preCgm
and
missmodel
. A default method operates on matrices and data frames.
impCgm(object, ...)
"preCgm"
or
"missmodel"
.
nimpute
which specifies the number of imputations. See the specific
method for a list of all arguments.
mi
(multiple imputations) object is returned;
see
and
for details.
Producing proper multiple imputations using data augmentation requires
that the sequence of parameters and imputations has converged to
stationarity. The concept of stationarity important to multiple
imputation is that the imputations are approximately independent draws
from the conditional distribution of the missing data, given the
observed data. You may assess stationarity using a variety of
S-PLUS
functions.
Once stationarity is reached,
impCgm
may be used to produce
multiple imputations in one of two ways: (1) one long chain or (2)
parallel chains.
Best, N. G., Cowles, M. K. and Vines, S. K. (1997),
CODA Convergence,
Diagnosis and Output Analysis Software for Gibbs sampling output ,
Version 0.4.,
Cambridge: Medical Research Council Biostatistics Unit.
Gilks, W.R., Richardson, S. and Spiegelhalter, D. J., editors (1996),
Markov Chain Monte Carlo in Practice ,
London: Chapman and Hall.
Schafer, J. L. (1997),
Analysis of Incomplete Multivariate Data ,
Chapman & Hall, London.
# See help files for emCgm() and daCgm() to create # language.da and language.em, which are # used as a starting values for the imputation algorithm. #Generate 10 imputations, using a single chain #with 250 cycles between imputations language.imp <- impCgm(language.da, nimpute = 10, control = list(niter = 250)) impCgm.missmodel(language.da, nimpute = 10, control = list(niter = 250)) #same miPrint(language.imp) #extract the second set of imputations miSubscript(language.imp, 2) impCgm(language.da, nimpute = 10, control = list(niter = 250))