}}
\usage{
construct.blocks(formulas = NULL, predictorMatrix = NULL)
}
\arguments{
\item{formulas}{A named list of formula's, or expressions that
can be converted into formula's by \code{as.formula}. List elements
correspond to blocks. The block to which the list element applies is
identified by its name, so list names must correspond to block names.
The \code{formulas} argument is an alternative to the
\code{predictorMatrix} argument that allows for more flexibility in
specifying imputation models, e.g., for specifying interaction terms.}
\item{predictorMatrix}{A numeric matrix of \code{length(blocks)} rows
and \code{ncol(data)} columns, containing 0/1 data specifying
the set of predictors to be used for each target column.
Each row corresponds to a variable block, i.e., a set of variables
to be imputed. A value of \code{1} means that the column
variable is used as a predictor for the target block (in the rows).
By default, the \code{predictorMatrix} is a square matrix of \code{ncol(data)}
rows and columns with all 1's, except for the diagonal.
Note: For two-level imputation models (which have \code{"2l"} in their names)
other codes (e.g, \code{2} or \code{-2}) are also allowed.}
}
\value{
A \code{blocks} object.
}
\description{
This helper function attempts to find blocks of variables in the
specification of the \code{formulas} and/or \code{predictorMatrix}
objects. Blocks specified by \code{formulas} may consist of
multiple variables. Blocks specified by \code{predictorMatrix} are
assumed to consist of single variables. Any duplicates in names are
removed, and the formula specification is preferred.
\code{predictorMatrix} and \code{formulas}. When both arguments
specify models for the same block, the model for the
\code{predictMatrix} is removed, and priority is given to the
specification given in \code{formulas}.
}
\examples{
form <- name.formulas(list(bmi + hyp ~ chl + age, chl ~ bmi))
pred <- make.predictorMatrix(nhanes[, c("age", "chl")])
construct.blocks(formulas = form, pred = pred)
}
\seealso{
\code{\link{make.blocks}}, \code{\link{name.blocks}}
}
mice/man/nhanes.Rd 0000644 0001762 0000144 00000001667 14330031606 013513 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/nhanes.R
\docType{data}
\name{nhanes}
\alias{nhanes}
\title{NHANES example - all variables numerical}
\format{
A data frame with 25 observations on the following 4 variables.
\describe{
\item{age}{Age group (1=20-39, 2=40-59, 3=60+)}
\item{bmi}{Body mass index (kg/m**2)}
\item{hyp}{Hypertensive (1=no,2=yes)}
\item{chl}{Total serum cholesterol (mg/dL)} }
}
\source{
Schafer, J.L. (1997). \emph{Analysis of Incomplete Multivariate
Data.} London: Chapman & Hall. Table 6.14.
}
\description{
A small data set with non-monotone missing values.
}
\details{
A small data set with all numerical variables. The data set \code{nhanes2} is
the same data set, but with \code{age} and \code{hyp} treated as factors.
}
\examples{
# create 5 imputed data sets
imp <- mice(nhanes)
# print the first imputed data set
complete(imp)
}
\seealso{
\code{\link{nhanes2}}
}
\keyword{datasets}
mice/man/mids-class.Rd 0000644 0001762 0000144 00000011375 14334445701 014304 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mids.R
\name{mids-class}
\alias{mids-class}
\alias{mids}
\title{Multiply imputed data set (\code{mids})}
\description{
The \code{mids} object contains a multiply imputed data set. The \code{mids} object is
generated by functions \code{mice()}, \code{mice.mids()}, \code{cbind.mids()},
\code{rbind.mids()} and \code{ibind.mids()}.
}
\details{
The \code{mids}
class of objects has methods for the following generic functions:
\code{print}, \code{summary}, \code{plot}.
The \code{loggedEvents} entry is a matrix with five columns containing a
record of automatic removal actions. It is \code{NULL} is no action was
made. At initialization the program does the following three actions:
\describe{
\item{1}{A variable that contains missing values, that is not imputed
and that is used as a predictor is removed}
\item{2}{A constant variable is removed}
\item{3}{A collinear variable is removed.}
}
During iteration, the program does the following
actions:
\describe{
\item{1}{One or more variables that are linearly dependent are removed
(for categorical data, a 'variable' corresponds to a dummy variable)}
\item{2}{Proportional odds regression imputation that does not converge
and is replaced by \code{polyreg}.}
}
Explanation of elements in \code{loggedEvents}:
\describe{
\item{\code{it}}{iteration number at which the record was added,}
\item{\code{im}}{imputation number,}
\item{\code{dep}}{name of the dependent variable,}
\item{\code{meth}}{imputation method used,}
\item{\code{out}}{a (possibly long) character vector with the
names of the altered or removed predictors.}
}
}
\note{
The \code{mice} package does not use
the S4 class definitions, and instead relies on the S3 list
equivalent \code{oldClass(obj) <- "mids"}.
}
\section{Slots}{
\describe{
\item{\code{.Data}:}{Object of class \code{"list"} containing the
following slots:}
\item{\code{data}:}{Original (incomplete) data set.}
\item{\code{imp}:}{A list of \code{ncol(data)} components with
the generated multiple imputations. Each list components is a
\code{data.frame} (\code{nmis[j]} by \code{m}) of imputed values
for variable \code{j}.}
\item{\code{m}:}{Number of imputations.}
\item{\code{where}:}{The \code{where} argument of the
\code{mice()} function.}
\item{\code{blocks}:}{The \code{blocks} argument of the
\code{mice()} function.}
\item{\code{call}:}{Call that created the object.}
\item{\code{nmis}:}{An array containing the number of missing
observations per column.}
\item{\code{method}:}{A vector of strings of \code{length(blocks}
specifying the imputation method per block.}
\item{\code{predictorMatrix}:}{A numerical matrix of containing
integers specifying the predictor set.}
\item{\code{visitSequence}:}{The sequence in which columns are visited.}
\item{\code{formulas}:}{A named list of formula's, or expressions that
can be converted into formula's by \code{as.formula}. List elements
correspond to blocks. The block to which the list element applies is
identified by its name, so list names must correspond to block names.}
\item{\code{post}:}{A vector of strings of length \code{length(blocks)}
with commands for post-processing.}
\item{\code{blots}:}{"Block dots". The \code{blots} argument to the \code{mice()}
function.}
\item{\code{ignore}:}{A logical vector of length \code{nrow(data)} indicating
the rows in \code{data} used to build the imputation model. (new in \code{mice 3.12.0})}
\item{\code{seed}:}{The seed value of the solution.}
\item{\code{iteration}:}{Last Gibbs sampling iteration number.}
\item{\code{lastSeedValue}:}{The most recent seed value.}
\item{\code{chainMean}:}{A list of \code{m} components. Each
component is a \code{length(visitSequence)} by \code{maxit} matrix
containing the mean of the generated multiple imputations.
The array can be used for monitoring convergence.
Note that observed data are not present in this mean.}
\item{\code{chainVar}:}{A list with similar structure of \code{chainMean},
containing the covariances of the imputed values.}
\item{\code{loggedEvents}:}{A \code{data.frame} with five columns
containing warnings, corrective actions, and other inside info.}
\item{\code{version}:}{Version number of \code{mice} package that
created the object.}
\item{\code{date}:}{Date at which the object was created.}
}
}
\references{
van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
}
\seealso{
\code{\link{mice}}, \code{\link[=mira-class]{mira}},
\code{\link{mipo}}
}
\author{
Stef van Buuren, Karin Groothuis-Oudshoorn, 2000
}
\keyword{classes}
mice/man/nelsonaalen.Rd 0000644 0001762 0000144 00000003534 14436640005 014537 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/nelsonaalen.R
\name{nelsonaalen}
\alias{nelsonaalen}
\alias{hazard}
\title{Cumulative hazard rate or Nelson-Aalen estimator}
\usage{
nelsonaalen(data, timevar, statusvar)
}
\arguments{
\item{data}{A data frame containing the data.}
\item{timevar}{The name of the time variable in \code{data}.}
\item{statusvar}{The name of the event variable, e.g. death in \code{data}.}
}
\value{
A vector with \code{nrow(data)} elements containing the Nelson-Aalen
estimates of the cumulative hazard function.
}
\description{
Calculates the cumulative hazard rate (Nelson-Aalen estimator)
}
\details{
This function is useful for imputing variables that depend on survival time.
White and Royston (2009) suggested using the cumulative hazard to the
survival time H0(T) rather than T or log(T) as a predictor in imputation
models. See section 7.1 of Van Buuren (2012) for an example.
}
\examples{
require(MASS)
leuk$status <- 1 ## no censoring occurs in leuk data (MASS)
ch <- nelsonaalen(leuk, time, status)
plot(x = leuk$time, y = ch, ylab = "Cumulative hazard", xlab = "Time")
### See example on http://www.engineeredsoftware.com/lmar/pe_cum_hazard_function.htm
time <- c(43, 67, 92, 94, 149, rep(149, 7))
status <- c(rep(1, 5), rep(0, 7))
eng <- data.frame(time, status)
ch <- nelsonaalen(eng, time, status)
plot(x = time, y = ch, ylab = "Cumulative hazard", xlab = "Time")
}
\references{
White, I. R., Royston, P. (2009). Imputing missing covariate
values for the Cox model. \emph{Statistics in Medicine}, \emph{28}(15),
1982-1998.
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-toomany.html#a-further-improvement-survival-as-predictor-variable}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\author{
Stef van Buuren, 2012
}
\keyword{misc}
mice/man/mdc.Rd 0000644 0001762 0000144 00000005223 14330031606 012772 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mdc.R
\name{mdc}
\alias{mdc}
\title{Graphical parameter for missing data plots}
\usage{
mdc(
r = "observed",
s = "symbol",
transparent = TRUE,
cso = grDevices::hcl(240, 100, 40, 0.7),
csi = grDevices::hcl(0, 100, 40, 0.7),
csc = "gray50",
clo = grDevices::hcl(240, 100, 40, 0.8),
cli = grDevices::hcl(0, 100, 40, 0.8),
clc = "gray50"
)
}
\arguments{
\item{r}{A numerical or character vector. The numbers 1-6 request colors as
follows: 1=\code{cso}, 2=\code{csi}, 3=\code{csc}, 4=\code{clo}, 5=\code{cli}
and 6=\code{clc}. Alternatively, \code{r} may contain the strings
' \code{observed}', '\code{missing}', or '\code{both}', or abbreviations
thereof.}
\item{s}{A character vector containing the strings '\code{symbol}' or
' \code{line}', or abbreviations thereof.}
\item{transparent}{A logical indicating whether alpha-transparency is
allowed. The default is \code{TRUE}.}
\item{cso}{The symbol color for the observed data. The default is a
transparent blue.}
\item{csi}{The symbol color for the missing or imputed data. The default is a
transparent red.}
\item{csc}{The symbol color for the combined observed and imputed data. The
default is a grey color.}
\item{clo}{The line color for the observed data. The default is a slightly
darker transparent blue.}
\item{cli}{The line color for the missing or imputed data. The default is a
slightly darker transparent red.}
\item{clc}{The line color for the combined observed and imputed data. The
default is a grey color.}
}
\value{
\code{mdc()} returns a vector containing color definitions. The length
of the output vector is calculate from the length of \code{r} and \code{s}.
Elements of the input vectors are repeated if needed.
}
\description{
\code{mdc} returns colors used to distinguish observed, missing and combined
data in plotting. \code{mice.theme} return a partial list of named objects
that can be used as a theme in \code{stripplot}, \code{bwplot},
\code{densityplot} and \code{xyplot}.
}
\details{
This function eases consistent use of colors in plots. The default follows
the Abayomi convention, which uses blue for observed data, red for missing or
imputed data, and black for combined data.
}
\examples{
# all six colors
mdc(1:6)
# lines color for observed and missing data
mdc(c("obs", "mis"), "lin")
}
\references{
Sarkar, Deepayan (2008) \emph{Lattice: Multivariate Data
Visualization with R}, Springer.
}
\seealso{
\code{\link{hcl}}, \code{\link{rgb}},
\code{\link{xyplot.mids}}, \code{\link[lattice:xyplot]{xyplot}},
\code{\link[lattice:trellis.par.get]{trellis.par.set}}
}
\author{
Stef van Buuren, sept 2012.
}
\keyword{hplot}
mice/man/pool.Rd 0000644 0001762 0000144 00000016524 14334522175 013220 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/pool.R
\name{pool}
\alias{pool}
\alias{pool.syn}
\title{Combine estimates by pooling rules}
\usage{
pool(object, dfcom = NULL, rule = NULL, custom.t = NULL)
pool.syn(object, dfcom = NULL, rule = "reiter2003")
}
\arguments{
\item{object}{An object of class \code{mira} (produced by \code{with.mids()}
or \code{as.mira()}), or a \code{list} with model fits.}
\item{dfcom}{A positive number representing the degrees of freedom in the
complete-data analysis. Normally, this would be the number of independent
observation minus the number of fitted parameters. The default
(\code{dfcom = NULL}) extract this information in the following
order: 1) the component
\code{residual.df} returned by \code{glance()} if a \code{glance()}
function is found, 2) the result of \code{df.residual(} applied to
the first fitted model, and 3) as \code{999999}.
In the last case, the warning \code{"Large sample assumed"} is printed.
If the degrees of freedom is incorrect, specify the appropriate value
manually.}
\item{rule}{A string indicating the pooling rule. Currently supported are
\code{"rubin1987"} (default, for missing data) and \code{"reiter2003"}
(for synthetic data created from a complete data set).}
\item{custom.t}{A custom character string to be parsed as a calculation rule
for the total variance \code{t}. The custom rule can use the other calculated
pooling statistics where the dimensions must come from \code{.data$}. The
default \code{t} calculation would have the form
\code{".data$ubar + (1 + 1 / .data$m) * .data$b"}.
See examples for an example.}
}
\value{
An object of class \code{mipo}, which stands for 'multiple imputation
pooled outcome'.
For rule \code{"reiter2003"} values for \code{lambda} and \code{fmi} are
set to `NA`, as these statistics do not apply for data synthesised from
fully observed data.
}
\description{
The \code{pool()} function combines the estimates from \code{m}
repeated complete data analyses. The typical sequence of steps to
perform a multiple imputation analysis is:
\enumerate{
\item Impute the missing data by the \code{mice()} function, resulting in
a multiple imputed data set (class \code{mids});
\item Fit the model of interest (scientific model) on each imputed data set
by the \code{with()} function, resulting an object of class \code{mira};
\item Pool the estimates from each model into a single set of estimates
and standard errors, resulting in an object of class \code{mipo};
\item Optionally, compare pooled estimates from different scientific models
by the \code{D1()} or \code{D3()} functions.
}
A common error is to reverse steps 2 and 3, i.e., to pool the
multiply-imputed data instead of the estimates. Doing so may severely bias
the estimates of scientific interest and yield incorrect statistical
intervals and p-values. The \code{pool()} function will detect
this case.
}
\details{
The \code{pool()} function averages the estimates of the complete
data model, computes the total variance over the repeated analyses
by Rubin's rules (Rubin, 1987, p. 76), and computes the following
diagnostic statistics per estimate:
\enumerate{
\item Relative increase in variance due to nonresponse {\code{r}};
\item Residual degrees of freedom for hypothesis testing {\code{df}};
\item Proportion of total variance due to missingness {\code{lambda}};
\item Fraction of missing information {\code{fmi}}.
}
The degrees of freedom calculation for the pooled estimates uses the
Barnard-Rubin adjustment for small samples (Barnard and Rubin, 1999).
The \code{pool.syn()} function combines estimates by Reiter's partially
synthetic data pooling rules (Reiter, 2003). This combination rule
assumes that the data that is synthesised is completely observed.
Pooling differs from Rubin's method in the calculation of the total
variance and the degrees of freedom.
Pooling requires the following input from each fitted model:
\enumerate{
\item the estimates of the model;
\item the standard error of each estimate;
\item the residual degrees of freedom of the model.
}
The \code{pool()} and \code{pool.syn()} functions rely on the
\code{broom::tidy} and \code{broom::glance} for extracting these
parameters.
Since \code{mice 3.0+}, the \code{broom}
package takes care of filtering out the relevant parts of the
complete-data analysis. It may happen that you'll see the messages
like \code{Error: No tidy method for objects of class ...} or
\code{Error: No glance method for objects of class ...}. The message
means that your complete-data method used in \code{with(imp, ...)} has
no \code{tidy} or \code{glance} method defined in the \code{broom} package.
The \code{broom.mixed} package contains \code{tidy} and \code{glance} methods
for mixed models. If you are using a mixed model, first run
\code{library(broom.mixed)} before calling \code{pool()}.
If no \code{tidy} or \code{glance} methods are defined for your analysis
tabulate the \code{m} parameter estimates and their variance
estimates (the square of the standard errors) from the \code{m} fitted
models stored in \code{fit$analyses}. For each parameter, run
\code{\link{pool.scalar}} to obtain the pooled parameters estimate, its variance, the
degrees of freedom, the relative increase in variance and the fraction of missing
information.
An alternative is to write your own \code{glance()} and \code{tidy()}
methods and add these to \code{broom} according to the specifications
given in \url{https://broom.tidymodels.org}.
In versions prior to \code{mice 3.0} pooling required that
\code{coef()} and \code{vcov()} methods were available for fitted
objects. \emph{This feature is no longer supported}. The reason is that
\code{vcov()} methods are inconsistent across packages, leading to
buggy behaviour of the \code{pool()} function.
Since \code{mice 3.13.2} function \code{pool()} uses the robust
the standard error estimate for pooling when it can extract
\code{robust.se} from the \code{tidy()} object.
}
\examples{
# impute missing data, analyse and pool using the classic MICE workflow
imp <- mice(nhanes, maxit = 2, m = 2)
fit <- with(data = imp, exp = lm(bmi ~ hyp + chl))
summary(pool(fit))
# generate fully synthetic data, analyse and pool
imp <- mice(cars,
maxit = 2, m = 2,
where = matrix(TRUE, nrow(cars), ncol(cars))
)
fit <- with(data = imp, exp = lm(speed ~ dist))
summary(pool.syn(fit))
# use a custom pooling rule for the total variance about the estimate
# e.g. use t = b + b/m instead of t = ubar + b + b/m
imp <- mice(nhanes, maxit = 2, m = 2)
fit <- with(data = imp, exp = lm(bmi ~ hyp + chl))
pool(fit, custom.t = ".data$b + .data$b / .data$m")
}
\references{
Barnard, J. and Rubin, D.B. (1999). Small sample degrees of
freedom with multiple imputation. \emph{Biometrika}, 86, 948-955.
Rubin, D.B. (1987). \emph{Multiple Imputation for Nonresponse in Surveys}.
New York: John Wiley and Sons.
Reiter, J.P. (2003). Inference for Partially Synthetic,
Public Use Microdata Sets. \emph{Survey Methodology}, \bold{29}, 181-189.
van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}: Multivariate
Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
}
\seealso{
\code{\link{with.mids}}, \code{\link{as.mira}}, \code{\link{pool.scalar}},
\code{\link[broom:reexports]{glance}}, \code{\link[broom:reexports]{tidy}}
\url{https://github.com/amices/mice/issues/142},
\url{https://github.com/amices/mice/issues/274}
}
mice/man/nic.Rd 0000644 0001762 0000144 00000001242 14330031606 012775 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/ncc.R
\name{nic}
\alias{nic}
\title{Number of incomplete cases}
\usage{
nic(x)
}
\arguments{
\item{x}{An \code{R} object. Currently supported are methods for the
following classes: \code{mids}, \code{data.frame} and \code{matrix}. Also,
\code{x} can be a vector.}
}
\value{
Number of elements in \code{x} with incomplete data.
}
\description{
Calculates the number of incomplete cases.
}
\examples{
nic(nhanes) # the remaining 12 rows
nic(nhanes[, c("bmi", "hyp")]) # number of cases with incomplete bmi and hyp
}
\seealso{
\code{\link{ncc}}, \code{\link{cci}}
}
\author{
Stef van Buuren, 2017
}
mice/man/glance.mipo.Rd 0000644 0001762 0000144 00000001214 13666252233 014433 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/tidiers.R
\name{glance.mipo}
\alias{glance.mipo}
\title{Glance method to extract information from a `mipo` object}
\usage{
\method{glance}{mipo}(x, ...)
}
\arguments{
\item{x}{An object with multiply-imputed models from `mice` (class: `mipo`)}
\item{...}{extra arguments (not used)}
}
\value{
a dataframe with one row and the following columns:
\itemize{
\item nimp
\item nobs
}
}
\description{
Glance method to extract information from a `mipo` object
}
\note{
If x contains `lm` models, R2 and Adj.R2 are included in the output
}
\concept{tidiers}
\keyword{internal}
mice/man/ic.Rd 0000644 0001762 0000144 00000001534 14330031606 012623 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/cc.R
\name{ic}
\alias{ic}
\title{Select incomplete cases}
\usage{
ic(x)
}
\arguments{
\item{x}{An \code{R} object. Methods are available for classes
\code{mids}, \code{data.frame} and \code{matrix}. Also, \code{x}
could be a vector.}
}
\value{
A \code{vector}, \code{matrix} or \code{data.frame} containing the data of the complete cases.
}
\description{
Extracts incomplete cases from a data set.
The companion function for selecting the complete cases is \code{\link{cc}}.
}
\examples{
ic(nhanes) # get the 12 rows with incomplete cases
ic(nhanes[1:10, ]) # incomplete cases within the first ten rows
ic(nhanes[, c("bmi", "hyp")]) # restrict extraction to variables bmi and hyp
}
\seealso{
\code{\link{cc}}, \code{\link{ici}}
}
\author{
Stef van Buuren, 2017.
}
\keyword{univar}
mice/man/as.mira.Rd 0000644 0001762 0000144 00000001070 14330031606 013555 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/as.R
\name{as.mira}
\alias{as.mira}
\title{Create a \code{mira} object from repeated analyses}
\usage{
as.mira(fitlist)
}
\arguments{
\item{fitlist}{A list containing $m$ fitted analysis objects}
}
\value{
An S3 object of class \code{mira}.
}
\description{
The \code{as.mira()} function takes the results of repeated
complete-data analysis stored as a list, and turns it
into a \code{mira} object that can be pooled.
}
\seealso{
\code{\link[=mira-class]{mira}}
}
\author{
Stef van Buuren
}
mice/man/selfreport.Rd 0000644 0001762 0000144 00000007162 14330031606 014420 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/selfreport.R
\docType{data}
\name{selfreport}
\alias{selfreport}
\alias{mgg}
\title{Self-reported and measured BMI}
\format{
A data frame with 2060 rows and 15 variables:
\describe{
\item{src}{Study, either \code{krul} or \code{mgg} (factor)}
\item{id}{Person identification number}
\item{pop}{Population, all \code{NL} (factor)}
\item{age}{Age of respondent in years}
\item{sex}{Sex of respondent (factor)}
\item{hm}{Height measured (cm)}
\item{wm}{Weight measured (kg)}
\item{hr}{Height reported (cm)}
\item{wr}{Weight reported (kg)}
\item{prg}{Pregnancy (factor), all \code{Not pregnant}}
\item{edu}{Educational level (factor)}
\item{etn}{Ethnicity (factor)}
\item{web}{Obtained through web survey (factor)}
\item{bm}{BMI measured (kg/m2)}
\item{br}{BMI reported (kg/m2)}
}
}
\source{
Krul, A., Daanen, H. A. M., Choi, H. (2010). Self-reported and
measured weight, height and body mass index (BMI) in Italy, The Netherlands
and North America. \emph{European Journal of Public Health}, \emph{21}(4),
414-419.
Van Keulen, H.M.,, Chorus, A.M.J., Verheijden, M.W. (2011). \emph{Monitor
Convenant Gezond Gewicht Nulmeting (determinanten van) beweeg- en eetgedrag
van kinderen (4-11 jaar), jongeren (12-17 jaar) en volwassenen (18+ jaar)}.
TNO/LS 2011.016. Leiden: TNO.
Van der Klauw, M., Van Keulen, H.M., Verheijden, M.W. (2011). \emph{Monitor
Convenant Gezond Gewicht Beweeg- en eetgedrag van kinderen (4-11 jaar),
jongeren (12-17 jaar) en volwassenen (18+ jaar) in 2010 en 2011.} TNO/LS
2011.055. Leiden: TNO. (in Dutch)
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-prevalence.html#sec:srcdata}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\description{
Dataset containing height and weight data (measured, self-reported) from two
studies.
}
\details{
This dataset combines two datasets: \code{krul} data (Krul, 2010) (1257
persons) and the \code{mgg} data (Van Keulen 2011; Van der Klauw 2011) (803
persons). The \code{krul} dataset contains height and weight (both measures
and self-reported) from 1257 Dutch adults, whereas the \code{mgg} dataset
contains self-reported height and weight for 803 Dutch adults. Section 7.3 in
Van Buuren (2012) shows how the missing measured data can be imputed in the
\code{mgg} data, so corrected prevalence estimates can be calculated.
}
\examples{
md.pattern(selfreport[, c("age", "sex", "hm", "hr", "wm", "wr")])
### FIMD Section 7.3.5 Application
bmi <- function(h, w) {
return(w / (h / 100)^2)
}
init <- mice(selfreport, maxit = 0)
meth <- init$meth
meth["bm"] <- "~bmi(hm,wm)"
pred <- init$pred
pred[, c("src", "id", "web", "bm", "br")] <- 0
imp <- mice(selfreport, pred = pred, meth = meth, seed = 66573, maxit = 2, m = 1)
## imp <- mice(selfreport, pred=pred, meth=meth, seed=66573, maxit=20, m=10)
### Like FIMD Figure 7.6
cd <- complete(imp, 1)
xy <- xy.coords(cd$bm, cd$br - cd$bm)
plot(xy,
col = mdc(2), xlab = "Measured BMI", ylab = "Reported - Measured BMI",
xlim = c(17, 45), ylim = c(-5, 5), type = "n", lwd = 0.7
)
polygon(x = c(30, 20, 30), y = c(0, 10, 10), col = "grey95", border = NA)
polygon(x = c(30, 40, 30), y = c(0, -10, -10), col = "grey95", border = NA)
abline(0, 0, lty = 2, lwd = 0.7)
idx <- cd$src == "krul"
xyc <- xy
xyc$x <- xy$x[idx]
xyc$y <- xy$y[idx]
xys <- xy
xys$x <- xy$x[!idx]
xys$y <- xy$y[!idx]
points(xyc, col = mdc(1), cex = 0.7)
points(xys, col = mdc(2), cex = 0.7)
lines(lowess(xyc), col = mdc(4), lwd = 2)
lines(lowess(xys), col = mdc(5), lwd = 2)
text(1:4, x = c(40, 28, 20, 32), y = c(4, 4, -4, -4), cex = 3)
box(lwd = 1)
}
\keyword{datasets}
mice/man/mice.impute.2lonly.norm.Rd 0000644 0001762 0000144 00000012662 14330031606 016643 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.2lonly.norm.R
\name{mice.impute.2lonly.norm}
\alias{mice.impute.2lonly.norm}
\alias{2lonly.norm}
\title{Imputation at level 2 by Bayesian linear regression}
\usage{
mice.impute.2lonly.norm(y, ry, x, type, wy = NULL, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{type}{Group identifier must be specified by '-2'. Predictors must be
specified by '1'.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{...}{Other named arguments.}
}
\value{
A vector of length \code{nmis} with imputations.
}
\description{
Imputes univariate missing data at level 2 using Bayesian linear regression
analysis. Variables are level 1 are aggregated at level 2. The group
identifier at level 2 must be indicated by \code{type = -2} in the
\code{predictorMatrix}.
}
\details{
This function allows in combination with \code{\link{mice.impute.2l.pan}}
switching regression imputation between level 1 and level 2 as described in
Yucel (2008) or Gelman and Hill (2007, p. 541).
The function checks for partial missing level-2 data. Level-2 data
are assumed to be constant within the same cluster. If one or more
entries are missing, then the procedure aborts with an error
message that identifies the cluster with incomplete level-2 data.
In such cases, one may first fill in the cluster mean (or mode) by
the \code{2lonly.mean} method to remove inconsistencies.
}
\note{
For a more general approach, see
\code{miceadds::mice.impute.2lonly.function()}.
}
\examples{
# simulate some data
# x,y ... level 1 variables
# v,w ... level 2 variables
G <- 250 # number of groups
n <- 20 # number of persons
beta <- .3 # regression coefficient
rho <- .30 # residual intraclass correlation
rho.miss <- .10 # correlation with missing response
missrate <- .50 # missing proportion
y1 <- rep(rnorm(G, sd = sqrt(rho)), each = n) + rnorm(G * n, sd = sqrt(1 - rho))
w <- rep(round(rnorm(G), 2), each = n)
v <- rep(round(runif(G, 0, 3)), each = n)
x <- rnorm(G * n)
y <- y1 + beta * x + .2 * w + .1 * v
dfr0 <- dfr <- data.frame("group" = rep(1:G, each = n), "x" = x, "y" = y, "w" = w, "v" = v)
dfr[rho.miss * x + rnorm(G * n, sd = sqrt(1 - rho.miss)) < qnorm(missrate), "y"] <- NA
dfr[rep(rnorm(G), each = n) < qnorm(missrate), "w"] <- NA
dfr[rep(rnorm(G), each = n) < qnorm(missrate), "v"] <- NA
# empty mice imputation
imp0 <- mice(as.matrix(dfr), maxit = 0)
predM <- imp0$predictorMatrix
impM <- imp0$method
# multilevel imputation
predM1 <- predM
predM1[c("w", "y", "v"), "group"] <- -2
predM1["y", "x"] <- 1 # fixed x effects imputation
impM1 <- impM
impM1[c("y", "w", "v")] <- c("2l.pan", "2lonly.norm", "2lonly.pmm")
# y ... imputation using pan
# w ... imputation at level 2 using norm
# v ... imputation at level 2 using pmm
imp1 <- mice(as.matrix(dfr),
m = 1, predictorMatrix = predM1,
method = impM1, maxit = 1, paniter = 500
)
# Demonstration that 2lonly.norm aborts for partial missing data.
# Better use 2lonly.mean for repair.
data <- data.frame(
patid = rep(1:4, each = 5),
sex = rep(c(1, 2, 1, 2), each = 5),
crp = c(
68, 78, 93, NA, 143,
5, 7, 9, 13, NA,
97, NA, 56, 52, 34,
22, 30, NA, NA, 45
)
)
pred <- make.predictorMatrix(data)
pred[, "patid"] <- -2
# only missing value (out of five) for patid == 1
data[3, "sex"] <- NA
\dontrun{
# The following fails because 2lonly.norm found partially missing
# level-2 data
# imp <- mice(data, method = c("", "2lonly.norm", "2l.pan"),
# predictorMatrix = pred, maxit = 1, m = 2)
# > iter imp variable
# > 1 1 sex crpError in .imputation.level2(y = y, ... :
# > Method 2lonly.norm found the following clusters with partially missing
# > level-2 data: 1
# > Method 2lonly.mean can fix such inconsistencies.
}
# In contrast, if all sex values are missing for patid == 1, it runs fine,
# except on r-patched-solaris-x86. I used dontrun to evade CRAN errors.
\dontrun{
data[1:5, "sex"] <- NA
imp <- mice(data,
method = c("", "2lonly.norm", "2l.pan"),
predictorMatrix = pred, maxit = 1, m = 2
)
}
}
\references{
Gelman, A. and Hill, J. (2007). \emph{Data analysis using
regression and multilevel/hierarchical models}. Cambridge, Cambridge
University Press.
Yucel, RM (2008). Multiple imputation inference for multivariate multilevel
continuous data with ignorable non-response. \emph{Philosophical
Transactions of the Royal Society A}, \bold{366}, 2389-2404.
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-level2pred.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\seealso{
\code{\link{mice.impute.norm}},
\code{\link{mice.impute.2lonly.pmm}}, \code{\link{mice.impute.2l.pan}},
\code{\link{mice.impute.2lonly.mean}}
Other univariate-2lonly:
\code{\link{mice.impute.2lonly.mean}()},
\code{\link{mice.impute.2lonly.pmm}()}
}
\author{
Alexander Robitzsch (IPN - Leibniz Institute for Science and
Mathematics Education, Kiel, Germany), \email{robitzsch@ipn.uni-kiel.de}
}
\concept{univariate-2lonly}
mice/man/glm.mids.Rd 0000644 0001762 0000144 00000003101 14330031606 013732 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/lm.R
\name{glm.mids}
\alias{glm.mids}
\title{Generalized linear model for \code{mids} object}
\usage{
glm.mids(formula, family = gaussian, data, ...)
}
\arguments{
\item{formula}{a formula expression as for other regression models, of the
form response ~ predictors. See the documentation of \code{\link{lm}} and
\code{\link{formula}} for details.}
\item{family}{The family of the glm model}
\item{data}{An object of type \code{mids}, which stands for 'multiply imputed
data set', typically created by function \code{mice()}.}
\item{\dots}{Additional parameters passed to \code{\link{glm}}.}
}
\value{
An objects of class \code{mira}, which stands for 'multiply imputed
repeated analysis'. This object contains \code{data$m} distinct
\code{glm.objects}, plus some descriptive information.
}
\description{
Applies \code{glm()} to a multiply imputed data set
}
\details{
This function is included for backward compatibility with V1.0. The function
is superseded by \code{\link{with.mids}}.
}
\examples{
imp <- mice(nhanes)
# logistic regression on the imputed data
fit <- glm.mids((hyp == 2) ~ bmi + chl, data = imp, family = binomial)
fit
}
\references{
Van Buuren, S., Groothuis-Oudshoorn, C.G.M. (2000)
\emph{Multivariate Imputation by Chained Equations: MICE V1.0 User's manual.}
Leiden: TNO Quality of Life.
}
\seealso{
\code{\link{with.mids}}, \code{\link{glm}}, \code{\link[=mids-class]{mids}},
\code{\link[=mira-class]{mira}}
}
\author{
Stef van Buuren, Karin Groothuis-Oudshoorn, 2000
}
\keyword{multivariate}
mice/man/is.mads.Rd 0000644 0001762 0000144 00000000515 13666252075 013604 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/is.R
\name{is.mads}
\alias{is.mads}
\title{Check for \code{mads} object}
\usage{
is.mads(x)
}
\arguments{
\item{x}{An object}
}
\value{
A logical indicating whether \code{x} is an object of class \code{mads}
}
\description{
Check for \code{mads} object
}
mice/man/extend.formulas.Rd 0000644 0001762 0000144 00000005254 14330031606 015351 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/formula.R
\name{extend.formulas}
\alias{extend.formulas}
\title{Extends formula's with predictor matrix settings}
\usage{
extend.formulas(
formulas,
data,
blocks,
predictorMatrix = NULL,
auxiliary = TRUE,
include.intercept = FALSE,
...
)
}
\arguments{
\item{formulas}{A named list of formula's, or expressions that
can be converted into formula's by \code{as.formula}. List elements
correspond to blocks. The block to which the list element applies is
identified by its name, so list names must correspond to block names.
The \code{formulas} argument is an alternative to the
\code{predictorMatrix} argument that allows for more flexibility in
specifying imputation models, e.g., for specifying interaction terms.}
\item{data}{A data frame or a matrix containing the incomplete data. Missing
values are coded as \code{NA}.}
\item{blocks}{List of vectors with variable names per block. List elements
may be named to identify blocks. Variables within a block are
imputed by a multivariate imputation method
(see \code{method} argument). By default each variable is placed
into its own block, which is effectively
fully conditional specification (FCS) by univariate models
(variable-by-variable imputation). Only variables whose names appear in
\code{blocks} are imputed. The relevant columns in the \code{where}
matrix are set to \code{FALSE} of variables that are not block members.
A variable may appear in multiple blocks. In that case, it is
effectively re-imputed each time that it is visited.}
\item{predictorMatrix}{A numeric matrix of \code{length(blocks)} rows
and \code{ncol(data)} columns, containing 0/1 data specifying
the set of predictors to be used for each target column.
Each row corresponds to a variable block, i.e., a set of variables
to be imputed. A value of \code{1} means that the column
variable is used as a predictor for the target block (in the rows).
By default, the \code{predictorMatrix} is a square matrix of \code{ncol(data)}
rows and columns with all 1's, except for the diagonal.
Note: For two-level imputation models (which have \code{"2l"} in their names)
other codes (e.g, \code{2} or \code{-2}) are also allowed.}
\item{auxiliary}{A logical that indicates whether the variables
listed in \code{predictors} should be added to the formula as main
effects. The default is \code{TRUE}.}
\item{include.intercept}{A logical that indicated whether the intercept
should be included in the result.}
\item{...}{Named arguments that are passed down to the univariate imputation
functions.}
}
\value{
A list of formula's
}
\description{
Extends formula's with predictor matrix settings
}
\keyword{internal}
mice/man/md.pairs.Rd 0000644 0001762 0000144 00000002750 14330031647 013753 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/md.pairs.R
\name{md.pairs}
\alias{md.pairs}
\title{Missing data pattern by variable pairs}
\usage{
md.pairs(data)
}
\arguments{
\item{data}{A data frame or a matrix containing the incomplete data. Missing
values are coded as \code{NA}.}
}
\value{
A list of four components named \code{rr}, \code{rm}, \code{mr} and
\code{mm}. Each component is square numerical matrix containing the number
observations within four missing data pattern.
}
\description{
Number of observations per variable pair.
}
\details{
The four components in the output value is have the following interpretation:
\describe{ \item{list('rr')}{response-response, both variables are observed}
\item{list('rm')}{response-missing, row observed, column missing}
\item{list('mr')}{missing -response, row missing, column observed}
\item{list('mm')}{missing -missing, both variables are missing} }
}
\examples{
pat <- md.pairs(nhanes)
pat
# show that these four matrices decompose the total sample size
# for each pair
pat$rr + pat$rm + pat$mr + pat$mm
# percentage of usable cases to impute row variable from column variable
round(100 * pat$mr / (pat$mr + pat$mm))
}
\references{
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
}
\author{
Stef van Buuren, Karin Groothuis-Oudshoorn, 2009
}
\keyword{univar}
mice/man/reexports.Rd 0000644 0001762 0000144 00000001001 14330031606 014250 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/filter.R, R/tidiers.R
\docType{import}
\name{reexports}
\alias{reexports}
\alias{filter}
\alias{tidy}
\alias{glance}
\title{Objects exported from other packages}
\keyword{internal}
\description{
These objects are imported from other packages. Follow the links
below to see their documentation.
\describe{
\item{dplyr}{\code{\link[dplyr]{filter}}}
\item{generics}{\code{\link[generics]{glance}}, \code{\link[generics]{tidy}}}
}}
mice/man/pool.r.squared.Rd 0000644 0001762 0000144 00000003655 14330031647 015117 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/pool.r.squared.R
\name{pool.r.squared}
\alias{pool.r.squared}
\title{Pools R^2 of m models fitted to multiply-imputed data}
\usage{
pool.r.squared(object, adjusted = FALSE)
}
\arguments{
\item{object}{An object of class 'mira' or 'mipo', produced by \code{lm.mids},
\code{with.mids}, or \code{pool} with \code{lm} as modeling function.}
\item{adjusted}{A logical value. If adjusted=TRUE then the adjusted R^2 is
calculated. The default value is FALSE.}
}
\value{
Returns a 1x4 table with components. Component \code{est} is the
pooled R^2 estimate. Component \code{lo95} is the 95 \% lower bound of the pooled R^2.
Component \code{hi95} is the 95 \% upper bound of the pooled R^2.
Component \code{fmi} is the fraction of missing information due to nonresponse.
}
\description{
The function pools the coefficients of determination R^2 or the adjusted
coefficients of determination (R^2_a) obtained with the \code{lm} modeling
function. For pooling it uses the Fisher \emph{z}-transformation.
}
\examples{
imp <- mice(nhanes, print = FALSE, seed = 16117)
fit <- with(imp, lm(chl ~ age + hyp + bmi))
# input: mira object
pool.r.squared(fit)
pool.r.squared(fit, adjusted = TRUE)
# input: mipo object
est <- pool(fit)
pool.r.squared(est)
pool.r.squared(est, adjusted = TRUE)
}
\references{
Harel, O (2009). The estimation of R^2 and adjusted R^2 in
incomplete data sets using multiple imputation, Journal of Applied Statistics,
36:1109-1118.
Rubin, D.B. (1987). Multiple Imputation for Nonresponse in Surveys. New
York: John Wiley and Sons.
van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}: Multivariate
Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
}
\seealso{
\code{\link{pool}},\code{\link{pool.scalar}}
}
\author{
Karin Groothuis-Oudshoorn and Stef van Buuren, 2009
}
\keyword{htest}
mice/man/bwplot.mids.Rd 0000644 0001762 0000144 00000017426 14330031647 014506 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/bwplot.R
\name{bwplot.mids}
\alias{bwplot.mids}
\alias{bwplot}
\title{Box-and-whisker plot of observed and imputed data}
\usage{
\method{bwplot}{mids}(
x,
data,
na.groups = NULL,
groups = NULL,
as.table = TRUE,
theme = mice.theme(),
mayreplicate = TRUE,
allow.multiple = TRUE,
outer = TRUE,
drop.unused.levels = lattice::lattice.getOption("drop.unused.levels"),
...,
subscripts = TRUE,
subset = TRUE
)
}
\arguments{
\item{x}{A \code{mids} object, typically created by \code{mice()} or
\code{mice.mids()}.}
\item{data}{Formula that selects the data to be plotted. This argument
follows the \pkg{lattice} rules for \emph{formulas}, describing the primary
variables (used for the per-panel display) and the optional conditioning
variables (which define the subsets plotted in different panels) to be used
in the plot.
The formula is evaluated on the complete data set in the \code{long} form.
Legal variable names for the formula include \code{names(x$data)} plus the
two administrative factors \code{.imp} and \code{.id}.
\bold{Extended formula interface:} The primary variable terms (both the LHS
\code{y} and RHS \code{x}) may consist of multiple terms separated by a
\sQuote{+} sign, e.g., \code{y1 + y2 ~ x | a * b}. This formula would be
taken to mean that the user wants to plot both \code{y1 ~ x | a * b} and
\code{y2 ~ x | a * b}, but with the \code{y1 ~ x} and \code{y2 ~ x} in
\emph{separate panels}. This behavior differs from standard \pkg{lattice}.
\emph{Only combine terms of the same type}, i.e. only factors or only
numerical variables. Mixing numerical and categorical data occasionally
produces odds labeling of vertical axis.
For convenience, in \code{stripplot()} and \code{bwplot} the formula
\code{y~.imp} may be abbreviated as \code{y}. This applies only to a single
\code{y}, and does not (yet) work for \code{y1+y2~.imp}.}
\item{na.groups}{An expression evaluating to a logical vector indicating
which two groups are distinguished (e.g. using different colors) in the
display. The environment in which this expression is evaluated in the
response indicator \code{is.na(x$data)}.
The default \code{na.group = NULL} contrasts the observed and missing data
in the LHS \code{y} variable of the display, i.e. groups created by
\code{is.na(y)}. The expression \code{y} creates the groups according to
\code{is.na(y)}. The expression \code{y1 & y2} creates groups by
\code{is.na(y1) & is.na(y2)}, and \code{y1 | y2} creates groups as
\code{is.na(y1) | is.na(y2)}, and so on.}
\item{groups}{This is the usual \code{groups} arguments in \pkg{lattice}. It
differs from \code{na.groups} because it evaluates in the completed data
\code{data.frame(complete(x, "long", inc=TRUE))} (as usual), whereas
\code{na.groups} evaluates in the response indicator. See
\code{\link{xyplot}} for more details. When both \code{na.groups} and
\code{groups} are specified, \code{na.groups} takes precedence, and
\code{groups} is ignored.}
\item{as.table}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{theme}{A named list containing the graphical parameters. The default
function \code{mice.theme} produces a short list of default colors, line
width, and so on. The extensive list may be obtained from
\code{trellis.par.get()}. Global graphical parameters like \code{col} or
\code{cex} in high-level calls are still honored, so first experiment with
the global parameters. Many setting consists of a pair. For example,
\code{mice.theme} defines two symbol colors. The first is for the observed
data, the second for the imputed data. The theme settings only exist during
the call, and do not affect the trellis graphical parameters.}
\item{mayreplicate}{A logical indicating whether color, line widths, and so
on, may be replicated. The graphical functions attempt to choose
"intelligent" graphical parameters. For example, the same color can be
replicated for different element, e.g. use all reds for the imputed data.
Replication may be switched off by setting the flag to \code{FALSE}, in order
to allow the user to gain full control.}
\item{allow.multiple}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{outer}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{drop.unused.levels}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{\dots}{Further arguments, usually not directly processed by the
high-level functions documented here, but instead passed on to other
functions.}
\item{subscripts}{See \code{\link[lattice:xyplot]{xyplot}}.}
\item{subset}{See \code{\link[lattice:xyplot]{xyplot}}.}
}
\value{
The high-level functions documented here, as well as other high-level
Lattice functions, return an object of class \code{"trellis"}. The
\code{\link[lattice:update.trellis]{update}} method can be used to
subsequently update components of the object, and the
\code{\link[lattice:print.trellis]{print}} method (usually called by default)
will plot it on an appropriate plotting device.
}
\description{
Plotting methods for imputed data using \pkg{lattice}. \code{bwplot}
produces box-and-whisker plots. The function
automatically separates the observed and imputed data. The
functions extend the usual features of \pkg{lattice}.
}
\details{
The argument \code{na.groups} may be used to specify (combinations of)
missingness in any of the variables. The argument \code{groups} can be used
to specify groups based on the variable values themselves. Only one of both
may be active at the same time. When both are specified, \code{na.groups}
takes precedence over \code{groups}.
Use the \code{subset} and \code{na.groups} together to plots parts of the
data. For example, select the first imputed data set by by
\code{subset=.imp==1}.
Graphical parameters like \code{col}, \code{pch} and \code{cex} can be
specified in the arguments list to alter the plotting symbols. If
\code{length(col)==2}, the color specification to define the observed and
missing groups. \code{col[1]} is the color of the 'observed' data,
\code{col[2]} is the color of the missing or imputed data. A convenient color
choice is \code{col=mdc(1:2)}, a transparent blue color for the observed
data, and a transparent red color for the imputed data. A good choice is
\code{col=mdc(1:2), pch=20, cex=1.5}. These choices can be set for the
duration of the session by running \code{mice.theme()}.
}
\note{
The first two arguments (\code{x} and \code{data}) are reversed
compared to the standard Trellis syntax implemented in \pkg{lattice}. This
reversal was necessary in order to benefit from automatic method dispatch.
In \pkg{mice} the argument \code{x} is always a \code{mids} object, whereas
in \pkg{lattice} the argument \code{x} is always a formula.
In \pkg{mice} the argument \code{data} is always a formula object, whereas in
\pkg{lattice} the argument \code{data} is usually a data frame.
All other arguments have identical interpretation.
}
\examples{
imp <- mice(boys, maxit = 1)
### box-and-whisker plot per imputation of all numerical variables
bwplot(imp)
### tv (testicular volume), conditional on region
bwplot(imp, tv ~ .imp | reg)
### same data, organized in a different way
bwplot(imp, tv ~ reg | .imp, theme = list())
}
\references{
Sarkar, Deepayan (2008) \emph{Lattice: Multivariate Data
Visualization with R}, Springer.
van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}: Multivariate
Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
}
\seealso{
\code{\link{mice}}, \code{\link{xyplot}}, \code{\link{densityplot}},
\code{\link{stripplot}}, \code{\link{lattice}} for an overview of the
package, as well as \code{\link[lattice:xyplot]{bwplot}},
\code{\link[lattice:panel.xyplot]{panel.bwplot}},
\code{\link[lattice:print.trellis]{print.trellis}},
\code{\link[lattice:trellis.par.get]{trellis.par.set}}
}
\author{
Stef van Buuren
}
\keyword{hplot}
mice/man/mice.impute.norm.Rd 0000644 0001762 0000144 00000006372 14330031647 015433 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.norm.R
\name{mice.impute.norm}
\alias{mice.impute.norm}
\alias{norm}
\title{Imputation by Bayesian linear regression}
\usage{
mice.impute.norm(y, ry, x, wy = NULL, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Calculates imputations for univariate missing data by Bayesian linear
regression, also known as the normal model.
}
\details{
Imputation of \code{y} by the normal model by the method defined by
Rubin (1987, p. 167). The procedure is as follows:
\enumerate{
\item{Calculate the cross-product matrix \eqn{S=X_{obs}'X_{obs}}.}
\item{Calculate \eqn{V = (S+{diag}(S)\kappa)^{-1}}, with some small ridge
parameter \eqn{\kappa}.}
\item{Calculate regression weights \eqn{\hat\beta = VX_{obs}'y_{obs}.}}
\item{Draw a random variable \eqn{\dot g \sim \chi^2_\nu} with \eqn{\nu=n_1 - q}.}
\item{Calculate \eqn{\dot\sigma^2 = (y_{obs} - X_{obs}\hat\beta)'(y_{obs} - X_{obs}\hat\beta)/\dot g.}}
\item{Draw \eqn{q} independent \eqn{N(0,1)} variates in vector \eqn{\dot z_1}.}
\item{Calculate \eqn{V^{1/2}} by Cholesky decomposition.}
\item{Calculate \eqn{\dot\beta = \hat\beta + \dot\sigma\dot z_1 V^{1/2}}.}
\item{Draw \eqn{n_0} independent \eqn{N(0,1)} variates in vector \eqn{\dot z_2}.}
\item{Calculate the \eqn{n_0} values \eqn{y_{imp} = X_{mis}\dot\beta + \dot z_2\dot\sigma}.}
}
Using \code{mice.impute.norm} for all columns emulates Schafer's NORM method (Schafer, 1997).
}
\references{
Rubin, D.B (1987). Multiple Imputation for Nonresponse in Surveys. New York: John Wiley & Sons.
Schafer, J.L. (1997). Analysis of incomplete multivariate data. London: Chapman & Hall.
}
\seealso{
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Stef van Buuren, Karin Groothuis-Oudshoorn
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/employee.Rd 0000644 0001762 0000144 00000002542 14330031606 014047 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/employee.R
\docType{data}
\name{employee}
\alias{employee}
\title{Employee selection data}
\format{
A data frame with 20 rows and 3 variables:
\describe{
\item{IQ}{candidate IQ score}
\item{wbeing}{candidate well-being score}
\item{jobperf}{candidate job performance score}
}
}
\source{
Enders (2010), Applied Missing Data Analysis, p. 218
}
\usage{
employee
}
\description{
A toy example from Craig Enders.
}
\details{
Enders describes these data as follows:
I designed these data to mimic an employee selection scenario in
which prospective employees complete an IQ test and a
psychological well-being questionnaire during their interview.
The company subsequently hires the applications that score in the
upper half of the IQ distribution, and a supervisor rates their
job performance following a 6-month probationary period.
Note that the job performance scores are missing at random (MAR)
(i.e. individuals in the lower half of the IQ distribution were
never hired, and thus have no performance rating). In addition,
I randomly deleted three of the well-being scores in order to
mimic a situation where the applicant's well-being questionnaire
is inadvertently lost.
A larger version of this data set in present as
\code{\link[miceadds:data.enders]{data.enders.employee}}.
}
\keyword{datasets}
mice/man/mice.impute.lasso.norm.Rd 0000644 0001762 0000144 00000006445 14330031647 016554 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.lasso.norm.R
\name{mice.impute.lasso.norm}
\alias{mice.impute.lasso.norm}
\alias{lasso.norm}
\title{Imputation by direct use of lasso linear regression}
\usage{
mice.impute.lasso.norm(y, ry, x, wy = NULL, nfolds = 10, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{nfolds}{The number of folds for the cross-validation of the lasso penalty.
The default is 10.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes univariate missing normal data using lasso linear regression with bootstrap.
}
\details{
The method consists of the following steps:
\enumerate{
\item For a given y variable under imputation, draw a bootstrap version y*
with replacement from the observed cases \code{y[ry]}, and stores in x* the
corresponding values from \code{x[ry, ]}.
\item Fit a regularised (lasso) linear regression with y* as the outcome,
and x* as predictors.
A vector of regression coefficients bhat is obtained.
All of these coefficients are considered random draws from the imputation model
parameters posterior distribution.
Same of these coefficients will be shrunken to 0.
\item Draw the imputed values from the predictive distribution defined by
the original (non-bootstrap) data, bhat, and estimated error variance.
}
The method is based on the Direct Use of Regularized Regression (DURR) proposed by
Zhao & Long (2016) and Deng et al (2016).
}
\references{
Deng, Y., Chang, C., Ido, M. S., & Long, Q. (2016). Multiple imputation for
general missing data patterns in the presence of high-dimensional data.
Scientific reports, 6(1), 1-10.
Zhao, Y., & Long, Q. (2016). Multiple imputation in the presence of
high-dimensional data. Statistical Methods in Medical Research, 25(5),
2021-2035.
}
\seealso{
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Edoardo Costantini, 2021
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/mice.impute.2lonly.pmm.Rd 0000644 0001762 0000144 00000011310 14330031606 016446 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.2lonly.pmm.R
\name{mice.impute.2lonly.pmm}
\alias{mice.impute.2lonly.pmm}
\alias{2lonly.pmm}
\title{Imputation at level 2 by predictive mean matching}
\usage{
mice.impute.2lonly.pmm(y, ry, x, type, wy = NULL, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{type}{Group identifier must be specified by '-2'. Predictors must be
specified by '1'.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{...}{Other named arguments.}
}
\value{
A vector of length \code{nmis} with imputations.
}
\description{
Imputes univariate missing data at level 2 using predictive mean matching.
Variables are level 1 are aggregated at level 2. The group identifier at
level 2 must be indicated by \code{type = -2} in the \code{predictorMatrix}.
}
\details{
This function allows in combination with \code{\link{mice.impute.2l.pan}}
switching regression imputation between level 1 and level 2 as described in
Yucel (2008) or Gelman and Hill (2007, p. 541).
The function checks for partial missing level-2 data. Level-2 data
are assumed to be constant within the same cluster. If one or more
entries are missing, then the procedure aborts with an error
message that identifies the cluster with incomplete level-2 data.
In such cases, one may first fill in the cluster mean (or mode) by
the \code{2lonly.mean} method to remove inconsistencies.
}
\note{
The extension to categorical variables transforms
a dependent factor variable by means of the \code{as.integer()}
function. This may make sense for categories that are
approximately ordered, but less so for pure nominal measures.
For a more general approach, see
\code{miceadds::mice.impute.2lonly.function()}.
}
\examples{
# simulate some data
# x,y ... level 1 variables
# v,w ... level 2 variables
G <- 250 # number of groups
n <- 20 # number of persons
beta <- .3 # regression coefficient
rho <- .30 # residual intraclass correlation
rho.miss <- .10 # correlation with missing response
missrate <- .50 # missing proportion
y1 <- rep(rnorm(G, sd = sqrt(rho)), each = n) + rnorm(G * n, sd = sqrt(1 - rho))
w <- rep(round(rnorm(G), 2), each = n)
v <- rep(round(runif(G, 0, 3)), each = n)
x <- rnorm(G * n)
y <- y1 + beta * x + .2 * w + .1 * v
dfr0 <- dfr <- data.frame("group" = rep(1:G, each = n), "x" = x, "y" = y, "w" = w, "v" = v)
dfr[rho.miss * x + rnorm(G * n, sd = sqrt(1 - rho.miss)) < qnorm(missrate), "y"] <- NA
dfr[rep(rnorm(G), each = n) < qnorm(missrate), "w"] <- NA
dfr[rep(rnorm(G), each = n) < qnorm(missrate), "v"] <- NA
# empty mice imputation
imp0 <- mice(as.matrix(dfr), maxit = 0)
predM <- imp0$predictorMatrix
impM <- imp0$method
# multilevel imputation
predM1 <- predM
predM1[c("w", "y", "v"), "group"] <- -2
predM1["y", "x"] <- 1 # fixed x effects imputation
impM1 <- impM
impM1[c("y", "w", "v")] <- c("2l.pan", "2lonly.norm", "2lonly.pmm")
# turn v into a categorical variable
dfr$v <- as.factor(dfr$v)
levels(dfr$v) <- LETTERS[1:4]
# y ... imputation using pan
# w ... imputation at level 2 using norm
# v ... imputation at level 2 using pmm
# skip imputation on solaris
is.solaris <- function() grepl("SunOS", Sys.info()["sysname"])
if (!is.solaris()) {
imp <- mice(dfr,
m = 1, predictorMatrix = predM1,
method = impM1, maxit = 1, paniter = 500
)
}
}
\references{
Gelman, A. and Hill, J. (2007). \emph{Data analysis using
regression and multilevel/hierarchical models}. Cambridge, Cambridge
University Press.
Yucel, RM (2008). Multiple imputation inference for multivariate multilevel
continuous data with ignorable non-response. \emph{Philosophical
Transactions of the Royal Society A}, \bold{366}, 2389-2404.
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-level2pred.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\seealso{
\code{\link{mice.impute.pmm}},
\code{\link{mice.impute.2lonly.norm}}, \code{\link{mice.impute.2l.pan}},
\code{\link{mice.impute.2lonly.mean}}
Other univariate-2lonly:
\code{\link{mice.impute.2lonly.mean}()},
\code{\link{mice.impute.2lonly.norm}()}
}
\author{
Alexander Robitzsch (IPN - Leibniz Institute for Science and
Mathematics Education, Kiel, Germany), \email{robitzsch@ipn.uni-kiel.de}
}
\concept{univariate-2lonly}
mice/man/mice.impute.logreg.Rd 0000644 0001762 0000144 00000006506 14330031647 015736 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.logreg.R
\name{mice.impute.logreg}
\alias{mice.impute.logreg}
\title{Imputation by logistic regression}
\usage{
mice.impute.logreg(y, ry, x, wy = NULL, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes univariate missing data using logistic regression.
}
\details{
Imputation for binary response variables by the Bayesian logistic regression
model (Rubin 1987, p. 169-170). The
Bayesian method consists of the following steps:
\enumerate{
\item Fit a logit, and find (bhat, V(bhat))
\item Draw BETA from N(bhat, V(bhat))
\item Compute predicted scores for m.d., i.e. logit-1(X BETA)
\item Compare the score to a random (0,1) deviate, and impute.
}
The method relies on the
standard \code{glm.fit} function. Warnings from \code{glm.fit} are
suppressed. Perfect prediction is handled by the data augmentation
method.
}
\references{
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
Brand, J.P.L. (1999). Development, Implementation and Evaluation of Multiple
Imputation Strategies for the Statistical Analysis of Incomplete Data Sets.
Ph.D. Thesis, TNO Prevention and Health/Erasmus University Rotterdam. ISBN
90-74479-08-1.
Venables, W.N. & Ripley, B.D. (1997). Modern applied statistics with S-Plus
(2nd ed). Springer, Berlin.
White, I., Daniel, R. and Royston, P (2010). Avoiding bias due to perfect
prediction in multiple imputation of incomplete categorical variables.
Computational Statistics and Data Analysis, 54:22672275.
}
\seealso{
\code{\link{mice}}, \code{\link{glm}}, \code{\link{glm.fit}}
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Stef van Buuren, Karin Groothuis-Oudshoorn
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/pmm.match.Rd 0000644 0001762 0000144 00000003373 14335404116 014124 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.pmm.R
\name{.pmm.match}
\alias{.pmm.match}
\title{Finds an imputed value from matches in the predictive metric (deprecated)}
\usage{
.pmm.match(z, yhat = yhat, y = y, donors = 5, ...)
}
\arguments{
\item{z}{A scalar containing the predicted value for the current case
to be imputed.}
\item{yhat}{A vector containing the predicted values for all cases with an observed
outcome.}
\item{y}{A vector of \code{length(yhat)} elements containing the observed outcome}
\item{donors}{The size of the donor pool among which a draw is made. The default is
\code{donors = 5}. Setting \code{donors = 1} always selects the closest match. Values
between 3 and 10 provide the best results. Note: This setting was changed from
3 to 5 in version 2.19, based on simulation work by Tim Morris (UCL).}
\item{\dots}{Other parameters (not used).}
}
\value{
A scalar containing the observed value of the selected donor.
}
\description{
This function finds matches among the observed data in the predictive
mean metric. It selects the \code{donors} closest matches, randomly
samples one of the donors, and returns the observed value of the
match.
}
\details{
This function is included for backward compatibility. It was
used up to \code{mice 2.21}. The current \code{mice.impute.pmm()}
function calls the faster \code{C} function \code{matcher} instead of
\code{.pmm.match()}.
}
\references{
Schenker N & Taylor JMG (1996) Partially parametric techniques
for multiple imputation. \emph{Computational Statistics and Data Analysis}, 22, 425-446.
Little RJA (1988) Missing-data adjustments in large surveys (with discussion).
\emph{Journal of Business Economics and Statistics}, 6, 287-301.
}
\author{
Stef van Buuren
}
mice/man/xyplot.mads.Rd 0000644 0001762 0000144 00000004376 14330031606 014521 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/xyplot.mads.R
\name{xyplot.mads}
\alias{xyplot.mads}
\title{Scatterplot of amputed and non-amputed data against weighted sum scores}
\usage{
\method{xyplot}{mads}(
x,
data,
which.pat = NULL,
standardized = TRUE,
layout = NULL,
colors = mdc(1:2),
...
)
}
\arguments{
\item{x}{A \code{mads} object, typically created by \code{\link{ampute}}.}
\item{data}{A string or vector of variable names that needs to be plotted. As
a default, all variables will be plotted.}
\item{which.pat}{A scalar or vector indicating which patterns need to be plotted.
As a default, all patterns are plotted.}
\item{standardized}{Logical. Whether the scatterplots need to be created
from standardized data or not. Default is TRUE.}
\item{layout}{A vector of two values indicating how the scatterplots of one
pattern should be divided over the plot. For example, \code{c(2, 3)} indicates
that the scatterplots of six variables need to be placed on 3 rows and 2 columns.
There are several defaults for different #variables. Note that for more than
9 variables, multiple plots will be created automatically.}
\item{colors}{A vector of two RGB values defining the colors of the non-amputed and
amputed data respectively. RGB values can be obtained with \code{\link{hcl}}.}
\item{\dots}{Not used, but for consistency with generic}
}
\value{
A list containing the scatterplots. Note that a new pattern
will always be shown in a new plot.
}
\description{
Plotting method to investigate relation between amputed data and the weighted sum
scores. Based on \code{\link{lattice}}. \code{xyplot} produces scatterplots.
The function plots the variables against the weighted sum scores. The function
automatically separates the amputed and non-amputed data to see the relation between
the amputation and the weighted sum scores.
}
\note{
The \code{mads} object contains all the information you need to
make any desired plots. Check \code{\link{mads-class}} or the vignette \emph{Multivariate
Amputation using Ampute} to understand the contents of class object \code{mads}.
}
\seealso{
\code{\link{ampute}}, \code{\link{bwplot}}, \code{\link{Lattice}} for
an overview of the package, \code{\link{mads-class}}
}
\author{
Rianne Schouten, 2016
}
mice/man/mira-class.Rd 0000644 0001762 0000144 00000004650 14330031647 014272 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mira.R
\docType{class}
\name{mira-class}
\alias{mira-class}
\alias{mira}
\title{Multiply imputed repeated analyses (\code{mira})}
\description{
The \code{mira} object is generated by the \code{with.mids()} function.
The \code{as.mira()}
function takes the results of repeated complete-data analysis stored as a
list, and turns it into a \code{mira} object that can be pooled.
}
\details{
In versions prior to \code{mice 3.0} pooling required only that
\code{coef()} and \code{vcov()} methods were available for fitted
objects. \emph{This feature is no longer supported}. The reason is that \code{vcov()}
methods are inconsistent across packages, leading to buggy behaviour
of the \code{pool()} function. Since \code{mice 3.0+}, the \code{broom}
package takes care of filtering out the relevant parts of the
complete-data analysis. It may happen that you'll see the messages
like \code{No method for tidying an S3 object of class ...} or
\code{Error: No glance method for objects of class ...}. The royal
way to solve this problem is to write your own \code{glance()} and \code{tidy()}
methods and add these to \code{broom} according to the specifications
given in \url{https://broom.tidymodels.org}.
#'The \code{mira} class of objects has methods for the
following generic functions: \code{print}, \code{summary}.
Many of the functions of the \code{mice} package do not use the
S4 class definitions, and instead rely on the S3 list equivalent
\code{oldClass(obj) <- "mira"}.
}
\section{Slots}{
\describe{
#' \item{\code{.Data}:}{Object of class \code{"list"} containing the
following slots:}
\item{\code{call}:}{The call that created the object.}
\item{\code{call1}:}{The call that created the \code{mids} object that was used
in \code{call}.}
\item{\code{nmis}:}{An array containing the number of missing observations per
column.}
\item{\code{analyses}:}{A list of \code{m} components containing the individual
fit objects from each of the \code{m} complete data analyses.}
}
}
\references{
van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
}
\seealso{
\code{\link{with.mids}}, \code{\link[=mids-class]{mids}}, \code{\link{mipo}}
}
\author{
Stef van Buuren, Karin Groothuis-Oudshoorn, 2000
}
\keyword{classes}
mice/man/fdgs.Rd 0000644 0001762 0000144 00000003650 14335404116 013161 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/fdgs.R
\docType{data}
\name{fdgs}
\alias{fdgs}
\title{Fifth Dutch growth study 2009}
\format{
\code{fdgs} is a data frame with 10030 rows and 8 columns:
\describe{
\item{id}{Person number}
\item{reg}{Region (factor, 5 levels)}
\item{age}{Age (years)}
\item{sex}{Sex (boy, girl)}
\item{hgt}{Height (cm)}
\item{wgt}{Weight (kg)}
\item{hgt.z}{Height Z-score}
\item{wgt.z}{Weight Z-score}
}
}
\source{
Schonbeck, Y., Talma, H., van Dommelen, P., Bakker, B., Buitendijk,
S. E., Hirasing, R. A., van Buuren, S. (2011). Increase in prevalence of
overweight in Dutch children and adolescents: A comparison of nationwide
growth studies in 1980, 1997 and 2009. \emph{PLoS ONE}, \emph{6}(11),
e27608.
Schonbeck, Y., Talma, H., van Dommelen, P., Bakker, B., Buitendijk, S. E.,
Hirasing, R. A., van Buuren, S. (2013). The world's tallest nation has
stopped growing taller: the height of Dutch children from 1955 to 2009.
\emph{Pediatric Research}, \emph{73}(3), 371-377.
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-nonresponse.html#fifth-dutch-growth-study}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Boca Raton, FL.: Chapman & Hall/CRC Press.
}
\description{
Age, height, weight and region of 10030 children measured within the Fifth
Dutch Growth Study 2009
}
\details{
The data set contains data from children of Dutch descent (biological parents
are born in the Netherlands). Children with growth-related diseases were
excluded. The data were used to construct new growth charts of children of
Dutch descent (Schonbeck 2013), and to calculate overweight and obesity
prevalence (Schonbeck 2011).
Some groups were underrepresented. Multiple imputation was used to create
synthetic cases that were used to correct for the nonresponse. See Van Buuren
(2012), chapter 8 for details.
}
\examples{
data <- data(fdgs)
summary(data)
}
\keyword{datasets}
mice/man/mice.impute.sample.Rd 0000644 0001762 0000144 00000002706 14330031647 015736 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.sample.R
\name{mice.impute.sample}
\alias{mice.impute.sample}
\title{Imputation by simple random sampling}
\usage{
mice.impute.sample(y, ry, x = NULL, wy = NULL, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes a random sample from the observed \code{y} data
}
\details{
This function takes a simple random sample from the observed values in
\code{y}, and returns these as imputations.
}
\references{
van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
}
\author{
Stef van Buuren, Karin Groothuis-Oudshoorn, 2000, 2017
}
\keyword{datagen}
mice/man/mice.impute.ri.Rd 0000644 0001762 0000144 00000005040 14330031647 015061 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.ri.R
\name{mice.impute.ri}
\alias{mice.impute.ri}
\alias{ri}
\title{Imputation by the random indicator method for nonignorable data}
\usage{
mice.impute.ri(y, ry, x, wy = NULL, ri.maxit = 10, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{ri.maxit}{Number of inner iterations}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes nonignorable missing data by the random indicator method.
}
\details{
The random indicator method estimates an offset between the
distribution of the observed and missing data using an algorithm
that iterates over the response and imputation models.
This routine assumes that the response model and imputation model
have same predictors.
For an MNAR alternative see also \code{\link{mice.impute.mnar.logreg}}.
}
\references{
Jolani, S. (2012).
\emph{Dual Imputation Strategies for Analyzing Incomplete Data}.
Dissertation. University of Utrecht, Dec 7 2012.
}
\seealso{
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()}
}
\author{
Shahab Jolani (University of Utrecht)
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/ncc.Rd 0000644 0001762 0000144 00000001113 14330031606 012764 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/ncc.R
\name{ncc}
\alias{ncc}
\title{Number of complete cases}
\usage{
ncc(x)
}
\arguments{
\item{x}{An \code{R} object. Currently supported are methods for the
following classes: \code{mids}, \code{data.frame} and \code{matrix}. Also,
\code{x} can be a vector.}
}
\value{
Number of elements in \code{x} with complete data.
}
\description{
Calculates the number of complete cases.
}
\examples{
ncc(nhanes) # 13 complete cases
}
\seealso{
\code{\link{nic}}, \code{\link{cci}}
}
\author{
Stef van Buuren, 2017
}
mice/man/version.Rd 0000644 0001762 0000144 00000000760 14330031606 013715 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/zzz.R
\name{version}
\alias{version}
\title{Echoes the package version number}
\usage{
version(pkg = "mice")
}
\arguments{
\item{pkg}{A character vector with the package name.}
}
\value{
A character vector containing the package name, version number and
installed directory.
}
\description{
Echoes the package version number
}
\examples{
version()
version("base")
}
\author{
Stef van Buuren, Oct 2010
}
\keyword{misc}
mice/man/popmis.Rd 0000644 0001762 0000144 00000001673 14330031606 013543 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/popmis.R
\docType{data}
\name{popmis}
\alias{popmis}
\title{Hox pupil popularity data with missing popularity scores}
\format{
A data frame with 2000 rows and 7 columns:
\describe{
\item{pupil}{Pupil number within school}
\item{school}{School number}
\item{popular}{Pupil popularity with 848 missing entries}
\item{sex}{Pupil gender}
\item{texp}{Teacher experience (years)}
\item{const}{Constant intercept term}
\item{teachpop}{Teacher popularity} }
}
\source{
Hox, J. J. (2002) \emph{Multilevel analysis. Techniques and
applications.} Mahwah, NJ: Lawrence Erlbaum.
}
\description{
Hox pupil popularity data with some missing popularity scores
}
\details{
The original, complete dataset was generated by Joop Hox as an example of
well-behaved multilevel data set. The distributed data contains missing data
in pupil popularity.
}
\examples{
popmis[1:3, ]
}
\keyword{datasets}
mice/man/plot.mids.Rd 0000644 0001762 0000144 00000003442 14330031606 014141 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/plot.R
\name{plot.mids}
\alias{plot.mids}
\title{Plot the trace lines of the MICE algorithm}
\usage{
\method{plot}{mids}(
x,
y = NULL,
theme = mice.theme(),
layout = c(2, 3),
type = "l",
col = 1:10,
lty = 1,
...
)
}
\arguments{
\item{x}{An object of class \code{mids}}
\item{y}{A formula that specifies which variables, stream and iterations are plotted.
If omitted, all streams, variables and iterations are plotted.}
\item{theme}{The trellis theme to applied to the graphs. The default is \code{mice.theme()}.}
\item{layout}{A vector of length 2 given the number of columns and rows in the plot.
The default is \code{c(2, 3)}.}
\item{type}{Parameter \code{type} of \code{\link{panel.xyplot}}.}
\item{col}{Parameter \code{col} of \code{\link{panel.xyplot}}.}
\item{lty}{Parameter \code{lty} of \code{\link{panel.xyplot}}.}
\item{...}{Extra arguments for \code{\link{xyplot}}.}
}
\value{
An object of class \code{"trellis"}.
}
\description{
Trace line plots portray the value of an estimate
against the iteration number. The estimate can be anything that you can calculate, but
typically are chosen as parameter of scientific interest. The \code{plot} method for
a \code{mids} object plots the mean and standard deviation of the imputed (not observed)
values against the iteration number for each of the $m$ replications. By default,
the function plot the development of the mean and standard deviation for each incomplete
variable. On convergence, the streams should intermingle and be free of any trend.
}
\examples{
imp <- mice(nhanes, print = FALSE)
plot(imp, bmi + chl ~ .it | .ms, layout = c(2, 1))
}
\seealso{
\code{\link{mice}}, \code{\link[=mids-class]{mids}},
\code{\link{xyplot}}
}
\author{
Stef van Buuren 2011
}
mice/man/convergence.Rd 0000644 0001762 0000144 00000005202 14433400023 014517 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/convergence.R
\name{convergence}
\alias{convergence}
\title{Computes convergence diagnostics for a \code{mids} object}
\usage{
convergence(data, diagnostic = "all", parameter = "mean", ...)
}
\arguments{
\item{data}{An object of class \code{mids} as created by the function
\code{mice()}.}
\item{diagnostic}{A keyword. One of the following keywords: \code{"ac"},
\code{"all"}, \code{"gr"} and \code{"psrf"}. See the Details section
for the interpretation.
The default is \code{diagnostic = "all"} which returns both the
autocorrelation and potential scale reduction factor per iteration.}
\item{parameter}{A keyword. One of the following keywords: \code{"mean"}
or \code{"sd"} to evaluate chain means or chain standard deviations,
respectively.}
\item{\dots}{Additional arguments. Not used.}
}
\value{
A \code{data.frame} with the autocorrelation and/or potential
scale reduction factor per iteration of the MICE algorithm.
}
\description{
Takes an object of class \code{mids}, computes the autocorrelation
and/or potential scale reduction factor, and returns a \code{data.frame}
with the specified diagnostic(s) per iteration.
}
\details{
The argument \code{diagnostic} can be length-1 character, which is
matched to one of the following keywords:
\describe{
\item{\code{"all"}}{computes both the lag-1 autocorrelation as well as
the potential scale reduction factor (cf. Vehtari et al., 2021) per
iteration of the MICE algorithm;}
\item{\code{"ac"}}{computes only the autocorrelation per iteration;}
\item{\code{"psrf"}}{computes only the potential scale reduction factor
per iteration;}
\item{\code{"gr"}}{same as \code{psrf}, the potential scale reduction
factor is colloquially called the Gelman-Rubin diagnostic.}
}
In the unlikely event of perfect convergence, the autocorrelation equals
zero and the potential scale reduction factor equals one. To interpret
the convergence diagnostic(s) in the output of the function, it is
recommended to plot the diagnostics (ac and/or psrf) against the
iteration number (.it) per imputed variable (vrb). A persistently
decreasing trend across iterations indicates potential non-convergence.
}
\examples{
\dontrun{
# obtain imputed data set
imp <- mice(nhanes2, print = FALSE)
# compute convergence diagnostics
convergence(imp)
}
}
\references{
Vehtari, A., Gelman, A., Simpson, D., Carpenter, B., & Burkner,
P.-C. (2021). Rank-Normalization, Folding, and Localization: An Improved
R for Assessing Convergence of MCMC. Bayesian Analysis, 1(1), 1-38.
https://doi.org/10.1214/20-BA1221
}
\seealso{
\code{\link{mice}}, \code{\link[=mids-class]{mids}}
}
\keyword{none}
mice/man/mnar_demo_data.Rd 0000644 0001762 0000144 00000001004 14330031606 015152 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mnar_demo_data.R
\docType{data}
\name{mnar_demo_data}
\alias{mnar_demo_data}
\title{MNAR demo data}
\format{
An object of class \code{data.frame} with 500 rows and 3 columns.
}
\source{
\url{https://github.com/moreno-betancur/NARFCS/blob/master/datmis.csv}
}
\usage{
mnar_demo_data
}
\description{
A toy example from Margarita Moreno-Betancur for checking NARFCS.
}
\details{
A small dataset with just three columns.
}
\keyword{datasets}
mice/man/squeeze.Rd 0000644 0001762 0000144 00000001604 14330031606 013707 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/squeeze.R
\name{squeeze}
\alias{squeeze}
\title{Squeeze the imputed values to be within specified boundaries.}
\usage{
squeeze(x, bounds = c(min(x[r]), max(x[r])), r = rep.int(TRUE, length(x)))
}
\arguments{
\item{x}{A numerical vector with values}
\item{bounds}{A numerical vector of length 2 containing the lower and upper bounds.
By default, the bounds are to the minimum and maximum values in \code{x}.}
\item{r}{A logical vector of length \code{length(x)} that is used to select a
subset in \code{x} before calculating automatic bounds.}
}
\value{
A vector of length \code{length(x)}.
}
\description{
This function replaces any values in \code{x} that are lower than
\code{bounds[1]} by \code{bounds[1]}, and replaces any values higher
than \code{bounds[2]} by \code{bounds[2]}.
}
\author{
Stef van Buuren, 2011.
}
mice/man/extractBS.Rd 0000644 0001762 0000144 00000000657 13666252075 014154 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/auxiliary.R
\name{extractBS}
\alias{extractBS}
\title{Extract broken stick estimates from a \code{lmer} object}
\usage{
extractBS(fit)
}
\arguments{
\item{fit}{An object of class \code{lmer}}
}
\value{
A matrix containing broken stick estimates
}
\description{
Extract broken stick estimates from a \code{lmer} object
}
\author{
Stef van Buuren, 2012
}
mice/man/mipo.Rd 0000644 0001762 0000144 00000005677 14330031647 013215 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mipo.R
\name{mipo}
\alias{mipo}
\alias{summary.mipo}
\alias{print.mipo}
\alias{print.mipo.summary}
\alias{process_mipo}
\title{\code{mipo}: Multiple imputation pooled object}
\usage{
mipo(mira.obj, ...)
\method{summary}{mipo}(
object,
type = c("tests", "all"),
conf.int = FALSE,
conf.level = 0.95,
exponentiate = FALSE,
...
)
\method{print}{mipo}(x, ...)
\method{print}{mipo.summary}(x, ...)
process_mipo(z, x, conf.int = FALSE, conf.level = 0.95, exponentiate = FALSE)
}
\arguments{
\item{mira.obj}{An object of class \code{mira}}
\item{\dots}{Arguments passed down}
\item{object}{An object of class \code{mipo}}
\item{conf.int}{Logical indicating whether to include
a confidence interval. The default is \code{FALSE}.}
\item{conf.level}{Confidence level of the interval, used only if
\code{conf.int = TRUE}. Number between 0 and 1.}
\item{exponentiate}{Flag indicating whether to exponentiate the
coefficient estimates and confidence intervals (typical for
logistic regression).}
\item{x}{An object of class \code{mipo}}
\item{z}{Data frame with a tidied version of a coefficient matrix}
}
\value{
The \code{summary} method returns a data frame with summary statistics of the pooled analysis.
}
\description{
The \code{mipo} object contains the results of the pooling step.
The function \code{\link{pool}} generates an object of class \code{mipo}.
}
\details{
An object class \code{mipo} is a \code{list} with
elements: \code{call}, \code{m}, \code{pooled} and \code{glanced}.
The \code{pooled} elements is a data frame with columns:
\tabular{ll}{
\code{estimate}\tab Pooled complete data estimate\cr
\code{ubar} \tab Within-imputation variance of \code{estimate}\cr
\code{b} \tab Between-imputation variance of \code{estimate}\cr
\code{t} \tab Total variance, of \code{estimate}\cr
\code{dfcom} \tab Degrees of freedom in complete data\cr
\code{df} \tab Degrees of freedom of $t$-statistic\cr
\code{riv} \tab Relative increase in variance\cr
\code{lambda} \tab Proportion attributable to the missingness\cr
\code{fmi} \tab Fraction of missing information\cr
}
The names of the terms are stored as \code{row.names(pooled)}.
The \code{glanced} elements is a \code{data.frame} with \code{m} rows.
The precise composition depends on the class of the complete-data analysis.
At least field \code{nobs} is expected to be present.
The \code{process_mipo} is a helper function to process a
tidied mipo object, and is normally not called directly.
It adds a confidence interval, and optionally exponentiates, the result.
}
\references{
van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
}
\seealso{
\code{\link{pool}},
\code{\link[=mids-class]{mids}}, \code{\link[=mira-class]{mira}}
}
\keyword{classes}
\keyword{internal}
mice/man/fdd.Rd 0000644 0001762 0000144 00000007750 14330031606 012773 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/fdd.R
\docType{data}
\name{fdd}
\alias{fdd}
\alias{fdd.pred}
\title{SE Fireworks disaster data}
\format{
\code{fdd} is a data frame with 52 rows and 65 columns:
\describe{
\item{id}{Client number}
\item{trt}{Treatment (E=EMDR, C=CBT)}
\item{pp}{Per protocol (Y/N)}
\item{trtp}{Number of parental treatments}
\item{sex}{Sex: M/F}
\item{etn}{Ethnicity: NL/OTHER}
\item{age}{Age (years)}
\item{trauma}{Trauma count (1-5)}
\item{prop1}{PROPS total score T1}
\item{prop2}{PROPS total score T2}
\item{prop3}{PROPS total score T3}
\item{crop1}{CROPS total score T1}
\item{crop2}{CROPS total score T2}
\item{crop3}{CROPS total score T3}
\item{masc1}{MASC score T1}
\item{masc2}{MASC score T2}
\item{masc3}{MASC score T3}
\item{cbcl1}{CBCL T1}
\item{cbcl3}{CBCL T3}
\item{prs1}{PRS total score T1}
\item{prs2}{PRS total score T2}
\item{prs3}{PRS total score T3}
\item{ypa1}{PTSD-RI B intrusive recollection parent T1}
\item{ypb1}{PTSD-RI C avoidant/numbing parent T1}
\item{ypc1}{PTSD-RI D hyper-arousal parent T1}
\item{yp1}{PTSD-RI B+C+D parent T1}
\item{ypa2}{PTSD-RI B intrusive recollection parent T2}
\item{ypb2}{PTSD-RI C avoidant/numbing parent T2}
\item{ypc2}{PTSD-RI D hyper-arousal parent T2}
\item{yp2}{PTSD-RI B+C+D parent T1}
\item{ypa3}{PTSD-RI B intrusive recollection parent T3}
\item{ypb3}{PTSD-RI C avoidant/numbing parent T3}
\item{ypc3}{PTSD-RI D hyper-arousal parent T3}
\item{yp3}{PTSD-RI B+C+D parent T3}
\item{yca1}{PTSD-RI B intrusive recollection child T1}
\item{ycb1}{PTSD-RI C avoidant/numbing child T1}
\item{ycc1}{PTSD-RI D hyper-arousal child T1}
\item{yc1}{PTSD-RI B+C+D child T1}
\item{yca2}{PTSD-RI B intrusive recollection child T2}
\item{ycb2}{PTSD-RI C avoidant/numbing child T2}
\item{ycc2}{PTSD-RI D hyper-arousal child T2}
\item{yc2}{PTSD-RI B+C+D child T2}
\item{yca3}{PTSD-RI B intrusive recollection child T3}
\item{ycb3}{PTSD-RI C avoidant/numbing child T3}
\item{ycc3}{PTSD-RI D hyper-arousal child T3}
\item{yc3}{PTSD-RI B+C+D child T3}
\item{ypf1}{PTSD-RI parent full T1}
\item{ypf2}{PTSD-RI parent full T2}
\item{ypf3}{PTSD-RI parent full T3}
\item{ypp1}{PTSD parent partial T1}
\item{ypp2}{PTSD parent partial T2}
\item{ypp3}{PTSD parent partial T3}
\item{ycf1}{PTSD child full T1}
\item{ycf2}{PTSD child full T2}
\item{ycf3}{PTSD child full T3}
\item{ycp1}{PTSD child partial T1}
\item{ycp2}{PTSD child partial T2}
\item{ycp3}{PTSD child partial T3}
\item{cbin1}{CBCL Internalizing T1}
\item{cbin3}{CBCL Internalizing T3}
\item{cbex1}{CBCL Externalizing T1}
\item{cbex3}{CBCL Externalizing T3}
\item{bir1}{Birlison T1}
\item{bir2}{Birlison T2}
\item{bir3}{Birlison T3}
}
\code{fdd.pred} is the 65 by 65 binary
predictor matrix used to impute \code{fdd}.
}
\source{
de Roos, C., Greenwald, R., den Hollander-Gijsman, M., Noorthoorn,
E., van Buuren, S., de Jong, A. (2011). A Randomised Comparison of Cognitive
Behavioral Therapy (CBT) and Eye Movement Desensitisation and Reprocessing
(EMDR) in disaster-exposed children. \emph{European Journal of
Psychotraumatology}, \emph{2}, 5694.
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-fdd.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
Boca Raton, FL.: Chapman & Hall/CRC Press.
}
\description{
Multiple outcomes of a randomized study to reduce post-traumatic stress.
}
\details{
Data from a randomized experiment to reduce post-traumatic stress by two
treatments: Eye Movement Desensitization and Reprocessing (EMDR)
(experimental treatment), and cognitive behavioral therapy (CBT) (control
treatment). 52 children were randomized to one of these two treatments.
Outcomes were measured at three time points: at baseline (pre-treatment, T1),
post-treatment (T2, 4-8 weeks), and at follow-up (T3, 3 months). For more
details, see de Roos et al (2011). Some person covariates were reshuffled.
The imputation methodology is explained in Chapter 9 of van Buuren (2012).
}
\examples{
data <- fdd
md.pattern(fdd)
}
\keyword{datasets}
mice/man/mids2spss.Rd 0000644 0001762 0000144 00000004725 14330031606 014164 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mids2spss.R
\name{mids2spss}
\alias{mids2spss}
\title{Export \code{mids} object to SPSS}
\usage{
mids2spss(
imp,
filename = "midsdata",
path = getwd(),
compress = FALSE,
silent = FALSE
)
}
\arguments{
\item{imp}{The \code{imp} argument is an object of class \code{mids},
typically produced by the \code{mice()} function.}
\item{filename}{A character string describing the name of the output data
file and its extension.}
\item{path}{A character string containing the path of the output file. The
value in \code{path} is appended to \code{filedat}. By
default, files are written to the current \code{R} working directory. If
\code{path=NULL} then no file path appending is done.}
\item{compress}{A logical flag stating whether the resulting SPSS set should
be a compressed \code{.zsav} file.}
\item{silent}{A logical flag stating whether the location of the saved file should be
printed.}
}
\value{
The return value is \code{NULL}.
}
\description{
Converts a \code{mids} object into a format recognized by SPSS, and writes
the data and the SPSS syntax files.
}
\details{
This function automates most of the work needed to export a \code{mids}
object to SPSS. It uses \code{haven::write_sav()} to facilitate the export to an
SPSS \code{.sav} or \code{.zsav} file.
Below are some things to pay attention to.
The \code{SPSS} syntax file has the proper file names and separators set, so
in principle it should run and read the data without alteration. \code{SPSS}
is more strict than \code{R} with respect to the paths. Always use the full
path, otherwise \code{SPSS} may not be able to find the data file.
Factors in \code{R} translate into categorical variables in \code{SPSS}. The
internal coding of factor levels used in \code{R} is exported. This is
generally acceptable for \code{SPSS}. However, when the data are to be
combined with existing \code{SPSS} data, watch out for any changes in the
factor levels codes.
\code{SPSS} will recognize the data set as a multiply imputed data set, and
do automatic pooling in procedures where that is supported. Note however that
pooling is an extra option only available to those who license the
\code{MISSING VALUES} module. Without this license, \code{SPSS} will still
recognize the structure of the data, but it will not pool the multiply imputed
estimates into a single inference.
}
\seealso{
\code{\link[=mids-class]{mids}}
}
\author{
Gerko Vink, dec 2020.
}
\keyword{manip}
mice/man/mice.impute.mean.Rd 0000644 0001762 0000144 00000005132 14330031647 015371 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.mean.R
\name{mice.impute.mean}
\alias{mice.impute.mean}
\title{Imputation by the mean}
\usage{
mice.impute.mean(y, ry, x = NULL, wy = NULL, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes the arithmetic mean of the observed data
}
\section{Warning}{
Imputing the mean of a variable is almost never
appropriate. See Little and Rubin (2002, p. 61-62) or
Van Buuren (2012, p. 10-11)
}
\references{
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
Little, R.J.A. and Rubin, D.B. (2002). Statistical Analysis with Missing
Data. New York: John Wiley and Sons.
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-simplesolutions.html#sec:meanimp}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\seealso{
\code{\link{mice}}, \code{\link{mean}}
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/mice.impute.lasso.select.norm.Rd 0000644 0001762 0000144 00000007612 14330031647 020027 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.lasso.select.norm.R
\name{mice.impute.lasso.select.norm}
\alias{mice.impute.lasso.select.norm}
\alias{lasso.select.norm}
\title{Imputation by indirect use of lasso linear regression}
\usage{
mice.impute.lasso.select.norm(y, ry, x, wy = NULL, nfolds = 10, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{nfolds}{The number of folds for the cross-validation of the lasso penalty.
The default is 10.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes univariate missing data using Bayesian linear regression following a
preprocessing lasso variable selection step.
}
\details{
The method consists of the following steps:
\enumerate{
\item For a given \code{y} variable under imputation, fit a linear regression with lasso
penalty using \code{y[ry]} as dependent variable and \code{x[ry, ]} as predictors.
Coefficients that are not shrunk to 0 define an active set of predictors
that will be used for imputation
\item Define a Bayesian linear model using \code{y[ry]} as the
dependent variable, the active set of \code{x[ry, ]} as predictors, and standard
non-informative priors
\item Draw parameter values for the intercept, regression weights, and error
variance from their posterior distribution
\item Draw imputations from the posterior predictive distribution
}
The user can specify a \code{predictorMatrix} in the \code{mice} call
to define which predictors are provided to this univariate imputation method.
The lasso regularization will select, among the variables indicated by
the user, the ones that are important for imputation at any given iteration.
Therefore, users may force the exclusion of a predictor from a given
imputation model by specifying a \code{0} entry.
However, a non-zero entry does not guarantee the variable will be used,
as this decision is ultimately made by the lasso variable selection
procedure.
The method is based on the Indirect Use of Regularized Regression (IURR) proposed by
Zhao & Long (2016) and Deng et al (2016).
}
\references{
Deng, Y., Chang, C., Ido, M. S., & Long, Q. (2016). Multiple imputation for
general missing data patterns in the presence of high-dimensional data.
Scientific reports, 6(1), 1-10.
Zhao, Y., & Long, Q. (2016). Multiple imputation in the presence of
high-dimensional data. Statistical Methods in Medical Research, 25(5),
2021-2035.
}
\seealso{
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Edoardo Costantini, 2021
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/mice.impute.cart.Rd 0000644 0001762 0000144 00000007031 14436637036 015415 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.cart.R
\name{mice.impute.cart}
\alias{mice.impute.cart}
\alias{cart}
\title{Imputation by classification and regression trees}
\usage{
mice.impute.cart(y, ry, x, wy = NULL, minbucket = 5, cp = 1e-04, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{minbucket}{The minimum number of observations in any terminal node used.
See \code{\link{rpart.control}} for details.}
\item{cp}{Complexity parameter. Any split that does not decrease the overall
lack of fit by a factor of cp is not attempted. See \code{\link{rpart.control}}
for details.}
\item{...}{Other named arguments passed down to \code{rpart()}.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
Numeric vector of length \code{sum(!ry)} with imputations
}
\description{
Imputes univariate missing data using classification and regression trees.
}
\details{
Imputation of \code{y} by classification and regression trees. The procedure
is as follows:
\enumerate{
\item Fit a classification or regression tree by recursive partitioning;
\item For each \code{ymis}, find the terminal node they end up according to the fitted tree;
\item Make a random draw among the member in the node, and take the observed value from that
draw as the imputation.
}
}
\examples{
imp <- mice(nhanes2, meth = "cart", minbucket = 4)
plot(imp)
}
\references{
Doove, L.L., van Buuren, S., Dusseldorp, E. (2014), Recursive partitioning
for missing data imputation in the presence of interaction Effects.
Computational Statistics & Data Analysis, 72, 92-104.
Breiman, L., Friedman, J. H., Olshen, R. A., and Stone, C. J.
(1984), Classification and regression trees, Monterey, CA: Wadsworth &
Brooks/Cole Advanced Books & Software.
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-cart.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\seealso{
\code{\link{mice}}, \code{\link{mice.impute.rf}},
\code{\link[rpart]{rpart}}, \code{\link[rpart]{rpart.control}}
Other univariate imputation functions:
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Lisa Doove, Stef van Buuren, Elise Dusseldorp, 2012
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/mcar.Rd 0000644 0001762 0000144 00000013024 14330032376 013154 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mcar.R
\name{mcar}
\alias{mcar}
\title{Jamshidian and Jalal's Non-Parametric MCAR Test}
\usage{
mcar(
x,
imputed = mice(x, method = "norm"),
min_n = 6,
method = "auto",
replications = 10000,
use_chisq = 30,
alpha = 0.05
)
}
\arguments{
\item{x}{An object for which a method exists; usually a \code{data.frame}.}
\item{imputed}{Either an object of class \code{mids}, as returned by
\code{\link[=mice]{mice()}}, or a list of \code{data.frame}s.}
\item{min_n}{Atomic numeric, must be greater than 1. When there are missing
data patterns with fewer than \code{min_n} cases, all cases with that pattern will
be removed from \code{x} and \code{imputed}.}
\item{method}{Atomic character. If it is known (or assumed) that data are
either multivariate normally distributed or not, then use either
\code{method = "hawkins"} or \code{method = "nonparametric"}, respectively.
The default argument \code{method = "auto"} follows the procedure outlined in the
Details section, and in Figure 7 of Jamshidian and Jalal (2010).}
\item{replications}{Number of replications used to simulate the Neyman
distribution when performing Hawkins' test. As this method is based on random
sampling, use a high number of \code{replications} (and optionally,
\code{\link[=set.seed]{set.seed()}}) to minimize Monte Carlo error and ensure reproducibility.}
\item{use_chisq}{Atomic integer, indicating the minimum number of cases
within a group \emph{k} that triggers the use of asymptotic Chi-square
distribution instead of the emprical distribution in the Neyman uniformity
test, which is performed as part of Hawkins' test.}
\item{alpha}{Atomic numeric, indicating the significance level of tests.}
}
\value{
An object of class \code{mcar_object}.
}
\description{
Test whether missingness is contingent upon the observed variables,
according to the methodology developed by Jamshidian and Jalal (2010) (see
Details).
}
\details{
Three types of missingness have been distinguished in the literature
(Rubin, 1976):
Missing completely at random (MCAR), which means that missingness is random;
missing at random (MAR), which means that missingness is contingent on the
\emph{observed};
and missing not at random (MNAR), which means that missingness is related to
unobserved data.
Jamshidian and Jalal's non-parametric MCAR test assumes that the missing data
are either MCAR or MAR, and tests whether the missingness is independent of
the observed values. If so, the covariance matrices of the imputed data will
be equal accross groups with different patterns of missingness. This test
consists of the following procedure:
\enumerate{
\item Data are imputed.
\item The imputed data are split into \emph{k} groups according to the
\emph{k} missing data patterns in the original data (see
\code{\link[=md.pattern]{md.pattern()}}).
\item Perform Hawkins' test for equality of covariances across the \emph{k}
groups.
\item If the test is \emph{not significant}, conclude that there is no evidence
against multivariate normality of the data, nor against MCAR.
\item If the test \emph{is significant}, and multivariate normality of the data
can be assumed, then it can be concluded that missingness is MAR.
\item If multivariate normality cannot be assumed, then perform the
Anderson-Darling non-parametric test for equality of covariances across the
\emph{k} groups.
\item If the Anderson-Darling test is \emph{not significant}, this is evidence
against multivariate normality - but no evidence against MCAR.
\item If the Anderson-Darling test \emph{is significant}, this is evidence
it can be concluded that missingness is MAR.
}
Note that, despite its name in common parlance, an MCAR test can only
indicate whether missingness is MCAR or MAR. The procedure cannot distinguish
MCAR from MNAR, so a non-significant result does not rule out MNAR.
This is a re-implementation of the function \code{TestMCARNormality}, which was
originally published in the R-packgage \code{MissMech}, which has been removed
from CRAN. This new implementation is faster, as its backend is written in
C++. It also enhances the functionality of the original:
\itemize{
\item Multiply imputed data can now be used; the median p-value and test
statistic across replications is then reported, as suggested by
Eekhout, Wiel, and Heymans (2017).
\item The printing method for an \code{mcar_object} gives a warning when at
least one p-value of either test was significant. In this case, it is
recommended to inspect the range of p-values, and consider potential
violations of MCAR.
\item A plotting method for an \code{mcar_object} is provided.
\item A plotting method for the \verb{$md.pattern} element of an \code{mcar_object}
is provided.
}
}
\examples{
res <- mcar(nhanes)
# Examine test results
res
# Plot p-values across imputed data sets
plot(res)
# Plot md patterns used for the test
plot(res, type = "md.pattern")
# Note difference with the raw md.patterns:
md.pattern(nhanes)
}
\references{
Rubin, D. B. (1976). Inference and Missing Data. Biometrika, Vol. 63, No. 3,
pp. 581-592. \doi{10.2307/2335739}
Eekhout, I., M. A. Wiel, & M. W. Heymans (2017). Methods for Significance
Testing of Categorical Covariates in Logistic Regression Models After
Multiple Imputation: Power and Applicability Analysis. BMC Medical Research
Methodology 17 (1): 129.
Jamshidian, M., & Jalal, S. (2010). Tests of homoscedasticity, normality, and
missing completely at random for incomplete multivariate data. Psychometrika,
75(4), 649–674. \doi{10.1007/s11336-010-9175-3}
}
\author{
Caspar J. Van Lissa
}
\keyword{internal}
mice/man/flux.Rd 0000644 0001762 0000144 00000005363 14330031606 013212 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/flux.R
\name{flux}
\alias{flux}
\title{Influx and outflux of multivariate missing data patterns}
\usage{
flux(data, local = names(data))
}
\arguments{
\item{data}{A data frame or a matrix containing the incomplete data. Missing
values are coded as NA's.}
\item{local}{A vector of names of columns of \code{data}. The default is to
include all columns in the calculations.}
}
\value{
A data frame with \code{ncol(data)} rows and six columns:
pobs = Proportion observed,
influx = Influx
outflux = Outflux
ainb = Average inbound statistic
aout = Average outbound statistic
fico = Fraction of incomplete cases among cases with \code{Yj} observed
}
\description{
Influx and outflux are statistics of the missing data pattern. These
statistics are useful in selecting predictors that should go into the
imputation model.
}
\details{
Infux and outflux have been proposed by Van Buuren (2018), chapter 4.
Influx is equal to the number of variable pairs \code{(Yj , Yk)} with
\code{Yj} missing and \code{Yk} observed, divided by the total number of
observed data cells. Influx depends on the proportion of missing data of the
variable. Influx of a completely observed variable is equal to 0, whereas for
completely missing variables we have influx = 1. For two variables with the
same proportion of missing data, the variable with higher influx is better
connected to the observed data, and might thus be easier to impute.
Outflux is equal to the number of variable pairs with \code{Yj} observed and
\code{Yk} missing, divided by the total number of incomplete data cells.
Outflux is an indicator of the potential usefulness of \code{Yj} for imputing
other variables. Outflux depends on the proportion of missing data of the
variable. Outflux of a completely observed variable is equal to 1, whereas
outflux of a completely missing variable is equal to 0. For two variables
having the same proportion of missing data, the variable with higher outflux
is better connected to the missing data, and thus potentially more useful for
imputing other variables.
FICO is an outbound statistic defined by the fraction of incomplete cases
among cases with \code{Yj} observed (White and Carlin, 2010).
}
\references{
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/missing-data-pattern.html#sec:flux}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
White, I.R., Carlin, J.B. (2010). Bias and efficiency of multiple imputation
compared with complete-case analysis for missing covariate values.
\emph{Statistics in Medicine}, \emph{29}, 2920-2931.
}
\seealso{
\code{\link{fluxplot}}, \code{\link{md.pattern}}, \code{\link{fico}}
}
\author{
Stef van Buuren, 2012
}
\keyword{misc}
mice/man/make.predictorMatrix.Rd 0000644 0001762 0000144 00000002226 14347615202 016333 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/predictorMatrix.R
\name{make.predictorMatrix}
\alias{make.predictorMatrix}
\title{Creates a \code{predictorMatrix} argument}
\usage{
make.predictorMatrix(data, blocks = make.blocks(data), predictorMatrix = NULL)
}
\arguments{
\item{data}{A \code{data.frame} with the source data}
\item{blocks}{An optional specification for blocks of variables in
the rows. The default assigns each variable in its own block.}
\item{predictorMatrix}{A predictor matrix from which rows with the same
names are copied into the output predictor matrix.}
}
\value{
A matrix
}
\description{
This helper function creates a valid \code{predictMatrix}. The
\code{predictorMatrix} is an argument to the \code{mice} function.
It specifies the target variable or block in the rows, and the
predictor variables on the columns. An entry of \code{0} means that
the column variable is NOT used to impute the row variable or block.
A nonzero value indicates that it is used.
}
\examples{
make.predictorMatrix(nhanes)
make.predictorMatrix(nhanes, blocks = make.blocks(nhanes, "collect"))
}
\seealso{
\code{\link{make.blocks}}
}
mice/man/norm.draw.Rd 0000644 0001762 0000144 00000002611 14330031606 014134 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.norm.R
\name{norm.draw}
\alias{norm.draw}
\alias{.norm.draw}
\title{Draws values of beta and sigma by Bayesian linear regression}
\usage{
norm.draw(y, ry, x, rank.adjust = TRUE, ...)
.norm.draw(y, ry, x, rank.adjust = TRUE, ...)
}
\arguments{
\item{y}{Incomplete data vector of length \code{n}}
\item{ry}{Vector of missing data pattern (\code{FALSE}=missing,
\code{TRUE}=observed)}
\item{x}{Matrix (\code{n} x \code{p}) of complete covariates.}
\item{rank.adjust}{Argument that specifies whether \code{NA}'s in the
coefficients need to be set to zero. Only relevant when \code{ls.meth = "qr"}
AND the predictor matrix is rank-deficient.}
\item{...}{Other named arguments.}
}
\value{
A \code{list} containing components \code{coef} (least squares estimate),
\code{beta} (drawn regression weights) and \code{sigma} (drawn value of the
residual standard deviation).
}
\description{
This function draws random values of beta and sigma under the Bayesian
linear regression model as described in Rubin (1987, p. 167). This function
can be called by user-specified imputation functions.
}
\references{
Rubin, D.B. (1987). \emph{Multiple imputation for nonresponse in surveys}. New York: Wiley.
}
\author{
Gerko Vink, 2018, for this version, based on earlier versions written
by Stef van Buuren, Karin Groothuis-Oudshoorn, 2017
}
mice/man/fluxplot.Rd 0000644 0001762 0000144 00000006522 14330031606 014107 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/flux.R
\name{fluxplot}
\alias{fluxplot}
\title{Fluxplot of the missing data pattern}
\usage{
fluxplot(
data,
local = names(data),
plot = TRUE,
labels = TRUE,
xlim = c(0, 1),
ylim = c(0, 1),
las = 1,
xlab = "Influx",
ylab = "Outflux",
main = paste("Influx-outflux pattern for", deparse(substitute(data))),
eqscplot = TRUE,
pty = "s",
lwd = 1,
...
)
}
\arguments{
\item{data}{A data frame or a matrix containing the incomplete data. Missing
values are coded as NA's.}
\item{local}{A vector of names of columns of \code{data}. The default is to
include all columns in the calculations.}
\item{plot}{Should a graph be produced?}
\item{labels}{Should the points be labeled?}
\item{xlim}{See \code{par}.}
\item{ylim}{See \code{par}.}
\item{las}{See \code{par}.}
\item{xlab}{See \code{par}.}
\item{ylab}{See \code{par}.}
\item{main}{See \code{par}.}
\item{eqscplot}{Should a square plot be produced?}
\item{pty}{See \code{par}.}
\item{lwd}{See \code{par}. Controls axis line thickness and diagonal}
\item{\dots}{Further arguments passed to \code{plot()} or \code{eqscplot()}.}
}
\value{
An invisible data frame with \code{ncol(data)} rows and six columns:
pobs = Proportion observed,
influx = Influx
outflux = Outflux
ainb = Average inbound statistic
aout = Average outbound statistic
fico = Fraction of incomplete cases among cases with \code{Yj} observed
}
\description{
Influx and outflux are statistics of the missing data pattern. These
statistics are useful in selecting predictors that should go into the
imputation model.
}
\details{
Infux and outflux have been proposed by Van Buuren (2012), chapter 4.
Influx is equal to the number of variable pairs \code{(Yj , Yk)} with
\code{Yj} missing and \code{Yk} observed, divided by the total number of
observed data cells. Influx depends on the proportion of missing data of the
variable. Influx of a completely observed variable is equal to 0, whereas for
completely missing variables we have influx = 1. For two variables with the
same proportion of missing data, the variable with higher influx is better
connected to the observed data, and might thus be easier to impute.
Outflux is equal to the number of variable pairs with \code{Yj} observed and
\code{Yk} missing, divided by the total number of incomplete data cells.
Outflux is an indicator of the potential usefulness of \code{Yj} for imputing
other variables. Outflux depends on the proportion of missing data of the
variable. Outflux of a completely observed variable is equal to 1, whereas
outflux of a completely missing variable is equal to 0. For two variables
having the same proportion of missing data, the variable with higher outflux
is better connected to the missing data, and thus potentially more useful for
imputing other variables.
}
\references{
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/missing-data-pattern.html#sec:flux}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
White, I.R., Carlin, J.B. (2010). Bias and efficiency of multiple imputation
compared with complete-case analysis for missing covariate values.
\emph{Statistics in Medicine}, \emph{29}, 2920-2931.
}
\seealso{
\code{\link{flux}}, \code{\link{md.pattern}}, \code{\link{fico}}
}
\author{
Stef van Buuren, 2012
}
\keyword{misc}
mice/man/mice.impute.2l.bin.Rd 0000644 0001762 0000144 00000005071 14330031606 015532 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.2l.bin.R
\name{mice.impute.2l.bin}
\alias{mice.impute.2l.bin}
\title{Imputation by a two-level logistic model using \code{glmer}}
\usage{
mice.impute.2l.bin(y, ry, x, type, wy = NULL, intercept = TRUE, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{type}{Vector of length \code{ncol(x)} identifying random and class
variables. Random variables are identified by a '2'. The class variable
(only one is allowed) is coded as '-2'. Fixed effects are indicated by
a '1'.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{intercept}{Logical determining whether the intercept is automatically
added.}
\item{\dots}{Arguments passed down to \code{glmer}}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes univariate systematically and sporadically missing data
using a two-level logistic model using \code{lme4::glmer()}
}
\details{
Data are missing systematically if they have not been measured, e.g., in the
case where we combine data from different sources. Data are missing sporadically
if they have been partially observed.
}
\examples{
library(tidyr)
library(dplyr)
data("toenail2")
data <- tidyr::complete(toenail2, patientID, visit) \%>\%
tidyr::fill(treatment) \%>\%
dplyr::select(-time) \%>\%
dplyr::mutate(patientID = as.integer(patientID))
\dontrun{
pred <- mice(data, print = FALSE, maxit = 0, seed = 1)$pred
pred["outcome", "patientID"] <- -2
imp <- mice(data, method = "2l.bin", pred = pred, maxit = 1, m = 1, seed = 1)
}
}
\references{
Jolani S., Debray T.P.A., Koffijberg H., van Buuren S., Moons K.G.M. (2015).
Imputation of systematically missing predictors in an individual
participant data meta-analysis: a generalized approach using MICE.
\emph{Statistics in Medicine}, 34:1841-1863.
}
\seealso{
Other univariate-2l:
\code{\link{mice.impute.2l.lmer}()},
\code{\link{mice.impute.2l.norm}()},
\code{\link{mice.impute.2l.pan}()}
}
\author{
Shahab Jolani, 2015; adapted to mice, SvB, 2018
}
\concept{univariate-2l}
\keyword{datagen}
mice/man/is.mira.Rd 0000644 0001762 0000144 00000000515 13666252075 013610 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/is.R
\name{is.mira}
\alias{is.mira}
\title{Check for \code{mira} object}
\usage{
is.mira(x)
}
\arguments{
\item{x}{An object}
}
\value{
A logical indicating whether \code{x} is an object of class \code{mira}
}
\description{
Check for \code{mira} object
}
mice/man/getfit.Rd 0000644 0001762 0000144 00000002762 14330031606 013516 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/getfit.R
\name{getfit}
\alias{getfit}
\title{Extract list of fitted models}
\usage{
getfit(x, i = -1L, simplify = FALSE)
}
\arguments{
\item{x}{An object of class \code{mira}, typically produced by a call
to \code{with()}.}
\item{i}{An integer between 1 and \code{x$m} signalling the index of the
repeated analysis. The default \code{i= -1} return a list with all analyses.}
\item{simplify}{Should the return value be unlisted?}
}
\value{
If \code{i = -1} an object of class \code{mira} containing
all analyses. If \code{i} selects one of the analyses, then it return
an object whose with class inherited from that element.
}
\description{
Function \code{getfit()} returns the list of objects containing the repeated analysis
results, or optionally, one of these fitted objects. The function looks for
a list element called \code{analyses}, and return this component as a list with
\code{mira} class. If element \code{analyses} is not found in \code{x}, then
it returns \code{x} as a \code{mira} object.
}
\details{
No checking is done for validity of objects. The function also processes
objects of class \code{mitml.result} from the \code{mitml} package.
}
\examples{
imp <- mice(nhanes, print = FALSE, seed = 21443)
fit <- with(imp, lm(bmi ~ chl + hyp))
f1 <- getfit(fit)
class(f1)
f2 <- getfit(fit, 2)
class(f2)
}
\seealso{
\code{\link[=mira-class]{mira}}, \code{\link{with.mids}}
}
\author{
Stef van Buuren, 2012, 2020
}
\keyword{manip}
mice/man/ampute.continuous.Rd 0000644 0001762 0000144 00000004076 14330031606 015734 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/ampute.continuous.R
\name{ampute.continuous}
\alias{ampute.continuous}
\title{Multivariate amputation based on continuous probability functions}
\usage{
ampute.continuous(P, scores, prop, type)
}
\arguments{
\item{P}{A vector containing the pattern numbers of the cases's candidacies.
For each case, a value between 1 and #patterns is given. For example, a
case with value 2 is candidate for missing data pattern 2.}
\item{scores}{A list containing vectors with the candidates's weighted sum scores,
the result of an underlying function in \code{\link{ampute}}.}
\item{prop}{A scalar specifying the proportion of missingness. Should be a value
between 0 and 1. Default is a missingness proportion of 0.5.}
\item{type}{A vector of strings containing the type of missingness for each
pattern. Either \code{"LEFT"}, \code{"MID"}, \code{"TAIL"} or '\code{"RIGHT"}.
If a single missingness type is entered, all patterns will be created by the same
type. If missingness types should differ over patterns, a vector of missingness
types should be entered. Default is RIGHT for all patterns and is the result of
\code{\link{ampute.default.type}}.}
}
\value{
A list containing vectors with \code{0} if a case should be made missing
and \code{1} if a case should remain complete. The first vector refers to the
first pattern, the second vector to the second pattern, etcetera.
}
\description{
This function creates a missing data indicator for each pattern. The continuous
probability distributions (Van Buuren, 2012, pp. 63, 64) will be induced on the
weighted sum scores, calculated earlier in the multivariate amputation function
\code{\link{ampute}}.
}
\references{
#'Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-linearnormal.html#sec:generateuni}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\seealso{
\code{\link{ampute}}, \code{\link{ampute.default.type}}
}
\author{
Rianne Schouten [aut, cre], Gerko Vink [aut], Peter Lugtig [ctb], 2016
}
\keyword{internal}
mice/man/mice.impute.pmm.Rd 0000644 0001762 0000144 00000015240 14335404116 015243 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.pmm.R
\name{mice.impute.pmm}
\alias{mice.impute.pmm}
\alias{pmm}
\title{Imputation by predictive mean matching}
\usage{
mice.impute.pmm(
y,
ry,
x,
wy = NULL,
donors = 5L,
matchtype = 1L,
exclude = -99999999,
ridge = 1e-05,
use.matcher = FALSE,
...
)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{donors}{The size of the donor pool among which a draw is made.
The default is \code{donors = 5L}. Setting \code{donors = 1L} always selects
the closest match, but is not recommended. Values between 3L and 10L
provide the best results in most cases (Morris et al, 2015).}
\item{matchtype}{Type of matching distance. The default choice
(\code{matchtype = 1L}) calculates the distance between
the \emph{predicted} value of \code{yobs} and
the \emph{drawn} values of \code{ymis} (called type-1 matching).
Other choices are \code{matchtype = 0L}
(distance between predicted values) and \code{matchtype = 2L}
(distance between drawn values).}
\item{exclude}{Value or vector of values to exclude from the imputation donor pool in \code{y}}
\item{ridge}{The ridge penalty used in \code{.norm.draw()} to prevent
problems with multicollinearity. The default is \code{ridge = 1e-05},
which means that 0.01 percent of the diagonal is added to the cross-product.
Larger ridges may result in more biased estimates. For highly noisy data
(e.g. many junk variables), set \code{ridge = 1e-06} or even lower to
reduce bias. For highly collinear data, set \code{ridge = 1e-04} or higher.}
\item{use.matcher}{Logical. Set \code{use.matcher = TRUE} to specify
the C function \code{matcher()}, the now deprecated matching function that
was default in versions
\code{2.22} (June 2014) to \code{3.11.7} (Oct 2020). Since version \code{3.12.0}
\code{mice()} uses the much faster \code{matchindex} C function. Use
the deprecated \code{matcher} function only for exact reproduction.}
\item{\dots}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputation by predictive mean matching
}
\details{
Imputation of \code{y} by predictive mean matching, based on
van Buuren (2012, p. 73). The procedure is as follows:
\enumerate{
\item{Calculate the cross-product matrix \eqn{S=X_{obs}'X_{obs}}.}
\item{Calculate \eqn{V = (S+{diag}(S)\kappa)^{-1}}, with some small ridge
parameter \eqn{\kappa}.}
\item{Calculate regression weights \eqn{\hat\beta = VX_{obs}'y_{obs}.}}
\item{Draw \eqn{q} independent \eqn{N(0,1)} variates in vector \eqn{\dot z_1}.}
\item{Calculate \eqn{V^{1/2}} by Cholesky decomposition.}
\item{Calculate \eqn{\dot\beta = \hat\beta + \dot\sigma\dot z_1 V^{1/2}}.}
\item{Calculate \eqn{\dot\eta(i,j)=|X_{{obs},[i]|}\hat\beta-X_{{mis},[j]}\dot\beta}
with \eqn{i=1,\dots,n_1} and \eqn{j=1,\dots,n_0}.}
\item{Construct \eqn{n_0} sets \eqn{Z_j}, each containing \eqn{d} candidate donors, from Y_{obs} such that \eqn{\sum_d\dot\eta(i,j)} is minimum for all \eqn{j=1,\dots,n_0}. Break ties randomly.}
\item{Draw one donor \eqn{i_j} from \eqn{Z_j} randomly for \eqn{j=1,\dots,n_0}.}
\item{Calculate imputations \eqn{\dot y_j = y_{i_j}} for \eqn{j=1,\dots,n_0}.}
}
The name \emph{predictive mean matching} was proposed by Little (1988).
}
\examples{
# We normally call mice.impute.pmm() from within mice()
# But we may call it directly as follows (not recommended)
set.seed(53177)
xname <- c("age", "hgt", "wgt")
r <- stats::complete.cases(boys[, xname])
x <- boys[r, xname]
y <- boys[r, "tv"]
ry <- !is.na(y)
table(ry)
# percentage of missing data in tv
sum(!ry) / length(ry)
# Impute missing tv data
yimp <- mice.impute.pmm(y, ry, x)
length(yimp)
hist(yimp, xlab = "Imputed missing tv")
# Impute all tv data
yimp <- mice.impute.pmm(y, ry, x, wy = rep(TRUE, length(y)))
length(yimp)
hist(yimp, xlab = "Imputed missing and observed tv")
plot(jitter(y), jitter(yimp),
main = "Predictive mean matching on age, height and weight",
xlab = "Observed tv (n = 224)",
ylab = "Imputed tv (n = 224)"
)
abline(0, 1)
cor(y, yimp, use = "pair")
# Use blots to exclude different values per column
# Create blots object
blots <- make.blots(boys)
# Exclude ml 1 through 5 from tv donor pool
blots$tv$exclude <- c(1:5)
# Exclude 100 random observed heights from tv donor pool
blots$hgt$exclude <- sample(unique(boys$hgt), 100)
imp <- mice(boys, method = "pmm", print = FALSE, blots = blots, seed=123)
blots$hgt$exclude \%in\% unlist(c(imp$imp$hgt)) # MUST be all FALSE
blots$tv$exclude \%in\% unlist(c(imp$imp$tv)) # MUST be all FALSE
}
\references{
Little, R.J.A. (1988), Missing data adjustments in large surveys
(with discussion), Journal of Business Economics and Statistics, 6, 287--301.
Morris TP, White IR, Royston P (2015). Tuning multiple imputation by predictive
mean matching and local residual draws. BMC Med Res Methodol. ;14:75.
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-pmm.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}: Multivariate
Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
}
\seealso{
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Gerko Vink, Stef van Buuren, Karin Groothuis-Oudshoorn
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/getqbar.Rd 0000644 0001762 0000144 00000000476 13666252075 013701 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/getfit.R
\name{getqbar}
\alias{getqbar}
\title{Extract estimate from \code{mipo} object}
\usage{
getqbar(x)
}
\arguments{
\item{x}{An object of class \code{mipo}}
}
\description{
\code{getqbar} returns a named vector of pooled estimates.
}
mice/man/mice.impute.norm.nob.Rd 0000644 0001762 0000144 00000006454 14330031647 016211 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.norm.nob.R
\name{mice.impute.norm.nob}
\alias{mice.impute.norm.nob}
\alias{norm.nob}
\title{Imputation by linear regression without parameter uncertainty}
\usage{
mice.impute.norm.nob(y, ry, x, wy = NULL, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes univariate missing data using linear regression analysis without
accounting for the uncertainty of the model parameters.
}
\details{
This function creates imputations using the spread around the
fitted linear regression line of \code{y} given \code{x}, as
fitted on the observed data.
This function is provided mainly to allow comparison between proper (e.g.,
as implemented in \code{mice.impute.norm} and improper (this function)
normal imputation methods.
For large data, having many rows, differences between proper and improper
methods are small, and in those cases one may opt for speed by using
\code{mice.impute.norm.nob}.
}
\section{Warning}{
The function does not incorporate the variability of the
regression weights, so it is not 'proper' in the sense of Rubin. For small
samples, variability of the imputed data is therefore underestimated.
}
\references{
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
Brand, J.P.L. (1999). Development, Implementation and Evaluation of Multiple
Imputation Strategies for the Statistical Analysis of Incomplete Data Sets.
Ph.D. Thesis, TNO Prevention and Health/Erasmus University Rotterdam.
}
\seealso{
\code{\link{mice}}, \code{\link{mice.impute.norm}}
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Gerko Vink, Stef van Buuren, Karin Groothuis-Oudshoorn, 2018
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/make.post.Rd 0000644 0001762 0000144 00000001211 14330031606 014121 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/post.R
\name{make.post}
\alias{make.post}
\title{Creates a \code{post} argument}
\usage{
make.post(data)
}
\arguments{
\item{data}{A data frame or a matrix containing the incomplete data. Missing
values are coded as \code{NA}.}
}
\value{
Character vector of \code{ncol(data)} element
}
\description{
This helper function creates a valid \code{post} vector. The
\code{post} vector is an argument to the \code{mice} function that
specifies post-processing for a variable after each iteration of imputation.
}
\examples{
make.post(nhanes2)
}
\seealso{
\code{\link{mice}}
}
mice/man/windspeed.Rd 0000644 0001762 0000144 00000002626 14330031606 014215 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/windspeed.R
\docType{data}
\name{windspeed}
\alias{windspeed}
\title{Subset of Irish wind speed data}
\format{
A data frame with 433 rows and 6 columns containing the daily average
wind speeds within the period 1961-1978 at meteorological stations in the
Republic of Ireland. The data are a random sample from a larger data set.
\describe{
\item{RochePt}{Roche Point}
\item{Rosslare}{Rosslare}
\item{Shannon}{Shannon}
\item{Dublin}{Dublin}
\item{Clones}{Clones}
\item{MalinHead}{Malin Head} }
}
\description{
Subset of Irish wind speed data
}
\details{
The original data set is much larger and was analyzed in detail by Haslett
and Raftery (1989). Van Buuren et al (2006) used this subset to investigate
the influence of extreme MAR mechanisms on the quality of imputation.
}
\examples{
windspeed[1:3, ]
}
\references{
Haslett, J. and Raftery, A. E. (1989). \emph{Space-time
Modeling with Long-memory Dependence: Assessing Ireland's Wind Power
Resource (with Discussion)}. Applied Statistics 38, 1-50.
\url{http://lib.stat.cmu.edu/datasets/wind.desc} and
\url{http://lib.stat.cmu.edu/datasets/wind.data}
van Buuren, S., Brand, J.P.L., Groothuis-Oudshoorn C.G.M., Rubin, D.B. (2006)
Fully conditional specification in multivariate imputation. \emph{Journal of
Statistical Computation and Simulation}, \bold{76}, 12, 1049--1064.
}
\keyword{datasets}
mice/man/brandsma.Rd 0000644 0001762 0000144 00000004267 14330031606 014025 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/brandsma.R
\docType{data}
\name{brandsma}
\alias{brandsma}
\title{Brandsma school data used Snijders and Bosker (2012)}
\format{
\code{brandsma} is a data frame with 4106 rows and 14 columns:
\describe{
\item{\code{sch}}{School number}
\item{\code{pup}}{Pupil ID}
\item{\code{iqv}}{IQ verbal}
\item{\code{iqp}}{IQ performal}
\item{\code{sex}}{Sex of pupil}
\item{\code{ses}}{SES score of pupil}
\item{\code{min}}{Minority member 0/1}
\item{\code{rpg}}{Number of repeated groups, 0, 1, 2}
\item{\code{lpr}}{language score PRE}
\item{\code{lpo}}{language score POST}
\item{\code{apr}}{Arithmetic score PRE}
\item{\code{apo}}{Arithmetic score POST}
\item{\code{den}}{Denomination classification 1-4 - at school level}
\item{\code{ssi}}{School SES indicator - at school level}
}
}
\source{
Constructed from \code{MLbook_2nded_total_4106-99.sav} from
\url{https://www.stats.ox.ac.uk/~snijders/mlbook.htm} by function
\code{data-raw/R/brandsma.R}
}
\description{
Dataset with raw data from Snijders and Bosker (2012) containing
data from 4106 pupils attending 216 schools. This dataset
includes all pupils and schools with missing data.
}
\note{
This dataset is constructed from the raw data. There are
a few differences with the data set used in Chapter 4 and 5
of Snijders and Bosker:
\enumerate{
\item All schools are included, including the five school with
missing values on \code{langpost}.
\item Missing \code{denomina} codes are left as missing.
\item Aggregates are undefined in the presence of missing data
in the underlying values.
Variables \code{ses}, \code{iqv} and \code{iqp} are in their
original scale, and not globally centered.
No aggregate variables at the school level are included.
\item There is a wider selection of original variables. Note
however that the source data contain an even wider set of
variables.
}
}
\references{
Brandsma, HP and Knuver, JWM (1989), Effects of school and
classroom characteristics on pupil progress in language and arithmetic.
International Journal of Educational Research, 13(7), 777 - 788.
Snijders, TAB and Bosker RJ (2012). Multilevel Analysis, 2nd Ed. Sage,
Los Angeles, 2012.
}
\keyword{datasets}
mice/man/ampute.discrete.Rd 0000644 0001762 0000144 00000004165 14330031606 015327 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/ampute.discrete.R
\name{ampute.discrete}
\alias{ampute.discrete}
\title{Multivariate amputation based on discrete probability functions}
\usage{
ampute.discrete(P, scores, prop, odds)
}
\arguments{
\item{P}{A vector containing the pattern numbers of candidates.
For each case, a value between 1 and #patterns is given. For example, a
case with value 2 is candidate for missing data pattern 2.}
\item{scores}{A list containing vectors with the candidates's weighted sum scores,
the result of an underlying function in \code{\link{ampute}}.}
\item{prop}{A scalar specifying the proportion of missingness. Should be a value
between 0 and 1. Default is a missingness proportion of 0.5.}
\item{odds}{A matrix where #patterns defines the #rows. Each row should contain
the odds of being missing for the corresponding pattern. The amount of odds values
defines in how many quantiles the sum scores will be divided. The values are
relative probabilities: a quantile with odds value 4 will have a probability of
being missing that is four times higher than a quantile with odds 1. The
#quantiles may differ between the patterns, specify NA for cells remaining empty.
Default is 4 quantiles with odds values 1, 2, 3 and 4, the result of
\code{\link{ampute.default.odds}}.}
}
\value{
A list containing vectors with \code{0} if a case should be made missing
and \code{1} if a case should remain complete. The first vector refers to the
first pattern, the second vector to the second pattern, etcetera.
}
\description{
This function creates a missing data indicator for each pattern. Odds probabilities
(Brand, 1999, pp. 110-113) will be induced on the weighted sum scores, calculated earlier
in the multivariate amputation function \code{\link{ampute}}.
}
\references{
Brand, J.P.L. (1999). \emph{Development, implementation and
evaluation of multiple imputation strategies for the statistical analysis of
incomplete data sets.} Dissertation. Rotterdam: Erasmus University.
}
\seealso{
\code{\link{ampute}}, \code{\link{ampute.default.odds}}
}
\author{
Rianne Schouten, 2016
}
\keyword{internal}
mice/man/mice.theme.Rd 0000644 0001762 0000144 00000001565 14330031606 014252 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.theme.R
\name{mice.theme}
\alias{mice.theme}
\title{Set the theme for the plotting Trellis functions}
\usage{
mice.theme(transparent = TRUE, alpha.fill = 0.3)
}
\arguments{
\item{transparent}{A logical indicating whether alpha-transparency is
allowed. The default is \code{TRUE}.}
\item{alpha.fill}{A numerical values between 0 and 1 that indicates the
default alpha value for fills.}
}
\value{
\code{mice.theme()} returns a named list that can be used as a theme in the functions in
\pkg{lattice}. By default, the \code{mice.theme()} function sets
\code{transparent <- TRUE} if the current device \code{.Device} supports
semi-transparent colors.
}
\description{
The \code{mice.theme()} function sets default choices for
Trellis plots that are built into \pkg{mice}.
}
\author{
Stef van Buuren 2011
}
mice/man/walking.Rd 0000644 0001762 0000144 00000004467 14330031606 013674 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/walking.R
\docType{data}
\name{walking}
\alias{walking}
\title{Walking disability data}
\format{
A data frame with 890 rows on the following 5 variables:
\describe{
\item{sex}{Sex of respondent (factor)}
\item{age}{Age of respondent}
\item{YA}{Item administered in samples A and E (factor)}
\item{YB}{Item administered in samples B and E (factor)}
\item{src}{Source: Sample A, B or E (factor)}
}
}
\description{
Two items YA and YB measuring walking disability in samples A, B and E.
}
\details{
Example dataset to demonstrate imputation of two items (YA and YB). Item YA
is administered to sample A and sample E, item YB is administered to sample B
and sample E, so sample E acts as a bridge study. Imputation using a bridge
study is better than simple equating or than imputation under independence.
Item YA corresponds to the HAQ8 item, and item YB corresponds to the GAR9
items from Van Buuren et al (2005). Sample E (as well as sample B) is the
Euridiss study (n=292), sample A is the ERGOPLUS study (n=306).
See Van Buuren (2018) section 9.4 for more details on the imputation
methodology.
}
\examples{
md.pattern(walking)
micemill <- function(n) {
for (i in 1:n) {
imp <<- mice.mids(imp) # global assignment
cors <- with(imp, cor(as.numeric(YA),
as.numeric(YB),
method = "kendall"
))
tau <<- rbind(tau, getfit(cors, s = TRUE)) # global assignment
}
}
plotit <- function() {
matplot(
x = 1:nrow(tau), y = tau,
ylab = expression(paste("Kendall's ", tau)),
xlab = "Iteration", type = "l", lwd = 1,
lty = 1:10, col = "black"
)
}
tau <- NULL
imp <- mice(walking, max = 0, m = 10, seed = 92786)
pred <- imp$pred
pred[, c("src", "age", "sex")] <- 0
imp <- mice(walking, max = 0, m = 3, seed = 92786, pred = pred)
micemill(5)
plotit()
### to get figure 9.8 van Buuren (2018) use m=10 and micemill(20)
}
\references{
van Buuren, S., Eyres, S., Tennant, A., Hopman-Rock, M. (2005).
Improving comparability of existing data by Response Conversion.
\emph{Journal of Official Statistics}, \bold{21}(1), 53-72.
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-codingsystems.html#sec:impbridge}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\keyword{datasets}
mice/man/mice.impute.midastouch.Rd 0000644 0001762 0000144 00000013736 14330031647 016622 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.midastouch.R
\name{mice.impute.midastouch}
\alias{mice.impute.midastouch}
\title{Imputation by predictive mean matching with distance aided donor selection}
\usage{
mice.impute.midastouch(
y,
ry,
x,
wy = NULL,
ridge = 1e-05,
midas.kappa = NULL,
outout = TRUE,
neff = NULL,
debug = NULL,
...
)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{ridge}{The ridge penalty used in \code{.norm.draw()} to prevent
problems with multicollinearity. The default is \code{ridge = 1e-05},
which means that 0.01 percent of the diagonal is added to the cross-product.
Larger ridges may result in more biased estimates. For highly noisy data
(e.g. many junk variables), set \code{ridge = 1e-06} or even lower to
reduce bias. For highly collinear data, set \code{ridge = 1e-04} or higher.}
\item{midas.kappa}{Scalar. If \code{NULL} (default) then the
optimal \code{kappa} gets selected automatically. Alternatively, the user
may specify a scalar. Siddique and Belin 2008 find \code{midas.kappa = 3}
to be sensible.}
\item{outout}{Logical. If \code{TRUE} (default) one model is estimated
for each donor (leave-one-out principle). For speedup choose
\code{outout = FALSE}, which estimates one model for all observations
leading to in-sample predictions for the donors and out-of-sample
predictions for the recipients. Mind the inappropriateness, though.}
\item{neff}{FOR EXPERTS. Null or character string. The name of an existing
environment in which the effective sample size of the donors for each
loop (CE iterations times multiple imputations) is supposed to be written.
The effective sample size is necessary to compute the correction for the
total variance as originally suggested by Parzen, Lipsitz and
Fitzmaurice 2005. The objectname is \code{midastouch.neff}.}
\item{debug}{FOR EXPERTS. Null or character string. The name of an existing
environment in which the input is supposed to be written. The objectname
is \code{midastouch.inputlist}.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of
length \code{sum(wy)}
}
\description{
Imputes univariate missing data using predictive mean matching.
}
\details{
Imputation of \code{y} by predictive mean matching, based on
Rubin (1987, p. 168, formulas a and b) and Siddique and Belin 2008.
The procedure is as follows:
\enumerate{
\item Draw a bootstrap sample from the donor pool.
\item Estimate a beta matrix on the bootstrap sample by the leave one out principle.
\item Compute type II predicted values for \code{yobs} (nobs x 1) and \code{ymis} (nmis x nobs).
\item Calculate the distance between all \code{yobs} and the corresponding \code{ymis}.
\item Convert the distances in drawing probabilities.
\item For each recipient draw a donor from the entire pool while considering the probabilities from the model.
\item Take its observed value in \code{y} as the imputation.
}
}
\examples{
# do default multiple imputation on a numeric matrix
imp <- mice(nhanes, method = "midastouch")
imp
# list the actual imputations for BMI
imp$imp$bmi
# first completed data matrix
complete(imp)
# imputation on mixed data with a different method per column
mice(nhanes2, method = c("sample", "midastouch", "logreg", "norm"))
}
\references{
Gaffert, P., Meinfelder, F., Bosch V. (2015) Towards an MI-proper
Predictive Mean Matching, Discussion Paper.
\url{https://www.uni-bamberg.de/fileadmin/uni/fakultaeten/sowi_lehrstuehle/statistik/Personen/Dateien_Florian/properPMM.pdf}
Little, R.J.A. (1988), Missing data adjustments in large
surveys (with discussion), Journal of Business Economics and
Statistics, 6, 287--301.
Parzen, M., Lipsitz, S. R., Fitzmaurice, G. M. (2005), A note on reducing
the bias of the approximate Bayesian bootstrap imputation variance estimator.
Biometrika \bold{92}, 4, 971--974.
Rubin, D.B. (1987), Multiple imputation for nonresponse in surveys. New York: Wiley.
Siddique, J., Belin, T.R. (2008), Multiple imputation using an iterative
hot-deck with distance-based donor selection. Statistics in medicine,
\bold{27}, 1, 83--102
Van Buuren, S., Brand, J.P.L., Groothuis-Oudshoorn C.G.M., Rubin, D.B. (2006),
Fully conditional specification in multivariate imputation.
\emph{Journal of Statistical Computation and Simulation}, \bold{76}, 12,
1049--1064.
Van Buuren, S., Groothuis-Oudshoorn, K. (2011), \code{mice}: Multivariate
Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}, 3, 1--67. \doi{10.18637/jss.v045.i03}
}
\seealso{
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Philipp Gaffert, Florian Meinfelder, Volker Bosch 2015
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/boys.Rd 0000644 0001762 0000144 00000005447 14330031606 013213 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/boys.R
\docType{data}
\name{boys}
\alias{boys}
\title{Growth of Dutch boys}
\format{
A data frame with 748 rows on the following 9 variables: \describe{
\item{age}{Decimal age (0-21 years)}
\item{hgt}{Height (cm)}
\item{wgt}{Weight (kg)}
\item{bmi}{Body mass index}
\item{hc}{Head circumference (cm)}
\item{gen}{Genital Tanner stage (G1-G5)}
\item{phb}{Pubic hair (Tanner P1-P6)}
\item{tv}{Testicular volume (ml)}
\item{reg}{Region (north, east, west, south, city)} }
}
\source{
Fredriks, A.M,, van Buuren, S., Burgmeijer, R.J., Meulmeester JF,
Beuker, R.J., Brugman, E., Roede, M.J., Verloove-Vanhorick, S.P., Wit, J.M.
(2000) Continuing positive secular growth change in The Netherlands
1955-1997. \emph{Pediatric Research}, \bold{47}, 316-323.
Fredriks, A.M., van Buuren, S., Wit, J.M., Verloove-Vanhorick, S.P. (2000).
Body index measurements in 1996-7 compared with 1980. \emph{Archives of
Disease in Childhood}, \bold{82}, 107-112.
}
\description{
Height, weight, head circumference and puberty of 748 Dutch boys.
}
\details{
Random sample of 10\% from the cross-sectional data used to construct the
Dutch growth references 1997. Variables \code{gen} and \code{phb} are ordered
factors. \code{reg} is a factor.
}
\examples{
# create two imputed data sets
imp <- mice(boys, m = 1, maxit = 2)
z <- complete(imp, 1)
# create imputations for age <8yrs
plot(z$age, z$gen,
col = mdc(1:2)[1 + is.na(boys$gen)],
xlab = "Age (years)", ylab = "Tanner Stage Genital"
)
# figure to show that the default imputation method does not impute BMI
# consistently
plot(z$bmi, z$wgt / (z$hgt / 100)^2,
col = mdc(1:2)[1 + is.na(boys$bmi)],
xlab = "Imputed BMI", ylab = "Calculated BMI"
)
# also, BMI distributions are somewhat different
oldpar <- par(mfrow = c(1, 2))
MASS::truehist(z$bmi[!is.na(boys$bmi)],
h = 1, xlim = c(10, 30), ymax = 0.25,
col = mdc(1), xlab = "BMI observed"
)
MASS::truehist(z$bmi[is.na(boys$bmi)],
h = 1, xlim = c(10, 30), ymax = 0.25,
col = mdc(2), xlab = "BMI imputed"
)
par(oldpar)
# repair the inconsistency problem by passive imputation
meth <- imp$meth
meth["bmi"] <- "~I(wgt/(hgt/100)^2)"
pred <- imp$predictorMatrix
pred["hgt", "bmi"] <- 0
pred["wgt", "bmi"] <- 0
imp2 <- mice(boys, m = 1, maxit = 2, meth = meth, pred = pred)
z2 <- complete(imp2, 1)
# show that new imputations are consistent
plot(z2$bmi, z2$wgt / (z2$hgt / 100)^2,
col = mdc(1:2)[1 + is.na(boys$bmi)],
ylab = "Calculated BMI"
)
# and compare distributions
oldpar <- par(mfrow = c(1, 2))
MASS::truehist(z2$bmi[!is.na(boys$bmi)],
h = 1, xlim = c(10, 30), ymax = 0.25, col = mdc(1),
xlab = "BMI observed"
)
MASS::truehist(z2$bmi[is.na(boys$bmi)],
h = 1, xlim = c(10, 30), ymax = 0.25, col = mdc(2),
xlab = "BMI imputed"
)
par(oldpar)
}
\keyword{datasets}
mice/man/mice.impute.norm.predict.Rd 0000644 0001762 0000144 00000006356 14330031647 017066 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.norm.predict.R
\name{mice.impute.norm.predict}
\alias{mice.impute.norm.predict}
\alias{norm.predict}
\title{Imputation by linear regression through prediction}
\usage{
mice.impute.norm.predict(y, ry, x, wy = NULL, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes the "best value" according to the linear regression model, also
known as \emph{regression imputation}.
}
\details{
Calculates regression weights from the observed data and returns predicted
values to as imputations. This
method is known as \emph{regression imputation}.
}
\section{Warning}{
THIS METHOD SHOULD NOT BE USED FOR DATA ANALYSIS.
This method is seductive because it imputes the most
likely value according to the model. However, it ignores the uncertainty
of the missing values and artificially
amplifies the relations between the columns of the data. Application of
richer models having more parameters does not help to evade these issues.
Stochastic regression methods, like \code{\link{mice.impute.pmm}} or
\code{\link{mice.impute.norm}}, are generally preferred.
At best, prediction can give reasonable estimates of the mean, especially
if normality assumptions are plausible. See Little and Rubin (2002, p. 62-64)
or Van Buuren (2012, p. 11-13, p. 45-46) for a discussion of this method.
}
\references{
Little, R.J.A. and Rubin, D.B. (2002). Statistical Analysis with Missing
Data. New York: John Wiley and Sons.
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-linearnormal.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\seealso{
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Gerko Vink, Stef van Buuren, 2018
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/make.blots.Rd 0000644 0001762 0000144 00000001631 14330031606 014265 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/blots.R
\name{make.blots}
\alias{make.blots}
\title{Creates a \code{blots} argument}
\usage{
make.blots(data, blocks = make.blocks(data))
}
\arguments{
\item{data}{A \code{data.frame} with the source data}
\item{blocks}{An optional specification for blocks of variables in
the rows. The default assigns each variable in its own block.}
}
\value{
A matrix
}
\description{
This helper function creates a valid \code{blots} object. The
\code{blots} object is an argument to the \code{mice} function.
The name \code{blots} is a contraction of blocks-dots.
Through \code{blots}, the user can specify any additional
arguments that are specifically passed down to the lowest level
imputation function.
}
\examples{
make.predictorMatrix(nhanes)
make.blots(nhanes, blocks = name.blocks(c("age", "hyp"), "xxx"))
}
\seealso{
\code{\link{make.blocks}}
}
mice/man/potthoffroy.Rd 0000644 0001762 0000144 00000004053 14330031606 014612 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/potthoffroy.R
\docType{data}
\name{potthoffroy}
\alias{potthoffroy}
\title{Potthoff-Roy data}
\format{
\code{tbs} is a data frame with 27 rows and 6 columns:
\describe{
\item{id}{Person number}
\item{sex}{Sex M/F}
\item{d8}{Distance at age 8 years}
\item{d10}{Distance at age 10 years}
\item{d12}{Distance at age 12 years}
\item{d14}{Distance at age 14 years}
}
}
\source{
Potthoff, R. F., Roy, S. N. (1964). A generalized multivariate
analysis of variance model usefully especially for growth curve problems.
\emph{Biometrika}, \emph{51}(3), 313-326.
Little, R. J. A., Rubin, D. B. (1987). \emph{Statistical Analysis with
Missing Data.} New York: John Wiley & Sons.
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/ex-ch-longitudinal.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
}
\description{
Data from Potthoff-Roy (1964) with repeated measures on dental fissures.
}
\details{
This data set is the famous Potthoff-Roy data, used to demonstrate MANOVA on
repeated measure data. Potthoff and Roy (1964) published classic data on a
study in 16 boys and 11 girls, who at ages 8, 10, 12, and 14 had the distance
(mm) from the center of the pituitary gland to the pteryomaxillary fissure
measured. Changes in pituitary-pteryomaxillary distances during growth is
important in orthodontic therapy. The goals of the study were to describe the
distance in boys and girls as simple functions of age, and then to compare
the functions for boys and girls. The data have been reanalyzed by many
authors including Jennrich and Schluchter (1986), Little and Rubin (1987),
Pinheiro and Bates (2000), Verbeke and Molenberghs (2000) and Molenberghs and
Kenward (2007). See Chapter 9 of Van Buuren (2012) for a challenging
exercise using these data.
}
\examples{
### create missing values at age 10 as in Little and Rubin (1987)
phr <- potthoffroy
idmis <- c(3, 6, 9, 10, 13, 16, 23, 24, 27)
phr[idmis, 4] <- NA
phr
md.pattern(phr)
}
\keyword{datasets}
mice/man/mice.impute.quadratic.Rd 0000644 0001762 0000144 00000010765 14436637251 016450 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.quadratic.R
\name{mice.impute.quadratic}
\alias{mice.impute.quadratic}
\alias{quadratic}
\title{Imputation of quadratic terms}
\usage{
mice.impute.quadratic(y, ry, x, wy = NULL, quad.outcome = NULL, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{quad.outcome}{The name of the outcome in the quadratic analysis as a
character string. For example, if the substantive model of interest is
\code{y ~ x + xx}, then \code{"y"} would be the \code{quad.outcome}}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes incomplete variable that appears as both
main effect and quadratic effect in the complete-data model.
}
\details{
This function implements the "polynomial combination" method.
First, the polynomial
combination \eqn{Z = Y \beta_1 + Y^2 \beta_2} is formed.
\eqn{Z} is imputed by
predictive mean matching, followed by a decomposition of the imputed
data \eqn{Z}
into components \eqn{Y} and \eqn{Y^2}.
See Van Buuren (2012, pp. 139-141) and Vink
et al (2012) for more details. The method ensures that 1) the imputed data
for \eqn{Y} and \eqn{Y^2} are mutually consistent, and 2) that provides unbiased
estimates of the regression weights in a complete-data linear regression that
use both \eqn{Y} and \eqn{Y^2}.
}
\note{
There are two situations to consider. If only the linear term \code{Y}
is present in the data, calculate the quadratic term \code{YY} after
imputation. If both the linear term \code{Y} and the the quadratic term
\code{YY} are variables in the data, then first impute \code{Y} by calling
\code{mice.impute.quadratic()} on \code{Y}, and then impute \code{YY} by
passive imputation as \code{meth["YY"] <- "~I(Y^2)"}. See example section
for details. Generally, we would like \code{YY} to be present in the data if
we need to preserve quadratic relations between \code{YY} and any third
variables in the multivariate incomplete data that we might wish to impute.
}
\examples{
# Create Data
B1 <- .5
B2 <- .5
X <- rnorm(1000)
XX <- X^2
e <- rnorm(1000, 0, 1)
Y <- B1 * X + B2 * XX + e
dat <- data.frame(x = X, xx = XX, y = Y)
# Impose 25 percent MCAR Missingness
dat[0 == rbinom(1000, 1, 1 - .25), 1:2] <- NA
# Prepare data for imputation
ini <- mice(dat, maxit = 0)
meth <- c("quadratic", "~I(x^2)", "")
pred <- ini$pred
pred[, "xx"] <- 0
# Impute data
imp <- mice(dat, meth = meth, pred = pred, quad.outcome = "y")
# Pool results
pool(with(imp, lm(y ~ x + xx)))
# Plot results
stripplot(imp)
plot(dat$x, dat$xx, col = mdc(1), xlab = "x", ylab = "xx")
cmp <- complete(imp)
points(cmp$x[is.na(dat$x)], cmp$xx[is.na(dat$x)], col = mdc(2))
}
\seealso{
\code{\link{mice.impute.pmm}}
Van Buuren, S. (2018).
\href{https://stefvanbuuren.name/fimd/sec-knowledge.html#sec:quadratic}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
Chapman & Hall/CRC. Boca Raton, FL.
Vink, G., van Buuren, S. (2013). Multiple Imputation of Squared Terms.
\emph{Sociological Methods & Research}, 42:598-607.
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Mingyang Cai and Gerko Vink
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/mice.impute.jomoImpute.Rd 0000644 0001762 0000144 00000006322 14436133175 016611 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.jomoImpute.R
\name{mice.impute.jomoImpute}
\alias{mice.impute.jomoImpute}
\title{Multivariate multilevel imputation using \code{jomo}}
\usage{
mice.impute.jomoImpute(
data,
formula,
type,
m = 1,
silent = TRUE,
format = "imputes",
...
)
}
\arguments{
\item{data}{A data frame containing incomplete and auxiliary variables,
the cluster indicator variable, and any other variables that should be
present in the imputed datasets.}
\item{formula}{A formula specifying the role of each variable
in the imputation model. The basic model is constructed
by \code{model.matrix}, thus allowing to include derived variables
in the imputation model using \code{I()}. See
\code{\link[mitml]{jomoImpute}}.}
\item{type}{An integer vector specifying the role of each variable
in the imputation model (see \code{\link[mitml]{jomoImpute}})}
\item{m}{The number of imputed data sets to generate. Default is 10.}
\item{silent}{(optional) Logical flag indicating if console output should be suppressed. Default is \code{FALSE}.}
\item{format}{A character vector specifying the type of object that should
be returned. The default is \code{format = "list"}. No other formats are
currently supported.}
\item{...}{Other named arguments: \code{n.burn}, \code{n.iter},
\code{group}, \code{prior}, \code{silent} and others.}
}
\value{
A list of imputations for all incomplete variables in the model,
that can be stored in the the \code{imp} component of the \code{mids}
object.
}
\description{
This function is a wrapper around the \code{jomoImpute} function
from the \code{mitml} package so that it can be called to
impute blocks of variables in \code{mice}. The \code{mitml::jomoImpute}
function provides an interface to the \code{jomo} package for
multiple imputation of multilevel data
\url{https://CRAN.R-project.org/package=jomo}.
Imputations can be generated using \code{type} or \code{formula},
which offer different options for model specification.
}
\note{
The number of imputations \code{m} is set to 1, and the function
is called \code{m} times so that it fits within the \code{mice}
iteration scheme.
This is a multivariate imputation function using a joint model.
}
\examples{
\dontrun{
# Note: Requires mitml 0.3-5.7
blocks <- list(c("bmi", "chl", "hyp"), "age")
method <- c("jomoImpute", "pmm")
ini <- mice(nhanes, blocks = blocks, method = method, maxit = 0)
pred <- ini$pred
pred["B1", "hyp"] <- -2
imp <- mice(nhanes, blocks = blocks, method = method, pred = pred, maxit = 1)
}
}
\references{
Grund S, Luedtke O, Robitzsch A (2016). Multiple
Imputation of Multilevel Missing Data: An Introduction to the R
Package \code{pan}. SAGE Open.
Quartagno M and Carpenter JR (2015).
Multiple imputation for IPD meta-analysis: allowing for heterogeneity
and studies with missing covariates. Statistics in Medicine,
35:2938-2954, 2015.
}
\seealso{
\code{\link[mitml]{jomoImpute}}
Other multivariate-2l:
\code{\link{mice.impute.panImpute}()}
}
\author{
Stef van Buuren, 2018, building on work of Simon Grund,
Alexander Robitzsch and Oliver Luedtke (authors of \code{mitml} package)
and Quartagno and Carpenter (authors of \code{jomo} package).
}
\concept{multivariate-2l}
\keyword{datagen}
mice/man/ifdo.Rd 0000644 0001762 0000144 00000000652 13666252075 013171 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/auxiliary.R
\name{ifdo}
\alias{ifdo}
\title{Conditional imputation helper}
\usage{
ifdo(cond, action)
}
\arguments{
\item{cond}{a condition}
\item{action}{the action to do}
}
\value{
Currently returns an error message.
}
\description{
Sorry, the \code{ifdo()} function is not yet implemented.
}
\author{
Stef van Buuren, 2012
}
\keyword{internal}
mice/man/ampute.default.odds.Rd 0000644 0001762 0000144 00000001543 14330031606 016076 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/ampute.default.R
\name{ampute.default.odds}
\alias{ampute.default.odds}
\title{Default \code{odds} in \code{ampute()}}
\usage{
ampute.default.odds(patterns)
}
\arguments{
\item{patterns}{A matrix of size #patterns by #variables where 0 indicates a
variable should have missing values and 1 indicates a variable should remain
complete. Could be the result of \code{\link{ampute.default.patterns}}.}
}
\value{
A matrix where #rows equals #patterns. Default is 4 quantiles with odds
values 1, 2, 3 and 4, for each pattern, imitating a RIGHT type of missingness.
}
\description{
Defines the default odds matrix for the multivariate amputation function
\code{ampute}.
}
\seealso{
\code{\link{ampute}}, \code{\link{ampute.default.patterns}}
}
\author{
Rianne Schouten, 2016
}
\keyword{internal}
mice/man/print.Rd 0000644 0001762 0000144 00000001531 13666252075 013401 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/print.R
\name{print.mids}
\alias{print.mids}
\alias{print.mira}
\alias{print.mice.anova}
\alias{print.mice.anova.summary}
\title{Print a \code{mids} object}
\usage{
\method{print}{mids}(x, ...)
\method{print}{mira}(x, ...)
\method{print}{mice.anova}(x, ...)
\method{print}{mice.anova.summary}(x, ...)
}
\arguments{
\item{x}{Object of class \code{mids}, \code{mira} or \code{mipo}}
\item{...}{Other parameters passed down to \code{print.default()}}
}
\value{
\code{NULL}
\code{NULL}
\code{NULL}
\code{NULL}
}
\description{
Print a \code{mids} object
Print a \code{mira} object
Print a \code{mice.anova} object
Print a \code{summary.mice.anova} object
}
\seealso{
\code{\link[=mids-class]{mids}}
\code{\link[=mira-class]{mira}}
\code{\link{mipo}}
\code{\link{mipo}}
}
mice/man/mids2mplus.Rd 0000644 0001762 0000144 00000003176 13666252075 014353 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mids2mplus.R
\name{mids2mplus}
\alias{mids2mplus}
\title{Export \code{mids} object to Mplus}
\usage{
mids2mplus(
imp,
file.prefix = "imp",
path = getwd(),
sep = "\\t",
dec = ".",
silent = FALSE
)
}
\arguments{
\item{imp}{The \code{imp} argument is an object of class \code{mids},
typically produced by the \code{mice()} function.}
\item{file.prefix}{A character string describing the prefix of the output
data files.}
\item{path}{A character string containing the path of the output file. By
default, files are written to the current \code{R} working directory.}
\item{sep}{The separator between the data fields.}
\item{dec}{The decimal separator for numerical data.}
\item{silent}{A logical flag stating whether the names of the files should be
printed.}
}
\value{
The return value is \code{NULL}.
}
\description{
Converts a \code{mids} object into a format recognized by Mplus, and writes
the data and the Mplus input files
}
\details{
This function automates most of the work needed to export a \code{mids}
object to \code{Mplus}. The function writes the multiple imputation datasets,
the file that contains the names of the multiple imputation data sets and an
\code{Mplus} input file. The \code{Mplus} input file has the proper file
names, so in principle it should run and read the data without alteration.
\code{Mplus} will recognize the data set as a multiply imputed data set, and
do automatic pooling in procedures where that is supported.
}
\seealso{
\code{\link[=mids-class]{mids}}, \code{\link{mids2spss}}
}
\author{
Gerko Vink, 2011.
}
\keyword{manip}
mice/man/ampute.mcar.Rd 0000644 0001762 0000144 00000003014 14330031606 014437 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/ampute.mcar.R
\name{ampute.mcar}
\alias{ampute.mcar}
\title{Multivariate amputation under a MCAR mechanism}
\usage{
ampute.mcar(P, patterns, prop)
}
\arguments{
\item{P}{A vector containing the pattern numbers of the cases' candidates.
For each case, a value between 1 and #patterns is given. For example, a
case with value 2 is candidate for missing data pattern 2.}
\item{patterns}{A matrix of size #patterns by #variables where \code{0} indicates
a variable should have missing values and \code{1} indicates a variable should
remain complete. The user may specify as many patterns as desired. One pattern
(a vector) is also possible. Could be the result of \code{\link{ampute.default.patterns}},
default will be a square matrix of size #variables where each pattern has missingness
on one variable only.}
\item{prop}{A scalar specifying the proportion of missingness. Should be a value
between 0 and 1. Default is a missingness proportion of 0.5.}
}
\value{
A list containing vectors with \code{0} if a case should be made missing
and \code{1} if a case should remain complete. The first vector refers to the
first pattern, the second vector to the second pattern, etcetera.
}
\description{
This function creates a missing data indicator for each pattern, based on a MCAR
missingness mechanism. The function is used in the multivariate amputation function
\code{\link{ampute}}.
}
\seealso{
\code{\link{ampute}}
}
\author{
Rianne Schouten, 2016
}
\keyword{internal}
mice/man/make.where.Rd 0000644 0001762 0000144 00000002171 14334522175 014266 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/where.R
\name{make.where}
\alias{make.where}
\title{Creates a \code{where} argument}
\usage{
make.where(data, keyword = c("missing", "all", "none", "observed"))
}
\arguments{
\item{data}{A \code{data.frame} with the source data}
\item{keyword}{An optional keyword, one of \code{"missing"} (missing
values are imputed), \code{"observed"} (observed values are imputed),
\code{"all"} and \code{"none"}. The default
is \code{keyword = "missing"}}
}
\value{
A matrix with logical
}
\description{
This helper function creates a valid \code{where} matrix. The
\code{where} matrix is an argument to the \code{mice} function.
It has the same size as \code{data} and specifies which values
are to be imputed (\code{TRUE}) or nor (\code{FALSE}).
}
\examples{
head(make.where(nhanes), 3)
# create & analyse synthetic data
where <- make.where(nhanes2, "all")
imp <- mice(nhanes2,
m = 10, where = where,
print = FALSE, seed = 123
)
fit <- with(imp, lm(chl ~ bmi + age + hyp))
summary(pool.syn(fit))
}
\seealso{
\code{\link{make.blocks}}, \code{\link{make.predictorMatrix}}
}
mice/man/bwplot.mads.Rd 0000644 0001762 0000144 00000004142 14330031606 014460 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/bwplot.mads.R
\name{bwplot.mads}
\alias{bwplot.mads}
\title{Box-and-whisker plot of amputed and non-amputed data}
\usage{
\method{bwplot}{mads}(
x,
data,
which.pat = NULL,
standardized = TRUE,
descriptives = TRUE,
layout = NULL,
...
)
}
\arguments{
\item{x}{A \code{mads} (\code{\link{mads-class}}) object, typically created by
\code{\link{ampute}}.}
\item{data}{A string or vector of variable names that needs to be plotted. As
a default, all variables will be plotted.}
\item{which.pat}{A scalar or vector indicating which patterns need to be plotted.
As a default, all patterns are plotted.}
\item{standardized}{Logical. Whether the box-and-whisker plots need to be created
from standardized data or not. Default is TRUE.}
\item{descriptives}{Logical. Whether the mean, variance and n of the variables
need to be printed. This is useful to examine the effect of the amputation.
Default is TRUE.}
\item{layout}{A vector of two values indicating how the boxplots of one pattern
should be divided over the plot. For example, \code{c(2, 3)} indicates that the
boxplots of six variables need to be placed on 3 rows and 2 columns. Default
is 1 row and an amount of columns equal to #variables. Note that for more than
6 variables, multiple plots will be created automatically.}
\item{\dots}{Not used, but for consistency with generic}
}
\value{
A list containing the box-and-whisker plots. Note that a new pattern
will always be shown in a new plot.
}
\description{
Plotting method to investigate the relation between the data variables and
the amputed data. The function shows how the amputed values are related
to the variable values.
}
\note{
The \code{mads} object contains all the information you need to
make any desired plots. Check \code{\link{mads-class}} or the vignette \emph{Multivariate
Amputation using Ampute} to understand the contents of class object \code{mads}.
}
\seealso{
\code{\link{ampute}}, \code{\link{bwplot}}, \code{\link{Lattice}} for
an overview of the package, \code{\link{mads-class}}
}
\author{
Rianne Schouten, 2016
}
mice/man/mice.impute.lda.Rd 0000644 0001762 0000144 00000007127 14330031647 015217 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.lda.R
\name{mice.impute.lda}
\alias{mice.impute.lda}
\title{Imputation by linear discriminant analysis}
\usage{
mice.impute.lda(y, ry, x, wy = NULL, ...)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{...}{Other named arguments. Not used.}
}
\value{
Vector with imputed data, of type factor, and of length
\code{sum(wy)}
}
\description{
Imputes univariate missing data using linear discriminant analysis
}
\details{
Imputation of categorical response variables by linear discriminant analysis.
This function uses the Venables/Ripley functions \code{lda()} and
\code{predict.lda()} to compute posterior probabilities for each incomplete
case, and draws the imputations from this posterior.
This function can be called from within the Gibbs sampler by specifying
\code{"lda"} in the \code{method} argument of \code{mice()}. This method is usually
faster and uses fewer resources than calling the function, but the statistical
properties may not be as good (Brand, 1999).
\code{\link{mice.impute.polyreg}}.
}
\section{Warning}{
The function does not incorporate the variability of the
discriminant weight, so it is not 'proper' in the sense of Rubin. For small
samples and rare categories in the \code{y}, variability of the imputed data
could therefore be underestimated.
Added: SvB June 2009 Tried to include bootstrap, but disabled since
bootstrapping may easily lead to constant variables within groups.
}
\references{
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
Brand, J.P.L. (1999). Development, Implementation and Evaluation of Multiple
Imputation Strategies for the Statistical Analysis of Incomplete Data Sets.
Ph.D. Thesis, TNO Prevention and Health/Erasmus University Rotterdam. ISBN
90-74479-08-1.
Venables, W.N. & Ripley, B.D. (1997). Modern applied statistics with S-PLUS
(2nd ed). Springer, Berlin.
}
\seealso{
\code{\link{mice}}, \code{link{mice.impute.polyreg}},
\code{\link[MASS]{lda}}
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.polyreg}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Stef van Buuren, Karin Groothuis-Oudshoorn, 2000
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/man/mice.impute.passive.Rd 0000644 0001762 0000144 00000002505 14330031647 016124 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.passive.R
\name{mice.impute.passive}
\alias{mice.impute.passive}
\title{Passive imputation}
\usage{
mice.impute.passive(data, func)
}
\arguments{
\item{data}{A data frame}
\item{func}{A \code{formula} specifying the transformations on data}
}
\value{
The result of applying \code{formula}
}
\description{
Calculate new variable during imputation
}
\details{
Passive imputation is a special internal imputation function. Using this
facility, the user can specify, at any point in the \code{mice} Gibbs
sampling algorithm, a function on the imputed data. This is useful, for
example, to compute a cubic version of a variable, a transformation like
\code{Q = W/H^2} based on two variables, or a mean variable like
\code{(x_1+x_2+x_3)/3}. The so derived variables might be used in other
places in the imputation model. The function allows to dynamically derive
virtually any function of the imputed data at virtually any time.
}
\references{
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
Statistical Software}, \bold{45}(3), 1-67.
\doi{10.18637/jss.v045.i03}
}
\seealso{
\code{\link{mice}}
}
\author{
Stef van Buuren, Karin Groothuis-Oudshoorn, 2000
}
\keyword{datagen}
mice/man/pool.compare.Rd 0000644 0001762 0000144 00000007351 14330031647 014636 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/pool.compare.R
\name{pool.compare}
\alias{pool.compare}
\title{Compare two nested models fitted to imputed data}
\usage{
pool.compare(fit1, fit0, method = c("wald", "likelihood"), data = NULL)
}
\arguments{
\item{fit1}{An object of class 'mira', produced by \code{with.mids()}.}
\item{fit0}{An object of class 'mira', produced by \code{with.mids()}. The
model in \code{fit0} is a nested fit0 of \code{fit1}.}
\item{method}{Either \code{"wald"} or \code{"likelihood"} specifying
the type of comparison. The default is \code{"wald"}.}
\item{data}{No longer used.}
}
\value{
A list containing several components. Component \code{call} is
the call to the \code{pool.compare} function. Component \code{call11} is
the call that created \code{fit1}. Component \code{call12} is the
call that created the imputations. Component \code{call01} is the
call that created \code{fit0}. Component \code{call02} is the
call that created the imputations. Components \code{method} is the
method used to compare two models: 'Wald' or 'likelihood'. Component
\code{nmis} is the number of missing entries for each variable.
Component \code{m} is the number of imputations.
Component \code{qhat1} is a matrix, containing the estimated coefficients of the
\emph{m} repeated complete data analyses from \code{fit1}.
Component \code{qhat0} is a matrix, containing the estimated coefficients of the
\emph{m} repeated complete data analyses from \code{fit0}.
Component \code{ubar1} is the mean of the variances of \code{fit1},
formula (3.1.3), Rubin (1987).
Component \code{ubar0} is the mean of the variances of \code{fit0},
formula (3.1.3), Rubin (1987).
Component \code{qbar1} is the pooled estimate of \code{fit1}, formula (3.1.2) Rubin
(1987).
Component \code{qbar0} is the pooled estimate of \code{fit0}, formula (3.1.2) Rubin
(1987).
Component \code{Dm} is the test statistic.
Component \code{rm} is the relative increase in variance due to nonresponse, formula
(3.1.7), Rubin (1987).
Component \code{df1}: df1 = under the null hypothesis it is assumed that \code{Dm} has an F
distribution with (df1,df2) degrees of freedom.
Component \code{df2}: df2.
Component \code{pvalue} is the P-value of testing whether the model \code{fit1} is
statistically different from the smaller \code{fit0}.
}
\description{
This function is deprecated in V3. Use \code{\link{D1}} or
\code{\link{D3}} instead.
}
\details{
Compares two nested models after m repeated complete data analysis
The function is based on the article of Meng and Rubin (1992). The
Wald-method can be found in paragraph 2.2 and the likelihood method can be
found in paragraph 3. One could use the Wald method for comparison of linear
models obtained with e.g. \code{lm} (in \code{with.mids()}). The likelihood
method should be used in case of logistic regression models obtained with
\code{glm()} in \code{with.mids()}.
The function assumes that \code{fit1} is the
larger model, and that model \code{fit0} is fully contained in \code{fit1}.
In case of \code{method='wald'}, the null hypothesis is tested that the extra
parameters are all zero.
}
\references{
Li, K.H., Meng, X.L., Raghunathan, T.E. and Rubin, D. B. (1991).
Significance levels from repeated p-values with multiply-imputed data.
Statistica Sinica, 1, 65-92.
Meng, X.L. and Rubin, D.B. (1992). Performing likelihood ratio tests with
multiple-imputed data sets. Biometrika, 79, 103-111.
van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}: Multivariate
Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
}
\seealso{
\code{\link{lm.mids}}, \code{\link{glm.mids}}
}
\author{
Karin Groothuis-Oudshoorn and Stef van Buuren, 2009
}
\keyword{htest}
mice/man/mice.impute.panImpute.Rd 0000644 0001762 0000144 00000006276 14436064333 016432 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.panImpute.R
\name{mice.impute.panImpute}
\alias{mice.impute.panImpute}
\title{Impute multilevel missing data using \code{pan}}
\usage{
mice.impute.panImpute(
data,
formula,
type,
m = 1,
silent = TRUE,
format = "imputes",
...
)
}
\arguments{
\item{data}{A data frame containing incomplete and auxiliary variables,
the cluster indicator variable, and any other variables that should be
present in the imputed datasets.}
\item{formula}{A formula specifying the role of each variable
in the imputation model. The basic model is constructed
by \code{model.matrix}, thus allowing to include derived variables
in the imputation model using \code{I()}. See
\code{\link[mitml]{panImpute}}.}
\item{type}{An integer vector specifying the role of each variable
in the imputation model (see \code{\link[mitml]{panImpute}})}
\item{m}{The number of imputed data sets to generate.}
\item{silent}{(optional) Logical flag indicating if console output should be suppressed. Default is to \code{FALSE}.}
\item{format}{A character vector specifying the type of object that should
be returned. The default is \code{format = "list"}. No other formats are
currently supported.}
\item{...}{Other named arguments: \code{n.burn}, \code{n.iter},
\code{group}, \code{prior}, \code{silent} and others.}
}
\value{
A list of imputations for all incomplete variables in the model,
that can be stored in the the \code{imp} component of the \code{mids}
object.
}
\description{
This function is a wrapper around the \code{panImpute} function
from the \code{mitml} package so that it can be called to
impute blocks of variables in \code{mice}. The \code{mitml::panImpute}
function provides an interface to the \code{pan} package for
multiple imputation of multilevel data (Schafer & Yucel, 2002).
Imputations can be generated using \code{type} or \code{formula},
which offer different options for model specification.
}
\note{
The number of imputations \code{m} is set to 1, and the function
is called \code{m} times so that it fits within the \code{mice}
iteration scheme.
This is a multivariate imputation function using a joint model.
}
\examples{
blocks <- list(c("bmi", "chl", "hyp"), "age")
method <- c("panImpute", "pmm")
ini <- mice(nhanes, blocks = blocks, method = method, maxit = 0)
pred <- ini$pred
pred["B1", "hyp"] <- -2
imp <- mice(nhanes, blocks = blocks, method = method, pred = pred, maxit = 1)
}
\references{
Grund S, Luedtke O, Robitzsch A (2016). Multiple
Imputation of Multilevel Missing Data: An Introduction to the R
Package \code{pan}. SAGE Open.
Schafer JL (1997). Analysis of Incomplete Multivariate Data. London:
Chapman & Hall.
Schafer JL, and Yucel RM (2002). Computational strategies for
multivariate linear mixed-effects models with missing values.
Journal of Computational and Graphical Statistics, 11, 437-457.
}
\seealso{
\code{\link[mitml]{panImpute}}
Other multivariate-2l:
\code{\link{mice.impute.jomoImpute}()}
}
\author{
Stef van Buuren, 2018, building on work of Simon Grund,
Alexander Robitzsch and Oliver Luedtke (authors of \code{mitml} package)
and Joe Schafer (author of \code{pan} package).
}
\concept{multivariate-2l}
\keyword{datagen}
mice/man/mice.impute.2l.pan.Rd 0000644 0001762 0000144 00000010664 14436064333 015556 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.2l.pan.R
\name{mice.impute.2l.pan}
\alias{mice.impute.2l.pan}
\alias{2l.pan}
\title{Imputation by a two-level normal model using \code{pan}}
\usage{
mice.impute.2l.pan(
y,
ry,
x,
type,
intercept = TRUE,
paniter = 500,
groupcenter.slope = FALSE,
...
)
}
\arguments{
\item{y}{Incomplete data vector of length \code{n}}
\item{ry}{Vector of missing data pattern (\code{FALSE}=missing,
\code{TRUE}=observed)}
\item{x}{Matrix (\code{n} x \code{p}) of complete covariates.}
\item{type}{Vector of length \code{ncol(x)} identifying random and class
variables. Random effects are identified by a '2'. The group variable (only
one is allowed) is coded as '-2'. Random effects also include the fixed
effect. If for a covariates X1 group means shall be calculated and included
as further fixed effects choose '3'. In addition to the effects in '3',
specification '4' also includes random effects of X1.}
\item{intercept}{Logical determining whether the intercept is automatically
added.}
\item{paniter}{Number of iterations in \code{pan}. Default is 500.}
\item{groupcenter.slope}{If \code{TRUE}, in case of group means (\code{type}
is '3' or'4') group mean centering for these predictors are conducted before
doing imputations. Default is \code{FALSE}.}
\item{...}{Other named arguments.}
}
\value{
A vector of length \code{nmis} with imputations.
}
\description{
Imputes univariate missing data using a two-level normal model with
homogeneous within group variances. Aggregated group effects (i.e. group
means) can be automatically created and included as predictors in the
two-level regression (see argument \code{type}). This function needs the
\code{pan} package.
}
\details{
Implements the Gibbs sampler for the linear two-level model with homogeneous
within group variances which is a special case of a multivariate linear mixed
effects model (Schafer & Yucel, 2002). For a two-level imputation with
heterogeneous within-group variances see \code{\link{mice.impute.2l.norm}}. %
The random intercept is automatically added in %
\code{mice.impute.2l.norm()}.
}
\note{
This function does not implement the \code{where} functionality. It
always produces \code{nmis} imputation, irrespective of the \code{where}
argument of the \code{mice} function.
}
\examples{
# simulate some data
# two-level regression model with fixed slope
# number of groups
G <- 250
# number of persons
n <- 20
# regression parameter
beta <- .3
# intraclass correlation
rho <- .30
# correlation with missing response
rho.miss <- .10
# missing proportion
missrate <- .50
y1 <- rep(rnorm(G, sd = sqrt(rho)), each = n) + rnorm(G * n, sd = sqrt(1 - rho))
x <- rnorm(G * n)
y <- y1 + beta * x
dfr0 <- dfr <- data.frame("group" = rep(1:G, each = n), "x" = x, "y" = y)
dfr[rho.miss * x + rnorm(G * n, sd = sqrt(1 - rho.miss)) < qnorm(missrate), "y"] <- NA
# empty imputation in mice
imp0 <- mice(as.matrix(dfr), maxit = 0)
predM <- imp0$predictorMatrix
impM <- imp0$method
# specify predictor matrix and method
predM1 <- predM
predM1["y", "group"] <- -2
predM1["y", "x"] <- 1 # fixed x effects imputation
impM1 <- impM
impM1["y"] <- "2l.pan"
# multilevel imputation
imp1 <- mice(as.matrix(dfr),
m = 1, predictorMatrix = predM1,
method = impM1, maxit = 1
)
# multilevel analysis
library(lme4)
mod <- lmer(y ~ (1 + x | group) + x, data = complete(imp1))
summary(mod)
# Examples of predictorMatrix specification
# random x effects
# predM1["y","x"] <- 2
# fixed x effects and group mean of x
# predM1["y","x"] <- 3
# random x effects and group mean of x
# predM1["y","x"] <- 4
}
\references{
Schafer J L, Yucel RM (2002). Computational strategies for multivariate
linear mixed-effects models with missing values. \emph{Journal of
Computational and Graphical Statistics}. \bold{11}, 437-457.
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}: Multivariate
Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
}
\seealso{
Other univariate-2l:
\code{\link{mice.impute.2l.bin}()},
\code{\link{mice.impute.2l.lmer}()},
\code{\link{mice.impute.2l.norm}()}
}
\author{
Alexander Robitzsch (IPN - Leibniz Institute for Science and
Mathematics Education, Kiel, Germany), \email{robitzsch@ipn.uni-kiel.de}
Alexander Robitzsch (IPN - Leibniz Institute for Science and
Mathematics Education, Kiel, Germany), \email{robitzsch@ipn.uni-kiel.de}.
}
\concept{univariate-2l}
mice/man/mice.impute.polyreg.Rd 0000644 0001762 0000144 00000007417 14330031647 016142 0 ustar ligges users % Generated by roxygen2: do not edit by hand
% Please edit documentation in R/mice.impute.polyreg.R
\name{mice.impute.polyreg}
\alias{mice.impute.polyreg}
\title{Imputation of unordered data by polytomous regression}
\usage{
mice.impute.polyreg(
y,
ry,
x,
wy = NULL,
nnet.maxit = 100,
nnet.trace = FALSE,
nnet.MaxNWts = 1500,
...
)
}
\arguments{
\item{y}{Vector to be imputed}
\item{ry}{Logical vector of length \code{length(y)} indicating the
the subset \code{y[ry]} of elements in \code{y} to which the imputation
model is fitted. The \code{ry} generally distinguishes the observed
(\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.}
\item{x}{Numeric design matrix with \code{length(y)} rows with predictors for
\code{y}. Matrix \code{x} may have no missing values.}
\item{wy}{Logical vector of length \code{length(y)}. A \code{TRUE} value
indicates locations in \code{y} for which imputations are created.}
\item{nnet.maxit}{Tuning parameter for \code{nnet()}.}
\item{nnet.trace}{Tuning parameter for \code{nnet()}.}
\item{nnet.MaxNWts}{Tuning parameter for \code{nnet()}.}
\item{...}{Other named arguments.}
}
\value{
Vector with imputed data, same type as \code{y}, and of length
\code{sum(wy)}
}
\description{
Imputes missing data in a categorical variable using polytomous regression
}
\details{
The function \code{mice.impute.polyreg()} imputes categorical response
variables by the Bayesian polytomous regression model. See J.P.L. Brand
(1999), Chapter 4, Appendix B.
By default, unordered factors with more than two levels are imputed by
\code{mice.impute.polyreg()}.
The method consists of the following steps:
\enumerate{
\item Fit categorical response as a multinomial model
\item Compute predicted categories
\item Add appropriate noise to predictions
}
The algorithm of \code{mice.impute.polyreg} uses the function
\code{multinom()} from the \code{nnet} package.
In order to avoid bias due to perfect prediction, the algorithm augment the
data according to the method of White, Daniel and Royston (2010).
}
\references{
Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}: Multivariate
Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
Brand, J.P.L. (1999) \emph{Development, implementation and evaluation of
multiple imputation strategies for the statistical analysis of incomplete
data sets.} Dissertation. Rotterdam: Erasmus University.
White, I.R., Daniel, R. Royston, P. (2010). Avoiding bias due to perfect
prediction in multiple imputation of incomplete categorical variables.
\emph{Computational Statistics and Data Analysis}, 54, 2267-2275.
Venables, W.N. & Ripley, B.D. (2002). \emph{Modern applied statistics with
S-Plus (4th ed)}. Springer, Berlin.
}
\seealso{
\code{\link{mice}}, \code{\link[nnet]{multinom}},
\code{\link[MASS]{polr}}
Other univariate imputation functions:
\code{\link{mice.impute.cart}()},
\code{\link{mice.impute.lasso.logreg}()},
\code{\link{mice.impute.lasso.norm}()},
\code{\link{mice.impute.lasso.select.logreg}()},
\code{\link{mice.impute.lasso.select.norm}()},
\code{\link{mice.impute.lda}()},
\code{\link{mice.impute.logreg.boot}()},
\code{\link{mice.impute.logreg}()},
\code{\link{mice.impute.mean}()},
\code{\link{mice.impute.midastouch}()},
\code{\link{mice.impute.mnar.logreg}()},
\code{\link{mice.impute.mpmm}()},
\code{\link{mice.impute.norm.boot}()},
\code{\link{mice.impute.norm.nob}()},
\code{\link{mice.impute.norm.predict}()},
\code{\link{mice.impute.norm}()},
\code{\link{mice.impute.pmm}()},
\code{\link{mice.impute.polr}()},
\code{\link{mice.impute.quadratic}()},
\code{\link{mice.impute.rf}()},
\code{\link{mice.impute.ri}()}
}
\author{
Stef van Buuren, Karin Groothuis-Oudshoorn, 2000-2010
}
\concept{univariate imputation functions}
\keyword{datagen}
mice/DESCRIPTION 0000644 0001762 0000144 00000007361 14437371702 012714 0 ustar ligges users Package: mice
Type: Package
Version: 3.16.0
Title: Multivariate Imputation by Chained Equations
Date: 2023-05-24
Authors@R: c(person("Stef", "van Buuren", role = c("aut","cre"),
email = "stef.vanbuuren@tno.nl"),
person("Karin", "Groothuis-Oudshoorn", role = "aut",
email = "c.g.m.oudshoorn@utwente.nl"),
person("Gerko","Vink", role = "ctb",
email = "g.vink@uu.nl"),
person("Rianne","Schouten", role = "ctb",
email = "R.M.Schouten@uu.nl"),
person("Alexander", "Robitzsch", role = "ctb",
email = "robitzsch@ipn.uni-kiel.de"),
person("Patrick", "Rockenschaub", role = "ctb",
email = "rockenschaub.patrick@gmail.com"),
person("Lisa","Doove", role = "ctb",
email = "lisa.doove@ppw.kuleuven.be"),
person("Shahab","Jolani", role = "ctb",
email = "s.jolani@maastrichtuniversity.nl"),
person("Margarita","Moreno-Betancur", role="ctb",
email = "margarita.moreno@mcri.edu.au"),
person("Ian", "White", role="ctb",
email = "ian.white@ucl.ac.uk"),
person("Philipp","Gaffert", role = "ctb",
email = "philipp.gaffert@gfk.com"),
person("Florian","Meinfelder", role = "ctb",
email = "florian.meinfelder@uni-bamberg.de"),
person("Bernie","Gray", role = "ctb",
email = "bfgray3@gmail.com"),
person("Vincent", "Arel-Bundock", role = "ctb",
email = "vincent.arel-bundock@umontreal.ca"),
person("Mingyang", "Cai", role = "ctb",
email = "m.cai@uu.nl"),
person("Thom", "Volker", role = "ctb",
email = "t.b.volker@uu.nl"),
person("Edoardo", "Costantini", role = "ctb",
email = "e.costantini@tilburguniversity.edu"),
person("Caspar", "van Lissa", role = "ctb",
email = "c.j.vanlissa@uu.nl"),
person("Hanne", "Oberman", role = "ctb",
email = "h.i.oberman@uu.nl"))
Maintainer: Stef van Buuren
Depends: R (>= 2.10.0)
Imports: broom, dplyr, generics, glmnet, graphics, grDevices, lattice,
methods, mitml, nnet, Rcpp, rpart, rlang, stats, tidyr, utils
Suggests: broom.mixed, future, furrr, haven, knitr, lme4, MASS,
miceadds, pan, parallelly, purrr, ranger, randomForest,
rmarkdown, rstan, survival, testthat
Description: Multiple imputation using Fully Conditional Specification (FCS)
implemented by the MICE algorithm as described in Van Buuren and
Groothuis-Oudshoorn (2011) . Each variable has
its own imputation model. Built-in imputation models are provided for
continuous data (predictive mean matching, normal), binary data (logistic
regression), unordered categorical data (polytomous logistic regression)
and ordered categorical data (proportional odds). MICE can also impute
continuous two-level data (normal model, pan, second-level variables).
Passive imputation can be used to maintain consistency between variables.
Various diagnostic plots are available to inspect the quality of the
imputations.
Encoding: UTF-8
LazyLoad: yes
LazyData: yes
URL: https://github.com/amices/mice, https://amices.org/mice/,
https://stefvanbuuren.name/fimd/
BugReports: https://github.com/amices/mice/issues
LinkingTo: cpp11, Rcpp
License: GPL (>= 2)
RoxygenNote: 7.2.3
NeedsCompilation: yes
Packaged: 2023-06-04 21:03:06 UTC; buurensv
Author: Stef van Buuren [aut, cre],
Karin Groothuis-Oudshoorn [aut],
Gerko Vink [ctb],
Rianne Schouten [ctb],
Alexander Robitzsch [ctb],
Patrick Rockenschaub [ctb],
Lisa Doove [ctb],
Shahab Jolani [ctb],
Margarita Moreno-Betancur [ctb],
Ian White [ctb],
Philipp Gaffert [ctb],
Florian Meinfelder [ctb],
Bernie Gray [ctb],
Vincent Arel-Bundock [ctb],
Mingyang Cai [ctb],
Thom Volker [ctb],
Edoardo Costantini [ctb],
Caspar van Lissa [ctb],
Hanne Oberman [ctb]
Repository: CRAN
Date/Publication: 2023-06-05 14:40:02 UTC
mice/build/ 0000755 0001762 0000144 00000000000 14437176011 012272 5 ustar ligges users mice/build/partial.rdb 0000644 0001762 0000144 00000000074 14437176011 014420 0 ustar ligges users b```b`a 00 FN ͚Z d@$ w7 mice/tests/ 0000755 0001762 0000144 00000000000 14437176012 012336 5 ustar ligges users mice/tests/testthat/ 0000755 0001762 0000144 00000000000 14437371702 014201 5 ustar ligges users mice/tests/testthat/test-mice.impute.iurr.logreg.R 0000644 0001762 0000144 00000007136 14433156031 021755 0 ustar ligges users context("mice.impute.lasso.select.logreg")
#########################
# TEST 1: Simple problem #
#########################
set.seed(123)
# generate data
n <- 1e3
y <- rnorm(n)
x <- y * .3 + rnorm(n, 0, .25)
x2 <- x + rnorm(n, 2, 3)
x <- cbind(x, x2)
y <- as.numeric(cut(y, 2)) - 1
# make missingness
y[sample(1:n, n * .3)] <- NA
ry <- !is.na(y)
wy <- !ry
# Use univariate imputation model
set.seed(123)
imps_t1 <- mice.impute.lasso.select.logreg(y, ry, x)
test_that("Returns requested length", {
expect_equal(length(imps_t1), sum(!ry))
})
test_that("Returns dichotomous imputations", {
expect_equal(length(unique(imps_t1)), 2)
})
#########################
# TEST 2: Nothing is important #
#########################
set.seed(20211013)
n <- 1e4
p <- 3
b0 <- 1
bs <- rep(0, p)
x <- cbind(matrix(rnorm(n * p), n, p))
y <- b0 + x %*% bs + rnorm(n)
y <- as.numeric(cut(y, 2)) - 1
# Missing values
y[sample(1:n, n * .3)] <- NA
ry <- !is.na(y)
wy <- !ry
# Use univariate imputation model
set.seed(123)
imps_t2 <- mice.impute.lasso.select.logreg(y, ry, x)
test_that("Returns dichotomous imputations", {
expect_equal(length(unique(imps_t2)), 2)
})
#########################
# TEST 3: Everything is important #
#########################
n <- 1e4
p <- 10
b0 <- 1
bs <- rep(1, p)
x <- cbind(matrix(rnorm(n * p), n, p))
y <- b0 + x %*% bs + rnorm(n)
y <- as.numeric(cut(y, 2)) - 1
# Missing values
y[sample(1:n, n * .3)] <- NA
ry <- !is.na(y)
wy <- !ry
# Use univariate imputation model
set.seed(123)
imps_t3 <- mice.impute.lasso.select.logreg(y, ry, x)
test_that("Works when all predictors are important", {
expect_equal(length(unique(imps_t3)), 2)
})
#########################
# TEST 4: Use it within mice call #
#########################
# Generate some dichotomous data
n <- 1e2
p <- 4
Sigma <- matrix(.7, nrow = p, ncol = p)
diag(Sigma) <- 1
X <- as.data.frame(MASS::mvrnorm(n, rep(0, p), Sigma))
# Discretize and impose miss
for (j in 1:2) {
X[, j] <- cut(X[, j], 2) # Make it discrete
X[sample(1:n, n * .3), j] <- NA # Impose missings
}
# Imputations
iurr_default <- mice(X,
m = 2, maxit = 2, method = "lasso.select.logreg",
print = FALSE
)
iurr_custom <- mice(X,
m = 2, maxit = 2, method = "lasso.select.logreg",
nfolds = 5,
print = FALSE
)
logreg_default <- mice(X,
m = 2, maxit = 2, method = "logreg",
print = FALSE
)
# Tests
test_that("mice call works", {
expect_equal(class(iurr_custom), "mids")
})
test_that("mice call works w/ custom arguments", {
expect_equal(class(iurr_custom), "mids")
})
test_that("same class as logreg default method", {
expect_equal(
class(complete(logreg_default)[, 1]),
class(complete(iurr_default)[, 1])
)
})
#########################
# TEST 5: Perfect Prediction / Complete Separation #
#########################
set.seed(123)
# Generate some dichotomous data
n <- 1e2
p <- 4
Sigma <- matrix(.7, nrow = p, ncol = p)
diag(Sigma) <- 1
x <- MASS::mvrnorm(n, rep(0, p), Sigma)
# Create Perfect Predictor
y <- factor(x[, 1] < 0, labels = c("y", "n"))
# Missing values
y[sample(1:n, n * .3)] <- NA
ry <- !is.na(y)
wy <- !ry
# Imputation well behaved
wellBehaved <- tryCatch(
expr = {
mice.impute.lasso.select.logreg(y = y, ry = ry, x = x[, -1])
},
error = function(e) {
e
},
warning = function(w) {
w
}
)
# Imputation perfect prediction
perfectPred <- tryCatch(
expr = {
mice.impute.lasso.select.logreg(y = y, ry = ry, x = x)
},
error = function(e) {
e
},
warning = function(w) {
w
}
)
# Test
test_that("Complete separation results in same class as well behaved case", {
expect_true(all.equal(class(wellBehaved), class(perfectPred)))
})
mice/tests/testthat/test-pool.r.squared.R 0000644 0001762 0000144 00000000744 14330031606 020146 0 ustar ligges users context("pool.r.squared")
data(nhanes)
imp <- mice::mice(nhanes, maxit = 2, m = 2, seed = 10, print = FALSE)
fit_mira <- with(data = imp, exp = lm(chl ~ age + bmi))
fit_mipo <- mice::pool(fit_mira)
test_that("pool.r.squared mira", {
result <- as.vector(pool.r.squared(fit_mira, adjusted = FALSE)[1, ])
expect_equal(length(result), 4L)
})
test_that("r.squared mipo", {
result <- as.vector(pool.r.squared(fit_mipo, adjusted = FALSE)[1, ])
expect_equal(length(result), 4L)
})
mice/tests/testthat/test-mice.R 0000644 0001762 0000144 00000016337 14334445701 016225 0 ustar ligges users context("mice: complete")
nhanes_mids <- mice(nhanes, m = 2, print = FALSE)
nhanes_complete <- complete(nhanes_mids)
test_that("No missing values remain in imputed nhanes data set", {
expect_gt(sum(is.na(nhanes)), 0)
expect_equal(sum(is.na(nhanes_complete)), 0)
})
test_that("Data set in returned mids object is identical to nhanes data set", {
expect_identical(nhanes_mids$data, nhanes)
})
context("mice: blocks")
test_that("blocks run as expected", {
expect_silent(imp1b <<- mice(nhanes,
blocks = list(c("age", "hyp"), chl = "chl", "bmi"),
print = FALSE, m = 1, maxit = 1, seed = 1
))
expect_silent(imp2b <<- mice(nhanes2,
blocks = list(c("age", "hyp", "bmi"), "chl", "bmi"),
print = FALSE, m = 1, maxit = 1, seed = 1
))
# expect_silent(imp3b <<- mice(nhanes2,
# blocks = list(c("hyp", "hyp", "hyp"), "chl", "bmi"),
# print = FALSE, m = 1, maxit = 1, seed = 1))
expect_silent(imp4b <<- mice(boys,
blocks = list(c("gen", "phb"), "tv"),
print = FALSE, m = 1, maxit = 1, seed = 1
))
expect_silent(imp5b <<- mice(nhanes,
blocks = list(c("age", "hyp")),
print = FALSE, m = 1, maxit = 1, seed = 1
))
})
test_that("Block names are generated automatically", {
expect_identical(names(imp1b$blocks), c("B1", "chl", "bmi"))
})
test_that("Method `pmm` is used for mixed variable types", {
expect_identical(unname(imp2b$method[1]), "pmm")
})
# test_that("Method `logreg` if all are binary", {
# expect_identical(unname(imp3b$method[1]), "logreg")
# })
test_that("Method `polr` if all are ordered", {
expect_identical(unname(imp4b$method[1]), "polr")
})
test_that("Method `polr` works with one block", {
expect_identical(unname(imp5b$method[1]), "pmm")
})
# check for equality of `scatter` and `collect` for univariate models
# the following models yield the same imputations
imp1 <- mice(nhanes,
blocks = make.blocks(nhanes, "scatter"),
print = FALSE, m = 1, maxit = 1, seed = 123
)
imp2 <- mice(nhanes,
blocks = make.blocks(nhanes, "collect"),
print = FALSE, m = 1, maxit = 1, seed = 123
)
imp3 <- mice(nhanes,
blocks = list("age", c("bmi", "hyp", "chl")),
print = FALSE, m = 1, maxit = 1, seed = 123
)
imp4 <- mice(nhanes,
blocks = list(c("bmi", "hyp", "chl"), "age"),
print = FALSE, m = 1, maxit = 1, seed = 123
)
test_that("Univariate yield same imputes for `scatter` and `collect`", {
expect_identical(complete(imp1), complete(imp2))
expect_identical(complete(imp1), complete(imp3))
expect_identical(complete(imp1), complete(imp4))
})
# potentially, we may also change the visitSequence, but mice
# is quite persistent in overwriting a user-specified
# visitSequence for complete columns, so this not
# currently not an option. Defer optimizing this to later.
# another trick is to specify where for age by hand, so it forces
# mice to impute age by pmm, but then, this would need to be
# done in both imp1 and imp2 models.
context("mice: formulas")
test_that("formulas run as expected", {
expect_silent(imp1f <<- mice(nhanes,
formulas = list(
age + hyp ~ chl + bmi,
chl ~ age + hyp + bmi,
bmi ~ age + hyp + chl
),
print = FALSE, m = 1, maxit = 1, seed = 1
))
expect_warning(imp2f <<- mice(nhanes2,
formulas = list(
age + hyp + bmi ~ chl + bmi,
chl ~ age + hyp + bmi + bmi,
bmi ~ age + hyp + bmi + chl
),
print = FALSE, m = 1, maxit = 1, seed = 1
))
# expect_silent(imp3f <<- mice(nhanes2,
# formulas = list( hyp + hyp + hyp ~ chl + bmi,
# chl ~ hyp + hyp + hyp + bmi,
# bmi ~ hyp + hyp + hyp + chl),
# print = FALSE, m = 1, maxit = 1, seed = 1))
expect_silent(imp4f <<- mice(boys,
formulas = list(
gen + phb ~ tv,
tv ~ gen + phb
),
print = FALSE, m = 1, maxit = 1, seed = 1
))
expect_silent(imp5f <<- mice(nhanes,
formulas = list(age + hyp ~ 1),
print = FALSE, m = 1, maxit = 1, seed = 1
))
})
test_that("Formula names are generated automatically", {
expect_identical(names(imp1f$blocks), c("F1", "chl", "bmi"))
})
test_that("Method `pmm` is used for mixed variable types", {
expect_identical(unname(imp2f$method[1]), "pmm")
})
# test_that("Method `logreg` if all are binary", {
# expect_identical(unname(imp3f$method[1]), "logreg")
# })
test_that("Method `polr` if all are ordered", {
expect_identical(unname(imp4f$method[1]), "polr")
})
test_that("Method `polr` works with one block", {
expect_identical(unname(imp5f$method[1]), "pmm")
})
context("mice: where")
# # all TRUE
imp1 <- mice(nhanes,
where = matrix(TRUE, nrow = 25, ncol = 4), maxit = 1,
m = 1, print = FALSE
)
# # all FALSE
imp2 <- mice(nhanes,
where = matrix(FALSE, nrow = 25, ncol = 4), maxit = 1,
m = 1, print = FALSE
)
# # alternate
imp3 <- mice(nhanes,
where = matrix(c(FALSE, TRUE), nrow = 25, ncol = 4),
maxit = 1, m = 1, print = FALSE
)
# # whacky situation where we expect no imputes for the incomplete cases
imp4 <- mice(nhanes2,
where = matrix(TRUE, nrow = 25, ncol = 4),
maxit = 1,
meth = c("pmm", "", "", ""), m = 1, print = FALSE
)
test_that("`where` produces correct number of imputes", {
expect_identical(nrow(imp1$imp$age), 25L)
expect_identical(nrow(imp2$imp$age), 0L)
expect_identical(nrow(imp3$imp$age), 12L)
expect_identical(sum(is.na(imp4$imp$age)), nrow(nhanes2) - sum(complete.cases(nhanes2)))
})
context("mice: ignore")
# # all TRUE
test_that("`ignore` throws appropriate errors and warnings", {
expect_error(
mice(nhanes, maxit = 1, m = 1, print = FALSE, seed = 1, ignore = TRUE),
"does not match"
)
expect_error(
mice(nhanes, maxit = 1, m = 1, print = FALSE, seed = 1, ignore = "string"),
"not a logical"
)
expect_warning(
mice(nhanes,
maxit = 1, m = 1, print = FALSE, seed = 1,
ignore = c(rep(FALSE, 9), rep(TRUE, nrow(nhanes) - 9))
),
"Fewer than 10 rows"
)
})
# Check that the ignore argument is taken into account when
# calculating the results
# # all FALSE
imp1 <- mice(nhanes,
maxit = 1, m = 1, print = FALSE, seed = 1,
ignore = rep(FALSE, nrow(nhanes))
)
# # NULL
imp2 <- mice(nhanes, maxit = 1, m = 1, print = FALSE, seed = 1)
# # alternate
alternate <- rep(c(TRUE, FALSE), nrow(nhanes))[1:nrow(nhanes)]
imp3 <- mice(nhanes,
maxit = 0, m = 1, print = FALSE, seed = 1,
ignore = alternate
)
test_that("`ignore` changes the imputation results", {
expect_identical(complete(imp1), complete(imp2))
expect_failure(expect_identical(complete(imp1), complete(imp3)))
})
# Check that rows flagged as ignored are indeed ignored by the
# univariate sampler in mice
artificial <- data.frame(
age = c(1, 1),
bmi = c(NA, 40.0),
hyp = c(1, 1),
chl = c(200, 200),
row.names = paste0("a", 1:2)
)
imp1 <- mice(
rbind(nhanes, artificial),
maxit = 1, m = 1, print = FALSE, seed = 1, donors = 1L, matchtype = 0
)
imp2 <- mice(
rbind(nhanes, artificial),
maxit = 1, m = 1, print = FALSE, seed = 1, donors = 1L, matchtype = 0,
ignore = c(rep(FALSE, nrow(nhanes)), rep(TRUE, nrow(artificial)))
)
test_that("`ignore` works with pmm", {
expect_equal(complete(imp1)["a1", "bmi"], 40.0)
expect_failure(expect_equal(complete(imp2)["a1", "bmi"], 40.0))
})
mice/tests/testthat/test-make.predictorMatrix.R 0000644 0001762 0000144 00000000375 14330031606 021366 0 ustar ligges users context("make.predictorMatrix")
blocks1 <- make.blocks(c("bmi", "chl", "hyp", "age"))
test_that("errors on invalid data arguments", {
expect_error(
make.predictorMatrix(data, blocks = blocks1),
"Data should be a matrix or data frame"
)
})
mice/tests/testthat/test-mice.impute.norm.R 0000644 0001762 0000144 00000011275 14330031606 020464 0 ustar ligges users context("mice.impute.norm")
#########################
# TEST 1: Simple problem #
#########################
set.seed(123)
# generate data
y <- rnorm(10)
x <- y * .3 + rnorm(10, 0, .25)
x2 <- x + rnorm(10, 2, 3)
x <- cbind(1, x, x2)
# make missingness
y[5:6] <- NA
ry <- !is.na(y)
set.seed(123)
svd <- .norm.draw(y, ry, x, ls.meth = "svd")
set.seed(123)
ridge <- .norm.draw(y, ry, x, ls.meth = "ridge")
set.seed(123)
qr <- .norm.draw(y, ry, x, ls.meth = "qr")
# tests for test1
test_that("Estimates are equal", {
expect_equal(svd$coef, matrix(qr$coef))
expect_equal(svd$beta, matrix(qr$beta))
expect_equal(svd$sigma, qr$sigma)
})
test_that("Correct estimation method used", {
expect_equal(svd$estimation, "svd")
expect_equal(qr$estimation, "qr")
expect_equal(ridge$estimation, "ridge")
})
# svd and qr deliver same estimates; ridge should be different!
#####################################
# TEST 2: extremely high correlation #
#####################################
x <- matrix(c(1:1000, seq(from = 2, to = 2000, by = 2)) + rnorm(1000), nrow = 1000, ncol = 2)
y <- t(c(5, 3) %*% t(x))
y[5:6] <- NA
ry <- !is.na(y)
svd <- .norm.draw(y, ry, x, ls.meth = "svd")
ridge <- .norm.draw(y, ry, x, ls.meth = "ridge")
qr <- .norm.draw(y, ry, x, ls.meth = "qr")
# tests for test2
test_that("Estimates are equal", {
expect_equal(svd$coef, matrix(qr$coef))
expect_equal(svd$beta, matrix(qr$beta))
expect_equal(svd$sigma, qr$sigma)
})
test_that("Correct estimation method used", {
expect_equal(svd$estimation, "svd")
expect_equal(qr$estimation, "qr")
expect_equal(ridge$estimation, "ridge")
})
# svd and qr deliver same estimates; ridge should be different!
#####################################
# TEST 3: correct imputation model #
#####################################
expect_warning(imp.qr <- mice(mammalsleep[, -1], ls.meth = "qr", seed = 123, print = FALSE, use.matcher = TRUE))
expect_warning(imp.svd <- mice(mammalsleep[, -1], ls.meth = "svd", seed = 123, print = FALSE, use.matcher = TRUE))
expect_warning(imp.ridge <- mice(mammalsleep[, -1], ls.meth = "ridge", seed = 123, print = FALSE, use.matcher = TRUE))
test_that("Imputations are equal", {
expect_equal(imp.qr$imp, imp.svd$imp)
expect_false(identical(imp.qr$imp, imp.ridge$imp))
})
#####################################
# TEST 4: exactly singular system #
#####################################
# test on faulty imputation model (exactly singular system)
expect_warning(imp.qr <- mice(mammalsleep, ls.meth = "qr", seed = 123, print = FALSE))
expect_warning(imp.svd <- mice(mammalsleep, ls.meth = "svd", seed = 123, print = FALSE))
expect_warning(imp.ridge <- mice(mammalsleep, ls.meth = "ridge", seed = 123, print = FALSE))
test_that("Imputations are not equal", {
expect_false(identical(imp.qr$imp, imp.svd$imp))
expect_false(identical(imp.qr$imp, imp.ridge$imp))
})
# difference stems from added ridge penalty when necessary (when and where depends
# on starting state of algorithm).
#####################################
# TEST 4: returns requested length #
#####################################
xname <- c("age", "hgt", "wgt")
br <- boys[c(1:10, 101:110, 501:510, 601:620, 701:710), ]
r <- stats::complete.cases(br[, xname])
x <- br[r, xname]
y <- br[r, "tv"]
ry <- !is.na(y)
wy1 <- !ry
wy2 <- rep(TRUE, length(y))
wy3 <- rep(FALSE, length(y))
wy4 <- rep(c(TRUE, FALSE), times = c(1, length(y) - 1))
test_that("Returns requested length", {
expect_equal(length(mice.impute.norm(y, ry, x)), sum(!ry))
expect_equal(length(mice.impute.norm(y, ry, x, wy = wy1)), sum(wy1))
expect_equal(length(mice.impute.norm(y, ry, x, wy = wy2)), sum(wy2))
expect_equal(length(mice.impute.norm(y, ry, x, wy = wy3)), sum(wy3))
expect_equal(length(mice.impute.norm(y, ry, x, wy = wy4)), sum(wy4))
})
###
x <- airquality[, c("Wind", "Temp", "Month")]
y <- airquality[, "Ozone"]
ry <- !is.na(y)
# do imputations depend on x column order?
x1 <- x[, c(1, 2, 3)]
x2 <- x[, c(1, 3, 2)]
set.seed(123)
pmm1 <- mice.impute.pmm(y, ry, x1)
set.seed(123)
pmm2 <- mice.impute.pmm(y, ry, x2)
set.seed(123)
norm1 <- mice.impute.norm(y, ry, x1)
set.seed(123)
norm2 <- mice.impute.norm(y, ry, x2)
set.seed(123)
norm.nob1 <- mice.impute.norm.nob(y, ry, x1)
set.seed(123)
norm.nob2 <- mice.impute.norm.nob(y, ry, x2)
set.seed(123)
norm.predict1 <- mice.impute.norm.predict(y, ry, x1)
set.seed(123)
norm.predict2 <- mice.impute.norm.predict(y, ry, x2)
set.seed(123)
norm.boot1 <- mice.impute.norm.boot(y, ry, x1)
set.seed(123)
norm.boot2 <- mice.impute.norm.boot(y, ry, x2)
test_that("Imputations are invariant to column order", {
# expect_equal(pmm1, pmm2)
# expect_equal(norm1, norm2)
expect_equal(norm.nob1, norm.nob2)
expect_equal(norm.predict1, norm.predict2)
expect_equal(norm.boot1, norm.boot2)
})
mice/tests/testthat/test-complete.R 0000644 0001762 0000144 00000001700 14347334322 017104 0 ustar ligges users context("complete")
imp <- mice(nhanes, maxit = 1, m = 2, seed = 123, print = FALSE)
lng <- subset(complete(imp, "long"), .imp == 1, select = c(age, bmi, hyp, chl))
all <- complete(imp, "all")[[1]]
test_that("long and all produce same data", {
expect_equal(lng, all)
})
# mids workflow using saved objects
imp <- mice(nhanes, seed = 123, print = FALSE)
fit <- with(imp, lm(chl ~ age + bmi + hyp))
est <- pool(fit)
est.mice <- est
# mild workflow using saved objects and base::lapply
idl <- complete(imp, "all")
fit <- lapply(idl, lm, formula = chl ~ age + bmi + hyp)
est <- pool(fit)
est.mild <- est
# long workflow using base::by
cmp <- complete(imp, "long")
fit <- by(cmp, as.factor(cmp$.imp), lm, formula = chl ~ age + bmi + hyp)
est <- pool(fit)
est.long <- est
test_that("workflow mids, mild and long produce same estimates", {
expect_identical(getqbar(est.mice), getqbar(est.mild))
expect_identical(getqbar(est.mice), getqbar(est.long))
})
mice/tests/testthat/test-mice.impute.2l.lmer.R 0000644 0001762 0000144 00000001235 13666252075 020777 0 ustar ligges users context("mice.impute.2l.lmer")
d <- brandsma[1:200, c("sch", "lpo")]
pred <- make.predictorMatrix(d)
pred["lpo", "sch"] <- -2
test_that("mice::mice.impute.2l.lmer() runs empty model", {
expect_silent(imp <- mice(d, method = "2l.lmer", print = FALSE, pred = pred, m = 1, maxit = 1))
expect_false(anyNA(complete(imp)))
})
d <- brandsma[1:200, c("sch", "lpo", "iqv")]
d[c(1, 11, 21), "iqv"] <- NA
pred <- make.predictorMatrix(d)
pred[c("lpo", "iqv"), "sch"] <- -2
test_that("2l.lmer() runs random intercept, one predictor", {
expect_silent(imp <- mice(d, method = "2l.lmer", print = FALSE, pred = pred, m = 1, maxit = 1))
expect_false(anyNA(complete(imp)))
})
mice/tests/testthat/test-mice.impute.rf.R 0000644 0001762 0000144 00000001406 14433400023 020110 0 ustar ligges users # Outcommented to reduce dependencies under the _R_CHECK_DEPENDS_ONLY=true flag
# context("mice.impute.rf")
#
# #####################################
# # TEST 1: runs with single miss val #
# #####################################
#
# data <- matrix(
# c(
# 1.0, 10.5, 1.5, 13.2, 1.8, 8.0, 1.7, 15.0, 23.0, 40.0,
# 2.0, 21.0, 3.3, 38.0, 4.5, -2.3, NA, -2.4
# ),
# nrow = 9, ncol = 2, byrow = TRUE
# )
# df <- data.frame(data)
#
# par <- list(
# y = df$X1,
# ry = !is.na(df$X1),
# x = df[, "X2", drop = FALSE]
# )
#
# test_that(
# "Runs with a single missing value",
# {
# expect_visible(do.call(mice.impute.rf, c(par, list(rfPackage = "ranger"))))
# expect_visible(do.call(mice.impute.rf, c(par, list(rfPackage = "randomForest"))))
# }
# )
mice/tests/testthat/test-mice.impute.logreg.R 0000644 0001762 0000144 00000001737 14330031647 020777 0 ustar ligges users context("test-mice.impute.logreg.R")
#########################
# TEST 1: Perfect Prediction / Complete Separation #
#########################
set.seed(123)
# Generate some dichotomous data
n <- 1e2
p <- 4
Sigma <- matrix(.7, nrow = p, ncol = p)
diag(Sigma) <- 1
x <- MASS::mvrnorm(n, rep(0, p), Sigma)
# Create Perfect Predictor
y <- factor(x[, 1] < 0, labels = c("y", "n"))
# Missing values
y[sample(1:n, n * .3)] <- NA
ry <- !is.na(y)
wy <- !ry
# Imputation well behaved
wellBehaved <- tryCatch(
expr = {
mice.impute.logreg(y = y, ry = ry, x = x[, -1])
},
error = function(e) {
e
},
warning = function(w) {
w
}
)
# Imputation perfect prediction
perfectPred <- tryCatch(
expr = {
mice.impute.logreg(y = y, ry = ry, x = x)
},
error = function(e) {
e
},
warning = function(w) {
w
}
)
# Test
test_that("Complete separation results in same class as well behaved case", {
expect_true(all.equal(class(wellBehaved), class(perfectPred)))
})
mice/tests/testthat/test-mice.impute.2l.bin.R 0000644 0001762 0000144 00000002267 14330031606 020576 0 ustar ligges users context("mice.impute.2l.bin")
# toenail: outcome is factor
data("toenail2")
data <- tidyr::complete(toenail2, patientID, visit) %>%
tidyr::fill(treatment) %>%
dplyr::select(-time) %>%
dplyr::mutate(patientID = as.integer(patientID))
summary(data)
# fit1 <- glm(outcome ~ treatment * month, data = toenail2, family = binomial)
# fit2 <- glm(outcome ~ treatment * visit, data = toenail2, family = binomial)
# fit3 <- lme4::glmer(outcome ~ treatment * visit + (1 | ID), data = data, family = binomial)
pred <- make.predictorMatrix(data)
pred["outcome", "patientID"] <- -2
test_that("mice::mice.impute.2l.bin() accepts factor outcome", {
expect_silent(imp <- mice(data, method = "2l.bin", print = FALSE, pred = pred, m = 1, maxit = 1))
expect_false(anyNA(complete(imp)))
})
# toenail: outcome is 0/1
data("toenail")
data <- tidyr::complete(toenail, ID, visit) %>%
tidyr::fill(treatment) %>%
dplyr::select(-month)
summary(data)
pred <- make.predictorMatrix(data)
pred["outcome", "ID"] <- -2
test_that("mice::mice.impute.2l.bin() accepts 0/1 outcome", {
expect_silent(imp <- mice(data, method = "2l.bin", print = FALSE, pred = pred, m = 1, maxit = 1))
expect_false(anyNA(complete(imp)))
})
mice/tests/testthat/test-mice.impute.panImpute.R 0000644 0001762 0000144 00000001311 14436064333 021453 0 ustar ligges users context("mice.impute.panImpute")
data <- boys[c(1:10, 101:110, 501:510, 601:620, 701:710), ]
type <- c(2, 0, 0, 0, -2, 0, 1, 1, 0)
names(type) <- names(data)
z1 <- mice.impute.panImpute(data = data, type = type, format = "native")
test_that("panImpute returns native class", {
expect_is(z1, "mitml")
})
blocks <- make.blocks(list(c("bmi", "chl", "hyp"), "age"))
method <- c("panImpute", "pmm")
pred <- make.predictorMatrix(nhanes, blocks)
pred["B1", "hyp"] <- -2
imp <- mice(nhanes,
blocks = blocks, method = method, pred = pred,
maxit = 1, seed = 1, print = FALSE
)
z <- complete(imp)
test_that("mice can call panImpute", {
expect_equal(sum(is.na(z$bmi)), 0)
expect_equal(sum(is.na(z$chl)), 0)
})
mice/tests/testthat/test-newdata.R 0000644 0001762 0000144 00000002706 14334522175 016727 0 ustar ligges users context("mice.mids: newdata")
# Check that mice.mids correctly appends the newdata to the
# existing mids object
init0 <- mice(nhanes, maxit = 0, m = 1, print = FALSE, seed = 1)
init1 <- mice(nhanes, maxit = 0, m = 1, print = FALSE)
init1$ignore <- rep(FALSE, nrow(nhanes))
init2 <- mice.mids(init0, newdata = nhanes, maxit = 0, print = FALSE)
test_that("`newdata` works like rbind with ignore", {
expect_equal(complete(rbind(init0, init1)), complete(init2))
})
imp <- mice(nhanes2, maxit = 0, m = 1, seed = 1)
test_that("`newdata` produces warning `invalid factor level, NA generated`", {
expect_silent(mice.mids(imp, newdata = nhanes2[1, ], print = FALSE))
})
# Check that rows flagged as ignored are indeed ignored by the
# univariate sampler in mice.mids
artificial <- data.frame(
age = c(1, 1),
bmi = c(NA, 40.0),
hyp = c(1, 1),
chl = c(200, 200),
row.names = paste0("a", 1:2)
)
imp1 <- mice(nhanes,
maxit = 1, m = 1, print = FALSE, seed = 1,
donors = 1L, matchtype = 0
)
imp2 <- mice.mids(imp1, newdata = artificial, maxit = 1, print = FALSE)
imp2b <- mice.mids(imp1, newdata = artificial, maxit = 1, print = FALSE)
test_that("`newdata` works with pmm", {
expect_failure(expect_equal(complete(imp2)["a1", "bmi"], 40.0))
})
test_that("`newdata` returns filtered mids object", {
expect_equal(nrow(complete(imp2)), nrow(artificial))
})
test_that("`newdata` uses a common seed", {
expect_true(identical(complete(imp2), complete(imp2b)))
})
mice/tests/testthat/test-blocks.R 0000644 0001762 0000144 00000001604 14344575371 016564 0 ustar ligges users context("blocks")
imp <- mice(nhanes, blocks = make.blocks(list(c("bmi", "chl"), "bmi", "age")), m = 10, print = FALSE)
# plot(imp)
test_that("removes variables from 'where'", {
expect_identical(sum(imp$where[, "hyp"]), 0L)
})
# reprex https://github.com/amices/mice/issues/326
imp1 <- mice(nhanes, seed = 1, m = 1, maxit = 2, print = FALSE)
imp2 <- mice(nhanes, blocks = list(c("bmi", "hyp"), "chl"), m = 1, maxit = 2, seed = 1, print = FALSE)
test_that("expands a univariate method to all variables in the block", {
expect_identical(complete(imp1, 1), complete(imp2, 1))
})
imp3 <- mice(nhanes, blocks = list(c("hyp", "bmi"), "chl"), m = 1, maxit = 2, seed = 1, print = FALSE)
imp4 <- mice(nhanes, visitSequence = c("hyp", "bmi", "chl"), m = 1, maxit = 2, seed = 1, print = FALSE)
test_that("blocks alter the visit sequence", {
expect_identical(complete(imp3, 1), complete(imp3, 1))
})
mice/tests/testthat/test-mice.impute.polr.R 0000644 0001762 0000144 00000000273 14330031606 020461 0 ustar ligges users context("polr")
# following halts with
# "Error in apply(draws, 2, sum) : dim(X) must have a positive length"
# imp1 <- mice(boys, blocks = list("gen"), print = FALSE, m = 1, maxit = 1)
mice/tests/testthat/test-rbind.R 0000644 0001762 0000144 00000010673 14330031606 016372 0 ustar ligges users context("rbind.mids")
expect_warning(imp1 <<- mice(nhanes[1:13, ], m = 2, maxit = 1, print = FALSE))
test_that("Constant variables are not imputed by default", {
expect_equal(sum(is.na(complete(imp1))), 6L)
})
expect_warning(imp1b <<- mice(nhanes[1:13, ], m = 2, maxit = 1, print = FALSE, remove.constant = FALSE))
test_that("Constant variables are imputed for remove.constant = FALSE", {
expect_equal(sum(is.na(complete(imp1b))), 0L)
})
imp2 <- mice(nhanes[14:25, ], m = 2, maxit = 1, print = FALSE)
imp3 <- mice(nhanes2, m = 2, maxit = 1, print = FALSE)
imp4 <- mice(nhanes2, m = 1, maxit = 1, print = FALSE)
expect_warning(imp5 <<- mice(nhanes[1:13, ], m = 2, maxit = 2, print = FALSE))
expect_error(imp6 <<- mice(nhanes[1:13, 2:3], m = 2, maxit = 2, print = FALSE), "`mice` detected constant and/or collinear variables. No predictors were left after their removal.")
nh3 <- nhanes
colnames(nh3) <- c("AGE", "bmi", "hyp", "chl")
imp7 <- mice(nh3[14:25, ], m = 2, maxit = 2, print = FALSE)
expect_warning(imp8 <<- mice(nhanes[1:13, ], m = 2, maxit = 2, print = FALSE))
imp9 <- mice(nhanes,
m = 2, maxit = 1, print = FALSE,
ignore = c(rep(FALSE, 20), rep(TRUE, 5))
)
mylist <- list(age = NA, bmi = NA, hyp = NA, chl = NA)
nhalf <- nhanes[13:25, ]
test_that("Expands number of rows and imputes", {
expect_equal(nrow(complete(rbind(imp1, imp2))), 25L)
expect_equal(nrow(rbind(imp1, imp2)$imp$bmi), 9L)
})
test_that("throws error", {
expect_error(rbind(imp1, imp3), "datasets have different factor variables")
expect_error(rbind(imp3, imp4), "number of imputations differ")
expect_error(rbind(imp1, imp7), "datasets have different variable names")
})
test_that("throws warning", {
expect_warning(
rbind(imp1, imp5),
"iterations differ, so no convergence diagnostics calculated"
)
expect_warning(
rbind(imp5, imp9),
"iterations differ, so no convergence diagnostics calculated"
)
})
r1 <- rbind(imp8, imp5)
r2 <- rbind(imp1, mylist)
r3 <- rbind(imp1, nhalf)
r4 <- rbind(imp1, imp2)
r5 <- rbind(imp2, imp9)
test_that("Produces longer imputed data", {
expect_identical(nrow(complete(r1)), 26L)
expect_identical(nrow(complete(r2)), 14L)
})
test_that("Constant variables are not imputed", {
expect_equal(sum(is.na(complete(r3))), 15L)
expect_equal(sum(is.na(complete(r4))), 6L)
})
test_that("`ignore` is correctly appended", {
expect_equal(r2$ignore, rep(FALSE, 14))
expect_equal(r5$ignore, c(rep(FALSE, 32), rep(TRUE, 5)))
})
# r11 <- mice.mids(rbind(imp1, imp5), print = FALSE)
# test_that("plot throws error on convergence diagnostics", {
# expect_error(plot(r11), "no convergence diagnostics found")
# })
r21 <- mice.mids(r2, print = FALSE)
r31 <- mice.mids(r3, print = FALSE)
# issue #59
set.seed <- 818
x <- rnorm(10)
D <- data.frame(x = x, y = 2 * x + rnorm(10))
D[c(2:4, 7), 1] <- NA
expect_error(D_mids <<- mice(D[1:5, ], print = FALSE), "`mice` detected constant and/or collinear variables. No predictors were left after their removal.")
expect_warning(D_mids <<- mice(D[1:5, ], print = FALSE, remove.collinear = FALSE))
D_rbind <- mice:::rbind.mids(D_mids, D[6:10, ])
cmp <- complete(D_rbind, 1)
test_that("Solves issue #59, rbind", expect_identical(cmp[6:10, ], D[6:10, ]))
test_that("rbind does not throw a warning (#114)", {
expect_silent(rbind(ordered(c(1, 2))))
})
# calculate chainMean and chainVar
# imp1 <- mice(nhanes[1:13, ], m = 5, maxit = 25, print = FALSE, seed = 123)
# imp2 <- mice(nhanes[14:25, ], m = 5, maxit = 25, print = FALSE, seed = 456)
# z <- rbind(imp1, imp2)
# plot(z)
#
# imp3 <- mice(nhanes, m = 5, maxit = 25, print = FALSE, seed = 123)
# plot(imp3)
#
# An interesting observation is that the SD(hyp, a) < SD(hyp, imp3). This is
# because SD(hyp, imp1) = 0.
# issue 319: https://github.com/amices/mice/issues/319
# impute a subset
# do not touch or impute non-selected rows
# return full data as mids object
# example: impute even rows only
data <- nhanes
odd <- as.logical((1:nrow(data)) %% 2)
# method 1: ignore + where
where <- make.where(data)
where[odd, ] <- FALSE
imp1 <- mice(nhanes, ignore = odd, where = where, seed = 1, m = 2, print = FALSE)
c1 <- complete(imp1, 2)
# method 2: filter + rbind
imp2 <- mice(data[!odd, ], seed = 1, m = 2, print = FALSE)
imp2 <- rbind(imp2, data[odd, ])
idx <- order(as.numeric(rownames(imp2$data)))
imp2$data <- imp2$data[idx, ]
imp2$where <- imp2$where[idx, ]
c2 <- complete(imp2, 2)
test_that("ignore + where is identical to filter + rbind (#319)", {
expect_identical(c1, c2)
})
mice/tests/testthat/test-mice.impute.durr.norm.R 0000644 0001762 0000144 00000001713 14433156031 021437 0 ustar ligges users context("mice.impute.lasso.norm")
#########################
# TEST 1: Simple problem #
#########################
set.seed(123)
# generate data
n <- 1e3
y <- rnorm(n)
x <- y * .3 + rnorm(n, 0, .25)
x2 <- x + rnorm(n, 2, 3)
x <- cbind(x, x2)
# make missingness
y[sample(1:n, n * .3)] <- NA
ry <- !is.na(y)
# Use univariate imputation model
set.seed(123)
imps <- mice.impute.lasso.norm(y, ry, x)
test_that("Returns requested length", {
expect_equal(length(imps), sum(!ry))
})
#########################
# TEST 2: Use it within mice call #
#########################
boys_cont <- boys[, 1:4]
durr_default <- mice(boys_cont,
m = 2, maxit = 2, method = "lasso.norm",
print = FALSE
)
durr_custom <- mice(boys_cont,
m = 2, maxit = 2, method = "lasso.norm",
nfolds = 5,
print = FALSE
)
test_that("mice call works", {
expect_equal(class(durr_custom), "mids")
})
test_that("mice call works w/ custom arguments", {
expect_equal(class(durr_custom), "mids")
})
mice/tests/testthat/test-ampute.R 0000644 0001762 0000144 00000027473 14335404116 016602 0 ustar ligges users context("ampute")
# make objects for testfunctions
sigma <- matrix(data = c(1, 0.2, 0.2, 0.2, 1, 0.2, 0.2, 0.2, 1), nrow = 3)
complete.data <- MASS::mvrnorm(n = 100, mu = c(5, 5, 5), Sigma = sigma)
test_that("all examples work", {
compl_boys <- cc(boys)[1:3]
expect_error(ampute(data = compl_boys), NA)
mads_boys <- ampute(data = compl_boys)
my_patterns <- mads_boys$patterns
my_patterns[1:3, 2] <- 0
my_weights <- mads_boys$weights
my_weights[2, 1] <- 2
my_weights[3, 1] <- 0.5
expect_error(ampute(
data = compl_boys, patterns = my_patterns,
freq = c(0.3, 0.3, 0.4), weights = my_weights,
type = c("RIGHT", "TAIL", "LEFT")
), NA)
})
test_that("all arguments work", {
set.seed(123)
# empty run
expect_error(ampute(data = complete.data, run = FALSE), NA)
# missingness by cells
expect_error(ampute(data = complete.data, prop = 0.1, bycases = FALSE), NA)
# prop with 3 dec, weigths with negative values, unequal odds matrix
expect_error(ampute(
data = complete.data, prop = 0.314,
freq = c(0.25, 0.4, 0.35),
patterns = matrix(
data = c(
1, 0, 1,
0, 1, 0,
0, 1, 1
),
nrow = 3, byrow = TRUE
),
weights = matrix(
data = c(
-1, 1, 0,
-4, -4, 1,
0, 0, -1
),
nrow = 3, byrow = TRUE
),
odds = matrix(
data = c(
1, 4, NA, NA,
0, 3, 3, NA,
4, 1, 1, 4
),
nrow = 3, byrow = TRUE
),
cont = FALSE
), NA)
# 1 pattern with vector for patterns and weights
expect_error(ampute(
data = complete.data, freq = 1, patterns = c(1, 0, 1),
weights = c(3, 3, 0)
), NA)
# multiple patterns given in vectors
expect_error(ampute(
data = complete.data, patterns = c(1, 0, 1, 1, 0, 0),
cont = TRUE, weights = c(1, 4, -2, 0, 1, 2),
type = c("LEFT", "TAIL")
), NA)
# one pattern with odds vector
expect_error(ampute(
data = complete.data, patterns = c(1, 0, 1),
weights = c(4, 1, 0), odds = c(2, 1), cont = FALSE
), NA)
# argument standardized
expect_error(ampute(data = complete.data, std = FALSE), NA)
# sum scores cannot be NaN
dich.data <- matrix(c(
0, 0, 0, 1, 0, 0, 0, 0, 0,
1, 0, 0, 0, 0, 0, 0, 0, 0
), ncol = 2, byrow = FALSE)
wss <- expect_warning(ampute(data = dich.data, mech = "MNAR")$scores)
check_na <- function(x) {
return(any(is.na(x)))
}
expect_false(any(unlist(lapply(wss, check_na))))
})
test_that("function works around unusual arguments", {
# data
nasty.data <- complete.data
nasty.data[, 1] <- rep(c("one", "two"), 50)
# when data is categorical and mech != mcar, warning is expected
expect_warning(
ampute(data = nasty.data),
"Data is made numeric because the calculation of weights requires numeric data"
)
# when data is categorical and mech = mcar, function can continue
expect_warning(ampute(data = nasty.data, mech = "MCAR"), NA)
# patterns
expect_error(ampute(data = complete.data, patterns = c(0, 0, 0), mech = "MCAR"), NA)
expect_error(ampute(data = complete.data, patterns = c(0, 0, 1, 0, 0, 0), mech = "MNAR"), NA)
expect_warning(ampute(data = complete.data, patterns = c(1, 1, 1, 0, 1, 0)))
# freq
expect_warning(ampute(data = complete.data, freq = c(0.8, 0.4)))
# prop
expect_warning(ampute(data = complete.data, prop = 1))
expect_error(ampute(data = complete.data, prop = 48.5), NA)
# mech, type and weights
expect_warning(
ampute(data = complete.data, mech = c("MCAR", "MAR")),
"Mechanism should contain merely MCAR, MAR or MNAR. First element is used"
)
expect_warning(
ampute(data = complete.data, type = c("LEFT", "RIGHT")),
"Type should either have length 1 or length equal to #patterns, first element is used for all patterns"
)
expect_warning(
ampute(
data = complete.data, mech = "MCAR",
odds = matrix(
data = c(
1, 4, NA, NA,
0, 3, 3, NA,
4, 1, 1, 4
),
nrow = 3, byrow = TRUE
), cont = FALSE
),
"Odds matrix is not used when mechanism is MCAR"
)
expect_warning(
ampute(
data = complete.data, mech = "MCAR",
weights = c(1, 3, 4)
),
"Weights matrix is not used when mechanism is MCAR"
)
expect_warning(ampute(data = complete.data, odds = matrix(
data = c(
1, 4, NA, NA,
0, 3, 3, NA,
4, 1, 1, 4
),
nrow = 3, byrow = TRUE
)))
expect_warning(ampute(data = complete.data, cont = FALSE, type = "LEFT"))
})
test_that("error messages work properly", {
# data
expect_error(
ampute(data = as.list(complete.data)),
"Data should be a matrix or data frame"
)
nasty.data <- complete.data
nasty.data[1:10, 1] <- NA
expect_error(ampute(data = nasty.data), "Data cannot contain NAs")
expect_error(
ampute(data = as.data.frame(complete.data[, 1])),
"Data should contain at least two columns"
)
# prop
expect_error(ampute(data = complete.data, prop = 104))
expect_error(
ampute(data = complete.data, prop = 0.9, bycases = FALSE),
"Proportion of missing cells is too large in combination with the desired number of missing variables"
)
# patterns
expect_error(
ampute(data = complete.data, patterns = c(1, 1, 1)),
"One pattern with merely ones results to no amputation at all, the procedure is therefore stopped"
)
expect_error(
ampute(data = complete.data, patterns = c(0, 0, 0), mech = "MAR"),
"Patterns object contains merely zeros and this kind of pattern is not possible when mechanism is MAR"
)
expect_error(
ampute(data = complete.data, patterns = c(1, 0, 1, 1)),
"Length of pattern vector does not match #variables"
)
expect_error(
ampute(data = complete.data, patterns = c(1, 0, 2)),
"Argument patterns can only contain 0 and 1, pattern 1 contains another element"
)
expect_error(
ampute(data = complete.data, mech = "MAR", patterns = c(0, 0, 1, 0, 0, 0)),
"Patterns object contains merely zeros and this kind of pattern is not possible when mechanism is MAR"
)
# mech, type, weights and odds
expect_error(
ampute(data = complete.data, mech = "MAAR"),
"Mechanism should be either MCAR, MAR or MNAR"
)
expect_error(
ampute(data = complete.data, type = "MARLEFT"),
"Type should contain LEFT, MID, TAIL or RIGHT"
)
expect_error(
ampute(data = complete.data, weights = c(1, 2, 1, 4)),
"Length of weight vector does not match #variables"
)
expect_error(ampute(
data = complete.data,
odds = matrix(c(1, 4, -3, 2, 1, 1), nrow = 3),
cont = FALSE
), "Odds matrix can only have positive values")
expect_error(
ampute(
data = complete.data,
patterns = matrix(
data = c(
1, 0, 1,
0, 1, 0,
0, 1, 1
),
nrow = 3, byrow = TRUE
),
weights = matrix(
data = c(
-1, 1, 0,
-4, -4, 1,
0, 0, -1,
1, 1, 0
),
nrow = 4, byrow = TRUE
)
),
"The objects patterns and weights are not matching"
)
expect_error(
ampute(
data = complete.data,
patterns = matrix(
data = c(
1, 0, 1,
0, 1, 0,
0, 1, 1
),
nrow = 3, byrow = TRUE
),
odds = matrix(
data = c(
1, 4, NA, NA,
0, 3, 3, 0
),
nrow = 2, byrow = TRUE
), cont = FALSE
),
"The objects patterns and odds are not matching"
)
})
# The following tests were created to evaluate the patterns and weights matrices in case of a pattern with only 1's (#449)
test_that("patterns and weights matrices have right dimensions", {
suppressWarnings(
expect_true(all(
ampute(data = complete.data, patterns = c(1, 1, 1, 0, 1, 0))$patterns == c(0, 1, 0)
))
)
suppressWarnings(
expect_true(all(
ampute(data = complete.data, patterns = c(0, 1, 0, 1, 1, 1))$patterns == c(0, 1, 0)
))
)
suppressWarnings(
expect_true(all(
ampute(data = complete.data, patterns = c(1, 1, 1, 0, 1, 0, 1, 1, 1))$patterns == c(0, 1, 0)
))
)
suppressWarnings(
expect_true(all(
ampute(
data = complete.data, patterns = c(1, 1, 1, 0, 1, 0),
weights = c(1, 0, 0, 0, 1, 0)
)$weights == c(0, 1, 0)
))
)
suppressWarnings(
expect_true(all(
ampute(
data = complete.data, patterns = c(0, 1, 0, 1, 1, 1),
weights = c(1, 0, 0, 0, 1, 0)
)$weights == c(1, 0, 0)
))
)
suppressWarnings(
expect_true(all(
ampute(
data = complete.data, patterns = c(0, 1, 0, 1, 1, 1),
weights = c(1, 0, 0)
)$weights == c(1, 0, 0)
))
)
suppressWarnings(
expect_true(all(
ampute(
data = complete.data, patterns = c(1, 1, 1, 0, 1, 0, 1, 1, 1),
weights = c(1, 0, 0)
)$weights == c(1, 0, 0)
))
)
suppressWarnings(
expect_true(all(
ampute(
data = complete.data, patterns = c(1, 1, 1, 0, 1, 0, 1, 1, 1),
weights = c(1, 0, 0, 0, 1, 0, 0, 0, 1)
)$weights == c(0, 1, 0)
))
)
})
test_that("prop and freq are properly adjusted when patterns contain only 1's", {
suppressWarnings(
expect_equal(ampute(data = complete.data, patterns = c(1, 1, 1, 0, 1, 0))$prop, 0.25)
)
suppressWarnings(
expect_equal(ampute(data = complete.data, patterns = c(1, 1, 1, 0, 1, 0))$freq, 1)
)
suppressWarnings(
expect_equal(
ampute(data = complete.data, patterns = c(1, 1, 1, 0, 1, 0, 0, 1, 0))$prop, 1 / 3
)
)
suppressWarnings(
expect_true(all(
ampute(data = complete.data, patterns = c(1, 1, 1, 0, 1, 0, 0, 1, 0))$freq == c(0.5, 0.5)
))
)
suppressWarnings(
expect_equal(
ampute(data = complete.data, patterns = c(1, 1, 1, 0, 1, 0, 1, 1, 1))$prop, 1 / 3 * 0.5
)
)
suppressWarnings(
expect_equal(
ampute(data = complete.data, patterns = c(1, 1, 1, 0, 1, 0, 1, 1, 1))$freq, 1
)
)
})
# The following test was created to evaluate warnings when not all patterns can be generated (#317)
test_that("warnings appear when not all patterns can be generated", {
set.seed(12032021)
binary.data <- lapply(
runif(10, 0.05, 0.15),
function(p, n) rbinom(n, 1, p),
n = 10
) %>%
do.call(what = "data.frame") %>%
rlang::set_names(paste0("type", LETTERS[1:ncol(.)]))
expect_warning(
ampute(
data = binary.data
)
)
df <- matrix(c(runif(1000, 0.5, 1), rep(0, 1000)), nrow = 1000, byrow = FALSE)
expect_warning(
ampute(df, pattern = c(0, 1)),
"The weighted sum scores of all candidates in pattern 1 are the same, they will be amputed with probability 0.5"
)
})
# The following test was contributed by Shangzhi-hong (#216) Dec 2019
context("ampute robust version")
set.seed(1)
# Set-up
# Dataset
NUM_OBS_DF <- 25
NUM_VAR_DF <- 10
data <- replicate(
n = NUM_VAR_DF,
expr = {
rnorm(n = NUM_OBS_DF, mean = 1, sd = 1)
},
simplify = "matrix"
)
# Ampute pattern
covNum <- NUM_VAR_DF - 1
misPatCov1 <- t(combn(
x = covNum, m = 1,
FUN = function(x) replace(rep(1, covNum), x, 0)
))
misPat1 <- cbind(rep(1, choose(covNum, 1)), misPatCov1)
misPatCov2 <- t(combn(
x = covNum, m = 2,
FUN = function(x) replace(rep(1, covNum), x, 0)
))
misPat2 <- cbind(rep(1, choose(covNum, 2)), misPatCov2)
patterns <- rbind(misPat1, misPat2)
weights <- matrix(0, nrow = nrow(patterns), ncol = ncol(patterns))
weights[, 1] <- 1
prop <- 0.5
mech <- "MAR"
type <- "RIGHT"
bycases <- TRUE
# Other params
# freq <- NULL
# std <- TRUE
# cont <- TRUE
# type <- NULL
# odds <- NULL
# run <- TRUE
test_that("ampute() works under extreme condition", {
expect_warning(
ampDf <- ampute(
data = data,
prop = prop,
mech = mech,
type = type,
bycases = bycases,
patterns = patterns,
weights = weights
)$amp
)
outProp <- sum(complete.cases(ampDf)) / NUM_OBS_DF
expect_true(outProp > 0.3 & outProp < 0.7)
})
# --- end test Shangzhi-hong (#216) Dec 2019
mice/tests/testthat/test-as.mids.R 0000644 0001762 0000144 00000004046 14330031606 016627 0 ustar ligges users context("as.mids")
nhanes3 <- nhanes
rownames(nhanes3) <- LETTERS[1:nrow(nhanes3)]
imp <- mice(nhanes3, m = 2, maxit = 1, print = FALSE)
X <- complete(imp, action = "long", include = TRUE)
# create dataset with .imp variable as numeric
X2 <- X
# nhanes example
test1 <- as.mids(X)
# nhanes example
test2 <- as.mids(X2)
# nhanes example, where we explicitly specify .id as column 2
test3 <- as.mids(X, .id = 2)
# nhanes example with .id where .imp is numeric
test4 <- as.mids(X2, .id = 2)
#' # example without an .id variable
#' # variable .id not preserved
test5 <- as.mids(X[, -2])
#' # reverse data order
rev <- ncol(X):1
test6 <- as.mids(X[, rev])
# as() syntax has fewer options
test7 <- as(X, "mids")
test8 <- as(X2, "mids")
test9 <- as(X2[, -2], "mids")
test10 <- as(X[, rev], "mids")
test_that("as.mids() produces a `mids` object", {
expect_is(test1, "mids")
expect_is(test2, "mids")
expect_is(test3, "mids")
expect_is(test4, "mids")
expect_is(test5, "mids")
expect_is(test7, "mids")
expect_is(test8, "mids")
expect_is(test9, "mids")
expect_is(test10, "mids")
expect_error(
as(X[-(1:10), ], "mids"),
"Unequal group sizes in imputation index `.imp`"
)
expect_error(
as(X[, -(1:2)], "mids"),
"Imputation index `.imp` not found"
)
})
test_that("complete() reproduces the original data", {
expect_true(all(complete(test1, action = "long", include = TRUE) == X, na.rm = TRUE))
expect_true(all(complete(test2, action = "long", include = TRUE) == X, na.rm = TRUE))
expect_true(all(complete(test3, action = "long", include = TRUE) == X, na.rm = TRUE))
expect_true(all(complete(test4, action = "long", include = TRUE) == X, na.rm = TRUE))
expect_true(all(complete(test5, action = "long", include = TRUE)[, -2] == X[, -2], na.rm = TRUE))
expect_true(all(complete(test6, action = "long", include = TRUE)[, -(1:2)] == X[, rev][, -(5:6)], na.rm = TRUE))
})
# works with dplyr
library(dplyr)
X3 <- X %>%
group_by(hyp) %>%
mutate(chlm = mean(chl, na.rm = TRUE))
test_that("handles grouped_df", {
expect_silent(as.mids(X3))
})
mice/tests/testthat/test-filter.R 0000644 0001762 0000144 00000001764 14330031606 016562 0 ustar ligges users
context("filter.mids")
imp <- mice(nhanes, m = 2, maxit = 1, print = FALSE, seed = 1)
imp_l <- imp
class(imp_l) <- "list"
test_that("throws error", {
# outcommented first two tests because they also throw a
# deprecated filter_ warning
# expect_error(filter("mids"))
# expect_error(filter(imp_l))
expect_error(filter(imp, "TRUE"))
expect_error(filter(imp, rep(1, nrow(nhanes))))
})
imp_f <- filter(imp, c(rep(TRUE, 13), rep(FALSE, 12)))
test_that("filtered mids is subset", {
expect_equal(complete(imp_f), complete(imp)[1:13, ])
expect_equal(imp_f$ignore, imp$ignore[1:13])
expect_equal(imp_f$where, imp$where[1:13, ])
expect_equal(imp_f$nmis, colSums(is.na(imp$data[1:13, ])))
expect_null(imp_f$chainMean)
})
imp_fa <- filter(imp, rep(TRUE, nrow(nhanes)))
imp2 <- mice.mids(imp, maxit = 1, printFlag = FALSE)
imp_fa2 <- mice.mids(imp_fa, maxit = 1, printFlag = FALSE)
test_that("other elements of mids are left unchanged", {
expect_equal(complete(imp2), complete(imp_fa2))
})
mice/tests/testthat/test-mice.impute.jomoImpute.R 0000644 0001762 0000144 00000001344 14330031606 021635 0 ustar ligges users context("mice.impute.jomoImpute")
data <- boys[c(1:10, 101:110, 501:510, 601:620, 701:710), ]
type <- c(2, 0, 0, 0, -2, 0, 1, 1, 0)
names(type) <- names(data)
z1 <- mice.impute.jomoImpute(data = data, type = type, format = "native")
test_that("jomoImpute returns native class", {
expect_is(z1, "mitml")
})
blocks <- make.blocks(list(c("bmi", "chl", "hyp"), "age"))
method <- c("jomoImpute", "pmm")
pred <- make.predictorMatrix(nhanes, blocks)
pred["B1", "hyp"] <- -2
# imp <- mice(nhanes, blocks = blocks, method = method, pred = pred,
# maxit = 1, seed = 1, print = FALSE)
# z <- complete(imp)
#
# test_that("mice can call jomoImpute", {
# expect_equal(sum(is.na(z$bmi)), 0)
# expect_equal(sum(is.na(z$chl)), 0)
# })
mice/tests/testthat/test-mice.impute.mpmm.R 0000644 0001762 0000144 00000001064 14347334322 020463 0 ustar ligges users context("mice.impute.mpmm")
set.seed(1)
beta2 <- beta1 <- .5
x <- rnorm(1000)
e <- rnorm(1000, 0, 1)
y <- beta1 * x + beta2 * x^2 + e
# dat <- data.frame(x = x, x2 = x^2, y = y) # worked
dat <- data.frame(y = y, x = x, x2 = x^2) # did not work
m <- as.logical(rbinom(1000, 1, 0.25))
dat[m, c("x", "x2")] <- NA
blk <- list("y", c("x", "x2"))
meth <- c("", "mpmm")
imp <- mice(dat, blocks = blk, method = meth, print = FALSE, m = 1, maxit = 1)
test_that("mpmm() works for any column order in data", {
expect_identical(complete(imp)$x^2, complete(imp)$x2)
})
mice/tests/testthat/test-mice.impute.pmm.R 0000644 0001762 0000144 00000002053 14334523631 020304 0 ustar ligges users context("mice.impute.pmm")
xname <- c("age", "hgt", "wgt")
br <- boys[c(1:10, 101:110, 501:510, 601:620, 701:710), ]
r <- stats::complete.cases(br[, xname])
x <- br[r, xname]
y <- br[r, "tv"]
ry <- !is.na(y)
wy1 <- !ry
wy2 <- rep(TRUE, length(y))
wy3 <- rep(FALSE, length(y))
wy4 <- rep(c(TRUE, FALSE), times = c(1, length(y) - 1))
test_that("Returns requested length", {
expect_equal(length(mice.impute.pmm(y, ry, x)), sum(!ry))
expect_equal(length(mice.impute.pmm(y, ry, x, wy = wy1)), sum(wy1))
expect_equal(length(mice.impute.pmm(y, ry, x, wy = wy2)), sum(wy2))
expect_equal(length(mice.impute.pmm(y, ry, x, wy = wy3)), sum(wy3))
expect_equal(length(mice.impute.pmm(y, ry, x, wy = wy4)), sum(wy4))
})
test_that("Excludes donors", {
expect_false(all(c(15:25) %in% mice.impute.pmm(y, ry, x, exclude = c(15:25))))
})
imp1 <- mice(nhanes, printFlag = FALSE, seed = 123)
imp2 <- mice(nhanes, printFlag = FALSE, seed = 123, exclude = c(-1, 1032))
test_that("excluding unobserved values does not impact pmm", {
expect_identical(imp1$imp, imp2$imp)
})
mice/tests/testthat/test-mice.impute.2lonly.mean.R 0000644 0001762 0000144 00000003432 14330031606 021643 0 ustar ligges users context("mice.impute.2lonly.mean")
set.seed(66322)
y <- popmis$texp
y[rbinom(length(y), size = 1, prob = 0.5) == 1] <- NA
x <- popmis[, c("pupil", "school", "sex")]
ry <- !is.na(y)
wy1 <- !ry
wy2 <- rep(TRUE, length(y))
wy3 <- rep(FALSE, length(y))
wy4 <- rep(c(TRUE, FALSE), times = c(1, length(y) - 1))
type <- c(1, -2, 1)
yn <- y
test_that("Returns requested length, for numeric", {
expect_equal(length(mice.impute.2lonly.mean(y, ry, x, type, wy1)), sum(wy1))
expect_equal(length(mice.impute.2lonly.mean(y, ry, x, type, wy2)), sum(wy2))
expect_equal(length(mice.impute.2lonly.mean(y, ry, x, type, wy3)), sum(wy3))
expect_equal(length(mice.impute.2lonly.mean(y, ry, x, type, wy4)), sum(wy4))
})
# test extension to factors
set.seed(66322)
y <- popmis$texp
y <- cut(y, breaks = c(0, 5, 10, 20, 30))
y[rbinom(length(y), size = 1, prob = 0.5) == 1] <- NA
test_that("Returns requested length, for factor", {
expect_equal(length(mice.impute.2lonly.mean(y, ry, x, type, wy1)), sum(wy1))
expect_equal(length(mice.impute.2lonly.mean(y, ry, x, type, wy2)), sum(wy2))
expect_equal(length(mice.impute.2lonly.mean(y, ry, x, type, wy3)), sum(wy3))
expect_equal(length(mice.impute.2lonly.mean(y, ry, x, type, wy4)), sum(wy4))
})
# check whether imputes for numeric and factor are identical
# tn <- mice.impute.2lonly.mean(yn, ry, x, type, wy1)
# tf <- mice.impute.2lonly.mean(y, ry, x, type, wy1)
# check what happens if all values within a class are missing
yn[1:100] <- NA
imn <- mice.impute.2lonly.mean(yn, ry, x, type, wy1)
zn <- table(imn, useNA = "al")
y[1:100] <- NA
imf <- mice.impute.2lonly.mean(y, ry, x, type, wy1)
zf <- table(imf, useNA = "al")
test_that("Return NA for classes without values", {
expect_equal(as.numeric(zn[length(zn)]), 39)
expect_equal(as.numeric(zf[length(zf)]), 39)
})
mice/tests/testthat/test-mice.impute.durr.logreg.R 0000644 0001762 0000144 00000005007 14433155521 021746 0 ustar ligges users context("mice.impute.lasso.norm")
#########################
# TEST 1: Simple problem #
#########################
set.seed(123)
# generate data
n <- 1e3
y <- rnorm(n)
x <- y * .3 + rnorm(n, 0, .25)
x2 <- x + rnorm(n, 2, 3)
x <- cbind(x, x2)
y <- as.numeric(cut(y, 2)) - 1
# make missingness
y[sample(1:n, n * .3)] <- NA
ry <- !is.na(y)
wy <- !ry
# Use univariate imputation model
set.seed(123)
imps <- mice.impute.lasso.logreg(y, ry, x)
test_that("Returns a matrix of dimensionality sum(wy) x 1", {
expect_true(is.matrix(imps))
expect_equal(dim(imps), c(sum(wy), 1))
})
#########################
# TEST 2: Use it within mice call #
#########################
# Generate some dichotomous data
n <- 1e2
p <- 4
Sigma <- matrix(.7, nrow = p, ncol = p)
diag(Sigma) <- 1
X <- as.data.frame(MASS::mvrnorm(n, rep(0, p), Sigma))
# Discretize and impose miss
for (j in 1:2) {
X[, j] <- cut(X[, j], 2) # Make it discrete
X[sample(1:n, n * .3), j] <- NA # Impose missings
}
# Imputations
durr_default <- mice(X,
m = 2, maxit = 2, method = "lasso.logreg",
print = FALSE
)
durr_custom <- mice(X,
m = 2, maxit = 2, method = "lasso.logreg",
nfolds = 5,
print = FALSE
)
logreg_default <- mice(X,
m = 2, maxit = 2, method = "logreg",
print = FALSE
)
# Tests
test_that("mice call works", {
expect_equal(class(durr_custom), "mids")
})
test_that("mice call works w/ custom arguments", {
expect_equal(class(durr_custom), "mids")
})
test_that("same class as logreg default method", {
expect_equal(
class(complete(logreg_default)[, 1]),
class(complete(durr_default)[, 1])
)
})
#########################
# TEST 3: Perfect Prediction / Complete Separation #
#########################
set.seed(123)
# Generate some dichotomous data
n <- 1e2
p <- 4
Sigma <- matrix(.7, nrow = p, ncol = p)
diag(Sigma) <- 1
x <- MASS::mvrnorm(n, rep(0, p), Sigma)
# Create Perfect Predictor
y <- factor(x[, 1] < 0, labels = c("y", "n"))
# Missing values
y[sample(1:n, n * .3)] <- NA
ry <- !is.na(y)
wy <- !ry
# Imputation well behaved
wellBehaved <- tryCatch(
expr = {
mice.impute.lasso.logreg(y = y, ry = ry, x = x[, -1])
},
error = function(e) {
e
},
warning = function(w) {
w
}
)
# Imputation perfect prediction
perfectPred <- tryCatch(
expr = {
mice.impute.lasso.logreg(y = y, ry = ry, x = x)
},
error = function(e) {
e
},
warning = function(w) {
w
}
)
# Test
test_that("Complete separation results in same class as well behaved case", {
expect_true(all.equal(class(wellBehaved), class(perfectPred)))
})
mice/tests/testthat/test-mice-initialize.R 0000644 0001762 0000144 00000020421 14430511431 020340 0 ustar ligges users context("mice-initialize")
data <- nhanes
# case A: no predictorMatrix, blocks or formulas arguments
imp1 <- mice(data, print = FALSE, m = 1, maxit = 1)
pred <- imp1$predictorMatrix
form <- imp1$formulas
test_that("Case A finds blocks", {
expect_identical(names(imp1$blocks), colnames(data))
})
test_that("Case A finds formulas", {
expect_identical(
attr(terms(form[["bmi"]]), "term.labels"),
names(pred["bmi", ])[pred["bmi", ] == 1]
)
})
# case B: only predictorMatrix argument
pred1 <- matrix(1, nrow = 4, ncol = 4)
pred2 <- matrix(1, nrow = 2, ncol = 2)
pred3 <- matrix(1,
nrow = 2, ncol = 2,
dimnames = list(c("bmi", "hyp"), c("bmi", "hyp"))
)
pred4 <- matrix(1,
nrow = 2, ncol = 3,
dimnames = list(c("bmi", "hyp"), c("bmi", "hyp", "chl"))
)
imp1 <- mice(data, predictorMatrix = pred1, print = FALSE, m = 1, maxit = 1)
imp3 <- mice(data, predictorMatrix = pred3, print = FALSE, m = 1, maxit = 1)
test_that("Case B tests the predictorMatrix", {
expect_equal(nrow(imp1$predictorMatrix), 4L)
expect_error(mice(data,
predictorMatrix = pred2,
"Missing row/column names in `predictorMatrix`."
))
expect_equal(nrow(imp3$predictorMatrix), 2L)
expect_error(mice(data, predictorMatrix = pred4))
})
pred <- imp3$predictorMatrix
blocks <- imp3$blocks
test_that("Case B finds blocks", {
expect_identical(names(blocks), c("bmi", "hyp"))
})
form <- imp3$formulas
test_that("Case B finds formulas", {
expect_identical(
attr(terms(form[["bmi"]]), "term.labels"),
names(pred["bmi", ])[pred["bmi", ] == 1]
)
})
# Case C: Only blocks argument
imp1.0 <- mice(data, blocks = list("bmi", "chl", "hyp"), m = 1, maxit = 0, seed = 11)
imp2.0 <- mice(data, blocks = list(c("bmi", "chl"), "hyp"), m = 1, maxit = 0, seed = 11)
imp3.0 <- mice(data, blocks = list(all = c("bmi", "chl", "hyp")), m = 1, maxit = 0, seed = 11)
test_that("Case C imputations are identical after initialization", {
expect_identical(complete(imp1.0), complete(imp2.0))
expect_identical(complete(imp1.0), complete(imp3.0))
})
imp1 <- mice(data, blocks = list("bmi", "chl", "hyp"), print = FALSE, m = 1, maxit = 1, seed = 11)
imp2 <- mice(data, blocks = list(c("bmi", "chl"), "hyp"), print = FALSE, m = 1, maxit = 1, seed = 11)
imp3 <- mice(data, blocks = list(all = c("bmi", "chl", "hyp")), print = FALSE, m = 1, maxit = 1, seed = 11)
test_that("Case C finds blocks", {
expect_identical(names(imp2$blocks), c("B1", "hyp"))
expect_identical(names(imp3$blocks), c("all"))
})
test_that("Case C finds predictorMatrix", {
expect_identical(imp2$predictorMatrix["hyp", "hyp"], 0)
expect_identical(dim(imp3$predictorMatrix), c(1L, 4L))
})
test_that("Case C finds formulas", {
expect_identical(sort(all.vars(imp2$formulas[["B1"]])), sort(colnames(data)))
})
test_that("Case C yields same imputations for FCS and multivariate", {
expect_identical(complete(imp1), complete(imp2))
expect_identical(complete(imp1), complete(imp3))
})
# Case D: Only formulas argument
# univariate models
form1 <- list(
bmi ~ age + hyp + chl,
hyp ~ age + bmi + chl,
chl ~ age + bmi + hyp
)
imp1 <- mice(data,
formulas = form1, method = "norm.nob",
print = FALSE, m = 1, maxit = 1, seed = 12199
)
# same model using dot notation
form2 <- list(bmi ~ ., hyp ~ ., chl ~ .)
imp2 <- mice(data,
formulas = form2, method = "norm.nob",
print = FALSE, m = 1, maxit = 1, seed = 12199
)
# multivariate models (= repeated univariate)
form3 <- list(
bmi + hyp ~ age + chl,
chl ~ age + bmi + hyp
)
imp3 <- mice(data,
formulas = form3, method = "norm.nob",
print = FALSE, m = 1, maxit = 1, seed = 12199
)
# same model using dot notation
form4 <- list(bmi + hyp ~ ., chl ~ .)
imp4 <- mice(data,
formulas = form4, method = "norm.nob",
print = FALSE, m = 1, maxit = 1, seed = 12199
)
test_that("Case D yields same imputations for dot notation", {
expect_identical(complete(imp1), complete(imp2))
expect_identical(complete(imp3), complete(imp4))
})
test_that("Case D yields same imputations for FCS and multivariate", {
expect_equal(complete(imp1), complete(imp3))
expect_equal(complete(imp2), complete(imp4))
})
# Case E: predictMatrix and blocks
blocks1 <- make.blocks(c("bmi", "chl", "hyp", "age"))
blocks2 <- make.blocks(list(c("bmi", "chl"), "hyp"))
blocks3 <- make.blocks(list(all = c("bmi", "chl", "hyp")))
pred1 <- make.predictorMatrix(data, blocks = blocks1)
pred2 <- make.predictorMatrix(data, blocks = blocks2)
pred3 <- make.predictorMatrix(data, blocks = blocks3)
imp1 <- mice(data, blocks = blocks1, pred = pred1, m = 1, maxit = 1, print = FALSE)
imp1a <- mice(data, blocks = blocks1, pred = matrix(1, nr = 4, nc = 4), m = 1, maxit = 1, print = FALSE)
imp2 <- mice(data, blocks = blocks2, pred = pred2, m = 1, maxit = 1, print = FALSE)
imp2a <- mice(data, blocks = blocks2, pred = matrix(1, nr = 2, nc = 4), m = 1, maxit = 1, print = FALSE)
imp3 <- mice(data, blocks = blocks3, pred = pred3, m = 1, maxit = 1, print = FALSE)
imp3a <- mice(data, blocks = blocks3, pred = matrix(1, nr = 1, nc = 4), m = 1, maxit = 1, print = FALSE)
test_that("Case E borrows rownames from blocks", {
expect_identical(rownames(imp1a$predictorMatrix), names(blocks1))
expect_identical(rownames(imp2a$predictorMatrix), names(blocks2))
expect_identical(rownames(imp3a$predictorMatrix), names(blocks3))
})
test_that("Case E borrows colnames from data", {
expect_identical(colnames(imp1a$predictorMatrix), names(data))
expect_identical(colnames(imp2a$predictorMatrix), names(data))
expect_identical(colnames(imp3a$predictorMatrix), names(data))
})
test_that("Case E name setting fails on incompatible sizes", {
expect_error(
mice(data, blocks = blocks2, pred = matrix(1, nr = 2, nc = 2)),
"Unable to set column names of predictorMatrix"
)
expect_error(
mice(data, blocks = blocks2, pred = matrix(1, nr = 1, nc = 4)),
"Unable to set row names of predictorMatrix"
)
expect_error(mice(data, blocks = blocks2, pred = matrix(1, nr = 4, nc = 4)))
})
colnames(pred1) <- c("A", "B", "chl", "bmi")
pred2a <- pred2[, -(1:4), drop = FALSE]
test_that("Case E detects incompatible arguments", {
expect_error(
mice(data, blocks = blocks1, pred = pred1),
"Names not found in data: A, B"
)
expect_error(
mice(data, blocks = blocks1, pred = pred2),
"Names not found in blocks: B1"
)
expect_error(
mice(data, blocks = blocks2, pred = matrix(1, nr = 1, nc = 4)),
"Unable to set row names of predictorMatrix"
)
expect_error(mice(data, blocks = blocks2, pred = matrix(1, nr = 4, nc = 4)))
expect_error(
mice(data, blocks = blocks2, pred = pred2a),
"predictorMatrix has no rows or columns"
)
})
# Case F: predictMatrix and formulas
blocks1 <- make.blocks(c("bmi", "chl", "hyp", "age"))
blocks2 <- make.blocks(list(c("bmi", "hyp"), "hyp"))
pred1 <- make.predictorMatrix(data, blocks = blocks1)
pred2 <- make.predictorMatrix(data, blocks = blocks2)
form1 <- list(
bmi ~ age + hyp + chl,
hyp ~ age + bmi + chl,
chl ~ age + bmi + hyp
)
form2 <- list(bmi ~ ., hyp ~ ., chl ~ .)
form3 <- list(
bmi + hyp ~ age + chl,
chl ~ age + bmi + hyp
)
form4 <- list(bmi + hyp ~ ., chl ~ .)
# blocks1 and form1 are compatible
imp1 <- mice(data, formulas = form1, pred = matrix(1, nr = 4, nc = 4), m = 1, maxit = 1, print = FALSE, seed = 3)
test_that("Case F combines forms and pred in blocks", {
expect_identical(unname(attr(imp1$blocks, "calltype")), c(rep("formula", 3), "type"))
})
# dots and unnamed predictorMatrix
imp2 <- mice(data, formulas = form2, pred = matrix(1, nr = 4, nc = 4), m = 1, maxit = 1, print = FALSE, seed = 3)
test_that("Case F dots and specified form produce same imputes", {
expect_identical(complete(imp1), complete(imp2))
})
# error
test_that("Case F generates error if it cannot handle non-square predictor", {
expect_error(
mice(data, formulas = form2, pred = pred2),
"If no blocks are specified, predictorMatrix must have same number of rows and columns"
)
})
## Error in formulas[[h]] : subscript out of bounds
imp3 <- mice(data, formulas = form3, pred = pred1, m = 1, maxit = 0, print = FALSE, seed = 3)
imp3a <- mice(data, formulas = form3, pred = pred1, m = 1, maxit = 1, print = FALSE, seed = 3)
# err on matrix columns
nh <- nhanes
nh$hyp <- as.matrix(nh$hyp)
test_that("MICE does not accept data.frames with embedded matrix ", {
expect_error(
mice(nh),
"Cannot handle columns with class matrix: hyp"
)
})
mice/tests/testthat/test-check.visitSequence.R 0000644 0001762 0000144 00000000436 13666252075 021213 0 ustar ligges users context("check.visitSequence")
data <- nhanes
test_that("mice() takes numerical and character visitSequence", {
expect_silent(imp <- mice(data, visitSequence = 4:1, m = 1, print = FALSE))
expect_silent(imp <- mice(data, visitSequence = rev(names(data)), m = 1, print = FALSE))
})
mice/tests/testthat/test-md.pattern.R 0000644 0001762 0000144 00000001151 14330031606 017337 0 ustar ligges users context("md.pattern")
test_that("patterns run as expected", {
# boys
expect_silent(md.pattern(boys, plot = FALSE))
# nhanes
expect_silent(md.pattern(nhanes, plot = FALSE))
# one whole column missing (single pattern) - should not produce output
expect_silent(md.pattern(cbind(na.omit(nhanes), NA), plot = FALSE))
# no missings (no pattern) - should produce output
expect_output(md.pattern(na.omit(nhanes), plot = FALSE))
# feed single column - expect error
expect_error(md.pattern(nhanes$bmi))
# feed other than dataframe/matrix - expect error
expect_error(md.pattern(as.list(nhanes)))
#
})
mice/tests/testthat/test-D1.R 0000644 0001762 0000144 00000003173 14330031606 015535 0 ustar ligges users context("D1")
imp <- mice(nhanes2, print = FALSE, m = 10, seed = 219)
fit1 <- with(data = imp, expr = glm(hyp == "yes" ~ age + chl, family = binomial))
fit0 <- with(data = imp, expr = glm(hyp == "yes" ~ 1, family = binomial))
empty <- with(data = imp, expr = glm(hyp == "yes" ~ 0, family = binomial))
# stat1 <- pool.compare(fit1, fit0, method = "wald")
# deprecated because it relies on full vcov, which is not present anymore
# in the mipo object
# the next tests were remove because they failed on many
# systems, not yet clear what the cause is (#128)
# This is solved in #132
# three new ways to compare fit1 to the intercept-only model
z1 <- D1(fit1, fit0)
z2 <- mitml::testModels(as.mitml.result(fit1), as.mitml.result(fit0), df.com = 21)
z3 <- D1(fit1)
test_that("compares fit1 to the intercept-only model", {
expect_identical(z1$result, z2$test)
expect_identical(z1$test, z3$test)
})
# two ways to compare fit1 to the empty model
z4 <- D1(fit1, empty)
z5 <- mitml::testModels(as.mitml.result(fit1), as.mitml.result(empty), df.com = 21)
test_that("compares fit1 to empty model", {
expect_identical(z4$result, z5$test)
})
context("D2")
z1 <- D2(fit1, fit0)
z2 <- mitml::testModels(as.mitml.result(fit1), as.mitml.result(fit0), method = "D2")
z3 <- D2(fit1)
test_that("compares fit1 to the intercept-only model", {
expect_identical(z1$result, z2$test)
expect_identical(z1$test, z3$test)
})
# two ways to compare fit1 to the empty model
z4 <- D2(fit1, empty)
z5 <- mitml::testModels(as.mitml.result(fit1), as.mitml.result(empty), method = "D2")
test_that("compares fit1 to empty model", {
expect_identical(z4$result, z5$test)
})
mice/tests/testthat/test-loggedEvents.R 0000644 0001762 0000144 00000000562 14330031606 017716 0 ustar ligges users context("mice: loggedEvents")
imp1 <- mice(nhanes, m = 2, print = FALSE)
# copy of data, different names
data2 <- cbind(nhanes, nhanes)
colnames(data2)[5:8] <- c("age2", "bmi2", "hyp2", "chl2")
imp2 <- suppressWarnings(mice(data2, m = 2, print = FALSE))
test_that("loggedEvents is NULL", {
expect_null(imp1$loggedEvents)
expect_type(imp2$loggedEvents, "list")
})
mice/tests/testthat/test-blots.R 0000644 0001762 0000144 00000001145 14330031606 016411 0 ustar ligges users context("blots")
# global change of donors argument
blocks1 <- name.blocks(list(c("bmi", "chl"), "hyp"))
imp0 <- mice(nhanes, blocks = blocks1, donors = 10, m = 1, maxit = 1, print = FALSE)
# vary donors, depending on block
blots1 <- list(B1 = list(donors = 10), hyp = list(donors = 1))
imp1 <- mice(nhanes, blocks = blocks1, blots = blots1, m = 1, maxit = 1, print = FALSE)
test_that("errors when mixing same global and local argument", {
expect_error(
mice(nhanes, blocks = blocks1, blots = blots1, donors = 7, print = FALSE),
'formal argument "donors" matched by multiple actual arguments'
)
})
mice/tests/testthat/test-anova.R 0000644 0001762 0000144 00000002260 14330031606 016371 0 ustar ligges users context("anova")
imp <- mice(nhanes2, m = 10, print = FALSE, seed = 71242)
m2 <- with(imp, lm(chl ~ age + bmi))
m1 <- with(imp, lm(chl ~ bmi))
m0 <- with(imp, lm(chl ~ 1))
# anova methods
test_that("anova.mira() produces silent D1 and D3", {
expect_silent(z1 <- anova(m2, m1, m0))
expect_silent(z3 <- anova(m2, m1, m0, method = "D3"))
})
test_that("anova.mira() produces silent with D2", {
expect_silent(z2a <- anova(m2, m1, m0, method = "D2"))
expect_silent(z2b <- anova(m2, m1, m0, method = "D2", use = "likelihood"))
})
context("Cox model tests")
library(survival)
set.seed(1)
data <- survival::lung
data$age[rbinom(nrow(data), size = 1, prob = 0.2) == 1] <- NA
data$sex[rbinom(nrow(data), size = 1, prob = 0.2) == 1] <- NA
data$ph.ecog[rbinom(nrow(data), size = 1, prob = 0.2) == 1] <- NA
imp <- mice(data, print = FALSE)
m1 <- with(imp, coxph(Surv(time, status) ~ age))
m2 <- with(imp, coxph(Surv(time, status) ~ age + sex))
m3 <- with(imp, coxph(Surv(time, status) ~ age + sex + ph.ecog))
test_that("runs tests for the Cox model", {
expect_silent(pool(m1))
expect_silent(D1(m2, m1))
expect_silent(D2(m2, m1))
expect_error(D3(m2, m1))
expect_silent(anova(m3, m2, m1))
})
mice/tests/testthat/test-check.formula.R 0000644 0001762 0000144 00000004534 14330031606 020014 0 ustar ligges users context("check.formulas")
data <- nhanes
where <- is.na(data)
# blocks <- name.blocks(list("bmi", "age", "chl"))
# ini <- mice(data, blocks = blocks, maxit = 0)
#
# # classic type specification
# setup <- list(blocks = blocks,
# predictorMatrix = ini$predictorMatrix,
# formulas = NULL)
# # v1 <- mice:::check.formulas(setup, data)
#
# # using a formula
# #formulas <- v1$formulas
# setup <- list(blocks = blocks,
# predictorMatrix = ini$predictorMatrix,
# formulas = formulas)
# #v2 <- mice:::check.formulas(setup, data)
# #v2$formulas
#
# test_that("updates `mode.formula` attribute", {
# # expect_false(identical(v2$formulas, v2$formulas.arg))
# # expect_identical(v2$formulas[[1]], v2$formulas.arg[[1]])
# })
#
# # try dot in formula
# formulas <- list(bmi ~ ., age ~ ., chl ~ .)
# formulas <- name.formulas(formulas)
# setup <- list(blocks = blocks,
# predictorMatrix = ini$predictorMatrix,
# formulas = formulas)
# #v3 <- mice:::check.formulas(setup, data)
# #v3$formulas
#
# # classic specification using predictorMatrix
# imp1 <- mice(nhanes, seed = 51212, print = FALSE, m = 1)
# cmp1 <- complete(imp1)
#
# # formula specification
# form <- list(age ~ ., bmi ~ ., hyp ~., chl ~ .)
# imp2 <- mice(nhanes, formulas = form, seed = 51212, print = FALSE, m = 1)
# cmp2 <- complete(imp2)
#
# test_that("predictorMatrix and formula yield same imputations", {
# expect_identical(cmp1, cmp2)
# expect_identical(imp1$imp, imp2$imp)
# })
# formula specification
form <- name.blocks(list(bmi ~ ., hyp ~ ., chl ~ .))
imp3 <- mice(nhanes, formulas = form, seed = 51212, print = FALSE, m = 1)
cmp3 <- complete(imp3)
# old.form <- c("", "bmi ~ chl + hyp", "hyp ~ bmi + chl", "chl ~ bmi + hyp")
# imp <- mice(nhanes, formula = old.form, m = 1, maxit = 2, print = FALSE)
#
# form1 <- list(bmi = ~ 1, chl = ~ 1, hyp = ~ 1)
# # impute given predictors
# imp1 <- mice(nhanes, formula = form1, m = 1, maxit = 2, method = "norm.predict",
# print = FALSE, seed = 1)
# # impute the mean
# imp2 <- mice(nhanes, formula = form1, m = 1, maxit = 2, method = "norm.predict",
# print = FALSE, seed = 1)
#
# form2 <- list(bmi = "hyp ~ 1", chl = "hyp ~ 1", hyp = "hyp ~ 1")
# imp3 <- mice(nhanes, formula = form2, m = 1, maxit = 2, method = "norm.predict",
# print = FALSE, seed = 1)
#
mice/tests/testthat/test-remove.lindep.R 0000644 0001762 0000144 00000003314 13666252075 020055 0 ustar ligges users context("remove.lindep")
set.seed(1)
td <- matrix(rnorm(20), nrow = 5, ncol = 4)
dimnames(td) <- list(1:5, LETTERS[1:4])
y <- td[, 1]
ry <- rep(TRUE, 5)
# data frame for storing the event log
state <- list(it = 0, im = 0, dep = "y", meth = "test", log = FALSE)
loggedEvents <- data.frame(it = 0, im = 0, dep = "", meth = "", out = "")
fr <- 2
state$meth <- "k1"
x <- td[, 2:4]
k1 <- mice:::remove.lindep(x, y, ry, frame = fr)
state$meth <- "k2"
x[, 2] <- x[, 1]
k2 <- mice:::remove.lindep(x, y, ry, frame = fr)
state$meth <- "k3"
x[, 3] <- 2 * x[, 1]
k3 <- mice:::remove.lindep(x, y, ry, frame = fr)
state$meth <- "k4"
x <- td[, 2:4]
y <- x[, 2]
k4 <- mice:::remove.lindep(x, y, ry, frame = fr)
state$meth <- "k5"
x <- td[, 2:4]
y <- x[, 2]
x[, 3] <- x[, 1] <- x[, 2]
k5 <- mice:::remove.lindep(x, y, ry, frame = fr)
# one column x, same as y --> FALSE
state$meth <- "k6"
x <- td[, 2, drop = FALSE]
y <- x[, 1]
k6 <- mice:::remove.lindep(x, y, ry, frame = fr)
# one column x, different from u --> TRUE
state$meth <- "k7"
x <- td[, 2, drop = FALSE]
y <- td[, 1]
k7 <- mice:::remove.lindep(x, y, ry, frame = fr)
# two columns, same x and y --> FALSE, FALSE
state$meth <- "k8"
x <- td[, 2:3, drop = FALSE]
x[, 2] <- x[, 1]
y <- x[, 1]
k8 <- mice:::remove.lindep(x, y, ry, frame = fr)
loggedEvents
test_that("removes copies", {
expect_identical(unname(k1), c(TRUE, TRUE, TRUE))
expect_identical(unname(k2), c(FALSE, TRUE, TRUE))
# expect_identical(unname(k3), c(FALSE, FALSE, TRUE))
expect_identical(unname(k4), c(TRUE, FALSE, TRUE))
expect_identical(unname(k5), c(FALSE, FALSE, FALSE))
expect_identical(unname(k6), c(FALSE))
expect_identical(unname(k7), c(TRUE))
expect_identical(unname(k8), c(FALSE, FALSE))
})
mice/tests/testthat/test-mira.R 0000644 0001762 0000144 00000000270 13666252075 016234 0 ustar ligges users context("mira")
imp <- mice(nhanes, print = FALSE, maxit = 1, seed = 121, m = 1)
fit <- with(imp, sd(bmi))
test_that("list prints without an error", {
expect_output(print(fit))
})
mice/tests/testthat/test-D3.R 0000644 0001762 0000144 00000007434 14433400023 015540 0 ustar ligges users context("D3")
# The following test was contributed by jawitte
# https://github.com/amices/mice/issues/226
set.seed(1)
A <- rnorm(100)
B <- 0.1 * A + rnorm(100)
fit1 <- lapply(1:5, function(m) {
lm(B ~ A)
})
fit0 <- lapply(1:5, function(m) {
lm(B ~ 1)
})
# outcomment to evade dependency of lmtest
# x1 <- lmtest::lrtest(fit1[[1]], fit0[[1]])
x1 <- structure(list(`#Df` = c(3, 2),
LogLik = c(-137.087912980007, -137.516434459951),
Df = c(NA, -1), Chisq = c(NA, 0.857042959888474),
`Pr(>Chisq)` = c(NA, 0.354567523408569)),
class = c("anova", "data.frame"),
row.names = c("1", "2"),
heading = c("Likelihood ratio test\n", "Model 1: B ~ A\nModel 2: B ~ 1"))
x2 <- D3(fit1 = fit1, fit0 = fit0)
x3 <- mitml::testModels(fit1, fit0, method = "D3")
# tests for complete data
test_that("lm, complete data: D3() and lrtest() calculate same test statistic", {
expect_equal(x1$Chisq[2], x2$result[1])
})
test_that("lm, complete data: testModels() and lrtest() calculate same test statistic", {
expect_equal(x1$Chisq[2], x3$test[1])
})
# FIXME:
# for imputed data, there are discrepancies between mitml and mice
# the tests below compare mitml and mice, but none of these seem to work
# so I have a outcommented the critical lines
imp <- mice(nhanes, print = FALSE, m = 10, seed = 219)
fit1 <- with(data = imp, expr = lm(hyp ~ age + chl))
fit0 <- with(data = imp, expr = lm(hyp ~ 1))
empty <- with(data = imp, expr = lm(hyp ~ 0))
# stat1 <- pool.compare(fit1, fit0, method = "likelihood")
z1 <- D3(fit1, fit0)
z2 <- mitml::testModels(as.mitml.result(fit1), as.mitml.result(fit0), method = "D3")
# This test fails
# FIXME
# test_that("lm: mice and mitml calculate same F", {
# expect_equal(z1$result[1], z2$test[1])
# })
# using lmer
suppressPackageStartupMessages(library(mitml, quietly = TRUE))
library(lme4, quietly = TRUE)
# library(broom.mixed, quietly = TRUE)
data(studentratings)
fml <- ReadDis + SES ~ ReadAchiev + (1 | ID)
set.seed(26262)
imp <- mitml::panImpute(studentratings,
formula = fml, n.burn = 1000, n.iter = 100, m = 5,
silent = TRUE
)
implist <- mitml::mitmlComplete(imp, print = 1:5)
fit0 <- with(implist, lmer(ReadAchiev ~ (1 | ID), REML = FALSE))
fit1 <- with(implist, lmer(ReadAchiev ~ ReadDis + SES + (1 | ID), REML = FALSE))
# outcommented to evade dependency on broom.mixed under the _R_CHECK_DEPENDS_ONLY=true flag
# # likelihood test
# z3 <- D3(fit1, fit0)
# z4 <- mitml::testModels(fit1, fit0, method = "D3")
# This test fails.
# FIXME
# test_that("lmer: mice and mitml calculate same F", {
# expect_equal(z3$result[1], z4$test[1])
# })
# glm
# imp <- mice(nhanes2, print = FALSE, m = 10, seed = 219)
#
# fit1 <- with(data = imp, expr = glm(hyp == "yes" ~ age + chl, family = binomial))
# fit0 <- with(data = imp, expr = glm(hyp == "yes" ~ 1, family = binomial))
# empty <- with(data = imp, expr = glm(hyp == "yes" ~ 0, family = binomial))
#
# model dev1.L does not look right, negative Dm, convergence problems
# FIXME
# z5 <- D3(fit1, fit0)
# mitml can't do this case
# z6 <- mitml::testModels(as.mitml.result(fit1), as.mitml.result(fit0), method = "D3")
# crashes on terms
# FIXME
# z5a <- D3(fit1, empty)
# This test fails.
# FIXME
# test_that("glm: mice and mitml calculate same F", {
# expect_equal(z5$result[1], z6$test[1])
# })
# data with factors
imp <- mice(nhanes2, print = FALSE, m = 10, seed = 219)
fit1 <- with(data = imp, expr = lm(bmi ~ age + chl + hyp))
fit0 <- with(data = imp, expr = lm(bmi ~ age))
empty <- with(data = imp, expr = lm(bmi ~ 0))
z7 <- D3(fit1, fit0)
z8 <- mitml::testModels(as.mitml.result(fit1), as.mitml.result(fit0), method = "D3")
# This test fails.
# FIXME
# test_that("factors: mice and mitml calculate same F", {
# expect_equal(z7$result[1], z8$test[1])
# })
mice/tests/testthat/test-with.R 0000644 0001762 0000144 00000001233 14433400023 016234 0 ustar ligges users # See #292: with.mids() using eval_tidy() breaks compatibility with metafor
# outcommented to evade dependency on metafor
# dat <- dat.bcg
# dat <- escalc(measure = "RR", ai = tpos, bi = tneg, ci = cpos, di = cneg, data = dat)
# dat$ablat[c(2, 4, 8)] <- NA
#
# predMatrix <- make.predictorMatrix(dat)
# predMatrix[, ] <- 0
# predMatrix["ablat", c("yi", "year")] <- 1
#
# impMethod <- make.method(dat)
# impMethod["ablat"] <- "pmm"
# impMethod
# imp <- mice(dat, print = FALSE, predictorMatrix = predMatrix, method = impMethod, seed = 1234)
# test_that("does not break metafor package", {
# expect_silent(fit <- with(imp, rma(yi, vi, mods = ~ ablat + year)))
# })
mice/tests/testthat/test-tidiers.R 0000644 0001762 0000144 00000002020 14334445701 016733 0 ustar ligges users context("tidiers")
data(nhanes)
imp <- mice::mice(nhanes, maxit = 2, m = 2, seed = 1, print = FALSE, use.matcher = TRUE)
fit_mira <- with(imp, lm(chl ~ age + bmi))
fit_mipo <- mice::pool(fit_mira)
test_that("glance.mipo: nhanes lm", {
tmp <- glance(fit_mipo)
expect_true(inherits(tmp, "data.frame"))
expect_equal(tmp$adj.r.squared[1], 0.4966436, tolerance = .00001)
expect_equal(tmp$r.squared[1], 0.539119, tolerance = .00001)
})
test_that("tidy.mipo: nhanes lm", {
tmp <- tidy(fit_mipo)
expect_true(inherits(tmp, "data.frame"))
expect_equal(dim(tmp), c(3, 13))
tmp <- tidy(fit_mipo, conf.int = TRUE)
expect_true(inherits(tmp, "data.frame"))
expect_equal(dim(tmp), c(3, 15))
expect_equal(tmp$conf.low, c(
-171.676808396086, -12.5277617578218,
3.42203157045941
))
tmp <- tidy(fit_mipo, conf.int = TRUE, conf.level = .99)
expect_true(inherits(tmp, "data.frame"))
expect_equal(dim(tmp), c(3, 15))
expect_equal(tmp$conf.low, c(
-216.910255354075, -63.8124944550467,
2.16193377446054
))
})
mice/tests/testthat/test-mice.impute.2l.norm.R 0000644 0001762 0000144 00000000511 13666252075 021007 0 ustar ligges users context("mice.impute.2l.norm")
d1 <- brandsma[1:200, c("sch", "lpo")]
pred <- make.predictorMatrix(d1)
pred["lpo", "sch"] <- -2
test_that("mice::mice.impute.2l.norm() runs empty model", {
expect_silent(imp <- mice(d1, method = "2l.norm", print = FALSE, pred = pred, m = 1, maxit = 1))
expect_false(anyNA(complete(imp)))
})
mice/tests/testthat/test-cbind.R 0000644 0001762 0000144 00000007710 14330031606 016351 0 ustar ligges users context("cbind.mids")
data <- nhanes
data1 <- data[, c("age", "bmi")]
data2 <- data[, c("hyp", "chl")]
imp1 <- mice(data1, m = 2, maxit = 1, print = FALSE)
imp2 <- mice(data2, m = 2, maxit = 1, print = FALSE)
imp <- cbind(imp1, imp2)
test_that("combines imputations", {
expect_identical(ncol(complete(imp)), 4L)
expect_identical(complete(imp1), complete(imp)[, c("age", "bmi")])
})
# when using blocks
data <- nhanes
data1 <- data[, c("age", "bmi")]
data2 <- data[, c("hyp", "chl")]
imp1 <- mice(data1, m = 1, maxit = 1, print = FALSE)
imp2 <- mice(data2, blocks = list(c("hyp", "chl")), m = 1, maxit = 1, print = FALSE)
imp <- cbind(imp1, imp2)
test_that("combines imputations with blocks", {
expect_identical(ncol(complete(imp)), 4L)
expect_identical(complete(imp1), complete(imp)[, c("age", "bmi")])
})
# handling of duplicate variable names
data <- nhanes
data1 <- data[, c("age", "bmi", "hyp")]
data2 <- data[, c("hyp", "chl")]
imp1 <- mice(data1, m = 1, maxit = 1, print = FALSE)
imp2 <- mice(data2, m = 1, maxit = 1, print = FALSE)
imp <- cbind(imp1, imp2)
impc <- mice.mids(imp, max = 2, print = FALSE)
test_that("duplicate variable adds a column", {
expect_identical(ncol(complete(impc)), 5L)
})
# handling of duplicate blocks
imp1 <- mice(data1, blocks = list(c("age", "bmi"), "hyp"), m = 1, maxit = 1, print = FALSE)
imp2 <- mice(data2, blocks = list(c("hyp", "chl")), m = 1, maxit = 1, print = FALSE)
imp <- cbind(imp1, imp2)
impc <- mice.mids(imp, max = 2, print = FALSE)
test_that("duplicate blocks names renames block", {
expect_identical(names(impc$blocks)[3], "B1.1")
})
# cbind - no second argument
imp1 <- mice(nhanes, blocks = list(c("bmi", "chl"), "hyp"), print = FALSE, maxit = 1, m = 1)
imp2 <- cbind(imp1)
imp3 <- cbind(imp1, NULL)
imp4 <- cbind(imp1, character(0))
test_that("returns imp1 object if there is nothing to bind", {
expect_identical(imp2, imp1)
expect_identical(imp3, imp1)
expect_identical(imp4, imp1)
})
# cbind - unnamed constant
imp2 <- cbind(imp1, 1)
imp3 <- cbind(imp1, NA)
imp4 <- cbind(imp1, "male")
test_that("replicates unnamed constant", {
expect_identical(ncol(complete(imp2)), 5L)
expect_identical(ncol(complete(imp3)), 5L)
expect_identical(ncol(complete(imp4)), 5L)
})
imp6 <- cbind(imp1, int = 51:75, out = 15, NA)
test_that("appends names vectors and constants", {
expect_identical(ncol(complete(imp6)), 7L)
expect_error(
cbind(imp1, c(NA, 9)),
"arguments imply differing number of rows: 25, 2"
)
})
# matrix, factor, data.frame
# NOTE: cbind() dispatches to wrong function if there is a data.frame
# so use cbind.mids()
imp8 <- mice:::cbind.mids(imp1,
ma = matrix(1:50, nrow = 25, ncol = 2),
age = nhanes2$age,
df = nhanes2[, c("hyp", "chl")]
)
test_that("appends matrix, factor and data.frame", {
expect_identical(ncol(complete(imp8)), 9L)
})
# impc <- mice.mids(imp8, max = 2, print = FALSE)
# NOTE: now using own version of cbind()
imp9 <- cbind(imp1,
ma = matrix(1:50, nrow = 25, ncol = 2),
age = nhanes2$age,
df = nhanes2[, c("hyp", "chl")]
)
test_that("appends matrix, factor and data.frame", {
expect_identical(ncol(complete(imp9)), 9L)
})
impc <- mice.mids(imp9, max = 2, print = FALSE)
test_that("combined object works as input to mice.mids", {
expect_true(is.mids(impc))
})
test_that("cbind does not throw a warning (#114)", {
expect_silent(cbind(ordered(c(1, 2))))
})
# # cbind data.frame (rename to age.1)
# imp1 <- mice(nhanes, blocks = list(c("bmi", "chl"), "hyp"), print = FALSE, maxit = 1, m = 1)
# agevar <- nhanes$age
# agevar[1:5] <- NA
# imp2 <- mice:::cbind.mids(imp1, data.frame(age = agevar, hyp = "test"))
# imp3 <- mice.mids(imp2, max = 2, print = FALSE)
# complete(imp3)
#
# # cbind data.frame (use quoted name)
# imp1 <- mice(nhanes, blocks = list(c("bmi", "chl"), "hyp"), print = FALSE, maxit = 1, m = 1)
# agevar <- nhanes$age
# agevar[1:5] <- NA
# imp2 <- mice:::cbind.mids(imp1, age = agevar, hyp = "test")
# imp3 <- mice.mids(imp2, max = 2, print = FALSE)
# complete(imp3)
#
mice/tests/testthat/test-mice.impute.2lonly.norm.R 0000644 0001762 0000144 00000002520 14330031606 021673 0 ustar ligges users context("mice.impute.2lonly.norm")
## https://stackoverflow.com/questions/58266785/mice-2l-pan-multilevel-multiple-imputation-error-missing-values-in-pred-not-all?sem=2
library("pan")
# Not multilevel to illustrate need
set.seed(100)
data <- data.frame(
patid = rep(1:4, each = 5),
sex = rep(c(1, 2, 1, 2), each = 5),
crp = c(
68, 78, 93, NA, 143,
5, 7, 9, 13, NA,
97, NA, 56, 52, 34,
22, 30, NA, NA, 45
)
)
pred <- make.predictorMatrix(data)
pred[, "patid"] <- -2
# only missing value (out of five) for patid == 1
data[3, "sex"] <- NA
test_that("2lonly.norm stops with partially missing level-2 data", {
expect_error(
mice(data,
method = c("", "2lonly.norm", "2l.pan"),
predictorMatrix = pred, maxit = 1, m = 2, print = FALSE
),
"Method 2lonly.norm found the following clusters with partially missing\n level-2 data: 1\n Method 2lonly.mean can fix such inconsistencies."
)
})
set.seed(66322)
y <- popmis[1:200, "texp"]
x <- popmis[1:200, c("pupil", "school", "sex")]
y[x$school %in% 1:3] <- NA
ry <- !is.na(y)
wy1 <- !ry
wy2 <- rep(TRUE, length(y))
wy3 <- rep(FALSE, length(y))
wy4 <- rep(c(TRUE, FALSE), times = c(1, length(y) - 1))
type <- c(1, -2, 1)
yn <- y
y <- as.numeric(y)
set.seed(1)
z1 <- mice.impute.2lonly.norm(y, ry = ry, x, type)
z2 <- mice.impute.2lonly.pmm(y, ry = ry, x, type)
mice/tests/testthat/test-formulas.R 0000644 0001762 0000144 00000001233 14330031606 017114 0 ustar ligges users context("formulas")
data <- nhanes
test_that("model.matrix() deletes incomplete cases", {
expect_identical(dim(model.matrix(~age, data)), c(25L, 2L))
expect_identical(dim(model.matrix(~chl, data)), c(15L, 2L))
expect_identical(dim(model.matrix(~ poly(age), data)), c(25L, 2L))
expect_error(
model.matrix(~ poly(chl), data),
"missing values are not allowed in 'poly'"
)
expect_identical(dim(model.matrix(~ poly(chl, raw = TRUE), data)), c(15L, 2L))
})
# in MICE we can now use poly()
form <- list(bmi ~ poly(chl, 2) + age + hyp)
pred <- make.predictorMatrix(nhanes)
imp1 <- mice(data, form = form, pred = pred, m = 1, maxit = 2, print = FALSE)
mice/tests/testthat/test-parlmice.R 0000644 0001762 0000144 00000005071 14335404116 017071 0 ustar ligges users # Same seed - multiple cores -
# Result: Imputations not equal between mice and parlmice
test_that("Warning and Imputations between mice and parlmice are unequal", {
skip_if_not(parallel::detectCores() > 2)
expect_warning(A <- parlmice(nhanes, m = 2, seed = 123))
B <- mice(nhanes, m = 2, print = FALSE, seed = 123)
expect_false(all(complete(A, "long") == complete(B, "long")))
})
# Same seed - single core -
# Result: Imputations equal between mice and parlmice
test_that("Imputations are equal between mice and parlmice", {
expect_warning(C <- parlmice(nhanes, n.core = 1, n.imp.core = 5, seed = 123))
D <- mice(nhanes, m = 5, print = FALSE, seed = 123)
expect_identical(complete(C, "long"), complete(D, "long"))
})
# Should return m = 8
test_that("Cores and n.imp.core specified. Override m", {
expect_warning(I <- parlmice(nhanes, n.core = 2, n.imp.core = 4))
expect_identical(I$m, 2 * 4)
})
# Should return m = 3x5=15
test_that("n.imp.core not specified", {
expect_warning(J <- parlmice(nhanes, n.core = 2))
expect_identical(J$m, 2 * 5)
})
# Should return m = 2x7=42
test_that("n.imp.core not specified", {
expect_warning(K <- parlmice(nhanes, n.core = 2, m = 7))
expect_identical(K$m, 2 * 7)
})
# Should return error
test_that("n.core larger than logical CPU cores", {
expect_error(suppresWarnings(parlmice(nhanes, n.core = parallel::detectCores() + 1)))
})
# # NOT RUN ON R CMD CHECK AND CRAN CHECK - TOO MANY PARALLEL PROCESSES SPAWNED
# # Should return m = n.imp.core * parallel::detectCores() - 1
# test_that("Warning because n.core not specified", {
# expect_warning(H <- parlmice(nhanes, n.imp.core = 3))
# expect_identical(H$m, 3 * (parallel::detectCores() - 1))
# })
#
# #Same cluster.seed - multiple cores
# #Result: Imputations equal between parlmice instances
# imp1 <- parlmice(nhanes, m=2, cluster.seed = 123)
# imp2 <- parlmice(nhanes, m=2, cluster.seed = 123)
# test_that("cluster.seed", {
# expect_equal(imp1, imp2)
# })
#
# #Should run without failure
# df <- boys
# meth <- make.method(df)
# pred <- make.predictorMatrix(df)
# visit <- 9:1
# imp3 <- parlmice(df, method = meth,
# predictorMatrix = pred,
# visitSequence = visit,
# n.core = 2,
# n.imp.core = 4,
# maxit = 3,
# cluster.seed = 123)
# test_that("Runs when overriding defaults", {
# expect_identical(imp3$pred, pred)
# expect_identical(imp3$iteration, 3)
# expect_identical(imp3$method, meth)
# expect_identical(imp3$visitSequence, names(df)[visit])
# expect_identical(imp3$m, 2*4)
# })
mice/tests/testthat/test-pool.R 0000644 0001762 0000144 00000013411 14436064333 016250 0 ustar ligges users context("pool")
# set the random generator to V3.5.0 to ensure that this test
# passes in V3.6.0 and later
# see mail Kurt Hornik, dated 06mar19
# FIXME: consider using the new generator once V3.6.0 is out,
# at the expense of breaking reproducibility of the examples in
# https://stefvanbuuren.name/fimd/
suppressWarnings(RNGversion("3.5.0"))
imp <- mice(nhanes2, print = FALSE, maxit = 2, seed = 121, use.matcher = TRUE)
fit <- with(imp, lm(bmi ~ chl + age + hyp))
est <- pool(fit)
# fitlist <- fit$analyses
# est <- mice:::pool.fitlist(fitlist)
mn <- c(18.76175, 0.05359003, -4.573652, -6.635969, 2.163629)
se <- c(4.002796, 0.02235067, 2.033986, 2.459769, 2.02898)
test_that("retains same numerical result", {
expect_equal(unname(getqbar(est)), mn, tolerance = 0.00001)
expect_equal(unname(summary(est)[, "std.error"]), se, tolerance = 0.00001)
})
imp <- mice(nhanes2, print = FALSE, m = 10, seed = 219)
fit0 <- with(data = imp, expr = glm(hyp == "yes" ~ 1, family = binomial))
fit1 <- with(data = imp, expr = glm(hyp == "yes" ~ chl + bmi, family = binomial))
D1(fit1, fit0)
D3(fit1, fit0)
# test_that("retains same numerical result", {
# expect_equal(round(as.vector(stat1$pvalue), 3), 0.188)
# expect_equal(stat2$pvalue, 0)
# })
#
# Venables & Ripley, 2nd, p 235
birthwt <- MASS::birthwt
bwt <- with(
birthwt,
data.frame(
low = factor(low),
age = age,
lwt = lwt,
race = factor(race, labels = c("white", "black", "other")),
smoke = smoke > 0,
ptd = factor(ptl > 0),
ht = ht > 0,
ui = ui > 0,
ftv = factor(ftv)
)
)
levels(bwt$ftv)[-(1:2)] <- "2"
birthwt.glm <- glm(low ~ ., family = binomial, data = bwt)
summary(birthwt.glm)
birthwt.step <- step(birthwt.glm, trace = FALSE)
LLlogistic <- function(formula, data, coefs) {
### Calculates -2 loglikelihood of a model.
logistic <- function(mu) exp(mu) / (1 + exp(mu))
Xb <- model.matrix(formula, data) %*% coefs
y <- model.frame(formula, data)[1][, 1]
if (is.factor(y)) y <- (0:1)[y]
p <- logistic(Xb)
## in case values of categorical var are other than 0 and 1.
y <- (y - min(y)) / (max(y) - min(y))
term1 <- term2 <- rep(0, length(y))
term1[y != 0] <- y[y != 0] * log(y[y != 0] / p[y != 0])
term2[y == 0] <- (1 - y[y == 0]) * log((1 - y[y == 0]) / (1 - p[y == 0]))
2 * sum(term1 + term2)
}
model1 <- glm(low ~ ., family = binomial, data = bwt)
model0 <- update(model1, formula = . ~ . - age - ftv)
model.null <- update(model1, formula = . ~ 1)
ll1 <- LLlogistic(formula = formula(model1), data = bwt, coefs = coef(model1))
ll0 <- LLlogistic(formula = formula(model0), data = bwt, coefs = coef(model0))
llnull <- LLlogistic(formula = formula(model.null), data = bwt, coefs = coef(model.null))
identical(deviance(model1), ll1, num.eq = FALSE)
identical(deviance(model0), ll0, num.eq = FALSE)
identical(deviance(model.null), llnull, num.eq = FALSE)
# try out coef.fix for binary data
f1 <- fix.coef(model1, beta = coef(model1))
broom::glance(model1)
broom::glance(f1)
identical(broom::glance(f1)$deviance, broom::glance(model1)$deviance)
beta <- coef(model1)
beta["age"] <- 0
beta["smokeTRUE"] <- 0
f2 <- fix.coef(model1, beta)
broom::glance(f2)$deviance
set.seed(123)
bwt.mis <- bwt
bwt.mis$smoke[runif(nrow(bwt)) < 0.001] <- NA
bwt.mis$lwt[runif(nrow(bwt)) < 0.01] <- NA
imp <- mice(bwt.mis, print = FALSE, m = 10)
fit1 <- with(data = imp, expr = glm(low ~ age + lwt + race + smoke + ptd + ht + ui + ftv, family = binomial))
fit0 <- with(data = imp, glm(low ~ lwt + race + smoke + ptd + ht + ui, family = binomial))
D1(fit1, fit0)
D3(fit1, fit0)
# --- test restriction of parameters
# all parameters estimated
fit <- lm(bmi ~ age + hyp + chl, data = nhanes)
coef(fit)
formula(fit)
newformula <- bmi ~ 0 + I(18.26966503 - 5.78652468 * age + 2.10467529 * hyp + 0.08044924 * chl)
newformula <- . ~ 0 + I(18.26966503 * 1L - 5.78652468 * age + 2.10467529 * hyp + 0.08044924 * chl)
fit2 <- update(fit, formula = newformula)
coef(fit2)
summary(fit)
summary(fit2)
cor(predict(fit), predict(fit) + residuals(fit))^2
cor(predict(fit2), predict(fit2) + residuals(fit2))^2
newformula <- bmi ~ 0 + offset(18.26966503 - 5.78652468 * age + 2.10467529 * hyp + 0.08044924 * chl)
fit3 <- update(fit, formula = newformula)
coef(fit3)
summary(fit3)
cor(predict(fit3), predict(fit3) + residuals(fit3))^2
# compare to mitml::anova.mitml.result
suppressPackageStartupMessages(library(mitml, quietly = TRUE))
library(lme4, quietly = TRUE)
data(studentratings)
fml <- ReadDis + SES ~ ReadAchiev + (1 | ID)
imp <- mitml::panImpute(studentratings,
formula = fml, n.burn = 1000, n.iter = 100, m = 5,
silent = TRUE
)
implist <- mitml::mitmlComplete(imp, print = 1:5)
# * Example 1: multiparameter hypothesis test for 'ReadDis' and 'SES'
# This tests the hypothesis that both effects are zero.
fit0 <- with(implist, lmer(ReadAchiev ~ (1 | ID), REML = FALSE))
fit1 <- with(implist, lmer(ReadAchiev ~ ReadDis + (1 | ID), REML = FALSE))
# apply Rubin's rules
testEstimates(fit1)
# Wald test
# multiparameter hypothesis test using D1 (default)
mitml::testModels(fit1, fit0)
# stats <- pool.compare(as.mira(fit1), as.mira(fit0), method = "wald")
# Is the same, but probably consequence of single parameter differerence
# Wald test - multiparameter difference - incorrect because now our
# ubar is vector, not a matrix anymore
fit0 <- with(implist, lmer(ReadAchiev ~ (1 | ID), REML = FALSE))
fit1 <- with(implist, lmer(ReadAchiev ~ ReadDis + SES + (1 | ID), REML = FALSE))
mitml::testModels(fit1, fit0)
# stats <- pool.compare(as.mira(fit1), as.mira(fit0), method = "wald")
# Is the same, but probably consequence of single parameter differerence
# likelihood test
mitml::testModels(fit1, fit0, method = "D3")
# stats <- pool.compare(as.mira(fit1), as.mira(fit0), method = "likelihood")
# ---
fit1 <- with(implist, lmer(ReadAchiev ~ ReadDis + SES + (1 | ID), REML = FALSE))
mice/tests/testthat/test-mice.impute.iurr.norm.R 0000644 0001762 0000144 00000003671 14433156031 021451 0 ustar ligges users context("mice.impute.lasso.select.norm")
#########################
# TEST 1: Simple problem #
#########################
set.seed(123)
# generate data
n <- 1e3
y <- rnorm(n)
x <- y * .3 + rnorm(n, 0, .25)
x2 <- x + rnorm(n, 2, 3)
x <- cbind(x, x2)
# make missingness
y[sample(1:n, n * .3)] <- NA
ry <- !is.na(y)
wy <- !ry
# Use univariate imputation model
set.seed(123)
imps_t1 <- mice.impute.lasso.select.norm(y, ry, x)
test_that("Returns requested length", {
expect_equal(length(imps_t1), sum(!ry))
})
#########################
# TEST 2: Nothing is important #
#########################
n <- 1e2
p <- 10
b0 <- 100
bs <- rep(0, p)
x <- matrix(rnorm(n * p), n, p)
y <- b0 + x %*% bs + rnorm(n)
# Missing values
y[sample(1:n, n * .3)] <- NA
ry <- !is.na(y)
wy <- !ry
# Use univariate imputation model
set.seed(123)
imps_t2 <- mice.impute.lasso.select.norm(y, ry, x)
test_that("Returns requested length w/ intercept only model", {
expect_equal(length(imps_t2), sum(!ry))
})
#########################
# TEST 3: Everything is important #
#########################
n <- 1e2
p <- 10
b0 <- 100
bs <- rep(1, p)
x <- matrix(rnorm(n * p), n, p)
y <- b0 + x %*% bs + rnorm(n)
# Missing values
y[sample(1:n, n * .3)] <- NA
ry <- !is.na(y)
wy <- !ry
# Use univariate imputation model
set.seed(123)
imps_t3 <- mice.impute.lasso.select.norm(y, ry, x)
test_that("Returns requested length when all predictors are important", {
expect_equal(length(imps_t3), sum(!ry))
})
#########################
# TEST 4: Use it within mice call #
#########################
boys_cont <- boys[, 1:4]
iurr_default <- mice(boys_cont,
m = 2, maxit = 2, method = "lasso.select.norm",
print = FALSE
)
iurr_custom <- mice(boys_cont,
m = 2, maxit = 2, method = "lasso.select.norm",
nfolds = 5,
print = FALSE
)
test_that("mice call works", {
expect_equal(class(iurr_custom), "mids")
})
test_that("mice call works w/ custom arguments", {
expect_equal(class(iurr_custom), "mids")
})
mice/tests/testthat.R 0000755 0001762 0000144 00000000064 13666251671 014334 0 ustar ligges users library(testthat)
library(mice)
test_check("mice")
mice/src/ 0000755 0001762 0000144 00000000000 14437176012 011763 5 ustar ligges users mice/src/match.cpp 0000644 0001762 0000144 00000003306 14335377210 013565 0 ustar ligges users #include
#include
using namespace std;
using namespace Rcpp;
// [[Rcpp::export]]
IntegerVector matcher(NumericVector obs, NumericVector mis, int k) {
// fast predictive mean matching algorithm
// for each of the n0 elements in mis
// 1) calculate the difference with obs
// 2) add small noise to break ties
// 3) find the k indices of the k closest predictors
// 4) randomly draw one index
// and return the vector of n0 matched positions
// SvB 26/01/2014
// declarations
int jj;
int n1 = obs.size();
int n0 = mis.size();
double dk = 0;
int count = 0;
int goal = 0;
NumericVector d(n1);
NumericVector d2(n1);
IntegerVector matched(n0);
// restrict 1 <= k <= n1
k = (k <= n1) ? k : n1;
k = (k >= 1) ? k : 1;
// in advance, uniform sample from k potential donors
NumericVector which = floor(runif(n0, 1, k + 1));
NumericVector mm = range(obs);
double small = (mm[1] - mm[0]) / 65536;
// loop over the missing values
for(int i = 0; i < n0; i++) {
// calculate the distance and add noise to break ties
d2 = runif(n1, 0, small);
dk = mis[i];
for (int j = 0; j < n1; j++) d[j] = std::abs(obs[j] - dk) + d2[j];
// find the k'th lowest value in d
for (int j = 0; j < n1; j++) d2[j] = d[j];
std::nth_element (d2.begin(), d2.begin() + k - 1, d2.end());
// find index of donor which[i]
dk = d2[k-1];
count = 0;
goal = (int) which[i];
for (jj = 0; jj < n1; jj++) {
if (d[jj] <= dk) count++;
if (count == goal) break;
}
// and store the result
matched[i] = jj;
}
// increase index to offset 1
return matched + 1;
}
mice/src/matchindex.cpp 0000644 0001762 0000144 00000013001 14335377156 014617 0 ustar ligges users #include
#include
#include
using namespace std;
using namespace Rcpp;
//' Find index of matched donor units
//'
//' @param d Numeric vector with values from donor cases.
//' @param t Numeric vector with values from target cases.
//' @param k Integer, number of unique donors from which a random draw is made.
//' For \code{k = 1} the function returns the index in \code{d} corresponding
//' to the closest unit. For multiple imputation, the
//' advice is to set values in the range of \code{k = 5} to \code{k = 10}.
//' @return An integer vector with \code{length(t)} elements. Each
//' element is an index in the array \code{d}.
//' @details
//' For each element in \code{t}, the method finds the \code{k} nearest
//' neighbours in \code{d}, randomly draws one of these neighbours, and
//' returns its position in vector \code{d}.
//'
//' Fast predictive mean matching algorithm in seven steps:
//'
//' 1. Shuffle records to remove effects of ties
//'
//' 2. Obtain sorting order on shuffled data
//'
//' 3. Calculate index on input data and sort it
//'
//' 4. Pre-sample vector \code{h} with values between 1 and \code{k}
//'
//' For each of the \code{n0} elements in \code{t}:
//'
//' 5. find the two adjacent neighbours
//'
//' 6. find the \code{h_i}'th nearest neighbour
//'
//' 7. store the index of that neighbour
//'
//' Return vector of \code{n0} positions in \code{d}.
//'
//' We may use the function to perform predictive mean matching under a given
//' predictive model. To do so, specify both \code{d} and \code{t} as
//' predictions from the same model. Suppose that \code{y} contains the observed
//' outcomes of the donor cases (in the same sequence as \code{d}), then
//' \code{y[matchindex(d, t)]} returns one matched outcome for every
//' target case.
//'
//' See \url{https://github.com/amices/mice/issues/236}.
//' This function is a replacement for the \code{matcher()} function that has
//' been in default in \code{mice} since version \code{2.22} (June 2014).
//' @examples
//' set.seed(1)
//'
//' # Inputs need not be sorted
//' d <- c(-5, 5, 0, 10, 12)
//' t <- c(-6, -4, 0, 2, 4, -2, 6)
//'
//' # Index (in vector a) of closest match
//' idx <- matchindex(d, t, 1)
//' idx
//'
//' # To check: show values of closest match
//'
//' # Random draw among indices of the 5 closest predictors
//' matchindex(d, t)
//'
//' # An example
//' train <- mtcars[1:20, ]
//' test <- mtcars[21:32, ]
//' fit <- lm(mpg ~ disp + cyl, data = train)
//' d <- fitted.values(fit)
//' t <- predict(fit, newdata = test) # note: not using mpg
//' idx <- matchindex(d, t)
//'
//' # Borrow values from train to produce 12 synthetic values for mpg in test.
//' # Synthetic values are plausible values that could have been observed if
//' # they had been measured.
//' train$mpg[idx]
//'
//' # Exercise: Create a distribution of 1000 plausible values for each of the
//' # twelve mpg entries in test, and count how many times the true value
//' # (which we know here) is located within the inter-quartile range of each
//' # distribution. Is your count anywhere close to 500? Why? Why not?
//' @author Stef van Buuren, Nasinski Maciej, Alexander Robitzsch
//' @export
// [[Rcpp::export]]
IntegerVector matchindex(NumericVector d, NumericVector t, int k = 5) {
Environment base("package:base");
Function sample = base["sample"];
// declarations
int n1 = d.size();
int n0 = t.size();
// 1. Shuffle records to remove effects of ties
// Suggested by Alexander Robitzsch
// https://github.com/stefvanbuuren/mice/issues/236
// Call base::sample() to advance .Random.seed
IntegerVector ishuf= sample(n1);
ishuf = ishuf - 1;
NumericVector yshuf(n1);
for (int i = 0; i < n1; i++) {yshuf(i) = d(ishuf(i));}
// 2. Obtain sorting order on shuffled data
// https://stackoverflow.com/questions/1577475/c-sorting-and-keeping-track-of-indexes
IntegerVector isort(n1);
iota(isort.begin(), isort.end(), 0);
stable_sort(isort.begin(), isort.end(),
[yshuf](int i1, int i2) {return yshuf[i1] < yshuf[i2];});
// 3. Calculate index on input data and sort
IntegerVector id(n1);
std::vector ysort(n1);
for (int i = 0; i < n1; i++) {
id(i) = ishuf(isort(i));
ysort[i] = d(id(i));
}
// 4. Pre-sample n0 values between 1 and k
// restrict 1 <= k <= n1
k = (k <= n1) ? k : n1;
k = (k >= 1) ? k : 1;
IntegerVector kv(k);
iota(kv.begin(), kv.end(), 1);
IntegerVector h = sample(kv, n0, Rcpp::_["replace"] = true);
IntegerVector idx(n0);
// loop over the target units
for (int i = 0; i < n0; i++) {
double val = t(i);
int hi = h(i);
int count = 0;
// 5. find the two adjacent neighbours
std::vector::iterator iter;
iter = std::lower_bound(ysort.begin(), ysort.end(), val);
int r = iter - ysort.begin();
int l = r - 1;
// 6. find the h_i'th nearest neighbour
// 7. store the index of that neighbour
// Compare elements on left and right of crossover
// point to find the h'th closest match
// Inspired on Polkas: https://github.com/Polkas/miceFast/issues/10
while (count < hi && l >= 0 && r < n1)
{
if (val - ysort[l] < ysort[r] - val)
{
idx(i) = id[l--];
} else {
idx(i) = id[r++];
}
count++;
}
// If right side is exhausted, take left elements
while (count < hi && l >= 0)
{
idx(i) = id[l--];
count++;
}
// If left side is exhausted, take right elements
while (count < hi && r < n1)
{
idx(i) = id[r++];
count++;
}
}
return idx + 1;
}
mice/src/Makevars 0000644 0001762 0000144 00000001666 14335377254 013500 0 ustar ligges users ## Use the R_HOME indirection to support installations of multiple R version
PKG_LIBS = `$(R_HOME)/bin/Rscript -e "Rcpp:::LdFlags()"`
## As an alternative, one can also add this code in a file 'configure'
##
## PKG_LIBS=`${R_HOME}/bin/Rscript -e "Rcpp:::LdFlags()"`
##
## sed -e "s|@PKG_LIBS@|${PKG_LIBS}|" \
## src/Makevars.in > src/Makevars
##
## which together with the following file 'src/Makevars.in'
##
## PKG_LIBS = @PKG_LIBS@
##
## can be used to create src/Makevars dynamically. This scheme is more
## powerful and can be expanded to also check for and link with other
## libraries. It should be complemented by a file 'cleanup'
##
## rm src/Makevars
##
## which removes the autogenerated file src/Makevars.
##
## Of course, autoconf can also be used to write configure files. This is
## done by a number of packages, but recommended only for more advanced users
## comfortable with autoconf and its related tools.
mice/src/Makevars.win 0000644 0001762 0000144 00000000241 14335377250 014254 0 ustar ligges users
## Use the R_HOME indirection to support installations of multiple R version
PKG_LIBS = $(shell "${R_HOME}/bin${R_ARCH_BIN}/Rscript.exe" -e "Rcpp:::LdFlags()")
mice/src/legendre.cpp 0000644 0001762 0000144 00000001266 14335377237 014272 0 ustar ligges users #include
using namespace Rcpp;
// [[Rcpp::export]]
Rcpp::NumericMatrix legendre(Rcpp::NumericVector x, int p) {
int n = x.size();
Rcpp::IntegerVector polies (p+1);
for(int i=0; i < polies.length(); i++)
polies[i] = (i*2+1);
Rcpp::NumericVector squirts = sqrt(polies);
Rcpp::NumericMatrix y(n, p);
for(int i = 0; i < n; ++i) {
y(i,0) = 2 * x[i] - 1;
y(i,1) = (3 * y(i,0) * y(i,0) - 1)/2;
}
for(int j=2; j < y.ncol(); j++){
for(int i = 0; i < n; ++i) {
y(i,j) = (polies[j] * y(i,0) * y(i,j-1) - j * y(i,j-2))/(j+1);
}
}
for(int j=0; j < y.ncol(); j++){
for(int i = 0; i < n; ++i)
y(i,j) = squirts[j+1] * y(i,j);
}
return y;
}
mice/src/RcppExports.cpp 0000644 0001762 0000144 00000004220 14335377244 014765 0 ustar ligges users // Generated by using Rcpp::compileAttributes() -> do not edit by hand
// Generator token: 10BE3573-1514-4C36-9D1C-5A225CD40393
#include
using namespace Rcpp;
#ifdef RCPP_USE_GLOBAL_ROSTREAM
Rcpp::Rostream& Rcpp::Rcout = Rcpp::Rcpp_cout_get();
Rcpp::Rostream& Rcpp::Rcerr = Rcpp::Rcpp_cerr_get();
#endif
// legendre
Rcpp::NumericMatrix legendre(Rcpp::NumericVector x, int p);
RcppExport SEXP _mice_legendre(SEXP xSEXP, SEXP pSEXP) {
BEGIN_RCPP
Rcpp::RObject rcpp_result_gen;
Rcpp::RNGScope rcpp_rngScope_gen;
Rcpp::traits::input_parameter< Rcpp::NumericVector >::type x(xSEXP);
Rcpp::traits::input_parameter< int >::type p(pSEXP);
rcpp_result_gen = Rcpp::wrap(legendre(x, p));
return rcpp_result_gen;
END_RCPP
}
// matcher
IntegerVector matcher(NumericVector obs, NumericVector mis, int k);
RcppExport SEXP _mice_matcher(SEXP obsSEXP, SEXP misSEXP, SEXP kSEXP) {
BEGIN_RCPP
Rcpp::RObject rcpp_result_gen;
Rcpp::RNGScope rcpp_rngScope_gen;
Rcpp::traits::input_parameter< NumericVector >::type obs(obsSEXP);
Rcpp::traits::input_parameter< NumericVector >::type mis(misSEXP);
Rcpp::traits::input_parameter< int >::type k(kSEXP);
rcpp_result_gen = Rcpp::wrap(matcher(obs, mis, k));
return rcpp_result_gen;
END_RCPP
}
// matchindex
IntegerVector matchindex(NumericVector d, NumericVector t, int k);
RcppExport SEXP _mice_matchindex(SEXP dSEXP, SEXP tSEXP, SEXP kSEXP) {
BEGIN_RCPP
Rcpp::RObject rcpp_result_gen;
Rcpp::RNGScope rcpp_rngScope_gen;
Rcpp::traits::input_parameter< NumericVector >::type d(dSEXP);
Rcpp::traits::input_parameter< NumericVector >::type t(tSEXP);
Rcpp::traits::input_parameter< int >::type k(kSEXP);
rcpp_result_gen = Rcpp::wrap(matchindex(d, t, k));
return rcpp_result_gen;
END_RCPP
}
static const R_CallMethodDef CallEntries[] = {
{"_mice_legendre", (DL_FUNC) &_mice_legendre, 2},
{"_mice_matcher", (DL_FUNC) &_mice_matcher, 3},
{"_mice_matchindex", (DL_FUNC) &_mice_matchindex, 3},
{NULL, NULL, 0}
};
RcppExport void R_init_mice(DllInfo *dll) {
R_registerRoutines(dll, NULL, CallEntries, NULL, NULL);
R_useDynamicSymbols(dll, FALSE);
}
mice/R/ 0000755 0001762 0000144 00000000000 14436642366 011406 5 ustar ligges users mice/R/with.R 0000644 0001762 0000144 00000004536 14330031647 012477 0 ustar ligges users #' Evaluate an expression in multiple imputed datasets
#'
#' Performs a computation of each of imputed datasets in data.
#'
#' @param data An object of type \code{mids}, which stands for 'multiply imputed
#' data set', typically created by a call to function \code{mice()}.
#' @param expr An expression to evaluate for each imputed data set. Formula's
#' containing a dot (notation for "all other variables") do not work.
#' @param \dots Not used
#' @return An object of S3 class \code{\link[=mira-class]{mira}}
#' @note Version 3.11.10 changed to tidy evaluation on a quosure. This change
#' should not affect any code that worked on previous versions.
#' It turned out that the latter statement was not true (#292).
#' Version 3.12.2 reverts to the old \code{with()} function.
#' @author Karin Oudshoorn, Stef van Buuren 2009, 2012, 2020
#' @seealso \code{\link[=mids-class]{mids}}, \code{\link[=mira-class]{mira}}, \code{\link{pool}},
#' \code{\link{D1}}, \code{\link{D3}}, \code{\link{pool.r.squared}}
#' @references van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#' @keywords multivariate
#' @examples
#' imp <- mice(nhanes2, m = 2, print = FALSE, seed = 14221)
#'
#' # descriptive statistics
#' getfit(with(imp, table(hyp, age)))
#'
#' # model fitting and testing
#' fit1 <- with(imp, lm(bmi ~ age + hyp + chl))
#' fit2 <- with(imp, glm(hyp ~ age + chl, family = binomial))
#' fit3 <- with(imp, anova(lm(bmi ~ age + chl)))
#' @method with mids
#' @export
with.mids <- function(data, expr, ...) {
call <- match.call()
if (!is.mids(data)) {
stop("The data must have class mids")
}
analyses <- as.list(seq_len(data$m))
# do the repeated analysis, store the result.
for (i in seq_along(analyses)) {
data.i <- complete(data, i)
analyses[[i]] <- eval(expr = substitute(expr), envir = data.i, enclos = parent.frame())
if (is.expression(analyses[[i]])) {
analyses[[i]] <- eval(expr = analyses[[i]], envir = data.i, enclos = parent.frame())
}
}
# return the complete data analyses as a list of length nimp
object <- list(call = call, call1 = data$call, nmis = data$nmis, analyses = analyses)
# formula=formula(analyses[[1]]$terms))
oldClass(object) <- c("mira", "matrix")
object
}
mice/R/md.pattern.R 0000644 0001762 0000144 00000010155 14330031647 013572 0 ustar ligges users #' Missing data pattern
#'
#' Display missing-data patterns.
#'
#' This function is useful for investigating any structure of missing
#' observations in the data. In specific case, the missing data pattern could be
#' (nearly) monotone. Monotonicity can be used to simplify the imputation model.
#' See Schafer (1997) for details. Also, the missing pattern could suggest which
#' variables could potentially be useful for imputation of missing entries.
#'
#' @param x A data frame or a matrix containing the incomplete data. Missing
#' values are coded as NA's.
#' @param plot Should the missing data pattern be made into a plot. Default is
#' `plot = TRUE`.
#' @param rotate.names Whether the variable names in the plot should be placed
#' horizontally or vertically. Default is `rotate.names = FALSE`.
#' @return A matrix with \code{ncol(x)+1} columns, in which each row corresponds
#' to a missing data pattern (1=observed, 0=missing). Rows and columns are
#' sorted in increasing amounts of missing information. The last column and row
#' contain row and column counts, respectively.
#' @author Gerko Vink, 2018, based on an earlier version of the same function by
#' Stef van Buuren, Karin Groothuis-Oudshoorn, 2000
#' @references Schafer, J.L. (1997), Analysis of multivariate incomplete data.
#' London: Chapman&Hall.
#'
#' Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}: Multivariate
#' Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
#' Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
#' @keywords univar
#' @examples
#' md.pattern(nhanes)
#' # age hyp bmi chl
#' # 13 1 1 1 1 0
#' # 1 1 1 0 1 1
#' # 3 1 1 1 0 1
#' # 1 1 0 0 1 2
#' # 7 1 0 0 0 3
#' # 0 8 9 10 27
#' @export
md.pattern <- function(x, plot = TRUE, rotate.names = FALSE) {
if (!(is.matrix(x) || is.data.frame(x))) {
stop("Data should be a matrix or dataframe")
}
if (ncol(x) < 2) {
stop("Data should have at least two columns")
}
R <- is.na(x)
nmis <- colSums(R)
# sort columnwise
R <- matrix(R[, order(nmis)], dim(x))
pat <- apply(R, 1, function(x) paste(as.numeric(x), collapse = ""))
# sort rowwise
sortR <- matrix(R[order(pat), ], dim(x))
if (nrow(x) == 1) {
mpat <- is.na(x)
} else {
mpat <- sortR[!duplicated(sortR), ]
}
# update row and column margins
if (all(!is.na(x))) {
cat(" /\\ /\\\n{ `---' }\n{ O O }\n==> V <==")
cat(" No need for mice. This data set is completely observed.\n")
cat(" \\ \\|/ /\n `-----'\n\n")
mpat <- t(as.matrix(mpat, byrow = TRUE))
rownames(mpat) <- table(pat)
} else {
if (is.null(dim(mpat))) {
mpat <- t(as.matrix(mpat))
}
rownames(mpat) <- table(pat)
}
r <- cbind(abs(mpat - 1), rowSums(mpat))
r <- rbind(r, c(nmis[order(nmis)], sum(nmis)))
if (plot) {
op <- par(mar = rep(0, 4))
on.exit(par(op))
plot.new()
if (is.null(dim(sortR[!duplicated(sortR), ]))) {
R <- t(as.matrix(r[1:nrow(r) - 1, 1:ncol(r) - 1]))
} else {
if (is.null(dim(R))) {
R <- t(as.matrix(R))
}
R <- r[1:nrow(r) - 1, 1:ncol(r) - 1]
}
if (rotate.names) {
adj <- c(0, 0.5)
srt <- 90
length_of_longest_colname <- max(nchar(colnames(r))) / 2.6
plot.window(
xlim = c(-1, ncol(R) + 1),
ylim = c(-1, nrow(R) + length_of_longest_colname),
asp = 1
)
} else {
adj <- c(0.5, 0)
srt <- 0
plot.window(
xlim = c(-1, ncol(R) + 1),
ylim = c(-1, nrow(R) + 1),
asp = 1
)
}
M <- cbind(c(row(R)), c(col(R))) - 1
shade <- ifelse(R[nrow(R):1, ], mdc(1), mdc(2))
rect(M[, 2], M[, 1], M[, 2] + 1, M[, 1] + 1, col = shade)
for (i in 1:ncol(R)) {
text(i - .5, nrow(R) + .3, colnames(r)[i], adj = adj, srt = srt)
text(i - .5, -.3, nmis[order(nmis)][i])
}
for (i in 1:nrow(R)) {
text(ncol(R) + .3, i - .5, r[(nrow(r) - 1):1, ncol(r)][i], adj = 0)
text(-.3, i - .5, rownames(r)[(nrow(r) - 1):1][i], adj = 1)
}
text(ncol(R) + .3, -.3, r[nrow(r), ncol(r)])
return(r)
} else {
return(r)
}
}
mice/R/leiden85.R 0000644 0001762 0000144 00000003165 14330031606 013131 0 ustar ligges users #' Leiden 85+ study
#'
#' Subset of data from the Leiden 85+ study
#'
#' The data set concerns of subset of 956 members of a very old (85+) cohort in
#' Leiden.
#'
#' Multiple imputation of this data set has been described in Boshuizen et al
#' (1998), Van Buuren et al (1999) and Van Buuren (2012), chapter 7.
#'
#' The data set is not available as part of \code{mice}.
#'
#' @name leiden85
#' @docType data
#' @format \code{leiden85} is a data frame with 956 rows and 336 columns.
#' @source
#'
#' Lagaay, A. M., van der Meij, J. C., Hijmans, W. (1992). Validation of
#' medical history taking as part of a population based survey in subjects aged
#' 85 and over. \emph{Brit. Med. J.}, \emph{304}(6834), 1091-1092.
#'
#' Izaks, G. J., van Houwelingen, H. C., Schreuder, G. M., Ligthart, G. J.
#' (1997). The association between human leucocyte antigens (HLA) and mortality
#' in community residents aged 85 and older. \emph{Journal of the American
#' Geriatrics Society}, \emph{45}(1), 56-60.
#'
#' Boshuizen, H. C., Izaks, G. J., van Buuren, S., Ligthart, G. J. (1998).
#' Blood pressure and mortality in elderly people aged 85 and older: Community
#' based study. \emph{Brit. Med. J.}, \emph{316}(7147), 1780-1784.
#'
#' Van Buuren, S., Boshuizen, H.C., Knook, D.L. (1999) Multiple imputation of
#' missing blood pressure covariates in survival analysis. \emph{Statistics in
#' Medicine}, \bold{18}, 681--694.
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-toomany.html#sec:leiden85cohort}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' @keywords datasets
NULL
mice/R/ampute.mcar.R 0000644 0001762 0000144 00000004306 14330031606 013726 0 ustar ligges users #' Multivariate amputation under a MCAR mechanism
#'
#' This function creates a missing data indicator for each pattern, based on a MCAR
#' missingness mechanism. The function is used in the multivariate amputation function
#' \code{\link{ampute}}.
#'
#' @param P A vector containing the pattern numbers of the cases' candidates.
#' For each case, a value between 1 and #patterns is given. For example, a
#' case with value 2 is candidate for missing data pattern 2.
#' @param patterns A matrix of size #patterns by #variables where \code{0} indicates
#' a variable should have missing values and \code{1} indicates a variable should
#' remain complete. The user may specify as many patterns as desired. One pattern
#' (a vector) is also possible. Could be the result of \code{\link{ampute.default.patterns}},
#' default will be a square matrix of size #variables where each pattern has missingness
#' on one variable only.
#' @param prop A scalar specifying the proportion of missingness. Should be a value
#' between 0 and 1. Default is a missingness proportion of 0.5.
#' @return A list containing vectors with \code{0} if a case should be made missing
#' and \code{1} if a case should remain complete. The first vector refers to the
#' first pattern, the second vector to the second pattern, etcetera.
#' @author Rianne Schouten, 2016
#' @seealso \code{\link{ampute}}
#' @keywords internal
#' @export
ampute.mcar <- function(P, patterns, prop) {
f <- function(i) {
# If there are no candidates in a certain pattern, the list will receive a 0
if (length(P[P == (i + 1)]) == 0) {
return(0)
} else {
# Otherwise, for all candidates in the pattern, the total proportion of
# missingness is used to define the probabilities to be missing.
nf <- length(P[P == (i + 1)])
R.temp <- 1 - rbinom(n = nf, size = 1, prob = prop)
# Based on the probabilities, each candidate will receive a missing data
# indicator 0, meaning he will be made missing or missing data indicator 1,
# meaning the candidate will remain complete.
R.temp <- replace(P, P == (i + 1), R.temp)
R.temp <- replace(R.temp, P != (i + 1), 1)
return(R.temp)
}
}
lapply(seq_len(nrow(patterns)), f)
}
mice/R/ampute.R 0000644 0001762 0000144 00000061146 14433402334 013016 0 ustar ligges users #' Generate missing data for simulation purposes
#'
#' This function generates multivariate missing data under a MCAR, MAR or MNAR
#' missing data mechanism. Imputation of data sets containing missing values can
#' be performed with \code{\link{mice}}.
#'
#' This function generates missing values in complete data sets. Amputation of complete
#' data sets is useful for the evaluation of imputation techniques, such as multiple
#' imputation (performed with function \code{\link{mice}} in this package).
#'
#' The basic strategy underlying multivariate imputation was suggested by
#' Don Rubin during discussions in the 90's. Brand (1997) created one particular
#' implementation, and his method found its way into the FCS paper
#' (Van Buuren et al, 2006).
#'
#' Until recently, univariate amputation procedures were used to generate missing
#' data in complete, simulated data sets. With this approach, variables are made
#' incomplete one variable at a time. When more than one variable needs to be amputed,
#' the procedure is repeated multiple times.
#'
#' With the univariate approach, it is difficult to relate the missingness on one
#' variable to the missingness on another variable. A multivariate amputation procedure
#' solves this issue and moreover, it does justice to the multivariate nature of
#' data sets. Hence, \code{ampute} is developed to perform multivariate amputation.
#'
#' The idea behind the function is the specification of several missingness
#' patterns. Each pattern is a combination of variables with and without missing
#' values (denoted by \code{0} and \code{1} respectively). For example, one might
#' want to create two missingness patterns on a data set with four variables. The
#' patterns could be something like: \code{0,0,1,1} and \code{1,0,1,0}.
#' Each combination of zeros and ones may occur.
#'
#' Furthermore, the researcher specifies the proportion of missingness, either the
#' proportion of missing cases or the proportion of missing cells, and the relative
#' frequency each pattern occurs. Consequently, the data is split into multiple subsets,
#' one subset per pattern. Now, each case is candidate for a certain missingness pattern,
#' but whether the case will have missing values eventually depends on other specifications.
#'
#' The first of these specifications is the missing mechanism. There are three possible
#' mechanisms: the missingness depends completely on chance (MCAR), the missingness
#' depends on the values of the observed variables (i.e. the variables that remain
#' complete) (MAR) or on the values of the variables that will be made incomplete (MNAR).
#' For a discussion on how missingness mechanisms are related to the observed data,
#' we refer to \doi{10.1177/0049124118799376}{Schouten and Vink, 2018}.
#'
#' When the user specifies the missingness mechanism to be \code{"MCAR"}, the candidates
#' have an equal probability of becoming incomplete. For a \code{"MAR"} or \code{"MNAR"} mechanism,
#' weighted sum scores are calculated. These scores are a linear combination of the
#' variables.
#'
#' In order to calculate the weighted sum scores, the data is standardized. For this reason,
#' the data has to be numeric. Second, for each case, the values in
#' the data set are multiplied with the weights, specified by argument \code{weights}.
#' These weighted scores will be summed, resulting in a weighted sum score for each case.
#'
#' The weights may differ between patterns and they may be negative or zero as well.
#' Naturally, in case of a MAR mechanism, the weights corresponding to the
#' variables that will be made incomplete, have a 0. Note that this may be
#' different for each pattern. In case of MNAR missingness, especially
#' the weights of the variables that will be made incomplete are of importance. However,
#' the other variables may be weighted as well.
#'
#' It is the relative difference between the weights that will result in an effect
#' in the sum scores. For example, for the first missing data
#' pattern mentioned above, the weights for the third and fourth variables could
#' be set to 2 and 4. However, weight values of 0.2 and 0.4 will have the exact
#' same effect on the weighted sum score: the fourth variable is weighted twice as
#' much as variable 3.
#'
#' Based on the weighted sum scores, either a discrete or continuous distribution
#' of probabilities is used to calculate whether a candidate will have missing values.
#'
#' For a discrete distribution of probabilities, the weighted sum scores are
#' divided into subgroups of equal size (quantiles). Thereafter, the user
#' specifies for each subgroup the odds of being missing. Both the number of
#' subgroups and the odds values are important for the generation of missing data.
#' For example, for a RIGHT-like mechanism, scoring in one of the
#' higher quantiles should have high missingness odds, whereas for a MID-like
#' mechanism, the central groups should have higher odds. Again, not the size of
#' the odds values are of importance, but the relative distance between the values.
#'
#' The continuous distributions of probabilities are based on the logistic distribution function.
#' The user can specify the type of missingness, which, again, may differ between patterns.
#'
#' For an example and more explanation about how the arguments interact with each other,
#' we refer to the vignette
#' \href{https://rianneschouten.github.io/mice_ampute/vignette/ampute.html}{Generate missing values with ampute}
#' The amputation methodology is published in
#' \doi{10.1080/00949655.2018.1491577}{Schouten, Lugtig and Vink, 2018}.
#'
#' @param data A complete data matrix or data frame. Values should be numeric.
#' Categorical variables should have been transformed to dummies.
#' @param prop A scalar specifying the proportion of missingness. Should be a value
#' between 0 and 1. Default is a missingness proportion of 0.5.
#' @param patterns A matrix or data frame of size #patterns by #variables where
#' \code{0} indicates that a variable should have missing values and \code{1} indicates
#' that a variable should remain complete. The user may specify as many patterns as
#' desired. One pattern (a vector) is possible as well. Default
#' is a square matrix of size #variables where each pattern has missingness on one
#' variable only (created with \code{\link{ampute.default.patterns}}). After the
#' amputation procedure, \code{\link{md.pattern}} can be used to investigate the
#' missing data patterns in the data.
#' @param freq A vector of length #patterns containing the relative frequency with
#' which the patterns should occur. For example, for three missing data patterns,
#' the vector could be \code{c(0.4, 0.4, 0.2)}, meaning that of all cases with
#' missing values, 40 percent should have pattern 1, 40 percent pattern 2 and 20
#' percent pattern 3. The vector should sum to 1. Default is an equal probability
#' for each pattern, created with \code{\link{ampute.default.freq}}.
#' @param mech A string specifying the missingness mechanism, either "MCAR"
#' (Missing Completely At Random), "MAR" (Missing At Random) or "MNAR" (Missing Not At
#' Random). Default is a MAR missingness mechanism.
#' @param weights A matrix or data frame of size #patterns by #variables. The matrix
#' contains the weights that will be used to calculate the weighted sum scores. For
#' a MAR mechanism, the weights of the variables that will be made incomplete should be
#' zero. For a MNAR mechanism, these weights could have any possible value. Furthermore,
#' the weights may differ between patterns and between variables. They may be negative
#' as well. Within each pattern, the relative size of the values are of importance.
#' The default weights matrix is made with \code{\link{ampute.default.weights}} and
#' returns a matrix with equal weights for all variables. In case of MAR, variables
#' that will be amputed will be weighted with \code{0}. For MNAR, variables
#' that will be observed will be weighted with \code{0}. If the mechanism is MCAR, the
#' weights matrix will not be used.
#' @param std Logical. Whether the weighted sum scores should be calculated with
#' standardized data or with non-standardized data. The latter is especially advised when
#' making use of train and test sets in order to prevent leakage.
#' @param cont Logical. Whether the probabilities should be based on a continuous
#' or a discrete distribution. If TRUE, the probabilities of being missing are based
#' on a continuous logistic distribution function. \code{\link{ampute.continuous}}
#' will be used to calculate and assign the probabilities. These probabilities will then
#' be based on the argument \code{type}. If FALSE, the probabilities of being missing are
#' based on a discrete distribution (\code{\link{ampute.discrete}}) based on the \code{odds}
#' argument. Default is TRUE.
#' @param type A string or vector of strings containing the type of missingness for each
#' pattern. Either \code{"LEFT"}, \code{"MID"}, \code{"TAIL"} or '\code{"RIGHT"}.
#' If a single missingness type is given, all patterns will be created with the same
#' type. If the missingness types should differ between patterns, a vector of missingness
#' types should be given. Default is RIGHT for all patterns and is the result of
#' \code{\link{ampute.default.type}}.
#' @param odds A matrix where #patterns defines the #rows. Each row should contain
#' the odds of being missing for the corresponding pattern. The number of odds values
#' defines in how many quantiles the sum scores will be divided. The odds values are
#' relative probabilities: a quantile with odds value 4 will have a probability of
#' being missing that is four times higher than a quantile with odds 1. The
#' number of quantiles may differ between the patterns, specify NA for cells remaining empty.
#' Default is 4 quantiles with odds values 1, 2, 3 and 4 and is created by
#' \code{\link{ampute.default.odds}}.
#' @param bycases Logical. If TRUE, the proportion of missingness is defined in
#' terms of cases. If FALSE, the proportion of missingness is defined in terms of
#' cells. Default is TRUE.
#' @param run Logical. If TRUE, the amputations are implemented. If FALSE, the
#' return object will contain everything except for the amputed data set.
#'
#' @return Returns an S3 object of class \code{\link{mads-class}} (multivariate
#' amputed data set)
#' @author Rianne Schouten [aut, cre], Gerko Vink [aut], Peter Lugtig [ctb], 2016
#' @seealso \code{\link{mads-class}}, \code{\link{bwplot}}, \code{\link{xyplot}},
#' \code{\link{mice}}
#'
#' @references Brand, J.P.L. (1999) \emph{Development, implementation and
#' evaluation of multiple imputation strategies for the statistical analysis of
#' incomplete data sets.} pp. 110-113. Dissertation. Rotterdam: Erasmus University.
#'
#' Schouten, R.M., Lugtig, P and Vink, G. (2018)
#' {Generating missing values for simulation purposes: A multivariate amputation procedure.}.
#' \emph{Journal of Statistical Computation and Simulation}, 88(15): 1909-1930.
#' \doi{10.1080/00949655.2018.1491577}
#'
#' Schouten, R.M. and Vink, G. (2018){The Dance of the Mechanisms: How Observed Information Influences the Validity of Missingness Assumptions}.
#' \emph{Sociological Methods and Research}, 50(3): 1243-1258.
#' \doi{10.1177/0049124118799376}
#'
#' Van Buuren, S., Brand, J.P.L., Groothuis-Oudshoorn, C.G.M., Rubin, D.B. (2006)
#' {Fully conditional specification in multivariate imputation.}
#' \emph{Journal of Statistical Computation and Simulation}, 76(12): 1049-1064.
#' \doi{10.1080/10629360600810434}
#'
#' Van Buuren, S. (2018) \href{https://stefvanbuuren.name/fimd/sec-FCS.html#sec:MICE}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#'
#' Vink, G. (2016) Towards a standardized evaluation of multiple imputation routines.
#' @examples
#' # start with a complete data set
#' compl_boys <- cc(boys)[1:3]
#'
#' # Perform amputation with default settings
#' mads_boys <- ampute(data = compl_boys)
#' mads_boys$amp
#'
#' # Change default matrices as desired
#' my_patterns <- mads_boys$patterns
#' my_patterns[1:3, 2] <- 0
#'
#' my_weights <- mads_boys$weights
#' my_weights[2, 1] <- 2
#' my_weights[3, 1] <- 0.5
#'
#' # Rerun amputation
#' my_mads_boys <- ampute(
#' data = compl_boys, patterns = my_patterns, freq =
#' c(0.3, 0.3, 0.4), weights = my_weights, type = c("RIGHT", "TAIL", "LEFT")
#' )
#' my_mads_boys$amp
#' @export
ampute <- function(data, prop = 0.5, patterns = NULL, freq = NULL,
mech = "MAR", weights = NULL, std = TRUE, cont = TRUE,
type = NULL, odds = NULL,
bycases = TRUE, run = TRUE) {
if (is.null(data)) {
stop("Argument data is missing, with no default", call. = FALSE)
}
data <- check.dataform(data)
if (anyNA(data)) {
stop("Data cannot contain NAs", call. = FALSE)
}
if (ncol(data) < 2) {
stop("Data should contain at least two columns", call. = FALSE)
}
data <- data.frame(data)
if (any(vapply(data, Negate(is.numeric), logical(1))) && mech != "MCAR") {
data <- as.data.frame(sapply(data, as.numeric))
warning("Data is made numeric because the calculation of weights requires numeric data",
call. = FALSE
)
}
if (prop < 0 || prop > 100) {
stop("Proportion of missingness should be a value between 0 and 1 (for a proportion) or between 1 and 100 (for a percentage)",
call. = FALSE
)
} else if (prop > 1) {
prop <- prop / 100
}
if (is.null(patterns)) {
patterns <- ampute.default.patterns(n = ncol(data))
} else if (is.vector(patterns) && (length(patterns) / ncol(data)) %% 1 == 0) {
patterns <- matrix(patterns, nrow = length(patterns) / ncol(data), byrow = TRUE)
if (nrow(patterns) == 1 && all(patterns[1, ] %in% 1)) {
stop("One pattern with merely ones results to no amputation at all, the procedure is therefore stopped", call. = FALSE)
}
} else if (is.vector(patterns)) {
stop("Length of pattern vector does not match #variables", call. = FALSE)
}
patterns <- data.frame(patterns)
if (is.null(freq)) {
freq <- ampute.default.freq(patterns = patterns)
}
if (!is.vector(freq)) {
freq <- as.vector(freq)
warning("Frequency should be a vector", call. = FALSE)
}
if (length(freq) != nrow(patterns)) {
if (length(freq) > nrow(patterns)) {
freq <- freq[seq_along(nrow(patterns))]
} else {
freq <- c(freq, rep.int(0.2, nrow(patterns) - length(freq)))
}
warning(paste("Length of vector with relative frequencies does not match #patterns and is therefore changed to", freq), call. = FALSE)
}
if (sum(freq) != 1) {
freq <- recalculate.freq(freq = freq)
}
check.pat <- check.patterns(
patterns = patterns,
freq = freq,
prop = prop
)
patterns.new <- check.pat[["patterns"]]
freq <- check.pat[["freq"]]
prop <- check.pat[["prop"]]
if (!bycases) {
prop <- recalculate.prop(
prop = prop,
freq = freq,
patterns = patterns.new,
k = ncol(data),
n = nrow(data)
)
}
if (any(!mech %in% c("MCAR", "MAR", "MNAR"))) {
stop("Mechanism should be either MCAR, MAR or MNAR", call. = FALSE)
}
if (!is.vector(mech)) {
mech <- as.vector(mech)
warning("Mechanism should contain merely MCAR, MAR or MNAR", call. = FALSE)
} else if (length(mech) > 1) {
mech <- mech[1]
warning("Mechanism should contain merely MCAR, MAR or MNAR. First element is used",
call. = FALSE
)
}
# Check if there is a pattern with merely zeroos
if (!is.null(check.pat[["row.zero"]]) && mech == "MAR") {
stop(paste("Patterns object contains merely zeros and this kind of pattern is not possible when mechanism is MAR"),
call. = FALSE
)
}
if (mech == "MCAR" && !is.null(weights)) {
weights <- NULL
warning("Weights matrix is not used when mechanism is MCAR", call. = FALSE)
}
if (mech == "MCAR" && !is.null(odds)) {
odds <- NULL
warning("Odds matrix is not used when mechanism is MCAR", call. = FALSE)
}
if (mech != "MCAR" && !is.null(weights)) {
if (is.vector(weights) && (length(weights) / ncol(data)) %% 1 == 0) {
weights <- matrix(weights, nrow = length(weights) / ncol(data), byrow = TRUE)
} else if (is.vector(weights)) {
stop("Length of weight vector does not match #variables", call. = FALSE)
} else if (!is.matrix(weights) && !is.data.frame(weights)) {
stop("Weights matrix should be a matrix", call. = FALSE)
}
}
if (is.null(weights)) {
weights <- ampute.default.weights(
patterns = patterns.new,
mech = mech
)
}
weights <- as.data.frame(weights)
if (!nrow(weights) == nrow(patterns.new)) {
if (!is.null(check.pat[["row.one"]])) {
weights <- weights[-check.pat[["row.one"]], ]
}
}
if (!nrow(weights) == nrow(patterns.new)) {
stop("The objects patterns and weights are not matching", call. = FALSE)
}
if (!is.vector(cont)) {
cont <- as.vector(cont)
warning("Continuous should contain merely TRUE or FALSE", call. = FALSE)
} else if (length(cont) > 1) {
cont <- cont[1]
warning("Continuous should contain merely TRUE or FALSE. First element is used",
call. = FALSE
)
}
if (!is.logical(cont)) {
stop("Continuous should contain TRUE or FALSE", call. = FALSE)
}
if (cont && !is.null(odds)) {
odds <- NULL
warning("Odds matrix is not used when continuous probabilities (cont == TRUE) are specified",
call. = FALSE
)
}
if (!cont && !is.null(type)) {
type <- NULL
warning("Type is not used when discrete probabilities (cont == FALSE) are specified",
call. = FALSE
)
}
if (is.null(type)) {
type <- ampute.default.type(patterns = patterns.new)
}
if (any(!type %in% c("LEFT", "MID", "TAIL", "RIGHT"))) {
stop("Type should contain LEFT, MID, TAIL or RIGHT",
call. = FALSE
)
}
if (!is.vector(type)) {
type <- as.vector(type)
warning("Type should be a vector of strings", call. = FALSE)
} else if (!length(type) %in% c(1, nrow(patterns), nrow(patterns.new))) {
type <- type[1]
warning("Type should either have length 1 or length equal to #patterns, first element is used for all patterns", call. = FALSE)
}
if (mech != "MCAR" && !is.null(odds) && !is.matrix(odds)) {
if (nrow(patterns.new) == 1 && is.vector(odds)) {
odds <- matrix(odds, nrow = 1)
} else {
stop("Odds matrix should be a matrix", call. = FALSE)
}
}
if (is.null(odds)) {
odds <- ampute.default.odds(patterns = patterns.new)
}
if (!cont) {
for (h in seq_len(nrow(odds))) {
if (any(!is.na(odds[h, ]) & odds[h, ] < 0)) {
stop("Odds matrix can only have positive values", call. = FALSE)
}
}
}
if (!nrow(odds) %in% c(nrow(patterns), nrow(patterns.new))) {
stop("The objects patterns and odds are not matching", call. = FALSE)
}
#
# Start using arguments
# Create empty objects
P <- NULL
scores <- NULL
missing.data <- NULL
# Apply function (run = TRUE) or merely return objects (run = FALSE)
if (run) {
# Assign cases to the patterns according probs
# Because 0 and 1 will be used for missingness,
# the numbering of the patterns will start from 2
P <- sample.int(
n = nrow(patterns.new), size = nrow(data),
replace = TRUE, prob = freq
) + 1
# Check whether cases are assigned to all patterns
non.used.patterns <- c(2:(nrow(patterns.new) + 1))[!c(2:(nrow(patterns.new) + 1)) %in% unique(P)]
if (length(non.used.patterns) > 0) {
warning(paste0("No records are assigned to patterns ", toString(non.used.patterns - 1), ". These patterns will not be generated. Consider reducing the number of patterns or increasing the dataset size."), call. = FALSE)
}
# Calculate missingness according MCAR or calculate weighted sum scores
# Standardized data is used to calculate weighted sum scores
if (mech == "MCAR") {
R <- ampute.mcar(
P = P,
patterns = patterns.new,
prop = prop
)
} else {
scores <- sumscores(
P = P,
data = data,
std = std,
weights = weights,
patterns = patterns
)
if (!cont) {
R <- ampute.discrete(
P = P,
scores = scores,
odds = odds,
prop = prop
)
} else if (cont) {
R <- ampute.continuous(
P = P,
scores = scores,
prop = round(prop, 3),
type = type
)
}
}
missing.data <- data
for (i in seq_len(nrow(patterns.new))) {
if (any(P == (i + 1))) {
missing.data[R[[i]] == 0, patterns.new[i, ] == 0] <- NA
}
}
}
#
# Create return object
names(patterns.new) <- names(data)
names(weights) <- names(data)
call <- match.call()
missing.data <- data.frame(missing.data)
result <- list(
call = call,
prop = prop,
patterns = patterns.new,
freq = freq,
mech = mech,
weights = weights,
cont = cont,
std = std,
type = type,
odds = odds,
amp = missing.data,
cand = P - 1,
scores = scores,
data = as.data.frame(data)
)
#
# Return result
oldClass(result) <- "mads"
result
}
# This is an underlying function of multivariate amputation function ampute().
# This function is used to calculate the weighted sum scores of the candidates.
# Based on the data, the weights matrix and the kind of mechanism, each case
# will obtain a certain score that will define his probability to be made missing.
# The calculation of the probabilities occur in the function ampute.mcar(),
# ampute.continuous() or ampute.discrete(), based on the kind of missingness.
sumscores <- function(P, data, std, weights, patterns) {
weights <- as.matrix(weights)
f <- function(i) {
if (length(P[P == (i + 1)]) == 0) {
return(0) # this will ensure length 1 which is used in ampute.continuous
} else {
candidates <- as.matrix(data[P == (i + 1), ])
# For each candidate in the pattern, a weighted sum score is calculated
if (std) {
length_unique <- function(x) {
return(length(unique(x)) == 1)
}
# shangzhi-hong, Feb 2020, #216
if (nrow(candidates) > 1 && !(any(apply(candidates, 2, length_unique)))) {
candidates <- scale(candidates)
}
}
scores <- apply(candidates, 1, function(x) weights[i, ] %*% x)
if (length(scores) > 1 && length(unique(scores)) != 1) {
scores <- scale(scores)
}
return(scores)
}
}
lapply(seq_len(nrow(patterns)), f)
}
# This is an underlying function of multivariate amputation function ampute().
# The function recalculates the proportion of missing cases for the desired
# #missing cells.
recalculate.prop <- function(prop, n, k, patterns, freq) {
miss <- prop * n * k # Desired #missing cells
# Calculate #cases according prop and #zeros in patterns
cases <- vapply(
seq_len(nrow(patterns)),
function(i) (miss * freq[i]) / length(patterns[i, ][patterns[i, ] == 0]),
numeric(1)
)
if (sum(cases) > n) {
stop("Proportion of missing cells is too large in combination with the desired number of missing variables",
call. = FALSE
)
} else {
prop <- sum(cases) / n
}
prop
}
# This is an underlying function of multivariate amputation function ampute().
# The function recalculates the frequency vector to make the sum equal to 1.
recalculate.freq <- function(freq) {
freq / sum(freq)
}
# This is an underlying function of multivariate amputation function ampute().
# The function checks whether there are patterns with merely ones or zeroos.
# In case of the first, these patterns will be removed, and argument prop
# and freq will be changed. In case there is a pattern with merely zeroos,
# this is ascertained and saved in the object row.zero.
check.patterns <- function(patterns, freq, prop) {
prop.one <- 0
row.one <- c()
for (h in seq_len(nrow(patterns))) {
if (any(!patterns[h, ] %in% c(0, 1))) {
stop(paste("Argument patterns can only contain 0 and 1, pattern", h, "contains another element"), call. = FALSE)
}
if (all(patterns[h, ] %in% 1)) {
prop.one <- prop.one + freq[h]
row.one <- c(row.one, h)
}
}
if (prop.one != 0) {
warning(paste("Proportion of missingness has changed from", prop, "to", (1 - prop.one) * prop, "because of pattern(s) with merely ones"), call. = FALSE)
prop <- (1 - prop.one) * prop
freq <- freq[-row.one]
freq <- recalculate.freq(freq)
patterns <- patterns[-row.one, ]
warning("Frequency vector and patterns matrix have changed because of pattern(s) with merely ones", call. = FALSE)
}
prop.zero <- 0
row.zero <- c()
for (h in seq_len(nrow(patterns))) {
if (all(patterns[h, ] %in% 0)) {
prop.zero <- prop.zero + freq[h]
row.zero <- c(row.zero, h)
}
}
objects <- list(
patterns = patterns,
prop = prop,
freq = freq,
row.one = row.one,
row.zero = row.zero
)
objects
}
mice/R/print.R 0000644 0001762 0000144 00000005003 14330031606 012641 0 ustar ligges users #' Print a \code{mids} object
#'
#' @rdname print
#' @param x Object of class \code{mids}, \code{mira} or \code{mipo}
#' @param ... Other parameters passed down to \code{print.default()}
#' @return \code{NULL}
#' @seealso \code{\link[=mids-class]{mids}}
#' @method print mids
#' @export
print.mids <- function(x, ...) {
cat("Class: mids\n")
cat("Number of multiple imputations: ", x$m, "\n")
cat("Imputation methods:\n")
print(x$method, ...)
cat("PredictorMatrix:\n")
print(head(x$predictorMatrix), ...)
if (!is.null(x$loggedEvents)) {
cat("Number of logged events: ", nrow(x$loggedEvents), "\n")
print(head(x$loggedEvents), ...)
}
invisible(x)
}
#' Print a \code{mira} object
#'
#' @rdname print
#' @return \code{NULL}
#' @seealso \code{\link[=mira-class]{mira}}
#' @method print mira
#' @export
print.mira <- function(x, ...) {
if (is.mira(x)) {
print.listof(x, ...)
} else {
print(x, ...)
}
invisible(x)
}
#' Print a \code{mice.anova} object
#'
#' @rdname print
#' @return \code{NULL}
#' @seealso \code{\link{mipo}}
#' @method print mice.anova
#' @export
print.mice.anova <- function(x, ...) {
z <- summary(x, ...)
print(z$comparisons, row.names = FALSE)
invisible(x)
}
#' Print a \code{summary.mice.anova} object
#'
#' @rdname print
#' @return \code{NULL}
#' @seealso \code{\link{mipo}}
#' @method print mice.anova.summary
#' @export
print.mice.anova.summary <- function(x, ...) {
cat("\nModels:\n")
print(x$models, row.names = FALSE)
cat("\nComparisons:\n")
print(x$comparisons, row.names = FALSE)
cat(
"\nNumber of imputations: ", x$m,
" Method", x$method
)
if (x$method == "D2") cat(" (", x$use, ")", sep = "")
cat("\n")
invisible(x)
}
#' Print a \code{mads} object
#'
#' @param x Object of class \code{mads}
#' @param ... Other parameters passed down to \code{print.default()}
#' @return \code{NULL}
#' @seealso \code{\link[=mads-class]{mads}}
#' @method print mads
#' @export
print.mads <- function(x, ...) {
if (is.mads(x)) {
cat("Multivariate Amputed Data Set")
cat("\nCall: ")
print(x$call)
cat("Class:", class(x))
cat("\nProportion of Missingness: ", x$prop)
cat("\nFrequency of Patterns: ", x$freq)
cat("\nPattern Matrix:\n")
print(x$patterns)
cat("Mechanism:")
print(x$mech)
cat("Weight Matrix:\n")
print(x$weights)
cat("Type Vector:\n")
print(x$type)
cat("Odds Matrix:\n")
print(x$odds)
cat("Head of Amputed Data Set\n")
print(head(x$amp))
} else {
print(x, ...)
}
invisible(x)
}
mice/R/mice.impute.2lonly.pmm.R 0000644 0001762 0000144 00000014166 14430474274 015761 0 ustar ligges users #' Imputation at level 2 by predictive mean matching
#'
#' Imputes univariate missing data at level 2 using predictive mean matching.
#' Variables are level 1 are aggregated at level 2. The group identifier at
#' level 2 must be indicated by \code{type = -2} in the \code{predictorMatrix}.
#'
#' @aliases 2lonly.pmm
#' @inheritParams mice.impute.pmm
#' @param type Group identifier must be specified by '-2'. Predictors must be
#' specified by '1'.
#' @param ... Other named arguments.
#' @return A vector of length \code{nmis} with imputations.
#' @author Alexander Robitzsch (IPN - Leibniz Institute for Science and
#' Mathematics Education, Kiel, Germany), \email{robitzsch@@ipn.uni-kiel.de}
#' @seealso \code{\link{mice.impute.pmm}},
#' \code{\link{mice.impute.2lonly.norm}}, \code{\link{mice.impute.2l.pan}},
#' \code{\link{mice.impute.2lonly.mean}}
#' @details
#' This function allows in combination with \code{\link{mice.impute.2l.pan}}
#' switching regression imputation between level 1 and level 2 as described in
#' Yucel (2008) or Gelman and Hill (2007, p. 541).
#'
#' The function checks for partial missing level-2 data. Level-2 data
#' are assumed to be constant within the same cluster. If one or more
#' entries are missing, then the procedure aborts with an error
#' message that identifies the cluster with incomplete level-2 data.
#' In such cases, one may first fill in the cluster mean (or mode) by
#' the \code{2lonly.mean} method to remove inconsistencies.
#' @references Gelman, A. and Hill, J. (2007). \emph{Data analysis using
#' regression and multilevel/hierarchical models}. Cambridge, Cambridge
#' University Press.
#'
#' Yucel, RM (2008). Multiple imputation inference for multivariate multilevel
#' continuous data with ignorable non-response. \emph{Philosophical
#' Transactions of the Royal Society A}, \bold{366}, 2389-2404.
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-level2pred.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#'
#' @note The extension to categorical variables transforms
#' a dependent factor variable by means of the \code{as.integer()}
#' function. This may make sense for categories that are
#' approximately ordered, but less so for pure nominal measures.
#'
#' For a more general approach, see
#' \code{miceadds::mice.impute.2lonly.function()}.
#' @family univariate-2lonly
#' @examples
#' # simulate some data
#' # x,y ... level 1 variables
#' # v,w ... level 2 variables
#'
#' G <- 250 # number of groups
#' n <- 20 # number of persons
#' beta <- .3 # regression coefficient
#' rho <- .30 # residual intraclass correlation
#' rho.miss <- .10 # correlation with missing response
#' missrate <- .50 # missing proportion
#' y1 <- rep(rnorm(G, sd = sqrt(rho)), each = n) + rnorm(G * n, sd = sqrt(1 - rho))
#' w <- rep(round(rnorm(G), 2), each = n)
#' v <- rep(round(runif(G, 0, 3)), each = n)
#' x <- rnorm(G * n)
#' y <- y1 + beta * x + .2 * w + .1 * v
#' dfr0 <- dfr <- data.frame("group" = rep(1:G, each = n), "x" = x, "y" = y, "w" = w, "v" = v)
#' dfr[rho.miss * x + rnorm(G * n, sd = sqrt(1 - rho.miss)) < qnorm(missrate), "y"] <- NA
#' dfr[rep(rnorm(G), each = n) < qnorm(missrate), "w"] <- NA
#' dfr[rep(rnorm(G), each = n) < qnorm(missrate), "v"] <- NA
#'
#' # empty mice imputation
#' imp0 <- mice(as.matrix(dfr), maxit = 0)
#' predM <- imp0$predictorMatrix
#' impM <- imp0$method
#'
#' # multilevel imputation
#' predM1 <- predM
#' predM1[c("w", "y", "v"), "group"] <- -2
#' predM1["y", "x"] <- 1 # fixed x effects imputation
#' impM1 <- impM
#' impM1[c("y", "w", "v")] <- c("2l.pan", "2lonly.norm", "2lonly.pmm")
#'
#' # turn v into a categorical variable
#' dfr$v <- as.factor(dfr$v)
#' levels(dfr$v) <- LETTERS[1:4]
#'
#' # y ... imputation using pan
#' # w ... imputation at level 2 using norm
#' # v ... imputation at level 2 using pmm
#'
#' # skip imputation on solaris
#' is.solaris <- function() grepl("SunOS", Sys.info()["sysname"])
#' if (!is.solaris()) {
#' imp <- mice(dfr,
#' m = 1, predictorMatrix = predM1,
#' method = impM1, maxit = 1, paniter = 500
#' )
#' }
#' @export
mice.impute.2lonly.pmm <- function(y, ry, x, type, wy = NULL, ...) {
.imputation.level2(
y = y, ry = ry, x = x, type = type, wy = wy,
method = "pmm", ...
)
}
# imputation function at level 2
# can be done with norm and pmm
.imputation.level2 <- function(y, ry, x, type, wy, method, ...) {
if (sum(type == -2L) != 1L) stop("No class variable")
if (is.null(wy)) wy <- !ry
# handle categorical data
ylev <- NULL
if (is.factor(y)) {
ylev <- levels(y)
y <- as.integer(y)
}
# extract cluster index
clusterx <- x[, type == -2L]
# clusters with one or more missing y's
cm <- unique(clusterx[!ry])
# clusters with one or more observed y's
co <- unique(clusterx[ry])
# cluster where all y's are observed
cobs <- setdiff(co, cm)
# clusters where some y's are missing
csom <- intersect(co, cm)
if (length(csom) > 0L) {
stop(paste0(
"Method 2lonly.", method, " found the following clusters with partially missing\n",
" level-2 data: ", paste(csom, collapse = ", "), "\n",
" Method 2lonly.mean can fix such inconsistencies."
))
}
# calculate aggregated values
x <- cbind(1, as.matrix(x[, type %in% c(1L, 2L)]))
a2 <- rowsum(cbind(x, y), clusterx, na.rm = TRUE)
a2 <- a2 / rowsum(1 * cbind(!is.na(x), ry), clusterx)
clusterx0 <- as.numeric(paste0(rownames(a2)))
a1 <- cbind(clusterx0, a2)
ry2 <- a1[, 1L] %in% cobs
wy2 <- !(a1[, 1L] %in% unique(clusterx[!wy]))
y2 <- a1[, ncol(a1)]
x2 <- as.matrix(a1[, -c(1L:2L, ncol(a1))])
# norm imputation at level 2
if (method == "norm") {
ximp2 <- mice.impute.norm(
y = y2, ry = ry2, x = x2,
wy = wy2, ...
)
}
# pmm imputation at level 2
if (method == "pmm") {
ximp2 <- mice.impute.pmm(
y = y2, ry = ry2, x = x2,
wy = wy2, ...
)
}
# expland to full matrix
cly2 <- a1[wy2, 1L]
i1 <- match(clusterx, cly2)
ximp <- (ximp2[i1])[wy]
# turn back into factor
if (!is.null(ylev)) {
ximp <- factor(as.integer(ximp),
levels = 1L:length(ylev),
labels = ylev)
}
ximp
}
mice/R/mice.impute.cart.R 0000644 0001762 0000144 00000007502 14436637016 014700 0 ustar ligges users #' Imputation by classification and regression trees
#'
#' Imputes univariate missing data using classification and regression trees.
#'
#' @aliases mice.impute.cart cart
#'
#' @inheritParams mice.impute.pmm
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @param minbucket The minimum number of observations in any terminal node used.
#' See \code{\link{rpart.control}} for details.
#' @param cp Complexity parameter. Any split that does not decrease the overall
#' lack of fit by a factor of cp is not attempted. See \code{\link{rpart.control}}
#' for details.
#' @param ... Other named arguments passed down to \code{rpart()}.
#' @return Numeric vector of length \code{sum(!ry)} with imputations
#' @details
#' Imputation of \code{y} by classification and regression trees. The procedure
#' is as follows:
#' \enumerate{
#' \item Fit a classification or regression tree by recursive partitioning;
#' \item For each \code{ymis}, find the terminal node they end up according to the fitted tree;
#' \item Make a random draw among the member in the node, and take the observed value from that
#' draw as the imputation.
#' }
#' @seealso \code{\link{mice}}, \code{\link{mice.impute.rf}},
#' \code{\link[rpart]{rpart}}, \code{\link[rpart]{rpart.control}}
#' @author Lisa Doove, Stef van Buuren, Elise Dusseldorp, 2012
#' @references
#'
#' Doove, L.L., van Buuren, S., Dusseldorp, E. (2014), Recursive partitioning
#' for missing data imputation in the presence of interaction Effects.
#' Computational Statistics & Data Analysis, 72, 92-104.
#'
#' Breiman, L., Friedman, J. H., Olshen, R. A., and Stone, C. J.
#' (1984), Classification and regression trees, Monterey, CA: Wadsworth &
#' Brooks/Cole Advanced Books & Software.
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-cart.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#'
#' @family univariate imputation functions
#' @examples
#' imp <- mice(nhanes2, meth = "cart", minbucket = 4)
#' plot(imp)
#' @keywords datagen
#' @export
mice.impute.cart <- function(y, ry, x, wy = NULL, minbucket = 5, cp = 1e-04,
...) {
if (is.null(wy)) {
wy <- !ry
}
minbucket <- max(1, minbucket)
if (dim(x)[2] == 0) {
x <- cbind(x, 1)
dimnames(x) <- list(NULL, "int")
}
xobs <- data.frame(x[ry, , drop = FALSE])
xmis <- data.frame(x[wy, , drop = FALSE])
yobs <- y[ry]
if (!is.factor(yobs)) {
fit <- rpart::rpart(yobs ~ .,
data = cbind(yobs, xobs), method = "anova",
control = rpart::rpart.control(minbucket = minbucket, cp = cp, ...)
)
leafnr <- floor(as.numeric(row.names(fit$frame[fit$where, ])))
fit$frame$yval <- as.numeric(row.names(fit$frame))
nodes <- predict(object = fit, newdata = xmis)
donor <- lapply(nodes, function(s) yobs[leafnr == s])
impute <- vapply(seq_along(donor), function(s) sample(donor[[s]], 1), numeric(1))
} else {
# escape with same impute if the dependent does not vary
cat.has.all.obs <- table(yobs) == sum(ry)
if (any(cat.has.all.obs)) {
return(rep(levels(yobs)[cat.has.all.obs], sum(wy)))
}
xy <- cbind(yobs, xobs)
xy <- droplevels(xy)
# FIXME: rpart fails to runs on empty categories in yobs,
# droplevels() removes empty levels, and this is
# likely to present problems further down the road
# potential problem case: table(yobs): 0 10 15, then
# droplevels may forget about category 1
fit <- rpart::rpart(yobs ~ .,
data = xy, method = "class",
control = rpart::rpart.control(minbucket = minbucket, cp = cp, ...)
)
nodes <- predict(object = fit, newdata = xmis)
impute <- apply(nodes,
MARGIN = 1,
FUN = function(s) {
sample(colnames(nodes),
size = 1, prob = s
)
}
)
}
impute
}
mice/R/potthoffroy.R 0000644 0001762 0000144 00000004035 14330031606 014074 0 ustar ligges users #' Potthoff-Roy data
#'
#' Data from Potthoff-Roy (1964) with repeated measures on dental fissures.
#'
#' This data set is the famous Potthoff-Roy data, used to demonstrate MANOVA on
#' repeated measure data. Potthoff and Roy (1964) published classic data on a
#' study in 16 boys and 11 girls, who at ages 8, 10, 12, and 14 had the distance
#' (mm) from the center of the pituitary gland to the pteryomaxillary fissure
#' measured. Changes in pituitary-pteryomaxillary distances during growth is
#' important in orthodontic therapy. The goals of the study were to describe the
#' distance in boys and girls as simple functions of age, and then to compare
#' the functions for boys and girls. The data have been reanalyzed by many
#' authors including Jennrich and Schluchter (1986), Little and Rubin (1987),
#' Pinheiro and Bates (2000), Verbeke and Molenberghs (2000) and Molenberghs and
#' Kenward (2007). See Chapter 9 of Van Buuren (2012) for a challenging
#' exercise using these data.
#'
#' @name potthoffroy
#' @docType data
#' @format \code{tbs} is a data frame with 27 rows and 6 columns:
#' \describe{
#' \item{id}{Person number}
#' \item{sex}{Sex M/F}
#' \item{d8}{Distance at age 8 years}
#' \item{d10}{Distance at age 10 years}
#' \item{d12}{Distance at age 12 years}
#' \item{d14}{Distance at age 14 years}
#' }
#' @source Potthoff, R. F., Roy, S. N. (1964). A generalized multivariate
#' analysis of variance model usefully especially for growth curve problems.
#' \emph{Biometrika}, \emph{51}(3), 313-326.
#'
#' Little, R. J. A., Rubin, D. B. (1987). \emph{Statistical Analysis with
#' Missing Data.} New York: John Wiley & Sons.
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/ex-ch-longitudinal.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' @keywords datasets
#' @examples
#' ### create missing values at age 10 as in Little and Rubin (1987)
#'
#' phr <- potthoffroy
#' idmis <- c(3, 6, 9, 10, 13, 16, 23, 24, 27)
#' phr[idmis, 4] <- NA
#' phr
#'
#' md.pattern(phr)
NULL
mice/R/check.R 0000644 0001762 0000144 00000004027 14330031606 012567 0 ustar ligges users check.data <- function(data, method) {
check.dataform(data)
}
check.dataform <- function(data) {
if (!(is.matrix(data) || is.data.frame(data))) {
stop("Data should be a matrix or data frame", call. = FALSE)
}
if (ncol(data) < 2) {
stop("Data should contain at least two columns", call. = FALSE)
}
data <- as.data.frame(data)
mat <- sapply(data, is.matrix)
df <- sapply(data, is.data.frame)
if (any(mat)) {
stop(
"Cannot handle columns with class matrix: ",
colnames(data)[mat]
)
}
if (any(df)) {
stop(
"Cannot handle columns with class data.frame: ",
colnames(data)[df]
)
}
dup <- duplicated(colnames(data))
if (any(dup)) {
stop(
"Duplicate names found: ",
paste(colnames(data)[dup], collapse = ", ")
)
}
data
}
check.m <- function(m) {
m <- m[1L]
if (!is.numeric(m)) {
stop("Argument m not numeric", call. = FALSE)
}
m <- floor(m)
if (m < 1L) {
stop("Number of imputations (m) lower than 1.", call. = FALSE)
}
m
}
check.cluster <- function(data, predictorMatrix) {
# stop if the cluster variable is a factor
isclassvar <- apply(predictorMatrix == -2, 2, any)
for (j in colnames(predictorMatrix)) {
if (isclassvar[j] && lapply(data, is.factor)[[j]]) {
stop("Convert cluster variable ", j, " to integer by as.integer()")
}
}
TRUE
}
check.ignore <- function(ignore, data) {
if (is.null(ignore)) {
return(rep(FALSE, nrow(data)))
}
if (!is.logical(ignore)) {
stop("Argument ignore not a logical.")
}
if (length(ignore) != nrow(data)) {
stop(
"length(ignore) (", length(ignore),
") does not match nrow(data) (", nrow(data), ")."
)
}
if (sum(!ignore) < 10L) {
warning(
"Fewer than 10 rows for fitting the imputation model. Are you sure?",
call. = FALSE
)
}
ignore
}
check.newdata <- function(newdata, data) {
if (is.null(newdata)) {
stop("No newdata found.")
}
if (!is.data.frame(newdata)) {
stop("newdata not a data.frame.")
}
newdata
}
mice/R/mice.mids.R 0000644 0001762 0000144 00000013371 14334522175 013376 0 ustar ligges users #' Multivariate Imputation by Chained Equations (Iteration Step)
#'
#' Takes a \code{mids} object, and produces a new object of class \code{mids}.
#'
#' This function enables the user to split up the computations of the Gibbs
#' sampler into smaller parts. This is useful for the following reasons:
#' \itemize{ \item RAM memory may become easily exhausted if the number of
#' iterations is large. Returning to prompt/session level may alleviate these
#' problems. \item The user can compute customized convergence statistics at
#' specific points, e.g. after each iteration, for monitoring convergence. -
#' For computing a 'few extra iterations'. } Note: The imputation model itself
#' is specified in the \code{mice()} function and cannot be changed with
#' \code{mice.mids}. The state of the random generator is saved with the
#' \code{mids} object.
#'
#' @param obj An object of class \code{mids}, typically produces by a previous
#' call to \code{mice()} or \code{mice.mids()}
#' @param newdata An optional \code{data.frame} for which multiple imputations
#' are generated according to the model in \code{obj}.
#' @param maxit The number of additional Gibbs sampling iterations.
#' @param printFlag A Boolean flag. If \code{TRUE}, diagnostic information
#' during the Gibbs sampling iterations will be written to the command window.
#' The default is \code{TRUE}.
#' @param ... Named arguments that are passed down to the univariate imputation
#' functions.
#' @author Stef van Buuren, Karin Groothuis-Oudshoorn, 2000
#' @seealso \code{\link{complete}}, \code{\link{mice}}, \code{\link{set.seed}},
#' \code{\link[=mids-class]{mids}}
#' @references Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#' @keywords iteration
#' @examples
#' imp1 <- mice(nhanes, maxit = 1, seed = 123)
#' imp2 <- mice.mids(imp1)
#'
#' # yields the same result as
#' imp <- mice(nhanes, maxit = 2, seed = 123)
#'
#' # verification
#' identical(imp$imp, imp2$imp)
#' #
#' @export
mice.mids <- function(obj, newdata = NULL, maxit = 1, printFlag = TRUE, ...) {
if (!is.mids(obj)) {
stop("Object should be of type mids.")
}
# Set seed to last seed after previous imputation
assign(".Random.seed", obj$lastSeedValue, pos = 1)
# obj contains training data, newdata contains test data
# overwrite obj with combined obj + imp.newdata
if (!is.null(newdata)) {
ignore <- rep(FALSE, nrow(obj$data))
if (!is.null(obj$ignore)) ignore <- obj$ignore
newdata <- check.newdata(newdata, obj$data)
imp.newdata <- mice(newdata,
m = obj$m, maxit = 0,
remove.collinear = FALSE,
remove.constant = FALSE
)
obj <- withCallingHandlers(
rbind.mids(obj, imp.newdata),
warning = function(w) {
if (grepl("iterations differ", w$message)) {
# Catch warnings concerning iterations, these differ by design
invokeRestart("muffleWarning")
}
}
)
# ignore newdata for model building, but do impute
obj$ignore <- c(ignore, rep(TRUE, nrow(newdata)))
}
if (maxit < 1) {
return(obj)
}
loggedEvents <- obj$loggedEvents
state <- list(
it = 0, im = 0, co = 0, dep = "", meth = "",
log = !is.null(loggedEvents)
)
if (is.null(loggedEvents)) {
loggedEvents <- data.frame(
it = 0, im = 0, co = 0, dep = "",
meth = "", out = ""
)
}
# Initialize local variables
call <- match.call()
imp <- obj$imp
where <- obj$where
if (is.null(where)) where <- is.na(obj$data)
blocks <- obj$blocks
if (is.null(blocks)) blocks <- make.blocks(obj$data)
## OK. Iterate.
sumIt <- obj$iteration + maxit
from <- obj$iteration + 1
to <- from + maxit - 1
q <- sampler(
obj$data, obj$m, obj$ignore, where, imp, blocks,
obj$method, obj$visitSequence, obj$predictorMatrix,
obj$formulas, obj$blots, obj$post,
c(from, to), printFlag, ...
)
imp <- q$imp
## combine with previous chainMean and chainVar
vnames <- unique(unlist(obj$blocks))
nvis <- length(vnames)
if (!is.null(obj$chainMean)) {
chainMean <- chainVar <- array(0,
dim = c(nvis, to, obj$m),
dimnames = list(
vnames,
seq_len(to), paste("Chain", seq_len(obj$m))
)
)
for (j in seq_len(nvis)) {
if (obj$iteration == 0) {
chainMean[j, , ] <- q$chainMean[j, , ]
chainVar[j, , ] <- q$chainVar[j, , ]
} else {
chainMean[j, seq_len(obj$iteration), ] <- obj$chainMean[j, , ]
chainVar[j, seq_len(obj$iteration), ] <- obj$chainVar[j, , ]
chainMean[j, from:to, ] <- q$chainMean[j, , ]
chainVar[j, from:to, ] <- q$chainVar[j, , ]
}
}
} else {
chainMean <- chainVar <- NULL
}
if (!state$log) {
loggedEvents <- NULL
}
if (state$log) {
row.names(loggedEvents) <- seq_len(nrow(loggedEvents))
}
## save, and return
midsobj <- list(
data = obj$data, imp = imp, m = obj$m,
where = where, blocks = obj$blocks,
call = call, nmis = obj$nmis,
method = obj$method,
predictorMatrix = obj$predictorMatrix,
visitSequence = obj$visitSequence,
formulas = obj$formulas, post = obj$post,
blots = obj$blots,
ignore = obj$ignore,
seed = obj$seed,
iteration = sumIt,
lastSeedValue = get(".Random.seed",
envir = globalenv(), mode = "integer",
inherits = FALSE
),
chainMean = chainMean,
chainVar = chainVar,
loggedEvents = loggedEvents,
version = packageVersion("mice"),
date = Sys.Date()
)
oldClass(midsobj) <- "mids"
if (!is.null(newdata)) {
include <- c(
rep(FALSE, nrow(midsobj$data) - nrow(newdata)),
rep(TRUE, nrow(newdata))
)
midsobj <- filter(midsobj, include)
}
midsobj
}
mice/R/selfreport.R 0000644 0001762 0000144 00000007370 14330031606 013703 0 ustar ligges users #' Self-reported and measured BMI
#'
#' Dataset containing height and weight data (measured, self-reported) from two
#' studies.
#'
#' This dataset combines two datasets: \code{krul} data (Krul, 2010) (1257
#' persons) and the \code{mgg} data (Van Keulen 2011; Van der Klauw 2011) (803
#' persons). The \code{krul} dataset contains height and weight (both measures
#' and self-reported) from 1257 Dutch adults, whereas the \code{mgg} dataset
#' contains self-reported height and weight for 803 Dutch adults. Section 7.3 in
#' Van Buuren (2012) shows how the missing measured data can be imputed in the
#' \code{mgg} data, so corrected prevalence estimates can be calculated.
#'
#' @name selfreport
#' @aliases selfreport mgg
#' @docType data
#' @format A data frame with 2060 rows and 15 variables:
#' \describe{
#' \item{src}{Study, either \code{krul} or \code{mgg} (factor)}
#' \item{id}{Person identification number}
#' \item{pop}{Population, all \code{NL} (factor)}
#' \item{age}{Age of respondent in years}
#' \item{sex}{Sex of respondent (factor)}
#' \item{hm}{Height measured (cm)}
#' \item{wm}{Weight measured (kg)}
#' \item{hr}{Height reported (cm)}
#' \item{wr}{Weight reported (kg)}
#' \item{prg}{Pregnancy (factor), all \code{Not pregnant}}
#' \item{edu}{Educational level (factor)}
#' \item{etn}{Ethnicity (factor)}
#' \item{web}{Obtained through web survey (factor)}
#' \item{bm}{BMI measured (kg/m2)}
#' \item{br}{BMI reported (kg/m2)}
#' }
#' @source Krul, A., Daanen, H. A. M., Choi, H. (2010). Self-reported and
#' measured weight, height and body mass index (BMI) in Italy, The Netherlands
#' and North America. \emph{European Journal of Public Health}, \emph{21}(4),
#' 414-419.
#'
#' Van Keulen, H.M.,, Chorus, A.M.J., Verheijden, M.W. (2011). \emph{Monitor
#' Convenant Gezond Gewicht Nulmeting (determinanten van) beweeg- en eetgedrag
#' van kinderen (4-11 jaar), jongeren (12-17 jaar) en volwassenen (18+ jaar)}.
#' TNO/LS 2011.016. Leiden: TNO.
#'
#' Van der Klauw, M., Van Keulen, H.M., Verheijden, M.W. (2011). \emph{Monitor
#' Convenant Gezond Gewicht Beweeg- en eetgedrag van kinderen (4-11 jaar),
#' jongeren (12-17 jaar) en volwassenen (18+ jaar) in 2010 en 2011.} TNO/LS
#' 2011.055. Leiden: TNO. (in Dutch)
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-prevalence.html#sec:srcdata}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' @keywords datasets
#' @examples
#' md.pattern(selfreport[, c("age", "sex", "hm", "hr", "wm", "wr")])
#'
#' ### FIMD Section 7.3.5 Application
#'
#' bmi <- function(h, w) {
#' return(w / (h / 100)^2)
#' }
#' init <- mice(selfreport, maxit = 0)
#' meth <- init$meth
#' meth["bm"] <- "~bmi(hm,wm)"
#' pred <- init$pred
#' pred[, c("src", "id", "web", "bm", "br")] <- 0
#' imp <- mice(selfreport, pred = pred, meth = meth, seed = 66573, maxit = 2, m = 1)
#' ## imp <- mice(selfreport, pred=pred, meth=meth, seed=66573, maxit=20, m=10)
#'
#' ### Like FIMD Figure 7.6
#'
#' cd <- complete(imp, 1)
#' xy <- xy.coords(cd$bm, cd$br - cd$bm)
#' plot(xy,
#' col = mdc(2), xlab = "Measured BMI", ylab = "Reported - Measured BMI",
#' xlim = c(17, 45), ylim = c(-5, 5), type = "n", lwd = 0.7
#' )
#' polygon(x = c(30, 20, 30), y = c(0, 10, 10), col = "grey95", border = NA)
#' polygon(x = c(30, 40, 30), y = c(0, -10, -10), col = "grey95", border = NA)
#' abline(0, 0, lty = 2, lwd = 0.7)
#'
#' idx <- cd$src == "krul"
#' xyc <- xy
#' xyc$x <- xy$x[idx]
#' xyc$y <- xy$y[idx]
#' xys <- xy
#' xys$x <- xy$x[!idx]
#' xys$y <- xy$y[!idx]
#' points(xyc, col = mdc(1), cex = 0.7)
#' points(xys, col = mdc(2), cex = 0.7)
#' lines(lowess(xyc), col = mdc(4), lwd = 2)
#' lines(lowess(xys), col = mdc(5), lwd = 2)
#' text(1:4, x = c(40, 28, 20, 32), y = c(4, 4, -4, -4), cex = 3)
#' box(lwd = 1)
NULL
mice/R/stripplot.R 0000644 0001762 0000144 00000031157 14330031647 013563 0 ustar ligges users #' Stripplot of observed and imputed data
#'
#' Plotting methods for imputed data using \pkg{lattice}.
#' \code{stripplot} produces one-dimensional
#' scatterplots. The function
#' automatically separates the observed and imputed data. The
#' functions extend the usual features of \pkg{lattice}.
#'
#' The argument \code{na.groups} may be used to specify (combinations of)
#' missingness in any of the variables. The argument \code{groups} can be used
#' to specify groups based on the variable values themselves. Only one of both
#' may be active at the same time. When both are specified, \code{na.groups}
#' takes precedence over \code{groups}.
#'
#' Use the \code{subset} and \code{na.groups} together to plots parts of the
#' data. For example, select the first imputed data set by by
#' \code{subset=.imp==1}.
#'
#' Graphical parameters like \code{col}, \code{pch} and \code{cex} can be
#' specified in the arguments list to alter the plotting symbols. If
#' \code{length(col)==2}, the color specification to define the observed and
#' missing groups. \code{col[1]} is the color of the 'observed' data,
#' \code{col[2]} is the color of the missing or imputed data. A convenient color
#' choice is \code{col=mdc(1:2)}, a transparent blue color for the observed
#' data, and a transparent red color for the imputed data. A good choice is
#' \code{col=mdc(1:2), pch=20, cex=1.5}. These choices can be set for the
#' duration of the session by running \code{mice.theme()}.
#'
#' @aliases stripplot
#' @param x A \code{mids} object, typically created by \code{mice()} or
#' \code{mice.mids()}.
#' @param data Formula that selects the data to be plotted. This argument
#' follows the \pkg{lattice} rules for \emph{formulas}, describing the primary
#' variables (used for the per-panel display) and the optional conditioning
#' variables (which define the subsets plotted in different panels) to be used
#' in the plot.
#'
#' The formula is evaluated on the complete data set in the \code{long} form.
#' Legal variable names for the formula include \code{names(x$data)} plus the
#' two administrative factors \code{.imp} and \code{.id}.
#'
#' \bold{Extended formula interface:} The primary variable terms (both the LHS
#' \code{y} and RHS \code{x}) may consist of multiple terms separated by a
#' \sQuote{+} sign, e.g., \code{y1 + y2 ~ x | a * b}. This formula would be
#' taken to mean that the user wants to plot both \code{y1 ~ x | a * b} and
#' \code{y2 ~ x | a * b}, but with the \code{y1 ~ x} and \code{y2 ~ x} in
#' \emph{separate panels}. This behavior differs from standard \pkg{lattice}.
#' \emph{Only combine terms of the same type}, i.e. only factors or only
#' numerical variables. Mixing numerical and categorical data occasionally
#' produces odds labeling of vertical axis.
#'
#' For convenience, in \code{stripplot()} and \code{bwplot} the formula
#' \code{y~.imp} may be abbreviated as \code{y}. This applies only to a single
#' \code{y}, and does not (yet) work for \code{y1+y2~.imp}.
#'
#' @param na.groups An expression evaluating to a logical vector indicating
#' which two groups are distinguished (e.g. using different colors) in the
#' display. The environment in which this expression is evaluated in the
#' response indicator \code{is.na(x$data)}.
#'
#' The default \code{na.group = NULL} contrasts the observed and missing data
#' in the LHS \code{y} variable of the display, i.e. groups created by
#' \code{is.na(y)}. The expression \code{y} creates the groups according to
#' \code{is.na(y)}. The expression \code{y1 & y2} creates groups by
#' \code{is.na(y1) & is.na(y2)}, and \code{y1 | y2} creates groups as
#' \code{is.na(y1) | is.na(y2)}, and so on.
#' @param groups This is the usual \code{groups} arguments in \pkg{lattice}. It
#' differs from \code{na.groups} because it evaluates in the completed data
#' \code{data.frame(complete(x, "long", inc=TRUE))} (as usual), whereas
#' \code{na.groups} evaluates in the response indicator. See
#' \code{\link{xyplot}} for more details. When both \code{na.groups} and
#' \code{groups} are specified, \code{na.groups} takes precedence, and
#' \code{groups} is ignored.
#' @param theme A named list containing the graphical parameters. The default
#' function \code{mice.theme} produces a short list of default colors, line
#' width, and so on. The extensive list may be obtained from
#' \code{trellis.par.get()}. Global graphical parameters like \code{col} or
#' \code{cex} in high-level calls are still honored, so first experiment with
#' the global parameters. Many setting consists of a pair. For example,
#' \code{mice.theme} defines two symbol colors. The first is for the observed
#' data, the second for the imputed data. The theme settings only exist during
#' the call, and do not affect the trellis graphical parameters.
#' @param jitter.data See \code{\link[lattice:panel.xyplot]{panel.xyplot}}.
#' @param horizontal See \code{\link[lattice:xyplot]{xyplot}}.
#' @param as.table See \code{\link[lattice:xyplot]{xyplot}}.
#' @param panel See \code{\link{xyplot}}.
#' @param default.prepanel See \code{\link[lattice:xyplot]{xyplot}}.
#' @param outer See \code{\link[lattice:xyplot]{xyplot}}.
#' @param allow.multiple See \code{\link[lattice:xyplot]{xyplot}}.
#' @param drop.unused.levels See \code{\link[lattice:xyplot]{xyplot}}.
#' @param subscripts See \code{\link[lattice:xyplot]{xyplot}}.
#' @param subset See \code{\link[lattice:xyplot]{xyplot}}.
#' @param \dots Further arguments, usually not directly processed by the
#' high-level functions documented here, but instead passed on to other
#' functions.
#' @return The high-level functions documented here, as well as other high-level
#' Lattice functions, return an object of class \code{"trellis"}. The
#' \code{\link[lattice:update.trellis]{update}} method can be used to
#' subsequently update components of the object, and the
#' \code{\link[lattice:print.trellis]{print}} method (usually called by default)
#' will plot it on an appropriate plotting device.
#' @note The first two arguments (\code{x} and \code{data}) are reversed
#' compared to the standard Trellis syntax implemented in \pkg{lattice}. This
#' reversal was necessary in order to benefit from automatic method dispatch.
#'
#' In \pkg{mice} the argument \code{x} is always a \code{mids} object, whereas
#' in \pkg{lattice} the argument \code{x} is always a formula.
#'
#' In \pkg{mice} the argument \code{data} is always a formula object, whereas in
#' \pkg{lattice} the argument \code{data} is usually a data frame.
#'
#' All other arguments have identical interpretation.
#'
#' @author Stef van Buuren
#' @seealso \code{\link{mice}}, \code{\link{xyplot}}, \code{\link{densityplot}},
#' \code{\link{bwplot}}, \code{\link{lattice}} for an overview of the
#' package, as well as \code{\link[lattice:xyplot]{stripplot}},
#' \code{\link[lattice:panel.stripplot]{panel.stripplot}},
#' \code{\link[lattice:print.trellis]{print.trellis}},
#' \code{\link[lattice:trellis.par.get]{trellis.par.set}}
#' @references Sarkar, Deepayan (2008) \emph{Lattice: Multivariate Data
#' Visualization with R}, Springer.
#'
#' van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}: Multivariate
#' Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
#' Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
#' @keywords hplot
#' @examples
#' imp <- mice(boys, maxit = 1)
#'
#' ### stripplot, all numerical variables
#' \dontrun{
#' stripplot(imp)
#' }
#'
#' ### same, but with improved display
#' \dontrun{
#' stripplot(imp, col = c("grey", mdc(2)), pch = c(1, 20))
#' }
#'
#' ### distribution per imputation of height, weight and bmi
#' ### labeled by their own missingness
#' \dontrun{
#' stripplot(imp, hgt + wgt + bmi ~ .imp,
#' cex = c(2, 4), pch = c(1, 20), jitter = FALSE,
#' layout = c(3, 1)
#' )
#' }
#'
#' ### same, but labeled with the missingness of wgt (just four cases)
#' \dontrun{
#' stripplot(imp, hgt + wgt + bmi ~ .imp,
#' na = wgt, cex = c(2, 4), pch = c(1, 20), jitter = FALSE,
#' layout = c(3, 1)
#' )
#' }
#'
#' ### distribution of age and height, labeled by missingness in height
#' ### most height values are missing for those around
#' ### the age of two years
#' ### some additional missings occur in region WEST
#' \dontrun{
#' stripplot(imp, age + hgt ~ .imp | reg, hgt,
#' col = c(grDevices::hcl(0, 0, 40, 0.2), mdc(2)), pch = c(1, 20)
#' )
#' }
#'
#' ### heavily jitted relation between two categorical variables
#' ### labeled by missingness of gen
#' ### aggregated over all imputed data sets
#' \dontrun{
#' stripplot(imp, gen ~ phb, factor = 2, cex = c(8, 1), hor = TRUE)
#' }
#'
#' ### circle fun
#' stripplot(imp, gen ~ .imp,
#' na = wgt, factor = 2, cex = c(8.6),
#' hor = FALSE, outer = TRUE, scales = "free", pch = c(1, 19)
#' )
#' @export
stripplot.mids <- function(x,
data,
na.groups = NULL,
groups = NULL,
as.table = TRUE,
theme = mice.theme(),
allow.multiple = TRUE,
outer = TRUE,
drop.unused.levels = lattice::lattice.getOption("drop.unused.levels"),
panel = lattice::lattice.getOption("panel.stripplot"),
default.prepanel = lattice::lattice.getOption("prepanel.default.stripplot"),
jitter.data = TRUE,
horizontal = FALSE,
...,
subscripts = TRUE,
subset = TRUE) {
call <- match.call()
if (!is.mids(x)) stop("Argument 'x' must be a 'mids' object")
## unpack data and response indicator
cd <- data.frame(complete(x, "long", include = TRUE))
r <- as.data.frame(is.na(x$data))
## evaluate na.group in response indicator
nagp <- eval(expr = substitute(na.groups), envir = r, enclos = parent.frame())
if (is.expression(nagp)) nagp <- eval(expr = nagp, envir = r, enclos = parent.frame())
## evaluate groups in imputed data
ngp <- eval(expr = substitute(groups), envir = cd, enclos = parent.frame())
if (is.expression(ngp)) ngp <- eval(expr = ngp, envir = cd, enclos = parent.frame())
groups <- ngp
## evaluate subset in imputed data
ss <- eval(expr = substitute(subset), envir = cd, enclos = parent.frame())
if (is.expression(ss)) ss <- eval(expr = ss, envir = cd, enclos = parent.frame())
subset <- ss
## evaluate further arguments before parsing
dots <- list(...)
args <- list(
panel = panel,
default.prepanel = default.prepanel,
allow.multiple = allow.multiple,
outer = outer,
drop.unused.levels = drop.unused.levels,
subscripts = subscripts,
as.table = as.table,
jitter.data = jitter.data,
horizontal = horizontal
)
## create formula if not given (in call$data !)
vnames <- names(cd)[-seq_len(2)]
allfactors <- unlist(lapply(cd, is.factor))[-seq_len(2)]
if (missing(data)) {
vnames <- vnames[!allfactors]
formula <- as.formula(paste0(paste0(vnames, collapse = "+"), "~ as.factor(.imp)"))
} else {
## pad abbreviated formula
abbrev <- !any(grepl("~", call$data))
if (abbrev) {
if (length(call$data) > 1) {
stop("Cannot pad extended formula.")
} else {
formula <- as.formula(paste(call$data, "~ as.factor(.imp)", sep = ""))
}
} else {
formula <- data
}
}
## determine the y-variables
form <- lattice::latticeParseFormula(
model = formula, data = cd, subset = subset,
groups = groups, multiple = allow.multiple,
outer = outer, subscripts = TRUE,
drop = drop.unused.levels
)
ynames <- unlist(lapply(strsplit(form$left.name, " \\+ "), rm.whitespace))
xnames <- unlist(lapply(strsplit(form$right.name, " \\+ "), rm.whitespace))
## calculate selection vector gp
nona <- is.null(call$na.groups)
if (!is.null(call$groups) && nona) {
gp <- call$groups
} else {
if (nona) {
na.df <- r[, ynames, drop = FALSE]
gp <- unlist(lapply(na.df, rep, x$m + 1))
} else {
gp <- rep(nagp, length(ynames) * (x$m + 1))
}
}
## change axis defaults of extended formula interface
if (is.null(call$xlab) && !is.na(match(".imp", xnames))) {
dots$xlab <- ""
if (length(xnames) == 1) dots$xlab <- "Imputation number"
}
if (is.null(call$ylab)) {
args$ylab <- ""
if (length(ynames) == 1) args$ylab <- ynames
}
if (is.null(call$scales)) {
args$scales <- list()
if (length(ynames) > 1) {
args$scales <- list(x = list(relation = "free"), y = list(relation = "free"))
}
}
## ready
args <- c(
x = formula, data = list(cd),
groups = list(gp),
args, dots, subset = call$subset
)
## go
tp <- do.call("stripplot", args)
update(tp, par.settings = theme)
}
mice/R/ampute.continuous.R 0000644 0001762 0000144 00000015310 14334522175 015221 0 ustar ligges users #' Multivariate amputation based on continuous probability functions
#'
#' This function creates a missing data indicator for each pattern. The continuous
#' probability distributions (Van Buuren, 2012, pp. 63, 64) will be induced on the
#' weighted sum scores, calculated earlier in the multivariate amputation function
#' \code{\link{ampute}}.
#'
#' @param P A vector containing the pattern numbers of the cases's candidacies.
#' For each case, a value between 1 and #patterns is given. For example, a
#' case with value 2 is candidate for missing data pattern 2.
#' @param scores A list containing vectors with the candidates's weighted sum scores,
#' the result of an underlying function in \code{\link{ampute}}.
#' @param prop A scalar specifying the proportion of missingness. Should be a value
#' between 0 and 1. Default is a missingness proportion of 0.5.
#' @param type A vector of strings containing the type of missingness for each
#' pattern. Either \code{"LEFT"}, \code{"MID"}, \code{"TAIL"} or '\code{"RIGHT"}.
#' If a single missingness type is entered, all patterns will be created by the same
#' type. If missingness types should differ over patterns, a vector of missingness
#' types should be entered. Default is RIGHT for all patterns and is the result of
#' \code{\link{ampute.default.type}}.
#' @return A list containing vectors with \code{0} if a case should be made missing
#' and \code{1} if a case should remain complete. The first vector refers to the
#' first pattern, the second vector to the second pattern, etcetera.
#' @author Rianne Schouten [aut, cre], Gerko Vink [aut], Peter Lugtig [ctb], 2016
#' @seealso \code{\link{ampute}}, \code{\link{ampute.default.type}}
#' @references
#' #'Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-linearnormal.html#sec:generateuni}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' @keywords internal
#' @export
ampute.continuous <- function(P, scores, prop, type) {
# For a test data set, the shift of the logit function is calculated
# in order to obtain the right proportion of missingness (area beneath the curve)
# The set-up for this is created in subsequent lines, it is executed within
# the for loop over i.
testset <- scale(rnorm(n = 10000, mean = 0, sd = 1))
logit <- function(x) exp(x) / (1 + exp(x))
# An empty list is created, type argument is given the right length
R <- vector(mode = "list", length = length(scores))
if (length(type) == 1) {
type <- rep.int(type, length(scores))
}
for (i in seq_along(scores)) {
# The desired function is chosen
formula <- switch(type[i],
LEFT = function(x, b) logit(mean(x) - x + b),
MID = function(x, b) logit(-abs(x - mean(x)) + 0.75 + b),
TAIL = function(x, b) logit(abs(x - mean(x)) - 0.75 + b),
function(x, b) logit(-mean(x) + x + b)
)
shift <- bin.search(
fun = function(shift) {
sum(formula(x = testset, b = shift)) / length(testset)
},
target = prop
)$where
if (length(shift) > 1) {
shift <- shift[1]
}
scores.temp <- scores[[i]]
# empty candidate group
if (length(scores.temp) == 1 && scores.temp == 0) {
R[[i]] <- 0
} else {
if (length(scores.temp) == 1) {
warning(paste("There is only 1 candidate for pattern", i, ",it will be amputed with probability", prop), call. = FALSE)
probs <- prop
} else if (length(unique(scores.temp)) == 1) {
warning(paste("The weighted sum scores of all candidates in pattern", i, "are the same, they will be amputed with probability", prop), call. = FALSE)
probs <- prop
} else {
probs <- formula(x = scores.temp, b = shift)
}
# Based on the probabilities, each candidate will receive a missing data
# indicator 0, meaning it will be made missing or missing data indicator 1,
# meaning the candidate will remain complete.
R.temp <- 1 - rbinom(n = length(scores.temp), size = 1, prob = probs)
R[[i]] <- replace(P, P == (i + 1), R.temp)
R[[i]] <- replace(R[[i]], P != (i + 1), 1)
}
}
R
}
# This is a custom adaptation of function binsearch from package gtools
# (version 3.5.0) that returns the adjustment of the probability curves used
# in the function ampute.continuous in ampute.
bin.search <- function(fun, range = c(-8, 8), ..., target = 0,
lower = ceiling(min(range)),
upper = floor(max(range)),
maxiter = 100, showiter = FALSE) {
lo <- lower
hi <- upper
counter <- 0
val.lo <- round(fun(lo, ...), 3)
val.hi <- round(fun(hi, ...), 3)
sign <- if (val.lo > val.hi) -1 else 1
if (target * sign < val.lo * sign) {
outside.range <- TRUE
} else if (target * sign > val.hi * sign) {
outside.range <- TRUE
} else {
outside.range <- FALSE
}
while (counter < maxiter && !outside.range) {
counter <- counter + 1
if (hi - lo <= (1 / (10^3)) || lo < lower || hi > upper) {
break
}
center <- round((hi - lo) / 2 + lo, 3)
val <- round(fun(center, ...), 3)
if (showiter) {
cat("--------------\n")
cat("Iteration #", counter, "\n")
cat("lo=", lo, "\n")
cat("hi=", hi, "\n")
cat("center=", center, "\n")
cat("fun(lo)=", val.lo, "\n")
cat("fun(hi)=", val.hi, "\n")
cat("fun(center)=", val, "\n")
}
if (val == target) {
val.lo <- val.hi <- val
lo <- hi <- center
break
} else if (sign * val < sign * target) {
lo <- center
val.lo <- val
} else {
hi <- center
val.hi <- val
}
if (showiter) {
cat("new lo=", lo, "\n")
cat("new hi=", hi, "\n")
cat("--------------\n")
}
}
retval <- list(call = match.call(), numiter = counter)
if (outside.range) {
if (target * sign < val.lo * sign) {
warning("The desired proportion of ", target, " is too small; ", val.lo, " is used instead.")
retval$flag <- "Lower Boundary"
retval$where <- lo
retval$value <- val.lo
} else {
warning("The desired proportion of ", target, " is too large; ", val.hi, " is used instead.")
retval$flag <- "Upper Boundary"
retval$where <- hi
retval$value <- val.hi
}
} else if (counter >= maxiter) {
retval$flag <- "Maximum number of iterations reached"
retval$where <- (lo + hi) / 2
retval$value <- (val.lo + val.hi) / 2
} else if (val.lo == target) {
retval$flag <- "Found"
retval$where <- lo
retval$value <- val.lo
} else if (val.hi == target) {
retval$flag <- "Found"
retval$where <- hi
retval$value <- val.hi
} else {
retval$flag <- "Between Elements"
retval$where <- (lo + hi) / 2
retval$value <- (val.lo + val.hi) / 2
}
retval
}
mice/R/blots.R 0000644 0001762 0000144 00000002471 14330031606 012636 0 ustar ligges users #' Creates a \code{blots} argument
#'
#' This helper function creates a valid \code{blots} object. The
#' \code{blots} object is an argument to the \code{mice} function.
#' The name \code{blots} is a contraction of blocks-dots.
#' Through \code{blots}, the user can specify any additional
#' arguments that are specifically passed down to the lowest level
#' imputation function.
#' @param data A \code{data.frame} with the source data
#' @param blocks An optional specification for blocks of variables in
#' the rows. The default assigns each variable in its own block.
#' @return A matrix
#' @seealso \code{\link{make.blocks}}
#' @examples
#' make.predictorMatrix(nhanes)
#' make.blots(nhanes, blocks = name.blocks(c("age", "hyp"), "xxx"))
#' @export
make.blots <- function(data, blocks = make.blocks(data)) {
data <- check.dataform(data)
blots <- vector("list", length(blocks))
for (i in seq_along(blots)) blots[[i]] <- alist()
names(blots) <- names(blocks)
blots
}
check.blots <- function(blots, data, blocks = NULL) {
data <- check.dataform(data)
if (is.null(blots)) {
return(make.blots(data, blocks))
}
blots <- as.list(blots)
for (i in seq_along(blots)) blots[[i]] <- as.list(blots[[i]])
if (length(blots) == length(blocks) && is.null(names(blots))) {
names(blots) <- names(blocks)
}
blots
}
mice/R/blocks.R 0000644 0001762 0000144 00000020105 14347606422 012776 0 ustar ligges users #' Creates a \code{blocks} argument
#'
#' This helper function generates a list of the type needed for
#' \code{blocks} argument in the \code{[=mice]{mice}} function.
#' @param data A \code{data.frame}, character vector with
#' variable names, or \code{list} with variable names.
#' @param partition A character vector of length 1 used to assign
#' variables to blocks when \code{data} is a \code{data.frame}. Value
#' \code{"scatter"} (default) will assign each column to it own
#' block. Value \code{"collect"} assigns all variables to one block,
#' whereas \code{"void"} produces an empty list.
#' @param calltype A character vector of \code{length(block)} elements
#' that indicates how the imputation model is specified. If
#' \code{calltype = "type"} (the default), the underlying imputation
#' model is called by means of the \code{type} argument. The
#' \code{type} argument for block \code{h} is equivalent to
#' row \code{h} in the \code{predictorMatrix}.
#' The alternative is \code{calltype = "formula"}. This will pass
#' \code{formulas[[h]]} to the underlying imputation
#' function for block \code{h}, together with the current data.
#' The \code{calltype} of a block is set automatically during
#' initialization. Where a choice is possible, calltype
#' \code{"formula"} is preferred over \code{"type"} since this is
#' more flexible and extendable. However, what precisely happens
#' depends also on the capabilities of the imputation
#' function that is called.
#' @return A named list of character vectors with variables names.
#' @details Choices \code{"scatter"} and \code{"collect"} represent to two
#' extreme scenarios for assigning variables to imputation blocks.
#' Use \code{"scatter"} to create an imputation model based on
#' \emph{fully conditionally specification} (FCS). Use \code{"collect"} to
#' gather all variables to be imputed by a \emph{joint model} (JM).
#' Scenario's in-between these two extremes represent
#' \emph{hybrid} imputation models that combine FCS and JM.
#'
#' Any variable not listed in will not be imputed.
#' Specification \code{"void"} represents the extreme scenario that
#' skips imputation of all variables.
#'
#' A variable may be a member of multiple blocks. The variable will be
#' re-imputed in each block, so the final imputations for variable
#' will come from the last block that was executed. This scenario
#' may be useful where the same complete background factors appear in
#' multiple imputation blocks.
#'
#' A variable may appear multiple times within a given block. If a univariate
#' imputation model is applied to such a block, then the variable is
#' re-imputed each time as it appears in the block.
#' @examples
#' make.blocks(nhanes)
#' make.blocks(c("age", "sex", "edu"))
#' @export
make.blocks <- function(data,
partition = c("scatter", "collect", "void"),
calltype = "type") {
if (is.vector(data) && !is.list(data)) {
v <- as.list(as.character(data))
names(v) <- as.character(data)
ct <- rep(calltype, length(v))
names(ct) <- names(v)
attr(v, "calltype") <- ct
return(v)
}
if (is.list(data) && !is.data.frame(data)) {
v <- name.blocks(data)
if (length(calltype) == 1L) {
ct <- rep(calltype, length(v))
names(ct) <- names(v)
attr(v, "calltype") <- ct
} else {
ct <- calltype
names(ct) <- names(v)
attr(v, "calltype") <- ct
}
return(v)
}
data <- as.data.frame(data)
partition <- match.arg(partition)
switch(partition,
scatter = {
v <- as.list(names(data))
names(v) <- names(data)
},
collect = {
v <- list(names(data))
names(v) <- "collect"
},
void = {
v <- list()
},
{
v <- as.list(names(data))
names(v) <- names(data)
}
)
if (length(calltype) == 1L) {
ct <- rep(calltype, length(v))
names(ct) <- names(v)
attr(v, "calltype") <- ct
} else {
ct <- calltype
names(ct) <- names(v)
attr(v, "calltype") <- ct
}
v
}
#' Name imputation blocks
#'
#' This helper function names any unnamed elements in the \code{blocks}
#' specification. This is a convenience function.
#' @inheritParams mice
#' @param prefix A character vector of length 1 with the prefix to
#' be using for naming any unnamed blocks with two or more variables.
#' @return A named list of character vectors with variables names.
#' @seealso \code{\link{mice}}
#' @details
#' This function will name any unnamed list elements specified in
#' the optional argument \code{blocks}. Unnamed blocks
#' consisting of just one variable will be named after this variable.
#' Unnamed blocks containing more than one variables will be named
#' by the \code{prefix} argument, padded by an integer sequence
#' stating at 1.
#' @examples
#' blocks <- list(c("hyp", "chl"), AGE = "age", c("bmi", "hyp"), "edu")
#' name.blocks(blocks)
#' @export
name.blocks <- function(blocks, prefix = "B") {
if (!is.list(blocks)) {
return(make.blocks(blocks))
}
if (is.null(names(blocks))) names(blocks) <- rep("", length(blocks))
inc <- 1
for (i in seq_along(blocks)) {
if (names(blocks)[i] != "") next
if (length(blocks[[i]]) == 1) {
names(blocks)[i] <- blocks[[i]][1]
} else {
names(blocks)[i] <- paste0(prefix, inc)
inc <- inc + 1
}
}
blocks
}
check.blocks <- function(blocks, data, calltype = "type") {
data <- check.dataform(data)
blocks <- name.blocks(blocks)
# check that all variable names exists in data
bv <- unique(unlist(blocks))
notFound <- !bv %in% colnames(data)
if (any(notFound)) {
stop(paste(
"The following names were not found in `data`:",
paste(bv[notFound], collapse = ", ")
))
}
if (length(calltype) == 1L) {
ct <- rep(calltype, length(blocks))
names(ct) <- names(blocks)
attr(blocks, "calltype") <- ct
} else {
ct <- calltype
names(ct) <- names(blocks)
attr(blocks, "calltype") <- ct
}
blocks
}
#' Construct blocks from \code{formulas} and \code{predictorMatrix}
#'
#' This helper function attempts to find blocks of variables in the
#' specification of the \code{formulas} and/or \code{predictorMatrix}
#' objects. Blocks specified by \code{formulas} may consist of
#' multiple variables. Blocks specified by \code{predictorMatrix} are
#' assumed to consist of single variables. Any duplicates in names are
#' removed, and the formula specification is preferred.
#' \code{predictorMatrix} and \code{formulas}. When both arguments
#' specify models for the same block, the model for the
#' \code{predictMatrix} is removed, and priority is given to the
#' specification given in \code{formulas}.
#' @inheritParams mice
#' @return A \code{blocks} object.
#' @seealso \code{\link{make.blocks}}, \code{\link{name.blocks}}
#' @examples
#' form <- name.formulas(list(bmi + hyp ~ chl + age, chl ~ bmi))
#' pred <- make.predictorMatrix(nhanes[, c("age", "chl")])
#' construct.blocks(formulas = form, pred = pred)
#' @export
construct.blocks <- function(formulas = NULL, predictorMatrix = NULL) {
blocks.f <- blocks.p <- NULL
if (!is.null(formulas)) {
if (!all(sapply(formulas, is.formula))) {
return(NULL)
}
blocks.f <- name.blocks(lapply(name.formulas(formulas), lhs))
ct <- rep("formula", length(blocks.f))
names(ct) <- names(blocks.f)
attr(blocks.f, "calltype") <- ct
if (is.null(predictorMatrix)) {
return(blocks.f)
}
}
if (!is.null(predictorMatrix)) {
if (is.null(row.names(predictorMatrix))) {
stop("No row names in predictorMatrix", call. = FALSE)
}
blocks.p <- name.blocks(row.names(predictorMatrix))
ct <- rep("type", length(blocks.p))
names(ct) <- names(blocks.p)
attr(blocks.p, "calltype") <- ct
if (is.null(formulas)) {
return(blocks.p)
}
}
# combine into unique blocks
blocknames <- unique(c(names(blocks.f), names(blocks.p)))
vars.f <- unlist(lapply(formulas, lhs))
keep <- setdiff(blocknames, vars.f)
add.p <- blocks.p[names(blocks.p) %in% keep]
blocks <- c(blocks.f, add.p)
ct <- c(
rep("formula", length(formulas)),
rep("type", length(add.p))
)
names(ct) <- names(blocks)
attr(blocks, "calltype") <- ct
blocks
}
mice/R/rm.whitespace.R 0000644 0001762 0000144 00000000417 14330031606 014262 0 ustar ligges users ## Authors: Gerko Vink, Stef van Buuren
rm.whitespace <- function(string, side = "both") {
side <- match.arg(side, c("left", "right", "both"))
pattern <- switch(side,
left = "^\\s+",
right = "\\s+$",
both = "^\\s+|\\s+$"
)
sub(pattern, "", string)
}
mice/R/pops.R 0000644 0001762 0000144 00000004673 14330031606 012502 0 ustar ligges users #' Project on preterm and small for gestational age infants (POPS)
#'
#' Subset of data from the POPS study, a national, prospective study on preterm
#' children, including all liveborn infants <32 weeks gestational age and/or <1500
#' g from 1983 (n = 1338).
#'
#' The data set concerns of subset of 959 children that survived up to the age
#' of 19 years.
#'
#' Hille et al (2005) divided the 959 survivors into three groups: Full
#' responders (examined at an outpatient clinic and completed the
#' questionnaires, n = 596), postal responders (only completed the mailed
#' questionnaires, n = 109), non-responders (did not respond to any of the
#' mailed requests or telephone calls, or could not be traced, n = 254).
#'
#' Compared to the postal and non-responders, the full response group consists
#' of more girls, contains more Dutch children, has higher educational and
#' social economic levels and has fewer handicaps. The responders form a highly
#' selective subgroup in the total cohort.
#'
#' Multiple imputation of this data set has been described in Hille et al (2007)
#' and Van Buuren (2012), chapter 8.
#' @note This dataset is not part of \code{mice}.
#' @name pops
#' @aliases pops pops.pred
#' @docType data
#' @format \code{pops} is a data frame with 959 rows and 86 columns.
#' \code{pops.pred} is the 86 by 86 binary predictor matrix used for specifying
#' the multiple imputation model.
#' @source
#' Hille, E. T. M., Elbertse, L., Bennebroek Gravenhorst, J., Brand, R.,
#' Verloove-Vanhorick, S. P. (2005). Nonresponse bias in a follow-up study of
#' 19-year-old adolescents born as preterm infants. Pediatrics, 116(5):662666.
#'
#' Hille, E. T. M., Weisglas-Kuperus, N., Van Goudoever, J. B., Jacobusse, G.
#' W., Ens-Dokkum, M. H., De Groot, L., Wit, J. M., Geven, W. B., Kok, J. H., De
#' Kleine, M. J. K., Kollee, L. A. A., Mulder, A. L. M., Van Straaten, H. L. M.,
#' De Vries, L. S., Van Weissenbruch, M. M., Verloove-Vanhorick, S. P. (2007).
#' Functional outcomes and participation in young adulthood for very preterm and
#' very low birth weight infants: The Dutch project on preterm and small for
#' gestational age infants at 19 years of age. Pediatrics, 120(3):587595.
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-selective.html#pops-study-19-years-follow-up}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' @keywords datasets
#' @examples
#' pops <- data(pops)
NULL
mice/R/df.residual.R 0000644 0001762 0000144 00000000533 14330031606 013710 0 ustar ligges users df.residual.mira <- function(object, ...) {
fit <- object$analyses[[1]]
df.residual(fit)
}
df.residual.lme <- function(object, ...) {
object$fixDF[["X"]][1]
}
df.residual.mer <- function(object, ...) {
sum(object@dims[2:4] * c(1, -1, -1)) + 1
}
df.residual.multinom <- function(object, ...) {
nrow(object$residuals) - object$edf
}
mice/R/mids.R 0000644 0001762 0000144 00000011676 14334445701 012467 0 ustar ligges users #' Multiply imputed data set (\code{mids})
#'
#' The \code{mids} object contains a multiply imputed data set. The \code{mids} object is
#' generated by functions \code{mice()}, \code{mice.mids()}, \code{cbind.mids()},
#' \code{rbind.mids()} and \code{ibind.mids()}.
#'
#' The \code{mids}
#' class of objects has methods for the following generic functions:
#' \code{print}, \code{summary}, \code{plot}.
#'
#' @section Slots:
#' \describe{
#' \item{\code{.Data}:}{Object of class \code{"list"} containing the
#' following slots:}
#' \item{\code{data}:}{Original (incomplete) data set.}
#' \item{\code{imp}:}{A list of \code{ncol(data)} components with
#' the generated multiple imputations. Each list components is a
#' \code{data.frame} (\code{nmis[j]} by \code{m}) of imputed values
#' for variable \code{j}.}
#' \item{\code{m}:}{Number of imputations.}
#' \item{\code{where}:}{The \code{where} argument of the
#' \code{mice()} function.}
#' \item{\code{blocks}:}{The \code{blocks} argument of the
#' \code{mice()} function.}
#' \item{\code{call}:}{Call that created the object.}
#' \item{\code{nmis}:}{An array containing the number of missing
#' observations per column.}
#' \item{\code{method}:}{A vector of strings of \code{length(blocks}
#' specifying the imputation method per block.}
#' \item{\code{predictorMatrix}:}{A numerical matrix of containing
#' integers specifying the predictor set.}
#' \item{\code{visitSequence}:}{The sequence in which columns are visited.}
#' \item{\code{formulas}:}{A named list of formula's, or expressions that
#' can be converted into formula's by \code{as.formula}. List elements
#' correspond to blocks. The block to which the list element applies is
#' identified by its name, so list names must correspond to block names.}
#' \item{\code{post}:}{A vector of strings of length \code{length(blocks)}
#' with commands for post-processing.}
#' \item{\code{blots}:}{"Block dots". The \code{blots} argument to the \code{mice()}
#' function.}
#' \item{\code{ignore}:}{A logical vector of length \code{nrow(data)} indicating
#' the rows in \code{data} used to build the imputation model. (new in \code{mice 3.12.0})}
#' \item{\code{seed}:}{The seed value of the solution.}
#' \item{\code{iteration}:}{Last Gibbs sampling iteration number.}
#' \item{\code{lastSeedValue}:}{The most recent seed value.}
#' \item{\code{chainMean}:}{A list of \code{m} components. Each
#' component is a \code{length(visitSequence)} by \code{maxit} matrix
#' containing the mean of the generated multiple imputations.
#' The array can be used for monitoring convergence.
#' Note that observed data are not present in this mean.}
#' \item{\code{chainVar}:}{A list with similar structure of \code{chainMean},
#' containing the covariances of the imputed values.}
#' \item{\code{loggedEvents}:}{A \code{data.frame} with five columns
#' containing warnings, corrective actions, and other inside info.}
#' \item{\code{version}:}{Version number of \code{mice} package that
#' created the object.}
#' \item{\code{date}:}{Date at which the object was created.}
#' }
#'
#' @details
#' The \code{loggedEvents} entry is a matrix with five columns containing a
#' record of automatic removal actions. It is \code{NULL} is no action was
#' made. At initialization the program does the following three actions:
#' \describe{
#' \item{1}{A variable that contains missing values, that is not imputed
#' and that is used as a predictor is removed}
#' \item{2}{A constant variable is removed}
#' \item{3}{A collinear variable is removed.}
#' }
#' During iteration, the program does the following
#' actions:
#' \describe{
#' \item{1}{One or more variables that are linearly dependent are removed
#' (for categorical data, a 'variable' corresponds to a dummy variable)}
#' \item{2}{Proportional odds regression imputation that does not converge
#' and is replaced by \code{polyreg}.}
#' }
#'
#' Explanation of elements in \code{loggedEvents}:
#' \describe{
#' \item{\code{it}}{iteration number at which the record was added,}
#' \item{\code{im}}{imputation number,}
#' \item{\code{dep}}{name of the dependent variable,}
#' \item{\code{meth}}{imputation method used,}
#' \item{\code{out}}{a (possibly long) character vector with the
#' names of the altered or removed predictors.}
#' }
#'
#' @note The \code{mice} package does not use
#' the S4 class definitions, and instead relies on the S3 list
#' equivalent \code{oldClass(obj) <- "mids"}.
#'
#' @name mids-class
#' @rdname mids-class
#' @aliases mids-class mids
#' @author Stef van Buuren, Karin Groothuis-Oudshoorn, 2000
#' @seealso \code{\link{mice}}, \code{\link[=mira-class]{mira}},
#' \code{\link{mipo}}
#' @references van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#' @keywords classes
NULL
mice/R/mice.impute.jomoImpute.R 0000644 0001762 0000144 00000006625 14436133166 016101 0 ustar ligges users #' Multivariate multilevel imputation using \code{jomo}
#'
#' This function is a wrapper around the \code{jomoImpute} function
#' from the \code{mitml} package so that it can be called to
#' impute blocks of variables in \code{mice}. The \code{mitml::jomoImpute}
#' function provides an interface to the \code{jomo} package for
#' multiple imputation of multilevel data
#' \url{https://CRAN.R-project.org/package=jomo}.
#' Imputations can be generated using \code{type} or \code{formula},
#' which offer different options for model specification.
#'
#' @name mice.impute.jomoImpute
#' @inheritParams mitml::jomoImpute
#' @param data A data frame containing incomplete and auxiliary variables,
#' the cluster indicator variable, and any other variables that should be
#' present in the imputed datasets.
#' @param type An integer vector specifying the role of each variable
#' in the imputation model (see \code{\link[mitml]{jomoImpute}})
#' @param formula A formula specifying the role of each variable
#' in the imputation model. The basic model is constructed
#' by \code{model.matrix}, thus allowing to include derived variables
#' in the imputation model using \code{I()}. See
#' \code{\link[mitml]{jomoImpute}}.
#' @param format A character vector specifying the type of object that should
#' be returned. The default is \code{format = "list"}. No other formats are
#' currently supported.
#' @param ... Other named arguments: \code{n.burn}, \code{n.iter},
#' \code{group}, \code{prior}, \code{silent} and others.
#' @return A list of imputations for all incomplete variables in the model,
#' that can be stored in the the \code{imp} component of the \code{mids}
#' object.
#' @seealso \code{\link[mitml]{jomoImpute}}
#' @note The number of imputations \code{m} is set to 1, and the function
#' is called \code{m} times so that it fits within the \code{mice}
#' iteration scheme.
#'
#' This is a multivariate imputation function using a joint model.
#' @author Stef van Buuren, 2018, building on work of Simon Grund,
#' Alexander Robitzsch and Oliver Luedtke (authors of \code{mitml} package)
#' and Quartagno and Carpenter (authors of \code{jomo} package).
#' @references
#' Grund S, Luedtke O, Robitzsch A (2016). Multiple
#' Imputation of Multilevel Missing Data: An Introduction to the R
#' Package \code{pan}. SAGE Open.
#'
#' Quartagno M and Carpenter JR (2015).
#' Multiple imputation for IPD meta-analysis: allowing for heterogeneity
#' and studies with missing covariates. Statistics in Medicine,
#' 35:2938-2954, 2015.
#'
#' @family multivariate-2l
#' @keywords datagen
#' @examples
#' \dontrun{
#' # Note: Requires mitml 0.3-5.7
#' blocks <- list(c("bmi", "chl", "hyp"), "age")
#' method <- c("jomoImpute", "pmm")
#' ini <- mice(nhanes, blocks = blocks, method = method, maxit = 0)
#' pred <- ini$pred
#' pred["B1", "hyp"] <- -2
#' imp <- mice(nhanes, blocks = blocks, method = method, pred = pred, maxit = 1)
#' }
#' @export
mice.impute.jomoImpute <- function(data, formula, type, m = 1, silent = TRUE,
format = "imputes", ...) {
install.on.demand("mitml", ...)
nat <- mitml::jomoImpute(
data = data, formula = formula, type = type,
m = m, silent = silent, ...
)
if (format == "native") {
return(nat)
}
cmp <- mitml::mitmlComplete(nat, print = 1)[, names(data)]
if (format == "complete") {
return(cmp)
}
if (format == "imputes") {
return(single2imputes(cmp, is.na(data)))
}
NULL
}
mice/R/fdgs.R 0000644 0001762 0000144 00000003670 14335404116 012445 0 ustar ligges users #' Fifth Dutch growth study 2009
#'
#' Age, height, weight and region of 10030 children measured within the Fifth
#' Dutch Growth Study 2009
#'
#' The data set contains data from children of Dutch descent (biological parents
#' are born in the Netherlands). Children with growth-related diseases were
#' excluded. The data were used to construct new growth charts of children of
#' Dutch descent (Schonbeck 2013), and to calculate overweight and obesity
#' prevalence (Schonbeck 2011).
#'
#' Some groups were underrepresented. Multiple imputation was used to create
#' synthetic cases that were used to correct for the nonresponse. See Van Buuren
#' (2012), chapter 8 for details.
#'
#' @name fdgs
#' @aliases fdgs
#' @docType data
#' @format \code{fdgs} is a data frame with 10030 rows and 8 columns:
#' \describe{
#' \item{id}{Person number}
#' \item{reg}{Region (factor, 5 levels)}
#' \item{age}{Age (years)}
#' \item{sex}{Sex (boy, girl)}
#' \item{hgt}{Height (cm)}
#' \item{wgt}{Weight (kg)}
#' \item{hgt.z}{Height Z-score}
#' \item{wgt.z}{Weight Z-score}
#' }
#' @source Schonbeck, Y., Talma, H., van Dommelen, P., Bakker, B., Buitendijk,
#' S. E., Hirasing, R. A., van Buuren, S. (2011). Increase in prevalence of
#' overweight in Dutch children and adolescents: A comparison of nationwide
#' growth studies in 1980, 1997 and 2009. \emph{PLoS ONE}, \emph{6}(11),
#' e27608.
#'
#' Schonbeck, Y., Talma, H., van Dommelen, P., Bakker, B., Buitendijk, S. E.,
#' Hirasing, R. A., van Buuren, S. (2013). The world's tallest nation has
#' stopped growing taller: the height of Dutch children from 1955 to 2009.
#' \emph{Pediatric Research}, \emph{73}(3), 371-377.
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-nonresponse.html#fifth-dutch-growth-study}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Boca Raton, FL.: Chapman & Hall/CRC Press.
#' @keywords datasets
#' @examples
#'
#'
#' data <- data(fdgs)
#' summary(data)
NULL
mice/R/visitSequence.R 0000644 0001762 0000144 00000003361 14334522175 014353 0 ustar ligges users #' Creates a \code{visitSequence} argument
#'
#' This helper function creates a valid \code{visitSequence}. The
#' \code{visitSequence} is an argument to the \code{mice} function that
#' specifies the sequence in which blocks are imputed.
#' @inheritParams mice
#' @return Vector containing block names
#' @seealso \code{\link{mice}}
#' @examples
#' make.visitSequence(nhanes)
#' @export
make.visitSequence <- function(data = NULL, blocks = NULL) {
if (!is.null(blocks)) {
blocks <- name.blocks(blocks)
return(names(blocks))
}
data <- check.dataform(data)
blocks <- make.blocks(data)
names(blocks)
}
check.visitSequence <- function(visitSequence = NULL,
data, where = NULL, blocks) {
if (is.null(names(blocks)) || any(is.na(names(blocks)))) {
stop("Missing names in `blocks`.")
}
if (is.null(visitSequence)) {
return(make.visitSequence(data, blocks))
}
if (is.null(where)) where <- is.na(data)
nimp <- nimp(where, blocks)
if (length(nimp) == 0) visitSequence <- nimp
if (length(visitSequence) == 1 && is.character(visitSequence)) {
code <- match.arg(visitSequence,
choices = c("roman", "arabic", "monotone", "revmonotone")
)
visitSequence <- switch(code,
roman = names(blocks)[nimp > 0],
arabic = rev(names(blocks)[nimp > 0]),
monotone = names(blocks)[order(nimp)],
revmonotone = rev(names(blocks)[order(nimp)])
)
}
# legacy handling
if (is.numeric(visitSequence)) {
visitSequence <- colnames(data)[visitSequence]
}
# check against names(blocks)
visitSequence <- visitSequence[is.element(visitSequence, names(blocks))]
# remove any blocks without missing data
visitSequence <- names((nimp > 0L)[visitSequence])
visitSequence
}
mice/R/parse.ums.R 0000644 0001762 0000144 00000002524 14334522175 013441 0 ustar ligges users parse.ums <- function(x, ums = NULL, umx = NULL, ...) {
if (is.null(ums)) stop("Unidentifiable model specification (ums) not found.")
if (!is.null(umx)) x <- base::cbind(x, umx)
## Unidentifiable part
# e.g. specified in blots as list(X = list(ums = "-3+2*bmi"))
mnar0 <- gsub("-", "+-", ums)
mnar0 <- unlist(strsplit(mnar0, "+", fixed = TRUE))
if (mnar0[1L] == "") mnar0 <- mnar0[-1L]
if (sum(!grepl("*", mnar0, fixed = TRUE)) == 0L) {
stop("An intercept (constant) term must be included in the expression")
} else if (sum(!grepl("*", mnar0, fixed = TRUE)) == 1L) {
mnar0[!grepl("*", mnar0, fixed = TRUE)] <- paste(
mnar0[!grepl("*", mnar0, fixed = TRUE)], "*intercept",
sep = ""
)
} else if (sum(!grepl("*", mnar0, fixed = TRUE)) > 1L) {
stop("Only one intercept term allowed")
}
mnar <- strsplit(mnar0, "*", fixed = TRUE)
# e.g. c("-3","2")
mnar.parm <- as.numeric(unlist(lapply(mnar, function(x) x[1L])))
# e.g. c("intercept","bmi")
mnar.vars <- unlist(lapply(mnar, function(x) x[2L]))
mnar.parm <- mnar.parm[c(which(mnar.vars == "intercept"), which(mnar.vars != "intercept"))]
mnar.vars <- mnar.vars[c(which(mnar.vars == "intercept"), which(mnar.vars != "intercept"))]
xmnar <- as.matrix(cbind(1, as.matrix(x[, mnar.vars[!mnar.vars == "intercept"]])))
list(delta = mnar.parm, x = xmnar)
}
mice/R/mice.impute.midastouch.R 0000644 0001762 0000144 00000017612 14330031647 016101 0 ustar ligges users #' Imputation by predictive mean matching with distance aided donor selection
#'
#' Imputes univariate missing data using predictive mean matching.
#' @aliases mice.impute.midastouch
#' @inheritParams mice.impute.pmm
#' @param midas.kappa Scalar. If \code{NULL} (default) then the
#' optimal \code{kappa} gets selected automatically. Alternatively, the user
#' may specify a scalar. Siddique and Belin 2008 find \code{midas.kappa = 3}
#' to be sensible.
#' @param outout Logical. If \code{TRUE} (default) one model is estimated
#' for each donor (leave-one-out principle). For speedup choose
#' \code{outout = FALSE}, which estimates one model for all observations
#' leading to in-sample predictions for the donors and out-of-sample
#' predictions for the recipients. Mind the inappropriateness, though.
#' @param neff FOR EXPERTS. Null or character string. The name of an existing
#' environment in which the effective sample size of the donors for each
#' loop (CE iterations times multiple imputations) is supposed to be written.
#' The effective sample size is necessary to compute the correction for the
#' total variance as originally suggested by Parzen, Lipsitz and
#' Fitzmaurice 2005. The objectname is \code{midastouch.neff}.
#' @param debug FOR EXPERTS. Null or character string. The name of an existing
#' environment in which the input is supposed to be written. The objectname
#' is \code{midastouch.inputlist}.
#' @return Vector with imputed data, same type as \code{y}, and of
#' length \code{sum(wy)}
#' @details Imputation of \code{y} by predictive mean matching, based on
#' Rubin (1987, p. 168, formulas a and b) and Siddique and Belin 2008.
#' The procedure is as follows:
#' \enumerate{
#' \item Draw a bootstrap sample from the donor pool.
#' \item Estimate a beta matrix on the bootstrap sample by the leave one out principle.
#' \item Compute type II predicted values for \code{yobs} (nobs x 1) and \code{ymis} (nmis x nobs).
#' \item Calculate the distance between all \code{yobs} and the corresponding \code{ymis}.
#' \item Convert the distances in drawing probabilities.
#' \item For each recipient draw a donor from the entire pool while considering the probabilities from the model.
#' \item Take its observed value in \code{y} as the imputation.
#' }
#' @examples
#' # do default multiple imputation on a numeric matrix
#' imp <- mice(nhanes, method = "midastouch")
#' imp
#'
#' # list the actual imputations for BMI
#' imp$imp$bmi
#'
#' # first completed data matrix
#' complete(imp)
#'
#' # imputation on mixed data with a different method per column
#' mice(nhanes2, method = c("sample", "midastouch", "logreg", "norm"))
#' @author Philipp Gaffert, Florian Meinfelder, Volker Bosch 2015
#' @references
#' Gaffert, P., Meinfelder, F., Bosch V. (2015) Towards an MI-proper
#' Predictive Mean Matching, Discussion Paper.
#' \url{https://www.uni-bamberg.de/fileadmin/uni/fakultaeten/sowi_lehrstuehle/statistik/Personen/Dateien_Florian/properPMM.pdf}
#'
#' Little, R.J.A. (1988), Missing data adjustments in large
#' surveys (with discussion), Journal of Business Economics and
#' Statistics, 6, 287--301.
#'
#' Parzen, M., Lipsitz, S. R., Fitzmaurice, G. M. (2005), A note on reducing
#' the bias of the approximate Bayesian bootstrap imputation variance estimator.
#' Biometrika \bold{92}, 4, 971--974.
#'
#' Rubin, D.B. (1987), Multiple imputation for nonresponse in surveys. New York: Wiley.
#'
#' Siddique, J., Belin, T.R. (2008), Multiple imputation using an iterative
#' hot-deck with distance-based donor selection. Statistics in medicine,
#' \bold{27}, 1, 83--102
#'
#' Van Buuren, S., Brand, J.P.L., Groothuis-Oudshoorn C.G.M., Rubin, D.B. (2006),
#' Fully conditional specification in multivariate imputation.
#' \emph{Journal of Statistical Computation and Simulation}, \bold{76}, 12,
#' 1049--1064.
#'
#' Van Buuren, S., Groothuis-Oudshoorn, K. (2011), \code{mice}: Multivariate
#' Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}, 3, 1--67. \doi{10.18637/jss.v045.i03}
#' @family univariate imputation functions
#' @keywords datagen
#' @export
mice.impute.midastouch <- function(y, ry, x, wy = NULL, ridge = 1e-05,
midas.kappa = NULL,
outout = TRUE, neff = NULL,
debug = NULL, ...) {
if (is.null(wy)) {
wy <- !ry
}
# auxiliaries
if (!is.null(debug)) {
midastouch.inputlist <- list(y = y, ry = ry, x = x, omega = NULL)
}
sminx <- .Machine$double.eps^(1 / 4)
# ensure data format
x <- data.matrix(x)
storage.mode(x) <- "numeric"
X <- cbind(1, x)
y <- as.numeric(y)
# get data dimensions
nobs <- sum(ry)
nmis <- sum(wy)
n <- length(ry)
obsind <- ry
misind <- wy
m <- ncol(X)
yobs <- y[obsind]
Xobs <- X[obsind, , drop = FALSE]
Xmis <- X[misind, , drop = FALSE]
# P-Step
# bootstrap
omega <- bootfunc.plain(nobs)
if (!is.null(debug)) {
midastouch.inputlist$omega <- omega
assign(
x = "midastouch.inputlist", value = midastouch.inputlist,
envir = get(debug)
)
}
# beta estimation
CX <- omega * Xobs
XCX <- crossprod(Xobs, CX)
if (ridge > 0) {
diag(XCX) <- diag(XCX) * (1 + c(0, rep(ridge, m - 1)))
}
# check if any diagonal element is exactly zero
diag0 <- diag(XCX) == 0
if (any(diag0)) {
diag(XCX)[diag0] <- max(sminx, ridge)
}
Xy <- crossprod(CX, yobs)
beta <- solve(XCX, Xy)
yhat.obs <- c(Xobs %*% beta)
# kappa estimation
if (is.null(midas.kappa)) {
mean.y <- as.vector(crossprod(yobs, omega) / nobs)
eps <- yobs - yhat.obs
r2 <- 1 - c(crossprod(omega, eps^2) / crossprod(omega, (yobs - mean.y)^2))
# slight deviation from the paper to ensure real results paper: a tiny
# delta is added to the denominator R Code: min function is used, note
# that this correction gets active for r2>.999 only
midas.kappa <- min((50 * r2 / (1 - r2))^(3 / 8), 100)
# if r2 cannot be determined (eg zero variance in yhat), use 3 as
# suggested by Siddique / Belin
if (is.na(midas.kappa)) {
midas.kappa <- 3
}
}
# I-Step
if (outout) {
# P-step if out of sample predictions for donors
# estimate one model per donor by leave-one-out
XXarray_pre <- t(t(apply(X = Xobs, MARGIN = 1, FUN = tcrossprod)) *
omega)
ridgeind <- c(1:(m - 1)) * (m + 1) + 1
if (ridge > 0) {
XXarray_pre[ridgeind, ] <- XXarray_pre[ridgeind, ] * (1 + ridge)
}
XXarray <- c(XCX) - XXarray_pre
# check if any diagonal element is exactly zero
diag0 <- XXarray[ridgeind, ] == 0
if (any(diag0)) {
XXarray[ridgeind, ][diag0] <- max(sminx, ridge)
}
Xyarray <- c(Xy) - t(Xobs * yobs * omega)
BETAarray <- apply(rbind(XXarray, Xyarray), 2, function(x, m) {
solve(a = matrix(head(x, m^2), m), b = tail(x, m))
}, m = m)
YHATdon <- rowSums(Xobs * t(BETAarray))
# each recipient has nobs different yhats
YHATrec <- Xmis %*% BETAarray
# distance calculations
dist.mat <- YHATdon - t(YHATrec)
} else {
yhat.mis <- c(Xmis %*% beta)
dist.mat <- yhat.obs - matrix(
data = yhat.mis, nrow = nobs, ncol = nmis,
byrow = TRUE
)
}
# convert distances to drawing probs // ensure real results
delta.mat <- 1 / ((abs(dist.mat))^midas.kappa)
delta.mat <- minmax(delta.mat)
probs <- delta.mat * omega
csums <- minmax(colSums(probs, na.rm = TRUE))
probs <- t(t(probs) / csums)
# calculate neff
if (!is.null(neff)) {
if (!exists("midastouch.neff", envir = get(neff))) {
assign(x = "midastouch.neff", value = list(), envir = get(neff))
}
midastouch.neff <- get("midastouch.neff", envir = get(neff))
midastouch.neff[[length(midastouch.neff) + 1]] <- mean(1 / rowSums((t(delta.mat) / csums)^2))
assign(x = "midastouch.neff", value = midastouch.neff, envir = get(neff))
}
# return result
index <- apply(probs, 2, sample, x = nobs, size = 1, replace = FALSE)
yimp <- y[obsind][index]
yimp
}
mice/R/xyplot.mads.R 0000644 0001762 0000144 00000010726 14330031606 013777 0 ustar ligges users #' Scatterplot of amputed and non-amputed data against weighted sum scores
#'
#' Plotting method to investigate relation between amputed data and the weighted sum
#' scores. Based on \code{\link{lattice}}. \code{xyplot} produces scatterplots.
#' The function plots the variables against the weighted sum scores. The function
#' automatically separates the amputed and non-amputed data to see the relation between
#' the amputation and the weighted sum scores.
#'
#' @param x A \code{mads} object, typically created by \code{\link{ampute}}.
#' @param data A string or vector of variable names that needs to be plotted. As
#' a default, all variables will be plotted.
#' @param which.pat A scalar or vector indicating which patterns need to be plotted.
#' As a default, all patterns are plotted.
#' @param standardized Logical. Whether the scatterplots need to be created
#' from standardized data or not. Default is TRUE.
#' @param layout A vector of two values indicating how the scatterplots of one
#' pattern should be divided over the plot. For example, \code{c(2, 3)} indicates
#' that the scatterplots of six variables need to be placed on 3 rows and 2 columns.
#' There are several defaults for different #variables. Note that for more than
#' 9 variables, multiple plots will be created automatically.
#' @param colors A vector of two RGB values defining the colors of the non-amputed and
#' amputed data respectively. RGB values can be obtained with \code{\link{hcl}}.
#' @param \dots Not used, but for consistency with generic
#' @return A list containing the scatterplots. Note that a new pattern
#' will always be shown in a new plot.
#' @note The \code{mads} object contains all the information you need to
#' make any desired plots. Check \code{\link{mads-class}} or the vignette \emph{Multivariate
#' Amputation using Ampute} to understand the contents of class object \code{mads}.
#' @author Rianne Schouten, 2016
#' @seealso \code{\link{ampute}}, \code{\link{bwplot}}, \code{\link{Lattice}} for
#' an overview of the package, \code{\link{mads-class}}
#' @export
xyplot.mads <- function(x, data, which.pat = NULL,
standardized = TRUE, layout = NULL,
colors = mdc(1:2), ...) {
if (!is.mads(x)) {
stop("Object is not of class mads")
}
if (missing(data)) data <- NULL
yvar <- data
if (is.null(yvar)) {
varlist <- colnames(x$amp)
} else {
varlist <- yvar
}
if (is.null(which.pat)) {
pat <- nrow(x$patterns)
which.pat <- seq_len(pat)
} else {
pat <- length(which.pat)
}
if (standardized) {
dat <- data.frame(scale(x$data))
xlab <- "Standardized values in pattern"
} else {
dat <- x$data
xlab <- "Data values in pattern"
}
data <- NULL
for (i in seq_len(pat)) {
can <- which(x$cand == which.pat[i])
mis <- matrix(NA, nrow = length(can), ncol = 3)
nc <- which(x$patterns[which.pat[i], ] == 0)
if (length(nc) > 1) {
mis[apply(is.na(x$amp[can, nc]), 1, all), 1] <- 1
mis[is.na(mis[, 1]), 1] <- 0
} else if (length(nc) == 1) {
mis[is.na(x$amp[can, nc]), 1] <- 1
mis[is.na(mis[, 1]), 1] <- 0
}
mis[, 2] <- rep.int(which.pat[i], length(can))
mis[, 3] <- unname(x$scores[[which.pat[i]]])
data <- rbind(data, cbind(mis, dat[can, ]))
}
colnames(data) <- c(".amp", ".pat", "scores", names(x$data))
data$.amp <- factor(data$.amp, levels = c(0, 1))
formula <- as.formula(paste0("scores ~ ", paste0(varlist, collapse = "+")))
if (is.null(layout)) {
if (length(varlist) > 6) {
layout <- c(3, 3)
} else if (length(varlist) > 4) {
layout <- c(3, 2)
} else if (length(varlist) > 2) {
layout <- c(2, 2)
} else if (length(varlist) > 1) {
layout <- c(2, 1)
}
}
theme <- list(
superpose.symbol = list(col = colors, pch = 1),
plot.symbol = list(col = colors, pch = 1),
strip.background = list(col = "grey95")
)
key <- list(
columns = 2, points = list(col = colors, pch = 1),
text = list(c("Non-Amputed Data", "Amputed Data"))
)
p <- stats::setNames(
vector(mode = "list", length = pat),
paste("Scatterplot Pattern", which.pat)
)
for (i in seq_len(pat)) {
p[[paste("Scatterplot Pattern", which.pat[i])]] <-
xyplot(
x = formula, data = data[data$.pat == which.pat[i], ],
groups = data$.amp, par.settings = theme,
multiple = TRUE, outer = TRUE, layout = layout, key = key,
ylab = "Weighted sum scores",
xlab = paste(xlab, which.pat[i])
)
}
p
}
mice/R/barnard.rubin.R 0000644 0001762 0000144 00000000425 14330031647 014244 0 ustar ligges users barnard.rubin <- function(m, b, t, dfcom = Inf) {
lambda <- (1 + 1 / m) * b / t
lambda[lambda < 1e-04] <- 1e-04
dfold <- (m - 1) / lambda^2
dfobs <- (dfcom + 1) / (dfcom + 3) * dfcom * (1 - lambda)
ifelse(is.infinite(dfcom), dfold, dfold * dfobs / (dfold + dfobs))
}
mice/R/zzz.R 0000644 0001762 0000144 00000000732 14330031606 012346 0 ustar ligges users #' Echoes the package version number
#'
#' @param pkg A character vector with the package name.
#' @return A character vector containing the package name, version number and
#' installed directory.
#' @author Stef van Buuren, Oct 2010
#' @keywords misc
#' @examples
#' version()
#' version("base")
#' @export
version <- function(pkg = "mice") {
lib <- dirname(system.file(package = pkg))
d <- packageDescription(pkg)
return(paste(d$Package, d$Version, d$Date, lib))
}
mice/R/pool.compare.R 0000644 0001762 0000144 00000015767 14330034531 014125 0 ustar ligges users #' Compare two nested models fitted to imputed data
#'
#' This function is deprecated in V3. Use \code{\link{D1}} or
#' \code{\link{D3}} instead.
#'
#' Compares two nested models after m repeated complete data analysis
#'
#' The function is based on the article of Meng and Rubin (1992). The
#' Wald-method can be found in paragraph 2.2 and the likelihood method can be
#' found in paragraph 3. One could use the Wald method for comparison of linear
#' models obtained with e.g. \code{lm} (in \code{with.mids()}). The likelihood
#' method should be used in case of logistic regression models obtained with
#' \code{glm()} in \code{with.mids()}.
#'
#' The function assumes that \code{fit1} is the
#' larger model, and that model \code{fit0} is fully contained in \code{fit1}.
#' In case of \code{method='wald'}, the null hypothesis is tested that the extra
#' parameters are all zero.
#'
#' @param fit1 An object of class 'mira', produced by \code{with.mids()}.
#' @param fit0 An object of class 'mira', produced by \code{with.mids()}. The
#' model in \code{fit0} is a nested fit0 of \code{fit1}.
#' @param method Either \code{"wald"} or \code{"likelihood"} specifying
#' the type of comparison. The default is \code{"wald"}.
#' @param data No longer used.
#' @return A list containing several components. Component \code{call} is
#' the call to the \code{pool.compare} function. Component \code{call11} is
#' the call that created \code{fit1}. Component \code{call12} is the
#' call that created the imputations. Component \code{call01} is the
#' call that created \code{fit0}. Component \code{call02} is the
#' call that created the imputations. Components \code{method} is the
#' method used to compare two models: 'Wald' or 'likelihood'. Component
#' \code{nmis} is the number of missing entries for each variable.
#' Component \code{m} is the number of imputations.
#' Component \code{qhat1} is a matrix, containing the estimated coefficients of the
#' \emph{m} repeated complete data analyses from \code{fit1}.
#' Component \code{qhat0} is a matrix, containing the estimated coefficients of the
#' \emph{m} repeated complete data analyses from \code{fit0}.
#' Component \code{ubar1} is the mean of the variances of \code{fit1},
#' formula (3.1.3), Rubin (1987).
#' Component \code{ubar0} is the mean of the variances of \code{fit0},
#' formula (3.1.3), Rubin (1987).
#' Component \code{qbar1} is the pooled estimate of \code{fit1}, formula (3.1.2) Rubin
#' (1987).
#' Component \code{qbar0} is the pooled estimate of \code{fit0}, formula (3.1.2) Rubin
#' (1987).
#' Component \code{Dm} is the test statistic.
#' Component \code{rm} is the relative increase in variance due to nonresponse, formula
#' (3.1.7), Rubin (1987).
#' Component \code{df1}: df1 = under the null hypothesis it is assumed that \code{Dm} has an F
#' distribution with (df1,df2) degrees of freedom.
#' Component \code{df2}: df2.
#' Component \code{pvalue} is the P-value of testing whether the model \code{fit1} is
#' statistically different from the smaller \code{fit0}.
#' @author Karin Groothuis-Oudshoorn and Stef van Buuren, 2009
#' @seealso \code{\link{lm.mids}}, \code{\link{glm.mids}}
#' @references Li, K.H., Meng, X.L., Raghunathan, T.E. and Rubin, D. B. (1991).
#' Significance levels from repeated p-values with multiply-imputed data.
#' Statistica Sinica, 1, 65-92.
#'
#' Meng, X.L. and Rubin, D.B. (1992). Performing likelihood ratio tests with
#' multiple-imputed data sets. Biometrika, 79, 103-111.
#'
#' van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}: Multivariate
#' Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
#' Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
#' @keywords htest
#' @export
pool.compare <- function(fit1, fit0, method = c("wald", "likelihood"),
data = NULL) {
.Deprecated("D1")
# Check the arguments
call <- match.call()
method <- match.arg(method)
fits1 <- getfit(fit1)
fits0 <- getfit(fit0)
if (length(fits1) != length(fits0)) {
stop("unequal number of imputations for 'fit1' and 'fit0'", call. = FALSE)
}
if (length(fits1) < 2L) {
stop("at least two imputations are needed", call. = FALSE)
}
m <- length(fits1)
est1 <- pool(fit1)
est0 <- pool(fit0)
dimQ1 <- length(getqbar(est1))
dimQ2 <- dimQ1 - length(getqbar(est0))
# Check: Only need the lm or lmer object
formula1 <- formula(getfit(fit1, 1L))
formula0 <- formula(getfit(fit0, 1L))
vars1 <- est1$pooled$term
vars0 <- est0$pooled$term
if (is.null(vars1) || is.null(vars0)) {
stop("coefficients do not have names", call. = FALSE)
}
if (dimQ2 < 1L) {
stop("Model 'fit1' not larger than 'fit0'", call. = FALSE)
}
if (!setequal(vars0, intersect(vars0, vars1))) {
stop("Model 'fit0' not contained in 'fit1'", call. = FALSE)
}
if (method == "wald") {
# Reference: paragraph 2.2, Article Meng & Rubin,
# Biometrika, 1992. When two objects are to be compared
# we need to calculate matrix Q
Q <- diag(dimQ1)
where_new_vars <- which(!(vars1 %in% vars0))
Q <- Q[where_new_vars, , drop = FALSE]
qbar <- Q %*% getqbar(est1)
Ubar <- Q %*% diag(est1$pooled$ubar) %*% (t(Q))
Bm <- Q %*% diag(est1$pooled$b) %*% (t(Q))
rm <- (1 + 1 / m) * sum(diag(Bm %*% (solve(Ubar)))) / dimQ2
Dm <- (t(qbar)) %*% (solve(Ubar)) %*% qbar / (dimQ2 * (1 + rm))
deviances <- NULL
}
if (method == "likelihood") {
# Calculate for each imputed dataset the deviance between the two
# models with its estimated coefficients
dev1.M <- lapply(fits1, glance) %>%
bind_rows() %>%
pull(.data$deviance)
dev0.M <- lapply(fits0, glance) %>%
bind_rows() %>%
pull(.data$deviance)
# Calculate for each imputed dataset the deviance between the two
# models with the pooled coefficients
qbar1 <- getqbar(pool(fits1))
mds1 <- lapply(fits1, fix.coef, beta = qbar1)
dev1.L <- lapply(mds1, glance) %>%
bind_rows() %>%
pull(.data$deviance)
qbar0 <- getqbar(pool(fits0))
mds0 <- lapply(fits0, fix.coef, beta = qbar0)
dev0.L <- lapply(mds0, glance) %>%
bind_rows() %>%
pull(.data$deviance)
deviances <- list(
dev1.M = dev1.M, dev0.M = dev0.M,
dev1.L = dev1.L, dev0.L = dev0.L
)
dev.M <- mean(dev0.M - dev1.M)
dev.L <- mean(dev0.L - dev1.L)
rm <- ((m + 1) / (dimQ2 * (m - 1))) * (dev.M - dev.L)
Dm <- dev.L / (dimQ2 * (1 + rm))
}
# Degrees of freedom for F distribution, same for both methods
v <- dimQ2 * (m - 1)
if (v > 4) {
# according to Li 1991
w <- 4 + (v - 4) * ((1 + (1 - 2 / v) * (1 / rm))^2)
} else {
w <- v * (1 + 1 / dimQ2) * ((1 + 1 / rm)^2) / 2
}
statistic <- list(
call = call, call11 = fit1$call, call12 = fit1$call1,
call01 = fit0$call, call02 = fit0$call1,
method = method, nmis = fit1$nmis, m = m,
qbar1 = getqbar(est1), qbar0 = getqbar(est0),
ubar1 = est1$pooled$ubar, ubar0 = est0$pooled$ubar,
deviances = deviances,
Dm = Dm, rm = rm, df1 = dimQ2, df2 = w,
pvalue = pf(Dm, dimQ2, w, lower.tail = FALSE)
)
statistic
}
mice/R/mice.impute.2l.bin.R 0000644 0001762 0000144 00000011103 14334522175 015017 0 ustar ligges users #' Imputation by a two-level logistic model using \code{glmer}
#'
#' Imputes univariate systematically and sporadically missing data
#' using a two-level logistic model using \code{lme4::glmer()}
#'
#' Data are missing systematically if they have not been measured, e.g., in the
#' case where we combine data from different sources. Data are missing sporadically
#' if they have been partially observed.
#'
#' @inheritParams mice.impute.2l.lmer
#' @param intercept Logical determining whether the intercept is automatically
#' added.
#' @param \dots Arguments passed down to \code{glmer}
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @author Shahab Jolani, 2015; adapted to mice, SvB, 2018
#' @references
#' Jolani S., Debray T.P.A., Koffijberg H., van Buuren S., Moons K.G.M. (2015).
#' Imputation of systematically missing predictors in an individual
#' participant data meta-analysis: a generalized approach using MICE.
#' \emph{Statistics in Medicine}, 34:1841-1863.
#' @family univariate-2l
#' @keywords datagen
#' @examples
#' library(tidyr)
#' library(dplyr)
#' data("toenail2")
#' data <- tidyr::complete(toenail2, patientID, visit) %>%
#' tidyr::fill(treatment) %>%
#' dplyr::select(-time) %>%
#' dplyr::mutate(patientID = as.integer(patientID))
#' \dontrun{
#' pred <- mice(data, print = FALSE, maxit = 0, seed = 1)$pred
#' pred["outcome", "patientID"] <- -2
#' imp <- mice(data, method = "2l.bin", pred = pred, maxit = 1, m = 1, seed = 1)
#' }
#' @export
mice.impute.2l.bin <- function(y, ry, x, type,
wy = NULL, intercept = TRUE, ...) {
install.on.demand("lme4", ...)
if (is.null(wy)) wy <- !ry
if (intercept) {
x <- cbind(1, as.matrix(x))
type <- c(2, type)
names(type)[1] <- colnames(x)[1] <- "(Intercept)"
}
clust <- names(type[type == -2])
rande <- names(type[type == 2])
fixe <- names(type[type > 0])
X <- x[, fixe, drop = FALSE]
Z <- x[, rande, drop = FALSE]
xobs <- x[ry, , drop = FALSE]
yobs <- y[ry]
# create formula, use [-1] to remove intercept
fr <- ifelse(length(rande) > 1,
paste("+ ( 1 +", paste(rande[-1L], collapse = "+")),
"+ ( 1 "
)
randmodel <- paste(
"yobs ~ ", paste(fixe[-1L], collapse = "+"),
fr, "|", clust, ")"
)
suppressWarnings(fit <- try(
lme4::glmer(formula(randmodel),
data = data.frame(yobs, xobs),
family = binomial, ...
),
silent = TRUE
))
if (!is.null(attr(fit, "class"))) {
if (attr(fit, "class") == "try-error") {
warning("glmer does not run. Simplify imputation model")
return(y[wy])
}
}
# draw beta*
beta <- lme4::fixef(fit)
rv <- t(chol(vcov(fit)))
beta.star <- beta + rv %*% rnorm(ncol(rv))
# calculate psi*
psi.hat <- matrix(lme4::VarCorr(fit)[[1L]],
nrow = dim(lme4::VarCorr(fit)[[1L]])[1L]
)
s <- nrow(psi.hat) * psi.hat
rancoef <- as.matrix(lme4::ranef(fit)[[1L]])
lambda <- t(rancoef) %*% rancoef
temp <- lambda + s
if (attr(suppressWarnings(chol(temp, pivot = TRUE)), "rank") != nrow(temp)) {
warning("The cov matrix is not full rank")
}
temp <- MASS::ginv(temp)
ev <- eigen(temp)
if (mode(ev$values) == "complex") {
ev$values <- suppressWarnings(as.numeric(ev$values))
ev$vectors <- suppressWarnings(matrix(as.numeric(ev$vectors),
nrow = length(ev$values)
))
warning("The cov matrix is complex")
}
if (sum(ev$values < 0) > 0) {
ev$values[ev$values < 0] <- 0
temp <- ev$vectors %*% diag(ev$values, nrow = length(ev$values)) %*% t(ev$vectors)
}
deco <- ev$vectors %*% diag(sqrt(ev$values), nrow = length(ev$values))
temp.psi.star <- stats::rWishart(
1,
nrow(rancoef) + nrow(psi.hat),
diag(nrow(psi.hat))
)[, , 1L]
psi.star <- MASS::ginv(deco %*% temp.psi.star %*% t(deco))
# psi.star positive definite?
if (!isSymmetric(psi.star)) psi.star <- (psi.star + t(psi.star)) / 2
valprop <- eigen(psi.star)
if (sum(valprop$values < 0) > 0) {
valprop$values[valprop$values < 0] <- 0
psi.star <- valprop$vectors %*% diag(valprop$values) %*% t(valprop$vectors)
}
# find clusters for which we need imputes
clmis <- x[wy, clust]
# the main imputation task
for (i in clmis) {
bi.star <- t(MASS::mvrnorm(
n = 1L, mu = rep(0, nrow(psi.star)),
Sigma = psi.star
))
idx <- wy & (x[, clust] == i)
logit <- X[idx, , drop = FALSE] %*% beta.star +
Z[idx, , drop = FALSE] %*% matrix(bi.star, ncol = 1)
vec <- rbinom(nrow(logit), 1, as.vector(1 / (1 + exp(-logit))))
if (is.factor(y)) {
vec <- factor(vec, c(0, 1), levels(y))
}
y[idx] <- vec
}
y[wy]
}
mice/R/brandsma.R 0000644 0001762 0000144 00000004320 14330031606 013275 0 ustar ligges users #' Brandsma school data used Snijders and Bosker (2012)
#'
#' Dataset with raw data from Snijders and Bosker (2012) containing
#' data from 4106 pupils attending 216 schools. This dataset
#' includes all pupils and schools with missing data.
#'
#' @name brandsma
#' @docType data
#' @format \code{brandsma} is a data frame with 4106 rows and 14 columns:
#' \describe{
#' \item{\code{sch}}{School number}
#' \item{\code{pup}}{Pupil ID}
#' \item{\code{iqv}}{IQ verbal}
#' \item{\code{iqp}}{IQ performal}
#' \item{\code{sex}}{Sex of pupil}
#' \item{\code{ses}}{SES score of pupil}
#' \item{\code{min}}{Minority member 0/1}
#' \item{\code{rpg}}{Number of repeated groups, 0, 1, 2}
#' \item{\code{lpr}}{language score PRE}
#' \item{\code{lpo}}{language score POST}
#' \item{\code{apr}}{Arithmetic score PRE}
#' \item{\code{apo}}{Arithmetic score POST}
#' \item{\code{den}}{Denomination classification 1-4 - at school level}
#' \item{\code{ssi}}{School SES indicator - at school level}
#' }
#'
#' @note This dataset is constructed from the raw data. There are
#' a few differences with the data set used in Chapter 4 and 5
#' of Snijders and Bosker:
#' \enumerate{
#' \item All schools are included, including the five school with
#' missing values on \code{langpost}.
#' \item Missing \code{denomina} codes are left as missing.
#' \item Aggregates are undefined in the presence of missing data
#' in the underlying values.
#' Variables \code{ses}, \code{iqv} and \code{iqp} are in their
#' original scale, and not globally centered.
#' No aggregate variables at the school level are included.
#' \item There is a wider selection of original variables. Note
#' however that the source data contain an even wider set of
#' variables.
#' }
#'
#' @source Constructed from \code{MLbook_2nded_total_4106-99.sav} from
#' \url{https://www.stats.ox.ac.uk/~snijders/mlbook.htm} by function
#' \code{data-raw/R/brandsma.R}
#'
#' @references
#' Brandsma, HP and Knuver, JWM (1989), Effects of school and
#' classroom characteristics on pupil progress in language and arithmetic.
#' International Journal of Educational Research, 13(7), 777 - 788.
#'
#' Snijders, TAB and Bosker RJ (2012). Multilevel Analysis, 2nd Ed. Sage,
#' Los Angeles, 2012.
#' @keywords datasets
NULL
mice/R/check.deprecated.R 0000644 0001762 0000144 00000001102 14330031606 014655 0 ustar ligges users # contributed by Simon Grund, #137
check.deprecated <- function(...) {
# print warnings for deprecated argument names
nms <- names(list(...))
replace.args <- list(
imputationMethod = "method",
defaultImputationMethod = "defaultMethod",
form = "formulas"
)
wrn <- names(replace.args) %in% nms
if (any(wrn)) {
for (i in which(wrn)) {
msg <- paste0(
"The '", names(replace.args)[i],
"' argument is no longer supported. Please use '",
replace.args[i], "' instead."
)
warning(msg)
}
}
invisible(NULL)
}
mice/R/mice.impute.mnar.norm.R 0000644 0001762 0000144 00000016475 14334522175 015663 0 ustar ligges users #' Imputation under MNAR mechanism by NARFCS
#'
#' Imputes univariate data under a user-specified MNAR mechanism by
#' linear or logistic regression and NARFCS. Sensitivity analysis under
#' different model specifications may shed light on the impact of
#' different MNAR assumptions on the conclusions.
#'
#' @rdname mice.impute.mnar
#' @aliases mice.impute.mnar.norm mnar.norm
#' mice.impute.mnar.logreg mnar.logreg
#' @inheritParams mice.impute.pmm
#' @param ums A string containing the specification of the
#' unidentifiable part of the imputation model (the *unidentifiable
#' model specification"), that is, the desired \eqn{\delta}-adjustment
#' (offset) as a function of other variables and values for the
#' corresponding deltas (sensitivity parameters). See details.
#' @param umx An auxiliary data matrix containing variables that do
#' not appear in the identifiable part of the imputation procedure
#' but that have been specified via \code{ums} as being predictors
#' in the unidentifiable part of the imputation model. See details.
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @details
#' This function imputes data that are thought to be Missing Not at
#' Random (MNAR) by the NARFCS method. The NARFCS procedure
#' (Tompsett et al, 2018) generalises the so-called
#' \eqn{\delta}-adjustment sensitivity analysis method of Van Buuren,
#' Boshuizen & Knook (1999) to the case with multiple incomplete
#' variables within the FCS framework. In practical terms, the
#' NARFCS procedure shifts the imputations drawn at each
#' iteration of \code{mice} by a user-specified quantity that can
#' vary across subjects, to reflect systematic departures of the
#' missing data from the data distribution imputed under MAR.
#'
#' Specification of the NARFCS model is done by the \code{blots}
#' argument of \code{mice()}. The \code{blots} parameter is a named
#' list. For each variable to be imputed by
#' \code{mice.impute.mnar.norm()} or \code{mice.impute.mnar.logreg()}
#' the corresponding element in \code{blots} is a list with
#' at least one argument \code{ums} and, optionally, a second
#' argument \code{umx}.
#' For example, the high-level call might like something like
#' \code{mice(nhanes[, c(2, 4)], method = c("pmm", "mnar.norm"),
#' blots = list(chl = list(ums = "-3+2*bmi")))}.
#'
#' The \code{ums} parameter is required, and might look like this:
#' \code{"-4+1*Y"}. The \code{ums} specifcation must have the
#' following characteristics:
#' \enumerate{
#' \item{A single term corresponding to the intercept (constant) term,
#' not multiplied by any variable name, must be included in the
#' expression;}
#' \item{Each term in the expression (corresponding to the intercept
#' or a predictor variable) must be separated by either a \code{"+"}
#' or \code{"-"} sign, depending on the sign of the sensitivity
#' parameter;}
#' \item{Within each non-intercept term, the sensitivity parameter
#' value comes first and the predictor variable comes second, and these
#' must be separated by a \code{"*"} sign;}
#' \item{For categorical predictors, for example a variable \code{Z}
#' with K + 1 categories \code{("Cat0","Cat1", ...,"CatK")}, K
#' category-specific terms are needed, and those not in \code{umx}
#' (see below) must be specified by concatenating the variable name
#' with the name of the category (e.g. \code{ZCat1}) as this is how
#' they are named in the design matrix (argument \code{x}) passed
#' to the univariate imputation function. An example is
#' \code{"2+1*ZCat1-3*ZCat2"}.}
#' }
#'
#' If given, the \code{umx} specification must have the following
#' characteristics:
#' \enumerate{
#' \item{It contains only complete variables, with no missing values;}
#' \item{It is a numeric matrix. In particular, categorical variables
#' must be represented as dummy indicators with names corresponding
#' to what is used in \code{ums} to refer to the category-specific terms
#' (see above);}
#' \item{It has the same number of rows as the \code{data} argument
#' passed on to the main \code{mice} function;}
#' \item{It does not contain variables that were already predictors
#' in the identifiable part of the model for the variable under
#' imputation.}
#' }
#'
#' Limitation: The present implementation can only condition on variables
#' that appear in the identifiable part of the imputation model (\code{x}) or
#' in complete auxiliary variables passed on via the \code{umx} argument.
#' It is not possible to specify models where the offset depends on
#' incomplete auxiliary variables.
#'
#' For an MNAR alternative see also \code{\link{mice.impute.ri}}.
#'
#' @author Margarita Moreno-Betancur, Stef van Buuren, Ian R. White, 2020.
#' @references
#' Tompsett, D. M., Leacy, F., Moreno-Betancur, M., Heron, J., &
#' White, I. R. (2018). On the use of the not-at-random fully
#' conditional specification (NARFCS) procedure in practice.
#' \emph{Statistics in Medicine}, \bold{37}(15), 2338-2353.
#' \doi{10.1002/sim.7643}.
#'
#' Van Buuren, S., Boshuizen, H.C., Knook, D.L. (1999) Multiple
#' imputation of missing blood pressure covariates in survival analysis.
#' \emph{Statistics in Medicine}, \bold{18}, 681--694.
#'
#' @family univariate imputation functions
#' @keywords datagen
#' @examples
#' # 1: Example with no auxiliary data: only pass unidentifiable model specification (ums)
#'
#' # Specify argument to pass on to mnar imputation functions via "blots" argument
#' mnar.blot <- list(X = list(ums = "-4"), Y = list(ums = "2+1*ZCat1-3*ZCat2"))
#'
#' # Run NARFCS by using mnar imputation methods and passing argument via blots
#' impNARFCS <- mice(mnar_demo_data,
#' method = c("mnar.logreg", "mnar.norm", ""),
#' blots = mnar.blot, seed = 234235, print = FALSE
#' )
#'
#' # Obtain MI results: Note they coincide with those from old version at
#' # https://github.com/moreno-betancur/NARFCS
#' pool(with(impNARFCS, lm(Y ~ X + Z)))$pooled$estimate
#'
#' # 2: Example passing also auxiliary data to MNAR procedure (umx)
#' # Assumptions:
#' # - Auxiliary data are complete, no missing values
#' # - Auxiliary data are a numeric matrix
#' # - Auxiliary data have same number of rows as x
#' # - Auxiliary data have no overlapping variable names with x
#'
#' # Specify argument to pass on to mnar imputation functions via "blots" argument
#' aux <- matrix(0:1, nrow = nrow(mnar_demo_data))
#' dimnames(aux) <- list(NULL, "even")
#' mnar.blot <- list(
#' X = list(ums = "-4"),
#' Y = list(ums = "2+1*ZCat1-3*ZCat2+0.5*even", umx = aux)
#' )
#'
#' # Run NARFCS by using mnar imputation methods and passing argument via blots
#' impNARFCS <- mice(mnar_demo_data,
#' method = c("mnar.logreg", "mnar.norm", ""),
#' blots = mnar.blot, seed = 234235, print = FALSE
#' )
#'
#' # Obtain MI results: As expected they differ (slightly) from those
#' # from old version at https://github.com/moreno-betancur/NARFCS
#' pool(with(impNARFCS, lm(Y ~ X + Z)))$pooled$estimate
#' @export
mice.impute.mnar.norm <- function(y, ry, x, wy = NULL,
ums = NULL, umx = NULL, ...) {
## Undentifiable part:
u <- parse.ums(x, ums = ums, umx = umx, ...)
## Identifiable part: exactly the same as mice.impute.norm
if (is.null(wy)) wy <- !ry
x <- cbind(1, as.matrix(x))
parm <- .norm.draw(y, ry, x, ...)
## Draw imputations
return(x[wy, , drop = FALSE] %*% parm$beta +
u$x[wy, , drop = FALSE] %*% u$delta +
rnorm(sum(wy)) * parm$sigma)
}
mice/R/mads.R 0000644 0001762 0000144 00000007336 14330031606 012444 0 ustar ligges users #' Multivariate amputed data set (\code{mads})
#'
#' The \code{mads} object contains an amputed data set. The \code{mads} object is
#' generated by the \code{ampute} function. The \code{mads} class of objects has
#' methods for the following generic functions: \code{print}, \code{summary},
#' \code{bwplot} and \code{xyplot}.
#'
#' @section Contents:
#' \describe{
#' \item{\code{call}:}{The function call.}
#' \item{\code{prop}:}{Proportion of cases with missing values. Note: even when
#' the proportion is entered as the proportion of missing cells (when
#' \code{bycases == TRUE}), this object contains the proportion of missing cases.}
#' \item{\code{patterns}:}{A data frame of size #patterns by #variables where \code{0}
#' indicates a variable has missing values and \code{1} indicates a variable remains
#' complete.}
#' \item{\code{freq}:}{A vector of length #patterns containing the relative
#' frequency with which the patterns occur. For example, if the vector is
#' \code{c(0.4, 0.4, 0.2)}, this means that of all cases with missing values,
#' 40 percent is candidate for pattern 1, 40 percent for pattern 2 and 20
#' percent for pattern 3. The vector sums to 1.}
#' \item{\code{mech}:}{A string specifying the missingness mechanism, either
#' \code{"MCAR"}, \code{"MAR"} or \code{"MNAR"}.}
#' \item{\code{weights}:}{A data frame of size #patterns by #variables. It contains
#' the weights that were used to calculate the weighted sum scores. The weights
#' may differ between patterns and between variables.}
#' \item{\code{cont}:}{Logical, whether probabilities are based on continuous logit
#' functions or on discrete odds distributions.}
#' \item{\code{type}:}{A vector of strings containing the type of missingness
#' for each pattern. Either \code{"LEFT"}, \code{"MID"}, \code{"TAIL"} or
#' \code{"RIGHT"}. The first type refers to the first pattern, the second type
#' to the second pattern, etc.}
#' \item{\code{odds}:}{A matrix where #patterns defines the #rows. Each row contains
#' the odds of being missing for the corresponding pattern. The amount of odds values
#' defines in how many quantiles the sum scores were divided. The values are
#' relative probabilities: a quantile with odds value 4 will have a probability of
#' being missing that is four times higher than a quantile with odds 1. The
#' #quantiles may differ between patterns, NA is used for cells remaining empty.}
#' \item{\code{amp}:}{A data frame containing the input data with NAs for the
#' amputed values.}
#' \item{\code{cand}:}{A vector that contains the pattern number for each case.
#' A value between 1 and #patterns is given. For example, a case with value 2 is
#' candidate for missing data pattern 2.}
#' \item{\code{scores}:}{A list containing vectors with weighted sum scores of the
#' candidates. The first vector refers to the candidates of the first pattern, the
#' second vector refers to the candidates of the second pattern, etc. The length
#' of the vectors differ because the number of candidates is different for each
#' pattern.}
#' \item{\code{data}:}{The complete data set that was entered in \code{ampute}.}
#' }
#' @note Many of the functions of the \code{mice} package do not use the S4 class
#' definitions, and instead rely on the S3 list equivalent
#' \code{oldClass(obj) <- "mads"}.
#' @author Rianne Schouten, 2016
#' @seealso \code{\link{ampute}}, Vignette titled "Multivariate Amputation using
#' Ampute".
#' @export
setClass("mads",
representation(
call = "call",
prop = "numeric",
patterns = "matrix",
freq = "numeric",
mech = "character",
weights = "matrix",
cont = "logical",
type = "character",
odds = "matrix",
amp = "data.frame",
cand = "integer",
scores = "list",
data = "data.frame"
),
contains = "list"
)
mice/R/mammalsleep.R 0000644 0001762 0000144 00000003612 14330031606 014006 0 ustar ligges users #' Mammal sleep data
#'
#' Dataset from Allison and Cicchetti (1976) of 62 mammal species on the
#' interrelationship between sleep, ecological, and constitutional variables.
#' The dataset contains missing values on five variables.
#'
#' Allison and Cicchetti (1976) investigated the interrelationship between
#' sleep, ecological, and constitutional variables. They assessed these
#' variables for 39 mammalian species. The authors concluded that slow-wave
#' sleep is negatively associated with a factor related to body size. This
#' suggests that large amounts of this sleep phase are disadvantageous in large
#' species. Also, paradoxical sleep (REM sleep) was associated with a factor
#' related to predatory danger, suggesting that large amounts of this sleep
#' phase are disadvantageous in prey species.
#'
#' @name mammalsleep
#' @aliases mammalsleep sleep
#' @docType data
#' @format \code{mammalsleep} is a data frame with 62 rows and 11 columns:
#' \describe{
#' \item{species}{Species of animal}
#' \item{bw}{Body weight (kg)}
#' \item{brw}{Brain weight (g)}
#' \item{sws}{Slow wave ("nondreaming") sleep (hrs/day)}
#' \item{ps}{Paradoxical ("dreaming") sleep (hrs/day)}
#' \item{ts}{Total sleep (hrs/day) (sum of slow wave and paradoxical sleep)}
#' \item{mls}{Maximum life span (years)}
#' \item{gt}{Gestation time (days)}
#' \item{pi}{Predation index (1-5), 1 = least likely to be preyed upon}
#' \item{sei}{Sleep exposure index (1-5), 1 = least exposed (e.g. animal sleeps in a
#' well-protected den), 5 = most exposed}
#' \item{odi}{Overall danger index (1-5) based on the above two indices and other information, 1 = least
#' danger (from other animals), 5 = most danger (from other animals)}
#' }
#' @source Allison, T., Cicchetti, D.V. (1976). Sleep in Mammals: Ecological and
#' Constitutional Correlates. Science, 194(4266), 732-734.
#' @keywords datasets
#' @examples
#' sleep <- data(mammalsleep)
NULL
mice/R/rbind.R 0000644 0001762 0000144 00000011571 14433400023 012607 0 ustar ligges users rbind.mids <- function(x, y = NULL, ...) {
call <- match.call()
if (is.mids(y)) {
return(rbind.mids.mids(x, y, call = call))
}
# Combine y and dots into data.frame
if (is.null(y)) {
y <- rbind.data.frame(...)
} else {
y <- rbind.data.frame(y, ...)
}
if (is.data.frame(y)) {
if (ncol(y) != ncol(x$data)) {
stop("datasets have different number of columns")
}
}
varnames <- colnames(x$data)
# Call is a vector, with first argument the mice statement and second argument the call to cbind.mids.
call <- c(x$call, call)
# The data in x (x$data) and y are combined together.
data <- rbind(x$data, y)
blocks <- x$blocks
# where argument: code all values as observed, including NA
wy <- matrix(FALSE, nrow = nrow(y), ncol = ncol(y))
rownames(wy) <- rownames(y)
where <- rbind(x$where, wy)
# ignore argument: include all new values
ignore <- c(x$ignore, rep(FALSE, nrow(y)))
# The number of imputations in the new midsobject is equal to that in x.
m <- x$m
# count the number of missing data in y and add them to x$nmis.
nmis <- x$nmis + colSums(is.na(y))
# The listelements method, post, predictorMatrix, visitSequence will be copied from x.
method <- x$method
post <- x$post
formulas <- x$formulas
blots <- x$blots
predictorMatrix <- x$predictorMatrix
visitSequence <- x$visitSequence
# Only x contributes imputations
imp <- x$imp
# seed, lastSeedValue, number of iterations, chainMean and chainVar is taken as in mids object x.
seed <- x$seed
lastSeedValue <- x$lastSeedValue
iteration <- x$iteration
chainMean <- x$chainMean
chainVar <- x$chainVar
loggedEvents <- x$loggedEvents
midsobj <- list(
data = data, imp = imp, m = m,
where = where, blocks = blocks,
call = call, nmis = nmis,
method = method,
predictorMatrix = predictorMatrix,
visitSequence = visitSequence,
formulas = formulas,
post = post,
blots = blots,
ignore = ignore,
seed = seed,
iteration = iteration,
lastSeedValue = lastSeedValue,
chainMean = chainMean,
chainVar = chainVar,
loggedEvents = loggedEvents,
version = packageVersion("mice"),
date = Sys.Date()
)
oldClass(midsobj) <- "mids"
midsobj
}
rbind.mids.mids <- function(x, y, call) {
if (!is.mids(y)) stop("argument `y` not a mids object")
if (ncol(y$data) != ncol(x$data)) {
stop("datasets have different number of columns")
}
if (!identical(colnames(x$data), colnames(y$data))) {
stop("datasets have different variable names")
}
if (!identical(sapply(x$data, is.factor), sapply(y$data, is.factor))) {
stop("datasets have different factor variables")
}
if (x$m != y$m) {
stop("number of imputations differ")
}
varnames <- colnames(x$data)
# Call is a vector, with first argument the mice statement and second argument the call to rbind.mids.
call <- match.call()
call <- c(x$call, call)
# The data in x (x$data) and y are combined together.
data <- rbind(x$data, y$data)
# Where argument
where <- rbind(x$where, y$where)
# The number of imputations in the new midsobject is equal to that in x.
m <- x$m
# count the number of missing data in y and add them to x$nmis.
nmis <- x$nmis + y$nmis
# The listelements method, post, predictorMatrix, visitSequence will be copied from x.
blocks <- x$blocks
method <- x$method
post <- x$post
formulas <- x$formulas
blots <- x$blots
ignore <- c(x$ignore, y$ignore)
predictorMatrix <- x$predictorMatrix
visitSequence <- x$visitSequence
# The original data of y will be binded into the multiple imputed dataset
# including the imputed values of y.
imp <- vector("list", ncol(x$data))
for (j in seq_len(ncol(x$data))) {
if (!is.null(x$imp[[j]]) || !is.null(y$imp[[j]])) {
imp[[j]] <- rbind(x$imp[[j]], y$imp[[j]])
}
}
names(imp) <- varnames
# seed, lastSeedValue, number of iterations
seed <- x$seed
lastSeedValue <- x$lastSeedValue
iteration <- x$iteration
if (x$iteration != y$iteration) {
warning("iterations differ, so no convergence diagnostics calculated",
call. = FALSE
)
chainMean <- NULL
chainVar <- NULL
} else {
w <- colSums(x$where) / colSums(where)
chainMean <- x$chainMean * w + y$chainMean * (1 - w)
chainVar <- x$chainVar * w + y$chainVar * (1 - w)
}
loggedEvents <- x$loggedEvents
midsobj <- list(
data = data, imp = imp, m = m,
where = where, blocks = blocks,
call = call, nmis = nmis,
method = method,
predictorMatrix = predictorMatrix,
visitSequence = visitSequence,
formulas = formulas,
post = post,
blots = blots,
ignore = ignore,
seed = seed,
iteration = iteration,
lastSeedValue = lastSeedValue,
chainMean = chainMean,
chainVar = chainVar,
loggedEvents = loggedEvents,
version = packageVersion("mice"),
date = Sys.Date()
)
oldClass(midsobj) <- "mids"
midsobj
}
mice/R/mice.theme.R 0000644 0001762 0000144 00000003533 14330031606 013531 0 ustar ligges users #' Set the theme for the plotting Trellis functions
#'
#' The \code{mice.theme()} function sets default choices for
#' Trellis plots that are built into \pkg{mice}.
#'
#' @aliases mice.theme
#' @param transparent A logical indicating whether alpha-transparency is
#' allowed. The default is \code{TRUE}.
#' @param alpha.fill A numerical values between 0 and 1 that indicates the
#' default alpha value for fills.
#' @return \code{mice.theme()} returns a named list that can be used as a theme in the functions in
#' \pkg{lattice}. By default, the \code{mice.theme()} function sets
#' \code{transparent <- TRUE} if the current device \code{.Device} supports
#' semi-transparent colors.
#' @author Stef van Buuren 2011
#' @export
mice.theme <- function(transparent = TRUE, alpha.fill = 0.3) {
filler <- function(transparent, alpha) {
if (transparent) {
return(c(
grDevices::hcl(240, 100, 40, alpha),
grDevices::hcl(0, 100, 40, alpha)
))
}
return(c(grDevices::hcl(240, 100, 40), grDevices::hcl(0, 100, 40)))
}
if (missing(transparent)) transparent <- supports.transparent()
if (missing(alpha.fill)) alpha.fill <- ifelse(transparent, 0.3, 0)
list(
superpose.symbol = list(
col = mdc(1:2),
fill = filler(transparent, alpha.fill),
pch = 1
),
superpose.line = list(
col = mdc(4:5),
lwd = 1
),
box.dot = list(
col = mdc(1:2)
),
box.rectangle = list(
col = mdc(4:5)
),
box.symbol = list(
col = mdc(1:2)
),
plot.symbol = list(
col = mdc(1:2),
fill = filler(transparent, alpha.fill),
pch = 1
),
plot.line = list(
col = mdc(4:5)
),
superpose.polygon = list(
col = filler(transparent, alpha.fill)
),
strip.background = list(
col = "grey95"
),
mice = list(
flag = TRUE
)
)
}
mice/R/ncc.R 0000644 0001762 0000144 00000002061 14330031606 012251 0 ustar ligges users #' Number of complete cases
#'
#' Calculates the number of complete cases.
#'
#' @param x An \code{R} object. Currently supported are methods for the
#' following classes: \code{mids}, \code{data.frame} and \code{matrix}. Also,
#' \code{x} can be a vector.
#' @return Number of elements in \code{x} with complete data.
#' @author Stef van Buuren, 2017
#' @seealso \code{\link{nic}}, \code{\link{cci}}
#' @examples
#' ncc(nhanes) # 13 complete cases
#' @export
ncc <- function(x) sum(cci(x))
#' Number of incomplete cases
#'
#' Calculates the number of incomplete cases.
#'
#' @param x An \code{R} object. Currently supported are methods for the
#' following classes: \code{mids}, \code{data.frame} and \code{matrix}. Also,
#' \code{x} can be a vector.
#' @return Number of elements in \code{x} with incomplete data.
#' @author Stef van Buuren, 2017
#' @seealso \code{\link{ncc}}, \code{\link{cci}}
#' @examples
#' nic(nhanes) # the remaining 12 rows
#' nic(nhanes[, c("bmi", "hyp")]) # number of cases with incomplete bmi and hyp
#' @export
nic <- function(x) sum(ici(x))
mice/R/is.R 0000644 0001762 0000144 00000002301 14330031606 012116 0 ustar ligges users #' Check for \code{mids} object
#'
#' @aliases is.mids
#' @param x An object
#' @return A logical indicating whether \code{x} is an object of class \code{mids}
#' @export
is.mids <- function(x) {
inherits(x, "mids")
}
#' Check for \code{mira} object
#'
#' @aliases is.mira
#' @param x An object
#' @return A logical indicating whether \code{x} is an object of class \code{mira}
#' @export
is.mira <- function(x) {
inherits(x, "mira")
}
#' Check for \code{mipo} object
#'
#' @aliases is.mipo
#' @param x An object
#' @return A logical indicating whether \code{x} is an object of class \code{mipo}
#' @export
is.mipo <- function(x) {
inherits(x, "mipo")
}
#' Check for \code{mitml.result} object
#'
#' @aliases is.mitml.result
#' @param x An object
#' @return A logical indicating whether \code{x} is an object of class \code{mitml.result}
#' @export
is.mitml.result <- function(x) {
inherits(x, "mitml.result")
}
is.passive <- function(string) {
"~" == substring(string, 1, 1)
}
#' Check for \code{mads} object
#'
#' @aliases is.mads
#' @param x An object
#' @return A logical indicating whether \code{x} is an object of class \code{mads}
#' @export
is.mads <- function(x) {
inherits(x, "mads")
}
mice/R/generics.R 0000644 0001762 0000144 00000020041 14433400023 013300 0 ustar ligges users #' Combine R objects by rows and columns
#'
#' Functions \code{cbind()} and \code{rbind()} are defined in
#' the \code{mice} package in order to
#' enable dispatch to \code{cbind.mids()} and \code{rbind.mids()}
#' when one of the arguments is a \code{data.frame}.
#'
#' The standard \code{base::cbind()} and \code{base::rbind()}
#' always dispatch to
#' \code{base::cbind.data.frame()} or \code{base::rbind.data.frame()}
#' if one of the arguments is a
#' \code{data.frame}. The versions defined in the \code{mice}
#' package intercept the user command
#' and test whether the first argument has class \code{"mids"}. If so,
#' function calls \code{cbind.mids()}, respectively \code{rbind.mids()}. In
#' all other cases, the call is forwarded to standard functions in the
#' \code{base} package.
#'
#' @inheritDotParams base::cbind
#' @details
#' The \code{cbind.mids()} function combines two \code{mids} objects
#' columnwise into a single
#' object of class \code{mids}, or combines a single \code{mids} object with
#' a \code{vector}, \code{matrix}, \code{factor} or \code{data.frame}
#' columnwise into a \code{mids} object.
#'
#' If both arguments of \code{cbind.mids()} are \code{mids}-objects, the
#' \code{data} list components should have the same number of rows. Also, the
#' number of imputations (\code{m}) should be identical.
#' If the second argument is a \code{matrix},
#' \code{factor} or \code{vector}, it is transformed into a
#' \code{data.frame}. The number of rows should match with the \code{data}
#' component of the first argument.
#'
#' The \code{cbind.mids()} function renames any duplicated variable or block names by
#' appending \code{".1"}, \code{".2"} to duplicated names.
#'
#' The \code{rbind.mids()} function combines two \code{mids} objects rowwise into a single
#' \code{mids} object, or combines a \code{mids} object with a vector, matrix,
#' factor or data frame rowwise into a \code{mids} object.
#'
#' If both arguments of \code{rbind.mids()} are \code{mids} objects,
#' then \code{rbind.mids()} requires that both have the same number of multiple
#' imputations. In addition, their \code{data} components should match.
#'
#' If the second argument of \code{rbind.mids()} is not a \code{mids} object,
#' the columns of the arguments should match. The \code{where} matrix for the
#' second argument is set to \code{FALSE}, signalling that any missing values in
#' that argument were not imputed. The \code{ignore} vector for the second argument is
#' set to \code{FALSE}. Rows inherited from the second argument will therefore
#' influence the parameter estimation of the imputation model in any future
#' iterations.
#
#' @note
#' The \code{cbind.mids()} function constructs the elements of the new \code{mids} object as follows:
#' \tabular{ll}{
#' \code{data} \tab Columnwise combination of the data in \code{x} and \code{y}\cr
#' \code{imp} \tab Combines the imputed values from \code{x} and \code{y}\cr
#' \code{m} \tab Taken from \code{x$m}\cr
#' \code{where} \tab Columnwise combination of \code{x$where} and \code{y$where}\cr
#' \code{blocks} \tab Combines \code{x$blocks} and \code{y$blocks}\cr
#' \code{call} \tab Vector, \code{call[1]} creates \code{x}, \code{call[2]}
#' is call to \code{cbind.mids()}\cr
#' \code{nmis} \tab Equals \code{c(x$nmis, y$nmis)}\cr
#' \code{method} \tab Combines \code{x$method} and \code{y$method}\cr
#' \code{predictorMatrix} \tab Combination with zeroes on the off-diagonal blocks\cr
#' \code{visitSequence} \tab Combined as \code{c(x$visitSequence, y$visitSequence)}\cr
#' \code{formulas} \tab Combined as \code{c(x$formulas, y$formulas)}\cr
#' \code{post} \tab Combined as \code{c(x$post, y$post)}\cr
#' \code{blots} \tab Combined as \code{c(x$blots, y$blots)}\cr
#' \code{ignore} \tab Taken from \code{x$ignore}\cr
#' \code{seed} \tab Taken from \code{x$seed}\cr
#' \code{iteration} \tab Taken from \code{x$iteration}\cr
#' \code{lastSeedValue} \tab Taken from \code{x$lastSeedValue}\cr
#' \code{chainMean} \tab Combined from \code{x$chainMean} and \code{y$chainMean}\cr
#' \code{chainVar} \tab Combined from \code{x$chainVar} and \code{y$chainVar}\cr
#' \code{loggedEvents} \tab Taken from \code{x$loggedEvents}\cr
#' \code{version} \tab Current package version\cr
#' \code{date} \tab Current date\cr
#' }
#'
#' The \code{rbind.mids()} function constructs the elements of the new \code{mids} object as follows:
#' \tabular{ll}{
#' \code{data} \tab Rowwise combination of the (incomplete) data in \code{x} and \code{y}\cr
#' \code{imp} \tab Equals \code{rbind(x$imp[[j]], y$imp[[j]])} if \code{y} is \code{mids} object; otherwise
#' the data of \code{y} will be copied\cr
#' \code{m} \tab Equals \code{x$m}\cr
#' \code{where} \tab Rowwise combination of \code{where} arguments\cr
#' \code{blocks} \tab Equals \code{x$blocks}\cr
#' \code{call} \tab Vector, \code{call[1]} creates \code{x}, \code{call[2]} is call to \code{rbind.mids}\cr
#' \code{nmis} \tab \code{x$nmis} + \code{y$nmis}\cr
#' \code{method} \tab Taken from \code{x$method}\cr
#' \code{predictorMatrix} \tab Taken from \code{x$predictorMatrix}\cr
#' \code{visitSequence} \tab Taken from \code{x$visitSequence}\cr
#' \code{formulas} \tab Taken from \code{x$formulas}\cr
#' \code{post} \tab Taken from \code{x$post}\cr
#' \code{blots} \tab Taken from \code{x$blots}\cr
#' \code{ignore} \tab Concatenate \code{x$ignore} and \code{y$ignore}\cr
#' \code{seed} \tab Taken from \code{x$seed}\cr
#' \code{iteration} \tab Taken from \code{x$iteration}\cr
#' \code{lastSeedValue} \tab Taken from \code{x$lastSeedValue}\cr
#' \code{chainMean} \tab Set to \code{NA}\cr
#' \code{chainVar} \tab Set to \code{NA}\cr
#' \code{loggedEvents} \tab Taken from \code{x$loggedEvents}\cr
#' \code{version} \tab Taken from \code{x$version}\cr
#' \code{date} \tab Taken from \code{x$date}
#' }
#' @return An S3 object of class \code{mids}
#' @author Karin Groothuis-Oudshoorn, Stef van Buuren
#' @seealso \code{\link[base:cbind]{cbind}}, \code{\link{ibind}},
#' \code{\link[=mids-class]{mids}}
#' @references van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#' @keywords manip
#' @examples
#' # --- cbind ---
#' # impute four variables at once (default)
#' imp <- mice(nhanes, m = 1, maxit = 1, print = FALSE)
#' imp$predictorMatrix
#'
#' # impute two by two
#' data1 <- nhanes[, c("age", "bmi")]
#' data2 <- nhanes[, c("hyp", "chl")]
#' imp1 <- mice(data1, m = 2, maxit = 1, print = FALSE)
#' imp2 <- mice(data2, m = 2, maxit = 1, print = FALSE)
#'
#' # Append two solutions
#' imp12 <- cbind(imp1, imp2)
#'
#' # This is a different imputation model
#' imp12$predictorMatrix
#'
#' # Append the other way around
#' imp21 <- cbind(imp2, imp1)
#' imp21$predictorMatrix
#'
#' # Append 'forgotten' variable chl
#' data3 <- nhanes[, 1:3]
#' imp3 <- mice(data3, maxit = 1, m = 2, print = FALSE)
#' imp4 <- cbind(imp3, chl = nhanes$chl)
#'
#' # Of course, chl was not imputed
#' head(complete(imp4))
#'
#' # Combine mids object with data frame
#' imp5 <- cbind(imp3, nhanes2)
#' head(complete(imp5))
#'
#' # --- rbind ---
#' imp1 <- mice(nhanes[1:13, ], m = 2, maxit = 1, print = FALSE)
#' imp5 <- mice(nhanes[1:13, ], m = 2, maxit = 2, print = FALSE)
#' mylist <- list(age = NA, bmi = NA, hyp = NA, chl = NA)
#'
#' nrow(complete(rbind(imp1, imp5)))
#' nrow(complete(rbind(imp1, mylist)))
#'
#' nrow(complete(rbind(imp1, data.frame(mylist))))
#' nrow(complete(rbind(imp1, complete(imp5))))
#' @export
cbind <- function(...) {
if (is.null(attr(list(...)[[1]], "class"))) {
return(base::cbind(...))
}
if ("mids" %in% attr(list(...)[[1]], "class")) {
cbind.mids(...)
} else {
base::cbind(...)
}
}
#' @rdname cbind
#' @export
rbind <- function(...) {
if (is.null(attr(list(...)[[1]], "class"))) {
return(base::rbind(...))
}
if ("mids" %in% attr(list(...)[[1]], "class")) {
rbind.mids(...)
} else {
base::rbind(...)
}
}
mice/R/getfit.R 0000644 0001762 0000144 00000003677 14330031606 013006 0 ustar ligges users #' Extract list of fitted models
#'
#' Function \code{getfit()} returns the list of objects containing the repeated analysis
#' results, or optionally, one of these fitted objects. The function looks for
#' a list element called \code{analyses}, and return this component as a list with
#' \code{mira} class. If element \code{analyses} is not found in \code{x}, then
#' it returns \code{x} as a \code{mira} object.
#'
#' No checking is done for validity of objects. The function also processes
#' objects of class \code{mitml.result} from the \code{mitml} package.
#'
#' @param x An object of class \code{mira}, typically produced by a call
#' to \code{with()}.
#' @param i An integer between 1 and \code{x$m} signalling the index of the
#' repeated analysis. The default \code{i= -1} return a list with all analyses.
#' @param simplify Should the return value be unlisted?
#' @return If \code{i = -1} an object of class \code{mira} containing
#' all analyses. If \code{i} selects one of the analyses, then it return
#' an object whose with class inherited from that element.
#' @author Stef van Buuren, 2012, 2020
#' @seealso \code{\link[=mira-class]{mira}}, \code{\link{with.mids}}
#' @keywords manip
#' @examples
#' imp <- mice(nhanes, print = FALSE, seed = 21443)
#' fit <- with(imp, lm(bmi ~ chl + hyp))
#' f1 <- getfit(fit)
#' class(f1)
#' f2 <- getfit(fit, 2)
#' class(f2)
#' @export
getfit <- function(x, i = -1L, simplify = FALSE) {
if (is.null(x$analyses)) {
ra <- x
} else {
ra <- x$analyses
}
if (i != -1L) {
return(ra[[i]])
}
if (simplify) ra <- unlist(ra)
class(ra) <- c("mira", "list")
ra
}
#' Extract estimate from \code{mipo} object
#'
#' \code{getqbar} returns a named vector of pooled estimates.
#'
#' @param x An object of class \code{mipo}
#' @export
getqbar <- function(x) {
if (!is.mipo(x)) stop("Not a mipo object")
qbar <- x$pooled$estimate
# note: not supported: component/y.values
names(qbar) <- x$pooled$term
qbar
}
mice/R/mice.impute.lasso.select.norm.R 0000644 0001762 0000144 00000006237 14330031647 017313 0 ustar ligges users #' Imputation by indirect use of lasso linear regression
#'
#' Imputes univariate missing data using Bayesian linear regression following a
#' preprocessing lasso variable selection step.
#'
#' @aliases mice.impute.lasso.select.norm lasso.select.norm
#' @inheritParams mice.impute.pmm
#' @param nfolds The number of folds for the cross-validation of the lasso penalty.
#' The default is 10.
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @details
#' The method consists of the following steps:
#' \enumerate{
#' \item For a given \code{y} variable under imputation, fit a linear regression with lasso
#' penalty using \code{y[ry]} as dependent variable and \code{x[ry, ]} as predictors.
#' Coefficients that are not shrunk to 0 define an active set of predictors
#' that will be used for imputation
#' \item Define a Bayesian linear model using \code{y[ry]} as the
#' dependent variable, the active set of \code{x[ry, ]} as predictors, and standard
#' non-informative priors
#' \item Draw parameter values for the intercept, regression weights, and error
#' variance from their posterior distribution
#' \item Draw imputations from the posterior predictive distribution
#' }
#' The user can specify a \code{predictorMatrix} in the \code{mice} call
#' to define which predictors are provided to this univariate imputation method.
#' The lasso regularization will select, among the variables indicated by
#' the user, the ones that are important for imputation at any given iteration.
#' Therefore, users may force the exclusion of a predictor from a given
#' imputation model by specifying a \code{0} entry.
#' However, a non-zero entry does not guarantee the variable will be used,
#' as this decision is ultimately made by the lasso variable selection
#' procedure.
#'
#' The method is based on the Indirect Use of Regularized Regression (IURR) proposed by
#' Zhao & Long (2016) and Deng et al (2016).
#' @author Edoardo Costantini, 2021
#' @references
#'
#' Deng, Y., Chang, C., Ido, M. S., & Long, Q. (2016). Multiple imputation for
#' general missing data patterns in the presence of high-dimensional data.
#' Scientific reports, 6(1), 1-10.
#'
#' Zhao, Y., & Long, Q. (2016). Multiple imputation in the presence of
#' high-dimensional data. Statistical Methods in Medical Research, 25(5),
#' 2021-2035.
#'
#' @family univariate imputation functions
#' @keywords datagen
#' @export
mice.impute.lasso.select.norm <- function(y, ry, x, wy = NULL, nfolds = 10, ...) {
install.on.demand("glmnet", ...)
# Body
if (is.null(wy)) wy <- !ry
x_glmnet <- cbind(1, x)
xobs <- x_glmnet[ry, , drop = FALSE]
xmis <- x[wy, ]
yobs <- y[ry]
# Train imputation model
# used later in the estiamtion require this.
cv_lasso <- glmnet::cv.glmnet(
x = xobs, y = yobs,
family = "gaussian",
nfolds = nfolds,
alpha = 1
)
# Define Active Set
glmnet_coefs <- as.matrix(coef(cv_lasso,
s = "lambda.min"
))[, 1]
AS <- which((glmnet_coefs != 0)[-1]) # Non-zero reg coefficinets
# Perform regular norm draw from Bayesian linear model
xas <- x_glmnet[, AS, drop = FALSE]
vec <- mice.impute.norm(
y = y, ry = ry, x = xas, wy = wy,
...
)
vec
}
mice/R/mice.impute.sample.R 0000644 0001762 0000144 00000001666 14330031647 015224 0 ustar ligges users #' Imputation by simple random sampling
#'
#' Imputes a random sample from the observed \code{y} data
#'
#' This function takes a simple random sample from the observed values in
#' \code{y}, and returns these as imputations.
#'
#' @inheritParams mice.impute.pmm
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @author Stef van Buuren, Karin Groothuis-Oudshoorn, 2000, 2017
#' @references van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#' @keywords datagen
#' @export
mice.impute.sample <- function(y, ry, x = NULL, wy = NULL, ...) {
if (is.null(wy)) {
wy <- !ry
}
yry <- y[ry]
if (length(yry) < 1) {
return(rnorm(sum(wy)))
}
if (length(yry) == 1) yry <- rep(yry, 2)
sample(yry, size = sum(wy), replace = TRUE)
}
mice/R/mice.impute.logreg.R 0000644 0001762 0000144 00000015045 14334522175 015223 0 ustar ligges users #' Imputation by logistic regression
#'
#' Imputes univariate missing data using logistic regression.
#'
#' @aliases mice.impute.logreg
#' @inheritParams mice.impute.pmm
#' @param ... Other named arguments.
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @author Stef van Buuren, Karin Groothuis-Oudshoorn
#' @details
#' Imputation for binary response variables by the Bayesian logistic regression
#' model (Rubin 1987, p. 169-170). The
#' Bayesian method consists of the following steps:
#' \enumerate{
#' \item Fit a logit, and find (bhat, V(bhat))
#' \item Draw BETA from N(bhat, V(bhat))
#' \item Compute predicted scores for m.d., i.e. logit-1(X BETA)
#' \item Compare the score to a random (0,1) deviate, and impute.
#' }
#' The method relies on the
#' standard \code{glm.fit} function. Warnings from \code{glm.fit} are
#' suppressed. Perfect prediction is handled by the data augmentation
#' method.
#'
#' @seealso \code{\link{mice}}, \code{\link{glm}}, \code{\link{glm.fit}}
#' @references Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#'
#' Brand, J.P.L. (1999). Development, Implementation and Evaluation of Multiple
#' Imputation Strategies for the Statistical Analysis of Incomplete Data Sets.
#' Ph.D. Thesis, TNO Prevention and Health/Erasmus University Rotterdam. ISBN
#' 90-74479-08-1.
#'
#' Venables, W.N. & Ripley, B.D. (1997). Modern applied statistics with S-Plus
#' (2nd ed). Springer, Berlin.
#'
#' White, I., Daniel, R. and Royston, P (2010). Avoiding bias due to perfect
#' prediction in multiple imputation of incomplete categorical variables.
#' Computational Statistics and Data Analysis, 54:22672275.
#' @family univariate imputation functions
#' @keywords datagen
#' @export
mice.impute.logreg <- function(y, ry, x, wy = NULL, ...) {
if (is.null(wy)) wy <- !ry
# augment data in order to evade perfect prediction
aug <- augment(y, ry, x, wy)
x <- aug$x
y <- aug$y
ry <- aug$ry
wy <- aug$wy
w <- aug$w
# fit model
x <- cbind(1, as.matrix(x))
expr <- expression(glm.fit(
x = x[ry, , drop = FALSE],
y = y[ry],
family = quasibinomial(link = logit),
weights = w[ry]
))
fit <- eval(expr)
fit.sum <- summary.glm(fit)
beta <- coef(fit)
rv <- t(chol(sym(fit.sum$cov.unscaled)))
beta.star <- beta + rv %*% rnorm(ncol(rv))
# draw imputations
p <- 1 / (1 + exp(-(x[wy, , drop = FALSE] %*% beta.star)))
vec <- (runif(nrow(p)) <= p)
vec[vec] <- 1
if (is.factor(y)) {
vec <- factor(vec, c(0, 1), levels(y))
}
vec
}
#' Imputation by logistic regression using the bootstrap
#'
#' Imputes univariate missing data using logistic regression
#' by a bootstrapped logistic regression model.
#' The bootstrap method draws a simple bootstrap sample with replacement
#' from the observed data \code{y[ry]} and \code{x[ry, ]}.
#'
#' @aliases mice.impute.logreg.boot
#' @inheritParams mice.impute.pmm
#' @param ... Other named arguments.
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @author Stef van Buuren, Karin Groothuis-Oudshoorn, 2000, 2011
#' @seealso \code{\link{mice}}, \code{\link{glm}}, \code{\link{glm.fit}}
#' @references Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-categorical.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' @family univariate imputation functions
#' @keywords datagen
#' @export
mice.impute.logreg.boot <- function(y, ry, x, wy = NULL, ...) {
if (is.null(wy)) wy <- !ry
# draw a bootstrap sample for yobs and xobs
xobs <- x[ry, , drop = FALSE]
yobs <- y[ry]
n1 <- sum(ry)
s <- sample(n1, n1, replace = TRUE)
doty <- y
doty[ry] <- yobs[s]
dotx <- x
dotx[ry, ] <- xobs[s, , drop = FALSE]
x <- dotx
y <- doty
# fit model
x <- cbind(1, as.matrix(x))
expr <- expression(glm.fit(
x = x[ry, , drop = FALSE],
y = y[ry],
family = binomial(link = logit)
))
fit <- suppressWarnings(eval(expr))
beta.star <- coef(fit)
# draw imputations
p <- 1 / (1 + exp(-(x[wy, ] %*% beta.star)))
vec <- (runif(nrow(p)) <= p)
vec[vec] <- 1
if (is.factor(y)) {
vec <- factor(vec, c(0, 1), levels(y))
}
vec
}
augment <- function(y, ry, x, wy, maxcat = 50) {
# define augmented data for stabilizing logreg and polyreg
# by the ad hoc procedure of White, Daniel & Royston, CSDA, 2010
# This function will prevent augmented data beyond the min and
# the max of the data
# Input:
# x: numeric data.frame (n rows)
# y: factor or numeric vector (lengt n)
# ry: logical vector (length n)
# Output:
# return a list with elements y, ry, x, and w with length n+2*(ncol(x))*length(levels(y))
# SvB May 2009
icod <- sort(unique(unclass(y)))
k <- length(icod)
if (k > maxcat) {
stop("Maximum number of categories (", maxcat, ") exceeded")
}
p <- ncol(x)
# skip augmentation if there are no predictors
if (p == 0) {
return(list(y = y, ry = ry, x = x, wy = wy, w = rep(1, length(y))))
}
## skip augmentation if there is only 1 missing value 12jul2012
## this need to be fixed 12jul2011
if (sum(!ry) == 1) {
return(list(y = y, ry = ry, x = x, wy = wy, w = rep(1, length(y))))
}
# calculate values to augment
mean <- apply(x, 2, mean, na.rm = TRUE)
sd <- sqrt(apply(x, 2, var, na.rm = TRUE))
minx <- apply(x, 2, min, na.rm = TRUE)
maxx <- apply(x, 2, max, na.rm = TRUE)
nr <- 2 * p * k
a <- matrix(mean, nrow = nr, ncol = p, byrow = TRUE)
b <- matrix(rep(c(rep.int(c(0.5, -0.5), k), rep.int(0, nr)), length = nr * p), nrow = nr, ncol = p, byrow = FALSE)
c <- matrix(sd, nrow = nr, ncol = p, byrow = TRUE)
d <- a + b * c
d <- pmax(matrix(minx, nrow = nr, ncol = p, byrow = TRUE), d, na.rm = TRUE)
d <- pmin(matrix(maxx, nrow = nr, ncol = p, byrow = TRUE), d, na.rm = TRUE)
e <- rep(rep(icod, each = 2), p)
dimnames(d) <- list(paste0("AUG", seq_len(nrow(d))), dimnames(x)[[2]])
xa <- rbind.data.frame(x, d)
# beware, concatenation of factors
ya <- if (is.factor(y)) as.factor(levels(y)[c(y, e)]) else c(y, e)
rya <- c(ry, rep.int(TRUE, nr))
wya <- c(wy, rep.int(FALSE, nr))
wa <- c(rep.int(1, length(y)), rep.int((p + 1) / nr, nr))
list(y = ya, ry = rya, x = xa, w = wa, wy = wya)
}
mice/R/squeeze.R 0000644 0001762 0000144 00000001725 14330031606 013175 0 ustar ligges users #' Squeeze the imputed values to be within specified boundaries.
#'
#' This function replaces any values in \code{x} that are lower than
#' \code{bounds[1]} by \code{bounds[1]}, and replaces any values higher
#' than \code{bounds[2]} by \code{bounds[2]}.
#'
#' @aliases squeeze
#' @param x A numerical vector with values
#' @param bounds A numerical vector of length 2 containing the lower and upper bounds.
#' By default, the bounds are to the minimum and maximum values in \code{x}.
#' @param r A logical vector of length \code{length(x)} that is used to select a
#' subset in \code{x} before calculating automatic bounds.
#' @return A vector of length \code{length(x)}.
#' @author Stef van Buuren, 2011.
#' @export
squeeze <- function(x, bounds = c(min(x[r]), max(x[r])),
r = rep.int(TRUE, length(x))) {
if (length(r) != length(x)) {
stop("Different length of vectors x and r")
}
x[x < bounds[1]] <- bounds[1]
x[x > bounds[2]] <- bounds[2]
x
}
mice/R/mice.impute.lda.R 0000644 0001762 0000144 00000005260 14330031647 014475 0 ustar ligges users #' Imputation by linear discriminant analysis
#'
#' Imputes univariate missing data using linear discriminant analysis
#'
#' @inheritParams mice.impute.pmm
#' @param ... Other named arguments. Not used.
#' @return Vector with imputed data, of type factor, and of length
#' \code{sum(wy)}
#' @details Imputation of categorical response variables by linear discriminant analysis.
#' This function uses the Venables/Ripley functions \code{lda()} and
#' \code{predict.lda()} to compute posterior probabilities for each incomplete
#' case, and draws the imputations from this posterior.
#'
#' This function can be called from within the Gibbs sampler by specifying
#' \code{"lda"} in the \code{method} argument of \code{mice()}. This method is usually
#' faster and uses fewer resources than calling the function, but the statistical
#' properties may not be as good (Brand, 1999).
#' \code{\link{mice.impute.polyreg}}.
#' @section Warning: The function does not incorporate the variability of the
#' discriminant weight, so it is not 'proper' in the sense of Rubin. For small
#' samples and rare categories in the \code{y}, variability of the imputed data
#' could therefore be underestimated.
#'
#' Added: SvB June 2009 Tried to include bootstrap, but disabled since
#' bootstrapping may easily lead to constant variables within groups.
#' @author Stef van Buuren, Karin Groothuis-Oudshoorn, 2000
#' @seealso \code{\link{mice}}, \code{link{mice.impute.polyreg}},
#' \code{\link[MASS]{lda}}
#' @references Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#'
#' Brand, J.P.L. (1999). Development, Implementation and Evaluation of Multiple
#' Imputation Strategies for the Statistical Analysis of Incomplete Data Sets.
#' Ph.D. Thesis, TNO Prevention and Health/Erasmus University Rotterdam. ISBN
#' 90-74479-08-1.
#'
#' Venables, W.N. & Ripley, B.D. (1997). Modern applied statistics with S-PLUS
#' (2nd ed). Springer, Berlin.
#' @family univariate imputation functions
#' @keywords datagen
#' @export
mice.impute.lda <- function(y, ry, x, wy = NULL, ...) {
install.on.demand("MASS", ...)
if (is.null(wy)) wy <- !ry
fy <- as.factor(y)
nc <- length(levels(fy))
# SvB June 2009 - take bootstrap sample of training data
# idx <- sample((1:length(y))[ry], size=sum(ry), replace=TRUE)
# x[ry,] <- x[idx,]
# y[ry] <- y[idx]
# end bootstrap
fit <- MASS::lda(x, fy, subset = ry)
post <- predict(fit, x[wy, , drop = FALSE])$posterior
un <- rep(runif(sum(wy)), each = nc)
idx <- 1 + apply(un > apply(post, 1, cumsum), 2, sum)
levels(fy)[idx]
}
mice/R/validate.arguments.R 0000644 0001762 0000144 00000000617 14330031606 015310 0 ustar ligges users validate.arguments <- function(y, ry, x, wy, allow.x.NULL = FALSE,
allow.x.NA = FALSE) {
# validate standard arguments of mice.impute functions
if (!allow.x.NULL && is.null(x)) {
stop("Cannot handle NULL value for `x`")
}
if (!allow.x.NA && anyNA(x)) {
stop("Cannot handle NA in `x`")
}
if (!is.vector(ry)) {
stop("`ry` is not a vector")
}
}
mice/R/ibind.R 0000644 0001762 0000144 00000006607 14433400023 012602 0 ustar ligges users #' Enlarge number of imputations by combining \code{mids} objects
#'
#' This function combines two \code{mids} objects \code{x} and \code{y} into a
#' single \code{mids} object, with the objective of increasing the number of
#' imputed data sets. If the number of imputations in \code{x} and \code{y} are
#' \code{m(x)} and \code{m(y)}, then the combined object will have
#' \code{m(x)+m(y)} imputations.
#'
#' The two \code{mids} objects are required to
#' have the same underlying multiple imputation model and should
#' be fitted on the same data.
#'
#' @param x A \code{mids} object.
#' @param y A \code{mids} object.
#' @return An S3 object of class \code{mids}
#' @author Karin Groothuis-Oudshoorn, Stef van Buuren
#' @seealso \code{\link[=mids-class]{mids}}
#' @keywords manip
#' @examples
#' data(nhanes)
#' imp1 <- mice(nhanes, m = 1, maxit = 2, print = FALSE)
#' imp1$m
#'
#' imp2 <- mice(nhanes, m = 3, maxit = 3, print = FALSE)
#' imp2$m
#'
#' imp12 <- ibind(imp1, imp2)
#' imp12$m
#' plot(imp12)
#' @export
ibind <- function(x, y) {
call <- match.call()
call <- c(x$call, call)
if (!is.mids(y) && !is.mids(x)) {
stop("Arguments `x` and `y` not of class `mids`")
}
if (!identical(is.na(x$data), is.na(y$data))) {
stop("Differences detected in the missing data pattern")
}
if (!identical(x$data[!is.na(x$data)], y$data[!is.na(y$data)])) {
stop("Differences detected in the observed data")
}
if (!identical(x$where, y$where)) {
stop("Differences detected between `x$where` and `y$where`")
}
if (!identical(x$blocks, y$blocks)) {
stop("Differences detected between `x$blocks` and `y$blocks`")
}
if (!identical(x$method, y$method)) {
stop("Differences detected between `x$method` and `y$method`")
}
if (!identical(x$predictorMatrix, y$predictorMatrix)) {
stop("Differences detected between `x$predictorMatrix` and `y$predictorMatrix`")
}
if (!identical(x$visitSequence, y$visitSequence)) {
stop("Differences detected between `x$visitSequence` and `y$visitSequence`")
}
if (!identical(x$post, y$post)) {
stop("Differences detected between `x$post` and `y$post`")
}
if (!identical(x$blots, y$blots)) {
stop("Differences detected between `x$blots` and `y$blots`")
}
visitSequence <- x$visitSequence
imp <- vector("list", ncol(x$data))
names(imp) <- names(x$data)
for (j in visitSequence) {
imp[[j]] <- cbind(x$imp[[j]], y$imp[[j]])
}
m <- (x$m + y$m)
iteration <- max(x$iteration, y$iteration)
chainMean <- chainVar <- initialize.chain(x$blocks, iteration, m)
for (j in seq_len(x$m)) {
chainMean[, seq_len(x$iteration), j] <- x$chainMean[, , j]
chainVar[, seq_len(x$iteration), j] <- x$chainVar[, , j]
}
for (j in seq_len(y$m)) {
chainMean[, seq_len(y$iteration), j + x$m] <- y$chainMean[, , j]
chainVar[, seq_len(y$iteration), j + x$m] <- y$chainVar[, , j]
}
midsobj <- list(
data = x$data, imp = imp, m = m,
where = x$where, blocks = x$blocks,
call = call, nmis = x$nmis,
method = x$method,
predictorMatrix = x$predictorMatrix,
visitSequence = visitSequence,
formulas = x$formulas, post = x$post,
blots = x$blots,
seed = x$seed,
iteration = iteration,
lastSeedValue = x$lastSeedValue,
chainMean = chainMean,
chainVar = chainVar,
loggedEvents = x$loggedEvents,
version = packageVersion("mice"),
date = Sys.Date()
)
oldClass(midsobj) <- "mids"
midsobj
}
mice/R/edit.setup.R 0000644 0001762 0000144 00000005330 14334522175 013606 0 ustar ligges users edit.setup <- function(data, setup,
allow.na = FALSE,
remove.constant = TRUE,
remove.collinear = TRUE,
remove_collinear = TRUE,
...) {
# legacy handling
if (!remove_collinear) remove.collinear <- FALSE
# edits the imputation model setup
# When it detec constant or collinear variables, write in loggedEvents
# and continues imputation with reduced model
pred <- setup$predictorMatrix
meth <- setup$method
vis <- setup$visitSequence
post <- setup$post
# FIXME: this function is not yet adapted to blocks
if (ncol(pred) != nrow(pred) || length(meth) != nrow(pred) ||
ncol(data) != nrow(pred)) {
return(setup)
}
varnames <- colnames(data)
# remove constant variables but leave passive variables untouched
for (j in seq_len(ncol(data))) {
if (!is.passive(meth[j])) {
d.j <- data[, j]
v <- if (is.character(d.j)) NA else var(as.numeric(d.j), na.rm = TRUE)
constant <- if (allow.na) {
if (is.na(v)) FALSE else v < 1000 * .Machine$double.eps
} else {
is.na(v) || v < 1000 * .Machine$double.eps
}
didlog <- FALSE
if (constant && any(pred[, j] != 0) && remove.constant) {
out <- varnames[j]
pred[, j] <- 0
updateLog(out = out, meth = "constant")
didlog <- TRUE
}
if (constant && meth[j] != "" && remove.constant) {
out <- varnames[j]
pred[j, ] <- 0
if (!didlog) {
updateLog(out = out, meth = "constant")
}
meth[j] <- ""
vis <- vis[vis != j]
post[j] <- ""
}
}
}
## remove collinear variables
ispredictor <- apply(pred != 0, 2, any)
if (any(ispredictor)) {
droplist <- find.collinear(data[, ispredictor, drop = FALSE], ...)
} else {
droplist <- NULL
}
if (length(droplist) > 0) {
for (k in seq_along(droplist)) {
j <- which(varnames %in% droplist[k])
didlog <- FALSE
if (any(pred[, j] != 0) && remove.collinear) {
# remove as predictor
out <- varnames[j]
pred[, j] <- 0
updateLog(out = out, meth = "collinear")
didlog <- TRUE
}
if (meth[j] != "" && remove.collinear) {
out <- varnames[j]
pred[j, ] <- 0
if (!didlog) {
updateLog(out = out, meth = "collinear")
}
meth[j] <- ""
vis <- vis[vis != j]
post[j] <- ""
}
}
}
if (all(pred == 0L)) {
stop("`mice` detected constant and/or collinear variables. No predictors were left after their removal.")
}
setup$predictorMatrix <- pred
setup$visitSequence <- vis
setup$post <- post
setup$method <- meth
setup
}
mice/R/mipo.R 0000644 0001762 0000144 00000013440 14433402232 012456 0 ustar ligges users #' \code{mipo}: Multiple imputation pooled object
#'
#' The \code{mipo} object contains the results of the pooling step.
#' The function \code{\link{pool}} generates an object of class \code{mipo}.
#'
#' @param x An object of class \code{mipo}
#' @param object An object of class \code{mipo}
#' @param mira.obj An object of class \code{mira}
#' @inheritParams broom::lm_tidiers
#' @param z Data frame with a tidied version of a coefficient matrix
#' @param conf.int Logical indicating whether to include
#' a confidence interval. The default is \code{FALSE}.
#' @param conf.level Confidence level of the interval, used only if
#' \code{conf.int = TRUE}. Number between 0 and 1.
#' @param exponentiate Flag indicating whether to exponentiate the
#' coefficient estimates and confidence intervals (typical for
#' logistic regression).
#' @param \dots Arguments passed down
#' @details An object class \code{mipo} is a \code{list} with
#' elements: \code{call}, \code{m}, \code{pooled} and \code{glanced}.
#'
#' The \code{pooled} elements is a data frame with columns:
#' \tabular{ll}{
#' \code{estimate}\tab Pooled complete data estimate\cr
#' \code{ubar} \tab Within-imputation variance of \code{estimate}\cr
#' \code{b} \tab Between-imputation variance of \code{estimate}\cr
#' \code{t} \tab Total variance, of \code{estimate}\cr
#' \code{dfcom} \tab Degrees of freedom in complete data\cr
#' \code{df} \tab Degrees of freedom of $t$-statistic\cr
#' \code{riv} \tab Relative increase in variance\cr
#' \code{lambda} \tab Proportion attributable to the missingness\cr
#' \code{fmi} \tab Fraction of missing information\cr
#' }
#' The names of the terms are stored as \code{row.names(pooled)}.
#'
#' The \code{glanced} elements is a \code{data.frame} with \code{m} rows.
#' The precise composition depends on the class of the complete-data analysis.
#' At least field \code{nobs} is expected to be present.
#'
#' The \code{process_mipo} is a helper function to process a
#' tidied mipo object, and is normally not called directly.
#' It adds a confidence interval, and optionally exponentiates, the result.
#' @seealso \code{\link{pool}},
#' \code{\link[=mids-class]{mids}}, \code{\link[=mira-class]{mira}}
#' @references van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#' @keywords classes
#' @name mipo
NULL
#' @rdname mipo
#' @export
mipo <- function(mira.obj, ...) {
if (!is.mira(mira.obj)) stop("`mira.obj` not of class `mira`")
structure(pool(mira.obj, ...), class = c("mipo"))
}
#' @return The \code{summary} method returns a data frame with summary statistics of the pooled analysis.
#' @rdname mipo
#' @export
summary.mipo <- function(object, type = c("tests", "all"),
conf.int = FALSE, conf.level = .95,
exponentiate = FALSE, ...) {
type <- match.arg(type)
m <- object$m
x <- object$pooled
std.error <- sqrt(x$t)
statistic <- x$estimate / std.error
p.value <- 2 * (pt(abs(statistic), pmax(x$df, 0.001), lower.tail = FALSE))
z <- data.frame(x,
std.error = std.error,
statistic = statistic,
p.value = p.value
)
z <- process_mipo(z, object,
conf.int = conf.int,
conf.level = conf.level,
exponentiate = exponentiate
)
parnames <- names(z)[1L:(pmatch("m", names(z)) - 1L)]
if (type == "tests") {
out <- c("m", "riv", "lambda", "fmi", "ubar", "b", "t", "dfcom")
keep <- base::setdiff(names(z), out)
z <- z[, keep]
}
class(z) <- c("mipo.summary", "data.frame")
z
}
#' @rdname mipo
#' @export
print.mipo <- function(x, ...) {
cat("Class: mipo m =", x$m, "\n")
print.data.frame(x$pooled, ...)
invisible(x)
}
#' @rdname mipo
#' @export
print.mipo.summary <- function(x, ...) {
print.data.frame(x, ...)
invisible(x)
}
#' @rdname mipo
#' @keywords internal
process_mipo <- function(z, x, conf.int = FALSE, conf.level = .95,
exponentiate = FALSE) {
if (exponentiate) {
# save transformation function for use on confidence interval
trans <- exp
} else {
trans <- identity
}
CI <- NULL
if (conf.int) {
# avoid "Waiting for profiling to be done..." message
CI <- suppressMessages(confint(x, level = conf.level))
}
z$estimate <- trans(z$estimate)
# combine and sort columns in desired order
parnames <- names(z)[1L:(pmatch("m", names(z)) - 1L)]
if (!is.null(CI)) {
z <- cbind(
z[, parnames, drop = FALSE],
z[, c("m", "estimate", "std.error", "statistic", "df", "p.value")],
trans(unrowname(CI)),
z[, c("riv", "lambda", "fmi", "ubar", "b", "t", "dfcom")]
)
} else {
z <- cbind(
z[, parnames, drop = FALSE],
z[, c("m", "estimate", "std.error", "statistic", "df", "p.value")],
z[, c("riv", "lambda", "fmi", "ubar", "b", "t", "dfcom")]
)
}
z
}
vcov.mipo <- function(object, ...) {
so <- diag(object$t)
dimnames(so) <- list(object$term, object$term)
so
}
confint.mipo <- function(object, parm, level = 0.95, ...) {
pooled <- object$pooled
cf <- getqbar(object)
df <- pooled$df
se <- sqrt(pooled$t)
pnames <- names(df) <- names(se) <- names(cf) <- row.names(pooled)
if (missing(parm)) {
parm <- pnames
} else if (is.numeric(parm)) {
parm <- pnames[parm]
}
a <- (1 - level) / 2
a <- c(a, 1 - a)
fac <- qt(a, df)
pct <- fmt.perc(a, 3)
ci <- array(NA,
dim = c(length(parm), 2L),
dimnames = list(parm, pct)
)
ci[, 1] <- cf[parm] + qt(a[1], df[parm]) * se[parm]
ci[, 2] <- cf[parm] + qt(a[2], df[parm]) * se[parm]
ci
}
unrowname <- function(x) {
rownames(x) <- NULL
x
}
fmt.perc <- function(probs, digits) {
paste(
format(100 * probs, trim = TRUE, scientific = FALSE, digits = digits),
"%"
)
}
mice/R/toenail2.R 0000644 0001762 0000144 00000004103 14330031606 013222 0 ustar ligges users #' Toenail data
#'
#' The toenail data come from a Multicenter study comparing two oral
#' treatments for toenail infection. Patients were evaluated for the
#' degree of separation of the nail. Patients were randomized into two
#' treatments and were followed over seven visits - four in the first
#' year and yearly thereafter. The patients have not been treated
#' prior to the first visit so this should be regarded as the
#' baseline.
#' @name toenail2
#' @docType data
#' @format A data frame with 1908 observations on the following 5 variables:
#' \describe{
#' \item{\code{patientID}}{a numeric vector giving the ID of patient}
#' \item{\code{outcome}}{a factor with 2 levels giving the response}
#' \item{\code{treatment}}{a factor with 2 levels giving the treatment group}
#' \item{\code{time}}{a numeric vector giving the time of the visit
#' (not exactly monthly intervals hence not round numbers)}
#' \item{\code{visit}}{an integer giving the number of the visit}
#' }
#' @source
#' De Backer, M., De Vroey, C., Lesaffre, E., Scheys, I., and De
#' Keyser, P. (1998). Twelve weeks of continuous oral therapy for
#' toenail onychomycosis caused by dermatophytes: A double-blind
#' comparative trial of terbinafine 250 mg/day versus itraconazole 200
#' mg/day. Journal of the American Academy of Dermatology, 38, 57-63.
#' @references
#' Lesaffre, E. and Spiessens, B. (2001). On the effect of the number of
#' quadrature points in a logistic random-effects model: An example.
#' Journal of the Royal Statistical Society, Series C, 50, 325-335.
#'
#' G. Fitzmaurice, N. Laird and J. Ware (2004) Applied Longitudinal Analysis,
#' Wiley and Sons, New York, USA.
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-catoutcome.html#example}{\emph{Flexible
#' Imputation of Missing Data. Second Edition.}} Chapman & Hall/CRC.
#' Boca Raton, FL.
#' @keywords datasets
#' @seealso \code{\link{toenail}}
#' @details Apart from formatting, this dataset is identical to
#' \code{toenail}. The formatting is taken identical to
#' \code{data("toenail", package = "HSAUR3")}.
NULL
mice/R/mnar_demo_data.R 0000644 0001762 0000144 00000000354 14330031606 014443 0 ustar ligges users #' MNAR demo data
#'
#' A toy example from Margarita Moreno-Betancur for checking NARFCS.
#'
#' A small dataset with just three columns.
#' @source \url{https://github.com/moreno-betancur/NARFCS/blob/master/datmis.csv}
"mnar_demo_data"
mice/R/mice.impute.polr.R 0000644 0001762 0000144 00000010620 14334522175 014712 0 ustar ligges users #' Imputation of ordered data by polytomous regression
#'
#' Imputes missing data in a categorical variable using polytomous regression
#' @aliases mice.impute.polr
#' @inheritParams mice.impute.pmm
#' @param nnet.maxit Tuning parameter for \code{nnet()}.
#' @param nnet.trace Tuning parameter for \code{nnet()}.
#' @param nnet.MaxNWts Tuning parameter for \code{nnet()}.
#' @param polr.to.loggedEvents A logical indicating whether each fallback
#' to the \code{multinom()} function should be written to \code{loggedEvents}.
#' The default is \code{FALSE}.
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @details
#' The function \code{mice.impute.polr()} imputes for ordered categorical response
#' variables by the proportional odds logistic regression (polr) model. The
#' function repeatedly applies logistic regression on the successive splits. The
#' model is also known as the cumulative link model.
#'
#' By default, ordered factors with more than two levels are imputed by
#' \code{mice.impute.polr}.
#'
#' The algorithm of \code{mice.impute.polr} uses the function \code{polr()} from
#' the \code{MASS} package.
#'
#' In order to avoid bias due to perfect prediction, the algorithm augment the
#' data according to the method of White, Daniel and Royston (2010).
#'
#' The call to \code{polr} might fail, usually because the data are very sparse.
#' In that case, \code{multinom} is tried as a fallback.
#' If the local flag \code{polr.to.loggedEvents} is set to TRUE,
#' a record is written
#' to the \code{loggedEvents} component of the \code{\link{mids}} object.
#' Use \code{mice(data, polr.to.loggedEvents = TRUE)} to set the flag.
#'
#' @note
#' In December 2019 Simon White alerted that the
#' \code{polr} could always fail silently. I can confirm this behaviour for
#' versions \code{mice 3.0.0 - mice 3.6.6}, so any method requests
#' for \code{polr} in these versions were in fact handled by \code{multinom}.
#' See \url{https://github.com/amices/mice/issues/206} for details.
#'
#' @author Stef van Buuren, Karin Groothuis-Oudshoorn, 2000-2010
#' @seealso \code{\link{mice}}, \code{\link[nnet]{multinom}},
#' \code{\link[MASS]{polr}}
#' @references
#'
#' Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}: Multivariate
#' Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
#' Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
#'
#' Brand, J.P.L. (1999) \emph{Development, implementation and evaluation of
#' multiple imputation strategies for the statistical analysis of incomplete
#' data sets.} Dissertation. Rotterdam: Erasmus University.
#'
#' White, I.R., Daniel, R. Royston, P. (2010). Avoiding bias due to perfect
#' prediction in multiple imputation of incomplete categorical variables.
#' \emph{Computational Statistics and Data Analysis}, 54, 2267-2275.
#'
#' Venables, W.N. & Ripley, B.D. (2002). \emph{Modern applied statistics with
#' S-Plus (4th ed)}. Springer, Berlin.
#' @family univariate imputation functions
#' @keywords datagen
#' @export
mice.impute.polr <- function(y, ry, x, wy = NULL, nnet.maxit = 100,
nnet.trace = FALSE, nnet.MaxNWts = 1500,
polr.to.loggedEvents = FALSE, ...) {
if (is.null(wy)) wy <- !ry
# augment data to evade issues with perfect prediction
x <- as.matrix(x)
aug <- augment(y, ry, x, wy)
x <- aug$x
y <- aug$y
ry <- aug$ry
wy <- aug$wy
w <- aug$w
xy <- cbind.data.frame(y = y, x = x)
## polr may fail on sparse data. We revert to multinom in such cases.
fit <- try(
suppressWarnings(MASS::polr(formula(xy),
data = xy[ry, , drop = FALSE],
weights = w[ry],
control = list(...)
)),
silent = TRUE
)
if (inherits(fit, "try-error")) {
if (polr.to.loggedEvents) {
updateLog(out = "polr falls back to multinom", frame = 6)
}
fit <- nnet::multinom(formula(xy),
data = xy[ry, , drop = FALSE],
weights = w[ry],
maxit = nnet.maxit, trace = nnet.trace,
MaxNWts = nnet.MaxNWts, ...
)
}
post <- predict(fit, xy[wy, , drop = FALSE], type = "probs")
if (sum(wy) == 1) {
post <- matrix(post, nrow = 1, ncol = length(post))
}
fy <- as.factor(y)
nc <- length(levels(fy))
un <- rep(runif(sum(wy)), each = nc)
if (is.vector(post)) {
post <- matrix(c(1 - post, post), ncol = 2)
}
draws <- un > apply(post, 1, cumsum)
idx <- 1 + apply(draws, 2, sum)
levels(fy)[idx]
}
mice/R/mice.impute.mnar.logreg.R 0000644 0001762 0000144 00000002112 14334522175 016146 0 ustar ligges users #' @rdname mice.impute.mnar
#' @export
mice.impute.mnar.logreg <- function(y, ry, x, wy = NULL,
ums = NULL, umx = NULL, ...) {
## Undentifiable part:
u <- parse.ums(x, ums = ums, umx = umx, ...)
if (is.null(wy)) wy <- !ry
wyold <- wy
## Identifiable part: exactly the same as mice.impute.logreg
# augment data in order to evade perfect prediction
aug <- augment(y, ry, x, wy)
x <- aug$x
y <- aug$y
ry <- aug$ry
wy <- aug$wy
w <- aug$w
# fit model
x <- cbind(1, as.matrix(x))
expr <- expression(glm.fit(
x = x[ry, , drop = FALSE],
y = y[ry],
family = quasibinomial(link = logit),
weights = w[ry]
))
fit <- eval(expr)
fit.sum <- summary.glm(fit)
beta <- coef(fit)
rv <- t(chol(sym(fit.sum$cov.unscaled)))
beta.star <- beta + rv %*% rnorm(ncol(rv))
## Draw imputations
p <- 1 / (1 + exp(-(x[wy, , drop = FALSE] %*% beta.star +
u$x[wyold, , drop = FALSE] %*% u$delta)))
vec <- (runif(nrow(p)) <= p)
vec[vec] <- 1
if (is.factor(y)) {
vec <- factor(vec, c(0, 1), levels(y))
}
vec
}
mice/R/initialize.imp.R 0000644 0001762 0000144 00000001737 14330031606 014444 0 ustar ligges users initialize.imp <- function(data, m, ignore, where, blocks, visitSequence,
method, nmis, data.init) {
imp <- vector("list", ncol(data))
names(imp) <- names(data)
r <- !is.na(data)
for (h in visitSequence) {
for (j in blocks[[h]]) {
y <- data[, j]
ry <- r[, j] & !ignore
wy <- where[, j]
imp[[j]] <- as.data.frame(matrix(NA, nrow = sum(wy), ncol = m))
dimnames(imp[[j]]) <- list(row.names(data)[wy], 1:m)
if (method[h] != "") {
for (i in seq_len(m)) {
if (nmis[j] < nrow(data) && is.null(data.init)) {
imp[[j]][, i] <- mice.impute.sample(y, ry, wy = wy)
} else if (!is.null(data.init)) {
imp[[j]][, i] <- data.init[wy, j]
} else {
if (is.factor(y)) {
imp[[j]][, i] <- sample(levels(y), nrow(data))
} else {
imp[[j]][, i] <- rnorm(nrow(data))
}
}
}
}
}
}
imp
}
mice/R/tbc.R 0000644 0001762 0000144 00000004734 14330031606 012267 0 ustar ligges users #' Terneuzen birth cohort
#'
#' Data of subset of the Terneuzen Birth Cohort data on child growth.
#'
#' This \code{tbc} data set is a random subset of persons from a much larger
#' collection of data from the Terneuzen Birth Cohort. The total cohort
#' comprises of 2604 unique persons, whereas the subset in \code{tbc} covers 306
#' persons. The \code{tbc.target} is an auxiliary data set containing two
#' outcomes at adult age. For more details, see De Kroon et al (2008, 2010,
#' 2011). The imputation methodology is explained in Chapter 9 of Van Buuren
#' (2012).
#'
#' @name tbc
#' @aliases tbc tbc.target terneuzen
#' @docType data
#' @format \code{tbs} is a data frame with 3951 rows and 11 columns:
#' \describe{
#' \item{id}{Person number}
#' \item{occ}{Occasion number}
#' \item{nocc}{Number of occasions}
#' \item{first}{Is this the first record for this person? (TRUE/FALSE)}
#' \item{typ}{Type of data (all observed)}
#' \item{age}{Age (years)}
#' \item{sex}{Sex 1=M, 2=F}
#' \item{hgt.z}{Height Z-score}
#' \item{wgt.z}{Weight Z-score}
#' \item{bmi.z}{BMI Z-score}
#' \item{ao}{Adult overweight (0=no, 1=yes)}
#' }
#'
#' \code{tbc.target} is a data frame with 2612 rows and 3 columns:
#' \describe{
#' \item{id}{Person number}
#' \item{ao}{Adult overweight (0=no, 1=yes)}
#' \item{bmi.z.jv}{BMI Z-score as young adult (18-29 years)}
#' }
#' @source De Kroon, M. L. A., Renders, C. M., Kuipers, E. C., van Wouwe, J. P.,
#' van Buuren, S., de Jonge, G. A., Hirasing, R. A. (2008). Identifying
#' metabolic syndrome without blood tests in young adults - The Terneuzen birth
#' cohort. \emph{European Journal of Public Health}, \emph{18}(6), 656-660.
#'
#' De Kroon, M. L. A., Renders, C. M., Van Wouwe, J. P., Van Buuren, S.,
#' Hirasing, R. A. (2010). The Terneuzen birth cohort: BMI changes between 2
#' and 6 years correlate strongest with adult overweight. \emph{PLoS ONE},
#' \emph{5}(2), e9155.
#'
#' De Kroon, M. L. A. (2011). \emph{The Terneuzen Birth Cohort. Detection and
#' Prevention of Overweight and Cardiometabolic Risk from Infancy Onward.}
#' Dissertation, Vrije Universiteit, Amsterdam.
#' \url{https://research.vu.nl/en/publications/the-terneuzen-birth-cohort-detection-and-prevention-of-overweight}
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-rastering.html#terneuzen-birth-cohort}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' @keywords datasets
#' @examples
#' data <- tbc
#' md.pattern(data)
NULL
mice/R/mice.impute.quadratic.R 0000644 0001762 0000144 00000012500 14436637243 015720 0 ustar ligges users #' Imputation of quadratic terms
#'
#' Imputes incomplete variable that appears as both
#' main effect and quadratic effect in the complete-data model.
#'
#' @aliases mice.impute.quadratic quadratic
#' @inheritParams mice.impute.pmm
#' @param quad.outcome The name of the outcome in the quadratic analysis as a
#' character string. For example, if the substantive model of interest is
#' \code{y ~ x + xx}, then \code{"y"} would be the \code{quad.outcome}
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @details
#' This function implements the "polynomial combination" method.
#' First, the polynomial
#' combination \eqn{Z = Y \beta_1 + Y^2 \beta_2} is formed.
#' \eqn{Z} is imputed by
#' predictive mean matching, followed by a decomposition of the imputed
#' data \eqn{Z}
#' into components \eqn{Y} and \eqn{Y^2}.
#' See Van Buuren (2012, pp. 139-141) and Vink
#' et al (2012) for more details. The method ensures that 1) the imputed data
#' for \eqn{Y} and \eqn{Y^2} are mutually consistent, and 2) that provides unbiased
#' estimates of the regression weights in a complete-data linear regression that
#' use both \eqn{Y} and \eqn{Y^2}.
#'
#' @note There are two situations to consider. If only the linear term \code{Y}
#' is present in the data, calculate the quadratic term \code{YY} after
#' imputation. If both the linear term \code{Y} and the the quadratic term
#' \code{YY} are variables in the data, then first impute \code{Y} by calling
#' \code{mice.impute.quadratic()} on \code{Y}, and then impute \code{YY} by
#' passive imputation as \code{meth["YY"] <- "~I(Y^2)"}. See example section
#' for details. Generally, we would like \code{YY} to be present in the data if
#' we need to preserve quadratic relations between \code{YY} and any third
#' variables in the multivariate incomplete data that we might wish to impute.
#' @author Mingyang Cai and Gerko Vink
#' @seealso \code{\link{mice.impute.pmm}}
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-knowledge.html#sec:quadratic}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#'
#' Vink, G., van Buuren, S. (2013). Multiple Imputation of Squared Terms.
#' \emph{Sociological Methods & Research}, 42:598-607.
#' @family univariate imputation functions
#' @keywords datagen
#' @examples
#' # Create Data
#' B1 <- .5
#' B2 <- .5
#' X <- rnorm(1000)
#' XX <- X^2
#' e <- rnorm(1000, 0, 1)
#' Y <- B1 * X + B2 * XX + e
#' dat <- data.frame(x = X, xx = XX, y = Y)
#'
#' # Impose 25 percent MCAR Missingness
#' dat[0 == rbinom(1000, 1, 1 - .25), 1:2] <- NA
#'
#' # Prepare data for imputation
#' ini <- mice(dat, maxit = 0)
#' meth <- c("quadratic", "~I(x^2)", "")
#' pred <- ini$pred
#' pred[, "xx"] <- 0
#'
#' # Impute data
#' imp <- mice(dat, meth = meth, pred = pred, quad.outcome = "y")
#'
#' # Pool results
#' pool(with(imp, lm(y ~ x + xx)))
#'
#' # Plot results
#' stripplot(imp)
#' plot(dat$x, dat$xx, col = mdc(1), xlab = "x", ylab = "xx")
#' cmp <- complete(imp)
#' points(cmp$x[is.na(dat$x)], cmp$xx[is.na(dat$x)], col = mdc(2))
#' @export
mice.impute.quadratic <- function(y, ry, x, wy = NULL, quad.outcome = NULL, ...) {
if (is.null(quad.outcome)) stop("Argument 'quad.outcome' for mice.impute.quadratic has not been specified")
if (!quad.outcome %in% colnames(x)) stop("The name specified for the outcome in 'quad.outcome' can not be found in the data")
if (is.null(wy)) {
wy <- !ry
}
x <- cbind(1, as.matrix(x))
# create the square of y
y2 <- y^2
# create z based on B1 * y + B2 * y^2
parm <- .norm.draw(x[, quad.outcome], ry, cbind(1, y, y2))
zobs <- cbind(y, y2) %*% parm$coef[-1]
# impute z
zmis <- mice.impute.pmm(zobs, ry, x[, -1])
zstar <- zobs
zstar[!ry] <- zmis
zstar <- as.vector(zstar)
# decompositions of z into roots
b1 <- parm$coef[2]
b2 <- parm$coef[3]
y.low <- -(1 / (2 * b2)) * (sqrt(4 * b2 * zstar + b1^2) + b1)
y.up <- (1 / (2 * b2)) * (sqrt(4 * b2 * zstar + b1^2) - b1)
# calculate the abscissa at the parabolic minimum/maximum
y.min <- -b1 / (2 * b2)
# data augmentation
data.augment <- data.frame(
V = c((y > y.min)[ry] * 1, 1, 1, 0, 0, 1, 1, 0, 0),
q = c(
x[ry, quad.outcome],
mean(x[ry, quad.outcome]) + sd(x[ry, quad.outcome]),
mean(x[ry, quad.outcome]) - sd(x[ry, quad.outcome]),
mean(x[ry, quad.outcome]) + sd(x[ry, quad.outcome]),
mean(x[ry, quad.outcome]) - sd(x[ry, quad.outcome]),
mean(x[ry, quad.outcome]), mean(x[ry, quad.outcome]),
mean(x[ry, quad.outcome]), mean(x[ry, quad.outcome])
),
zstar = c(
zstar[ry],
mean(zstar[ry]), mean(zstar[ry]),
mean(zstar[ry]), mean(zstar[ry]),
mean(zstar[ry]) + sd(zstar[ry]),
mean(zstar[ry]) - sd(zstar[ry]),
mean(zstar[ry]) + sd(zstar[ry]),
mean(zstar[ry]) - sd(zstar[ry])
)
)
w <- c(rep(1, nrow(data.augment) - 8), rep(3 / 8, 8))
# calculate regression parameters for
vobs <- glm(V ~ q + zstar + q * zstar,
family = quasibinomial,
data = data.augment, weights = w
)
# impute Vmis
newdata <- data.frame(q = x[wy, quad.outcome], zstar = zstar[wy])
prob <- predict(vobs,
newdata = newdata, type = "response",
na.action = na.exclude
)
idy <- rbinom(sum(wy), 1, prob = prob)
# create final imputation
ystar <- y.low[wy]
ystar[idy == 1] <- y.up[wy][idy == 1]
return(ystar)
}
mice/R/lm.R 0000644 0001762 0000144 00000007740 14330031647 012134 0 ustar ligges users #' Linear regression for \code{mids} object
#'
#' Applies \code{lm()} to multiply imputed data set
#'
#' This function is included for backward compatibility with V1.0. The function
#' is superseded by \code{\link{with.mids}}.
#'
#' @param formula a formula object, with the response on the left of a ~
#' operator, and the terms, separated by + operators, on the right. See the
#' documentation of \code{\link{lm}} and \code{\link{formula}} for details.
#' @param data An object of type 'mids', which stands for 'multiply imputed data
#' set', typically created by a call to function \code{mice()}.
#' @param \dots Additional parameters passed to \code{\link{lm}}
#' @return An objects of class \code{mira}, which stands for 'multiply imputed
#' repeated analysis'. This object contains \code{data$m} distinct
#' \code{lm.objects}, plus some descriptive information.
#' @author Stef van Buuren, Karin Groothuis-Oudshoorn, 2000
#' @seealso \code{\link{lm}}, \code{\link[=mids-class]{mids}}, \code{\link[=mira-class]{mira}}
#' @references Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#' @keywords multivariate
#' @examples
#' imp <- mice(nhanes)
#' fit <- lm.mids(bmi ~ hyp + chl, data = imp)
#' fit
#' @export
lm.mids <- function(formula, data, ...) {
.Deprecated("with",
msg = "Use with(imp, lm(yourmodel)."
)
# adapted 28/1/00 repeated complete data regression (lm) on a mids data set
call <- match.call()
if (!is.mids(data)) {
stop("The data must have class mids")
}
analyses <- lapply(seq_len(data$m), function(i) lm(formula, data = complete(data, i), ...))
# return the complete data analyses as a list of length nimp
object <- list(call = call, call1 = data$call, nmis = data$nmis, analyses = analyses)
oldClass(object) <- c("mira", "lm") ## FEH
object
}
#' Generalized linear model for \code{mids} object
#'
#' Applies \code{glm()} to a multiply imputed data set
#'
#' This function is included for backward compatibility with V1.0. The function
#' is superseded by \code{\link{with.mids}}.
#'
#' @param formula a formula expression as for other regression models, of the
#' form response ~ predictors. See the documentation of \code{\link{lm}} and
#' \code{\link{formula}} for details.
#' @param family The family of the glm model
#' @param data An object of type \code{mids}, which stands for 'multiply imputed
#' data set', typically created by function \code{mice()}.
#' @param \dots Additional parameters passed to \code{\link{glm}}.
#' @return An objects of class \code{mira}, which stands for 'multiply imputed
#' repeated analysis'. This object contains \code{data$m} distinct
#' \code{glm.objects}, plus some descriptive information.
#' @author Stef van Buuren, Karin Groothuis-Oudshoorn, 2000
#' @seealso \code{\link{with.mids}}, \code{\link{glm}}, \code{\link[=mids-class]{mids}},
#' \code{\link[=mira-class]{mira}}
#' @references Van Buuren, S., Groothuis-Oudshoorn, C.G.M. (2000)
#' \emph{Multivariate Imputation by Chained Equations: MICE V1.0 User's manual.}
#' Leiden: TNO Quality of Life.
#' @keywords multivariate
#' @examples
#'
#' imp <- mice(nhanes)
#'
#' # logistic regression on the imputed data
#' fit <- glm.mids((hyp == 2) ~ bmi + chl, data = imp, family = binomial)
#' fit
#' @export
glm.mids <- function(formula, family = gaussian, data, ...) {
.Deprecated("with",
msg = "Use with(imp, glm(yourmodel)."
)
# adapted 04/02/00 repeated complete data regression (glm) on a mids data set
call <- match.call()
if (!is.mids(data)) {
stop("The data must have class mids")
}
analyses <- lapply(
seq_len(data$m),
function(i) glm(formula, family = family, data = complete(data, i), ...)
)
# return the complete data analyses as a list of length nimp
object <- list(call = call, call1 = data$call, nmis = data$nmis, analyses = analyses)
oldClass(object) <- c("mira", "glm", "lm") ## FEH
object
}
mice/R/mice.impute.polyreg.R 0000644 0001762 0000144 00000006675 14330031647 015431 0 ustar ligges users #' Imputation of unordered data by polytomous regression
#'
#' Imputes missing data in a categorical variable using polytomous regression
#'
#' @aliases mice.impute.polyreg
#' @inheritParams mice.impute.pmm
#' @param nnet.maxit Tuning parameter for \code{nnet()}.
#' @param nnet.trace Tuning parameter for \code{nnet()}.
#' @param nnet.MaxNWts Tuning parameter for \code{nnet()}.
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @author Stef van Buuren, Karin Groothuis-Oudshoorn, 2000-2010
#' @details
#' The function \code{mice.impute.polyreg()} imputes categorical response
#' variables by the Bayesian polytomous regression model. See J.P.L. Brand
#' (1999), Chapter 4, Appendix B.
#'
#' By default, unordered factors with more than two levels are imputed by
#' \code{mice.impute.polyreg()}.
#'
#' The method consists of the following steps:
#' \enumerate{
#' \item Fit categorical response as a multinomial model
#' \item Compute predicted categories
#' \item Add appropriate noise to predictions
#' }
#'
#' The algorithm of \code{mice.impute.polyreg} uses the function
#' \code{multinom()} from the \code{nnet} package.
#'
#' In order to avoid bias due to perfect prediction, the algorithm augment the
#' data according to the method of White, Daniel and Royston (2010).
#' @seealso \code{\link{mice}}, \code{\link[nnet]{multinom}},
#' \code{\link[MASS]{polr}}
#' @references
#'
#' Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}: Multivariate
#' Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
#' Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
#'
#' Brand, J.P.L. (1999) \emph{Development, implementation and evaluation of
#' multiple imputation strategies for the statistical analysis of incomplete
#' data sets.} Dissertation. Rotterdam: Erasmus University.
#'
#' White, I.R., Daniel, R. Royston, P. (2010). Avoiding bias due to perfect
#' prediction in multiple imputation of incomplete categorical variables.
#' \emph{Computational Statistics and Data Analysis}, 54, 2267-2275.
#'
#' Venables, W.N. & Ripley, B.D. (2002). \emph{Modern applied statistics with
#' S-Plus (4th ed)}. Springer, Berlin.
#' @family univariate imputation functions
#' @keywords datagen
#' @export
mice.impute.polyreg <- function(y, ry, x, wy = NULL, nnet.maxit = 100,
nnet.trace = FALSE, nnet.MaxNWts = 1500, ...) {
if (is.null(wy)) {
wy <- !ry
}
# augment data to evade issues with perfect prediction
x <- as.matrix(x)
aug <- augment(y, ry, x, wy)
x <- aug$x
y <- aug$y
ry <- aug$ry
wy <- aug$wy
w <- aug$w
fy <- as.factor(y)
nc <- length(levels(fy))
un <- rep(runif(sum(wy)), each = nc)
xy <- cbind.data.frame(y = y, x = x)
if (ncol(x) == 0L) {
xy <- data.frame(xy, int = 1)
}
# escape with same impute if the dependent does not vary
cat.has.all.obs <- table(y[ry]) == sum(ry)
if (any(cat.has.all.obs)) {
return(rep(levels(fy)[cat.has.all.obs], sum(wy)))
}
fit <- nnet::multinom(formula(xy),
data = xy[ry, , drop = FALSE], weights = w[ry],
maxit = nnet.maxit, trace = nnet.trace, MaxNWts = nnet.MaxNWts,
...
)
post <- predict(fit, xy[wy, , drop = FALSE], type = "probs")
if (sum(wy) == 1) {
post <- matrix(post, nrow = 1, ncol = length(post))
}
if (is.vector(post)) {
post <- matrix(c(1 - post, post), ncol = 2)
}
draws <- un > apply(post, 1, cumsum)
idx <- 1 + apply(draws, 2, sum)
levels(fy)[idx]
}
mice/R/mdc.R 0000644 0001762 0000144 00000007553 14330031606 012264 0 ustar ligges users #' Graphical parameter for missing data plots
#'
#' \code{mdc} returns colors used to distinguish observed, missing and combined
#' data in plotting. \code{mice.theme} return a partial list of named objects
#' that can be used as a theme in \code{stripplot}, \code{bwplot},
#' \code{densityplot} and \code{xyplot}.
#'
#' This function eases consistent use of colors in plots. The default follows
#' the Abayomi convention, which uses blue for observed data, red for missing or
#' imputed data, and black for combined data.
#'
#' @aliases mdc
#' @param r A numerical or character vector. The numbers 1-6 request colors as
#' follows: 1=\code{cso}, 2=\code{csi}, 3=\code{csc}, 4=\code{clo}, 5=\code{cli}
#' and 6=\code{clc}. Alternatively, \code{r} may contain the strings
#'' \code{observed}', '\code{missing}', or '\code{both}', or abbreviations
#' thereof.
#' @param s A character vector containing the strings '\code{symbol}' or
#'' \code{line}', or abbreviations thereof.
#' @param transparent A logical indicating whether alpha-transparency is
#' allowed. The default is \code{TRUE}.
#' @param cso The symbol color for the observed data. The default is a
#' transparent blue.
#' @param csi The symbol color for the missing or imputed data. The default is a
#' transparent red.
#' @param csc The symbol color for the combined observed and imputed data. The
#' default is a grey color.
#' @param clo The line color for the observed data. The default is a slightly
#' darker transparent blue.
#' @param cli The line color for the missing or imputed data. The default is a
#' slightly darker transparent red.
#' @param clc The line color for the combined observed and imputed data. The
#' default is a grey color.
#' @return \code{mdc()} returns a vector containing color definitions. The length
#' of the output vector is calculate from the length of \code{r} and \code{s}.
#' Elements of the input vectors are repeated if needed.
#' @author Stef van Buuren, sept 2012.
#' @seealso \code{\link{hcl}}, \code{\link{rgb}},
#' \code{\link{xyplot.mids}}, \code{\link[lattice:xyplot]{xyplot}},
#' \code{\link[lattice:trellis.par.get]{trellis.par.set}}
#' @references Sarkar, Deepayan (2008) \emph{Lattice: Multivariate Data
#' Visualization with R}, Springer.
#' @keywords hplot
#' @examples
#' # all six colors
#' mdc(1:6)
#'
#' # lines color for observed and missing data
#' mdc(c("obs", "mis"), "lin")
#' @export
mdc <- function(r = "observed",
s = "symbol",
transparent = TRUE,
cso = grDevices::hcl(240, 100, 40, 0.7),
csi = grDevices::hcl(0, 100, 40, 0.7),
csc = "gray50",
clo = grDevices::hcl(240, 100, 40, 0.8),
cli = grDevices::hcl(0, 100, 40, 0.8),
clc = "gray50") {
# cso: blue symbol color for observed data
# csi: red symbol color for imputations
# csc: symbol color for combined data
# clo: blue line color for observed data
# cli: red line color for observed data
# clc: line color for combined data
if (missing(transparent)) {
if (!supports.transparent()) {
cso <- grDevices::hcl(240, 100, 40)
csi <- grDevices::hcl(0, 100, 40)
csc <- "black"
clo <- grDevices::hcl(240, 100, 40)
cli <- grDevices::hcl(0, 100, 40)
clc <- "black"
}
} else if (!transparent) {
cso <- grDevices::hcl(240, 100, 40)
csi <- grDevices::hcl(0, 100, 40)
csc <- "black"
clo <- grDevices::hcl(240, 100, 40)
cli <- grDevices::hcl(0, 100, 40)
clc <- "black"
}
fallback <- grDevices::palette()[1]
if (is.numeric(r)) {
idx <- floor(r)
idx[r < 1 | r > 6] <- 7
myc <- c(cso, csi, csc, clo, cli, clc, fallback)[idx]
return(myc)
}
rc <- pmatch(r, c("observed", "missing", "both"))
sc <- pmatch(s, c("symbol", "line"))
idx <- rc + (sc - 1) * 3
idx[is.na(idx)] <- 7
myc <- c(cso, csi, csc, clo, cli, clc, fallback)[idx]
myc
}
mice/R/design.R 0000644 0001762 0000144 00000000224 14433403072 012762 0 ustar ligges users obtain.design <- function(data, formula = ~.) {
mf <- model.frame(formula, data = data, na.action = na.pass)
model.matrix(formula, data = mf)
}
mice/R/parlmice.R 0000644 0001762 0000144 00000014407 14334522175 013323 0 ustar ligges users #' Wrapper function that runs MICE in parallel
#'
#' This function is included for backward compatibility. The function
#' is superseded by \code{\link{futuremice}}.
#'
#' This function relies on package \code{\link{parallel}}, which is a base
#' package for R versions 2.14.0 and later. We have chosen to use parallel function
#' \code{parLapply} to allow the use of \code{parlmice} on Mac, Linux and Windows
#' systems. For the same reason, we use the Parallel Socket Cluster (PSOCK) type by default.
#'
#' On systems other than Windows, it can be hugely beneficial to change the cluster type to
#' \code{FORK}, as it generally results in improved memory handling. When memory issues
#' arise on a Windows system, we advise to store the multiply imputed datasets,
#' clean the memory by using \code{\link{rm}} and \code{\link{gc}} and make another
#' run using the same settings.
#'
#' This wrapper function combines the output of \code{\link{parLapply}} with
#' function \code{\link{ibind}} in \code{\link{mice}}. A \code{mids} object is returned
#' and can be used for further analyses.
#'
#' Note that if a seed value is desired, the seed should be entered to this function
#' with argument \code{seed}. Seed values outside the wrapper function (in an
#' R-script or passed to \code{\link{mice}}) will not result to reproducible results.
#' We refer to the manual of \code{\link{parallel}} for an explanation on this matter.
#'
#' @aliases parlmice
#' @param data A data frame or matrix containing the incomplete data. Similar to
#' the first argument of \code{\link{mice}}.
#' @param m The number of desired imputated datasets. By default $m=5$ as with \code{mice}
#' @param seed A scalar to be used as the seed value for the mice algorithm within
#' each parallel stream. Please note that the imputations will be the same for all
#' streams and, hence, this should be used if and only if \code{n.core = 1} and
#' if it is desired to obtain the same output as under \code{mice}.
#' @param n.core A scalar indicating the number of cores that should be used.
#' @param n.imp.core A scalar indicating the number of imputations per core.
#' @param cluster.seed A scalar to be used as the seed value. It is recommended to put the
#' seed value here and not outside this function, as otherwise the parallel processes
#' will be performed with separate, random seeds.
#' @param cl.type The cluster type. Default value is \code{"PSOCK"}. Posix machines (linux, Mac)
#' generally benefit from much faster cluster computation if \code{type} is set to \code{type = "FORK"}.
#' @param ... Named arguments that are passed down to function \code{\link{mice}} or
#' \code{\link{makeCluster}}.
#'
#' @return A mids object as defined by \code{\link{mids-class}}
#'
#' @author Gerko Vink, Rianne Schouten
#' @seealso \code{\link{parallel}}, \code{\link{parLapply}}, \code{\link{makeCluster}},
#' \code{\link{mice}}, \code{\link{mids-class}}
#' @references
#' Schouten, R. and Vink, G. (2017). parlmice: faster, paraleller, micer.
#' \url{https://www.gerkovink.com/parlMICE/Vignette_parlMICE.html}
#'
#' #'Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/parallel-computation.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#'
#' @examples
#' # 150 imputations in dataset nhanes, performed by 3 cores
#' \dontrun{
#' imp1 <- parlmice(data = nhanes, n.core = 3, n.imp.core = 50)
#' # Making use of arguments in mice.
#' imp2 <- parlmice(data = nhanes, method = "norm.nob", m = 100)
#' imp2$method
#' fit <- with(imp2, lm(bmi ~ hyp))
#' pool(fit)
#' }
#'
#' @export
parlmice <- function(data, m = 5, seed = NA, cluster.seed = NA, n.core = NULL,
n.imp.core = NULL, cl.type = "PSOCK", ...) {
.Deprecated("futuremice")
# check form of data and m
data <- check.dataform(data)
m <- check.m(m)
# check if data complete
if (sum(is.na(data)) == 0) {
stop("Data has no missing values")
}
# check if arguments match CPU specifications
if (!is.null(n.core)) {
if (n.core > parallel::detectCores()) {
stop("Number of cores specified is greater than the number of logical cores in your CPU")
}
}
# determine course of action when not all arguments specified
if (!is.null(n.core) & is.null(n.imp.core)) {
n.imp.core <- m
warning(paste("Number of imputations per core not specified: n.imp.core = m =", m, "has been used"))
}
if (is.null(n.core) & !is.null(n.imp.core)) {
n.core <- parallel::detectCores() - 1
warning(paste("Number of cores not specified. Based on your machine a value of n.core =", parallel::detectCores() - 1, "is chosen"))
}
if (is.null(n.core) & is.null(n.imp.core)) {
specs <- match.cluster(n.core = parallel::detectCores() - 1, m = m)
n.core <- specs$cores
n.imp.core <- specs$imps
}
if (!is.na(seed)) {
if (n.core > 1) {
warning("Be careful; the specified seed is equal for all imputations. Please consider specifying cluster.seed instead.")
}
}
# create arguments to export to cluster
args <- match.call(mice, expand.dots = TRUE)
args[[1]] <- NULL
args$m <- n.imp.core
# make computing cluster
cl <- parallel::makeCluster(n.core, type = cl.type)
parallel::clusterExport(cl,
varlist = c(
"data", "m", "seed", "cluster.seed",
"n.core", "n.imp.core", "cl.type",
ls(parent.frame())
),
envir = environment()
)
parallel::clusterExport(cl,
varlist = "do.call"
)
parallel::clusterEvalQ(cl, library(mice))
if (!is.na(cluster.seed)) {
parallel::clusterSetRNGStream(cl, cluster.seed)
}
# generate imputations
imps <- parallel::parLapply(cl = cl, X = 1:n.core, function(x) do.call(mice, as.list(args), envir = environment()))
parallel::stopCluster(cl)
# postprocess clustered imputation into a mids object
imp <- imps[[1]]
if (length(imps) > 1) {
for (i in 2:length(imps)) {
imp <- ibind(imp, imps[[i]])
}
}
# let imputation matrix correspond to grand m
for (i in 1:length(imp$imp)) {
colnames(imp$imp[[i]]) <- 1:imp$m
}
imp
}
match.cluster <- function(n.core, m) {
cores <- 1:n.core
imps <- 1:m
out <- data.frame(
results = as.vector(cores %*% t(imps)),
cores = cores,
imps = rep(imps, each = n.core)
)
which <- out[out[, "results"] == m, ]
which[order(which$cores, decreasing = T), ][1, 2:3]
}
mice/R/mice.impute.ri.R 0000755 0001762 0000144 00000004361 14330031606 014346 0 ustar ligges users #' Imputation by the random indicator method for nonignorable data
#'
#' Imputes nonignorable missing data by the random indicator method.
#'
#' @aliases mice.impute.ri ri
#' @inheritParams mice.impute.pmm
#' @param ri.maxit Number of inner iterations
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @author Shahab Jolani (University of Utrecht)
#' @details
#' The random indicator method estimates an offset between the
#' distribution of the observed and missing data using an algorithm
#' that iterates over the response and imputation models.
#'
#' This routine assumes that the response model and imputation model
#' have same predictors.
#'
#' For an MNAR alternative see also \code{\link{mice.impute.mnar.logreg}}.
#' @references Jolani, S. (2012).
#' \emph{Dual Imputation Strategies for Analyzing Incomplete Data}.
#' Dissertation. University of Utrecht, Dec 7 2012.
#' @family univariate imputation functions
#' @keywords datagen
#' @export
mice.impute.ri <- function(y, ry, x, wy = NULL, ri.maxit = 10, ...) {
if (is.null(wy)) {
wy <- !ry
}
x <- cbind(1, as.matrix(x))
xy <- x
xr <- xy
y.dot <- y
y.dot[wy] <- mice.impute.sample(y, ry, wy = wy)
for (k in seq_len(ri.maxit)) {
r.dot <- .r.draw(y.dot, ry, xr, ...)
y.dot <- .y.draw(y, ry, r.dot, xy, wy, ...)
}
y.dot[wy]
}
# generting a realization of the response indicator r
.r.draw <- function(ydot, ry, xr, ...) {
n <- length(ry)
xr <- cbind(xr, ydot)
expr <- expression(glm.fit(xr, ry, family = binomial(link = logit)))
fit <- suppressWarnings(eval(expr))
fit.sum <- summary.glm(fit)
psi <- coef(fit)
rv <- t(chol(sym(fit.sum$cov.unscaled)))
psi.star <- psi + rv %*% rnorm(ncol(rv))
p <- 1 / (1 + exp(-(xr %*% psi.star)))
vec <- (runif(nrow(p)) <= p)
vec[vec] <- 1
vec[seq_len(n)]
}
# Imputation of y given rdot
.y.draw <- function(y, ry, rdot, xy, wy, ...) {
parm <- .norm.draw(y, ry, cbind(xy, rdot), ...)
if (all(rdot[ry] == 1) || all(rdot[ry] == 0)) parm$coef[length(parm$coef)] <- 0
ydot <- y
rydot <- as.logical(rdot)
ydot[wy] <- xy[wy, , drop = FALSE] %*% parm$beta[-length(parm$coef), ] +
rnorm(sum(wy)) * parm$sigma
ydot[wy & !rydot] <- ydot[wy & !rydot] - parm$coef[length(parm$coef)]
ydot
}
mice/R/xyplot.R 0000644 0001762 0000144 00000023724 14330031647 013063 0 ustar ligges users #' Scatterplot of observed and imputed data
#'
#' Plotting methods for imputed data using \pkg{lattice}.
#' \code{xyplot()} produces a conditional scatterplots. The function
#' automatically separates the observed (blue) and imputed (red) data. The
#' function extends the usual features of \pkg{lattice}.
#'
#' The argument \code{na.groups} may be used to specify (combinations of)
#' missingness in any of the variables. The argument \code{groups} can be used
#' to specify groups based on the variable values themselves. Only one of both
#' may be active at the same time. When both are specified, \code{na.groups}
#' takes precedence over \code{groups}.
#'
#' Use the \code{subset} and \code{na.groups} together to plots parts of the
#' data. For example, select the first imputed data set by by
#' \code{subset=.imp==1}.
#'
#' Graphical parameters like \code{col}, \code{pch} and \code{cex} can be
#' specified in the arguments list to alter the plotting symbols. If
#' \code{length(col)==2}, the color specification to define the observed and
#' missing groups. \code{col[1]} is the color of the 'observed' data,
#' \code{col[2]} is the color of the missing or imputed data. A convenient color
#' choice is \code{col=mdc(1:2)}, a transparent blue color for the observed
#' data, and a transparent red color for the imputed data. A good choice is
#' \code{col=mdc(1:2), pch=20, cex=1.5}. These choices can be set for the
#' duration of the session by running \code{mice.theme()}.
#'
#' @aliases xyplot
#' @param x A \code{mids} object, typically created by \code{mice()} or
#' \code{mice.mids()}.
#' @param data Formula that selects the data to be plotted. This argument
#' follows the \pkg{lattice} rules for \emph{formulas}, describing the primary
#' variables (used for the per-panel display) and the optional conditioning
#' variables (which define the subsets plotted in different panels) to be used
#' in the plot.
#'
#' The formula is evaluated on the complete data set in the \code{long} form.
#' Legal variable names for the formula include \code{names(x$data)} plus the
#' two administrative factors \code{.imp} and \code{.id}.
#'
#' \bold{Extended formula interface:} The primary variable terms (both the LHS
#' \code{y} and RHS \code{x}) may consist of multiple terms separated by a
#' \sQuote{+} sign, e.g., \code{y1 + y2 ~ x | a * b}. This formula would be
#' taken to mean that the user wants to plot both \code{y1 ~ x | a * b} and
#' \code{y2 ~ x | a * b}, but with the \code{y1 ~ x} and \code{y2 ~ x} in
#' \emph{separate panels}. This behavior differs from standard \pkg{lattice}.
#' \emph{Only combine terms of the same type}, i.e. only factors or only
#' numerical variables. Mixing numerical and categorical data occasionally
#' produces odds labeling of vertical axis.
#'
#' @param na.groups An expression evaluating to a logical vector indicating
#' which two groups are distinguished (e.g. using different colors) in the
#' display. The environment in which this expression is evaluated in the
#' response indicator \code{is.na(x$data)}.
#'
#' The default \code{na.group = NULL} contrasts the observed and missing data
#' in the LHS \code{y} variable of the display, i.e. groups created by
#' \code{is.na(y)}. The expression \code{y} creates the groups according to
#' \code{is.na(y)}. The expression \code{y1 & y2} creates groups by
#' \code{is.na(y1) & is.na(y2)}, and \code{y1 | y2} creates groups as
#' \code{is.na(y1) | is.na(y2)}, and so on.
#' @param groups This is the usual \code{groups} arguments in \pkg{lattice}. It
#' differs from \code{na.groups} because it evaluates in the completed data
#' \code{data.frame(complete(x, "long", inc=TRUE))} (as usual), whereas
#' \code{na.groups} evaluates in the response indicator. See
#' \code{\link{xyplot}} for more details. When both \code{na.groups} and
#' \code{groups} are specified, \code{na.groups} takes precedence, and
#' \code{groups} is ignored.
#' @param theme A named list containing the graphical parameters. The default
#' function \code{mice.theme} produces a short list of default colors, line
#' width, and so on. The extensive list may be obtained from
#' \code{trellis.par.get()}. Global graphical parameters like \code{col} or
#' \code{cex} in high-level calls are still honored, so first experiment with
#' the global parameters. Many setting consists of a pair. For example,
#' \code{mice.theme} defines two symbol colors. The first is for the observed
#' data, the second for the imputed data. The theme settings only exist during
#' the call, and do not affect the trellis graphical parameters.
#' @param as.table See \code{\link[lattice:xyplot]{xyplot}}.
#' @param outer See \code{\link[lattice:xyplot]{xyplot}}.
#' @param allow.multiple See \code{\link[lattice:xyplot]{xyplot}}.
#' @param drop.unused.levels See \code{\link[lattice:xyplot]{xyplot}}.
#' @param subscripts See \code{\link[lattice:xyplot]{xyplot}}.
#' @param subset See \code{\link[lattice:xyplot]{xyplot}}.
#' @param \dots Further arguments, usually not directly processed by the
#' high-level functions documented here, but instead passed on to other
#' functions.
#' @return The high-level functions documented here, as well as other high-level
#' Lattice functions, return an object of class \code{"trellis"}. The
#' \code{\link[lattice:update.trellis]{update}} method can be used to
#' subsequently update components of the object, and the
#' \code{\link[lattice:print.trellis]{print}} method (usually called by default)
#' will plot it on an appropriate plotting device.
#' @note The first two arguments (\code{x} and \code{data}) are reversed
#' compared to the standard Trellis syntax implemented in \pkg{lattice}. This
#' reversal was necessary in order to benefit from automatic method dispatch.
#'
#' In \pkg{mice} the argument \code{x} is always a \code{mids} object, whereas
#' in \pkg{lattice} the argument \code{x} is always a formula.
#'
#' In \pkg{mice} the argument \code{data} is always a formula object, whereas in
#' \pkg{lattice} the argument \code{data} is usually a data frame.
#'
#' All other arguments have identical interpretation.
#'
#' @author Stef van Buuren
#' @seealso \code{\link{mice}}, \code{\link{stripplot}}, \code{\link{densityplot}},
#' \code{\link{bwplot}}, \code{\link{lattice}} for an overview of the
#' package, as well as \code{\link[lattice:xyplot]{xyplot}},
#' \code{\link[lattice:panel.xyplot]{panel.xyplot}},
#' \code{\link[lattice:print.trellis]{print.trellis}},
#' \code{\link[lattice:trellis.par.get]{trellis.par.set}}
#' @references Sarkar, Deepayan (2008) \emph{Lattice: Multivariate Data
#' Visualization with R}, Springer.
#'
#' van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}: Multivariate
#' Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
#' Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
#' @keywords hplot
#' @examples
#' imp <- mice(boys, maxit = 1)
#'
#' # xyplot: scatterplot by imputation number
#' # observe the erroneous outlying imputed values
#' # (caused by imputing hgt from bmi)
#' xyplot(imp, hgt ~ age | .imp, pch = c(1, 20), cex = c(1, 1.5))
#'
#' # same, but label with missingness of wgt (four cases)
#' xyplot(imp, hgt ~ age | .imp, na.group = wgt, pch = c(1, 20), cex = c(1, 1.5))
#' @export
xyplot.mids <- function(x,
data,
na.groups = NULL,
groups = NULL,
as.table = TRUE,
theme = mice.theme(),
allow.multiple = TRUE,
outer = TRUE,
drop.unused.levels = lattice::lattice.getOption("drop.unused.levels"),
...,
subscripts = TRUE,
subset = TRUE) {
call <- match.call()
if (!is.mids(x)) stop("Argument 'x' must be a 'mids' object")
if (missing(data)) stop("Missing formula")
formula <- data
## unpack data and response indicator
cd <- data.frame(complete(x, "long", include = TRUE))
r <- as.data.frame(is.na(x$data))
## evaluate na.group in response indicator
nagp <- eval(expr = substitute(na.groups), envir = r, enclos = parent.frame())
if (is.expression(nagp)) nagp <- eval(expr = nagp, envir = r, enclos = parent.frame())
## evaluate groups in imputed data
ngp <- eval(expr = substitute(groups), envir = cd, enclos = parent.frame())
if (is.expression(ngp)) ngp <- eval(expr = ngp, envir = cd, enclos = parent.frame())
groups <- ngp
## evaluate subset in imputed data
ss <- eval(expr = substitute(subset), envir = cd, enclos = parent.frame())
if (is.expression(ss)) ss <- eval(expr = ss, envir = cd, enclos = parent.frame())
subset <- ss
## evaluate further arguments before parsing
dots <- list(...)
args <- list(
allow.multiple = allow.multiple,
outer = outer,
drop.unused.levels = drop.unused.levels,
subscripts = subscripts,
as.table = as.table
)
## determine the y-variables
form <- lattice::latticeParseFormula(
model = formula, data = cd, subset = subset,
groups = groups, multiple = allow.multiple,
outer = outer, subscripts = TRUE,
drop = drop.unused.levels
)
ynames <- unlist(lapply(strsplit(form$left.name, " \\+ "), rm.whitespace))
## calculate selection vector gp
nona <- is.null(call$na.groups)
if (!is.null(call$groups) && nona) {
gp <- call$groups
} else {
if (nona) {
na.df <- r[, ynames, drop = FALSE]
gp <- unlist(lapply(na.df, rep.int, x$m + 1))
} else {
gp <- rep.int(nagp, length(ynames) * (x$m + 1))
}
}
## change axis defaults of extended formula interface
if (is.null(call$ylab)) {
args$ylab <- ""
if (length(ynames) == 1) args$ylab <- ynames
}
if (is.null(call$scales)) {
args$scales <- list()
if (length(ynames) > 1) {
args$scales <- list(
x = list(relation = "free"),
y = list(relation = "free")
)
}
}
## ready
args <- c(
x = formula, data = list(cd),
groups = list(gp),
args, dots, subset = call$subset
)
## go
tp <- do.call("xyplot", args)
update(tp, par.settings = theme)
}
mice/R/install.on.demand.R 0000644 0001762 0000144 00000000720 14330031647 015023 0 ustar ligges users install.on.demand <- function(pkg, quiet = FALSE, ...) {
# internal function that checks whether package pkg is
# in the library. If not found, it asks the user permission
# to install from CRAN.
if (requireNamespace(pkg, quietly = TRUE)) {
return()
}
if (interactive()) {
answer <- askYesNo(paste("Package", pkg, "needed. Install from CRAN?"))
if (answer) install.packages(pkg, repos = "https://cloud.r-project.org/", quiet = quiet)
}
}
mice/R/popmis.R 0000644 0001762 0000144 00000001565 14330031606 013025 0 ustar ligges users #' Hox pupil popularity data with missing popularity scores
#'
#' Hox pupil popularity data with some missing popularity scores
#'
#' The original, complete dataset was generated by Joop Hox as an example of
#' well-behaved multilevel data set. The distributed data contains missing data
#' in pupil popularity.
#'
#' @name popmis
#' @docType data
#' @format A data frame with 2000 rows and 7 columns:
#' \describe{
#' \item{pupil}{Pupil number within school}
#' \item{school}{School number}
#' \item{popular}{Pupil popularity with 848 missing entries}
#' \item{sex}{Pupil gender}
#' \item{texp}{Teacher experience (years)}
#' \item{const}{Constant intercept term}
#' \item{teachpop}{Teacher popularity} }
#' @source Hox, J. J. (2002) \emph{Multilevel analysis. Techniques and
#' applications.} Mahwah, NJ: Lawrence Erlbaum.
#' @keywords datasets
#' @examples
#'
#' popmis[1:3, ]
NULL
mice/R/mice.impute.2lonly.mean.R 0000644 0001762 0000144 00000006102 14330031606 016062 0 ustar ligges users #' Imputation of most likely value within the class
#'
#' Method \code{2lonly.mean} replicates the most likely value within
#' a class of a second-level variable. It works for numeric and
#' factor data. The function is primarily useful as a quick fixup for
#' data in which the second-level variable is inconsistent.
#'
#' @aliases 2lonly.mean
#' @inheritParams mice.impute.pmm
#' @param type Vector of length \code{ncol(x)} identifying random and class
#' variables. The class variable (only one is allowed) is coded as \code{-2}.
#' @param ... Other named arguments.
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @details
#' Observed values in \code{y} are averaged within the class, and
#' replicated to the missing \code{y} within that class.
#' This function is primarily useful for repairing incomplete data
#' that are constant within the class, but vary over classes.
#'
#' For numeric variables, \code{mice.impute.2lonly.mean()} imputes the
#' class mean of \code{y}. If \code{y} is a second-level variable, then
#' conventionally all observed \code{y} will be identical within the
#' class, and the function just provides a quick fix for any
#' missing \code{y} by filling in the class mean.
#'
#' For factor variables, \code{mice.impute.2lonly.mean()} imputes the
#' most frequently occuring category within the class.
#'
#' If there are no observed \code{y} in the class, all entries of the
#' class are set to \code{NA}. Note that this may produce problems
#' later on in \code{mice} if imputation routines are called that
#' expects predictor data to be complete. Methods designed for
#' imputing this type of second-level variables include
#' \code{\link{mice.impute.2lonly.norm}} and
#' \code{\link{mice.impute.2lonly.pmm}}.
#'
#' @references
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-level2pred.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Boca Raton, FL.: Chapman & Hall/CRC Press.
#' @author Gerko Vink, Stef van Buuren, 2019
#' @family univariate-2lonly
#' @keywords datagen
#' @export
mice.impute.2lonly.mean <- function(y, ry, x, type, wy = NULL, ...) {
if (all(ry)) {
return(numeric(0))
}
if (is.null(wy)) {
wy <- !ry
}
yobs <- y[ry]
class <- x[, type == -2]
if (length(class) == 0) {
stop("No class variable")
}
classobs <- class[ry]
classmis <- class[wy]
# deal with empty classes (will be NaN)
empty.classes <- class[!class %in% classobs]
classobs <- c(classobs, empty.classes)
yobs <- c(yobs, rep.int(NA, length(empty.classes)))
# factor: return class levels corresponding to class median
if (is.factor(y)) {
ym <- aggregate(yobs, list(classobs), median, na.rm = TRUE)
ym$x <- as.integer(ym$x)
return(apply(as.matrix(classmis), 1,
function(z, y, lev) lev[y[z == y[, 1], 2]],
y = ym, lev = levels(y), ...
))
}
# otherwise: return the class means
ym <- aggregate(yobs, list(classobs), mean, na.rm = TRUE)
z <- apply(as.matrix(classmis), 1,
function(z, y) y[z == y[, 1], 2],
y = ym, ...
)
z[is.nan(z)] <- NA
z
}
mice/R/formula.R 0000644 0001762 0000144 00000017437 14430511431 013170 0 ustar ligges users #' Creates a \code{formulas} argument
#'
#' This helper function creates a valid \code{formulas} object. The
#' \code{formulas} object is an argument to the \code{mice} function.
#' It is a list of formula's that specifies the target variables and
#' the predictors by means of the standard \code{~} operator.
#' @param data A \code{data.frame} with the source data
#' @param blocks An optional specification for blocks of variables in
#' the rows. The default assigns each variable in its own block.
#' @param predictorMatrix A \code{predictorMatrix} specified by the user.
#' @return A list of formula's.
#' @seealso \code{\link{make.blocks}}, \code{\link{make.predictorMatrix}}
#' @examples
#' f1 <- make.formulas(nhanes)
#' f1
#' f2 <- make.formulas(nhanes, blocks = make.blocks(nhanes, "collect"))
#' f2
#'
#' # for editing, it may be easier to work with the character vector
#' c1 <- as.character(f1)
#' c1
#'
#' # fold it back into a formula list
#' f3 <- name.formulas(lapply(c1, as.formula))
#' f3
#' @export
make.formulas <- function(data, blocks = make.blocks(data),
predictorMatrix = NULL) {
data <- check.dataform(data)
formulas <- as.list(rep("~ 0", length(blocks)))
names(formulas) <- names(blocks)
for (h in names(blocks)) {
y <- blocks[[h]]
if (is.null(predictorMatrix)) {
predictors <- colnames(data)
} else {
type <- predictorMatrix[h, ]
predictors <- names(type)[type != 0]
}
x <- setdiff(predictors, y)
if (length(x) == 0) {
x <- "0"
}
formulas[[h]] <- paste(
paste(y, collapse = "+"), "~",
paste(x, collapse = "+")
)
}
formulas <- lapply(formulas, as.formula)
formulas
}
#' Name formula list elements
#'
#' This helper function names any unnamed elements in the \code{formula}
#' list. This is a convenience function.
#' @inheritParams mice
#' @param prefix A character vector of length 1 with the prefix to
#' be using for naming any unnamed blocks with two or more variables.
#' @return Named list of formulas
#' @seealso \code{\link{mice}}
#' @details
#' This function will name any unnamed list elements specified in
#' the optional argument \code{formula}. Unnamed formula's
#' consisting with just one response variable will be named
#' after this variable. Unnamed formula's containing more
#' than one variable will be named by the \code{prefix}
#' argument, padded by an integer sequence stating at 1.
#' @examples
#' # fully conditionally specified main effects model
#' form1 <- list(
#' bmi ~ age + chl + hyp,
#' hyp ~ age + bmi + chl,
#' chl ~ age + bmi + hyp
#' )
#' form1 <- name.formulas(form1)
#' imp1 <- mice(nhanes, formulas = form1, print = FALSE, m = 1, seed = 12199)
#'
#' # same model using dot notation
#' form2 <- list(bmi ~ ., hyp ~ ., chl ~ .)
#' form2 <- name.formulas(form2)
#' imp2 <- mice(nhanes, formulas = form2, print = FALSE, m = 1, seed = 12199)
#' identical(complete(imp1), complete(imp2))
#'
#' # same model using repeated multivariate imputation
#' form3 <- name.blocks(list(all = bmi + hyp + chl ~ .))
#' imp3 <- mice(nhanes, formulas = form3, print = FALSE, m = 1, seed = 12199)
#' cmp3 <- complete(imp3)
#' identical(complete(imp1), complete(imp3))
#'
#' # same model using predictorMatrix
#' imp4 <- mice(nhanes, print = FALSE, m = 1, seed = 12199, auxiliary = TRUE)
#' identical(complete(imp1), complete(imp4))
#'
#' # different model: multivariate imputation for chl and bmi
#' form5 <- list(chl + bmi ~ ., hyp ~ bmi + age)
#' form5 <- name.formulas(form5)
#' imp5 <- mice(nhanes, formulas = form5, print = FALSE, m = 1, seed = 71712)
#' @export
name.formulas <- function(formulas, prefix = "F") {
if (!is.list(formulas)) {
stop("Argument `formulas` not a list", call. = FALSE)
}
if (!all(sapply(formulas, is.formula) | sapply(formulas, is.list))) {
stop("Not all elements in `formulas` are a formula or a list")
}
if (is.null(names(formulas))) names(formulas) <- rep("", length(formulas))
inc <- 1
for (i in seq_along(formulas)) {
if (names(formulas)[i] != "") next
# if (hasdot(formulas[[i]]) && is.null(data))
# stop("Formula with dot requires `data` argument", call. = FALSE)
y <- lhs(formulas[[i]])
if (length(y) == 1) {
names(formulas)[i] <- y
} else {
names(formulas)[i] <- paste0(prefix, inc)
inc <- inc + 1
}
}
formulas
}
check.formulas <- function(formulas, data) {
formulas <- name.formulas(formulas)
formulas <- handle.oldstyle.formulas(formulas, data)
formulas <- lapply(formulas, expand.dots, data)
# escape if formula is list of two formula's
if (any(sapply(formulas, is.list))) {
return(formulas)
}
formulas <- lapply(formulas, as.formula)
formulas
}
#' Extends formula's with predictor matrix settings
#'
#' @inheritParams mice
#' @return A list of formula's
#' @param auxiliary A logical that indicates whether the variables
#' listed in \code{predictors} should be added to the formula as main
#' effects. The default is \code{TRUE}.
#' @param include.intercept A logical that indicated whether the intercept
#' should be included in the result.
#' @keywords internal
extend.formulas <- function(formulas, data, blocks, predictorMatrix = NULL,
auxiliary = TRUE,
include.intercept = FALSE,
...) {
# Extend formulas with predictorMatrix
if (is.null(predictorMatrix)) {
return(formulas)
}
for (h in names(blocks)) {
type <- predictorMatrix[h, ]
predictors <- names(type)[type != 0]
ff <- extend.formula(
formula = formulas[[h]],
predictors = predictors,
auxiliary = auxiliary,
include.intercept = include.intercept
)
formulas[[h]] <- ff
}
formulas
}
#' Extends a formula with predictors
#'
#' @param formula A formula. If it is
#' not a formula, the formula is internally reset to \code{~0}.
#' @param predictors A character vector of variable names.
#' @param auxiliary A logical that indicates whether the variables
#' listed in \code{predictors} should be added to the formula as main
#' effects. The default is \code{TRUE}.
#' @param include.intercept A logical that indicated whether the intercept
#' should be included in the result.
#' @return A formula
#' @keywords internal
extend.formula <- function(formula = ~0,
predictors = NULL,
auxiliary = TRUE,
include.intercept = FALSE, ...) {
if (!is.formula(formula)) formula <- ~0
# handle dot in RHS
if (hasdot(formula)) {
if (length(predictors) > 1) {
fr <- as.formula(c("~", paste(predictors, collapse = "+")))
} else {
fr <- ~0
}
} else {
fr <- reformulate(c(".", predictors))
}
if (auxiliary) formula <- update(formula, fr, ...)
if (include.intercept) formula <- update(formula, ~ . + 1, ...)
formula
}
handle.oldstyle.formulas <- function(formulas, data) {
# converts old-style character vector to formula list
oldstyle <- length(formulas) == ncol(data) && is.vector(formulas) &&
is.character(formulas)
if (!oldstyle) {
return(formulas)
}
formulas[formulas != ""] <- "~ 0"
fl <- as.list(formulas)
names(fl) <- names(formulas)
fl
}
is.empty.model.data <- function(x, data) {
tt <- terms(x, data = data)
(length(attr(tt, "factors")) == 0L) & (attr(tt, "intercept") == 0L)
}
lhs <- function(x) all.vars(update(x, . ~ 1))
is.formula <- function(x) {
inherits(x, "formula")
}
hasdot <- function(f) {
if (is.recursive(f)) {
return(any(sapply(as.list(f), hasdot)))
} else {
f == as.symbol(".")
}
}
expand.dots <- function(formula, data) {
if (!is.formula(formula)) {
return(formula)
}
if (!hasdot(formula)) {
return(formula)
}
y <- lhs(formula)
x <- setdiff(colnames(data), y)
fs <- paste(paste(y, collapse = "+"), "~", paste(x, collapse = "+"))
as.formula(fs)
}
mice/R/post.R 0000644 0001762 0000144 00000001464 14330031606 012501 0 ustar ligges users #' Creates a \code{post} argument
#'
#' This helper function creates a valid \code{post} vector. The
#' \code{post} vector is an argument to the \code{mice} function that
#' specifies post-processing for a variable after each iteration of imputation.
#' @inheritParams mice
#' @return Character vector of \code{ncol(data)} element
#' @seealso \code{\link{mice}}
#' @examples
#' make.post(nhanes2)
#' @export
make.post <- function(data) {
post <- vector("character", length = ncol(data))
names(post) <- colnames(data)
post
}
check.post <- function(post, data) {
if (is.null(post)) {
return(make.post(data))
}
# check
if (length(post) != ncol(data)) {
stop("length(post) does not match ncol(data)", call. = FALSE)
}
# change
if (is.null(names(post))) names(post) <- colnames(data)
post
}
mice/R/D2.R 0000644 0001762 0000144 00000004134 14436133147 011770 0 ustar ligges users #' Compare two nested models using D2-statistic
#'
#' The D2-statistic pools test statistics from the repeated analyses.
#' The method is less powerful than the D1- and D3-statistics.
#'
#' @inheritParams D1
#' @inheritParams mitml::testModels
#' @note Warning: `D2()` assumes that the order of the variables is the
#' same in different models. See
#' \url{https://github.com/amices/mice/issues/420} for details.
#' @references
#' Li, K. H., X. L. Meng, T. E. Raghunathan, and D. B. Rubin. 1991.
#' Significance Levels from Repeated p-Values with Multiply-Imputed Data.
#' \emph{Statistica Sinica} 1 (1): 65–92.
#'
#' \url{https://stefvanbuuren.name/fimd/sec-multiparameter.html#sec:chi}
#' @examples
#' # Compare two linear models:
#' imp <- mice(nhanes2, seed = 51009, print = FALSE)
#' mi1 <- with(data = imp, expr = lm(bmi ~ age + hyp + chl))
#' mi0 <- with(data = imp, expr = lm(bmi ~ age + hyp))
#' D2(mi1, mi0)
#' \dontrun{
#' # Compare two logistic regression models
#' imp <- mice(boys, maxit = 2, print = FALSE)
#' fit1 <- with(imp, glm(gen > levels(gen)[1] ~ hgt + hc + reg, family = binomial))
#' fit0 <- with(imp, glm(gen > levels(gen)[1] ~ hgt + hc, family = binomial))
#' D2(fit1, fit0)
#' }
#' @seealso \code{\link[mitml]{testModels}}
#' @export
D2 <- function(fit1, fit0 = NULL, use = "wald") {
install.on.demand("mitml")
# fit1: a fitlist or mira-object
# fit0: named numerical vector, character vector, or list
fit1 <- as.mitml.result(fit1)
est1 <- pool(fit1)
qbar1 <- getqbar(est1)
if (is.null(fit0)) {
# test all estimates equal to zero
beta <- rep(0, length(qbar1))
names(beta) <- names(qbar1)
fit0 <- lapply(fit1, fix.coef, beta = beta)
fit0 <- as.mitml.result(fit0)
} else if (is.mira(fit0)) {
fit0 <- as.mitml.result(fit0)
}
tmr <- mitml::testModels(fit1, fit0, method = "D2", use = use)
out <- list(
call = match.call(),
result = tmr$test,
formulas = list(
`1` = formula(fit1[[1L]]),
`2` = formula(fit0[[1L]])
),
m = tmr$m,
method = "D2",
use = use,
dfcom = NA
)
class(out) <- c("mice.anova", class(fit1))
out
}
mice/R/RcppExports.R 0000644 0001762 0000144 00000006465 14436642366 014035 0 ustar ligges users # Generated by using Rcpp::compileAttributes() -> do not edit by hand
# Generator token: 10BE3573-1514-4C36-9D1C-5A225CD40393
legendre <- function(x, p) {
.Call(`_mice_legendre`, x, p)
}
matcher <- function(obs, mis, k) {
.Call(`_mice_matcher`, obs, mis, k)
}
#' Find index of matched donor units
#'
#' @param d Numeric vector with values from donor cases.
#' @param t Numeric vector with values from target cases.
#' @param k Integer, number of unique donors from which a random draw is made.
#' For \code{k = 1} the function returns the index in \code{d} corresponding
#' to the closest unit. For multiple imputation, the
#' advice is to set values in the range of \code{k = 5} to \code{k = 10}.
#' @return An integer vector with \code{length(t)} elements. Each
#' element is an index in the array \code{d}.
#' @details
#' For each element in \code{t}, the method finds the \code{k} nearest
#' neighbours in \code{d}, randomly draws one of these neighbours, and
#' returns its position in vector \code{d}.
#'
#' Fast predictive mean matching algorithm in seven steps:
#'
#' 1. Shuffle records to remove effects of ties
#'
#' 2. Obtain sorting order on shuffled data
#'
#' 3. Calculate index on input data and sort it
#'
#' 4. Pre-sample vector \code{h} with values between 1 and \code{k}
#'
#' For each of the \code{n0} elements in \code{t}:
#'
#' 5. find the two adjacent neighbours
#'
#' 6. find the \code{h_i}'th nearest neighbour
#'
#' 7. store the index of that neighbour
#'
#' Return vector of \code{n0} positions in \code{d}.
#'
#' We may use the function to perform predictive mean matching under a given
#' predictive model. To do so, specify both \code{d} and \code{t} as
#' predictions from the same model. Suppose that \code{y} contains the observed
#' outcomes of the donor cases (in the same sequence as \code{d}), then
#' \code{y[matchindex(d, t)]} returns one matched outcome for every
#' target case.
#'
#' See \url{https://github.com/amices/mice/issues/236}.
#' This function is a replacement for the \code{matcher()} function that has
#' been in default in \code{mice} since version \code{2.22} (June 2014).
#' @examples
#' set.seed(1)
#'
#' # Inputs need not be sorted
#' d <- c(-5, 5, 0, 10, 12)
#' t <- c(-6, -4, 0, 2, 4, -2, 6)
#'
#' # Index (in vector a) of closest match
#' idx <- matchindex(d, t, 1)
#' idx
#'
#' # To check: show values of closest match
#'
#' # Random draw among indices of the 5 closest predictors
#' matchindex(d, t)
#'
#' # An example
#' train <- mtcars[1:20, ]
#' test <- mtcars[21:32, ]
#' fit <- lm(mpg ~ disp + cyl, data = train)
#' d <- fitted.values(fit)
#' t <- predict(fit, newdata = test) # note: not using mpg
#' idx <- matchindex(d, t)
#'
#' # Borrow values from train to produce 12 synthetic values for mpg in test.
#' # Synthetic values are plausible values that could have been observed if
#' # they had been measured.
#' train$mpg[idx]
#'
#' # Exercise: Create a distribution of 1000 plausible values for each of the
#' # twelve mpg entries in test, and count how many times the true value
#' # (which we know here) is located within the inter-quartile range of each
#' # distribution. Is your count anywhere close to 500? Why? Why not?
#' @author Stef van Buuren, Nasinski Maciej, Alexander Robitzsch
#' @export
matchindex <- function(d, t, k = 5L) {
.Call(`_mice_matchindex`, d, t, k)
}
mice/R/futuremice.R 0000644 0001762 0000144 00000016540 14422737141 013676 0 ustar ligges users #' Wrapper function that runs MICE in parallel
#'
#' This is a wrapper function for \code{\link{mice}}, using multiple cores to
#' execute \code{\link{mice}} in parallel. As a result, the imputation
#' procedure can be sped up, which may be useful in general. By default,
#' \code{\link{futuremice}} distributes the number of imputations \code{m}
#' about equally over the cores.
#'
#' This function relies on package \code{\link[furrr]{furrr}}, which is a
#' package for R versions 3.2.0 and later. We have chosen to use furrr function
#' \code{future_map} to allow the use of \code{futuremice} on Mac, Linux and
#' Windows systems.
#'
#'
#' This wrapper function combines the output of \code{\link[furrr]{future_map}} with
#' function \code{\link{ibind}} from the \code{\link{mice}} package. A
#' \code{mids} object is returned and can be used for further analyses.
#'
#' A seed value can be specified in the global environment, which will yield
#' reproducible results. A seed value can also be specified within the
#' \code{\link{futuremice}} call, through specifying the argument
#' \code{parallelseed}. If \code{parallelseed} is not specified, a seed value is
#' drawn randomly by default, and accessible through \code{$parallelseed} in the
#' output object. Hence, results will always be reproducible, regardless of
#' whether the seed is specified in the global environment, or by setting the
#' same seed within the function (potentially by extracting the seed from the
#' \code{futuremice} output object.
#'
#' @aliases futuremice
#' @param data A data frame or matrix containing the incomplete data. Similar to
#' the first argument of \code{\link{mice}}.
#' @param m The number of desired imputated datasets. By default $m=5$ as with
#' \code{mice}
#' @param parallelseed A scalar to be used to obtain reproducible results over
#' the futures. The default \code{parallelseed = NA} will result in a seed value
#' that is randomly drawn between -999999999 and 999999999.
#' @param n.core A scalar indicating the number of cores that should be used.
#' @param seed A scalar to be used as the seed value for the mice algorithm
#' within each parallel stream. Please note that the imputations will be the
#' same for all streams and, hence, this should be used if and only if
#' \code{n.core = 1} and if it is desired to obtain the same output as under
#' \code{mice}.
#' @param use.logical A logical indicating whether logical (\code{TRUE}) or
#' physical (\code{FALSE}) CPU's on machine should be used.
#' @param future.plan A character indicating how \code{future}s are resolved.
#' The default \code{multisession} resolves futures asynchronously (in parallel)
#' in separate \code{R} sessions running in the background. See
#' \code{\link[future]{plan}} for more information on future plans.
#' @param packages A character vector with additional packages to be used in
#' \code{mice} (e.g., for using external imputation functions).
#' @param globals A character string with additional functions to be exported to
#' each future (e.g., user-written imputation functions).
#' @param ... Named arguments that are passed down to function \code{\link{mice}}.
#'
#' @return A mids object as defined by \code{\link{mids-class}}
#'
#' @author Thom Benjamin Volker, Gerko Vink
#' @seealso \code{\link[future]{future}}, \code{\link[furrr]{furrr}}, \code{\link[furrr]{future_map}},
#' \code{\link[future]{plan}}, \code{\link{mice}}, \code{\link{mids-class}}
#' @references
#' Volker, T.B. and Vink, G. (2022). futuremice: The future starts today.
#' \url{https://www.gerkovink.com/miceVignettes/futuremice/Vignette_futuremice.html}
#'
#' #'Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/parallel-computation.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#'
#' @examples
#' # 150 imputations in dataset nhanes, performed by 3 cores
#' \dontrun{
#' imp1 <- futuremice(data = nhanes, m = 150, n.core = 3)
#' # Making use of arguments in mice.
#' imp2 <- futuremice(data = nhanes, m = 100, method = "norm.nob")
#' imp2$method
#' fit <- with(imp2, lm(bmi ~ hyp))
#' pool(fit)
#' }
#'
#' @export
futuremice <- function(data, m = 5, parallelseed = NA, n.core = NULL, seed = NA,
use.logical = TRUE, future.plan = "multisession",
packages = NULL, globals = NULL, ...) {
# check if pacakages available
install.on.demand("parallelly", ...)
install.on.demand("furrr", ...)
install.on.demand("future", ...)
# check form of data and m
data <- check.dataform(data)
m <- check.m(m)
# check if data complete
if (sum(is.na(data)) == 0) {
stop("Data has no missing values")
}
# number of available cores
available <- parallelly::availableCores(logical = use.logical)
# set the number of cores
n.core <- check.cores(n.core, available, m)
if (n.core > 1) {
dist.core <- cut(1:m, n.core, labels = paste0("core", 1:n.core))
} else {
dist.core <- rep("core1", m)
}
n.imp.core <- as.vector(table(dist.core))
if (!is.na(seed)) {
if (n.core > 1) {
if (interactive()) {
msg <- "Be careful; specifying seed rather than parallelseed results in duplicate imputations.\nDo you want to continue?\n"
ask <- askYesNo(msg, prompts = getOption("askYesNo", gettext(c("Yes", "No, ignore seed", "Cancel"))))
if (isTRUE(ask)) {
seed <- seed
warning("Be careful; the imputations will be the same over the cores.")
} else if (isFALSE(ask)) {
seed <- NA
message("Parallelseed is specified for you, and is accessible in the output object under $parallelseed.")
} else if (is.na(ask)) {
stop("You stopped futuremice. To obtain unique, but reproducible imputations, specify parallelseed.")
}
} else {
warning("Be careful; the imputations will be identical over the cores. Perhaps you want to specify parallelseed, for unique, but reproducible results.")
}
}
}
if (!is.na(parallelseed)) {
set.seed(parallelseed)
} else {
if(!exists(".Random.seed")) {
set.seed(NULL)
}
parallelseed <- get(
".Random.seed",
envir = globalenv(),
mode = "integer",
inherits = FALSE
)
}
# start multisession
future::plan(future.plan,
workers = n.core
)
# begin future
imps <- furrr::future_map(
n.imp.core,
function(x) {
mice(data = data,
m = x,
printFlag = FALSE,
seed = seed,
...
)},
.options = furrr::furrr_options(
seed = TRUE,
globals = globals,
packages = packages
)
)
# end multisession
future::plan(future::sequential)
# stitch future into mids
# postprocess clustered imputation into a mids object
imp <- imps[[1]]
if (length(imps) > 1) {
for (i in 2:length(imps)) {
imp <- ibind(imp, imps[[i]])
}
}
# let imputation matrix correspond to grand m
for (i in 1:length(imp$imp)) {
colnames(imp$imp[[i]]) <- 1:imp$m
}
imp$parallelseed <- parallelseed
return(imp)
}
check.cores <- function(n.core, available, m) {
if (is.null(n.core)) {
n.core <- min(available - 1, m)
} else {
if (n.core > available | n.core > m) {
warning(paste("'n.core' exceeds the maximum number of available cores on your machine or the number of imputations, and is set to", min(available - 1, m)))
}
n.core <- min(available - 1, m, n.core)
}
n.core
} mice/R/anova.R 0000644 0001762 0000144 00000003364 14330031606 012621 0 ustar ligges users #' Compare several nested models
#'
#' @rdname anova
#' @param object Two or more objects of class \code{mira}
#' @param method Either \code{"D1"}, \code{"D2"} or \code{"D3"}
#' @param use An character indicating the test statistic
#' @param ... Other parameters passed down to \code{D1()}, \code{D2()},
#' \code{D3()} and \code{mitml::testModels}.
#' @return Object of class \code{mice.anova}
#' @export
anova.mira <- function(object, ..., method = "D1", use = "wald") {
modlist <- list(object, ...)
first <- lapply(modlist, getfit, 1L) %>% sapply(glance)
if (is.null(names(modlist))) {
names(modlist) <- names(first) <- 1L:length(modlist)
} else {
names(first) <- names(modlist)
}
# order by model complexity
dfcom <- rep(NA, ncol(first))
for (j in 1:ncol(first)) dfcom[j] <- get.dfcom(modlist[[j]])
idx <- order(dfcom, decreasing = FALSE)
modlist <- modlist[idx]
dfcom <- dfcom[idx]
names(dfcom) <- names(modlist)
# get model formulas
formulas <- lapply(modlist, getfit, 1L) %>% lapply(formula)
names(formulas) <- names(modlist)
# test successive models
nm <- length(modlist)
out <- vector("list", nm - 1L)
names(out) <- paste(names(modlist), lead(names(modlist)), sep = " ~~ ")[-nm]
for (j in seq_along(out)) {
if (method == "D2") {
args <- alist(fit1 = modlist[[j]], fit0 = modlist[[j + 1L]], use = use)
} else {
args <- alist(
fit1 = modlist[[j]], fit0 = modlist[[j + 1L]],
dfcom = as.numeric(unlist(dfcom[j]))
)
}
out[[j]] <- do.call(method, args = args)
}
obj <- list(
call = match.call(),
out = out,
formulas = formulas,
m = length(getfit(modlist[[1L]])),
method = method,
use = use
)
class(obj) <- c("mice.anova", class(first))
obj
}
mice/R/fix.coef.R 0000644 0001762 0000144 00000006427 14433402442 013225 0 ustar ligges users #' Fix coefficients and update model
#'
#' Refits a model with a specified set of coefficients.
#'
#' @param model An R model, e.g., produced by \code{lm} or \code{glm}
#' @param beta A numeric vector with \code{length(coef)} model coefficients.
#' If the vector is not named, the coefficients should be
#' given in the same order as in \code{coef(model)}. If the vector is named,
#' the procedure attempts to match on names.
#' @return An updated R model object
#' @author Stef van Buuren, 2018
#' @details
#' The function calculates the linear predictor using the new coefficients,
#' and reformulates the model using the \code{offset}
#' argument. The linear predictor is called
#' \code{offset}, and its coefficient will be \code{1} by definition.
#' The new model only fits the intercept, which should be \code{0}
#' if we set \code{beta = coef(model)}.
#' @examples
#' model0 <- lm(Volume ~ Girth + Height, data = trees)
#' formula(model0)
#' coef(model0)
#' deviance(model0)
#'
#' # refit same model
#' model1 <- fix.coef(model0)
#' formula(model1)
#' coef(model1)
#' deviance(model1)
#'
#' # change the beta's
#' model2 <- fix.coef(model0, beta = c(-50, 5, 1))
#' coef(model2)
#' deviance(model2)
#'
#' # compare predictions
#' plot(predict(model0), predict(model1))
#' abline(0, 1)
#' plot(predict(model0), predict(model2))
#' abline(0, 1)
#'
#' # compare proportion explained variance
#' cor(predict(model0), predict(model0) + residuals(model0))^2
#' cor(predict(model1), predict(model1) + residuals(model1))^2
#' cor(predict(model2), predict(model2) + residuals(model2))^2
#'
#' # extract offset from constrained model
#' summary(model2$offset)
#'
#' # it also works with factors and missing data
#' model0 <- lm(bmi ~ age + hyp + chl, data = nhanes2)
#' model1 <- fix.coef(model0)
#' model2 <- fix.coef(model0, beta = c(15, -8, -8, 2, 0.2))
#' @export
fix.coef <- function(model, beta = NULL) {
oldcoef <- clean.coef(model)
if (is.null(beta)) beta <- oldcoef
if (length(oldcoef) != length(beta)) {
stop("incorrect length of 'beta'", call. = FALSE)
}
# handle naming
if (is.null(names(oldcoef))) {
names(oldcoef) <- make.names(seq_along(oldcoef))
}
if (is.null(names(beta))) {
names(beta) <- names(oldcoef)
} else {
diff <- setdiff(names(oldcoef), names(beta))
if (length(diff) > 0) {
stop("names not found in 'beta': ", diff, call. = FALSE)
}
diff <- setdiff(names(beta), names(oldcoef))
if (length(diff) > 0) {
stop("names not found in 'coef(model)': ", diff, call. = FALSE)
}
}
beta <- beta[names(oldcoef)]
# re-calculate model for new beta's
data <- model.frame(formula = formula(model), data = model.frame(model))
mm <- model.matrix(formula(model, fixed.only = TRUE), data = data)
# Problem: offset cannot be calculated for the Cox model because that does
# not include the intercept
if (inherits(model, "coxph")) {
stop("D3 does not support the Cox model.", call. = FALSE)
}
offset <- as.vector(mm %*% beta)
uf <- . ~ 1
if (inherits(model, "merMod")) uf <- formula(model, random.only = TRUE)
upd <- update(model,
formula. = uf,
data = cbind(data, offset = offset),
offset = offset
)
upd
}
clean.coef <- function(model) {
est <- tidy(model, effects = "fixed")
coef <- est$estimate
names(coef) <- est$term
coef
}
mice/R/nelsonaalen.R 0000644 0001762 0000144 00000004256 14436637751 014041 0 ustar ligges users #' Cumulative hazard rate or Nelson-Aalen estimator
#'
#' Calculates the cumulative hazard rate (Nelson-Aalen estimator)
#'
#' This function is useful for imputing variables that depend on survival time.
#' White and Royston (2009) suggested using the cumulative hazard to the
#' survival time H0(T) rather than T or log(T) as a predictor in imputation
#' models. See section 7.1 of Van Buuren (2012) for an example.
#'
#' @aliases nelsonaalen hazard
#' @param data A data frame containing the data.
#' @param timevar The name of the time variable in \code{data}.
#' @param statusvar The name of the event variable, e.g. death in \code{data}.
#' @return A vector with \code{nrow(data)} elements containing the Nelson-Aalen
#' estimates of the cumulative hazard function.
#' @author Stef van Buuren, 2012
#' @references White, I. R., Royston, P. (2009). Imputing missing covariate
#' values for the Cox model. \emph{Statistics in Medicine}, \emph{28}(15),
#' 1982-1998.
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-toomany.html#a-further-improvement-survival-as-predictor-variable}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' @keywords misc
#' @examples
#' require(MASS)
#'
#' leuk$status <- 1 ## no censoring occurs in leuk data (MASS)
#' ch <- nelsonaalen(leuk, time, status)
#' plot(x = leuk$time, y = ch, ylab = "Cumulative hazard", xlab = "Time")
#'
#' ### See example on http://www.engineeredsoftware.com/lmar/pe_cum_hazard_function.htm
#' time <- c(43, 67, 92, 94, 149, rep(149, 7))
#' status <- c(rep(1, 5), rep(0, 7))
#' eng <- data.frame(time, status)
#' ch <- nelsonaalen(eng, time, status)
#' plot(x = time, y = ch, ylab = "Cumulative hazard", xlab = "Time")
#' @export
nelsonaalen <- function(data, timevar, statusvar) {
if (!is.data.frame(data)) {
stop("Data must be a data frame")
}
timevar <- as.character(substitute(timevar))
statusvar <- as.character(substitute(statusvar))
time <- data[, timevar, drop = TRUE]
status <- data[, statusvar, drop = TRUE]
hazard <- survival::basehaz(survival::coxph(survival::Surv(time, status) ~ 1))
idx <- match(time, hazard[, "time"])
hazard[idx, "hazard"]
}
mice/R/flux.R 0000644 0001762 0000144 00000021402 14330031606 012464 0 ustar ligges users #' Influx and outflux of multivariate missing data patterns
#'
#' Influx and outflux are statistics of the missing data pattern. These
#' statistics are useful in selecting predictors that should go into the
#' imputation model.
#'
#' Infux and outflux have been proposed by Van Buuren (2018), chapter 4.
#'
#' Influx is equal to the number of variable pairs \code{(Yj , Yk)} with
#' \code{Yj} missing and \code{Yk} observed, divided by the total number of
#' observed data cells. Influx depends on the proportion of missing data of the
#' variable. Influx of a completely observed variable is equal to 0, whereas for
#' completely missing variables we have influx = 1. For two variables with the
#' same proportion of missing data, the variable with higher influx is better
#' connected to the observed data, and might thus be easier to impute.
#'
#' Outflux is equal to the number of variable pairs with \code{Yj} observed and
#' \code{Yk} missing, divided by the total number of incomplete data cells.
#' Outflux is an indicator of the potential usefulness of \code{Yj} for imputing
#' other variables. Outflux depends on the proportion of missing data of the
#' variable. Outflux of a completely observed variable is equal to 1, whereas
#' outflux of a completely missing variable is equal to 0. For two variables
#' having the same proportion of missing data, the variable with higher outflux
#' is better connected to the missing data, and thus potentially more useful for
#' imputing other variables.
#'
#' FICO is an outbound statistic defined by the fraction of incomplete cases
#' among cases with \code{Yj} observed (White and Carlin, 2010).
#'
#' @aliases flux
#' @param data A data frame or a matrix containing the incomplete data. Missing
#' values are coded as NA's.
#' @param local A vector of names of columns of \code{data}. The default is to
#' include all columns in the calculations.
#' @return A data frame with \code{ncol(data)} rows and six columns:
#' pobs = Proportion observed,
#' influx = Influx
#' outflux = Outflux
#' ainb = Average inbound statistic
#' aout = Average outbound statistic
#' fico = Fraction of incomplete cases among cases with \code{Yj} observed
#' @seealso \code{\link{fluxplot}}, \code{\link{md.pattern}}, \code{\link{fico}}
#' @author Stef van Buuren, 2012
#' @references
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/missing-data-pattern.html#sec:flux}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#'
#' White, I.R., Carlin, J.B. (2010). Bias and efficiency of multiple imputation
#' compared with complete-case analysis for missing covariate values.
#' \emph{Statistics in Medicine}, \emph{29}, 2920-2931.
#' @keywords misc
#' @export
flux <- function(data, local = names(data)) {
.avg <- function(row) sum(row, na.rm = TRUE) / (length(row) - 1)
## calculates influx and outflux statistics
## of the missing data pattern
x <- colMeans(!is.na(data))
pat <- md.pairs(data)
pat$rr <- pat$rr[local, , drop = FALSE]
pat$rm <- pat$rm[local, , drop = FALSE]
pat$mr <- pat$mr[local, , drop = FALSE]
pat$mm <- pat$mm[local, , drop = FALSE]
ainb <- apply(pat$mr / (pat$mr + pat$mm), 1, .avg)
aout <- apply(pat$rm / (pat$rm + pat$rr), 1, .avg)
fico <- fico(data)
outflux <- rowSums(pat$rm) / (rowSums(pat$rm + pat$mm))
influx <- rowSums(pat$mr) / (rowSums(pat$mr + pat$rr))
data.frame(pobs = x, influx = influx, outflux = outflux, ainb = ainb, aout = aout, fico = fico)
}
#' Fluxplot of the missing data pattern
#'
#' Influx and outflux are statistics of the missing data pattern. These
#' statistics are useful in selecting predictors that should go into the
#' imputation model.
#'
#' Infux and outflux have been proposed by Van Buuren (2012), chapter 4.
#'
#' Influx is equal to the number of variable pairs \code{(Yj , Yk)} with
#' \code{Yj} missing and \code{Yk} observed, divided by the total number of
#' observed data cells. Influx depends on the proportion of missing data of the
#' variable. Influx of a completely observed variable is equal to 0, whereas for
#' completely missing variables we have influx = 1. For two variables with the
#' same proportion of missing data, the variable with higher influx is better
#' connected to the observed data, and might thus be easier to impute.
#'
#' Outflux is equal to the number of variable pairs with \code{Yj} observed and
#' \code{Yk} missing, divided by the total number of incomplete data cells.
#' Outflux is an indicator of the potential usefulness of \code{Yj} for imputing
#' other variables. Outflux depends on the proportion of missing data of the
#' variable. Outflux of a completely observed variable is equal to 1, whereas
#' outflux of a completely missing variable is equal to 0. For two variables
#' having the same proportion of missing data, the variable with higher outflux
#' is better connected to the missing data, and thus potentially more useful for
#' imputing other variables.
#'
#' @aliases fluxplot
#' @param data A data frame or a matrix containing the incomplete data. Missing
#' values are coded as NA's.
#' @param local A vector of names of columns of \code{data}. The default is to
#' include all columns in the calculations.
#' @param plot Should a graph be produced?
#' @param labels Should the points be labeled?
#' @param xlim See \code{par}.
#' @param ylim See \code{par}.
#' @param las See \code{par}.
#' @param xlab See \code{par}.
#' @param ylab See \code{par}.
#' @param main See \code{par}.
#' @param eqscplot Should a square plot be produced?
#' @param pty See \code{par}.
#' @param lwd See \code{par}. Controls axis line thickness and diagonal
#' @param \dots Further arguments passed to \code{plot()} or \code{eqscplot()}.
#' @return An invisible data frame with \code{ncol(data)} rows and six columns:
#' pobs = Proportion observed,
#' influx = Influx
#' outflux = Outflux
#' ainb = Average inbound statistic
#' aout = Average outbound statistic
#' fico = Fraction of incomplete cases among cases with \code{Yj} observed
#' @seealso \code{\link{flux}}, \code{\link{md.pattern}}, \code{\link{fico}}
#' @author Stef van Buuren, 2012
#' @references
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/missing-data-pattern.html#sec:flux}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#'
#' White, I.R., Carlin, J.B. (2010). Bias and efficiency of multiple imputation
#' compared with complete-case analysis for missing covariate values.
#' \emph{Statistics in Medicine}, \emph{29}, 2920-2931.
#' @keywords misc
#' @export
fluxplot <- function(data, local = names(data),
plot = TRUE, labels = TRUE,
xlim = c(0, 1), ylim = c(0, 1), las = 1,
xlab = "Influx", ylab = "Outflux",
main = paste("Influx-outflux pattern for", deparse(substitute(data))),
eqscplot = TRUE, pty = "s",
lwd = 1,
...) {
f <- flux(data, local)
if (plot) {
if (eqscplot) {
MASS::eqscplot(
x = f$influx, y = f$outflux, type = "n",
main = main,
xlab = xlab, ylab = ylab,
xlim = xlim, ylim = ylim,
pty = pty, lwd = lwd, axes = FALSE, ...
)
} else {
plot(
x = f$influx, y = f$outflux, type = "n",
main = main,
xlab = xlab, ylab = ylab,
xlim = xlim, ylim = ylim,
pty = pty, lwd = lwd, axes = FALSE, ...
)
}
axis(1, lwd = lwd, las = las)
axis(2, lwd = lwd, las = las)
abline(1, -1, lty = 2, lwd = lwd)
if (labels) {
text(x = f$influx, y = f$outflux, label = names(data), ...)
} else {
points(x = f$influx, y = f$outflux, ...)
}
box(lwd = lwd)
}
invisible(data.frame(f))
}
#' Fraction of incomplete cases among cases with observed
#'
#' FICO is an outbound statistic defined by the fraction of incomplete cases
#' among cases with \code{Yj} observed (White and Carlin, 2010).
#'
#' @aliases fico
#' @param data A data frame or a matrix containing the incomplete data. Missing
#' values are coded as NA's.
#' @return A vector of length \code{ncol(data)} of FICO statistics.
#' @seealso \code{\link{fluxplot}}, \code{\link{flux}}, \code{\link{md.pattern}}
#' @author Stef van Buuren, 2012
#' @references
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/missing-data-pattern.html#sec:flux}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#'
#' White, I.R., Carlin, J.B. (2010). Bias and efficiency of multiple imputation
#' compared with complete-case analysis for missing covariate values.
#' \emph{Statistics in Medicine}, \emph{29}, 2920-2931.
#' @keywords misc
#' @export
fico <- function(data) {
ic <- ici(data)
unlist(lapply(data, FUN = function(x) sum((!is.na(x)) & ic) / sum(!is.na(x))))
}
mice/R/fdd.R 0000644 0001762 0000144 00000010222 14330031606 012241 0 ustar ligges users #' SE Fireworks disaster data
#'
#' Multiple outcomes of a randomized study to reduce post-traumatic stress.
#'
#' Data from a randomized experiment to reduce post-traumatic stress by two
#' treatments: Eye Movement Desensitization and Reprocessing (EMDR)
#' (experimental treatment), and cognitive behavioral therapy (CBT) (control
#' treatment). 52 children were randomized to one of these two treatments.
#' Outcomes were measured at three time points: at baseline (pre-treatment, T1),
#' post-treatment (T2, 4-8 weeks), and at follow-up (T3, 3 months). For more
#' details, see de Roos et al (2011). Some person covariates were reshuffled.
#' The imputation methodology is explained in Chapter 9 of van Buuren (2012).
#'
#' @name fdd
#' @aliases fdd fdd.pred
#' @docType data
#' @format \code{fdd} is a data frame with 52 rows and 65 columns:
#' \describe{
#' \item{id}{Client number}
#' \item{trt}{Treatment (E=EMDR, C=CBT)}
#' \item{pp}{Per protocol (Y/N)}
#' \item{trtp}{Number of parental treatments}
#' \item{sex}{Sex: M/F}
#' \item{etn}{Ethnicity: NL/OTHER}
#' \item{age}{Age (years)}
#' \item{trauma}{Trauma count (1-5)}
#' \item{prop1}{PROPS total score T1}
#' \item{prop2}{PROPS total score T2}
#' \item{prop3}{PROPS total score T3}
#' \item{crop1}{CROPS total score T1}
#' \item{crop2}{CROPS total score T2}
#' \item{crop3}{CROPS total score T3}
#' \item{masc1}{MASC score T1}
#' \item{masc2}{MASC score T2}
#' \item{masc3}{MASC score T3}
#' \item{cbcl1}{CBCL T1}
#' \item{cbcl3}{CBCL T3}
#' \item{prs1}{PRS total score T1}
#' \item{prs2}{PRS total score T2}
#' \item{prs3}{PRS total score T3}
#' \item{ypa1}{PTSD-RI B intrusive recollection parent T1}
#' \item{ypb1}{PTSD-RI C avoidant/numbing parent T1}
#' \item{ypc1}{PTSD-RI D hyper-arousal parent T1}
#' \item{yp1}{PTSD-RI B+C+D parent T1}
#' \item{ypa2}{PTSD-RI B intrusive recollection parent T2}
#' \item{ypb2}{PTSD-RI C avoidant/numbing parent T2}
#' \item{ypc2}{PTSD-RI D hyper-arousal parent T2}
#' \item{yp2}{PTSD-RI B+C+D parent T1}
#' \item{ypa3}{PTSD-RI B intrusive recollection parent T3}
#' \item{ypb3}{PTSD-RI C avoidant/numbing parent T3}
#' \item{ypc3}{PTSD-RI D hyper-arousal parent T3}
#' \item{yp3}{PTSD-RI B+C+D parent T3}
#' \item{yca1}{PTSD-RI B intrusive recollection child T1}
#' \item{ycb1}{PTSD-RI C avoidant/numbing child T1}
#' \item{ycc1}{PTSD-RI D hyper-arousal child T1}
#' \item{yc1}{PTSD-RI B+C+D child T1}
#' \item{yca2}{PTSD-RI B intrusive recollection child T2}
#' \item{ycb2}{PTSD-RI C avoidant/numbing child T2}
#' \item{ycc2}{PTSD-RI D hyper-arousal child T2}
#' \item{yc2}{PTSD-RI B+C+D child T2}
#' \item{yca3}{PTSD-RI B intrusive recollection child T3}
#' \item{ycb3}{PTSD-RI C avoidant/numbing child T3}
#' \item{ycc3}{PTSD-RI D hyper-arousal child T3}
#' \item{yc3}{PTSD-RI B+C+D child T3}
#' \item{ypf1}{PTSD-RI parent full T1}
#' \item{ypf2}{PTSD-RI parent full T2}
#' \item{ypf3}{PTSD-RI parent full T3}
#' \item{ypp1}{PTSD parent partial T1}
#' \item{ypp2}{PTSD parent partial T2}
#' \item{ypp3}{PTSD parent partial T3}
#' \item{ycf1}{PTSD child full T1}
#' \item{ycf2}{PTSD child full T2}
#' \item{ycf3}{PTSD child full T3}
#' \item{ycp1}{PTSD child partial T1}
#' \item{ycp2}{PTSD child partial T2}
#' \item{ycp3}{PTSD child partial T3}
#' \item{cbin1}{CBCL Internalizing T1}
#' \item{cbin3}{CBCL Internalizing T3}
#' \item{cbex1}{CBCL Externalizing T1}
#' \item{cbex3}{CBCL Externalizing T3}
#' \item{bir1}{Birlison T1}
#' \item{bir2}{Birlison T2}
#' \item{bir3}{Birlison T3}
#' }
#' \code{fdd.pred} is the 65 by 65 binary
#' predictor matrix used to impute \code{fdd}.
#' @source de Roos, C., Greenwald, R., den Hollander-Gijsman, M., Noorthoorn,
#' E., van Buuren, S., de Jong, A. (2011). A Randomised Comparison of Cognitive
#' Behavioral Therapy (CBT) and Eye Movement Desensitisation and Reprocessing
#' (EMDR) in disaster-exposed children. \emph{European Journal of
#' Psychotraumatology}, \emph{2}, 5694.
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-fdd.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' Boca Raton, FL.: Chapman & Hall/CRC Press.
#' @keywords datasets
#' @examples
#'
#'
#' data <- fdd
#' md.pattern(fdd)
NULL
mice/R/method.R 0000644 0001762 0000144 00000010111 14330031606 012761 0 ustar ligges users #' Creates a \code{method} argument
#'
#' This helper function creates a valid \code{method} vector. The
#' \code{method} vector is an argument to the \code{mice} function that
#' specifies the method for each block.
#' @inheritParams mice
#' @return Vector of \code{length(blocks)} element with method names
#' @seealso \code{\link{mice}}
#' @examples
#' make.method(nhanes2)
#' @export
make.method <- function(data,
where = make.where(data),
blocks = make.blocks(data),
defaultMethod = c("pmm", "logreg", "polyreg", "polr")) {
method <- rep("", length(blocks))
names(method) <- names(blocks)
for (j in names(blocks)) {
yvar <- blocks[[j]]
y <- data[, yvar]
def <- sapply(y, assign.method)
k <- ifelse(all(diff(def) == 0), k <- def[1], 1)
method[j] <- defaultMethod[k]
}
nimp <- nimp(where, blocks)
method[nimp == 0] <- ""
method
}
check.method <- function(method, data, where, blocks, defaultMethod) {
if (is.null(method)) {
return(make.method(
data = data,
where = where,
blocks = blocks,
defaultMethod = defaultMethod
))
}
nimp <- nimp(where, blocks)
# expand user's imputation method to all visited columns
# single string supplied by user (implicit assumption of two columns)
if (length(method) == 1) {
if (is.passive(method)) {
stop("Cannot have a passive imputation method for every column.")
}
method <- rep(method, length(blocks))
method[nimp == 0] <- ""
}
# check the length of the argument
if (length(method) != length(blocks)) {
stop("Length of method differs from number of blocks", call. = FALSE)
}
# add names to method
names(method) <- names(blocks)
# check whether the requested imputation methods are on the search path
active.check <- !is.passive(method) & nimp > 0 & method != ""
passive.check <- is.passive(method) & nimp > 0 & method != ""
check <- all(active.check) & any(passive.check)
if (check) {
fullNames <- rep.int("mice.impute.passive", length(method[passive.check]))
} else {
fullNames <- paste("mice.impute", method[active.check], sep = ".")
if (length(method[active.check]) == 0) fullNames <- character(0)
}
# type checks on built-in imputation methods
for (j in names(blocks)) {
vname <- blocks[[j]]
y <- data[, vname, drop = FALSE]
mj <- method[j]
mlist <- list(
m1 = c("logreg", "logreg.boot", "polyreg", "lda", "polr"),
m2 = c(
"norm", "norm.nob", "norm.predict", "norm.boot",
"mean", "2l.norm", "2l.pan",
"2lonly.norm", "2lonly.pan",
"quadratic", "ri"
),
m3 = c(
"norm", "norm.nob", "norm.predict", "norm.boot",
"mean", "2l.norm", "2l.pan",
"2lonly.norm", "2lonly.pan",
"quadratic", "logreg", "logreg.boot"
)
)
cond1 <- sapply(y, is.numeric)
cond2 <- sapply(y, is.factor) & sapply(y, nlevels) == 2
cond3 <- sapply(y, is.factor) & sapply(y, nlevels) > 2
if (any(cond1) && mj %in% mlist$m1) {
warning("Type mismatch for variable(s): ",
paste(vname[cond1], collapse = ", "),
"\nImputation method ", mj, " is for categorical data.",
call. = FALSE
)
}
if (any(cond2) && mj %in% mlist$m2) {
warning("Type mismatch for variable(s): ",
paste(vname[cond2], collapse = ", "),
"\nImputation method ", mj, " is not for factors.",
call. = FALSE
)
}
if (any(cond3) && mj %in% mlist$m3) {
warning("Type mismatch for variable(s): ",
paste(vname[cond3], collapse = ", "),
"\nImputation method ", mj, " is not for factors with >2 levels.",
call. = FALSE
)
}
}
method[nimp == 0] <- ""
unlist(method)
}
# assign methods based on type,
# use method 1 if there is no single method within the block
assign.method <- function(y) {
if (is.numeric(y)) {
return(1)
}
if (nlevels(y) == 2) {
return(2)
}
if (is.ordered(y) && nlevels(y) > 2) {
return(4)
}
if (nlevels(y) > 2) {
return(3)
}
if (is.logical(y)) {
return(2)
}
1
}
mice/R/toenail.R 0000644 0001762 0000144 00000004111 14330031606 013137 0 ustar ligges users #' Toenail data
#'
#' The toenail data come from a Multicenter study comparing two oral
#' treatments for toenail infection. Patients were evaluated for the
#' degree of separation of the nail. Patients were randomized into two
#' treatments and were followed over seven visits - four in the first
#' year and yearly thereafter. The patients have not been treated
#' prior to the first visit so this should be regarded as the
#' baseline.
#' @name toenail
#' @docType data
#' @format A data frame with 1908 observations on the following 5 variables:
#' \describe{
#' \item{\code{ID}}{a numeric vector giving the ID of patient}
#' \item{\code{outcome}}{a numeric vector giving the response
#' (0=none or mild seperation, 1=moderate or severe)}
#' \item{\code{treatment}}{a numeric vector giving the treatment group}
#' \item{\code{month}}{a numeric vector giving the time of the visit
#' (not exactly monthly intervals hence not round numbers)}
#' \item{\code{visit}}{a numeric vector giving the number of the visit}
#' }
#' @source
#' De Backer, M., De Vroey, C., Lesaffre, E., Scheys, I., and De
#' Keyser, P. (1998). Twelve weeks of continuous oral therapy for
#' toenail onychomycosis caused by dermatophytes: A double-blind
#' comparative trial of terbinafine 250 mg/day versus itraconazole 200
#' mg/day. Journal of the American Academy of Dermatology, 38, 57-63.
#' @references
#' Lesaffre, E. and Spiessens, B. (2001). On the effect of the number of
#' quadrature points in a logistic random-effects model: An example.
#' Journal of the Royal Statistical Society, Series C, 50, 325-335.
#'
#' G. Fitzmaurice, N. Laird and J. Ware (2004) Applied Longitudinal Analysis,
#' Wiley and Sons, New York, USA.
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-catoutcome.html#example}{\emph{Flexible
#' Imputation of Missing Data. Second Edition.}} Chapman & Hall/CRC.
#' Boca Raton, FL.
#' @keywords datasets
#' @seealso \code{\link{toenail2}}
#' @details This dataset was copied from the \code{DPpackage}, which is
#' scheduled to be discontinued from CRAN in August 2019.
NULL
mice/R/mice.impute.pmm.R 0000644 0001762 0000144 00000022154 14335404116 014527 0 ustar ligges users #' Imputation by predictive mean matching
#'
#' @aliases mice.impute.pmm pmm
#' @param y Vector to be imputed
#' @param ry Logical vector of length \code{length(y)} indicating the
#' the subset \code{y[ry]} of elements in \code{y} to which the imputation
#' model is fitted. The \code{ry} generally distinguishes the observed
#' (\code{TRUE}) and missing values (\code{FALSE}) in \code{y}.
#' @param x Numeric design matrix with \code{length(y)} rows with predictors for
#' \code{y}. Matrix \code{x} may have no missing values.
#' @param exclude Value or vector of values to exclude from the imputation donor pool in \code{y}
#' @param wy Logical vector of length \code{length(y)}. A \code{TRUE} value
#' indicates locations in \code{y} for which imputations are created.
#' @param donors The size of the donor pool among which a draw is made.
#' The default is \code{donors = 5L}. Setting \code{donors = 1L} always selects
#' the closest match, but is not recommended. Values between 3L and 10L
#' provide the best results in most cases (Morris et al, 2015).
#' @param matchtype Type of matching distance. The default choice
#' (\code{matchtype = 1L}) calculates the distance between
#' the \emph{predicted} value of \code{yobs} and
#' the \emph{drawn} values of \code{ymis} (called type-1 matching).
#' Other choices are \code{matchtype = 0L}
#' (distance between predicted values) and \code{matchtype = 2L}
#' (distance between drawn values).
#' @param ridge The ridge penalty used in \code{.norm.draw()} to prevent
#' problems with multicollinearity. The default is \code{ridge = 1e-05},
#' which means that 0.01 percent of the diagonal is added to the cross-product.
#' Larger ridges may result in more biased estimates. For highly noisy data
#' (e.g. many junk variables), set \code{ridge = 1e-06} or even lower to
#' reduce bias. For highly collinear data, set \code{ridge = 1e-04} or higher.
#' @param use.matcher Logical. Set \code{use.matcher = TRUE} to specify
#' the C function \code{matcher()}, the now deprecated matching function that
#' was default in versions
#' \code{2.22} (June 2014) to \code{3.11.7} (Oct 2020). Since version \code{3.12.0}
#' \code{mice()} uses the much faster \code{matchindex} C function. Use
#' the deprecated \code{matcher} function only for exact reproduction.
#' @param \dots Other named arguments.
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @author Gerko Vink, Stef van Buuren, Karin Groothuis-Oudshoorn
#' @details
#' Imputation of \code{y} by predictive mean matching, based on
#' van Buuren (2012, p. 73). The procedure is as follows:
#'
#' \enumerate{
#' \item{Calculate the cross-product matrix \eqn{S=X_{obs}'X_{obs}}.}
#' \item{Calculate \eqn{V = (S+{diag}(S)\kappa)^{-1}}, with some small ridge
#' parameter \eqn{\kappa}.}
#' \item{Calculate regression weights \eqn{\hat\beta = VX_{obs}'y_{obs}.}}
#' \item{Draw \eqn{q} independent \eqn{N(0,1)} variates in vector \eqn{\dot z_1}.}
#' \item{Calculate \eqn{V^{1/2}} by Cholesky decomposition.}
#' \item{Calculate \eqn{\dot\beta = \hat\beta + \dot\sigma\dot z_1 V^{1/2}}.}
#' \item{Calculate \eqn{\dot\eta(i,j)=|X_{{obs},[i]|}\hat\beta-X_{{mis},[j]}\dot\beta}
#' with \eqn{i=1,\dots,n_1} and \eqn{j=1,\dots,n_0}.}
#' \item{Construct \eqn{n_0} sets \eqn{Z_j}, each containing \eqn{d} candidate donors, from Y_{obs} such that \eqn{\sum_d\dot\eta(i,j)} is minimum for all \eqn{j=1,\dots,n_0}. Break ties randomly.}
#' \item{Draw one donor \eqn{i_j} from \eqn{Z_j} randomly for \eqn{j=1,\dots,n_0}.}
#' \item{Calculate imputations \eqn{\dot y_j = y_{i_j}} for \eqn{j=1,\dots,n_0}.}
#' }
#'
#' The name \emph{predictive mean matching} was proposed by Little (1988).
#'
#' @references Little, R.J.A. (1988), Missing data adjustments in large surveys
#' (with discussion), Journal of Business Economics and Statistics, 6, 287--301.
#'
#' Morris TP, White IR, Royston P (2015). Tuning multiple imputation by predictive
#' mean matching and local residual draws. BMC Med Res Methodol. ;14:75.
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-pmm.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#'
#' Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}: Multivariate
#' Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
#' Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
#' @family univariate imputation functions
#' @keywords datagen
#' @examples
#' # We normally call mice.impute.pmm() from within mice()
#' # But we may call it directly as follows (not recommended)
#'
#' set.seed(53177)
#' xname <- c("age", "hgt", "wgt")
#' r <- stats::complete.cases(boys[, xname])
#' x <- boys[r, xname]
#' y <- boys[r, "tv"]
#' ry <- !is.na(y)
#' table(ry)
#'
#' # percentage of missing data in tv
#' sum(!ry) / length(ry)
#'
#' # Impute missing tv data
#' yimp <- mice.impute.pmm(y, ry, x)
#' length(yimp)
#' hist(yimp, xlab = "Imputed missing tv")
#'
#' # Impute all tv data
#' yimp <- mice.impute.pmm(y, ry, x, wy = rep(TRUE, length(y)))
#' length(yimp)
#' hist(yimp, xlab = "Imputed missing and observed tv")
#' plot(jitter(y), jitter(yimp),
#' main = "Predictive mean matching on age, height and weight",
#' xlab = "Observed tv (n = 224)",
#' ylab = "Imputed tv (n = 224)"
#' )
#' abline(0, 1)
#' cor(y, yimp, use = "pair")
#'
#' # Use blots to exclude different values per column
#' # Create blots object
#' blots <- make.blots(boys)
#' # Exclude ml 1 through 5 from tv donor pool
#' blots$tv$exclude <- c(1:5)
#' # Exclude 100 random observed heights from tv donor pool
#' blots$hgt$exclude <- sample(unique(boys$hgt), 100)
#' imp <- mice(boys, method = "pmm", print = FALSE, blots = blots, seed=123)
#' blots$hgt$exclude %in% unlist(c(imp$imp$hgt)) # MUST be all FALSE
#' blots$tv$exclude %in% unlist(c(imp$imp$tv)) # MUST be all FALSE
#' @export
mice.impute.pmm <- function(y, ry, x, wy = NULL, donors = 5L,
matchtype = 1L, exclude = -99999999, ridge = 1e-05,
use.matcher = FALSE, ...) {
id.ex <- !ry | !y %in% exclude # id vector for exclusion
y <- y[id.ex] # leave out the exclude vector y's
# allow for one-dimensional x-space
if(!is.null(dim(x))){
x <- x[id.ex, ]
} else {
x <- x[id.ex]
}
# leave out the exclude vector x's
ry <- ry[id.ex] # leave out the exclude vector indicator
{
if (is.null(wy)) {
wy <- !ry
} else {
wy <- wy[id.ex] # if applicable adjust wy to match exclude
}
}
x <- cbind(1, as.matrix(x))
ynum <- y
if (is.factor(y)) {
ynum <- as.integer(y)
}
parm <- .norm.draw(ynum, ry, x, ridge = ridge, ...)
if (matchtype == 0L) {
yhatobs <- x[ry, , drop = FALSE] %*% parm$coef
yhatmis <- x[wy, , drop = FALSE] %*% parm$coef
}
if (matchtype == 1L) {
yhatobs <- x[ry, , drop = FALSE] %*% parm$coef
yhatmis <- x[wy, , drop = FALSE] %*% parm$beta
}
if (matchtype == 2L) {
yhatobs <- x[ry, , drop = FALSE] %*% parm$beta
yhatmis <- x[wy, , drop = FALSE] %*% parm$beta
}
if (use.matcher) {
idx <- matcher(yhatobs, yhatmis, k = donors)
} else {
idx <- matchindex(yhatobs, yhatmis, donors)
}
return(y[ry][idx])
}
#' Finds an imputed value from matches in the predictive metric (deprecated)
#'
#' This function finds matches among the observed data in the predictive
#' mean metric. It selects the \code{donors} closest matches, randomly
#' samples one of the donors, and returns the observed value of the
#' match.
#'
#' This function is included for backward compatibility. It was
#' used up to \code{mice 2.21}. The current \code{mice.impute.pmm()}
#' function calls the faster \code{C} function \code{matcher} instead of
#' \code{.pmm.match()}.
#'
#' @aliases .pmm.match
#' @param z A scalar containing the predicted value for the current case
#' to be imputed.
#' @param yhat A vector containing the predicted values for all cases with an observed
#' outcome.
#' @param y A vector of \code{length(yhat)} elements containing the observed outcome
#' @param donors The size of the donor pool among which a draw is made. The default is
#' \code{donors = 5}. Setting \code{donors = 1} always selects the closest match. Values
#' between 3 and 10 provide the best results. Note: This setting was changed from
#' 3 to 5 in version 2.19, based on simulation work by Tim Morris (UCL).
#' @param \dots Other parameters (not used).
#' @return A scalar containing the observed value of the selected donor.
#' @author Stef van Buuren
#' @rdname pmm.match
#' @references
#' Schenker N & Taylor JMG (1996) Partially parametric techniques
#' for multiple imputation. \emph{Computational Statistics and Data Analysis}, 22, 425-446.
#'
#' Little RJA (1988) Missing-data adjustments in large surveys (with discussion).
#' \emph{Journal of Business Economics and Statistics}, 6, 287-301.
#'
#' @export
.pmm.match <- function(z, yhat = yhat, y = y, donors = 5, ...) {
d <- abs(yhat - z)
f <- d > 0
a1 <- ifelse(any(f), min(d[f]), 1)
d <- d + runif(length(d), 0, a1 / 10^10)
if (donors == 1) {
return(y[which.min(d)])
}
donors <- min(donors, length(d))
donors <- max(donors, 1)
ds <- sort.int(d, partial = donors)
m <- sample(y[d <= ds[donors]], 1)
return(m)
}
mice/R/pattern1.R 0000644 0001762 0000144 00000003252 14436640305 013260 0 ustar ligges users #' Datasets with various missing data patterns
#'
#' Four simple datasets with various missing data patterns
#'
#' Van Buuren (2012) uses these four artificial datasets to illustrate various
#' missing data patterns.
#'
#' @name pattern
#' @aliases pattern1 pattern2 pattern3 pattern4
#' @docType data
#' @format \describe{ \item{list("pattern1")}{Data with a univariate missing
#' data pattern} \item{list("pattern2")}{Data with a monotone missing data
#' pattern} \item{list("pattern3")}{Data with a file matching missing data
#' pattern} \item{list("pattern4")}{Data with a general missing data pattern} }
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/missing-data-pattern.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' @keywords datasets
#' @examples
#' pattern4
#'
#' data <- rbind(pattern1, pattern2, pattern3, pattern4)
#' mdpat <- cbind(expand.grid(rec = 8:1, pat = 1:4, var = 1:3), r = as.numeric(as.vector(is.na(data))))
#'
#' types <- c("Univariate", "Monotone", "File matching", "General")
#' tp41 <- lattice::levelplot(r ~ var + rec | as.factor(pat),
#' data = mdpat,
#' as.table = TRUE, aspect = "iso",
#' shrink = c(0.9),
#' col.regions = mdc(1:2),
#' colorkey = FALSE,
#' scales = list(draw = FALSE),
#' xlab = "", ylab = "",
#' between = list(x = 1, y = 0),
#' strip = lattice::strip.custom(
#' bg = "grey95", style = 1,
#' factor.levels = types
#' )
#' )
#' print(tp41)
#'
#' md.pattern(pattern4)
#' p <- md.pairs(pattern4)
#' p
#'
#' ### proportion of usable cases
#' p$mr / (p$mr + p$mm)
#'
#' ### outbound statistics
#' p$rm / (p$rm + p$rr)
#'
#'
#' fluxplot(pattern2)
NULL
mice/R/quickpred.R 0000644 0001762 0000144 00000013272 14330031647 013510 0 ustar ligges users #' Quick selection of predictors from the data
#'
#' Selects predictors according to simple statistics
#'
#' This function creates a predictor matrix using the variable selection
#' procedure described in Van Buuren et al.~(1999, p.~687--688). The function is
#' designed to aid in setting up a good imputation model for data with many
#' variables.
#'
#' Basic workings: The procedure calculates for each variable pair (i.e.
#' target-predictor pair) two correlations using all available cases per pair.
#' The first correlation uses the values of the target and the predictor
#' directly. The second correlation uses the (binary) response indicator of the
#' target and the values of the predictor. If the largest (in absolute value) of
#' these correlations exceeds \code{mincor}, the predictor will be added to the
#' imputation set. The default value for \code{mincor} is 0.1.
#'
#' In addition, the procedure eliminates predictors whose proportion of usable
#' cases fails to meet the minimum specified by \code{minpuc}. The default value
#' is 0, so predictors are retained even if they have no usable case.
#'
#' Finally, the procedure includes any predictors named in the \code{include}
#' argument (which is useful for background variables like age and sex) and
#' eliminates any predictor named in the \code{exclude} argument. If a variable
#' is listed in both \code{include} and \code{exclude} arguments, the
#' \code{include} argument takes precedence.
#'
#' Advanced topic: \code{mincor} and \code{minpuc} are typically specified as
#' scalars, but vectors and squares matrices of appropriate size will also work.
#' Each element of the vector corresponds to a row of the predictor matrix, so
#' the procedure can effectively differentiate between different target
#' variables. Setting a high values for can be useful for auxiliary, less
#' important, variables. The set of predictor for those variables can remain
#' relatively small. Using a square matrix extends the idea to the columns, so
#' that one can also apply cellwise thresholds.
#'
#' @note \code{quickpred()} uses \code{\link[base]{data.matrix}} to convert
#' factors to numbers through their internal codes. Especially for unordered
#' factors the resulting quantification may not make sense.
#'
#' @param data Matrix or data frame with incomplete data.
#' @param mincor A scalar, numeric vector (of size \code{ncol(data))} or numeric
#' matrix (square, of size \code{ncol(data)} specifying the minimum
#' threshold(s) against which the absolute correlation in the data is compared.
#' @param minpuc A scalar, vector (of size \code{ncol(data))} or matrix (square,
#' of size \code{ncol(data)} specifying the minimum threshold(s) for the
#' proportion of usable cases.
#' @param include A string or a vector of strings containing one or more
#' variable names from \code{names(data)}. Variables specified are always
#' included as a predictor.
#' @param exclude A string or a vector of strings containing one or more
#' variable names from \code{names(data)}. Variables specified are always
#' excluded as a predictor.
#' @param method A string specifying the type of correlation. Use
#' \code{'pearson'} (default), \code{'kendall'} or \code{'spearman'}. Can be
#' abbreviated.
#' @return A square binary matrix of size \code{ncol(data)}.
#' @author Stef van Buuren, Aug 2009
#' @seealso \code{\link{mice}}, \code{\link[=mids-class]{mids}}
#' @references van Buuren, S., Boshuizen, H.C., Knook, D.L. (1999) Multiple
#' imputation of missing blood pressure covariates in survival analysis.
#' \emph{Statistics in Medicine}, \bold{18}, 681--694.
#'
#' van Buuren, S. and Groothuis-Oudshoorn, K. (2011). \code{mice}: Multivariate
#' Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
#' Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
#' @keywords misc
#' @examples
#' # default: include all predictors with absolute correlation over 0.1
#' quickpred(nhanes)
#'
#' # all predictors with absolute correlation over 0.4
#' quickpred(nhanes, mincor = 0.4)
#'
#' # include age and bmi, exclude chl
#' quickpred(nhanes, mincor = 0.4, inc = c("age", "bmi"), exc = "chl")
#'
#' # only include predictors with at least 30% usable cases
#' quickpred(nhanes, minpuc = 0.3)
#'
#' # use low threshold for bmi, and high thresholds for hyp and chl
#' pred <- quickpred(nhanes, mincor = c(0, 0.1, 0.5, 0.5))
#' pred
#'
#' # use it directly from mice
#' imp <- mice(nhanes, pred = quickpred(nhanes, minpuc = 0.25, include = "age"))
#' @export
quickpred <- function(data, mincor = 0.1, minpuc = 0, include = "",
exclude = "", method = "pearson") {
data <- check.dataform(data)
# initialize
nvar <- ncol(data)
predictorMatrix <- matrix(0, nrow = nvar, ncol = nvar, dimnames = list(names(data), names(data)))
x <- data.matrix(data)
r <- !is.na(x)
# include predictors with
# 1) pairwise correlation among data
# 2) pairwise correlation of data with response indicator higher than mincor
suppressWarnings(v <- abs(cor(x, use = "pairwise.complete.obs", method = method)))
v[is.na(v)] <- 0
suppressWarnings(u <- abs(cor(y = x, x = r, use = "pairwise.complete.obs", method = method)))
u[is.na(u)] <- 0
maxc <- pmax(v, u)
predictorMatrix[maxc > mincor] <- 1
# exclude predictors with a percentage usable cases below minpuc
p <- md.pairs(data)
puc <- p$mr / (p$mr + p$mm)
predictorMatrix[puc < minpuc] <- 0
# exclude predictors listed in the exclude argument
yz <- pmatch(exclude, names(data))
predictorMatrix[, yz] <- 0
# include predictors listed in the include argument
yz <- pmatch(include, names(data))
predictorMatrix[, yz] <- 1
# some final processing
diag(predictorMatrix) <- 0
predictorMatrix[colSums(!r) == 0, ] <- 0
predictorMatrix
}
mice/R/cc.R 0000644 0001762 0000144 00000004173 14330031606 012101 0 ustar ligges users #' Select complete cases
#'
#' Extracts the complete cases, also known as \emph{listwise deletion}.
#' \code{cc(x)} is similar to
#' \code{na.omit(x)}, but returns an object of the same class
#' as the input data. Dimensions are not dropped. For extracting
#' incomplete cases, use \code{\link{ici}}.
#'
#' @param x An \code{R} object. Methods are available for classes
#' \code{mids}, \code{data.frame} and \code{matrix}. Also, \code{x}
#' could be a vector.
#' @return A \code{vector}, \code{matrix} or \code{data.frame} containing the data of the complete cases.
#' @author Stef van Buuren, 2017.
#' @seealso \code{\link{na.omit}}, \code{\link{cci}}, \code{\link{ici}}
#' @keywords univar
#' @examples
#'
#' # cc(nhanes) # get the 13 complete cases
#' # cc(nhanes$bmi) # extract complete bmi
#' @export
cc <- function(x) UseMethod("cc", x)
#' @export
cc.mids <- function(x) {
x$data[cci(x), , drop = FALSE]
}
#' @export
cc.matrix <- function(x) {
x[cci(x), , drop = FALSE]
}
#' @export
cc.data.frame <- function(x) {
x[cci(x), , drop = FALSE]
}
#' @export
cc.default <- function(x) {
x[cci(x)]
}
#' Select incomplete cases
#'
#' Extracts incomplete cases from a data set.
#' The companion function for selecting the complete cases is \code{\link{cc}}.
#'
#' @param x An \code{R} object. Methods are available for classes
#' \code{mids}, \code{data.frame} and \code{matrix}. Also, \code{x}
#' could be a vector.
#' @return A \code{vector}, \code{matrix} or \code{data.frame} containing the data of the complete cases.
#' @author Stef van Buuren, 2017.
#' @seealso \code{\link{cc}}, \code{\link{ici}}
#' @keywords univar
#' @examples
#'
#' ic(nhanes) # get the 12 rows with incomplete cases
#' ic(nhanes[1:10, ]) # incomplete cases within the first ten rows
#' ic(nhanes[, c("bmi", "hyp")]) # restrict extraction to variables bmi and hyp
#' @export
ic <- function(x) UseMethod("ic", x)
#' @export
ic.mids <- function(x) {
x$data[ici(x), , drop = FALSE]
}
#' @export
ic.matrix <- function(x) {
x[ici(x), , drop = FALSE]
}
#' @export
ic.data.frame <- function(x) {
x[ici(x), , drop = FALSE]
}
#' @export
ic.default <- function(x) {
x[ici(x)]
}
mice/R/pool.r.squared.R 0000644 0001762 0000144 00000007363 14330031647 014401 0 ustar ligges users #' Pools R^2 of m models fitted to multiply-imputed data
#'
#' The function pools the coefficients of determination R^2 or the adjusted
#' coefficients of determination (R^2_a) obtained with the \code{lm} modeling
#' function. For pooling it uses the Fisher \emph{z}-transformation.
#'
#' @param object An object of class 'mira' or 'mipo', produced by \code{lm.mids},
#' \code{with.mids}, or \code{pool} with \code{lm} as modeling function.
#' @param adjusted A logical value. If adjusted=TRUE then the adjusted R^2 is
#' calculated. The default value is FALSE.
#' @return Returns a 1x4 table with components. Component \code{est} is the
#' pooled R^2 estimate. Component \code{lo95} is the 95 \% lower bound of the pooled R^2.
#' Component \code{hi95} is the 95 \% upper bound of the pooled R^2.
#' Component \code{fmi} is the fraction of missing information due to nonresponse.
#' @author Karin Groothuis-Oudshoorn and Stef van Buuren, 2009
#' @seealso \code{\link{pool}},\code{\link{pool.scalar}}
#' @references Harel, O (2009). The estimation of R^2 and adjusted R^2 in
#' incomplete data sets using multiple imputation, Journal of Applied Statistics,
#' 36:1109-1118.
#'
#' Rubin, D.B. (1987). Multiple Imputation for Nonresponse in Surveys. New
#' York: John Wiley and Sons.
#'
#' van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}: Multivariate
#' Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
#' Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
#'
#' @keywords htest
#' @examples
#' imp <- mice(nhanes, print = FALSE, seed = 16117)
#' fit <- with(imp, lm(chl ~ age + hyp + bmi))
#'
#' # input: mira object
#' pool.r.squared(fit)
#' pool.r.squared(fit, adjusted = TRUE)
#'
#' # input: mipo object
#' est <- pool(fit)
#' pool.r.squared(est)
#' pool.r.squared(est, adjusted = TRUE)
#' @export
pool.r.squared <- function(object, adjusted = FALSE) {
call <- match.call()
if (!is.mira(object) & !is.mipo(object)) {
stop("The object must have class 'mira' or 'mipo'")
}
if (is.mira(object)) {
if ((m <- length(object$analyses)) < 2) {
stop("At least two imputations are needed for pooling.\n")
}
if (class((object$analyses[[1]]))[1] != "lm") {
stop("r^2 can only be calculated for results of the 'lm' modeling function")
}
glanced <- summary(object, type = "glance")
}
if (is.mipo(object)) {
if (nrow(object$glanced) < 2) {
stop("At least two imputations are needed for pooling.\n")
}
if (!"r.squared" %in% colnames(object$glanced)) {
stop("r^2 can only be calculated for results of the 'lm' modeling function")
}
glanced <- object$glanced
}
# Set up array r2 to store R2 values, Fisher z-transformations of R2 values and its variance.
m <- nrow(glanced)
r2 <- matrix(NA, nrow = m, ncol = 3, dimnames = list(seq_len(m), c("R^2", "Fisher trans F^2", "se()")))
# Fill arrays
for (i in seq_len(m)) {
r2[i, 1] <- if (!adjusted) sqrt(glanced$r.squared[i]) else sqrt(glanced$adj.r.squared[i])
r2[i, 2] <- 0.5 * log((r2[i, 1] + 1) / (1 - r2[i, 1]))
r2[i, 3] <- 1 / (glanced$nobs[i] - 3)
}
# Compute within, between and total variances following Rubin's rules with function pool.scalar().
fit <- pool.scalar(r2[, 2], r2[, 3])
# Make table with results.
qbar <- fit$qbar
table <- array(((exp(2 * qbar) - 1) / (1 + exp(2 * qbar)))^2,
dim = c(1, 4)
)
dimnames(table) <- if (!adjusted) {
list("R^2", c("est", "lo 95", "hi 95", "fmi"))
} else {
list("adj R^2", c("est", "lo 95", "hi 95", "fmi"))
}
table[, 2] <- ((exp(2 * (qbar - 1.96 * sqrt(fit$t))) - 1) / (1 + exp(2 * (qbar - 1.96 * sqrt(fit$t)))))^2
table[, 3] <- ((exp(2 * (qbar + 1.96 * sqrt(fit$t))) - 1) / (1 + exp(2 * (qbar + 1.96 * sqrt(fit$t)))))^2
table[, 4] <- fit$f
table
}
mice/R/pool.R 0000644 0001762 0000144 00000025123 14334522175 012475 0 ustar ligges users #' Combine estimates by pooling rules
#'
#' The \code{pool()} function combines the estimates from \code{m}
#' repeated complete data analyses. The typical sequence of steps to
#' perform a multiple imputation analysis is:
#' \enumerate{
#' \item Impute the missing data by the \code{mice()} function, resulting in
#' a multiple imputed data set (class \code{mids});
#' \item Fit the model of interest (scientific model) on each imputed data set
#' by the \code{with()} function, resulting an object of class \code{mira};
#' \item Pool the estimates from each model into a single set of estimates
#' and standard errors, resulting in an object of class \code{mipo};
#' \item Optionally, compare pooled estimates from different scientific models
#' by the \code{D1()} or \code{D3()} functions.
#' }
#' A common error is to reverse steps 2 and 3, i.e., to pool the
#' multiply-imputed data instead of the estimates. Doing so may severely bias
#' the estimates of scientific interest and yield incorrect statistical
#' intervals and p-values. The \code{pool()} function will detect
#' this case.
#'
#' @details
#' The \code{pool()} function averages the estimates of the complete
#' data model, computes the total variance over the repeated analyses
#' by Rubin's rules (Rubin, 1987, p. 76), and computes the following
#' diagnostic statistics per estimate:
#' \enumerate{
#' \item Relative increase in variance due to nonresponse {\code{r}};
#' \item Residual degrees of freedom for hypothesis testing {\code{df}};
#' \item Proportion of total variance due to missingness {\code{lambda}};
#' \item Fraction of missing information {\code{fmi}}.
#' }
#' The degrees of freedom calculation for the pooled estimates uses the
#' Barnard-Rubin adjustment for small samples (Barnard and Rubin, 1999).
#'
#' The \code{pool.syn()} function combines estimates by Reiter's partially
#' synthetic data pooling rules (Reiter, 2003). This combination rule
#' assumes that the data that is synthesised is completely observed.
#' Pooling differs from Rubin's method in the calculation of the total
#' variance and the degrees of freedom.
#'
#' Pooling requires the following input from each fitted model:
#' \enumerate{
#' \item the estimates of the model;
#' \item the standard error of each estimate;
#' \item the residual degrees of freedom of the model.
#' }
#' The \code{pool()} and \code{pool.syn()} functions rely on the
#' \code{broom::tidy} and \code{broom::glance} for extracting these
#' parameters.
#'
#' Since \code{mice 3.0+}, the \code{broom}
#' package takes care of filtering out the relevant parts of the
#' complete-data analysis. It may happen that you'll see the messages
#' like \code{Error: No tidy method for objects of class ...} or
#' \code{Error: No glance method for objects of class ...}. The message
#' means that your complete-data method used in \code{with(imp, ...)} has
#' no \code{tidy} or \code{glance} method defined in the \code{broom} package.
#'
#' The \code{broom.mixed} package contains \code{tidy} and \code{glance} methods
#' for mixed models. If you are using a mixed model, first run
#' \code{library(broom.mixed)} before calling \code{pool()}.
#'
#' If no \code{tidy} or \code{glance} methods are defined for your analysis
#' tabulate the \code{m} parameter estimates and their variance
#' estimates (the square of the standard errors) from the \code{m} fitted
#' models stored in \code{fit$analyses}. For each parameter, run
#' \code{\link{pool.scalar}} to obtain the pooled parameters estimate, its variance, the
#' degrees of freedom, the relative increase in variance and the fraction of missing
#' information.
#'
#' An alternative is to write your own \code{glance()} and \code{tidy()}
#' methods and add these to \code{broom} according to the specifications
#' given in \url{https://broom.tidymodels.org}.
#' In versions prior to \code{mice 3.0} pooling required that
#' \code{coef()} and \code{vcov()} methods were available for fitted
#' objects. \emph{This feature is no longer supported}. The reason is that
#' \code{vcov()} methods are inconsistent across packages, leading to
#' buggy behaviour of the \code{pool()} function.
#'
#' Since \code{mice 3.13.2} function \code{pool()} uses the robust
#' the standard error estimate for pooling when it can extract
#' \code{robust.se} from the \code{tidy()} object.
#'
#' @param object An object of class \code{mira} (produced by \code{with.mids()}
#' or \code{as.mira()}), or a \code{list} with model fits.
#' @param dfcom A positive number representing the degrees of freedom in the
#' complete-data analysis. Normally, this would be the number of independent
#' observation minus the number of fitted parameters. The default
#' (\code{dfcom = NULL}) extract this information in the following
#' order: 1) the component
#' \code{residual.df} returned by \code{glance()} if a \code{glance()}
#' function is found, 2) the result of \code{df.residual(} applied to
#' the first fitted model, and 3) as \code{999999}.
#' In the last case, the warning \code{"Large sample assumed"} is printed.
#' If the degrees of freedom is incorrect, specify the appropriate value
#' manually.
#' @param rule A string indicating the pooling rule. Currently supported are
#' \code{"rubin1987"} (default, for missing data) and \code{"reiter2003"}
#' (for synthetic data created from a complete data set).
#' @param custom.t A custom character string to be parsed as a calculation rule
#' for the total variance \code{t}. The custom rule can use the other calculated
#' pooling statistics where the dimensions must come from \code{.data$}. The
#' default \code{t} calculation would have the form
#' \code{".data$ubar + (1 + 1 / .data$m) * .data$b"}.
#' See examples for an example.
#' @return An object of class \code{mipo}, which stands for 'multiple imputation
#' pooled outcome'.
#' For rule \code{"reiter2003"} values for \code{lambda} and \code{fmi} are
#' set to `NA`, as these statistics do not apply for data synthesised from
#' fully observed data.
#' @seealso \code{\link{with.mids}}, \code{\link{as.mira}}, \code{\link{pool.scalar}},
#' \code{\link[broom:reexports]{glance}}, \code{\link[broom:reexports]{tidy}}
#' \url{https://github.com/amices/mice/issues/142},
#' \url{https://github.com/amices/mice/issues/274}
#' @references
#' Barnard, J. and Rubin, D.B. (1999). Small sample degrees of
#' freedom with multiple imputation. \emph{Biometrika}, 86, 948-955.
#'
#' Rubin, D.B. (1987). \emph{Multiple Imputation for Nonresponse in Surveys}.
#' New York: John Wiley and Sons.
#'
#' Reiter, J.P. (2003). Inference for Partially Synthetic,
#' Public Use Microdata Sets. \emph{Survey Methodology}, \bold{29}, 181-189.
#'
#' van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}: Multivariate
#' Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
#' Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
#' @examples
#' # impute missing data, analyse and pool using the classic MICE workflow
#' imp <- mice(nhanes, maxit = 2, m = 2)
#' fit <- with(data = imp, exp = lm(bmi ~ hyp + chl))
#' summary(pool(fit))
#'
#' # generate fully synthetic data, analyse and pool
#' imp <- mice(cars,
#' maxit = 2, m = 2,
#' where = matrix(TRUE, nrow(cars), ncol(cars))
#' )
#' fit <- with(data = imp, exp = lm(speed ~ dist))
#' summary(pool.syn(fit))
#'
#' # use a custom pooling rule for the total variance about the estimate
#' # e.g. use t = b + b/m instead of t = ubar + b + b/m
#' imp <- mice(nhanes, maxit = 2, m = 2)
#' fit <- with(data = imp, exp = lm(bmi ~ hyp + chl))
#' pool(fit, custom.t = ".data$b + .data$b / .data$m")
#'
#' @export
pool <- function(object, dfcom = NULL, rule = NULL, custom.t = NULL) {
call <- match.call()
if (!is.list(object)) stop("Argument 'object' not a list", call. = FALSE)
object <- as.mira(object)
m <- length(object$analyses)
if (m == 1) {
warning("Number of multiple imputations m = 1. No pooling done.")
return(getfit(object, 1))
}
dfcom <- get.dfcom(object, dfcom)
pooled <- pool.fitlist(getfit(object), dfcom = dfcom, rule = rule, custom.t = custom.t)
# mipo object
rr <- list(
call = call, m = m,
pooled = pooled,
glanced = get.glanced(object)
)
class(rr) <- c("mipo", "data.frame")
rr
}
pool.fitlist <- function(fitlist, dfcom = NULL,
rule = c("rubin1987", "reiter2003"), custom.t = NULL) {
# rubin1987: Rubin's rules for scalar estimates
# reiter2003: Reiter's rules for partially synthetic data
rule <- match.arg(rule)
w <- summary(fitlist, type = "tidy", exponentiate = FALSE)
grp <- intersect(names(w), c("parameter", "term", "contrast", "y.level", "component"))
# Note: group_by() changes the order of the terms, which is undesirable
# We convert any parameter terms to factor to preserve ordering
if ("term" %in% names(w)) w$term <- factor(w$term, levels = unique(w$term))
if ("contrast" %in% names(w)) w$contrast <- factor(w$contrast, levels = unique(w$contrast))
if ("y.level" %in% names(w)) w$y.level <- factor(w$y.level, levels = unique(w$y.level))
if ("component" %in% names(w)) w$component <- factor(w$component, levels = unique(w$component))
# https://github.com/amices/mice/issues/310
# Prefer using robust.se when tidy object contains it
if ("robust.se" %in% names(w)) w$std.error <- w$robust.se
if (rule == "rubin1987") {
pooled <- w %>%
group_by(!!!syms(grp)) %>%
summarize(
m = n(),
qbar = mean(.data$estimate),
ubar = mean(.data$std.error^2),
b = var(.data$estimate),
t = ifelse(is.null(custom.t),
.data$ubar + (1 + 1 / .data$m) * .data$b,
eval(parse(text = custom.t))
),
dfcom = dfcom,
df = barnard.rubin(.data$m, .data$b, .data$t, .data$dfcom),
riv = (1 + 1 / .data$m) * .data$b / .data$ubar,
lambda = (1 + 1 / .data$m) * .data$b / .data$t,
fmi = (.data$riv + 2 / (.data$df + 3)) / (.data$riv + 1)
)
}
if (rule == "reiter2003") {
pooled <- w %>%
group_by(!!!syms(grp)) %>%
summarize(
m = n(),
qbar = mean(.data$estimate),
ubar = mean(.data$std.error^2),
b = var(.data$estimate),
t = ifelse(is.null(custom.t),
.data$ubar + (1 / .data$m) * .data$b,
eval(parse(text = custom.t))
),
dfcom = dfcom,
df = (.data$m - 1) * (1 + (.data$ubar / (.data$b / .data$m)))^2,
riv = (1 + 1 / .data$m) * .data$b / .data$ubar,
lambda = NA_real_,
fmi = NA_real_
)
}
pooled <- data.frame(pooled)
names(pooled)[names(pooled) == "qbar"] <- "estimate"
pooled
}
#' @rdname pool
#' @export
pool.syn <- function(object, dfcom = NULL, rule = "reiter2003") {
pool(object = object, dfcom = dfcom, rule = rule)
}
mice/R/mice.impute.norm.boot.R 0000644 0001762 0000144 00000002235 14330031647 015651 0 ustar ligges users #' Imputation by linear regression, bootstrap method
#'
#' Imputes univariate missing data using linear regression with bootstrap
#'
#' @aliases mice.impute.norm.boot norm.boot
#' @inheritParams mice.impute.pmm
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @details
#' Draws a bootstrap sample from \code{x[ry,]} and \code{y[ry]}, calculates
#' regression weights and imputes with normal residuals.
#' @author Gerko Vink, Stef van Buuren, 2018
#' @references Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#' @family univariate imputation functions
#' @keywords datagen
#' @export
mice.impute.norm.boot <- function(y, ry, x, wy = NULL, ...) {
if (is.null(wy)) wy <- !ry
x <- cbind(1, as.matrix(x))
n1 <- sum(ry)
s <- sample(n1, n1, replace = TRUE)
ss <- s
dotxobs <- x[ry, , drop = FALSE][s, ]
dotyobs <- y[ry][s]
p <- estimice(dotxobs, dotyobs, ...)
sigma <- sqrt((sum(p$r^2)) / (n1 - ncol(x) - 1))
x[wy, ] %*% p$c + rnorm(sum(wy)) * sigma
}
mice/R/bwplot.mads.R 0000644 0001762 0000144 00000013010 14330031606 013734 0 ustar ligges users #' Box-and-whisker plot of amputed and non-amputed data
#'
#' Plotting method to investigate the relation between the data variables and
#' the amputed data. The function shows how the amputed values are related
#' to the variable values.
#'
#' @param x A \code{mads} (\code{\link{mads-class}}) object, typically created by
#' \code{\link{ampute}}.
#' @param data A string or vector of variable names that needs to be plotted. As
#' a default, all variables will be plotted.
#' @param which.pat A scalar or vector indicating which patterns need to be plotted.
#' As a default, all patterns are plotted.
#' @param standardized Logical. Whether the box-and-whisker plots need to be created
#' from standardized data or not. Default is TRUE.
#' @param descriptives Logical. Whether the mean, variance and n of the variables
#' need to be printed. This is useful to examine the effect of the amputation.
#' Default is TRUE.
#' @param layout A vector of two values indicating how the boxplots of one pattern
#' should be divided over the plot. For example, \code{c(2, 3)} indicates that the
#' boxplots of six variables need to be placed on 3 rows and 2 columns. Default
#' is 1 row and an amount of columns equal to #variables. Note that for more than
#' 6 variables, multiple plots will be created automatically.
#' @param \dots Not used, but for consistency with generic
#' @return A list containing the box-and-whisker plots. Note that a new pattern
#' will always be shown in a new plot.
#' @note The \code{mads} object contains all the information you need to
#' make any desired plots. Check \code{\link{mads-class}} or the vignette \emph{Multivariate
#' Amputation using Ampute} to understand the contents of class object \code{mads}.
#' @author Rianne Schouten, 2016
#' @seealso \code{\link{ampute}}, \code{\link{bwplot}}, \code{\link{Lattice}} for
#' an overview of the package, \code{\link{mads-class}}
#' @export
bwplot.mads <- function(x, data, which.pat = NULL, standardized = TRUE,
descriptives = TRUE, layout = NULL, ...) {
if (!is.mads(x)) {
stop("Object is not of class mads")
}
if (missing(data)) data <- NULL
yvar <- data
if (is.null(yvar)) {
varlist <- colnames(x$amp)
} else {
varlist <- yvar
}
if (is.null(which.pat)) {
pat <- nrow(x$patterns)
which.pat <- seq_len(pat)
} else {
pat <- length(which.pat)
}
formula <- as.formula(paste0(paste0(varlist, collapse = "+"), "~ factor(.amp)"))
data <- NULL
if (standardized) {
dat <- data.frame(scale(x$data))
} else {
dat <- x$data
}
if (is.null(layout)) {
if (ceiling(length(varlist) / 2) > 6) {
layout <- c(6, 1)
} else {
layout <- c(length(varlist), 1)
}
}
for (i in seq_len(pat)) {
can <- which(x$cand == which.pat[i])
mis <- matrix(NA, nrow = length(can), ncol = 2)
nc <- which(x$patterns[which.pat[i], ] == 0)
if (length(nc) > 1) {
mis[apply(is.na(x$amp[can, nc]), 1, all), 1] <- "Amp"
mis[is.na(mis[, 1]), 1] <- "Non-Amp"
} else if (length(nc) == 1) {
mis[is.na(x$amp[can, nc]), 1] <- "Amp"
mis[is.na(mis[, 1]), 1] <- "Non-Amp"
}
mis[, 2] <- rep.int(which.pat[i], length(can))
data <- rbind(data, cbind(mis, dat[can, ]))
}
colnames(data) <- c(".amp", ".pat", varlist)
p <- list()
vec1 <- c()
vec3 <- c()
for (i in seq_along(which.pat)) {
vec1[((i * 2) - 1):(i * 2)] <- rep.int(paste(which.pat[i]), 2)
}
vec3 <- paste("", varlist)
var <- length(varlist)
if (descriptives) {
desc <- array(NA,
dim = c(2 * length(which.pat), 4, var),
dimnames = list(
Pattern = vec1,
Descriptives = c("Amp", "Mean", "Var", "N"),
Variable = vec3
)
)
desc[, 1, ] <- rep.int(rep.int(c(1, 0), length(which.pat)), var)
for (i in seq_along(which.pat)) {
wp <- which.pat[i]
desc[(i * 2) - 1, 2, ] <-
round(vapply(varlist, function(x) {
mean(data[data$.pat == wp & data$.amp == "Amp", x])
}, numeric(1)), 5)
desc[(i * 2), 2, ] <-
round(vapply(varlist, function(x) {
mean(data[data$.pat == wp & data$.amp == "Non-Amp", x])
}, numeric(1)), 5)
desc[(i * 2) - 1, 3, ] <-
round(vapply(varlist, function(x) {
var(data[data$.pat == wp & data$.amp == "Amp", x])
}, numeric(1)), 5)
desc[(i * 2), 3, ] <-
round(vapply(varlist, function(x) {
var(data[data$.pat == wp & data$.amp == "Non-Amp", x])
}, numeric(1)), 5)
desc[(i * 2) - 1, 4, ] <-
vapply(varlist, function(x) {
length(data[data$.pat == wp & data$.amp == "Amp", x])
}, numeric(1))
desc[(i * 2), 4, ] <-
vapply(varlist, function(x) {
length(data[data$.pat == wp & data$.amp == "Non-Amp", x])
}, numeric(1))
}
p[["Descriptives"]] <- desc
}
theme <- list(
superpose.symbol = list(col = "black", pch = 1),
superpose.line = list(col = "black", lwd = 1),
box.dot = list(col = "black"),
box.rectangle = list(col = "black"),
box.umbrella = list(col = "black"),
box.symbol = list(col = "black"),
plot.symbol = list(col = "black", pch = 1),
plot.line = list(col = "black"),
strip.background = list(col = "grey95")
)
for (i in seq_len(pat)) {
p[[paste("Boxplot pattern", which.pat[i])]] <-
bwplot(
x = formula, data = data[data$.pat == which.pat[i], ],
multiple = TRUE, outer = TRUE, layout = layout,
ylab = "", par.settings = theme,
xlab = paste("Data distributions in pattern", which.pat[i])
)
}
p
}
mice/R/mice.impute.passive.R 0000644 0001762 0000144 00000002343 14330031647 015406 0 ustar ligges users #' Passive imputation
#'
#' Calculate new variable during imputation
#'
#' @param data A data frame
#' @param func A \code{formula} specifying the transformations on data
#' @return The result of applying \code{formula}
#' @details
#' Passive imputation is a special internal imputation function. Using this
#' facility, the user can specify, at any point in the \code{mice} Gibbs
#' sampling algorithm, a function on the imputed data. This is useful, for
#' example, to compute a cubic version of a variable, a transformation like
#' \code{Q = W/H^2} based on two variables, or a mean variable like
#' \code{(x_1+x_2+x_3)/3}. The so derived variables might be used in other
#' places in the imputation model. The function allows to dynamically derive
#' virtually any function of the imputed data at virtually any time.
#' @author Stef van Buuren, Karin Groothuis-Oudshoorn, 2000
#' @seealso \code{\link{mice}}
#' @references Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#' @keywords datagen
#' @export
mice.impute.passive <- function(data, func) {
model.frame(func, data)
}
mice/R/mice.impute.lasso.norm.R 0000644 0001762 0000144 00000004700 14330031647 016026 0 ustar ligges users #' Imputation by direct use of lasso linear regression
#'
#' Imputes univariate missing normal data using lasso linear regression with bootstrap.
#'
#' @aliases mice.impute.lasso.norm lasso.norm
#' @inheritParams mice.impute.norm.boot
#' @param nfolds The number of folds for the cross-validation of the lasso penalty.
#' The default is 10.
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @details
#' The method consists of the following steps:
#' \enumerate{
#' \item For a given y variable under imputation, draw a bootstrap version y*
#' with replacement from the observed cases \code{y[ry]}, and stores in x* the
#' corresponding values from \code{x[ry, ]}.
#' \item Fit a regularised (lasso) linear regression with y* as the outcome,
#' and x* as predictors.
#' A vector of regression coefficients bhat is obtained.
#' All of these coefficients are considered random draws from the imputation model
#' parameters posterior distribution.
#' Same of these coefficients will be shrunken to 0.
#' \item Draw the imputed values from the predictive distribution defined by
#' the original (non-bootstrap) data, bhat, and estimated error variance.
#' }
#' The method is based on the Direct Use of Regularized Regression (DURR) proposed by
#' Zhao & Long (2016) and Deng et al (2016).
#' @author Edoardo Costantini, 2021
#' @references
#'
#' Deng, Y., Chang, C., Ido, M. S., & Long, Q. (2016). Multiple imputation for
#' general missing data patterns in the presence of high-dimensional data.
#' Scientific reports, 6(1), 1-10.
#'
#' Zhao, Y., & Long, Q. (2016). Multiple imputation in the presence of
#' high-dimensional data. Statistical Methods in Medical Research, 25(5),
#' 2021-2035.
#'
#' @family univariate imputation functions
#' @keywords datagen
#' @export
mice.impute.lasso.norm <- function(y, ry, x, wy = NULL, nfolds = 10, ...) {
install.on.demand("glmnet", ...)
# Bootstrap sample
if (is.null(wy)) wy <- !ry
n1 <- sum(ry)
s <- sample(n1, n1, replace = TRUE)
x_glmnet <- cbind(1, x)
dotxobs <- x_glmnet[ry, , drop = FALSE][s, , drop = FALSE]
dotyobs <- y[ry][s]
# Train imputation model
cv_lasso <- glmnet::cv.glmnet(
x = dotxobs, y = dotyobs,
family = "gaussian",
nfolds = nfolds,
alpha = 1
)
# Obtain imputations
s2hat <- mean((predict(cv_lasso, dotxobs, s = "lambda.min") - dotyobs)^2)
as.vector(predict(cv_lasso, x_glmnet[wy, ], s = "lambda.min")) +
rnorm(sum(wy), 0, sqrt(s2hat))
}
mice/R/supports.transparent.R 0000644 0001762 0000144 00000001361 14330031606 015747 0 ustar ligges users #' Supports semi-transparent foreground colors?
#'
#' This function is used by \code{mdc()} to find out whether the current device
#' supports semi-transparent foreground colors.
#'
#' The function calls the function \code{dev.capabilities()} from the package
#' \code{grDevices}. The function return \code{FALSE} if the status of the
#' current device is unknown.
#'
#' @aliases supports.transparent transparent
#' @return \code{TRUE} or \code{FALSE}
#' @seealso \code{\link{mdc}} \code{\link{dev.capabilities}}
#' @keywords hplot
#' @examples
#'
#' supports.transparent()
#' @export
supports.transparent <- function() {
query <- grDevices::dev.capabilities("semiTransparency")$semiTransparency
if (is.na(query)) {
query <- FALSE
}
query
}
mice/R/complete.R 0000644 0001762 0000144 00000012520 14334522175 013331 0 ustar ligges users #' Extracts the completed data from a \code{mids} object
#'
#' Takes an object of class \code{mids}, fills in the missing data, and returns
#' the completed data in a specified format.
#'
#' @aliases complete
#' @param data An object of class \code{mids} as created by the function
#' \code{mice()}.
#' @param action A numeric vector or a keyword. Numeric
#' values between 1 and \code{data$m} return the data with
#' imputation number \code{action} filled in. The value of \code{action = 0}
#' return the original data, with missing values. \code{action} can
#' also be one of the following keywords: \code{"all"}, \code{"long"},
#' \code{"broad"} and \code{"repeated"}. See the Details section
#' for the interpretation.
#' The default is \code{action = 1L} returns the first imputed data set.
#' @param include A logical to indicate whether the original data with the missing
#' values should be included.
#' @param mild A logical indicating whether the return value should
#' always be an object of class \code{mild}. Setting \code{mild = TRUE}
#' overrides \code{action} keywords \code{"long"}, \code{"broad"}
#' and \code{"repeated"}. The default is \code{FALSE}.
#' @param \dots Additional arguments. Not used.
#' @return Complete data set with missing values replaced by imputations.
#' A \code{data.frame}, or a list of data frames of class \code{mild}.
#' @details
#' The argument \code{action} can be length-1 character, which is
#' matched to one of the following keywords:
#' \describe{
#' \item{\code{"all"}}{produces a \code{mild} object of imputed data sets. When
#' \code{include = TRUE}, then the original data are appended as the first list
#' element;}
#' \item{\code{"long"}}{ produces a data set where imputed data sets
#' are stacked vertically. The columns are added: 1) \code{.imp}, integer,
#' referring the imputation number, and 2) \code{.id}, character, the row
#' names of \code{data$data};}
#' \item{\code{"stacked"}}{ same as \code{"long"} but without the two
#' additional columns;}
#' \item{\code{"broad"}}{ produces a data set with where imputed data sets
#' are stacked horizontally. Columns are ordered as in the original data.
#' The imputation number is appended to each column name;}
#' \item{\code{"repeated"}}{ same as \code{"broad"}, but with
#' columns in a different order.}
#' }
#' @note
#' Technical note: \code{mice 3.7.5} renamed the \code{complete()} function
#' to \code{complete.mids()} and exported it as an S3 method of the
#' generic \code{tidyr::complete()}. Name clashes between
#' \code{mice::complete()} and \code{tidyr::complete()} should no
#' longer occur.
#' @seealso \code{\link{mice}}, \code{\link[=mids-class]{mids}}
#' @keywords manip
#' @examples
#'
#' # obtain first imputed data set
#' sum(is.na(nhanes2))
#' imp <- mice(nhanes2, print = FALSE, maxit = 1)
#' dat <- complete(imp)
#' sum(is.na(dat))
#'
#' # obtain stacked third and fifth imputation
#' dat <- complete(imp, c(3, 5))
#'
#' # obtain all datasets, with additional identifiers
#' head(complete(imp, "long"))
#'
#' # same, but now as list, mild object
#' dslist <- complete(imp, "all")
#' length(dslist)
#'
#' # same, but also include the original data
#' dslist <- complete(imp, "all", include = TRUE)
#' length(dslist)
#'
#' # select original + 3 + 5, store as mild
#' dslist <- complete(imp, c(0, 3, 5), mild = TRUE)
#' names(dslist)
#' @export
complete.mids <- function(data, action = 1L, include = FALSE,
mild = FALSE, ...) {
if (!is.mids(data)) stop("'data' not of class 'mids'")
m <- as.integer(data$m)
if (is.numeric(action)) {
action <- as.integer(action)
idx <- action[action >= 0L & action <= m]
if (include && all(idx != 0L)) idx <- c(0L, idx)
shape <- ifelse(mild, "mild", "stacked")
} else if (is.character(action)) {
if (include) idx <- 0L:m else idx <- 1L:m
shape <- match.arg(action, c("all", "long", "broad", "repeated", "stacked"))
shape <- ifelse(shape == "all" || mild, "mild", shape)
} else {
stop("'action' not recognized")
}
mylist <- vector("list", length = length(idx))
for (j in seq_along(idx)) {
mylist[[j]] <- single.complete(data$data, data$where, data$imp, idx[j])
}
if (shape == "stacked") {
return(bind_rows(mylist))
}
if (shape == "mild") {
names(mylist) <- as.character(idx)
class(mylist) <- c("mild", "list")
return(mylist)
}
if (shape == "long") {
cmp <- bind_rows(mylist)
cmp <- data.frame(
.imp = rep(idx, each = nrow(data$data)),
.id = rep.int(1L:nrow(data$data), length(idx)),
cmp
)
if (is.integer(attr(data$data, "row.names"))) {
row.names(cmp) <- seq_len(nrow(cmp))
} else {
row.names(cmp) <- as.character(seq_len(nrow(cmp)))
}
return(cmp)
}
# must be broad or repeated
cmp <- bind_cols(mylist)
names(cmp) <- paste(rep.int(names(data$data), m),
rep.int(idx, rep.int(ncol(data$data), length(idx))),
sep = "."
)
if (shape == "broad") {
return(cmp)
} else {
return(cmp[, order(rep.int(seq_len(ncol(data$data)), length(idx)))])
}
}
single.complete <- function(data, where, imp, ell) {
if (ell == 0L) {
return(data)
}
if (is.null(where)) {
where <- is.na(data)
}
idx <- seq_len(ncol(data))[apply(where, 2, any)]
for (j in idx) {
if (is.null(imp[[j]])) {
data[where[, j], j] <- NA
} else {
data[where[, j], j] <- imp[[j]][, ell]
}
}
data
}
mice/R/ampute.default.R 0000644 0001762 0000144 00000010670 14330031606 014431 0 ustar ligges users #' Default \code{patterns} in \code{ampute}
#'
#' This function creates a default pattern matrix for the multivariate
#' amputation function \code{ampute()}.
#'
#' @param n A scalar specifying the number of variables in the data.
#' @return A square matrix of size \code{n} where \code{0} indicates a variable
# should have missing values and \code{1} indicates a variable should remain
# complete. Each pattern has missingness on one variable only.
#' @seealso \code{\link{ampute}}, \code{\link{md.pattern}}
#' @author Rianne Schouten, 2016
#' @keywords internal
#' @export
ampute.default.patterns <- function(n) {
patterns.list <- lapply(
seq_len(n),
function(i) c(rep.int(1, i - 1), 0, rep.int(1, n - i))
)
do.call(rbind, patterns.list)
}
#' Default \code{freq} in \code{ampute}
#'
#' Defines the default relative frequency vector for the multivariate
#' amputation function \code{ampute}.
#'
#' @param patterns A matrix of size #patterns by #variables where \code{0} indicates
#' a variable should have missing values and \code{1} indicates a variable should
#' remain complete. Could be the result of \code{\link{ampute.default.patterns}}.
#' @return A vector of length #patterns containing the relative frequencies with
#' which the patterns should occur. An equal probability is given to each pattern.
#' @seealso \code{\link{ampute}}, \code{\link{ampute.default.patterns}}
#' @author Rianne Schouten, 2016
#' @keywords internal
#' @export
ampute.default.freq <- function(patterns) {
rep.int(1 / nrow(patterns), nrow(patterns))
}
#' Default \code{weights} in \code{ampute}
#'
#' Defines the default weights matrix for the multivariate amputation function
#' \code{ampute}.
#'
#' @param patterns A matrix of size #patterns by #variables where \code{0} indicates
#' a variable should have missing values and \code{1} indicates a variable should
#' remain complete. Could be the result of \code{\link{ampute.default.patterns}}.
#' @param mech A string specifying the missingness mechanism.
#' @return A matrix of size #patterns by #variables containing the weights that
#' will be used to calculate the weighted sum scores. Equal weights are given to
#' all variables. When mechanism is MAR, variables that will be amputed will be
#' weighted with \code{0}. If it is MNAR, variables that will be observed
#' will be weighted with \code{0}. If mechanism is MCAR, the weights matrix will
#' not be used. A default MAR matrix will be returned.
#' @seealso \code{\link{ampute}}, \code{\link{ampute.default.patterns}}
#' @author Rianne Schouten, 2016
#' @keywords internal
#' @export
ampute.default.weights <- function(patterns, mech) {
weights <- matrix(data = 1, nrow = nrow(patterns), ncol = ncol(patterns))
if (mech != "MNAR") {
weights <- matrix(data = 1, nrow = nrow(patterns), ncol = ncol(patterns))
weights[patterns == 0] <- 0
} else {
weights <- matrix(data = 0, nrow = nrow(patterns), ncol = ncol(patterns))
weights[patterns == 0] <- 1
}
weights
}
#' Default \code{type} in \code{ampute()}
#'
#' Defines the default type vector for the multivariate amputation function
#' \code{ampute}.
#'
#' @param patterns A matrix of size #patterns by #variables where 0 indicates a
#' variable should have missing values and 1 indicates a variable should remain
#' complete. Could be the result of \code{\link{ampute.default.patterns}}.
#' @return A string vector of length #patterns containing the missingness types.
#' Each pattern will be amputed with a "RIGHT" missingness.
#' @seealso \code{\link{ampute}}, \code{\link{ampute.default.patterns}}
#' @author Rianne Schouten, 2016
#' @keywords internal
#' @export
ampute.default.type <- function(patterns) {
rep.int("RIGHT", nrow(patterns))
}
#' Default \code{odds} in \code{ampute()}
#'
#' Defines the default odds matrix for the multivariate amputation function
#' \code{ampute}.
#'
#' @param patterns A matrix of size #patterns by #variables where 0 indicates a
#' variable should have missing values and 1 indicates a variable should remain
#' complete. Could be the result of \code{\link{ampute.default.patterns}}.
#' @return A matrix where #rows equals #patterns. Default is 4 quantiles with odds
#' values 1, 2, 3 and 4, for each pattern, imitating a RIGHT type of missingness.
#' @seealso \code{\link{ampute}}, \code{\link{ampute.default.patterns}}
#' @author Rianne Schouten, 2016
#' @keywords internal
#' @export
ampute.default.odds <- function(patterns) {
matrix(seq_len(4), nrow = nrow(patterns), ncol = 4, byrow = TRUE)
}
mice/R/bwplot.R 0000644 0001762 0000144 00000027724 14330031647 013037 0 ustar ligges users #' Box-and-whisker plot of observed and imputed data
#'
#' Plotting methods for imputed data using \pkg{lattice}. \code{bwplot}
#' produces box-and-whisker plots. The function
#' automatically separates the observed and imputed data. The
#' functions extend the usual features of \pkg{lattice}.
#'
#' The argument \code{na.groups} may be used to specify (combinations of)
#' missingness in any of the variables. The argument \code{groups} can be used
#' to specify groups based on the variable values themselves. Only one of both
#' may be active at the same time. When both are specified, \code{na.groups}
#' takes precedence over \code{groups}.
#'
#' Use the \code{subset} and \code{na.groups} together to plots parts of the
#' data. For example, select the first imputed data set by by
#' \code{subset=.imp==1}.
#'
#' Graphical parameters like \code{col}, \code{pch} and \code{cex} can be
#' specified in the arguments list to alter the plotting symbols. If
#' \code{length(col)==2}, the color specification to define the observed and
#' missing groups. \code{col[1]} is the color of the 'observed' data,
#' \code{col[2]} is the color of the missing or imputed data. A convenient color
#' choice is \code{col=mdc(1:2)}, a transparent blue color for the observed
#' data, and a transparent red color for the imputed data. A good choice is
#' \code{col=mdc(1:2), pch=20, cex=1.5}. These choices can be set for the
#' duration of the session by running \code{mice.theme()}.
#'
#' @aliases bwplot
#' @param x A \code{mids} object, typically created by \code{mice()} or
#' \code{mice.mids()}.
#' @param data Formula that selects the data to be plotted. This argument
#' follows the \pkg{lattice} rules for \emph{formulas}, describing the primary
#' variables (used for the per-panel display) and the optional conditioning
#' variables (which define the subsets plotted in different panels) to be used
#' in the plot.
#'
#' The formula is evaluated on the complete data set in the \code{long} form.
#' Legal variable names for the formula include \code{names(x$data)} plus the
#' two administrative factors \code{.imp} and \code{.id}.
#'
#' \bold{Extended formula interface:} The primary variable terms (both the LHS
#' \code{y} and RHS \code{x}) may consist of multiple terms separated by a
#' \sQuote{+} sign, e.g., \code{y1 + y2 ~ x | a * b}. This formula would be
#' taken to mean that the user wants to plot both \code{y1 ~ x | a * b} and
#' \code{y2 ~ x | a * b}, but with the \code{y1 ~ x} and \code{y2 ~ x} in
#' \emph{separate panels}. This behavior differs from standard \pkg{lattice}.
#' \emph{Only combine terms of the same type}, i.e. only factors or only
#' numerical variables. Mixing numerical and categorical data occasionally
#' produces odds labeling of vertical axis.
#'
#' For convenience, in \code{stripplot()} and \code{bwplot} the formula
#' \code{y~.imp} may be abbreviated as \code{y}. This applies only to a single
#' \code{y}, and does not (yet) work for \code{y1+y2~.imp}.
#'
#' @param na.groups An expression evaluating to a logical vector indicating
#' which two groups are distinguished (e.g. using different colors) in the
#' display. The environment in which this expression is evaluated in the
#' response indicator \code{is.na(x$data)}.
#'
#' The default \code{na.group = NULL} contrasts the observed and missing data
#' in the LHS \code{y} variable of the display, i.e. groups created by
#' \code{is.na(y)}. The expression \code{y} creates the groups according to
#' \code{is.na(y)}. The expression \code{y1 & y2} creates groups by
#' \code{is.na(y1) & is.na(y2)}, and \code{y1 | y2} creates groups as
#' \code{is.na(y1) | is.na(y2)}, and so on.
#' @param groups This is the usual \code{groups} arguments in \pkg{lattice}. It
#' differs from \code{na.groups} because it evaluates in the completed data
#' \code{data.frame(complete(x, "long", inc=TRUE))} (as usual), whereas
#' \code{na.groups} evaluates in the response indicator. See
#' \code{\link{xyplot}} for more details. When both \code{na.groups} and
#' \code{groups} are specified, \code{na.groups} takes precedence, and
#' \code{groups} is ignored.
#' @param theme A named list containing the graphical parameters. The default
#' function \code{mice.theme} produces a short list of default colors, line
#' width, and so on. The extensive list may be obtained from
#' \code{trellis.par.get()}. Global graphical parameters like \code{col} or
#' \code{cex} in high-level calls are still honored, so first experiment with
#' the global parameters. Many setting consists of a pair. For example,
#' \code{mice.theme} defines two symbol colors. The first is for the observed
#' data, the second for the imputed data. The theme settings only exist during
#' the call, and do not affect the trellis graphical parameters.
#' @param mayreplicate A logical indicating whether color, line widths, and so
#' on, may be replicated. The graphical functions attempt to choose
#' "intelligent" graphical parameters. For example, the same color can be
#' replicated for different element, e.g. use all reds for the imputed data.
#' Replication may be switched off by setting the flag to \code{FALSE}, in order
#' to allow the user to gain full control.
#' @param as.table See \code{\link[lattice:xyplot]{xyplot}}.
#' @param outer See \code{\link[lattice:xyplot]{xyplot}}.
#' @param allow.multiple See \code{\link[lattice:xyplot]{xyplot}}.
#' @param drop.unused.levels See \code{\link[lattice:xyplot]{xyplot}}.
#' @param subscripts See \code{\link[lattice:xyplot]{xyplot}}.
#' @param subset See \code{\link[lattice:xyplot]{xyplot}}.
#' @param \dots Further arguments, usually not directly processed by the
#' high-level functions documented here, but instead passed on to other
#' functions.
#' @return The high-level functions documented here, as well as other high-level
#' Lattice functions, return an object of class \code{"trellis"}. The
#' \code{\link[lattice:update.trellis]{update}} method can be used to
#' subsequently update components of the object, and the
#' \code{\link[lattice:print.trellis]{print}} method (usually called by default)
#' will plot it on an appropriate plotting device.
#' @note The first two arguments (\code{x} and \code{data}) are reversed
#' compared to the standard Trellis syntax implemented in \pkg{lattice}. This
#' reversal was necessary in order to benefit from automatic method dispatch.
#'
#' In \pkg{mice} the argument \code{x} is always a \code{mids} object, whereas
#' in \pkg{lattice} the argument \code{x} is always a formula.
#'
#' In \pkg{mice} the argument \code{data} is always a formula object, whereas in
#' \pkg{lattice} the argument \code{data} is usually a data frame.
#'
#' All other arguments have identical interpretation.
#'
#' @author Stef van Buuren
#' @seealso \code{\link{mice}}, \code{\link{xyplot}}, \code{\link{densityplot}},
#' \code{\link{stripplot}}, \code{\link{lattice}} for an overview of the
#' package, as well as \code{\link[lattice:xyplot]{bwplot}},
#' \code{\link[lattice:panel.xyplot]{panel.bwplot}},
#' \code{\link[lattice:print.trellis]{print.trellis}},
#' \code{\link[lattice:trellis.par.get]{trellis.par.set}}
#' @references Sarkar, Deepayan (2008) \emph{Lattice: Multivariate Data
#' Visualization with R}, Springer.
#'
#' van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}: Multivariate
#' Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
#' Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
#' @keywords hplot
#' @examples
#'
#' imp <- mice(boys, maxit = 1)
#'
#' ### box-and-whisker plot per imputation of all numerical variables
#' bwplot(imp)
#'
#' ### tv (testicular volume), conditional on region
#' bwplot(imp, tv ~ .imp | reg)
#'
#' ### same data, organized in a different way
#' bwplot(imp, tv ~ reg | .imp, theme = list())
#' @export
bwplot.mids <- function(x,
data,
na.groups = NULL,
groups = NULL,
as.table = TRUE,
theme = mice.theme(),
mayreplicate = TRUE,
allow.multiple = TRUE,
outer = TRUE,
drop.unused.levels = lattice::lattice.getOption("drop.unused.levels"),
...,
subscripts = TRUE,
subset = TRUE) {
call <- match.call()
if (!is.mids(x)) stop("Argument 'x' must be a 'mids' object")
## unpack data and response indicator
cd <- data.frame(complete(x, "long", include = TRUE))
cd$.imp <- as.factor(cd$.imp)
r <- as.data.frame(is.na(x$data))
## evaluate na.group in response indicator
nagp <- eval(expr = substitute(na.groups), envir = r, enclos = parent.frame())
if (is.expression(nagp)) nagp <- eval(expr = nagp, envir = r, enclos = parent.frame())
## evaluate groups in imputed data
ngp <- eval(expr = substitute(groups), envir = cd, enclos = parent.frame())
if (is.expression(ngp)) ngp <- eval(expr = ngp, envir = cd, enclos = parent.frame())
groups <- ngp
## evaluate subset in imputed data
ss <- eval(expr = substitute(subset), envir = cd, enclos = parent.frame())
if (is.expression(ss)) ss <- eval(expr = ss, envir = cd, enclos = parent.frame())
subset <- ss
## evaluate further arguments before parsing
dots <- list(...)
args <- list(
allow.multiple = allow.multiple,
outer = outer,
drop.unused.levels = drop.unused.levels,
subscripts = subscripts,
as.table = as.table
)
## create formula if not given (in call$data !)
vnames <- names(cd)[-seq_len(2)]
allfactors <- unlist(lapply(cd, is.factor))[-seq_len(2)]
if (missing(data)) {
vnames <- vnames[!allfactors]
formula <- as.formula(paste(paste(vnames, collapse = "+", sep = ""), "~.imp", sep = ""))
} else {
## pad abbreviated formula
abbrev <- length(grep("~", call$data)) == 0
if (abbrev) {
if (length(call$data) > 1) {
stop("Cannot pad extended formula.")
} else {
formula <- as.formula(paste(call$data, "~.imp", sep = ""))
}
} else {
formula <- data
}
}
## determine the y-variables
form <- lattice::latticeParseFormula(
model = formula, data = cd, subset = subset,
groups = groups, multiple = allow.multiple,
outer = outer, subscripts = TRUE,
drop = drop.unused.levels
)
ynames <- unlist(lapply(strsplit(form$left.name, " \\+ "), rm.whitespace))
xnames <- unlist(lapply(strsplit(form$right.name, " \\+ "), rm.whitespace))
## groups is not useful in bwplot
## in order to force subgroup analysis,
## make the observed data in .imp>0 missing data
nona <- is.null(call$na.groups)
if (!is.null(call$groups) & nona) {
gp <- call$groups
} else {
if (nona) {
for (i in seq_along(ynames)) {
yvar <- ynames[i]
select <- cd$.imp != 0 & !r[, yvar]
cd[select, yvar] <- NA
}
} else {
for (i in seq_along(ynames)) {
yvar <- ynames[i]
select <- cd$.imp != 0 & !nagp
cd[select, yvar] <- NA
}
}
}
## replicate color 2 if .imp is part of xnames
mustreplicate <- !is.na(match(".imp", xnames)) & mayreplicate
if (mustreplicate) {
theme$box.dot$col <- rep(theme$box.dot$col[1:2], c(1, x$m))
theme$box.rectangle$col <- rep(theme$box.rectangle$col[1:2], c(1, x$m))
theme$box.umbrella$col <- rep(theme$box.rectangle$col[1:2], c(1, x$m))
theme$plot.symbol$col <- mdc(3)
theme$plot.symbol$pch <- 1
}
## change axis defaults of extended formula interface
if (is.null(call$xlab) && !is.na(match(".imp", xnames))) {
dots$xlab <- ""
if (length(xnames) == 1) dots$xlab <- "Imputation number"
}
if (is.null(call$ylab)) {
args$ylab <- ""
if (length(ynames) == 1) args$ylab <- ynames
}
if (is.null(call$scales)) {
args$scales <- list()
if (length(ynames) > 1) {
args$scales <- list(x = list(relation = "free"), y = list(relation = "free"))
}
}
## ready
args <- c(
x = formula, data = list(cd),
groups = list(groups),
args, dots, subset = call$subset
)
## go
tp <- do.call("bwplot", args)
update(tp, par.settings = theme)
}
mice/R/sampler.R 0000644 0001762 0000144 00000017001 14347612445 013167 0 ustar ligges users # The sampler controls the actual Gibbs sampling iteration scheme.
# This function is called by mice and mice.mids
sampler <- function(data, m, ignore, where, imp, blocks, method,
visitSequence, predictorMatrix, formulas, blots,
post, fromto, printFlag, ...) {
from <- fromto[1]
to <- fromto[2]
maxit <- to - from + 1
r <- !is.na(data)
# set up array for convergence checking
chainMean <- chainVar <- initialize.chain(blocks, maxit, m)
## THE MAIN LOOP: GIBBS SAMPLER ##
if (maxit < 1) iteration <- 0
if (maxit >= 1) {
if (printFlag) {
cat("\n iter imp variable")
}
for (k in from:to) {
# begin k loop : main iteration loop
iteration <- k
for (i in seq_len(m)) {
# begin i loop: repeated imputation loop
if (printFlag) {
cat("\n ", iteration, " ", i)
}
# prepare the i'th imputation
# do not overwrite any observed data
for (h in visitSequence) {
for (j in blocks[[h]]) {
y <- data[, j]
ry <- r[, j]
wy <- where[, j]
data[(!ry) & wy, j] <- imp[[j]][(!ry)[wy], i]
}
}
# impute block-by-block
for (h in visitSequence) {
ct <- attr(blocks, "calltype")
calltype <- ifelse(length(ct) == 1, ct[1], ct[h])
b <- blocks[[h]]
if (calltype == "formula") ff <- formulas[[h]] else ff <- NULL
type <- predictorMatrix[h, ]
user <- blots[[h]]
# univariate/multivariate logic
theMethod <- method[h]
empt <- theMethod == ""
univ <- !empt && !is.passive(theMethod) &&
!handles.format(paste0("mice.impute.", theMethod))
mult <- !empt && !is.passive(theMethod) &&
handles.format(paste0("mice.impute.", theMethod))
pass <- !empt && is.passive(theMethod) && length(blocks[[h]]) == 1
if (printFlag & !empt) cat(" ", b)
## store current state
oldstate <- get("state", pos = parent.frame())
newstate <- list(
it = k, im = i,
dep = h,
meth = theMethod,
log = oldstate$log
)
assign("state", newstate, pos = parent.frame(), inherits = TRUE)
# (repeated) univariate imputation - type method
if (univ) {
for (j in b) {
imp[[j]][, i] <-
sampler.univ(
data = data, r = r, where = where,
type = type, formula = ff,
method = theMethod,
yname = j, k = k,
calltype = calltype,
user = user, ignore = ignore,
...
)
data[(!r[, j]) & where[, j], j] <-
imp[[j]][(!r[, j])[where[, j]], i]
# optional post-processing
cmd <- post[j]
if (cmd != "") {
eval(parse(text = cmd))
data[(!r[, j]) & where[, j], j] <-
imp[[j]][(!r[, j])[where[, j]], i]
}
}
}
# multivariate imputation - type and formula
if (mult) {
mis <- !r
mis[, setdiff(colnames(data), b)] <- FALSE
data[mis] <- NA
fm <- paste("mice.impute", theMethod, sep = ".")
if (calltype == "formula") {
imputes <- do.call(fm, args = list(
data = data,
formula = ff, ...
))
} else if (calltype == "type") {
imputes <- do.call(fm, args = list(
data = data,
type = type, ...
))
} else {
stop("Cannot call function of type ", calltype,
call. = FALSE
)
}
if (is.null(imputes)) {
stop("No imputations from ", theMethod,
h,
call. = FALSE
)
}
for (j in names(imputes)) {
imp[[j]][, i] <- imputes[[j]]
data[!r[, j], j] <- imp[[j]][, i]
}
}
# passive imputation
# applies to all rows, so no ignore needed
if (pass) {
for (j in b) {
wy <- where[, j]
ry <- r[, j]
imp[[j]][, i] <- model.frame(as.formula(theMethod), data[wy, ],
na.action = na.pass
)
data[(!ry) & wy, j] <- imp[[j]][(!ry)[wy], i]
}
}
} # end h loop (blocks)
} # end i loop (imputation number)
# store means and sd of m imputes
k2 <- k - from + 1L
if (length(visitSequence) > 0L) {
for (h in visitSequence) {
for (j in blocks[[h]]) {
if (!is.factor(data[, j])) {
chainVar[j, k2, ] <- apply(imp[[j]], 2L, var, na.rm = TRUE)
chainMean[j, k2, ] <- colMeans(as.matrix(imp[[j]]), na.rm = TRUE)
}
if (is.factor(data[, j])) {
for (mm in seq_len(m)) {
nc <- as.integer(factor(imp[[j]][, mm], levels = levels(data[, j])))
chainVar[j, k2, mm] <- var(nc, na.rm = TRUE)
chainMean[j, k2, mm] <- mean(nc, na.rm = TRUE)
}
}
}
}
}
} # end main iteration
if (printFlag) {
r <- get("loggedEvents", parent.frame(1))
ridge.used <- any(grepl("A ridge penalty", r$out))
if (ridge.used) {
cat("\n * Please inspect the loggedEvents \n")
} else {
cat("\n")
}
}
}
list(iteration = maxit, imp = imp, chainMean = chainMean, chainVar = chainVar)
}
sampler.univ <- function(data, r, where, type, formula, method, yname, k,
calltype = "type", user, ignore, ...) {
j <- yname[1L]
if (calltype == "type") {
vars <- colnames(data)[type != 0]
pred <- setdiff(vars, j)
if (length(pred) > 0L) {
formula <- reformulate(pred, response = j)
formula <- update(formula, ". ~ . ")
} else {
formula <- as.formula(paste0(j, " ~ 1"))
}
}
if (calltype == "formula") {
# move terms other than j from lhs to rhs
ymove <- setdiff(lhs(formula), j)
formula <- update(formula, paste(j, " ~ . "))
if (length(ymove) > 0L) {
formula <- update(formula, paste("~ . + ", paste(ymove, collapse = "+")))
}
}
# get the model matrix
x <- obtain.design(data, formula)
# expand type vector to model matrix, remove intercept
if (calltype == "type") {
type <- type[labels(terms(formula))][attr(x, "assign")]
x <- x[, -1L, drop = FALSE]
names(type) <- colnames(x)
}
if (calltype == "formula") {
x <- x[, -1L, drop = FALSE]
type <- rep(1L, length = ncol(x))
names(type) <- colnames(x)
}
# define y, ry and wy
y <- data[, j]
ry <- complete.cases(x, y) & r[, j] & !ignore
wy <- complete.cases(x) & where[, j]
# nothing to impute
if (all(!wy)) {
return(numeric(0))
}
cc <- wy[where[, j]]
if (k == 1L) check.df(x, y, ry)
# remove linear dependencies
keep <- remove.lindep(x, y, ry, ...)
x <- x[, keep, drop = FALSE]
type <- type[keep]
if (ncol(x) != length(type)) {
stop("Internal error: length(type) != number of predictors")
}
# here we go
f <- paste("mice.impute", method, sep = ".")
imputes <- data[wy, j]
imputes[!cc] <- NA
args <- c(list(y = y, ry = ry, x = x, wy = wy, type = type), user, list(...))
imputes[cc] <- do.call(f, args = args)
imputes
}
mice/R/tidiers.R 0000644 0001762 0000144 00000004550 14330035710 013156 0 ustar ligges users #' @importFrom generics tidy
#' @export
generics::tidy
#' @importFrom generics glance
#' @export
generics::glance
#' Tidy method to extract results from a `mipo` object
#'
#' @param x An object of class \code{mipo}
#' @param conf.int Logical. Should confidence intervals be returned?
#' @param conf.level Confidence level for intervals. Defaults to .95
#' @param ... extra arguments (not used)
#' @export
#' @keywords internal
#' @return A dataframe withh these columns:
#' \itemize{
#' \item term
#' \item estimate
#' \item ubar
#' \item b
#' \item t
#' \item dfcom
#' \item df
#' \item riv
#' \item lambda
#' \item fmi
#' \item p.value
#' \item conf.low (if called with conf.int = TRUE)
#' \item conf.high (if called with conf.int = TRUE)
#' }
tidy.mipo <- function(x, conf.int = FALSE, conf.level = .95, ...) {
out <- summary(x,
type = "all",
conf.int = conf.int,
conf.level = conf.level,
...
)
if ("term" %in% names(out)) out$term <- as.character(out$term)
if ("contrast" %in% names(out)) out$contrast <- as.character(out$contrast)
# needed for broom <= 0.5.6
# rename variables if present
idx <- grepl("%", names(out))
names(out)[idx] <- c("conf.low", "conf.high")
idx <- names(out) == "t"
names(out)[idx] <- "statistic"
# order columns
cols_a <- c(
"term", "estimate", "std.error", "statistic", "p.value",
"conf.low", "conf.high"
)
cols_a <- base::intersect(cols_a, colnames(out))
cols_b <- sort(base::setdiff(colnames(out), cols_a))
out[, c(cols_a, cols_b)]
}
#' Glance method to extract information from a `mipo` object
#'
#' @param x An object with multiply-imputed models from `mice` (class: `mipo`)
#' @param ... extra arguments (not used)
#' @return a dataframe with one row and the following columns:
#' \itemize{
#' \item nimp
#' \item nobs
#' }
#' @note If x contains `lm` models, R2 and Adj.R2 are included in the output
#' @export
#' @keywords internal
#' @family tidiers
glance.mipo <- function(x, ...) {
out <- data.frame(nimp = nrow(x$glanced))
out$nobs <- tryCatch(x$glanced$nobs[1],
error = function(e) NULL
)
# R2 in lm models
out$r.squared <- tryCatch(pool.r.squared(x, adjusted = FALSE)[1],
error = function(e) NULL
)
out$adj.r.squared <- tryCatch(pool.r.squared(x, adjusted = TRUE)[1],
error = function(e) NULL
)
out
}
mice/R/initialize.chain.R 0000644 0001762 0000144 00000000356 14330031606 014735 0 ustar ligges users initialize.chain <- function(blocks, maxit, m) {
vars <- unique(unlist(blocks))
chain <- array(NA, dim = c(length(vars), maxit, m))
dimnames(chain) <- list(
vars,
seq_len(maxit),
paste("Chain", seq_len(m))
)
chain
}
mice/R/mice.impute.2l.lmer.R 0000644 0001762 0000144 00000014243 14334522175 015216 0 ustar ligges users #' Imputation by a two-level normal model using \code{lmer}
#'
#' Imputes univariate systematically and sporadically missing data using a
#' two-level normal model using \code{lme4::lmer()}.
#'
#' Data are missing systematically if they have not been measured, e.g., in the
#' case where we combine data from different sources. Data are missing sporadically
#' if they have been partially observed.
#'
#' While the method is fully Bayesian, it may fix parameters of the
#' variance-covariance matrix or the random effects to their estimated
#' value in cases where creating draws from the posterior is not
#' possible. The procedure throws a warning when this happens.
#'
#' If \code{lme4::lmer()} fails, the procedure prints the warning
#' \code{"lmer does not run. Simplify imputation model"} and returns the
#' current imputation. If that happens we see flat lines in the
#' trace line plots. Thus, the appearance of flat trace lines should be taken
#' as an additional alert to a problem with imputation model fitting.
#' @name mice.impute.2l.lmer
#' @inheritParams mice.impute.pmm
#' @param type Vector of length \code{ncol(x)} identifying random and class
#' variables. Random variables are identified by a '2'. The class variable
#' (only one is allowed) is coded as '-2'. Fixed effects are indicated by
#' a '1'.
#' @param intercept Logical determining whether the intercept is automatically
#' added.
#' @param \dots Arguments passed down to \code{lmer}
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @author Shahab Jolani, 2017
#' @references
#' Jolani S. (2017) Hierarchical imputation of systematically and
#' sporadically missing data: An approximate Bayesian approach using
#' chained equations. Forthcoming.
#'
#' Jolani S., Debray T.P.A., Koffijberg H., van Buuren S., Moons K.G.M. (2015).
#' Imputation of systematically missing predictors in an individual
#' participant data meta-analysis: a generalized approach using MICE.
#' \emph{Statistics in Medicine}, 34:1841-1863.
#'
#' Van Buuren, S. (2011) Multiple imputation of multilevel data. In Hox, J.J.
#' and and Roberts, J.K. (Eds.), \emph{The Handbook of Advanced Multilevel
#' Analysis}, Chapter 10, pp. 173--196. Milton Park, UK: Routledge.
#' @family univariate-2l
#' @keywords datagen
#' @export
mice.impute.2l.lmer <- function(y, ry, x, type, wy = NULL, intercept = TRUE, ...) {
install.on.demand("lme4", ...)
if (is.null(wy)) wy <- !ry
if (intercept) {
x <- cbind(1, as.matrix(x))
type <- c(2, type)
names(type)[1] <- colnames(x)[1] <- "(Intercept)"
}
clust <- names(type[type == -2])
rande <- names(type[type == 2])
fixe <- names(type[type > 0])
lev <- unique(x[, clust])
X <- x[, fixe, drop = FALSE]
Z <- x[, rande, drop = FALSE]
xobs <- x[ry, , drop = FALSE]
yobs <- y[ry]
Xobs <- X[ry, , drop = FALSE]
Zobs <- Z[ry, , drop = FALSE]
# create formula
fr <- ifelse(length(rande) > 1,
paste("+ ( 1 +", paste(rande[-1L], collapse = "+")),
"+ ( 1 "
)
randmodel <- paste(
"yobs ~ ", paste(fixe[-1L], collapse = "+"),
fr, "|", clust, ")"
)
suppressWarnings(fit <- try(
lme4::lmer(formula(randmodel),
data = data.frame(yobs, xobs),
...
),
silent = TRUE
))
if (inherits(fit, "try-error")) {
warning("lmer does not run. Simplify imputation model")
return(y[wy])
}
# taken from lme4
sigma <- function(object, ...) {
dc <- object@devcomp
dd <- dc$dims
if (dd[["useSc"]]) {
dc$cmp[[if (dd[["REML"]]) {
"sigmaREML"
} else {
"sigmaML"
}]]
} else {
1
}
}
# draw sigma*
sigmahat <- sigma(fit)
df <- nrow(fit@frame) - length(fit@beta)
sigma2star <- df * sigmahat^2 / rchisq(1, df)
# draw beta*
beta <- lme4::fixef(fit)
RX <- lme4::getME(fit, "RX")
# cov-matrix, i.e., vcov(fit)
covmat <- sigma2star * chol2inv(RX)
rv <- t(chol(covmat))
beta.star <- beta + rv %*% rnorm(ncol(rv))
# draw psi*
# applying the standard Wishart prior
rancoef <- as.matrix(lme4::ranef(fit)[[1]])
lambda <- t(rancoef) %*% rancoef
df.psi <- nrow(rancoef)
temp.psi.star <- stats::rWishart(1, df.psi, diag(nrow(lambda)))[, , 1]
temp <- MASS::ginv(lambda)
ev <- eigen(temp)
if (sum(ev$values > 0) == length(ev$values)) {
deco <- ev$vectors %*% diag(sqrt(ev$values), nrow = length(ev$values))
psi.star <- MASS::ginv(deco %*% temp.psi.star %*% t(deco))
} else {
try(temp.svd <- svd(lambda))
if (!inherits(temp.svd, "try-error")) {
deco <- temp.svd$u %*% diag(sqrt(temp.svd$d), nrow = length(temp.svd$d))
psi.star <- MASS::ginv(deco %*% temp.psi.star %*% t(deco))
} else {
psi.star <- temp
warning("psi fixed to estimate")
}
}
# Calculate myi, vyi and drawing bi per cluster
for (jj in lev) {
if (jj %in% unique(xobs[, clust])) {
Xi <- Xobs[xobs[, clust] == jj, ]
Zi <- as.matrix(Zobs[xobs[, clust] == jj, ])
yi <- yobs[xobs[, clust] == jj]
sigma2 <- diag(sigma2star, nrow = nrow(Zi))
Mi <- psi.star %*% t(Zi) %*% MASS::ginv(Zi %*% psi.star %*% t(Zi) + sigma2)
myi <- Mi %*% (yi - Xi %*% beta.star)
vyi <- psi.star - Mi %*% Zi %*% psi.star
} else {
myi <- matrix(0, nrow = nrow(psi.star), ncol = 1)
vyi <- psi.star
}
vyi <- vyi - upper.tri(vyi) * vyi + t(lower.tri(vyi) * vyi)
# generating bi.star using eigenvalues
deco1 <- eigen(vyi)
if (sum(deco1$values > 0) == length(deco1$values)) {
A <- deco1$vectors %*% sqrt(diag(deco1$values, nrow = length(deco1$values)))
bi.star <- myi + A %*% rnorm(length(myi))
} else {
# generating bi.star using svd
try(deco1 <- svd(vyi))
if (!inherits(deco1, "try-error")) {
A <- deco1$u %*% sqrt(diag(deco1$d, nrow = length(deco1$d)))
bi.star <- myi + A %*% rnorm(length(myi))
} else {
bi.star <- myi
warning("b_", jj, " fixed to estimate")
}
}
# imputation
y[wy & x[, clust] == jj] <- as.vector(
as.matrix(X[wy & x[, clust] == jj, , drop = FALSE]) %*% beta.star +
as.matrix(Z[wy & x[, clust] == jj, , drop = FALSE]) %*% as.matrix(bi.star) +
rnorm(sum(wy & x[, clust] == jj)) * sqrt(sigma2star)
)
}
y[wy]
}
mice/R/mice.impute.rf.R 0000644 0001762 0000144 00000011257 14436640532 014355 0 ustar ligges users #' Imputation by random forests
#'
#' Imputes univariate missing data using random forests.
#'
#' @aliases mice.impute.rf
#' @inheritParams mice.impute.pmm
#' @param ntree The number of trees to grow. The default is 10.
#' @param rfPackage A single string specifying the backend for estimating the
#' random forest. The default backend is the \code{ranger} package. The only
#' alternative currently implemented is the \code{randomForest} package, which
#' used to be the default in mice 3.13.10 and earlier.
#' @param \dots Other named arguments passed down to
#' \code{mice:::install.on.demand()}, \code{randomForest::randomForest()} and
#' \code{randomForest:::randomForest.default()}.
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @details
#' Imputation of \code{y} by random forests. The method
#' calls \code{randomForrest()} which implements Breiman's random forest
#' algorithm (based on Breiman and Cutler's original Fortran code)
#' for classification and regression. See Appendix A.1 of Doove et al.
#' (2014) for the definition of the algorithm used.
#' @note An alternative implementation was independently
#' developed by Shah et al (2014). This were available as
#' functions \code{CALIBERrfimpute::mice.impute.rfcat} and
#' \code{CALIBERrfimpute::mice.impute.rfcont} (now archived).
#' Simulations by Shah (Feb 13, 2014) suggested that
#' the quality of the imputation for 10 and 100 trees was identical,
#' so mice 2.22 changed the default number of trees from \code{ntree = 100} to
#' \code{ntree = 10}.
#' @author Lisa Doove, Stef van Buuren, Elise Dusseldorp, 2012; Patrick Rockenschaub, 2021
#' @references
#'
#' Doove, L.L., van Buuren, S., Dusseldorp, E. (2014), Recursive partitioning
#' for missing data imputation in the presence of interaction Effects.
#' Computational Statistics & Data Analysis, 72, 92-104.
#'
#' Shah, A.D., Bartlett, J.W., Carpenter, J., Nicholas, O., Hemingway, H. (2014),
#' Comparison of random forest and parametric imputation models for
#' imputing missing data using MICE: A CALIBER study. American Journal
#' of Epidemiology, \doi{10.1093/aje/kwt312}.
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-cart.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' @seealso \code{\link{mice}}, \code{\link{mice.impute.cart}},
#' \code{\link[randomForest]{randomForest}}
#' \code{\link[ranger]{ranger}}
#' @family univariate imputation functions
#' @keywords datagen
#' @examples
#' \dontrun{
#' imp <- mice(nhanes2, meth = "rf", ntree = 3)
#' plot(imp)
#' }
#' @export
mice.impute.rf <- function(y, ry, x, wy = NULL, ntree = 10,
rfPackage = c("ranger", "randomForest"), ...) {
rfPackage <- match.arg(rfPackage)
if (is.null(wy)) wy <- !ry
ntree <- max(1, ntree) # safety
nmis <- sum(wy)
xobs <- x[ry, , drop = FALSE]
xmis <- x[wy, , drop = FALSE]
yobs <- y[ry]
# Find eligible donors
f <- switch(rfPackage,
randomForest = .randomForest.donors,
ranger = .ranger.donors
)
forest <- f(xobs, xmis, yobs, ntree, ...)
# Sample from donors
if (nmis == 1) forest <- array(forest, dim = c(1, ntree))
apply(forest, MARGIN = 1, FUN = function(s) sample(unlist(s), 1))
}
# Find eligible donors using the randomForest package (default)
.randomForest.donors <- function(xobs, xmis, yobs, ntree, ...) {
install.on.demand("randomForest", ...)
onetree <- function(xobs, xmis, yobs, ...) {
# Function to fit a single tree
fit <- randomForest::randomForest(
x = xobs,
y = yobs,
ntree = 1, ...
)
leafnr <- predict(object = fit, newdata = xobs, nodes = TRUE)
leafnr <- as.vector(attr(leafnr, "nodes"))
nodes <- predict(object = fit, newdata = xmis, nodes = TRUE)
nodes <- as.vector(attr(nodes, "nodes"))
donor <- lapply(nodes, function(s) yobs[leafnr == s])
return(donor)
}
sapply(seq_len(ntree), FUN = function(s) onetree(xobs, xmis, yobs, ...))
}
# Find eligible donors using the ranger package
.ranger.donors <- function(xobs, xmis, yobs, ntree, ...) {
install.on.demand("ranger", ...)
# Fit all trees at once
fit <- ranger::ranger(x = xobs, y = yobs, num.trees = ntree)
nodes <- predict(
object = fit, data = rbind(xobs, xmis),
type = "terminalNodes", predict.all = TRUE
)
nodes <- ranger::predictions(nodes)
nodes_obs <- nodes[1:nrow(xobs), , drop = FALSE]
nodes_mis <- nodes[(nrow(xobs) + 1):nrow(nodes), , drop = FALSE]
select_donors <- function(i) {
# Function to extract all eligible donors for each missing value
donors <- split(yobs, nodes_obs[, i])
donors[as.character(nodes_mis[, i])]
}
sapply(seq_len(ntree), FUN = select_donors)
}
mice/R/walking.R 0000644 0001762 0000144 00000004570 14330031606 013151 0 ustar ligges users #' Walking disability data
#'
#' Two items YA and YB measuring walking disability in samples A, B and E.
#'
#' Example dataset to demonstrate imputation of two items (YA and YB). Item YA
#' is administered to sample A and sample E, item YB is administered to sample B
#' and sample E, so sample E acts as a bridge study. Imputation using a bridge
#' study is better than simple equating or than imputation under independence.
#'
#' Item YA corresponds to the HAQ8 item, and item YB corresponds to the GAR9
#' items from Van Buuren et al (2005). Sample E (as well as sample B) is the
#' Euridiss study (n=292), sample A is the ERGOPLUS study (n=306).
#'
#' See Van Buuren (2018) section 9.4 for more details on the imputation
#' methodology.
#'
#' @name walking
#' @aliases walking
#' @docType data
#' @format A data frame with 890 rows on the following 5 variables:
#' \describe{
#' \item{sex}{Sex of respondent (factor)}
#' \item{age}{Age of respondent}
#' \item{YA}{Item administered in samples A and E (factor)}
#' \item{YB}{Item administered in samples B and E (factor)}
#' \item{src}{Source: Sample A, B or E (factor)}
#' }
#' @references van Buuren, S., Eyres, S., Tennant, A., Hopman-Rock, M. (2005).
#' Improving comparability of existing data by Response Conversion.
#' \emph{Journal of Official Statistics}, \bold{21}(1), 53-72.
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-codingsystems.html#sec:impbridge}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' @keywords datasets
#' @examples
#' md.pattern(walking)
#'
#' micemill <- function(n) {
#' for (i in 1:n) {
#' imp <<- mice.mids(imp) # global assignment
#' cors <- with(imp, cor(as.numeric(YA),
#' as.numeric(YB),
#' method = "kendall"
#' ))
#' tau <<- rbind(tau, getfit(cors, s = TRUE)) # global assignment
#' }
#' }
#'
#' plotit <- function() {
#' matplot(
#' x = 1:nrow(tau), y = tau,
#' ylab = expression(paste("Kendall's ", tau)),
#' xlab = "Iteration", type = "l", lwd = 1,
#' lty = 1:10, col = "black"
#' )
#' }
#'
#' tau <- NULL
#' imp <- mice(walking, max = 0, m = 10, seed = 92786)
#' pred <- imp$pred
#' pred[, c("src", "age", "sex")] <- 0
#' imp <- mice(walking, max = 0, m = 3, seed = 92786, pred = pred)
#' micemill(5)
#' plotit()
#'
#' ### to get figure 9.8 van Buuren (2018) use m=10 and micemill(20)
NULL
mice/R/mice.impute.lasso.logreg.R 0000644 0001762 0000144 00000004704 14330031647 016336 0 ustar ligges users #' Imputation by direct use of lasso logistic regression
#'
#' Imputes univariate missing binary data using lasso logistic regression with bootstrap.
#'
#' @aliases mice.impute.lasso.logreg lasso.logreg
#' @inheritParams mice.impute.pmm
#' @param nfolds The number of folds for the cross-validation of the lasso penalty.
#' The default is 10.
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @details
#' The method consists of the following steps:
#' \enumerate{
#' \item For a given y variable under imputation, draw a bootstrap version y*
#' with replacement from the observed cases \code{y[ry]}, and stores in x* the
#' corresponding values from \code{x[ry, ]}.
#' \item Fit a regularised (lasso) logistic regression with y* as the outcome,
#' and x* as predictors.
#' A vector of regression coefficients bhat is obtained.
#' All of these coefficients are considered random draws from the imputation model
#' parameters posterior distribution.
#' Same of these coefficients will be shrunken to 0.
#' \item Compute predicted scores for m.d., i.e. logit-1(X bhat)
#' \item Compare the score to a random (0,1) deviate, and impute.
#' }
#' The method is based on the Direct Use of Regularized Regression (DURR) proposed by
#' Zhao & Long (2016) and Deng et al (2016).
#' @author Edoardo Costantini, 2021
#' @references
#'
#' Deng, Y., Chang, C., Ido, M. S., & Long, Q. (2016). Multiple imputation for
#' general missing data patterns in the presence of high-dimensional data.
#' Scientific reports, 6(1), 1-10.
#'
#' Zhao, Y., & Long, Q. (2016). Multiple imputation in the presence of
#' high-dimensional data. Statistical Methods in Medical Research, 25(5),
#' 2021-2035.
#'
#' @family univariate imputation functions
#' @keywords datagen
#' @export
mice.impute.lasso.logreg <- function(y, ry, x, wy = NULL, nfolds = 10, ...) {
install.on.demand("glmnet", ...)
if (is.null(wy)) wy <- !ry
# Bootstrap sample
n1 <- sum(ry)
s <- sample(n1, n1, replace = TRUE)
x_glmnet <- cbind(1, x)
dotxobs <- x_glmnet[ry, , drop = FALSE][s, , drop = FALSE]
dotyobs <- y[ry][s]
# Train imputation model
cv_lasso <- glmnet::cv.glmnet(
x = dotxobs, y = dotyobs,
family = "binomial",
nfolds = nfolds,
alpha = 1
)
# Obtain imputation
p <- 1 / (1 + exp(-predict(cv_lasso, x_glmnet[wy, ], s = "lambda.min")))
vec <- (runif(nrow(p)) <= p)
vec[vec] <- 1
if (is.factor(y)) {
vec <- factor(vec, c(0, 1), levels(y))
}
vec
}
mice/R/mice.impute.2lonly.norm.R 0000644 0001762 0000144 00000012054 14330031606 016120 0 ustar ligges users #' Imputation at level 2 by Bayesian linear regression
#'
#' Imputes univariate missing data at level 2 using Bayesian linear regression
#' analysis. Variables are level 1 are aggregated at level 2. The group
#' identifier at level 2 must be indicated by \code{type = -2} in the
#' \code{predictorMatrix}.
#'
#' @aliases 2lonly.norm
#' @inheritParams mice.impute.pmm
#' @param type Group identifier must be specified by '-2'. Predictors must be
#' specified by '1'.
#' @param ... Other named arguments.
#' @return A vector of length \code{nmis} with imputations.
#' @author Alexander Robitzsch (IPN - Leibniz Institute for Science and
#' Mathematics Education, Kiel, Germany), \email{robitzsch@@ipn.uni-kiel.de}
#' @seealso \code{\link{mice.impute.norm}},
#' \code{\link{mice.impute.2lonly.pmm}}, \code{\link{mice.impute.2l.pan}},
#' \code{\link{mice.impute.2lonly.mean}}
#' @details
#' This function allows in combination with \code{\link{mice.impute.2l.pan}}
#' switching regression imputation between level 1 and level 2 as described in
#' Yucel (2008) or Gelman and Hill (2007, p. 541).
#'
#' The function checks for partial missing level-2 data. Level-2 data
#' are assumed to be constant within the same cluster. If one or more
#' entries are missing, then the procedure aborts with an error
#' message that identifies the cluster with incomplete level-2 data.
#' In such cases, one may first fill in the cluster mean (or mode) by
#' the \code{2lonly.mean} method to remove inconsistencies.
#'
#' @references Gelman, A. and Hill, J. (2007). \emph{Data analysis using
#' regression and multilevel/hierarchical models}. Cambridge, Cambridge
#' University Press.
#'
#' Yucel, RM (2008). Multiple imputation inference for multivariate multilevel
#' continuous data with ignorable non-response. \emph{Philosophical
#' Transactions of the Royal Society A}, \bold{366}, 2389-2404.
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-level2pred.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' @family univariate-2lonly
#' @note
#' For a more general approach, see
#' \code{miceadds::mice.impute.2lonly.function()}.
#' @examples
#' # simulate some data
#' # x,y ... level 1 variables
#' # v,w ... level 2 variables
#'
#' G <- 250 # number of groups
#' n <- 20 # number of persons
#' beta <- .3 # regression coefficient
#' rho <- .30 # residual intraclass correlation
#' rho.miss <- .10 # correlation with missing response
#' missrate <- .50 # missing proportion
#' y1 <- rep(rnorm(G, sd = sqrt(rho)), each = n) + rnorm(G * n, sd = sqrt(1 - rho))
#' w <- rep(round(rnorm(G), 2), each = n)
#' v <- rep(round(runif(G, 0, 3)), each = n)
#' x <- rnorm(G * n)
#' y <- y1 + beta * x + .2 * w + .1 * v
#' dfr0 <- dfr <- data.frame("group" = rep(1:G, each = n), "x" = x, "y" = y, "w" = w, "v" = v)
#' dfr[rho.miss * x + rnorm(G * n, sd = sqrt(1 - rho.miss)) < qnorm(missrate), "y"] <- NA
#' dfr[rep(rnorm(G), each = n) < qnorm(missrate), "w"] <- NA
#' dfr[rep(rnorm(G), each = n) < qnorm(missrate), "v"] <- NA
#'
#' # empty mice imputation
#' imp0 <- mice(as.matrix(dfr), maxit = 0)
#' predM <- imp0$predictorMatrix
#' impM <- imp0$method
#'
#' # multilevel imputation
#' predM1 <- predM
#' predM1[c("w", "y", "v"), "group"] <- -2
#' predM1["y", "x"] <- 1 # fixed x effects imputation
#' impM1 <- impM
#' impM1[c("y", "w", "v")] <- c("2l.pan", "2lonly.norm", "2lonly.pmm")
#'
#' # y ... imputation using pan
#' # w ... imputation at level 2 using norm
#' # v ... imputation at level 2 using pmm
#'
#' imp1 <- mice(as.matrix(dfr),
#' m = 1, predictorMatrix = predM1,
#' method = impM1, maxit = 1, paniter = 500
#' )
#'
#' # Demonstration that 2lonly.norm aborts for partial missing data.
#' # Better use 2lonly.mean for repair.
#' data <- data.frame(
#' patid = rep(1:4, each = 5),
#' sex = rep(c(1, 2, 1, 2), each = 5),
#' crp = c(
#' 68, 78, 93, NA, 143,
#' 5, 7, 9, 13, NA,
#' 97, NA, 56, 52, 34,
#' 22, 30, NA, NA, 45
#' )
#' )
#' pred <- make.predictorMatrix(data)
#' pred[, "patid"] <- -2
#' # only missing value (out of five) for patid == 1
#' data[3, "sex"] <- NA
#' \dontrun{
#' # The following fails because 2lonly.norm found partially missing
#' # level-2 data
#' # imp <- mice(data, method = c("", "2lonly.norm", "2l.pan"),
#' # predictorMatrix = pred, maxit = 1, m = 2)
#' # > iter imp variable
#' # > 1 1 sex crpError in .imputation.level2(y = y, ... :
#' # > Method 2lonly.norm found the following clusters with partially missing
#' # > level-2 data: 1
#' # > Method 2lonly.mean can fix such inconsistencies.
#' }
#'
#' # In contrast, if all sex values are missing for patid == 1, it runs fine,
#' # except on r-patched-solaris-x86. I used dontrun to evade CRAN errors.
#' \dontrun{
#' data[1:5, "sex"] <- NA
#' imp <- mice(data,
#' method = c("", "2lonly.norm", "2l.pan"),
#' predictorMatrix = pred, maxit = 1, m = 2
#' )
#' }
#' @export
mice.impute.2lonly.norm <- function(y, ry, x, type, wy = NULL, ...) {
imp <- .imputation.level2(
y = y, ry = ry, x = x, type = type, wy = wy,
method = "norm", ...
)
imp
}
mice/R/summary.R 0000644 0001762 0000144 00000005520 14334522175 013220 0 ustar ligges users #' Summary of a \code{mira} object
#'
#' @rdname summary
#' @param object A \code{mira} object
#' @param type A length-1 character vector indicating the
#' type of summary. There are three choices: \code{type = "tidy"}
#' return the parameters estimates of each analyses as a data frame.
#' \code{type = "glance"} return the fit statistics of each analysis
#' as a data frame. \code{type = "summary"} returns a list of
#' length \code{m} with the analysis results. The default is
#' \code{"tidy"}.
#' @param ... Other parameters passed down to \code{print()} and \code{summary()}
#' @return \code{NULL}
#' @seealso \code{\link[=mira-class]{mira}}
#' @method summary mira
#' @export
summary.mira <- function(object,
type = c("tidy", "glance", "summary"),
...) {
type <- match.arg(type)
fitlist <- getfit(object)
if (type == "tidy") {
v <- lapply(fitlist, tidy, effects = "fixed", parametric = TRUE, ...) %>% bind_rows()
}
if (type == "glance") {
v <- lapply(fitlist, glance, ...) %>% bind_rows()
}
# nobs is needed for pool.r.squared
# not supplied by broom <= 0.5.6
if (!"nobs" %in% colnames(v)) {
v$nobs <- tryCatch(length(stats::residuals(getfit(object)[[1]])),
error = function(e) NULL
)
}
if (type == "summary") {
v <- lapply(fitlist, summary, ...)
}
v
}
#' Summary of a \code{mids} object
#'
#' @rdname summary
#' @return \code{NULL}
#' @seealso \code{\link[=mids-class]{mids}}
#' @method summary mids
#' @export
summary.mids <- function(object, ...) {
print(object, ...)
invisible(object)
}
#' Summary of a \code{mads} object
#'
#' @rdname summary
#' @return \code{NULL}
#' @seealso \code{\link[=mads-class]{mads}}
#' @export
summary.mads <- function(object, ...) {
print(object, ...)
invisible(object)
}
#' Print a \code{mice.anova} object
#'
#' @rdname summary
#' @return \code{NULL}
#' @seealso \code{\link{mipo}}
#' @method summary mice.anova
#' @export
summary.mice.anova <- function(object, ...) {
# handle objects from anova
out <- object$out
# handle objects from D1, D2 and D3
if (is.null(out)) {
out <- list(`1 ~~ 2` = list(
result = object$result,
dfcom = object$dfcom
))
}
test <- names(out)
dfcom <- vapply(out, function(x) x$dfcom, numeric(1))
results <- t(vapply(out, function(x) x$result, numeric(5)))
rf <- data.frame(
test = test,
statistic = results[, 1],
df1 = results[, 2],
df2 = results[, 3],
dfcom = dfcom,
p.value = results[, 4],
riv = results[, 5],
row.names = NULL
)
formulas <- object$formulas
ff <- data.frame(
model = names(formulas),
formula = as.character(formulas)
)
structure(
list(
models = ff, comparisons = rf,
m = object$m, method = object$method, use = object$use
),
class = c("mice.anova.summary", class(object))
)
}
mice/R/plot.R 0000644 0001762 0000144 00000010313 14330031606 012463 0 ustar ligges users #' Plot the trace lines of the MICE algorithm
#'
#' Trace line plots portray the value of an estimate
#' against the iteration number. The estimate can be anything that you can calculate, but
#' typically are chosen as parameter of scientific interest. The \code{plot} method for
#' a \code{mids} object plots the mean and standard deviation of the imputed (not observed)
#' values against the iteration number for each of the $m$ replications. By default,
#' the function plot the development of the mean and standard deviation for each incomplete
#' variable. On convergence, the streams should intermingle and be free of any trend.
#'
#' @param x An object of class \code{mids}
#' @param y A formula that specifies which variables, stream and iterations are plotted.
#' If omitted, all streams, variables and iterations are plotted.
#' @param theme The trellis theme to applied to the graphs. The default is \code{mice.theme()}.
#' @param layout A vector of length 2 given the number of columns and rows in the plot.
#' The default is \code{c(2, 3)}.
#' @param type Parameter \code{type} of \code{\link{panel.xyplot}}.
#' @param col Parameter \code{col} of \code{\link{panel.xyplot}}.
#' @param lty Parameter \code{lty} of \code{\link{panel.xyplot}}.
#' @param ... Extra arguments for \code{\link{xyplot}}.
#' @return An object of class \code{"trellis"}.
#' @author Stef van Buuren 2011
#' @seealso \code{\link{mice}}, \code{\link[=mids-class]{mids}},
#' \code{\link{xyplot}}
#' @method plot mids
#' @examples
#' imp <- mice(nhanes, print = FALSE)
#' plot(imp, bmi + chl ~ .it | .ms, layout = c(2, 1))
#' @export
plot.mids <- function(x, y = NULL, theme = mice.theme(), layout = c(2, 3),
type = "l", col = 1:10, lty = 1, ...) {
strip.combined <- function(which.given, which.panel, factor.levels, ...) {
if (which.given == 1) {
lattice::panel.rect(0, 0, 1, 1,
col = theme$strip.background$col, border = 1
)
lattice::panel.text(
x = 0, y = 0.5, pos = 4,
lab = factor.levels[which.panel[which.given]]
)
}
if (which.given == 2) {
lattice::panel.text(
x = 1, y = 0.5, pos = 2,
lab = factor.levels[which.panel[which.given]]
)
}
}
call <- match.call()
if (!is.mids(x)) {
stop("argument 'x' must be a 'mids' object", call. = FALSE)
}
if (is.null(x$chainMean)) {
stop("no convergence diagnostics found", call. = FALSE)
}
mn <- x$chainMean
sm <- sqrt(x$chainVar)
# select subset of nonmissing entries
obs <- apply(!(is.nan(mn) | is.na(mn)), 1, all)
varlist <- names(obs)[obs]
## create formula if not given in y
if (missing(y)) {
formula <- as.formula(paste0(
paste0(varlist, collapse = "+"),
"~.it|.ms"
))
} else {
formula <- NULL
if (is.null(y)) {
formula <- as.formula(paste0(
paste0(varlist, collapse = "+"),
"~.it|.ms"
))
}
if (is.character(y)) {
formula <- if (length(y) == 1) {
as.formula(paste0(y, "~.it|.ms"))
} else {
as.formula(paste0(paste0(y, collapse = "+"), "~.it|.ms"))
}
}
if (is.integer(y) || is.logical(y)) {
vars <- varlist[y]
formula <- if (length(vars) == 1) {
as.formula(paste0(vars, "~.it|.ms"))
} else {
as.formula(paste0(paste0(vars, collapse = "+"), "~.it|.ms"))
}
}
if (is.null(formula)) {
formula <- as.formula(y)
}
}
m <- x$m
it <- x$iteration
mn <- matrix(aperm(mn[varlist, , , drop = FALSE], c(2, 3, 1)), nrow = m * it)
sm <- matrix(aperm(sm[varlist, , , drop = FALSE], c(2, 3, 1)), nrow = m * it)
adm <- expand.grid(seq_len(it), seq_len(m), c("mean", "sd"))
data <- cbind(adm, rbind(mn, sm))
colnames(data) <- c(".it", ".m", ".ms", varlist)
## Dummy to trick R CMD check
.m <- NULL
rm(.m)
tp <- xyplot(
x = formula, data = data, groups = .m,
type = type, lty = lty, col = col, layout = layout,
scales = list(
y = list(relation = "free"),
x = list(alternating = FALSE)
),
as.table = TRUE,
xlab = "Iteration",
ylab = "",
strip = strip.combined,
par.strip.text = list(lines = 0.5),
...
)
update(tp, par.settings = theme)
}
mice/R/nimp.R 0000644 0001762 0000144 00000001356 14330031606 012457 0 ustar ligges users #' Number of imputations per block
#'
#' Calculates the number of cells within a block for which imputation
#' is requested.
#' @inheritParams mice
#' @return A numeric vector of length \code{length(blocks)} containing
#' the number of cells that need to be imputed within a block.
#' @seealso \code{\link{mice}}
#' @export
#' @examples
#' where <- is.na(nhanes)
#'
#' # standard FCS
#' nimp(where)
#'
#' # user-defined blocks
#' nimp(where, blocks = name.blocks(list(c("bmi", "hyp"), "age", "chl")))
nimp <- function(where, blocks = make.blocks(where)) {
nwhere <- apply(where, 2, sum)
nimp <- vector("integer", length = length(blocks))
names(nimp) <- names(blocks)
for (i in seq_along(blocks)) nimp[i] <- sum(nwhere[blocks[[i]]])
nimp
}
mice/R/mice.impute.panImpute.R 0000644 0001762 0000144 00000006601 14436064333 015704 0 ustar ligges users #' Impute multilevel missing data using \code{pan}
#'
#' This function is a wrapper around the \code{panImpute} function
#' from the \code{mitml} package so that it can be called to
#' impute blocks of variables in \code{mice}. The \code{mitml::panImpute}
#' function provides an interface to the \code{pan} package for
#' multiple imputation of multilevel data (Schafer & Yucel, 2002).
#' Imputations can be generated using \code{type} or \code{formula},
#' which offer different options for model specification.
#'
#' @name mice.impute.panImpute
#' @inheritParams mitml::panImpute
#' @param data A data frame containing incomplete and auxiliary variables,
#' the cluster indicator variable, and any other variables that should be
#' present in the imputed datasets.
#' @param type An integer vector specifying the role of each variable
#' in the imputation model (see \code{\link[mitml]{panImpute}})
#' @param formula A formula specifying the role of each variable
#' in the imputation model. The basic model is constructed
#' by \code{model.matrix}, thus allowing to include derived variables
#' in the imputation model using \code{I()}. See
#' \code{\link[mitml]{panImpute}}.
#' @param format A character vector specifying the type of object that should
#' be returned. The default is \code{format = "list"}. No other formats are
#' currently supported.
#' @param ... Other named arguments: \code{n.burn}, \code{n.iter},
#' \code{group}, \code{prior}, \code{silent} and others.
#' @return A list of imputations for all incomplete variables in the model,
#' that can be stored in the the \code{imp} component of the \code{mids}
#' object.
#' @seealso \code{\link[mitml]{panImpute}}
#' @note The number of imputations \code{m} is set to 1, and the function
#' is called \code{m} times so that it fits within the \code{mice}
#' iteration scheme.
#'
#' This is a multivariate imputation function using a joint model.
#' @author Stef van Buuren, 2018, building on work of Simon Grund,
#' Alexander Robitzsch and Oliver Luedtke (authors of \code{mitml} package)
#' and Joe Schafer (author of \code{pan} package).
#' @references
#' Grund S, Luedtke O, Robitzsch A (2016). Multiple
#' Imputation of Multilevel Missing Data: An Introduction to the R
#' Package \code{pan}. SAGE Open.
#'
#' Schafer JL (1997). Analysis of Incomplete Multivariate Data. London:
#' Chapman & Hall.
#'
#' Schafer JL, and Yucel RM (2002). Computational strategies for
#' multivariate linear mixed-effects models with missing values.
#' Journal of Computational and Graphical Statistics, 11, 437-457.
#' @family multivariate-2l
#' @keywords datagen
#' @examples
#' blocks <- list(c("bmi", "chl", "hyp"), "age")
#' method <- c("panImpute", "pmm")
#' ini <- mice(nhanes, blocks = blocks, method = method, maxit = 0)
#' pred <- ini$pred
#' pred["B1", "hyp"] <- -2
#' imp <- mice(nhanes, blocks = blocks, method = method, pred = pred, maxit = 1)
#' @export
mice.impute.panImpute <- function(data, formula, type, m = 1, silent = TRUE,
format = "imputes", ...) {
install.on.demand("mitml", ...)
nat <- mitml::panImpute(
data = data, formula = formula, type = type,
m = m, silent = silent, ...
)
if (format == "native") {
return(nat)
}
cmp <- mitml::mitmlComplete(nat, print = 1)[, names(data)]
if (format == "complete") {
return(cmp)
}
if (format == "imputes") {
return(single2imputes(cmp, is.na(data)))
}
NULL
}
mice/R/mice.impute.mean.R 0000644 0001762 0000144 00000002256 14330031647 014657 0 ustar ligges users #' Imputation by the mean
#'
#' Imputes the arithmetic mean of the observed data
#'
#' @inheritParams mice.impute.pmm
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @section Warning: Imputing the mean of a variable is almost never
#' appropriate. See Little and Rubin (2002, p. 61-62) or
#' Van Buuren (2012, p. 10-11)
#' @seealso \code{\link{mice}}, \code{\link{mean}}
#' @references Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#'
#' Little, R.J.A. and Rubin, D.B. (2002). Statistical Analysis with Missing
#' Data. New York: John Wiley and Sons.
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-simplesolutions.html#sec:meanimp}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' @family univariate imputation functions
#' @keywords datagen
#' @export
mice.impute.mean <- function(y, ry, x = NULL, wy = NULL, ...) {
if (is.null(wy)) {
wy <- !ry
}
rep.int(mean(y[ry]), times = sum(wy))
}
mice/R/D1.R 0000644 0001762 0000144 00000005605 14436064333 011773 0 ustar ligges users #' Compare two nested models using D1-statistic
#'
#' The D1-statistics is the multivariate Wald test.
#'
#' @param fit1 An object of class \code{mira}, produced by \code{with()}.
#' @param fit0 An object of class \code{mira}, produced by \code{with()}. The
#' model in \code{fit0} is a nested within \code{fit1}. The default null
#' model \code{fit0 = NULL} compares \code{fit1} to the intercept-only model.
#' @param dfcom A single number denoting the
#' complete-data degrees of freedom of model \code{fit1}. If not specified,
#' it is set equal to \code{df.residual} of model \code{fit1}. If that cannot
#' be done, the procedure assumes (perhaps incorrectly) a large sample.
#' @param df.com Deprecated
#' @note Warning: `D1()` assumes that the order of the variables is the
#' same in different models. See
#' \url{https://github.com/amices/mice/issues/420} for details.
#' @references
#' Li, K. H., T. E. Raghunathan, and D. B. Rubin. 1991.
#' Large-Sample Significance Levels from Multiply Imputed Data Using
#' Moment-Based Statistics and an F Reference Distribution.
#' \emph{Journal of the American Statistical Association}, 86(416): 1065–73.
#'
#' \url{https://stefvanbuuren.name/fimd/sec-multiparameter.html#sec:wald}
#' @examples
#' # Compare two linear models:
#' imp <- mice(nhanes2, seed = 51009, print = FALSE)
#' mi1 <- with(data = imp, expr = lm(bmi ~ age + hyp + chl))
#' mi0 <- with(data = imp, expr = lm(bmi ~ age + hyp))
#' D1(mi1, mi0)
#' \dontrun{
#' # Compare two logistic regression models
#' imp <- mice(boys, maxit = 2, print = FALSE)
#' fit1 <- with(imp, glm(gen > levels(gen)[1] ~ hgt + hc + reg, family = binomial))
#' fit0 <- with(imp, glm(gen > levels(gen)[1] ~ hgt + hc, family = binomial))
#' D1(fit1, fit0)
#' }
#' @seealso \code{\link[mitml]{testModels}}
#' @export
D1 <- function(fit1, fit0 = NULL, dfcom = NULL, df.com = NULL) {
install.on.demand("mitml")
# legacy handling
if (!missing(df.com)) {
warning("argument df.com is deprecated; please use dfcom instead.",
call. = FALSE
)
dfcom <- df.com
}
dfcom <- get.dfcom(fit1, dfcom)
# fit1: a fitlist or mira-object
# fit0: named numerical vector, character vector, or list
fit1 <- as.mitml.result(fit1)
est1 <- pool(fit1, dfcom = dfcom)
qbar1 <- getqbar(est1)
if (is.null(fit0)) {
# test all estimates equal to zero, except intercept
beta <- rep(0, length(qbar1))
names(beta) <- names(qbar1)
fit0 <- lapply(fit1, fix.coef, beta = beta)
fit0 <- as.mitml.result(fit0)
} else if (is.mira(fit0)) {
fit0 <- as.mitml.result(fit0)
}
tmr <- mitml::testModels(fit1, fit0, method = "D1", df.com = dfcom)
out <- list(
call = match.call(),
result = tmr$test,
formulas = list(
`1` = formula(fit1[[1L]]),
`2` = formula(fit0[[1L]])
),
m = tmr$m,
method = "D1",
use = NULL,
dfcom = tmr$df.com
)
class(out) <- c("mice.anova", class(fit1))
out
}
mice/R/employee.R 0000644 0001762 0000144 00000002347 14330031606 013334 0 ustar ligges users #' Employee selection data
#'
#' A toy example from Craig Enders.
#'
#' Enders describes these data as follows:
#' I designed these data to mimic an employee selection scenario in
#' which prospective employees complete an IQ test and a
#' psychological well-being questionnaire during their interview.
#' The company subsequently hires the applications that score in the
#' upper half of the IQ distribution, and a supervisor rates their
#' job performance following a 6-month probationary period.
#' Note that the job performance scores are missing at random (MAR)
#' (i.e. individuals in the lower half of the IQ distribution were
#' never hired, and thus have no performance rating). In addition,
#' I randomly deleted three of the well-being scores in order to
#' mimic a situation where the applicant's well-being questionnaire
#' is inadvertently lost.
#'
#' A larger version of this data set in present as
#' \code{\link[miceadds:data.enders]{data.enders.employee}}.
#'
#' @format A data frame with 20 rows and 3 variables:
#' \describe{
#' \item{IQ}{candidate IQ score}
#' \item{wbeing}{candidate well-being score}
#' \item{jobperf}{candidate job performance score}
#' }
#' @source Enders (2010), Applied Missing Data Analysis, p. 218
"employee"
mice/R/cci.R 0000644 0001762 0000144 00000003340 14330031606 012245 0 ustar ligges users #' Complete case indicator
#'
#'
#' The complete case indicator is useful for extracting the subset of complete cases. The function
#' \code{cci(x)} calls \code{complete.cases(x)}.
#' The companion function \code{ici()} selects the incomplete cases.
#'
#' @name cci
#' @param x An \code{R} object. Currently supported are methods for the
#' following classes: \code{mids}.
#' @return Logical vector indicating the complete cases.
#' @author Stef van Buuren, 2017.
#' @seealso \code{\link{complete.cases}}, \code{\link{ici}}, \code{\link{cc}}
#' @keywords univar
#' @examples
#' cci(nhanes) # indicator for 13 complete cases
#' cci(mice(nhanes, maxit = 0))
#' f <- cci(nhanes[, c("bmi", "hyp")]) # complete data for bmi and hyp
#' nhanes[f, ] # obtain all data from those with complete bmi and hyp
#' @export
cci <- function(x) UseMethod("cci", x)
#' @export
cci.mids <- function(x) {
complete.cases(x$data)
}
#' @export
cci.default <- function(x) {
complete.cases(x)
}
#' Incomplete case indicator
#'
#' This array is useful for extracting the subset of incomplete cases.
#' The companion function \code{cci()} selects the complete cases.
#'
#' @name ici
#' @aliases ici ici,data.frame-method ici,matrix-method ici,mids-method
#' @param x An \code{R} object. Currently supported are methods for the
#' following classes: \code{mids}.
#' @return Logical vector indicating the incomplete cases,
#' @author Stef van Buuren, 2017.
#' @seealso \code{\link{cci}}, \code{\link{ic}}
#' @keywords univar
#' @examples
#'
#' ici(nhanes) # indicator for 12 rows with incomplete cases
#' @export
ici <- function(x) UseMethod("ici", x)
#' @export
ici.mids <- function(x) {
!complete.cases(x$data)
}
#' @export
ici.default <- function(x) {
!complete.cases(x)
}
mice/R/mice.impute.norm.nob.R 0000644 0001762 0000144 00000004313 14330031647 015463 0 ustar ligges users #' Imputation by linear regression without parameter uncertainty
#'
#' Imputes univariate missing data using linear regression analysis without
#' accounting for the uncertainty of the model parameters.
#'
#' @aliases mice.impute.norm.nob norm.nob
#' @inheritParams mice.impute.pmm
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @details
#' This function creates imputations using the spread around the
#' fitted linear regression line of \code{y} given \code{x}, as
#' fitted on the observed data.
#'
#' This function is provided mainly to allow comparison between proper (e.g.,
#' as implemented in \code{mice.impute.norm} and improper (this function)
#' normal imputation methods.
#'
#' For large data, having many rows, differences between proper and improper
#' methods are small, and in those cases one may opt for speed by using
#' \code{mice.impute.norm.nob}.
#' @section Warning: The function does not incorporate the variability of the
#' regression weights, so it is not 'proper' in the sense of Rubin. For small
#' samples, variability of the imputed data is therefore underestimated.
#' @author Gerko Vink, Stef van Buuren, Karin Groothuis-Oudshoorn, 2018
#' @seealso \code{\link{mice}}, \code{\link{mice.impute.norm}}
#' @references Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#'
#' Brand, J.P.L. (1999). Development, Implementation and Evaluation of Multiple
#' Imputation Strategies for the Statistical Analysis of Incomplete Data Sets.
#' Ph.D. Thesis, TNO Prevention and Health/Erasmus University Rotterdam.
#' @family univariate imputation functions
#' @keywords datagen
#' @export
mice.impute.norm.nob <- function(y, ry, x, wy = NULL, ...) {
if (is.null(wy)) wy <- !ry
x <- cbind(1, as.matrix(x))
parm <- .norm.fix(y, ry, x, ...)
x[wy, ] %*% parm$beta + rnorm(sum(wy)) * parm$sigma
}
.norm.fix <- function(y, ry, x, ...) {
p <- estimice(x[ry, , drop = FALSE], y[ry], ...)
sigma <- sqrt((sum(p$r^2)) / (sum(ry) - ncol(x) - 1))
parm <- list(p$c, sigma)
names(parm) <- c("beta", "sigma")
parm
}
mice/R/cbind.R 0000644 0001762 0000144 00000020255 14433400023 012567 0 ustar ligges users cbind.mids <- function(x, y = NULL, ...) {
call <- match.call()
dots <- list(...)
if (is.mids(y)) {
return(cbind.mids.mids(x, y, call = call))
}
if ((is.null(y) || length(y) == 0L) && length(dots) == 0L) {
return(x)
}
n <- nrow(x$data)
if (length(y) == 1L) y <- rep(y, n)
if (length(y) == 0L && length(dots) > 0L) {
y <- cbind.data.frame(dots)
} else if (length(y) > 0L && length(dots) == 0L) {
y <- cbind.data.frame(y)
} else {
y <- cbind.data.frame(y, dots)
}
# Call is a vector, with first argument the mice statement
# and second argument the call to cbind.mids.
call <- c(x$call, call)
if (nrow(y) != nrow(x$data)) {
stop(
"arguments imply differing number of rows: ",
c(nrow(x$data), ", ", nrow(y))
)
}
data <- cbind(x$data, y)
varnames <- make.unique(colnames(data))
colnames(data) <- varnames
# where argument
where <- cbind(x$where, matrix(FALSE, nrow = nrow(x$where), ncol = ncol(y)))
colnames(where) <- varnames
# blocks: no renaming needed because all block definition will
# refer to varnames[1:ncol(x$data)] only, and are hence unique
# but we do need to rename duplicate block names
yblocks <- vector("list", length = ncol(y))
blocks <- c(x$blocks, yblocks)
xynames <- c(names(x$blocks), colnames(y))
blocknames <- make.unique(xynames)
names(blocknames) <- xynames
names(blocks) <- blocknames
ct <- c(attr(x$blocks, "calltype"), rep("type", ncol(y)))
names(ct) <- blocknames
attr(blocks, "calltype") <- ct
m <- x$m
# count the number of missing data in y
nmis <- c(x$nmis, colSums(is.na(y)))
names(nmis) <- varnames
# imp: original data of y will be copied into the multiple imputed dataset,
# including the missing values of y.
r <- (!is.na(y))
f <- function(j) {
m <- matrix(NA,
nrow = sum(!r[, j]),
ncol = x$m,
dimnames = list(row.names(y)[!r[, j]], seq_len(m))
)
as.data.frame(m)
}
imp <- lapply(seq_len(ncol(y)), f)
imp <- c(x$imp, imp)
names(imp) <- varnames
# The imputation method for (columns in) y will be set to ''.
method <- c(x$method, rep.int("", ncol(y)))
names(method) <- blocknames
# The variable(s) in y are included in the predictorMatrix.
# y is not used as predictor as well as not imputed.
predictorMatrix <- rbind(
x$predictorMatrix,
matrix(0,
ncol = ncol(x$predictorMatrix),
nrow = ncol(y)
)
)
predictorMatrix <- cbind(
predictorMatrix,
matrix(0,
ncol = ncol(y),
nrow = nrow(x$predictorMatrix) + ncol(y)
)
)
rownames(predictorMatrix) <- blocknames
colnames(predictorMatrix) <- varnames
visitSequence <- x$visitSequence
formulas <- x$formulas
post <- c(x$post, rep.int("", ncol(y)))
names(post) <- varnames
blots <- x$blots
ignore <- x$ignore
# seed, lastSeedValue, number of iterations, chainMean and chainVar
# is taken as in mids object x.
seed <- x$seed
lastSeedValue <- x$lastSeedValue
iteration <- x$iteration
chainMean <- x$chainMean
chainVar <- x$chainVar
loggedEvents <- x$loggedEvents
## save, and return
midsobj <- list(
data = data, imp = imp, m = m,
where = where, blocks = blocks,
call = call, nmis = nmis,
method = method,
predictorMatrix = predictorMatrix,
visitSequence = visitSequence,
formulas = formulas,
post = post,
blots = blots,
ignore = ignore,
seed = seed,
iteration = iteration,
lastSeedValue = lastSeedValue,
chainMean = chainMean,
chainVar = chainVar,
loggedEvents = loggedEvents,
version = packageVersion("mice"),
date = Sys.Date()
)
oldClass(midsobj) <- "mids"
midsobj
}
cbind.mids.mids <- function(x, y, call) {
if (!is.mids(y)) stop("Argument `y` not a mids object")
if (nrow(y$data) != nrow(x$data)) {
stop("The two datasets do not have the same length\n")
}
if (x$m != y$m) {
stop("The two mids objects should have the same number of imputations\n")
}
# Call is a vector, with first argument the mice statement
# and second argument the call to cbind.mids.
call <- c(x$call, call)
# The data in x$data and y$data are combined together.
# make variable names unique
data <- cbind(x$data, y$data)
xynames <- c(colnames(x$data), colnames(y$data))
varnames <- make.unique(xynames)
names(varnames) <- xynames
names(data) <- varnames
where <- cbind(x$where, y$where)
colnames(where) <- varnames
# rename variable names within each x$blocks and y$blocks
xnew <- varnames[1:ncol(x$data)]
ynew <- varnames[-(1:ncol(x$data))]
xblocks <- x$blocks
yblocks <- y$blocks
for (i in names(xblocks)) xblocks[[i]] <- unname(xnew[xblocks[[i]]])
for (i in names(yblocks)) yblocks[[i]] <- unname(ynew[yblocks[[i]]])
blocks <- c(xblocks, yblocks)
xynames <- c(names(xblocks), names(yblocks))
blocknames <- make.unique(xynames)
names(blocknames) <- xynames
names(blocks) <- blocknames
ct <- c(attr(xblocks, "calltype"), attr(yblocks, "calltype"))
names(ct) <- blocknames
attr(blocks, "calltype") <- ct
m <- x$m
nmis <- c(x$nmis, y$nmis)
names(nmis) <- varnames
imp <- c(x$imp, y$imp)
names(imp) <- varnames
method <- c(x$method, y$method)
names(method) <- blocknames
# The predictorMatrices of x and y are combined with zero matrices
# on the off diagonal blocks.
predictorMatrix <- rbind(
x$predictorMatrix,
matrix(0,
ncol = ncol(x$predictorMatrix),
nrow = nrow(y$predictorMatrix)
)
)
predictorMatrix <- cbind(
predictorMatrix,
rbind(
matrix(0,
ncol = ncol(y$predictorMatrix),
nrow = nrow(x$predictorMatrix)
),
y$predictorMatrix
)
)
rownames(predictorMatrix) <- blocknames
colnames(predictorMatrix) <- varnames
# As visitSequence is taken first the order for x and after that from y.
# take care that duplicate names need to be renamed
xnew <- blocknames[1:length(x$blocks)]
ynew <- blocknames[-(1:length(x$blocks))]
visitSequence <- unname(c(xnew[x$visitSequence], ynew[y$visitSequence]))
formulas <- c(x$formulas, y$formulas)
names(formulas) <- blocknames
post <- c(x$post, y$post)
names(post) <- varnames
blots <- c(x$blots, y$blots)
names(blots) <- blocknames
ignore <- x$ignore
# For the elements seed, lastSeedValue and iteration the values
# from midsobject x are copied.
seed <- x$seed
lastSeedValue <- x$lastSeedValue
iteration <- x$iteration
# the chainMean and chainVar vectors for x and y are combined.
chainMean <- array(
data = NA,
dim = c(dim(x$chainMean)[1] + dim(y$chainMean)[1], iteration, m),
dimnames = list(
c(
dimnames(x$chainMean)[[1]],
dimnames(y$chainMean)[[1]]
),
dimnames(x$chainMean)[[2]],
dimnames(x$chainMean)[[3]]
)
)
chainMean[seq_len(dim(x$chainMean)[1]), , ] <- x$chainMean
if (iteration <= dim(y$chainMean)[2]) {
chainMean[(dim(x$chainMean)[1] + 1):dim(chainMean)[1], , ] <- y$chainMean[, seq_len(iteration), ]
} else {
chainMean[(dim(x$chainMean)[1] + 1):dim(chainMean)[1], seq_len(dim(y$chainMean)[2]), ] <- y$chainMean
}
chainVar <- array(
data = NA, dim = c(dim(x$chainVar)[1] + dim(y$chainVar)[1], iteration, m),
dimnames = list(
c(
dimnames(x$chainVar)[[1]],
dimnames(y$chainVar)[[1]]
),
dimnames(x$chainVar)[[2]],
dimnames(x$chainVar)[[3]]
)
)
chainVar[seq_len(dim(x$chainVar)[1]), , ] <- x$chainVar
if (iteration <= dim(y$chainVar)[2]) {
chainVar[(dim(x$chainVar)[1] + 1):dim(chainVar)[1], , ] <- y$chainVar[, seq_len(iteration), ]
} else {
chainVar[(dim(x$chainVar)[1] + 1):dim(chainVar)[1], seq_len(dim(y$chainVar)[2]), ] <- y$chainVar
}
loggedEvents <- x$loggedEvents
midsobj <- list(
data = data, imp = imp, m = m,
where = where, blocks = blocks,
call = call, nmis = nmis,
method = method,
predictorMatrix = predictorMatrix,
visitSequence = visitSequence,
formulas = formulas,
post = post,
blots = blots,
ignore = ignore,
seed = seed,
iteration = iteration,
lastSeedValue = lastSeedValue,
chainMean = chainMean,
chainVar = chainVar,
loggedEvents = loggedEvents,
version = packageVersion("mice"),
date = Sys.Date()
)
oldClass(midsobj) <- "mids"
midsobj
}
mice/R/mice.impute.2l.pan.R 0000644 0001762 0000144 00000017301 14436064333 015033 0 ustar ligges users # Usage is an extension of 2l.norm
# -2 ... group identifier
# 1 ... fixed effects
# 2 ... fixed and random effects
# 3 ... introduce aggregated effects (i.e. group means)
# 4 ... fixed, random and aggregated effects
#' Imputation by a two-level normal model using \code{pan}
#'
#' Imputes univariate missing data using a two-level normal model with
#' homogeneous within group variances. Aggregated group effects (i.e. group
#' means) can be automatically created and included as predictors in the
#' two-level regression (see argument \code{type}). This function needs the
#' \code{pan} package.
#'
#' Implements the Gibbs sampler for the linear two-level model with homogeneous
#' within group variances which is a special case of a multivariate linear mixed
#' effects model (Schafer & Yucel, 2002). For a two-level imputation with
#' heterogeneous within-group variances see \code{\link{mice.impute.2l.norm}}. %
#' The random intercept is automatically added in %
#' \code{mice.impute.2l.norm()}.
#'
#' @aliases mice.impute.2l.pan 2l.pan
#' @author Alexander Robitzsch (IPN - Leibniz Institute for Science and
#' Mathematics Education, Kiel, Germany), \email{robitzsch@@ipn.uni-kiel.de}
#' @name mice.impute.2l.pan
#' @param y Incomplete data vector of length \code{n}
#' @param ry Vector of missing data pattern (\code{FALSE}=missing,
#' \code{TRUE}=observed)
#' @param x Matrix (\code{n} x \code{p}) of complete covariates.
#' @param type Vector of length \code{ncol(x)} identifying random and class
#' variables. Random effects are identified by a '2'. The group variable (only
#' one is allowed) is coded as '-2'. Random effects also include the fixed
#' effect. If for a covariates X1 group means shall be calculated and included
#' as further fixed effects choose '3'. In addition to the effects in '3',
#' specification '4' also includes random effects of X1.
#' @param intercept Logical determining whether the intercept is automatically
#' added.
#' @param paniter Number of iterations in \code{pan}. Default is 500.
#' @param groupcenter.slope If \code{TRUE}, in case of group means (\code{type}
#' is '3' or'4') group mean centering for these predictors are conducted before
#' doing imputations. Default is \code{FALSE}.
#' @param ... Other named arguments.
#' @return A vector of length \code{nmis} with imputations.
#' @author Alexander Robitzsch (IPN - Leibniz Institute for Science and
#' Mathematics Education, Kiel, Germany), \email{robitzsch@@ipn.uni-kiel.de}.
#' @note This function does not implement the \code{where} functionality. It
#' always produces \code{nmis} imputation, irrespective of the \code{where}
#' argument of the \code{mice} function.
#' @family univariate-2l
#' @references
#'
#' Schafer J L, Yucel RM (2002). Computational strategies for multivariate
#' linear mixed-effects models with missing values. \emph{Journal of
#' Computational and Graphical Statistics}. \bold{11}, 437-457.
#'
#' Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}: Multivariate
#' Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
#' Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
#' @examples
#' # simulate some data
#' # two-level regression model with fixed slope
#'
#' # number of groups
#' G <- 250
#' # number of persons
#' n <- 20
#' # regression parameter
#' beta <- .3
#' # intraclass correlation
#' rho <- .30
#' # correlation with missing response
#' rho.miss <- .10
#' # missing proportion
#' missrate <- .50
#' y1 <- rep(rnorm(G, sd = sqrt(rho)), each = n) + rnorm(G * n, sd = sqrt(1 - rho))
#' x <- rnorm(G * n)
#' y <- y1 + beta * x
#' dfr0 <- dfr <- data.frame("group" = rep(1:G, each = n), "x" = x, "y" = y)
#' dfr[rho.miss * x + rnorm(G * n, sd = sqrt(1 - rho.miss)) < qnorm(missrate), "y"] <- NA
#'
#' # empty imputation in mice
#' imp0 <- mice(as.matrix(dfr), maxit = 0)
#' predM <- imp0$predictorMatrix
#' impM <- imp0$method
#'
#' # specify predictor matrix and method
#' predM1 <- predM
#' predM1["y", "group"] <- -2
#' predM1["y", "x"] <- 1 # fixed x effects imputation
#' impM1 <- impM
#' impM1["y"] <- "2l.pan"
#'
#' # multilevel imputation
#' imp1 <- mice(as.matrix(dfr),
#' m = 1, predictorMatrix = predM1,
#' method = impM1, maxit = 1
#' )
#'
#' # multilevel analysis
#' library(lme4)
#' mod <- lmer(y ~ (1 + x | group) + x, data = complete(imp1))
#' summary(mod)
#'
#' # Examples of predictorMatrix specification
#'
#' # random x effects
#' # predM1["y","x"] <- 2
#'
#' # fixed x effects and group mean of x
#' # predM1["y","x"] <- 3
#'
#' # random x effects and group mean of x
#' # predM1["y","x"] <- 4
#' @export
mice.impute.2l.pan <- function(y, ry, x, type, intercept = TRUE, paniter = 500,
groupcenter.slope = FALSE, ...) {
install.on.demand("pan", ...)
## append intercept
if (intercept) {
x <- cbind(1, as.matrix(x))
type <- c(2, type)
}
# add groupmeans in the regression model
if (any(type %in% c(3, 4))) {
x0 <- as.matrix(cbind(x[, type == -2], x[, type %in% c(3, 4)]))
colnames(x0) <- c(colnames(x)[type == -2], colnames(x)[type %in% c(3, 4)])
type0 <- c(-2, rep.int(1, ncol(x0) - 1))
x0.aggr <- as.matrix(.mice.impute.2l.groupmean(
y = y, ry = ry, x = x0,
type = type0, grmeanwarning = FALSE, ...
))
colnames(x0.aggr) <- paste0("M.", colnames(x0)[-1])
# groupcentering
if (groupcenter.slope) {
x0.aggr1 <- as.matrix(x0.aggr)
colnames(x0.aggr1) <- colnames(x0)[-1]
x0cent <- x0[, -1] - x0.aggr1
x[, colnames(x0cent)] <- x0cent
}
# combine covariate matrix
x <- cbind(x, x0.aggr)
# add type
type1 <- c(type, rep.int(1, ncol(x0.aggr)))
names(type1) <- c(names(type), colnames(x0.aggr))
type1[type1 == 3] <- 1
type1[type1 == 4] <- 2
type <- type1
}
#############################
# pan imputation
# define cluster
group <- x[, type == -2]
subj <- match(group, unique(group))
# is group resorting necessary? (need this for pan)
sortgroups <- any(diff(subj) < 0)
if (sortgroups) {
dfr <- data.frame(
"group" = group, "ry" = ry,
"index" = seq(1, length(ry))
)
dfr <- dfr[order(dfr$group), ]
group <- group[dfr$index]
y <- y[dfr$index]
x <- x[dfr$index, ]
ry <- ry[dfr$index]
subj <- subj[dfr$index]
# stop( "Sort group identifiers in increasing order!\n")
}
y1 <- matrix(as.numeric(y), ncol = 1)
y1[!ry, 1] <- NA
# specify predictors
pred <- x[, type != -2, drop = FALSE] ## fixed SvB 1feb2013
# columns fixed effects
xcol <- seq(1, ncol(pred))
type1 <- type[type != -2]
zcol <- which(type1 == 2)
# noninformative priors
prior <- list(
a = ncol(y1), Binv = diag(rep(1, ncol(y1))),
c = ncol(y1) * length(zcol), Dinv = diag(rep(1, ncol(y1) * length(zcol)))
)
if (length(subj) != nrow(y1)) stop("No class variable")
# pan imputation
ii <- 0
while (ii == 0) {
s1 <- round(runif(1, 1, 10^7))
imput <- pan::pan(y1, subj, pred, xcol, zcol, prior, seed = s1, iter = paniter)
res <- imput$y
ii <- 1 - any(is.na(res))
# check for invalid imputations: pan occasionally produces NaNs
}
if (sortgroups) {
dfr <- cbind(res, dfr)
dfr <- dfr[order(dfr$index), ]
res <- dfr[!dfr$ry, "res"]
} else {
res <- res[!ry]
}
flush.console()
res
}
# compute cluster groupmean
.mice.impute.2l.groupmean <- function(y, ry, x, type, grmeanwarning = TRUE, ...) {
if ((ncol(x) > 2) & grmeanwarning) warning("\nMore than one variable is requested to be aggregated.\n")
# calculate aggregated values
a1 <- aggregate(x[, type %in% c(1, 2)], list(x[, type == -2]), mean, na.rm = TRUE)
i1 <- match(x[, type == -2], a1[, 1])
ximp <- as.matrix(a1[i1, -1])
colnames(ximp) <- paste(names(type)[type %in% c(1, 2)], names(type)[type == -2], sep = ".")
return(ximp)
}
mice/R/md.pairs.R 0000644 0001762 0000144 00000003626 14330031647 013240 0 ustar ligges users #' Missing data pattern by variable pairs
#'
#' Number of observations per variable pair.
#'
#' The four components in the output value is have the following interpretation:
#' \describe{ \item{list('rr')}{response-response, both variables are observed}
#' \item{list('rm')}{response-missing, row observed, column missing}
#' \item{list('mr')}{missing -response, row missing, column observed}
#' \item{list('mm')}{missing -missing, both variables are missing} }
#'
#' @param data A data frame or a matrix containing the incomplete data. Missing
#' values are coded as \code{NA}.
#' @return A list of four components named \code{rr}, \code{rm}, \code{mr} and
#' \code{mm}. Each component is square numerical matrix containing the number
#' observations within four missing data pattern.
#' @author Stef van Buuren, Karin Groothuis-Oudshoorn, 2009
#' @references Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#' @keywords univar
#' @examples
#' pat <- md.pairs(nhanes)
#' pat
#'
#' # show that these four matrices decompose the total sample size
#' # for each pair
#' pat$rr + pat$rm + pat$mr + pat$mm
#'
#' # percentage of usable cases to impute row variable from column variable
#' round(100 * pat$mr / (pat$mr + pat$mm))
#' @export
md.pairs <- function(data) {
# calculates pairwise missing data statistics
# rr: response-response pairs
# rm: response-missing pairs
# mr: missing-response pairs
# mm: missing-missing pairs
if (!(is.matrix(data) || is.data.frame(data))) {
stop("Data should be a matrix or dataframe")
}
if (ncol(data) < 2) {
stop("Data should have at least two columns")
}
r <- !is.na(data)
rr <- t(r) %*% r
mm <- t(!r) %*% (!r)
mr <- t(!r) %*% r
rm <- t(r) %*% (!r)
list(rr = rr, rm = rm, mr = mr, mm = mm)
}
mice/R/pool.scalar.R 0000644 0001762 0000144 00000007645 14334522175 013752 0 ustar ligges users #' Multiple imputation pooling: univariate version
#'
#' Pools univariate estimates of m repeated complete data analysis
#'
#' The function averages the univariate estimates of the complete data model,
#' computes the total variance over the repeated analyses, and computes the
#' relative increase in variance due to missing data or data synthesisation
#' and the fraction of missing information.
#'
#' @param Q A vector of univariate estimates of \code{m} repeated complete data
#' analyses.
#' @param U A vector containing the corresponding \code{m} variances of the univariate
#' estimates.
#' @param n A number providing the sample size. If nothing is specified,
#' an infinite sample \code{n = Inf} is assumed.
#' @param k A number indicating the number of parameters to be estimated.
#' By default, \code{k = 1} is assumed.
#' @inheritParams pool
#' @return Returns a list with components.
#' \describe{
#' \item{\code{m}:}{Number of imputations.}
#' \item{\code{qhat}:}{The \code{m} univariate estimates of repeated complete-data analyses.}
#' \item{\code{u}:}{The corresponding \code{m} variances of the univariate estimates.}
#' \item{\code{qbar}:}{The pooled univariate estimate, formula (3.1.2) Rubin (1987).}
#' \item{\code{ubar}:}{The mean of the variances (i.e. the pooled within-imputation variance),
#' formula (3.1.3) Rubin (1987).}
#' \item{\code{b}:}{The between-imputation variance, formula (3.1.4) Rubin (1987).}
#' \item{\code{t}:}{The total variance of the pooled estimated, formula (3.1.5)
#' Rubin (1987).}
#' \item{\code{r}:}{The relative increase in variance due to nonresponse, formula
#' (3.1.7) Rubin (1987).}
#' \item{\code{df}:}{The degrees of freedom for t reference distribution by the
#' method of Barnard-Rubin (1999).}
#' \item{\code{fmi}:}{The fraction missing information due to nonresponse,
#' formula (3.1.10) Rubin (1987). (Not defined for synthetic data.)}
#' }
#' @author Karin Groothuis-Oudshoorn and Stef van Buuren, 2009; Thom Volker, 2021
#' @seealso \code{\link{pool}}
#' @references
#' Rubin, D.B. (1987). Multiple Imputation for Nonresponse in
#' Surveys. New York: John Wiley and Sons.
#'
#' Reiter, J.P. (2003). Inference for Partially Synthetic,
#' Public Use Microdata Sets. \emph{Survey Methodology}, \bold{29}, 181-189.
#' @examples
#' # missing data imputation with with manual pooling
#' imp <- mice(nhanes, maxit = 2, m = 2, print = FALSE, seed = 18210)
#' fit <- with(data = imp, lm(bmi ~ age))
#'
#' # manual pooling
#' summary(fit$analyses[[1]])
#' summary(fit$analyses[[2]])
#' pool.scalar(Q = c(-1.5457, -1.428), U = c(0.9723^2, 1.041^2), n = 25, k = 2)
#'
#' # check: automatic pooling using broom
#' pool(fit)
#'
#' # manual pooling for synthetic data created from complete data
#' imp <- mice(cars,
#' maxit = 2, m = 2, print = FALSE, seed = 18210,
#' where = matrix(TRUE, nrow(cars), ncol(cars))
#' )
#' fit <- with(data = imp, lm(speed ~ dist))
#'
#' # manual pooling: extract Q and U
#' summary(fit$analyses[[1]])
#' summary(fit$analyses[[2]])
#' pool.scalar.syn(Q = c(0.12182, 0.13209), U = c(0.02121^2, 0.02516^2), n = 50, k = 2)
#'
#' # check: automatic pooling using broom
#' pool.syn(fit)
#' @export
pool.scalar <- function(Q, U, n = Inf, k = 1, rule = c("rubin1987", "reiter2003")) {
rule <- match.arg(rule)
m <- length(Q)
qbar <- mean(Q)
ubar <- mean(U)
b <- var(Q)
r <- (1 + 1 / m) * b / ubar
if (rule == "rubin1987") {
t <- ubar + (m + 1) * b / m
df <- barnard.rubin(m, b, t, dfcom = n - k)
fmi <- (r + 2 / (df + 3)) / (r + 1)
}
if (rule == "reiter2003") {
t <- ubar + b / m
df <- (m - 1) * (1 + (ubar / (b / m)))^2
fmi <- NA_real_
}
list(
m = m, qhat = Q, u = U, qbar = qbar, ubar = ubar, b = b, t = t,
df = df, r = r, fmi = fmi
)
}
#' @rdname pool.scalar
#' @export
pool.scalar.syn <- function(Q, U, n = Inf, k = 1, rule = "reiter2003") {
pool.scalar(Q, U, n = Inf, k = 1, rule = rule)
}
mice/R/windspeed.R 0000644 0001762 0000144 00000002541 14330031606 013473 0 ustar ligges users #' Subset of Irish wind speed data
#'
#' Subset of Irish wind speed data
#'
#' The original data set is much larger and was analyzed in detail by Haslett
#' and Raftery (1989). Van Buuren et al (2006) used this subset to investigate
#' the influence of extreme MAR mechanisms on the quality of imputation.
#'
#' @name windspeed
#' @docType data
#' @format A data frame with 433 rows and 6 columns containing the daily average
#' wind speeds within the period 1961-1978 at meteorological stations in the
#' Republic of Ireland. The data are a random sample from a larger data set.
#' \describe{
#' \item{RochePt}{Roche Point}
#' \item{Rosslare}{Rosslare}
#' \item{Shannon}{Shannon}
#' \item{Dublin}{Dublin}
#' \item{Clones}{Clones}
#' \item{MalinHead}{Malin Head} }
#' @references Haslett, J. and Raftery, A. E. (1989). \emph{Space-time
#' Modeling with Long-memory Dependence: Assessing Ireland's Wind Power
#' Resource (with Discussion)}. Applied Statistics 38, 1-50.
#' \url{http://lib.stat.cmu.edu/datasets/wind.desc} and
#' \url{http://lib.stat.cmu.edu/datasets/wind.data}
#'
#' van Buuren, S., Brand, J.P.L., Groothuis-Oudshoorn C.G.M., Rubin, D.B. (2006)
#' Fully conditional specification in multivariate imputation. \emph{Journal of
#' Statistical Computation and Simulation}, \bold{76}, 12, 1049--1064.
#' @keywords datasets
#' @examples
#'
#' windspeed[1:3, ]
NULL
mice/R/mice.impute.2l.norm.R 0000644 0001762 0000144 00000012360 14334522175 015230 0 ustar ligges users #' Imputation by a two-level normal model
#'
#' Imputes univariate missing data using a two-level normal model
#'
#' Implements the Gibbs sampler for the linear multilevel model with
#' heterogeneous with-class variance (Kasim and Raudenbush, 1998). Imputations
#' are drawn as an extra step to the algorithm. For simulation work see Van
#' Buuren (2011).
#'
#' The random intercept is automatically added in \code{mice.impute.2L.norm()}.
#' A model within a random intercept can be specified by \code{mice(...,
#' intercept = FALSE)}.
#'
#' @name mice.impute.2l.norm
#' @inheritParams mice.impute.pmm
#' @param type Vector of length \code{ncol(x)} identifying random and class
#' variables. Random variables are identified by a '2'. The class variable
#' (only one is allowed) is coded as '-2'. Random variables also include the
#' fixed effect.
#' @param intercept Logical determining whether the intercept is automatically
#' added.
#' @param ... Other named arguments.
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @note Added June 25, 2012: The currently implemented algorithm does not
#' handle predictors that are specified as fixed effects (type=1). When using
#' \code{mice.impute.2l.norm()}, the current advice is to specify all predictors
#' as random effects (type=2).
#'
#' Warning: The assumption of heterogeneous variances requires that in every
#' class at least one observation has a response in \code{y}.
#' @author Roel de Jong, 2008
#' @references
#'
#' Kasim RM, Raudenbush SW. (1998). Application of Gibbs sampling to nested
#' variance components models with heterogeneous within-group variance. Journal
#' of Educational and Behavioral Statistics, 23(2), 93--116.
#'
#' Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}: Multivariate
#' Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
#' Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
#'
#' Van Buuren, S. (2011) Multiple imputation of multilevel data. In Hox, J.J.
#' and and Roberts, J.K. (Eds.), \emph{The Handbook of Advanced Multilevel
#' Analysis}, Chapter 10, pp. 173--196. Milton Park, UK: Routledge.
#' @family univariate-2l
#' @keywords datagen
#' @export
mice.impute.2l.norm <- function(y, ry, x, type, wy = NULL, intercept = TRUE, ...) {
if (intercept) {
x <- cbind(1, as.matrix(x))
type <- c(2, type)
}
## Initialize
n.iter <- 100
if (is.null(wy)) wy <- !ry
n.class <- length(unique(x[, type == -2]))
if (n.class == 0) stop("No class variable")
gf.full <- factor(x[, type == -2], labels = seq_len(n.class))
gf <- gf.full[ry]
XG <- split.data.frame(as.matrix(x[ry, type == 2]), gf)
X.SS <- lapply(XG, crossprod)
yg <- split(as.vector(y[ry]), gf)
n.g <- tabulate(gf)
n.rc <- ncol(XG[[1]])
bees <- matrix(0, nrow = n.class, ncol = n.rc)
ss <- vector(mode = "numeric", length = n.class)
mu <- rep.int(0, n.rc)
inv.psi <- diag(1, n.rc, n.rc)
inv.sigma2 <- rep.int(1, n.class)
sigma2.0 <- 1
theta <- 1
## Execute Gibbs sampler
for (iter in seq_len(n.iter)) {
## Draw bees
for (class in seq_len(n.class)) {
vv <- symridge(inv.sigma2[class] * X.SS[[class]] + inv.psi, ...)
bees.var <- chol2inv(chol(vv))
bees[class, ] <- drop(bees.var %*% (crossprod(inv.sigma2[class] * XG[[class]], yg[[class]]) + inv.psi %*% mu)) +
drop(rnorm(n = n.rc) %*% chol(symridge(bees.var, ...)))
ss[class] <- crossprod(yg[[class]] - XG[[class]] %*% bees[class, ])
}
## Draw mu
mu <- colMeans(bees) + drop(rnorm(n = n.rc) %*%
chol(chol2inv(chol(symridge(inv.psi, ...))) / n.class))
## Draw psi
inv.psi <- rwishart(
df = n.class - n.rc - 1,
SqrtSigma = chol(chol2inv(chol(symridge(crossprod(t(t(bees) - mu)), ...))))
)
## Draw sigma2
inv.sigma2 <- rgamma(n.class, n.g / 2 + 1 / (2 * theta), scale = 2 * theta / (ss * theta + sigma2.0))
## Draw sigma2.0
H <- 1 / mean(inv.sigma2) # Harmonic mean
sigma2.0 <- rgamma(1, n.class / (2 * theta) + 1, scale = 2 * theta * H / n.class)
## Draw theta
G <- exp(mean(log(1 / inv.sigma2))) # Geometric mean
theta <- 1 / rgamma(1, n.class / 2 - 1, scale = 2 / (n.class * (sigma2.0 / H - log(sigma2.0) + log(G) - 1)))
}
## Generate imputations
imps <- rnorm(n = sum(wy), sd = sqrt(1 / inv.sigma2[gf.full[wy]])) + rowSums(as.matrix(x[wy, type == 2, drop = FALSE]) * bees[gf.full[wy], ])
imps
}
rwishart <- function(df, p = nrow(SqrtSigma), SqrtSigma = diag(p)) {
## rwishart, written by Bill Venables
Z <- matrix(0, p, p)
diag(Z) <- sqrt(rchisq(p, df:(df - p + 1)))
if (p > 1) {
pseq <- seq_len(p - 1)
Z[rep(p * pseq, pseq) + unlist(lapply(pseq, seq))] <- rnorm(p * (p - 1) / 2)
}
crossprod(Z %*% SqrtSigma)
}
force.chol <- function(x, warn = TRUE) {
z <- 0
repeat {
lambda <- 0.1 * z
XT <- x + diag(x = lambda, nrow = nrow(x))
XT <- (XT + t(XT)) / 2
s <- try(expr = chol(XT), silent = TRUE)
if (!inherits(s, "try-error")) {
break
}
z <- z + 1
}
attr(s, "forced") <- (z > 0)
if (warn && z > 0) {
warning("Cholesky decomposition had to be forced", call. = FALSE)
}
s
}
symridge <- function(x, ridge = 0.0001, ...) {
x <- (x + t(x)) / 2
if (nrow(x) == 1L) {
return(x)
}
x + diag(diag(x) * ridge)
}
mice/R/where.R 0000644 0001762 0000144 00000004204 14334522175 012633 0 ustar ligges users #' Creates a \code{where} argument
#'
#' This helper function creates a valid \code{where} matrix. The
#' \code{where} matrix is an argument to the \code{mice} function.
#' It has the same size as \code{data} and specifies which values
#' are to be imputed (\code{TRUE}) or nor (\code{FALSE}).
#' @param data A \code{data.frame} with the source data
#' @param keyword An optional keyword, one of \code{"missing"} (missing
#' values are imputed), \code{"observed"} (observed values are imputed),
#' \code{"all"} and \code{"none"}. The default
#' is \code{keyword = "missing"}
#' @return A matrix with logical
#' @seealso \code{\link{make.blocks}}, \code{\link{make.predictorMatrix}}
#' @examples
#' head(make.where(nhanes), 3)
#'
#' # create & analyse synthetic data
#' where <- make.where(nhanes2, "all")
#' imp <- mice(nhanes2,
#' m = 10, where = where,
#' print = FALSE, seed = 123
#' )
#' fit <- with(imp, lm(chl ~ bmi + age + hyp))
#' summary(pool.syn(fit))
#' @export
make.where <- function(data,
keyword = c("missing", "all", "none", "observed")) {
keyword <- match.arg(keyword)
data <- check.dataform(data)
where <- switch(keyword,
missing = is.na(data),
all = matrix(TRUE, nrow = nrow(data), ncol = ncol(data)),
none = matrix(FALSE, nrow = nrow(data), ncol = ncol(data)),
observed = !is.na(data)
)
dimnames(where) <- dimnames(data)
where
}
check.where <- function(where, data, blocks) {
if (is.null(where)) {
where <- make.where(data, keyword = "missing")
}
if (!(is.matrix(where) || is.data.frame(where))) {
if (is.character(where)) {
return(make.where(data, keyword = where))
} else {
stop("Argument `where` not a matrix or data frame", call. = FALSE)
}
}
if (!all(dim(data) == dim(where))) {
stop("Arguments `data` and `where` not of same size", call. = FALSE)
}
where <- as.logical(as.matrix(where))
if (anyNA(where)) {
stop("Argument `where` contains missing values", call. = FALSE)
}
where <- matrix(where, nrow = nrow(data), ncol = ncol(data))
dimnames(where) <- dimnames(data)
where[, !colnames(where) %in% unlist(blocks)] <- FALSE
where
}
mice/R/ampute.discrete.R 0000644 0001762 0000144 00000010727 14334522175 014624 0 ustar ligges users #' Multivariate amputation based on discrete probability functions
#'
#' This function creates a missing data indicator for each pattern. Odds probabilities
#' (Brand, 1999, pp. 110-113) will be induced on the weighted sum scores, calculated earlier
#' in the multivariate amputation function \code{\link{ampute}}.
#'
#' @param P A vector containing the pattern numbers of candidates.
#' For each case, a value between 1 and #patterns is given. For example, a
#' case with value 2 is candidate for missing data pattern 2.
#' @param scores A list containing vectors with the candidates's weighted sum scores,
#' the result of an underlying function in \code{\link{ampute}}.
#' @param prop A scalar specifying the proportion of missingness. Should be a value
#' between 0 and 1. Default is a missingness proportion of 0.5.
#' @param odds A matrix where #patterns defines the #rows. Each row should contain
#' the odds of being missing for the corresponding pattern. The amount of odds values
#' defines in how many quantiles the sum scores will be divided. The values are
#' relative probabilities: a quantile with odds value 4 will have a probability of
#' being missing that is four times higher than a quantile with odds 1. The
#' #quantiles may differ between the patterns, specify NA for cells remaining empty.
#' Default is 4 quantiles with odds values 1, 2, 3 and 4, the result of
#' \code{\link{ampute.default.odds}}.
#' @return A list containing vectors with \code{0} if a case should be made missing
#' and \code{1} if a case should remain complete. The first vector refers to the
#' first pattern, the second vector to the second pattern, etcetera.
#' @author Rianne Schouten, 2016
#' @seealso \code{\link{ampute}}, \code{\link{ampute.default.odds}}
#' @references Brand, J.P.L. (1999). \emph{Development, implementation and
#' evaluation of multiple imputation strategies for the statistical analysis of
#' incomplete data sets.} Dissertation. Rotterdam: Erasmus University.
#' @keywords internal
#' @export
ampute.discrete <- function(P, scores, prop, odds) {
R <- vector(mode = "list", length = nrow(odds))
for (i in seq_len(nrow(odds))) {
if (scores[[i]][[1]] == 0) {
R[[i]] <- 0
} else {
# The scores are divided into quantiles
# Specify #quantiles by #odds values
ng <- length(odds[i, ][!is.na(odds[i, ])])
quantiles <- quantile(scores[[i]], probs = seq.int(0, 1, by = 1 / ng))
if (anyDuplicated(quantiles) || anyNA(quantiles)) {
stop("Division of sum scores into quantiles did not succeed. Possibly
the sum scores contain too few different observations (in case of
categorical or dummy variables). Try using more variables to
calculate the sum scores or diminish the number of quantiles in the
odds matrix", call. = FALSE)
}
# For each candidate the quantile number is specified
R.temp <- rep.int(NA, length(scores[[i]]))
for (k in seq_len(ng)) {
R.temp <- replace(R.temp, scores[[i]] >= quantiles[k] &
scores[[i]] <= quantiles[k + 1], k)
}
# For each candidate, a random value between 0 and 1 is compared with the
# odds probability of being missing. If random value <= prob, the candidate
# will receive missing data indicator 0, meaning he will be made missing
# according the pattern; if random value > prob, the candidate will receive
# missing data indicator 1, meaning the candidate will remain complete.
for (l in seq_len(ng)) {
prob <- (ng * prop * odds[i, l]) / sum(odds[i, ], na.rm = TRUE)
if (prob >= 1.0) {
warning("Combination of odds matrix and desired proportion of
missingness results to small quantile groups, probably
decreasing the obtained proportion of missingness",
call. = FALSE
)
}
gn <- length(R.temp[R.temp == l])
if (gn != 0) {
random <- runif(n = gn, min = 0, max = 1)
Q <- c()
for (m in seq_len(gn)) {
if (random[m] <= prob) {
Q[m] <- 0 # Candidate will be made missing
} else {
Q[m] <- 1 # Candidate will be kept complete
}
}
# Give the result to the right candidate
R.temp <- replace(R.temp, R.temp == l, Q)
}
}
# Give the result to the right cases in the data
R[[i]] <- replace(P, P == (i + 1), R.temp)
R[[i]] <- replace(R[[i]], P != (i + 1), 1)
}
}
R
}
mice/R/nhanes.R 0000644 0001762 0000144 00000001556 14330031606 012772 0 ustar ligges users #' NHANES example - all variables numerical
#'
#' A small data set with non-monotone missing values.
#'
#' A small data set with all numerical variables. The data set \code{nhanes2} is
#' the same data set, but with \code{age} and \code{hyp} treated as factors.
#'
#' @name nhanes
#' @docType data
#' @format A data frame with 25 observations on the following 4 variables.
#' \describe{
#' \item{age}{Age group (1=20-39, 2=40-59, 3=60+)}
#' \item{bmi}{Body mass index (kg/m**2)}
#' \item{hyp}{Hypertensive (1=no,2=yes)}
#' \item{chl}{Total serum cholesterol (mg/dL)} }
#' @seealso \code{\link{nhanes2}}
#' @source Schafer, J.L. (1997). \emph{Analysis of Incomplete Multivariate
#' Data.} London: Chapman & Hall. Table 6.14.
#' @keywords datasets
#' @examples
#' # create 5 imputed data sets
#' imp <- mice(nhanes)
#'
#' # print the first imputed data set
#' complete(imp)
NULL
mice/R/convergence.R 0000644 0001762 0000144 00000011232 14433400023 014001 0 ustar ligges users #' Computes convergence diagnostics for a \code{mids} object
#'
#' Takes an object of class \code{mids}, computes the autocorrelation
#' and/or potential scale reduction factor, and returns a \code{data.frame}
#' with the specified diagnostic(s) per iteration.
#'
#' @param data An object of class \code{mids} as created by the function
#' \code{mice()}.
#' @param diagnostic A keyword. One of the following keywords: \code{"ac"},
#' \code{"all"}, \code{"gr"} and \code{"psrf"}. See the Details section
#' for the interpretation.
#' The default is \code{diagnostic = "all"} which returns both the
#' autocorrelation and potential scale reduction factor per iteration.
#' @param parameter A keyword. One of the following keywords: \code{"mean"}
#' or \code{"sd"} to evaluate chain means or chain standard deviations,
#' respectively.
#' @param \dots Additional arguments. Not used.
#' @return A \code{data.frame} with the autocorrelation and/or potential
#' scale reduction factor per iteration of the MICE algorithm.
#' @details
#' The argument \code{diagnostic} can be length-1 character, which is
#' matched to one of the following keywords:
#' \describe{
#' \item{\code{"all"}}{computes both the lag-1 autocorrelation as well as
#' the potential scale reduction factor (cf. Vehtari et al., 2021) per
#' iteration of the MICE algorithm;}
#' \item{\code{"ac"}}{computes only the autocorrelation per iteration;}
#' \item{\code{"psrf"}}{computes only the potential scale reduction factor
#' per iteration;}
#' \item{\code{"gr"}}{same as \code{psrf}, the potential scale reduction
#' factor is colloquially called the Gelman-Rubin diagnostic.}
#' }
#' In the unlikely event of perfect convergence, the autocorrelation equals
#' zero and the potential scale reduction factor equals one. To interpret
#' the convergence diagnostic(s) in the output of the function, it is
#' recommended to plot the diagnostics (ac and/or psrf) against the
#' iteration number (.it) per imputed variable (vrb). A persistently
#' decreasing trend across iterations indicates potential non-convergence.
#'
#' @seealso \code{\link{mice}}, \code{\link[=mids-class]{mids}}
#' @keywords none
#' @references Vehtari, A., Gelman, A., Simpson, D., Carpenter, B., & Burkner,
#' P.-C. (2021). Rank-Normalization, Folding, and Localization: An Improved
#' R for Assessing Convergence of MCMC. Bayesian Analysis, 1(1), 1-38.
#' https://doi.org/10.1214/20-BA1221
#' @examples
#' \dontrun{
#' # obtain imputed data set
#' imp <- mice(nhanes2, print = FALSE)
#' # compute convergence diagnostics
#' convergence(imp)
#' }
#' @export
convergence <- function(data, diagnostic = "all", parameter = "mean", ...) {
# process inputs
install.on.demand("rstan", ...)
install.on.demand("purrr", ...)
if (!is.mids(data)) {
stop("'data' not of class 'mids'")
}
if (data$m < 2) {
stop("the number of imputations should be at least two (m > 1)")
}
if (data$iteration < 3) {
stop("the number of iterations should be at least three (maxit > 2)")
}
if (is.null(data$chainMean)) {
stop("no convergence diagnostics found", call. = FALSE)
}
if (parameter != "mean" & parameter != "sd") {
stop("'parameter' not recognized, please use 'mean' or 'sd'")
}
vrbs <- names(data$data)
p <- length(vrbs)
m <- as.integer(data$m)
t <- as.integer(data$iteration)
out <- expand.grid(.it = seq_len(t), vrb = vrbs)
# extract chain means or chain standard deviations
if (parameter == "mean") {
param <- lapply(seq(p), function(x) {
aperm(data$chainMean[vrbs, , , drop = FALSE], c(2, 3, 1))[, , x]
})
}
if (parameter == "sd") {
param <- lapply(seq(p), function(x) {
aperm(sqrt(data$chainVar)[vrbs, , , drop = FALSE], c(2, 3, 1))[, , x]
})
}
names(param) <- vrbs
# compute autocorrelation
if (diagnostic == "all" | diagnostic == "ac") {
ac <-
purrr::map(vrbs, function(.vrb) {
c(NA, dplyr::cummean(dplyr::coalesce(
purrr::map_dbl(2:t, function(.itr) {
suppressWarnings(stats::cor(
param[[.vrb]][.itr - 1, ],
param[[.vrb]][.itr, ],
use = "pairwise.complete.obs"
))
}), 0
))) + 0 * param[[.vrb]][, 1]
})
out <- base::cbind(out, ac = unlist(ac))
}
# compute potential scale reduction factor
if (diagnostic == "all" | diagnostic == "psrf" | diagnostic == "gr") {
psrf <- purrr::map_dfr(param, ~ {
purrr::map_dfr(1:t, function(.itr) {
data.frame(psrf = rstan::Rhat(.[1:.itr, ]))
})
})
out <- base::cbind(out, psrf)
}
out[is.nan(out)] <- NA
return(out)
}
# function to extend is.nan() to data.frame objects
is.nan.data.frame <- function(x) do.call(cbind, lapply(x, is.nan))
mice/R/get.df.R 0000644 0001762 0000144 00000002456 14330031606 012665 0 ustar ligges users get.dfcom <- function(object, dfcom = NULL) {
# residual degrees of freedom of model fitted on hypothetically complete data
# assumed to be the same across imputations
if (!is.null(dfcom)) {
return(max(dfcom, 1L))
}
glanced <- get.glanced(object)
# try to extract from df.residual
if (!is.null(glanced)) {
if ("df.residual" %in% colnames(glanced)) {
return(glanced$df.residual[1L])
}
}
# try n - p (or nevent - p for Cox model)
if (!is.null(glanced)) {
if ("nobs" %in% colnames(glanced)) {
model <- getfit(object, 1L)
if (inherits(model, "coxph")) {
return(max(model$nevent - length(coef(model)), 1L))
}
return(max(glanced$nobs[1L] - length(coef(model)), 1L))
}
}
# not found
warning("Infinite sample size assumed.")
Inf
}
get.glanced <- function(object) {
if (!is.list(object)) stop("Argument 'object' not a list", call. = FALSE)
object <- as.mira(object)
glanced <- try(data.frame(summary(getfit(object), type = "glance")), silent = TRUE)
if (inherits(glanced, "data.frame")) {
# nobs is needed for pool.r.squared
# broom <= 0.5.6 does not supply it
if (!"nobs" %in% colnames(glanced)) {
glanced$nobs <- length(stats::residuals(object$analyses[[1]]))
}
} else {
glanced <- NULL
}
glanced
}
mice/R/mids2spss.R 0000644 0001762 0000144 00000006275 14330031606 013450 0 ustar ligges users #' Export \code{mids} object to SPSS
#'
#' Converts a \code{mids} object into a format recognized by SPSS, and writes
#' the data and the SPSS syntax files.
#'
#' This function automates most of the work needed to export a \code{mids}
#' object to SPSS. It uses \code{haven::write_sav()} to facilitate the export to an
#' SPSS \code{.sav} or \code{.zsav} file.
#'
#' Below are some things to pay attention to.
#'
#' The \code{SPSS} syntax file has the proper file names and separators set, so
#' in principle it should run and read the data without alteration. \code{SPSS}
#' is more strict than \code{R} with respect to the paths. Always use the full
#' path, otherwise \code{SPSS} may not be able to find the data file.
#'
#' Factors in \code{R} translate into categorical variables in \code{SPSS}. The
#' internal coding of factor levels used in \code{R} is exported. This is
#' generally acceptable for \code{SPSS}. However, when the data are to be
#' combined with existing \code{SPSS} data, watch out for any changes in the
#' factor levels codes.
#'
#' \code{SPSS} will recognize the data set as a multiply imputed data set, and
#' do automatic pooling in procedures where that is supported. Note however that
#' pooling is an extra option only available to those who license the
#' \code{MISSING VALUES} module. Without this license, \code{SPSS} will still
#' recognize the structure of the data, but it will not pool the multiply imputed
#' estimates into a single inference.
#'
#' @param imp The \code{imp} argument is an object of class \code{mids},
#' typically produced by the \code{mice()} function.
#' @param filename A character string describing the name of the output data
#' file and its extension.
#' @param path A character string containing the path of the output file. The
#' value in \code{path} is appended to \code{filedat}. By
#' default, files are written to the current \code{R} working directory. If
#' \code{path=NULL} then no file path appending is done.
#' @param compress A logical flag stating whether the resulting SPSS set should
#' be a compressed \code{.zsav} file.
#' @param silent A logical flag stating whether the location of the saved file should be
#' printed.
#' @return The return value is \code{NULL}.
#' @author Gerko Vink, dec 2020.
#' @seealso \code{\link[=mids-class]{mids}}
#' @keywords manip
#' @export
mids2spss <- function(imp, filename = "midsdata",
path = getwd(), compress = FALSE, silent = FALSE) {
.id <- NULL # avoid empty global variable binding
install.on.demand("haven")
# extract a completed dataset (long format - all imputations stacked)
# rename the .imp variable to imputation_, such that SPSS can identify a multiply imputed dataset
out <-
imp %>%
complete(action = "long", include = TRUE) %>%
dplyr::select(-.id) %>%
dplyr::rename("Imputation_" = ".imp")
# write the data to a .sav file with package haven and print (optional) the saved location
if (!compress) {
whereto <- paste(path, "/", filename, ".sav", sep = "")
} else {
whereto <- paste(path, "/", filename, ".zsav", sep = "")
}
haven::write_sav(data = out, path = whereto, compress = compress)
if (!silent) {
cat("SPSS file written to", whereto, "\n")
}
}
mice/R/densityplot.R 0000644 0001762 0000144 00000031243 14330031647 014075 0 ustar ligges users #' Density plot of observed and imputed data
#'
#' Plotting methods for imputed data using \pkg{lattice}. \code{densityplot}
#' produces plots of the densities. The function
#' automatically separates the observed and imputed data. The
#' functions extend the usual features of \pkg{lattice}.
#'
#' The argument \code{na.groups} may be used to specify (combinations of)
#' missingness in any of the variables. The argument \code{groups} can be used
#' to specify groups based on the variable values themselves. Only one of both
#' may be active at the same time. When both are specified, \code{na.groups}
#' takes precedence over \code{groups}.
#'
#' Use the \code{subset} and \code{na.groups} together to plots parts of the
#' data. For example, select the first imputed data set by by
#' \code{subset=.imp==1}.
#'
#' Graphical parameters like \code{col}, \code{pch} and \code{cex} can be
#' specified in the arguments list to alter the plotting symbols. If
#' \code{length(col)==2}, the color specification to define the observed and
#' missing groups. \code{col[1]} is the color of the 'observed' data,
#' \code{col[2]} is the color of the missing or imputed data. A convenient color
#' choice is \code{col=mdc(1:2)}, a transparent blue color for the observed
#' data, and a transparent red color for the imputed data. A good choice is
#' \code{col=mdc(1:2), pch=20, cex=1.5}. These choices can be set for the
#' duration of the session by running \code{mice.theme()}.
#'
#' @aliases densityplot
#' @param x A \code{mids} object, typically created by \code{mice()} or
#' \code{mice.mids()}.
#' @param data Formula that selects the data to be plotted. This argument
#' follows the \pkg{lattice} rules for \emph{formulas}, describing the primary
#' variables (used for the per-panel display) and the optional conditioning
#' variables (which define the subsets plotted in different panels) to be used
#' in the plot.
#'
#' The formula is evaluated on the complete data set in the \code{long} form.
#' Legal variable names for the formula include \code{names(x$data)} plus the
#' two administrative factors \code{.imp} and \code{.id}.
#'
#' \bold{Extended formula interface:} The primary variable terms (both the LHS
#' \code{y} and RHS \code{x}) may consist of multiple terms separated by a
#' \sQuote{+} sign, e.g., \code{y1 + y2 ~ x | a * b}. This formula would be
#' taken to mean that the user wants to plot both \code{y1 ~ x | a * b} and
#' \code{y2 ~ x | a * b}, but with the \code{y1 ~ x} and \code{y2 ~ x} in
#' \emph{separate panels}. This behavior differs from standard \pkg{lattice}.
#' \emph{Only combine terms of the same type}, i.e. only factors or only
#' numerical variables. Mixing numerical and categorical data occasionally
#' produces odds labeling of vertical axis.
#'
#' The function \code{densityplot} does not use the \code{y} terms in the
#' formula. Density plots for \code{x1} and \code{x2} are requested as \code{~
#' x1 + x2}.
#' @param na.groups An expression evaluating to a logical vector indicating
#' which two groups are distinguished (e.g. using different colors) in the
#' display. The environment in which this expression is evaluated in the
#' response indicator \code{is.na(x$data)}.
#'
#' The default \code{na.group = NULL} contrasts the observed and missing data
#' in the LHS \code{y} variable of the display, i.e. groups created by
#' \code{is.na(y)}. The expression \code{y} creates the groups according to
#' \code{is.na(y)}. The expression \code{y1 & y2} creates groups by
#' \code{is.na(y1) & is.na(y2)}, and \code{y1 | y2} creates groups as
#' \code{is.na(y1) | is.na(y2)}, and so on.
#' @param groups This is the usual \code{groups} arguments in \pkg{lattice}. It
#' differs from \code{na.groups} because it evaluates in the completed data
#' \code{data.frame(complete(x, "long", inc=TRUE))} (as usual), whereas
#' \code{na.groups} evaluates in the response indicator. See
#' \code{\link{xyplot}} for more details. When both \code{na.groups} and
#' \code{groups} are specified, \code{na.groups} takes precedence, and
#' \code{groups} is ignored.
#' @param plot.points A logical used in \code{densityplot} that signals whether
#' the points should be plotted.
#' @param theme A named list containing the graphical parameters. The default
#' function \code{mice.theme} produces a short list of default colors, line
#' width, and so on. The extensive list may be obtained from
#' \code{trellis.par.get()}. Global graphical parameters like \code{col} or
#' \code{cex} in high-level calls are still honored, so first experiment with
#' the global parameters. Many setting consists of a pair. For example,
#' \code{mice.theme} defines two symbol colors. The first is for the observed
#' data, the second for the imputed data. The theme settings only exist during
#' the call, and do not affect the trellis graphical parameters.
#' @param mayreplicate A logical indicating whether color, line widths, and so
#' on, may be replicated. The graphical functions attempt to choose
#' "intelligent" graphical parameters. For example, the same color can be
#' replicated for different element, e.g. use all reds for the imputed data.
#' Replication may be switched off by setting the flag to \code{FALSE}, in order
#' to allow the user to gain full control.
#' @param thicker Used in \code{densityplot}. Multiplication factor of the line
#' width of the observed density. \code{thicker=1} uses the same thickness for
#' the observed and imputed data.
#' @param as.table See \code{\link[lattice:xyplot]{xyplot}}.
#' @param panel See \code{\link{xyplot}}.
#' @param default.prepanel See \code{\link[lattice:xyplot]{xyplot}}.
#' @param outer See \code{\link[lattice:xyplot]{xyplot}}.
#' @param allow.multiple See \code{\link[lattice:xyplot]{xyplot}}.
#' @param drop.unused.levels See \code{\link[lattice:xyplot]{xyplot}}.
#' @param subscripts See \code{\link[lattice:xyplot]{xyplot}}.
#' @param subset See \code{\link[lattice:xyplot]{xyplot}}.
#' @param \dots Further arguments, usually not directly processed by the
#' high-level functions documented here, but instead passed on to other
#' functions.
#' @return The high-level functions documented here, as well as other high-level
#' Lattice functions, return an object of class \code{"trellis"}. The
#' \code{\link[lattice:update.trellis]{update}} method can be used to
#' subsequently update components of the object, and the
#' \code{\link[lattice:print.trellis]{print}} method (usually called by default)
#' will plot it on an appropriate plotting device.
#' @note The first two arguments (\code{x} and \code{data}) are reversed
#' compared to the standard Trellis syntax implemented in \pkg{lattice}. This
#' reversal was necessary in order to benefit from automatic method dispatch.
#'
#' In \pkg{mice} the argument \code{x} is always a \code{mids} object, whereas
#' in \pkg{lattice} the argument \code{x} is always a formula.
#'
#' In \pkg{mice} the argument \code{data} is always a formula object, whereas in
#' \pkg{lattice} the argument \code{data} is usually a data frame.
#'
#' All other arguments have identical interpretation.
#'
#' \code{densityplot} errs on empty groups, which occurs if all observations in
#' the subgroup contain \code{NA}. The relevant error message is: \code{Error in
#' density.default: ... need at least 2 points to select a bandwidth
#' automatically}. There is yet no workaround for this problem. Use the more
#' robust \code{bwplot} or \code{stripplot} as a replacement.
#' @author Stef van Buuren
#' @seealso \code{\link{mice}}, \code{\link{xyplot}}, \code{\link{stripplot}},
#' \code{\link{bwplot}}, \code{\link{lattice}} for an overview of the
#' package, as well as \code{\link[lattice:histogram]{densityplot}},
#' \code{\link[lattice:panel.densityplot]{panel.densityplot}},
#' \code{\link[lattice:print.trellis]{print.trellis}},
#' \code{\link[lattice:trellis.par.get]{trellis.par.set}}
#' @references Sarkar, Deepayan (2008) \emph{Lattice: Multivariate Data
#' Visualization with R}, Springer.
#'
#' van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}: Multivariate
#' Imputation by Chained Equations in \code{R}. \emph{Journal of Statistical
#' Software}, \bold{45}(3), 1-67. \doi{10.18637/jss.v045.i03}
#' @keywords hplot
#' @examples
#' imp <- mice(boys, maxit = 1)
#'
#' ### density plot of head circumference per imputation
#' ### blue is observed, red is imputed
#' densityplot(imp, ~ hc | .imp)
#'
#' ### All combined in one panel.
#' densityplot(imp, ~hc)
#' @export
densityplot.mids <- function(x,
data,
na.groups = NULL,
groups = NULL,
as.table = TRUE,
plot.points = FALSE,
theme = mice.theme(),
mayreplicate = TRUE,
thicker = 2.5,
allow.multiple = TRUE,
outer = TRUE,
drop.unused.levels = lattice::lattice.getOption("drop.unused.levels"),
panel = lattice::lattice.getOption("panel.densityplot"),
default.prepanel = lattice::lattice.getOption("prepanel.default.densityplot"),
...,
subscripts = TRUE,
subset = TRUE) {
call <- match.call()
if (!is.mids(x)) stop("Argument 'x' must be a 'mids' object")
## unpack data and response indicator
cd <- data.frame(complete(x, "long", include = TRUE))
r <- as.data.frame(is.na(x$data))
## evaluate na.group in response indicator
nagp <- eval(expr = substitute(na.groups), envir = r, enclos = parent.frame())
if (is.expression(nagp)) nagp <- eval(expr = nagp, envir = r, enclos = parent.frame())
## evaluate groups in imputed data
ngp <- eval(expr = substitute(groups), envir = cd, enclos = parent.frame())
if (is.expression(ngp)) ngp <- eval(expr = ngp, envir = cd, enclos = parent.frame())
groups <- ngp
## evaluate subset in imputed data
ss <- eval(expr = substitute(subset), envir = cd, enclos = parent.frame())
if (is.expression(ss)) ss <- eval(expr = ss, envir = cd, enclos = parent.frame())
subset <- ss
## evaluate further arguments before parsing
dots <- list(...)
args <- list(
panel = panel,
default.prepanel = default.prepanel,
allow.multiple = allow.multiple,
outer = outer,
drop.unused.levels = drop.unused.levels,
subscripts = subscripts,
as.table = as.table,
plot.points = plot.points
)
## create formula if not given (in call$data !)
vnames <- names(cd)[-seq_len(2)]
allfactors <- vapply(cd, is.factor, logical(1))[-seq_len(2)]
if (missing(data)) {
vnames <- vnames[!allfactors & x$nmis > 2 & x$nmis < nrow(x$data) - 1]
formula <- as.formula(paste("~", paste(vnames, collapse = "+", sep = ""), sep = ""))
} else {
formula <- data
}
## determine the y-variables
form <- lattice::latticeParseFormula(
model = formula, data = cd, subset = subset,
groups = groups, multiple = allow.multiple,
outer = outer, subscripts = TRUE,
drop = drop.unused.levels
)
xnames <- unlist(lapply(strsplit(form$right.name, " \\+ "), rm.whitespace)) ## Jul2011
## calculate selection vector gp
nona <- is.null(call$na.groups)
if (!is.null(call$groups) && nona) {
gp <- call$groups
} else {
if (nona) {
for (i in seq_along(xnames)) {
xvar <- xnames[i]
select <- cd$.imp != 0 & !r[, xvar]
cd[select, xvar] <- NA
}
gp <- rep.int(cd$.imp, length(xnames))
} else {
for (i in seq_along(xnames)) {
xvar <- xnames[i]
select <- cd$.imp != 0 & !nagp
cd[select, xvar] <- NA
}
gp <- rep.int(cd$.imp, length(xnames))
}
}
## replicate color 2 if group=.imp is part of xnames
mustreplicate <- !(!is.null(call$groups) && nona) && mayreplicate
if (mustreplicate) {
theme$superpose.line$col <- rep(theme$superpose.line$col[seq_len(2)], c(1, x$m))
theme$superpose.line$lwd <- rep(c(theme$superpose.line$lwd[1] * thicker, theme$superpose.line$lwd[1]), c(1, x$m))
theme$superpose.symbol$col <- rep(theme$superpose.symbol$col[seq_len(2)], c(1, x$m))
theme$superpose.symbol$pch <- c(NA, 49:(49 + x$m - 1))
}
## change axis defaults of extended formula interface
if (is.null(call$xlab)) {
args$xlab <- ""
if (length(xnames) == 1) args$xlab <- xnames
}
if (is.null(call$scales)) {
args$scales <- list()
if (length(xnames) > 1) {
args$scales <- list(x = list(relation = "free"), y = list(relation = "free"))
}
}
## ready
args <- c(
x = formula, data = list(cd),
groups = list(gp),
args, dots, subset = call$subset
)
## go
tp <- do.call("densityplot", args)
tp <- update(tp, par.settings = theme)
return(tp)
}
mice/R/filter.R 0000644 0001762 0000144 00000011156 14334447167 013021 0 ustar ligges users #' @importFrom dplyr filter
#' @export
dplyr::filter
#' Subset rows of a \code{mids} object
#'
#' This function takes a \code{mids} object and returns a new
#' \code{mids} object that pertains to the subset of the data
#' identified by the expression in \dots. The expression may use
#' column values from the incomplete data in \code{.data$data}.
#'
#' @param .data A \code{mids} object.
#' @param ... Expressions that return a
#' logical value, and are defined in terms of the variables in \code{.data$data}.
#' If multiple expressions are specified, they are combined with the \code{&} operator.
#' Only rows for which all conditions evaluate to \code{TRUE} are kept.
#' @inheritParams dplyr::filter
#' @seealso \code{\link[dplyr]{filter}}
#' @return An S3 object of class \code{mids}
#' @note The function calculates a logical vector \code{include} of length \code{nrow(.data$data)}.
#' The function constructs the elements of the filtered \code{mids} object as follows:
#' \tabular{ll}{
#' \code{data} \tab Select rows in \code{.data$data} for which \code{include == TRUE}\cr
#' \code{imp} \tab Select rows each imputation \code{data.frame} in \code{.data$imp} for which \code{include == TRUE}\cr
#' \code{m} \tab Equals \code{.data$m}\cr
#' \code{where} \tab Select rows in \code{.data$where} for which \code{include == TRUE}\cr
#' \code{blocks} \tab Equals \code{.data$blocks}\cr
#' \code{call} \tab Equals \code{.data$call}\cr
#' \code{nmis} \tab Recalculate \code{nmis} based on the selected \code{data} rows\cr
#' \code{method} \tab Equals \code{.data$method}\cr
#' \code{predictorMatrix} \tab Equals \code{.data$predictorMatrix}\cr
#' \code{visitSequence} \tab Equals \code{.data$visitSequence}\cr
#' \code{formulas} \tab Equals \code{.data$formulas}\cr
#' \code{post} \tab Equals \code{.data$post}\cr
#' \code{blots} \tab Equals \code{.data$blots}\cr
#' \code{ignore} \tab Select positions in \code{.data$ignore} for which \code{include == TRUE}\cr
#' \code{seed} \tab Equals \code{.data$seed}\cr
#' \code{iteration} \tab Equals \code{.data$iteration}\cr
#' \code{lastSeedValue} \tab Equals \code{.data$lastSeedValue}\cr
#' \code{chainMean} \tab Set to \code{NULL}\cr
#' \code{chainVar} \tab Set to \code{NULL}\cr
#' \code{loggedEvents} \tab Equals \code{.data$loggedEvents}\cr
#' \code{version} \tab Replaced with current version\cr
#' \code{date} \tab Replaced with current date
#' }
#' @author Patrick Rockenschaub
#' @keywords manip
#' @examples
#' imp <- mice(nhanes, m = 2, maxit = 1, print = FALSE)
#'
#' # example with external logical vector
#' imp_f <- filter(imp, c(rep(TRUE, 13), rep(FALSE, 12)))
#'
#' nrow(complete(imp))
#' nrow(complete(imp_f))
#'
#' # example with calculated include vector
#' imp_f2 <- filter(imp, age >= 2 & hyp == 1)
#' nrow(complete(imp_f2)) # should be 5
#' @export
filter.mids <- function(.data, ..., .preserve = FALSE) {
if (!is.mids(.data)) {
stop("Argument `.data` should be of class mids.")
}
rows <- .data$data %>%
mutate(.rownumber = row_number()) %>%
filter(...) %>%
pull(".rownumber")
include <- 1L:nrow(.data$data) %in% rows
# Components that stay the same after filtering
m <- .data$m
call <- .data$call
blocks <- .data$blocks
method <- .data$method
predictorMatrix <- .data$predictorMatrix
visitSequence <- .data$visitSequence
formulas <- .data$formulas
blots <- .data$blots
post <- .data$post
seed <- .data$seed
iteration <- .data$iteration
lastSeedValue <- .data$lastSeedValue
loggedEvents <- .data$loggedEvents
# Components that need to be subset
data <- .data$data[include, ]
ignore <- .data$ignore[include]
where <- .data$where[include, ]
imp <- vector("list", length(.data$imp))
names(imp) <- names(.data$imp)
for (i in names(.data$imp)) {
wy <- .data$where[, i]
iy <- .data$where[, i] & include
imp[[i]] <- .data$imp[[i]][iy[wy], , drop = FALSE]
}
# Components that need to be recalculated/reset
nmis <- colSums(is.na(data))
chainMean <- NULL
chainVar <- NULL
# Create subset mids object
midsobj <- list(
data = data, imp = imp, m = m,
where = where, blocks = blocks,
call = call, nmis = nmis,
method = method,
predictorMatrix = predictorMatrix,
visitSequence = visitSequence,
formulas = formulas,
post = post,
blots = blots,
ignore = ignore,
seed = seed,
iteration = iteration,
lastSeedValue = lastSeedValue,
chainMean = chainMean,
chainVar = chainVar,
loggedEvents = loggedEvents,
version = packageVersion("mice"),
date = Sys.Date()
)
oldClass(midsobj) <- "mids"
midsobj
}
mice/R/internal.R 0000644 0001762 0000144 00000007374 14334522175 013350 0 ustar ligges users keep.in.model <- function(y, ry, x, wy) {
(complete.cases(y, x) & ry) | (complete.cases(x) & wy)
}
impute.with.na <- function(x, wy) !complete.cases(x) & wy
check.df <- function(x, y, ry) {
# if needed, writes the df warning message to the log
df <- sum(ry) - ncol(x) - 1
mess <- paste("df set to 1. # observed cases:", sum(ry), " # predictors:", ncol(x) + 1)
if (df < 1 && sum(ry) > 0) {
updateLog(out = mess, frame = 4)
}
}
remove.lindep <- function(x, y, ry, eps = 1e-04, maxcor = 0.99,
allow.na = TRUE, frame = 4, ...) {
# returns a logical vector of length ncol(x)
if (ncol(x) == 0) {
return(NULL)
}
# setting eps = 0 bypasses remove.lindep()
if (eps == 0) {
return(rep.int(TRUE, ncol(x)))
}
if (eps < 0) {
stop("\n Argument 'eps' must be positive.")
}
# Keep all predictors if we allow imputation of fully missing y
if (allow.na && sum(ry) == 0) {
return(rep.int(TRUE, ncol(x)))
}
xobs <- x[ry, , drop = FALSE]
yobs <- as.numeric(y[ry])
if (var(yobs) < eps) {
return(rep(FALSE, ncol(xobs)))
}
keep <- unlist(apply(xobs, 2, var) > eps)
keep[is.na(keep)] <- FALSE
highcor <- suppressWarnings(unlist(apply(xobs, 2, cor, yobs) < maxcor))
keep <- keep & highcor
if (all(!keep)) {
updateLog(
out = "All predictors are constant or have too high correlation.",
frame = frame
)
}
# no need to calculate correlations, so return
k <- sum(keep)
if (k <= 1L) {
return(keep)
} # at most one TRUE
# correlation between x's
cx <- cor(xobs[, keep, drop = FALSE], use = "all.obs")
eig <- eigen(cx, symmetric = TRUE)
ncx <- cx
while (eig$values[k] / eig$values[1] < eps) {
j <- seq_len(k)[order(abs(eig$vectors[, k]), decreasing = TRUE)[1]]
keep[keep][j] <- FALSE
ncx <- cx[keep[keep], keep[keep], drop = FALSE]
k <- k - 1
eig <- eigen(ncx)
}
if (!all(keep)) {
out <- paste(dimnames(x)[[2]][!keep], collapse = ", ")
updateLog(out = out, frame = frame)
}
return(keep)
}
## make list of collinear variables to remove
find.collinear <- function(x, threshold = 0.999, ...) {
nvar <- ncol(x)
x <- data.matrix(x)
r <- !is.na(x)
nr <- apply(r, 2, sum, na.rm = TRUE)
ord <- order(nr, decreasing = TRUE)
xo <- x[, ord, drop = FALSE] ## SvB 24mar2011
varnames <- dimnames(xo)[[2]]
z <- suppressWarnings(cor(xo, use = "pairwise.complete.obs"))
hit <- outer(seq_len(nvar), seq_len(nvar), "<") & (abs(z) >= threshold)
out <- apply(hit, 2, any, na.rm = TRUE)
return(varnames[out])
}
updateLog <- function(out = NULL, meth = NULL, frame = 1) {
# find structures defined a mice() level
pos_state <- ma_exists("state", frame)$pos
pos_loggedEvents <- ma_exists("loggedEvents", frame)$pos
s <- get("state", pos_state)
r <- get("loggedEvents", pos_loggedEvents)
rec <- data.frame(
it = s$it,
im = s$im,
dep = s$dep,
meth = if (is.null(meth)) s$meth else meth,
out = if (is.null(out)) "" else out
)
if (s$log) {
rec <- rbind(r, rec)
}
s$log <- TRUE
assign("state", s, pos = pos_state, inherits = TRUE)
assign("loggedEvents", rec, pos = pos_loggedEvents, inherits = TRUE)
return()
}
sym <- function(x) {
(x + t(x)) / 2
}
# This helper function was copied from
# https://github.com/alexanderrobitzsch/miceadds/blob/master/R/ma_exists.R
ma_exists <- function(x, pos, n_index = 1:8) {
n_index <- n_index + 1
is_there <- exists(x, where = pos)
obj <- NULL
if (is_there) {
obj <- get(x, pos)
}
if (!is_there) {
for (nn in n_index) {
pos <- parent.frame(n = nn)
is_there <- exists(x, where = pos)
if (is_there) {
obj <- get(x, pos)
break
}
}
}
#--- output
res <- list(is_there = is_there, obj = obj, pos = pos)
return(res)
}
mice/R/mice.impute.norm.R 0000644 0001762 0000144 00000020050 14330031606 014675 0 ustar ligges users #' Imputation by Bayesian linear regression
#'
#' Calculates imputations for univariate missing data by Bayesian linear
#' regression, also known as the normal model.
#'
#' @aliases mice.impute.norm norm
#' @inheritParams mice.impute.pmm
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @author Stef van Buuren, Karin Groothuis-Oudshoorn
#' @details
#' Imputation of \code{y} by the normal model by the method defined by
#' Rubin (1987, p. 167). The procedure is as follows:
#'
#' \enumerate{
#' \item{Calculate the cross-product matrix \eqn{S=X_{obs}'X_{obs}}.}
#' \item{Calculate \eqn{V = (S+{diag}(S)\kappa)^{-1}}, with some small ridge
#' parameter \eqn{\kappa}.}
#' \item{Calculate regression weights \eqn{\hat\beta = VX_{obs}'y_{obs}.}}
#' \item{Draw a random variable \eqn{\dot g \sim \chi^2_\nu} with \eqn{\nu=n_1 - q}.}
#' \item{Calculate \eqn{\dot\sigma^2 = (y_{obs} - X_{obs}\hat\beta)'(y_{obs} - X_{obs}\hat\beta)/\dot g.}}
#' \item{Draw \eqn{q} independent \eqn{N(0,1)} variates in vector \eqn{\dot z_1}.}
#' \item{Calculate \eqn{V^{1/2}} by Cholesky decomposition.}
#' \item{Calculate \eqn{\dot\beta = \hat\beta + \dot\sigma\dot z_1 V^{1/2}}.}
#' \item{Draw \eqn{n_0} independent \eqn{N(0,1)} variates in vector \eqn{\dot z_2}.}
#' \item{Calculate the \eqn{n_0} values \eqn{y_{imp} = X_{mis}\dot\beta + \dot z_2\dot\sigma}.}
#' }
#'
#' Using \code{mice.impute.norm} for all columns emulates Schafer's NORM method (Schafer, 1997).
#' @references
#' Rubin, D.B (1987). Multiple Imputation for Nonresponse in Surveys. New York: John Wiley & Sons.
#'
#' Schafer, J.L. (1997). Analysis of incomplete multivariate data. London: Chapman & Hall.
#' @family univariate imputation functions
#' @keywords datagen
#' @export
mice.impute.norm <- function(y, ry, x, wy = NULL, ...) {
if (is.null(wy)) wy <- !ry
x <- cbind(1, as.matrix(x))
parm <- .norm.draw(y, ry, x, ...)
x[wy, ] %*% parm$beta + rnorm(sum(wy)) * parm$sigma
}
#' Draws values of beta and sigma by Bayesian linear regression
#'
#' This function draws random values of beta and sigma under the Bayesian
#' linear regression model as described in Rubin (1987, p. 167). This function
#' can be called by user-specified imputation functions.
#'
#' @aliases norm.draw .norm.draw
#' @param y Incomplete data vector of length \code{n}
#' @param ry Vector of missing data pattern (\code{FALSE}=missing,
#' \code{TRUE}=observed)
#' @param x Matrix (\code{n} x \code{p}) of complete covariates.
#' @param rank.adjust Argument that specifies whether \code{NA}'s in the
#' coefficients need to be set to zero. Only relevant when \code{ls.meth = "qr"}
#' AND the predictor matrix is rank-deficient.
#' @param ... Other named arguments.
#' @return A \code{list} containing components \code{coef} (least squares estimate),
#' \code{beta} (drawn regression weights) and \code{sigma} (drawn value of the
#' residual standard deviation).
#' @references
#' Rubin, D.B. (1987). \emph{Multiple imputation for nonresponse in surveys}. New York: Wiley.
#' @author Gerko Vink, 2018, for this version, based on earlier versions written
#' by Stef van Buuren, Karin Groothuis-Oudshoorn, 2017
#' @export
norm.draw <- function(y, ry, x, rank.adjust = TRUE, ...) {
return(.norm.draw(y, ry, x, rank.adjust = TRUE, ...))
}
###' @rdname norm.draw
###' @export
.norm.draw <- function(y, ry, x, rank.adjust = TRUE, ...) {
p <- estimice(x[ry, , drop = FALSE], y[ry], ...)
sigma.star <- sqrt(sum((p$r)^2) / rchisq(1, p$df))
beta.star <- p$c + (t(chol(sym(p$v))) %*% rnorm(ncol(x))) * sigma.star
parm <- list(p$c, beta.star, sigma.star, p$ls.meth)
names(parm) <- c("coef", "beta", "sigma", "estimation")
if (any(is.na(parm$coef)) & rank.adjust) {
parm$coef[is.na(parm$coef)] <- 0
parm$beta[is.na(parm$beta)] <- 0
}
parm
}
#' Computes least squares parameters
#'
#' This function computes least squares estimates, variance/covariance matrices,
#' residuals and degrees of freedom according to ridge regression, QR decomposition
#' or Singular Value Decomposition. This function is internally called by .norm.draw(),
#' but can be called by any user-specified imputation function.
#'
#' When calculating the inverse of the crossproduct of the predictor matrix,
#' problems may arise. For example, taking the inverse is not possible when the
#' predictor matrix is rank deficient, or when the estimation problem is
#' computationally singular. This function detects such error cases and
#' automatically falls back to adding a ridge penalty to the diagonal of the
#' crossproduct to allow for proper calculation of the inverse.
#'
#' @note
#' This functions adds a star to variable names in the mice iteration
#' history to signal that a ridge penalty was added. In that case, it
#' also adds an entry to \code{loggedEvents}.
#'
#' @aliases estimice
#' @param x Matrix (\code{n} x \code{p}) of complete covariates.
#' @param y Incomplete data vector of length \code{n}
#' @param ls.meth the method to use for obtaining the least squares estimates. By
#' default parameters are drawn by means of QR decomposition.
#' @param ridge A small numerical value specifying the size of the ridge used.
#' The default value \code{ridge = 1e-05} represents a compromise between stability
#' and unbiasedness. Decrease \code{ridge} if the data contain many junk variables.
#' Increase \code{ridge} for highly collinear data.
#' @param ... Other named arguments.
#' @return A \code{list} containing components \code{c} (least squares estimate),
#' \code{r} (residuals), \code{v} (variance/covariance matrix) and \code{df}
#' (degrees of freedom).
#' @author Gerko Vink, 2018
#' @export
estimice <- function(x, y, ls.meth = "qr", ridge = 1e-05, ...) {
df <- max(length(y) - ncol(x), 1)
if (ls.meth == "qr") {
qr <- lm.fit(x = x, y = y)
c <- t(qr$coef)
f <- qr$fitted.values
r <- t(qr$residuals)
v <- try(solve(as.matrix(crossprod(qr.R(qr$qr)))), silent = TRUE)
if (inherits(v, "try-error")) {
xtx <- as.matrix(crossprod(qr.R(qr$qr)))
# calculate ridge penalty
pen <- diag(xtx) * ridge
# add ridge penalty to allow inverse of v
v <- solve(xtx + diag(pen))
mess <- paste0(
"mice detected that your data are (nearly) multi-collinear.\n",
"It applied a ridge penalty to continue calculations, but the results can be unstable.\n",
"Does your dataset contain duplicates, linear transformation, or factors with unique respondent names?"
)
updateLog(out = mess, frame = 6)
if (get.printFlag()) {
cat("*")
} # indicator of added ridge penalty in the printed iteration history
}
return(list(c = t(c), r = t(r), v = v, df = df, ls.meth = ls.meth))
}
if (ls.meth == "ridge") {
xtx <- crossprod(x)
pen <- ridge * diag(xtx)
if (length(pen) == 1) {
pen <- matrix(pen)
}
v <- solve(xtx + diag(pen))
c <- t(y) %*% x %*% v
r <- y - x %*% t(c)
return(list(c = t(c), r = r, v = v, df = df, ls.meth = ls.meth))
}
if (ls.meth == "svd") {
s <- svd(x)
c <- s$v %*% ((t(s$u) %*% y) / s$d)
f <- x %*% c
r <- f - y
v <- try(solve(s$v %*% diag(s$d)^2 %*% t(s$v)), silent = TRUE)
if (inherits(v, "try-error")) {
xtx <- s$v %*% diag(s$d)^2 %*% t(s$v)
# calculate ridge penalty
pen <- diag(xtx) * ridge
# add ridge penalty to allow inverse of v
v <- solve(xtx + diag(pen))
mess <- paste0(
"mice detected that your data are (nearly) multi-collinear.\n",
"It applied a ridge penalty to continue calculations, but the results can be unstable.\n",
"Does your dataset contain duplicates, linear transformation, or factors with unique respondent names?"
)
updateLog(out = mess, frame = 6)
if (get.printFlag()) {
cat("*")
} # indicator of added ridge penalty in the printed iteration history
}
return(list(c = c, r = r, v = v, df = df, ls.meth = ls.meth))
}
}
get.printFlag <- function(start = 4) {
while (inherits(
try(get("printFlag", parent.frame(start)), silent = TRUE),
"try-error"
)) {
start <- start + 1
}
get("printFlag", parent.frame(start))
}
mice/R/mids2mplus.R 0000644 0001762 0000144 00000005336 14330031606 013615 0 ustar ligges users #' Export \code{mids} object to Mplus
#'
#' Converts a \code{mids} object into a format recognized by Mplus, and writes
#' the data and the Mplus input files
#'
#' This function automates most of the work needed to export a \code{mids}
#' object to \code{Mplus}. The function writes the multiple imputation datasets,
#' the file that contains the names of the multiple imputation data sets and an
#' \code{Mplus} input file. The \code{Mplus} input file has the proper file
#' names, so in principle it should run and read the data without alteration.
#' \code{Mplus} will recognize the data set as a multiply imputed data set, and
#' do automatic pooling in procedures where that is supported.
#'
#' @param imp The \code{imp} argument is an object of class \code{mids},
#' typically produced by the \code{mice()} function.
#' @param file.prefix A character string describing the prefix of the output
#' data files.
#' @param path A character string containing the path of the output file. By
#' default, files are written to the current \code{R} working directory.
#' @param sep The separator between the data fields.
#' @param dec The decimal separator for numerical data.
#' @param silent A logical flag stating whether the names of the files should be
#' printed.
#' @return The return value is \code{NULL}.
#' @author Gerko Vink, 2011.
#' @seealso \code{\link[=mids-class]{mids}}, \code{\link{mids2spss}}
#' @keywords manip
#' @export
mids2mplus <- function(imp, file.prefix = "imp", path = getwd(), sep = "\t", dec = ".", silent = FALSE) {
m <- imp$m
file.list <- matrix(0, m, 1)
script <- matrix(0, 3, 1)
for (i in seq_len(m)) {
write.table(complete(imp, i),
file = file.path(path, paste0(file.prefix, i, ".dat")),
sep = sep, dec = dec, col.names = FALSE, row.names = FALSE
)
file.list[i, ] <- paste0(file.prefix, i, ".dat")
}
write.table(file.list,
file = file.path(path, paste0(file.prefix, "list.dat")),
sep = sep, dec = dec, col.names = FALSE, row.names = FALSE, quote = FALSE
)
names <- paste(colnames(complete(imp, 1)), collapse = " ")
script[1, ] <- paste0("DATA: FILE IS ", file.prefix, "list.dat;")
script[2, ] <- "TYPE = IMPUTATION;"
script[3, ] <- paste0("VARIABLE: NAMES ARE ", names, ";")
write.table(script,
file = file.path(path, paste0(file.prefix, "list.inp")),
sep = sep, dec = dec, col.names = FALSE, row.names = FALSE, quote = FALSE
)
if (!silent) {
cat(
"Data values written to", file.path(path, paste0(file.prefix, 1, ".dat")),
"through", paste0(file.prefix, m, ".dat"), "\n"
)
cat("Data names written to", file.path(path, paste0(file.prefix, "list.dat")), "\n")
cat("Mplus code written to", file.path(path, paste0(file.prefix, "list.inp")), "\n")
}
}
mice/R/predictorMatrix.R 0000644 0001762 0000144 00000011714 14347615172 014711 0 ustar ligges users #' Creates a \code{predictorMatrix} argument
#'
#' This helper function creates a valid \code{predictMatrix}. The
#' \code{predictorMatrix} is an argument to the \code{mice} function.
#' It specifies the target variable or block in the rows, and the
#' predictor variables on the columns. An entry of \code{0} means that
#' the column variable is NOT used to impute the row variable or block.
#' A nonzero value indicates that it is used.
#' @param data A \code{data.frame} with the source data
#' @param blocks An optional specification for blocks of variables in
#' the rows. The default assigns each variable in its own block.
#' @param predictorMatrix A predictor matrix from which rows with the same
#' names are copied into the output predictor matrix.
#' @return A matrix
#' @seealso \code{\link{make.blocks}}
#' @examples
#' make.predictorMatrix(nhanes)
#' make.predictorMatrix(nhanes, blocks = make.blocks(nhanes, "collect"))
#' @export
make.predictorMatrix <- function(data, blocks = make.blocks(data),
predictorMatrix = NULL) {
input.predictorMatrix <- predictorMatrix
data <- check.dataform(data)
predictorMatrix <- matrix(1, nrow = length(blocks), ncol = ncol(data))
dimnames(predictorMatrix) <- list(names(blocks), colnames(data))
for (i in row.names(predictorMatrix)) {
predictorMatrix[i, colnames(predictorMatrix) %in% i] <- 0
}
# preserve any user setting in predictorMatrix specification
if (!is.null(input.predictorMatrix)) {
for (i in row.names(predictorMatrix)) {
if (i %in% row.names(input.predictorMatrix)) {
predictorMatrix[i, ] <- input.predictorMatrix[i, ]
}
}
}
predictorMatrix
}
check.predictorMatrix <- function(predictorMatrix,
data,
blocks = NULL) {
data <- check.dataform(data)
if (!is.matrix(predictorMatrix)) {
stop("predictorMatrix not a matrix", call. = FALSE)
}
if (any(dim(predictorMatrix) == 0L)) {
stop("predictorMatrix has no rows or columns", call. = FALSE)
}
# if we have no blocks, restrict to square predictorMatrix
if (is.null(blocks)) {
if (nrow(predictorMatrix) != ncol(predictorMatrix)) {
stop(
paste(
"If no blocks are specified, predictorMatrix must",
"have same number of rows and columns"
),
call. = FALSE
)
}
if (is.null(dimnames(predictorMatrix))) {
if (ncol(predictorMatrix) == ncol(data)) {
dimnames(predictorMatrix) <- list(colnames(data), colnames(data))
} else {
stop("Missing row/column names in predictorMatrix", call. = FALSE)
}
}
for (i in row.names(predictorMatrix)) {
predictorMatrix[i, grep(paste0("^", i, "$"), colnames(predictorMatrix))] <- 0
}
return(predictorMatrix)
}
# check conforming arguments
if (nrow(predictorMatrix) > length(blocks)) {
stop(
paste0(
"predictorMatrix has more rows (", nrow(predictorMatrix),
") than blocks (", length(blocks), ")"
),
call. = FALSE
)
}
# borrow rownames from blocks if needed
if (is.null(rownames(predictorMatrix)) &&
nrow(predictorMatrix) == length(blocks)) {
rownames(predictorMatrix) <- names(blocks)
}
if (is.null(rownames(predictorMatrix))) {
stop("Unable to set row names of predictorMatrix", call. = FALSE)
}
# borrow blocknames from predictorMatrix if needed
if (is.null(names(blocks)) &&
nrow(predictorMatrix) == length(blocks)) {
names(blocks) <- rownames(predictorMatrix)
}
if (is.null(names(blocks))) {
stop("Unable to set names of blocks", call. = FALSE)
}
# check existence of row names in blocks
found <- rownames(predictorMatrix) %in% names(blocks)
if (!all(found)) {
stop("Names not found in blocks: ",
paste(rownames(predictorMatrix)[!found], collapse = ", "),
call. = FALSE
)
}
# borrow colnames from data if needed
if (is.null(colnames(predictorMatrix)) &&
ncol(predictorMatrix) == ncol(data)) {
colnames(predictorMatrix) <- names(data)
}
if (is.null(colnames(predictorMatrix))) {
stop("Unable to set column names of predictorMatrix", call. = FALSE)
}
# check existence of variable names on data
found <- colnames(predictorMatrix) %in% names(data)
if (!all(found)) {
stop("Names not found in data: ",
paste(colnames(predictorMatrix)[!found], collapse = ", "),
call. = FALSE
)
}
list(
predictorMatrix = predictorMatrix,
blocks = blocks
)
}
edit.predictorMatrix <- function(predictorMatrix,
visitSequence,
user.visitSequence,
maxit) {
# edit predictorMatrix to a monotone pattern
if (maxit == 1L && !is.null(user.visitSequence) && user.visitSequence == "monotone") {
for (i in 1L:length(visitSequence)) {
predictorMatrix[visitSequence[i], visitSequence[i:length(visitSequence)]] <- 0
}
}
predictorMatrix
}
mice/R/mira.R 0000644 0001762 0000144 00000005130 14330031647 012443 0 ustar ligges users #' Multiply imputed repeated analyses (\code{mira})
#'
#' The \code{mira} object is generated by the \code{with.mids()} function.
#' The \code{as.mira()}
#' function takes the results of repeated complete-data analysis stored as a
#' list, and turns it into a \code{mira} object that can be pooled.
#'
#' @section Slots:
#' \describe{
#' #' \item{\code{.Data}:}{Object of class \code{"list"} containing the
#' following slots:}
#' \item{\code{call}:}{The call that created the object.}
#' \item{\code{call1}:}{The call that created the \code{mids} object that was used
#' in \code{call}.}
#' \item{\code{nmis}:}{An array containing the number of missing observations per
#' column.}
#' \item{\code{analyses}:}{A list of \code{m} components containing the individual
#' fit objects from each of the \code{m} complete data analyses.}
#' }
#'
#' @details
#' In versions prior to \code{mice 3.0} pooling required only that
#' \code{coef()} and \code{vcov()} methods were available for fitted
#' objects. \emph{This feature is no longer supported}. The reason is that \code{vcov()}
#' methods are inconsistent across packages, leading to buggy behaviour
#' of the \code{pool()} function. Since \code{mice 3.0+}, the \code{broom}
#' package takes care of filtering out the relevant parts of the
#' complete-data analysis. It may happen that you'll see the messages
#' like \code{No method for tidying an S3 object of class ...} or
#' \code{Error: No glance method for objects of class ...}. The royal
#' way to solve this problem is to write your own \code{glance()} and \code{tidy()}
#' methods and add these to \code{broom} according to the specifications
#' given in \url{https://broom.tidymodels.org}.
#'
#' #'The \code{mira} class of objects has methods for the
#' following generic functions: \code{print}, \code{summary}.
#'
#' Many of the functions of the \code{mice} package do not use the
#' S4 class definitions, and instead rely on the S3 list equivalent
#' \code{oldClass(obj) <- "mira"}.
#'
#' @name mira-class
#' @rdname mira-class
#' @aliases mira-class mira
#' @author Stef van Buuren, Karin Groothuis-Oudshoorn, 2000
#' @seealso \code{\link{with.mids}}, \code{\link[=mids-class]{mids}}, \code{\link{mipo}}
#' @references van Buuren S and Groothuis-Oudshoorn K (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#' @keywords classes
#' @export
setClass("mira",
representation(
call = "call",
call1 = "call",
nmis = "integer",
analyses = "list"
),
contains = "list"
)
mice/R/mice.impute.mpmm.R 0000644 0001762 0000144 00000006512 14347334322 014710 0 ustar ligges users #' Imputation by multivariate predictive mean matching
#'
#' Imputes multivariate incomplete data among which there are specific relations,
#' for instance, polynomials, interactions, range restrictions and sum scores.
#' @aliases mice.impute.mpmm mpmm
#' @param data matrix with exactly two missing data patterns
#' @param format A character vector specifying the type of object that should
#' be returned. The default is \code{format = "imputes"}.
#' @param ... Other named arguments.
#' @return A matrix with imputed data, which has \code{ncol(y)} columns and
#' \code{sum(wy)} rows.
#' @details
#' This function implements the predictive mean matching and applies canonical
#' regression analysis to select donors fora set of missing variables. In general,
#' canonical regressionanalysis looks for a linear combination of covariates that
#' predicts a linear combination of outcomes (a set of missing variables)
#' optimally in a least-square sense (Israels, 1987). The predicted
#' value of the linear combination of the set of missing variables
#' would be applied to perform predictive mean matching.
#'
#' @note
#' The function requires variables in the block have the same missingness pattern.
#' If there are more than one missingness pattern, the function will return
#' a warning.
#' @author Mingyang Cai and Gerko Vink
# @author Mingyang Cai (University of Utrecht), \email{g.vink#uu.nl}
#' @seealso \code{\link{mice.impute.pmm}}
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-knowledge.html#sec:quadratic}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' @family univariate imputation functions
#' @keywords datagen
#' @examples
#' # simulate data
#' beta2 <- beta1 <- .5
#' x <- rnorm(1000)
#' e <- rnorm(1000, 0, 1)
#' y <- beta1 * x + beta2 * x^2 + e
#' dat <- data.frame(y = y, x = x, x2 = x^2)
#' m <- as.logical(rbinom(1000, 1, 0.25))
#' dat[m, c("x", "x2")] <- NA
#'
#' # impute
#' blk <- list("y", c("x", "x2"))
#' meth <- c("", "mpmm")
#' imp <- mice(dat, blocks = blk, method = meth, print = FALSE,
#' m = 2, maxit = 2)
#'
#' # analyse and check
#' summary(pool(with(imp, lm(y ~ x + x2))))
#' with(dat, plot(x, x2, col = mdc(1)))
#' with(complete(imp), points(x[m], x2[m], col = mdc(2)))
#' @export
#'
mice.impute.mpmm <- function(data, format = "imputes", ...) {
order <- dimnames(data)[[1]]
res <- mpmm.impute(data, ...)
return(single2imputes(res[order, ], is.na(data)))
}
mpmm.impute <- function(data, ...) {
data <- as.data.frame(data)
r <- !is.na(data)
mpat <- apply(r, 1, function(x) paste(as.numeric(x), collapse = ""))
nmpat <- length(unique(mpat))
if (nmpat != 2) stop("There are more than one missingness patterns")
r <- unique(r)
r <- r[rowSums(r) < ncol(r), ]
y <- data[, which(r == FALSE), drop = FALSE]
ry <- !is.na(y)[, 1]
x <- data[, which(r == TRUE), drop = FALSE]
wy <- !ry
ES <- eigen(solve(cov(y[ry, , drop = FALSE], y[ry, , drop = FALSE])) %*% cov(y[ry, , drop = FALSE], x[ry, , drop = FALSE])
%*% solve(cov(x[ry, , drop = FALSE], x[ry, , drop = FALSE])) %*% cov(x[ry, , drop = FALSE], y[ry, , drop = FALSE]))
parm <- as.matrix(ES$vectors[, 1])
z <- as.matrix(y) %*% parm
imp <- mice.impute.pmm(z, ry, x)
zstar <- as.matrix(imp)
y[wy, ] <- y[ry, , drop = FALSE][match(zstar, z[ry]), ]
data[colnames(y)] <- y
return(data)
}
mice/R/nhanes2.R 0000644 0001762 0000144 00000001621 14330031606 013045 0 ustar ligges users #' NHANES example - mixed numerical and discrete variables
#'
#' A small data set with non-monotone missing values.
#'
#' A small data set with missing data and mixed numerical and discrete
#' variables. The data set \code{nhanes} is the same data set, but with all data
#' treated as numerical.
#'
#' @name nhanes2
#' @docType data
#' @format A data frame with 25 observations on the following 4 variables.
#' \describe{
#' \item{age}{Age group (1=20-39, 2=40-59, 3=60+)}
#' \item{bmi}{Body mass index (kg/m**2)}
#' \item{hyp}{Hypertensive (1=no,2=yes)}
#' \item{chl}{Total serum cholesterol (mg/dL)} }
#' @seealso \code{\link{nhanes}}
#' @source Schafer, J.L. (1997). \emph{Analysis of Incomplete Multivariate
#' Data.} London: Chapman & Hall. Table 6.14.
#' @keywords datasets
#' @examples
#' # create 5 imputed data sets
#' imp <- mice(nhanes2)
#'
#' # print the first imputed data set
#' complete(imp)
NULL
mice/R/as.R 0000644 0001762 0000144 00000014665 14330031625 012127 0 ustar ligges users #' Converts an imputed dataset (long format) into a \code{mids} object
#'
#' This function converts imputed data stored in long format into
#' an object of class \code{mids}. The original incomplete dataset
#' needs to be available so that we know where the missing data are.
#' The function is useful to convert back operations applied to
#' the imputed data back in a \code{mids} object. It may also be
#' used to store multiply imputed data sets from other software
#' into the format used by \code{mice}.
#' @note The function expects the input data \code{long} to be sorted by
#' imputation number (variable \code{".imp"} by default), and in the
#' same sequence within each imputation block.
#' @param long A multiply imputed data set in long format, for example
#' produced by a call to \code{complete(..., action = 'long', include = TRUE)},
#' or by other software.
#' @param .imp An optional column number or column name in \code{long},
#' indicating the imputation index. The values are assumed to be consecutive
#' integers between 0 and \code{m}. Values \code{1} through \code{m}
#' correspond to the imputation index, value \code{0} indicates
#' the original data (with missings).
#' By default, the procedure will search for a variable named \code{".imp"}.
#' @param .id An optional column number or column name in \code{long},
#' indicating the subject identification. If not specified, then the
#' function searches for a variable named \code{".id"}. If this variable
#' is found, the values in the column will define the row names in
#' the \code{data} element of the resulting \code{mids} object.
#' @inheritParams mice
#' @return An object of class \code{mids}
#' @author Gerko Vink
#' @examples
#' # impute the nhanes dataset
#' imp <- mice(nhanes, print = FALSE)
#' # extract the data in long format
#' X <- complete(imp, action = "long", include = TRUE)
#' # create dataset with .imp variable as numeric
#' X2 <- X
#'
#' # nhanes example without .id
#' test1 <- as.mids(X)
#' is.mids(test1)
#' identical(complete(test1, action = "long", include = TRUE), X)
#'
#' # nhanes example without .id where .imp is numeric
#' test2 <- as.mids(X2)
#' is.mids(test2)
#' identical(complete(test2, action = "long", include = TRUE), X)
#'
#' # nhanes example, where we explicitly specify .id as column 2
#' test3 <- as.mids(X, .id = ".id")
#' is.mids(test3)
#' identical(complete(test3, action = "long", include = TRUE), X)
#'
#' # nhanes example with .id where .imp is numeric
#' test4 <- as.mids(X2, .id = 2)
#' is.mids(test4)
#' identical(complete(test4, action = "long", include = TRUE), X)
#'
#' # example without an .id variable
#' # variable .id not preserved
#' X3 <- X[, -2]
#' test5 <- as.mids(X3)
#' is.mids(test5)
#' identical(complete(test5, action = "long", include = TRUE)[, -2], X[, -2])
#'
#' # as() syntax has fewer options
#' test7 <- as(X, "mids")
#' test8 <- as(X2, "mids")
#' test9 <- as(X2[, -2], "mids")
#' rev <- ncol(X):1
#' test10 <- as(X[, rev], "mids")
#'
#' # where argument copies also observed data into $imp element
#' where <- matrix(TRUE, nrow = nrow(nhanes), ncol = ncol(nhanes))
#' colnames(where) <- colnames(nhanes)
#' test11 <- as.mids(X, where = where)
#' identical(complete(test11, action = "long", include = TRUE), X)
#' @keywords mids
#' @export
as.mids <- function(long, where = NULL, .imp = ".imp", .id = ".id") {
if (is.numeric(.imp)) .imp <- names(long)[.imp]
if (is.numeric(.id)) .id <- names(long)[.id]
if (!.imp %in% names(long)) stop("Imputation index `.imp` not found")
# no missings allowed in .imp
imps <- unlist(long[, .imp], use.names = FALSE)
if (anyNA(imps)) stop("Missing values in imputation index `.imp`")
# number of records within .imp should be the same
if (any(diff(table(imps))) != 0) {
stop("Unequal group sizes in imputation index `.imp`")
}
# get original data part
keep <- setdiff(names(long), na.omit(c(.imp, .id)))
data <- long[imps == 0, keep, drop = FALSE]
n <- nrow(data)
if (n == 0) {
stop("Original data not found.\n Use `complete(..., action = 'long', include = TRUE)` to save original data.")
}
# determine m
m <- length(unique(imps)) - 1
# use mice to get info on data
if (is.null(where)) where <- is.na(data)
ini <- mice(data,
m = m, where = where, maxit = 0,
remove.collinear = FALSE, allow.na = TRUE
)
# store any .id as row names
if (!is.na(.id)) {
rownames(ini$data) <- unlist(long[imps == 0, .id], use.names = FALSE)
}
# copy imputations from long into proper ini$imp elements
names <- names(ini$imp)
for (i in seq_along(names)) {
varname <- names[i]
if (!is.null(ini$imp[[varname]])) {
for (j in seq_len(m)) {
idx <- imps == j & where[, varname]
ini$imp[[varname]][j] <- long[idx, varname]
}
}
}
ini
}
#' Create a \code{mira} object from repeated analyses
#'
#' The \code{as.mira()} function takes the results of repeated
#' complete-data analysis stored as a list, and turns it
#' into a \code{mira} object that can be pooled.
#' @param fitlist A list containing $m$ fitted analysis objects
#' @return An S3 object of class \code{mira}.
#' @seealso \code{\link[=mira-class]{mira}}
#' @author Stef van Buuren
#' @export
as.mira <- function(fitlist) {
if (is.mira(fitlist)) {
return(fitlist)
}
if (is.mids(fitlist)) {
stop("as.mira() cannot convert class 'mids' into 'mira'. Use with() instead.")
}
call <- match.call()
if (!is.list(fitlist)) {
stop("Argument 'fitlist' is not a list")
}
class(fitlist) <- "list"
object <- list(call = call, call1 = NULL, nmis = NULL, analyses = fitlist)
oldClass(object) <- c("mira", "matrix")
object
}
#' Converts into a \code{mitml.result} object
#'
#' The \code{as.mitml.result()} function takes the results of repeated
#' complete-data analysis stored as a list, and turns it
#' into an object of class \code{mitml.result}.
#' @param x An object of class \code{mira}
#' @return An S3 object of class \code{mitml.result}, a list
#' containing $m$ fitted analysis objects.
#' @seealso \code{\link[mitml]{with.mitml.list}}
#' @author Stef van Buuren
#' @export
as.mitml.result <- function(x) {
if (inherits(x, "mitml.result")) {
return(x)
}
z <- NULL
if (is.mira(x)) {
z <- getfit(x)
} else if (is.list(x)) z <- x
class(z) <- c("mitml.result", "list")
z
}
setOldClass(c("mids", "mira"))
setAs(
from = "data.frame", to = "mids",
def = function(from) {
as.mids(from)
}
)
setAs(
from = "list", to = "mira",
def = function(from) {
as.mira(from)
}
)
mice/R/mice-package.R 0000644 0001762 0000144 00000011563 14330031647 014030 0 ustar ligges users #' \pkg{mice}: Multivariate Imputation by Chained Equations
#'
#' The \pkg{mice} package implements a method to deal with missing data.
#' The package creates multiple imputations (replacement values) for
#' multivariate missing data. The method is based on Fully Conditional
#' Specification, where each incomplete variable is imputed by a separate
#' model. The MICE algorithm can impute mixes of continuous, binary,
#' unordered categorical and ordered categorical data. In addition, MICE
#' can impute continuous two-level data, and maintain consistency between
#' imputations by means of passive imputation. Many diagnostic plots are
#' implemented to inspect the quality of the imputations.
#'
#' The \pkg{mice} package contains functions to
#' \itemize{
#' \item Inspect the missing data pattern
#' \item Impute the missing data \emph{m} times, resulting in \emph{m} completed data sets
#' \item Diagnose the quality of the imputed values
#' \item Analyze each completed data set
#' \item Pool the results of the repeated analyses
#' \item Store and export the imputed data in various formats
#' \item Generate simulated incomplete data
#' \item Incorporate custom imputation methods
#' }
#'
#' @section Functions:
#'
#' The main functions are:
#' \tabular{ll}{
#' \code{mice()} \tab Impute the missing data *m* times\cr
#' \code{with()} \tab Analyze completed data sets\cr
#' \code{pool()} \tab Combine parameter estimates\cr
#' \code{complete()} \tab Export imputed data\cr
#' \code{ampute()} \tab Generate missing data\cr}
#'
#' @section Vignettes:
#'
#' There is a detailed series of
#' six online vignettes that walk you through solving realistic inference
#' problems with mice.
#'
#' We suggest going through these vignettes in the following order
#' \enumerate{
#' \item \href{https://www.gerkovink.com/miceVignettes/Ad_hoc_and_mice/Ad_hoc_methods.html}{Ad hoc methods and the MICE algorithm}
#' \item \href{https://www.gerkovink.com/miceVignettes/Convergence_pooling/Convergence_and_pooling.html}{Convergence and pooling}
#' \item \href{https://www.gerkovink.com/miceVignettes/Missingness_inspection/Missingness_inspection.html}{Inspecting how the observed data and missingness are related}
#' \item \href{https://www.gerkovink.com/miceVignettes/Passive_Post_processing/Passive_imputation_post_processing.html}{Passive imputation and post-processing}
#' \item \href{https://www.gerkovink.com/miceVignettes/Multi_level/Multi_level_data.html}{Imputing multilevel data}
#' \item \href{https://www.gerkovink.com/miceVignettes/Sensitivity_analysis/Sensitivity_analysis.html}{Sensitivity analysis with \pkg{mice}}
#' }
#'
#' #'Van Buuren, S. (2018).
#' Boca Raton, FL.: Chapman & Hall/CRC Press.
#' The book
#' \href{https://stefvanbuuren.name/fimd/}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' contains a lot of \href{https://github.com/stefvanbuuren/fimdbook/tree/master/R}{example code}.
#'
#' @section Methodology:
#'
#' The \pkg{mice} software was published in the {Journal of Statistical Software} (Van Buuren and Groothuis-Oudshoorn, 2011). \doi{10.18637/jss.v045.i03}
#' The first application of the method
#' concerned missing blood pressure data (Van Buuren et. al., 1999).
#' The term \emph{Fully Conditional Specification} was introduced in 2006 to describe a general class of methods that specify imputations model for multivariate data as a set of conditional distributions (Van Buuren et. al., 2006). Further details on mixes of variables and applications can be found in the book
#' \href{https://stefvanbuuren.name/fimd/}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#'
#' @section Enhanced linear algebra:
#'
#' Updating the BLAS can improve speed of R, sometime considerably. The details
#' depend on the operating system. See the discussion in the
#' "R Installation and Administration" guide for further information.
#'
#' @docType package
#' @name mice
#' @seealso \code{\link{mice}}, \code{\link{with.mids}},
#' \code{\link{pool}}, \code{\link{complete}}, \code{\link{ampute}}
#' @references
#' van Buuren, S., Boshuizen, H.C., Knook, D.L. (1999) Multiple
#' imputation of missing blood pressure covariates in survival analysis.
#' \emph{Statistics in Medicine}, \bold{18}, 681--694.
#'
#' van Buuren, S., Brand, J.P.L., Groothuis-Oudshoorn C.G.M., Rubin, D.B. (2006)
#' Fully conditional specification in multivariate imputation. \emph{Journal of
#' Statistical Computation and Simulation}, \bold{76}, 12, 1049--1064.
#'
#' van Buuren, S., Groothuis-Oudshoorn, K. (2011). {\code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1--67. \doi{10.18637/jss.v045.i03}
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' @useDynLib mice, .registration = TRUE
NULL
mice/R/mice.impute.lasso.select.logreg.R 0000644 0001762 0000144 00000006030 14330031647 017606 0 ustar ligges users #' Imputation by indirect use of lasso logistic regression
#'
#' Imputes univariate missing data using logistic regression following a
#' preprocessing lasso variable selection step.
#'
#' @aliases mice.impute.lasso.select.logreg lasso.select.logreg
#' @inheritParams mice.impute.pmm
#' @param nfolds The number of folds for the cross-validation of the lasso penalty.
#' The default is 10.
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @details
#' The method consists of the following steps:
#' \enumerate{
#' \item For a given \code{y} variable under imputation, fit a linear regression with lasso
#' penalty using \code{y[ry]} as dependent variable and \code{x[ry, ]} as predictors.
#' The coefficients that are not shrunk to 0 define the active set of predictors
#' that will be used for imputation.
#' \item Fit a logit with the active set of predictors, and find (bhat, V(bhat))
#' \item Draw BETA from N(bhat, V(bhat))
#' \item Compute predicted scores for m.d., i.e. logit-1(X BETA)
#' \item Compare the score to a random (0,1) deviate, and impute.
#' }
#' The user can specify a \code{predictorMatrix} in the \code{mice} call
#' to define which predictors are provided to this univariate imputation method.
#' The lasso regularization will select, among the variables indicated by
#' the user, the ones that are important for imputation at any given iteration.
#' Therefore, users may force the exclusion of a predictor from a given
#' imputation model by speficing a \code{0} entry.
#' However, a non-zero entry does not guarantee the variable will be used,
#' as this decision is ultimately made by the lasso variable selection
#' procedure.
#'
#' The method is based on the Indirect Use of Regularized Regression (IURR) proposed by
#' Zhao & Long (2016) and Deng et al (2016).
#' @author Edoardo Costantini, 2021
#' @references
#'
#' Deng, Y., Chang, C., Ido, M. S., & Long, Q. (2016). Multiple imputation for
#' general missing data patterns in the presence of high-dimensional data.
#' Scientific reports, 6(1), 1-10.
#'
#' Zhao, Y., & Long, Q. (2016). Multiple imputation in the presence of
#' high-dimensional data. Statistical Methods in Medical Research, 25(5),
#' 2021-2035.
#'
#' @family univariate imputation functions
#' @keywords datagen
#' @export
mice.impute.lasso.select.logreg <- function(y, ry, x, wy = NULL, nfolds = 10, ...) {
install.on.demand("glmnet", ...)
# Body
if (is.null(wy)) wy <- !ry
x_glmnet <- cbind(1, x)
xobs <- x_glmnet[ry, , drop = FALSE]
xmis <- x[wy, ]
yobs <- y[ry]
# Train imputation model
# used later in the estiamtion require this.
cv_lasso <- glmnet::cv.glmnet(
x = xobs, y = yobs,
family = "binomial",
nfolds = nfolds,
alpha = 1
)
# Define Active Set
glmnet_coefs <- as.matrix(coef(cv_lasso,
s = "lambda.min"
))[, 1]
AS <- which((glmnet_coefs != 0)[-1]) # Non-zero reg coefficinets
# Perform regular logreg draw
xas <- x_glmnet[, AS, drop = FALSE]
vec <- mice.impute.logreg(
y = y, ry = ry, x = xas, wy = wy,
...
)
vec
}
mice/R/imports.R 0000644 0001762 0000144 00000002746 14436636423 013234 0 ustar ligges users #' @import methods
#' @importFrom broom glance tidy
#' @importFrom dplyr %>% bind_cols bind_rows filter group_by lead
#' mutate n pull row_number select summarize
#' @importFrom glmnet cv.glmnet
#' @importFrom graphics abline axis box par plot plot.new plot.window
#' points rect text
#' @importFrom lattice bwplot densityplot stripplot xyplot
#' @importFrom mitml jomoImpute mitmlComplete panImpute testModels
#' @importFrom nnet multinom
#' @importFrom Rcpp evalCpp
#' @importFrom rlang .data syms
#' @importFrom rpart rpart rpart.control
#' @importFrom stats C aggregate as.formula binomial coef
#' complete.cases confint
#' contr.treatment cor df.residual fitted
#' formula gaussian getCall
#' glm is.empty.model lm lm.fit
#' median model.frame model.matrix
#' na.exclude na.omit na.pass
#' pf predict pt qt quantile quasibinomial
#' rbinom rchisq reformulate rgamma rnorm runif
#' sd summary.glm terms update var vcov
#' @importFrom tidyr complete
#' @importFrom utils askYesNo flush.console head install.packages methods
#' packageDescription packageVersion
#' tail write.table
#' @export bwplot
#' @export densityplot
#' @export stripplot
#' @export xyplot
#' @export complete
NULL
mice/R/boys.R 0000644 0001762 0000144 00000005625 14330031606 012473 0 ustar ligges users #' Growth of Dutch boys
#'
#' Height, weight, head circumference and puberty of 748 Dutch boys.
#'
#' Random sample of 10\% from the cross-sectional data used to construct the
#' Dutch growth references 1997. Variables \code{gen} and \code{phb} are ordered
#' factors. \code{reg} is a factor.
#'
#' @name boys
#' @docType data
#' @format A data frame with 748 rows on the following 9 variables: \describe{
#' \item{age}{Decimal age (0-21 years)}
#' \item{hgt}{Height (cm)}
#' \item{wgt}{Weight (kg)}
#' \item{bmi}{Body mass index}
#' \item{hc}{Head circumference (cm)}
#' \item{gen}{Genital Tanner stage (G1-G5)}
#' \item{phb}{Pubic hair (Tanner P1-P6)}
#' \item{tv}{Testicular volume (ml)}
#' \item{reg}{Region (north, east, west, south, city)} }
#' @source Fredriks, A.M,, van Buuren, S., Burgmeijer, R.J., Meulmeester JF,
#' Beuker, R.J., Brugman, E., Roede, M.J., Verloove-Vanhorick, S.P., Wit, J.M.
#' (2000) Continuing positive secular growth change in The Netherlands
#' 1955-1997. \emph{Pediatric Research}, \bold{47}, 316-323.
#'
#' Fredriks, A.M., van Buuren, S., Wit, J.M., Verloove-Vanhorick, S.P. (2000).
#' Body index measurements in 1996-7 compared with 1980. \emph{Archives of
#' Disease in Childhood}, \bold{82}, 107-112.
#' @keywords datasets
#' @examples
#'
#' # create two imputed data sets
#' imp <- mice(boys, m = 1, maxit = 2)
#' z <- complete(imp, 1)
#'
#' # create imputations for age <8yrs
#' plot(z$age, z$gen,
#' col = mdc(1:2)[1 + is.na(boys$gen)],
#' xlab = "Age (years)", ylab = "Tanner Stage Genital"
#' )
#'
#' # figure to show that the default imputation method does not impute BMI
#' # consistently
#' plot(z$bmi, z$wgt / (z$hgt / 100)^2,
#' col = mdc(1:2)[1 + is.na(boys$bmi)],
#' xlab = "Imputed BMI", ylab = "Calculated BMI"
#' )
#'
#' # also, BMI distributions are somewhat different
#' oldpar <- par(mfrow = c(1, 2))
#' MASS::truehist(z$bmi[!is.na(boys$bmi)],
#' h = 1, xlim = c(10, 30), ymax = 0.25,
#' col = mdc(1), xlab = "BMI observed"
#' )
#' MASS::truehist(z$bmi[is.na(boys$bmi)],
#' h = 1, xlim = c(10, 30), ymax = 0.25,
#' col = mdc(2), xlab = "BMI imputed"
#' )
#' par(oldpar)
#'
#' # repair the inconsistency problem by passive imputation
#' meth <- imp$meth
#' meth["bmi"] <- "~I(wgt/(hgt/100)^2)"
#' pred <- imp$predictorMatrix
#' pred["hgt", "bmi"] <- 0
#' pred["wgt", "bmi"] <- 0
#' imp2 <- mice(boys, m = 1, maxit = 2, meth = meth, pred = pred)
#' z2 <- complete(imp2, 1)
#'
#' # show that new imputations are consistent
#' plot(z2$bmi, z2$wgt / (z2$hgt / 100)^2,
#' col = mdc(1:2)[1 + is.na(boys$bmi)],
#' ylab = "Calculated BMI"
#' )
#'
#' # and compare distributions
#' oldpar <- par(mfrow = c(1, 2))
#' MASS::truehist(z2$bmi[!is.na(boys$bmi)],
#' h = 1, xlim = c(10, 30), ymax = 0.25, col = mdc(1),
#' xlab = "BMI observed"
#' )
#' MASS::truehist(z2$bmi[is.na(boys$bmi)],
#' h = 1, xlim = c(10, 30), ymax = 0.25, col = mdc(2),
#' xlab = "BMI imputed"
#' )
#' par(oldpar)
NULL
mice/R/mice.impute.norm.predict.R 0000644 0001762 0000144 00000003636 14330031606 016341 0 ustar ligges users #' Imputation by linear regression through prediction
#'
#' Imputes the "best value" according to the linear regression model, also
#' known as \emph{regression imputation}.
#'
#' @aliases mice.impute.norm.predict norm.predict
#' @inheritParams mice.impute.pmm
#' @return Vector with imputed data, same type as \code{y}, and of length
#' \code{sum(wy)}
#' @details
#' Calculates regression weights from the observed data and returns predicted
#' values to as imputations. This
#' method is known as \emph{regression imputation}.
#' @section Warning: THIS METHOD SHOULD NOT BE USED FOR DATA ANALYSIS.
#' This method is seductive because it imputes the most
#' likely value according to the model. However, it ignores the uncertainty
#' of the missing values and artificially
#' amplifies the relations between the columns of the data. Application of
#' richer models having more parameters does not help to evade these issues.
#' Stochastic regression methods, like \code{\link{mice.impute.pmm}} or
#' \code{\link{mice.impute.norm}}, are generally preferred.
#'
#' At best, prediction can give reasonable estimates of the mean, especially
#' if normality assumptions are plausible. See Little and Rubin (2002, p. 62-64)
#' or Van Buuren (2012, p. 11-13, p. 45-46) for a discussion of this method.
#' @author Gerko Vink, Stef van Buuren, 2018
#' @references
#' Little, R.J.A. and Rubin, D.B. (2002). Statistical Analysis with Missing
#' Data. New York: John Wiley and Sons.
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-linearnormal.html}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#' @family univariate imputation functions
#' @keywords datagen
#' @export
mice.impute.norm.predict <- function(y, ry, x, wy = NULL, ...) {
if (is.null(wy)) wy <- !ry
x <- cbind(1, as.matrix(x))
p <- estimice(x[ry, , drop = FALSE], y[ry], ...)
x[wy, , drop = FALSE] %*% p$c
}
mice/R/auxiliary.R 0000644 0001762 0000144 00000005713 14344615604 013537 0 ustar ligges users #' Conditional imputation helper
#'
#' Sorry, the \code{ifdo()} function is not yet implemented.
#' @aliases ifdo
#' @param cond a condition
#' @param action the action to do
#' @return Currently returns an error message.
#' @author Stef van Buuren, 2012
#' @keywords internal
ifdo <- function(cond, action) {
cat("Function ifdo() not yet implemented.\n")
}
#' Appends specified break to the data
#'
#' A custom function to insert rows in long data with new pseudo-observations
#' that are being done on the specified break ages. There should be a
#' column called \code{first} in \code{data} with logical data that codes whether
#' the current row is the first for subject \code{id}. Furthermore,
#' the function assumes that columns \code{age}, \code{occ},
#' \code{hgt.z}, \code{wgt.z} and
#' \code{bmi.z} are available. This function is used on the \code{tbc}
#' data in FIMD chapter 9. Check that out to see it in action.
#' @aliases appendbreak
#' @param data A data frame in the long long format
#' @param brk A vector of break ages
#' @param warp.model A time warping model
#' @param id The subject identifier
#' @param typ Label to signal that this is a newly added observation
#' @return A long data frame with additional rows for the break ages
#' @export
appendbreak <- function(data, brk, warp.model = warp.model, id = NULL, typ = "pred") {
k <- length(brk)
app <- data[data$first, ]
if (!is.null(id)) {
idx <- app$id %in% id
app <- app[idx, ]
}
nap <- nrow(app)
## update administrative variables
app$first <- FALSE
app$typ <- typ
app$occ <- NA
app <- app[rep.int(seq_len(nap), length(brk)), ]
## update age variables
app$age <- rep(brk, each = nap)
app$age2 <- predict(warp.model, newdata = app)
X <- splines::bs(app$age,
knots = brk,
Boundary.knots = c(brk[1], brk[k] + 0.0001),
degree = 1
)
X <- X[, -(k + 1)]
app[, paste0("x", seq_len(ncol(X)))] <- X
## update outcome variable (set to missing)
app[, c("hgt.z", "wgt.z", "bmi.z")] <- NA
app <- rbind(data, app)
app[order(app$id, app$age), ]
}
#' Extract broken stick estimates from a \code{lmer} object
#'
#' @param fit An object of class \code{lmer}
#' @return A matrix containing broken stick estimates
#' @author Stef van Buuren, 2012
#' @export
extractBS <- function(fit) {
siz <- t(lme4::ranef(fit)[[1]]) + lme4::fixef(fit)
matrix(siz, nrow = nrow(siz) * ncol(siz), ncol = 1)
}
## used by mice.impute.midastouch
bootfunc.plain <- function(n) {
random <- sample.int(n, replace = TRUE)
as.numeric(table(factor(random, levels = seq_len(n))))
}
minmax <- function(x, domin = TRUE, domax = TRUE) {
maxx <- sqrt(.Machine$double.xmax)
minx <- sqrt(.Machine$double.eps)
if (domin) {
x <- pmin(x, maxx)
}
if (domax) {
x <- pmax(x, minx)
}
x
}
single2imputes <- function(single, mis) {
nmis <- colSums(mis)
vars <- names(single)[nmis > 0]
z <- vector("list", length(vars))
names(z) <- vars
for (j in vars) z[[j]] <- single[mis[, j], j]
z
}
mice/R/mice.R 0000644 0001762 0000144 00000056532 14430511431 012437 0 ustar ligges users #' Multivariate Imputation by Chained Equations (MICE)
#'
#' Generates Multivariate Imputations by Chained Equations (MICE)
#'
#' Generates multiple imputations for incomplete multivariate data by Gibbs
#' sampling. Missing data can occur anywhere in the data. The algorithm imputes
#' an incomplete column (the target column) by generating 'plausible' synthetic
#' values given other columns in the data. Each incomplete column must act as a
#' target column, and has its own specific set of predictors. The default set of
#' predictors for a given target consists of all other columns in the data. For
#' predictors that are incomplete themselves, the most recently generated
#' imputations are used to complete the predictors prior to imputation of the
#' target column.
#'
#' A separate univariate imputation model can be specified for each column. The
#' default imputation method depends on the measurement level of the target
#' column. In addition to these, several other methods are provided. You can
#' also write their own imputation functions, and call these from within the
#' algorithm.
#'
#' The data may contain categorical variables that are used in a regressions on
#' other variables. The algorithm creates dummy variables for the categories of
#' these variables, and imputes these from the corresponding categorical
#' variable.
#'
#' Built-in univariate imputation methods are:
#'
#' \tabular{lll}{
#' \code{pmm} \tab any \tab Predictive mean matching\cr
#' \code{midastouch} \tab any \tab Weighted predictive mean matching\cr
#' \code{sample} \tab any \tab Random sample from observed values\cr
#' \code{cart} \tab any \tab Classification and regression trees\cr
#' \code{rf} \tab any \tab Random forest imputations\cr
#' \code{mean} \tab numeric \tab Unconditional mean imputation\cr
#' \code{norm} \tab numeric \tab Bayesian linear regression\cr
#' \code{norm.nob} \tab numeric \tab Linear regression ignoring model error\cr
#' \code{norm.boot} \tab numeric \tab Linear regression using bootstrap\cr
#' \code{norm.predict} \tab numeric \tab Linear regression, predicted values\cr
#' \code{lasso.norm} \tab numeric \tab Lasso linear regression\cr
#' \code{lasso.select.norm} \tab numeric \tab Lasso select + linear regression\cr
#' \code{quadratic} \tab numeric \tab Imputation of quadratic terms\cr
#' \code{ri} \tab numeric \tab Random indicator for nonignorable data\cr
#' \code{logreg} \tab binary \tab Logistic regression\cr
#' \code{logreg.boot} \tab binary \tab Logistic regression with bootstrap\cr
#' \code{lasso.logreg} \tab binary \tab Lasso logistic regression\cr
#' \code{lasso.select.logreg}\tab binary \tab Lasso select + logistic regression\cr
#' \code{polr} \tab ordered \tab Proportional odds model\cr
#' \code{polyreg} \tab unordered\tab Polytomous logistic regression\cr
#' \code{lda} \tab unordered\tab Linear discriminant analysis\cr
#' \code{2l.norm} \tab numeric \tab Level-1 normal heteroscedastic\cr
#' \code{2l.lmer} \tab numeric \tab Level-1 normal homoscedastic, lmer\cr
#' \code{2l.pan} \tab numeric \tab Level-1 normal homoscedastic, pan\cr
#' \code{2l.bin} \tab binary \tab Level-1 logistic, glmer\cr
#' \code{2lonly.mean} \tab numeric \tab Level-2 class mean\cr
#' \code{2lonly.norm} \tab numeric \tab Level-2 class normal\cr
#' \code{2lonly.pmm} \tab any \tab Level-2 class predictive mean matching
#' }
#'
#' These corresponding functions are coded in the \code{mice} library under
#' names \code{mice.impute.method}, where \code{method} is a string with the
#' name of the univariate imputation method name, for example \code{norm}. The
#' \code{method} argument specifies the methods to be used. For the \code{j}'th
#' column, \code{mice()} calls the first occurrence of
#' \code{paste('mice.impute.', method[j], sep = '')} in the search path. The
#' mechanism allows uses to write customized imputation function,
#' \code{mice.impute.myfunc}. To call it for all columns specify
#' \code{method='myfunc'}. To call it only for, say, column 2 specify
#' \code{method=c('norm','myfunc','logreg',\dots{})}.
#'
#' \emph{Skipping imputation:} The user may skip imputation of a column by
#' setting its entry to the empty method: \code{""}. For complete columns without
#' missing data \code{mice} will automatically set the empty method. Setting t
#' he empty method does not produce imputations for the column, so any missing
#' cells remain \code{NA}. If column A contains \code{NA}'s and is used as
#' predictor in the imputation model for column B, then \code{mice} produces no
#' imputations for the rows in B where A is missing. The imputed data
#' for B may thus contain \code{NA}'s. The remedy is to remove column A from
#' the imputation model for the other columns in the data. This can be done
#' by setting the entire column for variable A in the \code{predictorMatrix}
#' equal to zero.
#'
#' \emph{Passive imputation:} \code{mice()} supports a special built-in method,
#' called passive imputation. This method can be used to ensure that a data
#' transform always depends on the most recently generated imputations. In some
#' cases, an imputation model may need transformed data in addition to the
#' original data (e.g. log, quadratic, recodes, interaction, sum scores, and so
#' on).
#'
#' Passive imputation maintains consistency among different transformations of
#' the same data. Passive imputation is invoked if \code{~} is specified as the
#' first character of the string that specifies the univariate method.
#' \code{mice()} interprets the entire string, including the \code{~} character,
#' as the formula argument in a call to \code{model.frame(formula,
#' data[!r[,j],])}. This provides a simple mechanism for specifying deterministic
#' dependencies among the columns. For example, suppose that the missing entries
#' in variables \code{data$height} and \code{data$weight} are imputed. The body
#' mass index (BMI) can be calculated within \code{mice} by specifying the
#' string \code{'~I(weight/height^2)'} as the univariate imputation method for
#' the target column \code{data$bmi}. Note that the \code{~} mechanism works
#' only on those entries which have missing values in the target column. You
#' should make sure that the combined observed and imputed parts of the target
#' column make sense. An easy way to create consistency is by coding all entries
#' in the target as \code{NA}, but for large data sets, this could be
#' inefficient. Note that you may also need to adapt the default
#' \code{predictorMatrix} to evade linear dependencies among the predictors that
#' could cause errors like \code{Error in solve.default()} or \code{Error:
#' system is exactly singular}. Though not strictly needed, it is often useful
#' to specify \code{visitSequence} such that the column that is imputed by the
#' \code{~} mechanism is visited each time after one of its predictors was
#' visited. In that way, deterministic relation between columns will always be
#' synchronized.
#'
#' #'A new argument \code{ls.meth} can be parsed to the lower level
#' \code{.norm.draw} to specify the method for generating the least squares
#' estimates and any subsequently derived estimates. Argument \code{ls.meth}
#' takes one of three inputs: \code{"qr"} for QR-decomposition, \code{"svd"} for
#' singular value decomposition and \code{"ridge"} for ridge regression.
#' \code{ls.meth} defaults to \code{ls.meth = "qr"}.
#'
#' \emph{Auxiliary predictors in formulas specification: }
#' For a given block, the \code{formulas} specification takes precedence over
#' the corresponding row in the \code{predictMatrix} argument. This
#' precedence is, however, restricted to the subset of variables
#' specified in the terms of the block formula. Any
#' variables not specified by \code{formulas} are imputed
#' according to the \code{predictMatrix} specification. Variables with
#' non-zero \code{type} values in the \code{predictMatrix} will
#' be added as main effects to the \code{formulas}, which will
#' act as supplementary covariates in the imputation model. It is possible
#' to turn off this behavior by specifying the
#' argument \code{auxiliary = FALSE}.
#'
#' @param data A data frame or a matrix containing the incomplete data. Missing
#' values are coded as \code{NA}.
#' @param m Number of multiple imputations. The default is \code{m=5}.
#' @param method Can be either a single string, or a vector of strings with
#' length \code{length(blocks)}, specifying the imputation method to be
#' used for each column in data. If specified as a single string, the same
#' method will be used for all blocks. The default imputation method (when no
#' argument is specified) depends on the measurement level of the target column,
#' as regulated by the \code{defaultMethod} argument. Columns that need
#' not be imputed have the empty method \code{""}. See details.
#' @param predictorMatrix A numeric matrix of \code{length(blocks)} rows
#' and \code{ncol(data)} columns, containing 0/1 data specifying
#' the set of predictors to be used for each target column.
#' Each row corresponds to a variable block, i.e., a set of variables
#' to be imputed. A value of \code{1} means that the column
#' variable is used as a predictor for the target block (in the rows).
#' By default, the \code{predictorMatrix} is a square matrix of \code{ncol(data)}
#' rows and columns with all 1's, except for the diagonal.
#' Note: For two-level imputation models (which have \code{"2l"} in their names)
#' other codes (e.g, \code{2} or \code{-2}) are also allowed.
#' @param ignore A logical vector of \code{nrow(data)} elements indicating
#' which rows are ignored when creating the imputation model. The default
#' \code{NULL} includes all rows that have an observed value of the variable
#' to imputed. Rows with \code{ignore} set to \code{TRUE} do not influence the
#' parameters of the imputation model, but are still imputed. We may use the
#' \code{ignore} argument to split \code{data} into a training set (on which the
#' imputation model is built) and a test set (that does not influence the
#' imputation model estimates).
#' Note: Multivariate imputation methods, like \code{mice.impute.jomoImpute()}
#' or \code{mice.impute.panImpute()}, do not honour the \code{ignore} argument.
#' @param where A data frame or matrix with logicals of the same dimensions
#' as \code{data} indicating where in the data the imputations should be
#' created. The default, \code{where = is.na(data)}, specifies that the
#' missing data should be imputed. The \code{where} argument may be used to
#' overimpute observed data, or to skip imputations for selected missing values.
#' Note: Imputation methods that generate imptutations outside of
#' \code{mice}, like \code{mice.impute.panImpute()} may depend on a complete
#' predictor space. In that case, a custom \code{where} matrix can not be
#' specified.
#' @param blocks List of vectors with variable names per block. List elements
#' may be named to identify blocks. Variables within a block are
#' imputed by a multivariate imputation method
#' (see \code{method} argument). By default each variable is placed
#' into its own block, which is effectively
#' fully conditional specification (FCS) by univariate models
#' (variable-by-variable imputation). Only variables whose names appear in
#' \code{blocks} are imputed. The relevant columns in the \code{where}
#' matrix are set to \code{FALSE} of variables that are not block members.
#' A variable may appear in multiple blocks. In that case, it is
#' effectively re-imputed each time that it is visited.
#' @param visitSequence A vector of block names of arbitrary length, specifying the
#' sequence of blocks that are imputed during one iteration of the Gibbs
#' sampler. A block is a collection of variables. All variables that are
#' members of the same block are imputed
#' when the block is visited. A variable that is a member of multiple blocks
#' is re-imputed within the same iteration.
#' The default \code{visitSequence = "roman"} visits the blocks (left to right)
#' in the order in which they appear in \code{blocks}.
#' One may also use one of the following keywords: \code{"arabic"}
#' (right to left), \code{"monotone"} (ordered low to high proportion
#' of missing data) and \code{"revmonotone"} (reverse of monotone).
#' \emph{Special case}: If you specify both \code{visitSequence = "monotone"} and
#' \code{maxit = 1}, then the procedure will edit the \code{predictorMatrix}
#' to conform to the monotone pattern. Realize that convergence in one
#' iteration is only guaranteed if the missing data pattern is actually
#' monotone. The procedure does not check this.
#' @param formulas A named list of formula's, or expressions that
#' can be converted into formula's by \code{as.formula}. List elements
#' correspond to blocks. The block to which the list element applies is
#' identified by its name, so list names must correspond to block names.
#' The \code{formulas} argument is an alternative to the
#' \code{predictorMatrix} argument that allows for more flexibility in
#' specifying imputation models, e.g., for specifying interaction terms.
#' @param blots A named \code{list} of \code{alist}'s that can be used
#' to pass down arguments to lower level imputation function. The entries
#' of element \code{blots[[blockname]]} are passed down to the function
#' called for block \code{blockname}.
#' @param post A vector of strings with length \code{ncol(data)} specifying
#' expressions as strings. Each string is parsed and
#' executed within the \code{sampler()} function to post-process
#' imputed values during the iterations.
#' The default is a vector of empty strings, indicating no post-processing.
#' Multivariate (block) imputation methods ignore the \code{post} parameter.
#' @param defaultMethod A vector of length 4 containing the default
#' imputation methods for 1) numeric data, 2) factor data with 2 levels, 3)
#' factor data with > 2 unordered levels, and 4) factor data with > 2
#' ordered levels. By default, the method uses
#' \code{pmm}, predictive mean matching (numeric data) \code{logreg}, logistic
#' regression imputation (binary data, factor with 2 levels) \code{polyreg},
#' polytomous regression imputation for unordered categorical data (factor > 2
#' levels) \code{polr}, proportional odds model for (ordered, > 2 levels).
#' @param maxit A scalar giving the number of iterations. The default is 5.
#' @param printFlag If \code{TRUE}, \code{mice} will print history on console.
#' Use \code{print=FALSE} for silent computation.
#' @param seed An integer that is used as argument by the \code{set.seed()} for
#' offsetting the random number generator. Default is to leave the random number
#' generator alone.
#' @param data.init A data frame of the same size and type as \code{data},
#' without missing data, used to initialize imputations before the start of the
#' iterative process. The default \code{NULL} implies that starting imputation
#' are created by a simple random draw from the data. Note that specification of
#' \code{data.init} will start all \code{m} Gibbs sampling streams from the same
#' imputation.
#' @param \dots Named arguments that are passed down to the univariate imputation
#' functions.
#'
#' @return Returns an S3 object of class \code{\link[=mids-class]{mids}}
#' (multiply imputed data set)
#' @author Stef van Buuren \email{stef.vanbuuren@@tno.nl}, Karin
#' Groothuis-Oudshoorn \email{c.g.m.oudshoorn@@utwente.nl}, 2000-2010, with
#' contributions of Alexander Robitzsch, Gerko Vink, Shahab Jolani,
#' Roel de Jong, Jason Turner, Lisa Doove,
#' John Fox, Frank E. Harrell, and Peter Malewski.
#' @seealso \code{\link[=mids-class]{mids}}, \code{\link{with.mids}},
#' \code{\link{set.seed}}, \code{\link{complete}}
#' @references Van Buuren, S., Groothuis-Oudshoorn, K. (2011). \code{mice}:
#' Multivariate Imputation by Chained Equations in \code{R}. \emph{Journal of
#' Statistical Software}, \bold{45}(3), 1-67.
#' \doi{10.18637/jss.v045.i03}
#'
#' Van Buuren, S. (2018).
#' \href{https://stefvanbuuren.name/fimd/sec-FCS.html#sec:MICE}{\emph{Flexible Imputation of Missing Data. Second Edition.}}
#' Chapman & Hall/CRC. Boca Raton, FL.
#'
#' Van Buuren, S., Brand, J.P.L., Groothuis-Oudshoorn C.G.M., Rubin, D.B. (2006)
#' Fully conditional specification in multivariate imputation. \emph{Journal of
#' Statistical Computation and Simulation}, \bold{76}, 12, 1049--1064.
#'
#' Van Buuren, S. (2007) Multiple imputation of discrete and continuous data by
#' fully conditional specification. \emph{Statistical Methods in Medical
#' Research}, \bold{16}, 3, 219--242.
#'
#' Van Buuren, S., Boshuizen, H.C., Knook, D.L. (1999) Multiple imputation of
#' missing blood pressure covariates in survival analysis. \emph{Statistics in
#' Medicine}, \bold{18}, 681--694.
#'
#' Brand, J.P.L. (1999) \emph{Development, implementation and evaluation of
#' multiple imputation strategies for the statistical analysis of incomplete
#' data sets.} Dissertation. Rotterdam: Erasmus University.
#' @keywords iteration
#' @examples
#' # do default multiple imputation on a numeric matrix
#' imp <- mice(nhanes)
#' imp
#'
#' # list the actual imputations for BMI
#' imp$imp$bmi
#'
#' # first completed data matrix
#' complete(imp)
#'
#' # imputation on mixed data with a different method per column
#' mice(nhanes2, meth = c("sample", "pmm", "logreg", "norm"))
#'
#' \dontrun{
#' # example where we fit the imputation model on the train data
#' # and apply the model to impute the test data
#' set.seed(123)
#' ignore <- sample(c(TRUE, FALSE), size = 25, replace = TRUE, prob = c(0.3, 0.7))
#'
#' # scenario 1: train and test in the same dataset
#' imp <- mice(nhanes2, m = 2, ignore = ignore, print = FALSE, seed = 22112)
#' imp.test1 <- filter(imp, ignore)
#' imp.test1$data
#' complete(imp.test1, 1)
#' complete(imp.test1, 2)
#'
#' # scenario 2: train and test in separate datasets
#' traindata <- nhanes2[!ignore, ]
#' testdata <- nhanes2[ignore, ]
#' imp.train <- mice(traindata, m = 2, print = FALSE, seed = 22112)
#' imp.test2 <- mice.mids(imp.train, newdata = testdata)
#' complete(imp.test2, 1)
#' complete(imp.test2, 2)
#' }
#' @export
mice <- function(data,
m = 5,
method = NULL,
predictorMatrix,
ignore = NULL,
where = NULL,
blocks,
visitSequence = NULL,
formulas,
blots = NULL,
post = NULL,
defaultMethod = c("pmm", "logreg", "polyreg", "polr"),
maxit = 5,
printFlag = TRUE,
seed = NA,
data.init = NULL,
...) {
call <- match.call()
check.deprecated(...)
if (!is.na(seed)) set.seed(seed)
# check form of data and m
data <- check.dataform(data)
m <- check.m(m)
# determine input combination: predictorMatrix, blocks, formulas
mp <- missing(predictorMatrix)
mb <- missing(blocks)
mf <- missing(formulas)
# case A
if (mp & mb & mf) {
# blocks lead
blocks <- make.blocks(colnames(data))
predictorMatrix <- make.predictorMatrix(data, blocks)
formulas <- make.formulas(data, blocks)
}
# case B
if (!mp & mb & mf) {
# predictorMatrix leads
predictorMatrix <- check.predictorMatrix(predictorMatrix, data)
blocks <- make.blocks(colnames(predictorMatrix), partition = "scatter")
formulas <- make.formulas(data, blocks, predictorMatrix = predictorMatrix)
}
# case C
if (mp & !mb & mf) {
# blocks leads
blocks <- check.blocks(blocks, data)
predictorMatrix <- make.predictorMatrix(data, blocks)
formulas <- make.formulas(data, blocks)
}
# case D
if (mp & mb & !mf) {
# formulas leads
formulas <- check.formulas(formulas, data)
blocks <- construct.blocks(formulas)
predictorMatrix <- make.predictorMatrix(data, blocks)
}
# case E
if (!mp & !mb & mf) {
# predictor leads
blocks <- check.blocks(blocks, data)
z <- check.predictorMatrix(predictorMatrix, data, blocks)
predictorMatrix <- z$predictorMatrix
blocks <- z$blocks
formulas <- make.formulas(data, blocks, predictorMatrix = predictorMatrix)
}
# case F
if (!mp & mb & !mf) {
# formulas lead
formulas <- check.formulas(formulas, data)
predictorMatrix <- check.predictorMatrix(predictorMatrix, data)
blocks <- construct.blocks(formulas, predictorMatrix)
predictorMatrix <- make.predictorMatrix(data, blocks, predictorMatrix)
}
# case G
if (mp & !mb & !mf) {
# blocks lead
blocks <- check.blocks(blocks, data, calltype = "formula")
formulas <- check.formulas(formulas, blocks)
predictorMatrix <- make.predictorMatrix(data, blocks)
}
# case H
if (!mp & !mb & !mf) {
# blocks lead
blocks <- check.blocks(blocks, data)
formulas <- check.formulas(formulas, data)
predictorMatrix <- check.predictorMatrix(predictorMatrix, data, blocks)
}
chk <- check.cluster(data, predictorMatrix)
where <- check.where(where, data, blocks)
# check visitSequence, edit predictorMatrix for monotone
user.visitSequence <- visitSequence
visitSequence <- check.visitSequence(visitSequence,
data = data, where = where, blocks = blocks
)
predictorMatrix <- edit.predictorMatrix(
predictorMatrix = predictorMatrix,
visitSequence = visitSequence,
user.visitSequence = user.visitSequence,
maxit = maxit
)
method <- check.method(
method = method, data = data, where = where,
blocks = blocks, defaultMethod = defaultMethod
)
post <- check.post(post, data)
blots <- check.blots(blots, data, blocks)
ignore <- check.ignore(ignore, data)
# data frame for storing the event log
state <- list(it = 0, im = 0, dep = "", meth = "", log = FALSE)
loggedEvents <- data.frame(it = 0, im = 0, dep = "", meth = "", out = "")
# edit imputation setup
setup <- list(
method = method,
predictorMatrix = predictorMatrix,
visitSequence = visitSequence,
post = post
)
setup <- edit.setup(data, setup, ...)
method <- setup$method
predictorMatrix <- setup$predictorMatrix
visitSequence <- setup$visitSequence
post <- setup$post
# initialize imputations
nmis <- apply(is.na(data), 2, sum)
imp <- initialize.imp(
data, m, ignore, where, blocks, visitSequence,
method, nmis, data.init
)
# and iterate...
from <- 1
to <- from + maxit - 1
q <- sampler(
data, m, ignore, where, imp, blocks, method,
visitSequence, predictorMatrix, formulas, blots,
post, c(from, to), printFlag, ...
)
if (!state$log) loggedEvents <- NULL
if (state$log) row.names(loggedEvents) <- seq_len(nrow(loggedEvents))
## save, and return
midsobj <- list(
data = data,
imp = q$imp,
m = m,
where = where,
blocks = blocks,
call = call,
nmis = nmis,
method = method,
predictorMatrix = predictorMatrix,
visitSequence = visitSequence,
formulas = formulas,
post = post,
blots = blots,
ignore = ignore,
seed = seed,
iteration = q$iteration,
lastSeedValue = get(".Random.seed",
envir = globalenv(), mode = "integer",
inherits = FALSE
),
chainMean = q$chainMean,
chainVar = q$chainVar,
loggedEvents = loggedEvents,
version = packageVersion("mice"),
date = Sys.Date()
)
oldClass(midsobj) <- "mids"
if (!is.null(midsobj$loggedEvents)) {
warning("Number of logged events: ", nrow(midsobj$loggedEvents),
call. = FALSE
)
}
midsobj
}
mice/R/mcar.R 0000644 0001762 0000144 00000045561 14334522175 012456 0 ustar ligges users #' Jamshidian and Jalal's Non-Parametric MCAR Test
#'
#' Test whether missingness is contingent upon the observed variables,
#' according to the methodology developed by Jamshidian and Jalal (2010) (see
#' Details).
#' @param x An object for which a method exists; usually a `data.frame`.
#' @param imputed Either an object of class `mids`, as returned by
#' [mice::mice()], or a list of `data.frame`s.
#' @param min_n Atomic numeric, must be greater than 1. When there are missing
#' data patterns with fewer than `min_n` cases, all cases with that pattern will
#' be removed from `x` and `imputed`.
#' @param method Atomic character. If it is known (or assumed) that data are
#' either multivariate normally distributed or not, then use either
#' `method = "hawkins"` or `method = "nonparametric"`, respectively.
#' The default argument `method = "auto"` follows the procedure outlined in the
#' Details section, and in Figure 7 of Jamshidian and Jalal (2010).
#' @param replications Number of replications used to simulate the Neyman
#' distribution when performing Hawkins' test. As this method is based on random
#' sampling, use a high number of `replications` (and optionally,
#' [set.seed()]) to minimize Monte Carlo error and ensure reproducibility.
#' @param use_chisq Atomic integer, indicating the minimum number of cases
#' within a group *k* that triggers the use of asymptotic Chi-square
#' distribution instead of the emprical distribution in the Neyman uniformity
#' test, which is performed as part of Hawkins' test.
#' @param alpha Atomic numeric, indicating the significance level of tests.
#' @details Three types of missingness have been distinguished in the literature
#' (Rubin, 1976):
#' Missing completely at random (MCAR), which means that missingness is random;
#' missing at random (MAR), which means that missingness is contingent on the
#' *observed*;
#' and missing not at random (MNAR), which means that missingness is related to
#' unobserved data.
#'
#' Jamshidian and Jalal's non-parametric MCAR test assumes that the missing data
#' are either MCAR or MAR, and tests whether the missingness is independent of
#' the observed values. If so, the covariance matrices of the imputed data will
#' be equal accross groups with different patterns of missingness. This test
#' consists of the following procedure:
#' \enumerate{
#' \item Data are imputed.
#' \item The imputed data are split into *k* groups according to the
#' *k* missing data patterns in the original data (see
#' [mice::md.pattern()]).
#' \item Perform Hawkins' test for equality of covariances across the *k*
#' groups.
#' \item If the test is *not significant*, conclude that there is no evidence
#' against multivariate normality of the data, nor against MCAR.
#' \item If the test *is significant*, and multivariate normality of the data
#' can be assumed, then it can be concluded that missingness is MAR.
#' \item If multivariate normality cannot be assumed, then perform the
#' Anderson-Darling non-parametric test for equality of covariances across the
#' *k* groups.
#' \item If the Anderson-Darling test is *not significant*, this is evidence
#' against multivariate normality - but no evidence against MCAR.
#' \item If the Anderson-Darling test *is significant*, this is evidence
#' it can be concluded that missingness is MAR.
#' }
#'
#' Note that, despite its name in common parlance, an MCAR test can only
#' indicate whether missingness is MCAR or MAR. The procedure cannot distinguish
#' MCAR from MNAR, so a non-significant result does not rule out MNAR.
#'
#' This is a re-implementation of the function `TestMCARNormality`, which was
#' originally published in the R-packgage `MissMech`, which has been removed
#' from CRAN. This new implementation is faster, as its backend is written in
#' C++. It also enhances the functionality of the original:
#' \itemize{
#' \item Multiply imputed data can now be used; the median p-value and test
#' statistic across replications is then reported, as suggested by
#' Eekhout, Wiel, and Heymans (2017).
#' \item The printing method for an `mcar_object` gives a warning when at
#' least one p-value of either test was significant. In this case, it is
#' recommended to inspect the range of p-values, and consider potential
#' violations of MCAR.
#' \item A plotting method for an `mcar_object` is provided.
#' \item A plotting method for the `$md.pattern` element of an `mcar_object`
#' is provided.
#' }
#'
#' @return An object of class `mcar_object`.
#' @author Caspar J. Van Lissa
#' @references
#' Rubin, D. B. (1976). Inference and Missing Data. Biometrika, Vol. 63, No. 3,
#' pp. 581-592. \doi{10.2307/2335739}
#'
#' Eekhout, I., M. A. Wiel, & M. W. Heymans (2017). Methods for Significance
#' Testing of Categorical Covariates in Logistic Regression Models After
#' Multiple Imputation: Power and Applicability Analysis. BMC Medical Research
#' Methodology 17 (1): 129.
#'
#' Jamshidian, M., & Jalal, S. (2010). Tests of homoscedasticity, normality, and
#' missing completely at random for incomplete multivariate data. Psychometrika,
#' 75(4), 649–674. \doi{10.1007/s11336-010-9175-3}
#' @keywords internal
#' @examples
#' res <- mcar(nhanes)
#' # Examine test results
#' res
#' # Plot p-values across imputed data sets
#' plot(res)
#' # Plot md patterns used for the test
#' plot(res, type = "md.pattern")
#' # Note difference with the raw md.patterns:
#' md.pattern(nhanes)
#' @export
#' @importFrom stats cov pchisq spline
#' @md
mcar <- function(x,
imputed = mice(x, method = "norm"),
min_n = 6,
method = "auto",
replications = 10000,
use_chisq = 30,
alpha = 0.05) {
UseMethod("mcar", x)
}
#' @method mcar data.frame
#' @export
mcar.data.frame <- function(x,
imputed = mice(x, method = "norm"),
min_n = 6,
method = "auto",
replications = 10000,
use_chisq = 30,
alpha = 0.05) {
anyfact <- sapply(x, inherits, what = "factor")
if (any(anyfact)) {
stop("Be advised that this MCAR test has not been validated for categorical variables.")
}
if (min_n < 1) {
stop("Argument 'min_n' must be greater than 1.")
}
out <-
list(
hawk_chisq = NULL,
hawk_df = NULL,
hawk_p = NULL,
ad_value = NULL,
ad_p = NULL,
alpha = alpha,
method = method,
md.pattern = NULL,
removed_rows = NULL,
removed_patterns = NULL
)
if (inherits(imputed, "mids")) {
imputed <- mice::complete(imputed, action = "all")
} else if (inherits(imputed, "list")) {
if (!inherits(imputed[[1]], "data.frame")) {
imputed <- tryCatch(lapply(imputed, data.frame), error = function(e) {
stop("Argument 'imputed' must be a list of data.frames or an object of class 'mids'. Could not coerce argument 'imputed' to class 'data.frame'.")
})
message("Argument 'imputed' must be an object of class 'mids', or a list of 'data.frame's. Coerced argument 'imputed' to data.frame.")
}
}
if (!all(dim(x) == dim(imputed[[1]]))) {
stop("Imputed data must have the same dimensions as the incomplete data.")
}
if (any(is.na(imputed[[1]]))) {
stop("Imputed data contain missing values.")
}
missings <- is.na(x)
rowmis <- rowSums(missings)
colmis <- colSums(missings)
if (any(rowmis == ncol(x))) {
warning("Note that there were some rows with all missing data. Tests may be invalid for these rows. Consider removing them.")
# x <- x[!rowmis == ncol(x), , drop = FALSE]
}
if (any(colmis == nrow(x))) {
stop("Some columns contain all missing data. This will result in invalid results.")
}
univals <- sapply(x, function(i) {
length(unique(i))
})
if (any(univals < 2)) {
stop("Some columns are constant, not variable. This will result in invalid results.")
}
newdata <- x
missings <- is.na(x)
pats <- mice::md.pattern(x, plot = FALSE)
if (nrow(pats) < 4L) {
stop("Two or more missing data patterns are required.")
}
remove_pats <- as.numeric(rownames(pats))[-nrow(pats)] <= min_n
if (any(remove_pats)) {
out$removed_patterns <- pats[remove_pats, ]
pats <- pats[-nrow(pats), colnames(missings)]
idmiss <- do.call(paste, as.data.frame(missings))
idpats <- do.call(paste, as.data.frame(pats == 0))
remove_these <- idmiss %in% idpats[remove_pats]
out$removed_rows <- remove_these
newdata <- x[!remove_these, , drop = FALSE]
imputed <- lapply(imputed, `[`, i = !remove_these, j = colnames(newdata), drop = FALSE)
missings <- is.na(newdata)
pats <- mice::md.pattern(newdata, plot = FALSE)
}
class(pats) <- c("md.pattern", class(pats))
out$md.pattern <- pats
pat_n <- as.numeric(rownames(pats))[-nrow(pats)]
pats <- pats[-nrow(pats), colnames(missings)]
idpats <- idpats[!remove_pats]
idmiss <- do.call(paste, as.data.frame(missings))
which_pat <- apply(sapply(idpats, `==`, idmiss), 1, which)
numpat <- length(pat_n)
if (nrow(newdata) == 0) {
stop("No valid rows of data left.")
}
if (numpat == 1) {
stop("Only one missing data pattern.")
}
if (any(pat_n < 2)) {
stop("At least 2 cases needed in each missing data pattern.")
}
# Perform Hawkins test ----------------------------------------------------
hawklist <- lapply(imputed, function(thisimp) {
hawkins(thisimp, which_pat)
})
if (method %in% c("auto", "hawkins")) {
pvalsn <- sapply(hawklist, function(thisimp) {
sapply(thisimp[["a"]], function(thistail) {
ni <- length(thistail)
pn <- p_neyman(thistail, replications, use_chisq)
pn + (pn == 0) / replications
})
})
out$hawk_chisq <- -2 * colSums(log(pvalsn))
out$hawk_df <- 2 * numpat
out$hawk_p <- pchisq(out$hawk_chisq, out$hawk_df, lower.tail = FALSE)
}
# Perform Anderson-Darling test -------------------------------------------
if ((method == "auto" & any(out$hawk_p < alpha)) | method == "nonparametric") {
adout <- sapply(hawklist, function(thisimp) {
anderson_darling(thisimp[["fij"]]) # First row is p
})
out$ad_value <- colSums(adout[-1, , drop = FALSE])
out$ad_p <- adout[1, ]
}
class(out) <- c("mcar_object", class(out))
out
}
#' @method print mcar_object
#' @export
print.mcar_object <- function(x, ...) {
ni <- x$pat_n
out <- "\nInterpretation of results:\n"
cat("\nMissing data patterns:", (nrow(x$md.pattern) - 1))
if (!is.null(x$removed_patterns)) cat(" used,", nrow(x$removed_patterns), "removed.")
cat("\nCases used:", sum(!x$removed_rows), "\n\n")
if (!is.null(x$hawk_p)) {
if (length(x$hawk_p) > 1) {
hawkp <- median(x$hawk_p)
cat("Hawkins' test: median chi^2 (", x$hawk_df, ") = ", median(x$hawk_chisq), ", median p = ", hawkp, sep = "")
if (any(x$hawk_p < x$alpha) & !hawkp < x$alpha) cat(". Some p-values for Hawkins' test were significant; please inspect their values, e.g., using `plot(", deparse(substitute(x)), ")`", sep = "")
} else {
hawkp <- x$hawk_p
cat("Hawkins' test: chi^2 (", x$hawk_df, ") = ", x$hawk_chisq, ", p = ", x$hawk_p, sep = "")
}
cat("\n\n")
if (x$method == "auto") {
out <- c(
out,
c(
"Hawkins' test is not significant; there is no evidence to reject the assumptions of multivariate normality and MCAR.\n",
"Hawkins' test is significant; if multivariate normality can be assumed, then reject the assumption that missingness is MCAR.\n"
)[(hawkp < x$alpha) + 1]
)
}
}
if (!is.null(x$ad_p)) {
if (length(x$ad_p) > 1) {
adp <- median(x$ad_p)
cat("Anderson-Darling rank test: median T = ", median(x$ad_value), ", median p = ", median(x$ad_p), sep = "")
if (any(x$ad_p < x$alpha) & !median(x$ad_p) < x$alpha) cat(". Some p-values for the Anderson-Darling test were significant; please inspect their values, e.g., using `plot(", deparse(substitute(x)), ")`", sep = "")
} else {
adp <- x$ad_p
cat("Anderson-Darling rank test: T = ", x$ad_value, ", p = ", x$ad_p, sep = "")
}
cat("\n")
if (x$method == "auto") {
out <- c(
out,
c(
"Anderson-Darling test is not significant. There is thus evidence against multivariate normality, but not against MCAR.\n",
"Anderson-Darling test is significant. Reject the assumption that missingness is MCAR."
)[(adp < x$alpha) + 1]
)
}
}
if (x$method == "auto") {
cat(out)
}
}
#' @importFrom graphics hist
#' @importFrom grDevices dev.off
#' @method plot mcar_object
#' @export
plot.mcar_object <- function(x, y, type = NULL, ...) {
if (isTRUE(type == "md.pattern")) {
plot(x[["md.pattern"]])
} else {
op <- par(mar = rep(0, 4))
on.exit(par(op))
dev.off()
if (!is.null(x$hawk_p) & !is.null(x$ad_p)) par(mfrow = c(2, 1))
if (!is.null(x$hawk_p)) {
pct <- sum(x$hawk_p < x$alpha) / length(x$hawk_p)
hist(x$hawk_p, main = NULL, xlab = paste0("Hawkins p-values, ", round(pct * 100), "% significant"), ylab = NULL)
abline(v = x$alpha, col = "red")
}
if (!is.null(x$ad_p)) {
pct <- sum(x$ad_p < x$alpha) / length(x$ad_p)
hist(x$ad_p, main = NULL, xlab = paste0("Anderson-Darling p-values, ", round(pct * 100), "% significant"), ylab = NULL)
abline(v = x$alpha, col = "red")
}
}
}
hawkins <- function(x, grouping) {
p <- ncol(x)
n <- nrow(x)
x <- split(x, factor(grouping))
g <- length(x)
S_pooled <- lapply(x, function(i) {
(nrow(i) - 1) * cov(i, use = "complete.obs")
})
S_pooled <- Reduce("+", S_pooled)
S_pooled <- S_pooled / (n - g)
S_pooled <- solve(S_pooled)
f <- lapply(x, function(i) {
i_centered <- scale(i, center = TRUE, scale = FALSE)
i_centered <- apply(i_centered %*% S_pooled * i_centered, 1, sum)
i_centered <- i_centered * nrow(i)
((n - g - p) * i_centered) / (p * ((nrow(i) - 1) * (n - g) - i_centered))
})
a <- lapply(f, function(thisf) {
pf(thisf, p, (n - g - p), lower.tail = FALSE)
})
list(fij = f, a = a, ni = matrix(sapply(x, nrow), ncol = 1))
}
p_neyman <- function(x, replications = 10000, use_chisq = 30) {
n <- length(x)
n4 <- sum(colSums(legendre(x, 4))^2) / n
if (n < use_chisq) {
sum(sim_neyman(n, replications) > n4) / replications
} else {
pchisq(n4, 4, lower.tail = FALSE)
}
}
sim_neyman <- function(n, replications) {
x <- matrix(runif(replications * n), ncol = replications)
pi <- apply(x, 2, function(i) {
sum(colSums(legendre(i, 4))^2) / n
})
sort(pi)
}
anderson_darling <- function(fij) {
x <- unlist(fij)
ni <- sapply(fij, length)
if (length(ni) < 2) {
stop("At least 2 groups required for Anderson-Darling test.")
}
k <- length(ni)
n <- length(x)
x.sort <- sort(x)[-n]
hj <- rle(x.sort)$lengths
hn <- cumsum(hj)
zj <- x.sort[which(!duplicated(x.sort))]
adk.all <- sapply(fij, function(fi) {
ni <- length(fi)
combs <- expand.grid(zj, fi)
b <- combs[, 1] == combs[, 2]
thisfij <- rowSums(matrix(b, length(zj)))
mij <- cumsum(thisfij)
num <- (n * mij - ni * hn)^2
den <- hn * (n - hn)
(1 / ni * sum(hj * (num / den)))
})
adk <- sum(adk.all) / n
adk.all <- adk.all / n
j <- sum(1 / ni)
h <- sum(1 / seq(1:(n - 1)))
g <- sum(sapply(1:(n - 2), function(i) {
(1 / (n - i)) * sum(1 / seq((i + 1), (n - 1)))
}))
a <- (4 * g - 6) * (k - 1) + (10 - 6 * g) * j
b <- (2 * g - 4) * k^2 + 8 * h * k + (2 * g - 14 * h - 4) *
j - 8 * h + 4 * g - 6
c <- (6 * h + 2 * g - 2) * k^2 + (4 * h - 4 * g + 6) * k +
(2 * h - 6) * j + 4 * h
d <- (2 * h + 6) * k^2 - 4 * h * k
var.adk <- max(((a * n^3) + (b * n^2) + (c * n) + d) / ((n - 1) *
(n - 2) * (n - 3)), 0)
adk.s <- (adk - (k - 1)) / sqrt(var.adk)
b0 <- c(0.675, 1.281, 1.645, 1.96, 2.326)
b1 <- c(-0.245, 0.25, 0.678, 1.149, 1.822)
b2 <- c(-0.105, -0.305, -0.362, -0.391, -0.396)
c0 <- c(
1.09861228866811, 2.19722457733622, 2.94443897916644, 3.66356164612965,
4.59511985013459
)
qnt <- b0 + b1 / sqrt(k - 1) + b2 / (k - 1)
ind <- seq(1:4) + (adk.s <= qnt[3])
yy <- spline(qnt[ind], c0[ind], xout = adk.s)$y
p <- 1 / (1 + exp(yy))
c(p, adk.all)
}
ad <- function(fij) {
x <- unlist(fij)
ni <- sapply(fij, length)
if (length(ni) < 2) {
stop("At least 2 groups required for Anderson-Darling test.")
}
k <- length(ni)
n <- length(x)
x.sort <- sort(x)[-n]
hj <- rle(x.sort)$lengths
hn <- cumsum(hj)
zj <- x.sort[which(!duplicated(x.sort))]
adk.all <- sapply(fij, function(fi) {
ni <- length(fi)
combs <- expand.grid(zj, fi)
b <- combs[, 1] == combs[, 2]
thisfij <- rowSums(matrix(b, length(zj)))
mij <- cumsum(thisfij)
num <- (n * mij - ni * hn)^2
den <- hn * (n - hn)
(1 / ni * sum(hj * (num / den)))
})
adk <- sum(adk.all) / n
adk.all <- adk.all / n
j <- sum(1 / ni)
h <- sum(1 / seq(1:(n - 1)))
g <- sum(sapply(1:(n - 2), function(i) {
(1 / (n - i)) * sum(1 / seq((i + 1), (n - 1)))
}))
a <- (4 * g - 6) * (k - 1) + (10 - 6 * g) * j
b <- (2 * g - 4) * k^2 + 8 * h * k + (2 * g - 14 * h - 4) *
j - 8 * h + 4 * g - 6
c <- (6 * h + 2 * g - 2) * k^2 + (4 * h - 4 * g + 6) * k +
(2 * h - 6) * j + 4 * h
d <- (2 * h + 6) * k^2 - 4 * h * k
var.adk <- max(((a * n^3) + (b * n^2) + (c * n) + d) / ((n - 1) *
(n - 2) * (n - 3)), 0)
adk.s <- (adk - (k - 1)) / sqrt(var.adk)
b0 <- c(0.675, 1.281, 1.645, 1.96, 2.326)
b1 <- c(-0.245, 0.25, 0.678, 1.149, 1.822)
b2 <- c(-0.105, -0.305, -0.362, -0.391, -0.396)
c0 <- c(
1.09861228866811, 2.19722457733622, 2.94443897916644, 3.66356164612965,
4.59511985013459
)
qnt <- b0 + b1 / sqrt(k - 1) + b2 / (k - 1)
ind <- seq(1:4) + (adk.s <= qnt[3])
yy <- spline(qnt[ind], c0[ind], xout = adk.s)$y
p <- 1 / (1 + exp(yy))
list(pn = p, adk.all = adk.all, adk = adk, var.sdk = var.adk)
}
#' @method plot md.pattern
#' @export
plot.md.pattern <- function(x, y, rotate.names = FALSE, ...) {
op <- par(mar = rep(0, 4))
on.exit(par(op))
plot.new()
R <- x[1:nrow(x) - 1, 1:ncol(x) - 1]
nmis <- x[nrow(x), 1:ncol(x) - 1]
if (rotate.names) {
adj <- c(0, 0.5)
srt <- 90
length_of_longest_colname <- max(nchar(colnames(x))) / 2.6
plot.window(
xlim = c(-1, ncol(R) + 1),
ylim = c(-1, nrow(R) + length_of_longest_colname),
asp = 1
)
} else {
adj <- c(0.5, 0)
srt <- 0
plot.window(
xlim = c(-1, ncol(R) + 1),
ylim = c(-1, nrow(R) + 1),
asp = 1
)
}
M <- cbind(c(row(R)), c(col(R))) - 1
shade <- ifelse(R[nrow(R):1, ], mdc(1), mdc(2))
rect(M[, 2], M[, 1], M[, 2] + 1, M[, 1] + 1, col = shade)
for (i in 1:ncol(R)) {
text(i - .5,
nrow(R) + .3,
colnames(x)[i],
adj = adj,
srt = srt
)
text(i - .5, -.3, nmis[order(nmis)][i])
}
for (i in 1:nrow(R)) {
text(ncol(R) + .3, i - .5, x[(nrow(x) - 1):1, ncol(x)][i], adj = 0)
text(-.3, i - .5, rownames(x)[(nrow(x) - 1):1][i], adj = 1)
}
text(ncol(R) + .3, -.3, x[nrow(x), ncol(x)])
}
mice/R/handles.R 0000644 0001762 0000144 00000000470 14330031606 013126 0 ustar ligges users handles.arg <- function(f, a = "data") {
# determine whether function f handles argument a
if (!is.function(f)) {
return(FALSE)
}
a %in% names(formals(f))
}
handles.format <- function(fn) {
# determine whether function fn handles the `format` argument
f <- get(fn)
handles.arg(f, "format")
}
mice/R/D3.R 0000644 0001762 0000144 00000010266 14436133157 011775 0 ustar ligges users #' Compare two nested models using D3-statistic
#'
#' The D3-statistic is a likelihood-ratio test statistic.
#'
#' @details
#' The \code{D3()} function implement the LR-method by
#' Meng and Rubin (1992). The implementation of the method relies
#' on the \code{broom} package, the standard \code{update} mechanism
#' for statistical models in \code{R} and the \code{offset} function.
#'
#' The function calculates \code{m} repetitions of the full
#' (or null) models, calculates the mean of the estimates of the
#' (fixed) parameter coefficients \eqn{\beta}. For each imputed
#' imputed dataset, it calculates the likelihood for the model with
#' the parameters constrained to \eqn{\beta}.
#'
#' The \code{mitml::testModels()} function offers similar functionality
#' for a subset of statistical models. Results of \code{mice::D3()} and
#' \code{mitml::testModels()} differ in multilevel models because the
#' \code{testModels()} also constrains the variance components parameters.
#' For more details on
#'
#' @seealso \code{\link{fix.coef}}
#' @inheritParams D1
#' @return An object of class \code{mice.anova}
#' @references
#' Meng, X. L., and D. B. Rubin. 1992.
#' Performing Likelihood Ratio Tests with Multiply-Imputed Data Sets.
#' \emph{Biometrika}, 79 (1): 103–11.
#'
#' \url{https://stefvanbuuren.name/fimd/sec-multiparameter.html#sec:likelihoodratio}
#'
#' \url{http://bbolker.github.io/mixedmodels-misc/glmmFAQ.html#setting-residual-variances-to-a-fixed-value-zero-or-other}
#' @examples
#' # Compare two linear models:
#' imp <- mice(nhanes2, seed = 51009, print = FALSE)
#' mi1 <- with(data = imp, expr = lm(bmi ~ age + hyp + chl))
#' mi0 <- with(data = imp, expr = lm(bmi ~ age + hyp))
#' D3(mi1, mi0)
#' \dontrun{
#' # Compare two logistic regression models
#' imp <- mice(boys, maxit = 2, print = FALSE)
#' fit1 <- with(imp, glm(gen > levels(gen)[1] ~ hgt + hc + reg, family = binomial))
#' fit0 <- with(imp, glm(gen > levels(gen)[1] ~ hgt + hc, family = binomial))
#' D3(fit1, fit0)
#' }
#' @export
D3 <- function(fit1, fit0 = NULL, dfcom = NULL, df.com = NULL) {
if (!missing(df.com)) {
warning("argument df.com is deprecated; please use dfcom instead.",
call. = FALSE
)
dfcom <- df.com
}
dfcom <- get.dfcom(fit1, dfcom)
call <- match.call()
fit1 <- getfit(fit1)
m <- length(fit1)
est1 <- pool(fit1, dfcom = dfcom)
qbar1 <- getqbar(est1)
if (is.null(fit0)) {
# test all estimates equal to zero
beta <- rep(0, length(qbar1))
names(beta) <- names(qbar1)
fit0 <- lapply(fit1, fix.coef, beta = beta)
} else {
fit0 <- getfit(fit0)
}
est0 <- pool(fit0, dfcom = dfcom)
qbar0 <- getqbar(est0)
k <- length(qbar1) - length(qbar0)
# For each imputed dataset, calculate the deviance between the two
# models as fitted
dev1.M <- -2 * lapply(fit1, glance) %>%
bind_rows() %>%
pull(.data$logLik)
dev0.M <- -2 * lapply(fit0, glance) %>%
bind_rows() %>%
pull(.data$logLik)
# For each imputed dataset, calculate the deviance between the two
# models with coefficients restricted to qbar
mds1 <- lapply(fit1, fix.coef, beta = qbar1)
dev1.L <- -2 * lapply(mds1, glance) %>%
bind_rows() %>%
pull(.data$logLik)
mds0 <- lapply(fit0, fix.coef, beta = qbar0)
dev0.L <- -2 * lapply(mds0, glance) %>%
bind_rows() %>%
pull(.data$logLik)
deviances <- list(
dev1.M = dev1.M, dev0.M = dev0.M,
dev1.L = dev1.L, dev0.L = dev0.L
)
# scaled deviance, as fitted
dev.M <- mean(dev0.M - dev1.M)
# scaled deviance, restricted
dev.L <- mean(dev0.L - dev1.L)
rm <- ((m + 1) / (k * (m - 1))) * (dev.M - dev.L)
Dm <- dev.L / (k * (1 + rm))
# Degrees of freedom for F distribution
v <- k * (m - 1)
if (v > 4) {
w <- 4 + (v - 4) * ((1 + (1 - 2 / v) * (1 / rm))^2)
} else {
w <- v * (1 + 1 / k) * ((1 + 1 / rm)^2) / 2
}
pvalue <- pf(Dm, k, w, lower.tail = FALSE)
test <-
out <- list(
call = match.call(),
result = c(Dm, k, w, pvalue, rm),
formulas = list(
`1` = formula(getfit(fit1, 1L)),
`2` = formula(getfit(fit0, 1L))
),
m = m,
method = "D3",
use = NULL,
dfcom = dfcom,
deviances = deviances
)
class(out) <- c("mice.anova", class(fit1))
out
}
mice/NEWS.md 0000644 0001762 0000144 00000107332 14433407421 012275 0 ustar ligges users # mice 3.16.0
### Major changes
* Expands `futuremice()` functionality by allowing for external packages and user-written functions (#550). Contributed @thomvolker
* Adds GH issue templates `bug_report`, `feature_request` and `help_wanted` (#560). Contributed @hanneoberman
### Minor changes
* Removes documentation files for `rbind.mids()` and `cbind.mids()` to conform to CRAN policy
* Adds `mitml` and `glmnet` to imports so that test code conforms to `_R_CHECK_DEPENDS_ONLY=true` flag in `R CMD check`
* Initializes random number generator in `futuremice()` if there is no `.Random.seed` yet.
* Updates GitHub actions for package checking and site building
* Preserves user settings in `predictorMatrix` for case F by adding a `predictorMatrix` argument to `make.predictorMatrix()`
* Polishes `mice.impute.mpmm()` example code
### Bug fixes
* Adds proper support for factors to `mice.impute.2lonly.pmm()` (#555)
* Solves function naming problems for S3 generic functions `tidy()`, `update()`, `format()` and `sum()`
* Out-comments and weeds example&test code to silence `R CMD check` with `_R_CHECK_DEPENDS_ONLY=true`
* Fixes small bug in `futuremice()` that throws an error when the number of cores is not specified, but the number of available cores is greater than the number of imputations.
* Solves a bug in `mice.impute.mpmm()` that changed the column order of the data
# mice 3.15.0
### Major changes
* Adds a function `futuremice()` with support for parallel imputation using the `future` package (#504). Contributed @thomvolker, @gerkovink
* Adds multivariate predictive mean matching `mice.impute.mpmm()`. (#460). Contributed @Mingyang-Cai
* Adds `convergence()` for convergence evaluation (#484). Contributed @hanneoberman
* Reverts the internal seed behaviour back to `mice 3.13.10` (#515). #432 introduced new local seed in response to #426. However, various issues arose with this facility (#459, #492, #502, #505). This version restores the old behaviour using global `.Random.seed`. Contributed @gerkovink
* Adds a `custom.t` argument to `pool()` that allows the advanced user to specify a custom rule for calculating the total variance $T$. Contributed @gerkovink
* Adds new argument `exclude` to `mice.impute.pmm()` that excludes a user-specified vector of values from matching. Excluded values will not appear in the imputations. Since the observed values are not imputed, the user-specified values are still being used to fit the imputation model (#392, #519). Contributed @gerkovink
### Minor changes
* Styles all `.R` and `.Rmd` files
* Makes post-processing assignment consistent with lines 85/86 in `sampler.R` (#511)
* Edit test broken on R<4 (#501). Contributed @MichaelChirico
* Adds support for models reporting contrasts rather than terms (#498). Contributed @LukasWallrich
* Applies edits to autocorrelation function (#491). Contributed @hanneoberman
* Changes p-value calculation to more robust alternative (#494). Contributed @AndrewLawrence
* Uses `inherits()` to check on class membership
* Adds decprecation notices to `parlmice()`
* Adapt `prop`, `patterns` and `weights` matrices for pattern with only 1's
* Adds warning when patterns cannot be generated (#449, #317, #451)
* Adds warning on the order of model terms in `D1()` and `D2()` (#420)
* Adds example code to fit model on train data and apply to test data to `mice()`
* Adds example code on synthetic data generation and analysis in `make.where()`
* Adds testfile `test-mice.impute.rf.R`(#448)
### Bug fixes
* Replaces `.Random.seed` reads from the `.GlobalEnv` by `get(".Random.seed", envir = globalenv(), mode = "integer", inherits = FALSE)`
* Repairs capitalisation problems with `lastSeedValue` variable name
* Solves `x$lastSeedValue` problem in `cbind.mids()` (#502)
* Fixes problems with `ampute()`
* Preserves stochastic nature of `mice()` by smarter random seed initialisation (#459)
* Repairs a `drop = FALSE` buglet in `mice.impute.rf()` (#447, #448)
* @str-amg reported that the new dependency on `withr` package should have version 2.4.0 (published in January 2021) or higher. Versions `withr 2.3.0` and before may give `Error: object 'local_seed' is not exported by 'namespace:withr'`. Either update manually, or install the patched version `mice 3.14.1` from GitHub. (#445). NOTE: `withr` is no longer needed in `mice 3.15.0`
# mice 3.14.0
### Major changes
* Adds four new univariate functions using the lasso for automatic variable selection:
| Function | Description |
| --------------------------------- | --------------------------------- |
|`mice.impute.lasso.norm()` | Lasso linear regression |
|`mice.impute.lasso.logreg()` | Lasso logistic regression |
|`mice.impute.lasso.select.norm()` | Lasso selector + linear regression |
|`mice.impute.lasso.select.logreg()`| Lasso selector + logistic regression |
Contributed by @EdoardoCostantini (#438).
* Adds Jamshidian && Jalal's non-parametric MCAR test, `mice::MCAR()` and associated plot method. Contributed by @cjvanlissa (#423).
* Adds two new functions `pool.syn()` and `pool.scalar.syn()` that specialise pooling estimates from synthetic data. The `"reiter2003"` pooling rule assumes that synthetic data were created from complete data. Thanks Thom Volker (#436).
* Avoids changing the global `.Random.seed` (#426, #432) by implementing `withr::local_preserve_seed()` and `withr::local_seed()`. This change provides stabler behavior in complex scripts. The change does not appear to break reproducibility when `mice()` was run with a seed. Nevertheless, if you run into a reproducibility problem, install `mice 3.13.12` or before.
* Improves the imputation of parabolic data in `mice.impute.quadratic()`, adds a parameter `quad.outcome` containing the name of the outcome variable in the complete-data model. Contributed @Mingyang-Cai, @gerkovink (#408)
* By default, `mice.impute.rf()` now uses the faster `ranger` package as back-end instead of `randomForest` package. If you want the old behaviour specify the `rfPackage = "randomForest"` argument to the `mice(...)` call. Contributed @prockenschaub (#431).
* Generalises `pool()` so that it processes the parameters from all `gamlss` sub-models. Thanks Marcio Augusto Diniz (#406, #405)
* Uses the robust standard error estimate for pooling when `pool()` can extract `robust.se` from the object returned by `broom::tidy()` (#310)
### Bug fixes
* Contains an emergency solution as `install.on.demand()` broke the standard CRAN workflow. mice 3.14.0 does not call `install.on.demand()` anymore for recommended packages. Also, `install.on.demand()` will not run anymore in non-interactive mode.
* Repairs an error in the `mice:::barnard.rubin()` function for infinite `dfcom`. Thanks @huftis (#441).
* Solves problem with `Xi <- as.matrix(...)` in `mice.impute.2l.lmer()` that occurred when a cluster contains only one observation (#384)
* Edits the `predictorMatrix` to a monotone pattern if `visitSequence = "monotone"` and `maxit = 1` (#316)
* Solves a problem with the plot produced by `md.pattern()` (#318, #323)
* Fixes the intercept in `make.formulas()` (#305, #324)
* Fixes seed when using `newdata` in `mice.mids()` (#313, #325)
* Solves a problem with row names of the `where` element created in `rbind()` (#319)
* Solves a bug in mnar imputation routine. Contributed by Margarita Moreno Betancur.
### Minor changes
* Replaces URL to jstatsoft with DOI
* Update reference to literature (#442)
* Informs the user that `pool()` cannot take a `mids` object (#433)
* Updates documentation for post-processing functionality (#387)
* Adds Rcpp necessities
* Solves a problem with "last resort" initialisation of factors (#410)
* Documents the "flat-line behaviour" of `mice.impute.2l.lmer()` to indicate a problem in fitting the imputation model (#385)
* Add reprex to test (#326)
* Documents that multivariate imputation methods do not support the `post` parameter (#326)
# mice 3.13.0
### Major changes
* Updated `mids2spss()` replaces the `foreign` by `haven` package. Contributed Gerko Vink (#291)
### Minor changes
* Repairs an error in `tests\testhat\test-D1.R` that failed on `mitml 0.4-0`
* Reverts `with.mids()` function to old version because the change in commit 4634094 broke downstream package `metafor` (#292)
* Solves a glitch in `mice.impute.rf()` in finding candidate donors (#288, #289)
# mice 3.12.0
### Much faster predictive mean matching
* The new `matchindex` C function makes predictive mean matching **50 to 600 times faster**.
The speed of `pmm` is now on par with normal imputation (`mice.impute.norm()`)
and with the `miceFast` package, without compromising on the statistical quality of
the imputations. Thanks to Polkas
