preGauss
and
missmodel
. A default method operates on matrices and data frames.
impGauss(object, ...)
preGauss
or
missmodel
.
nimpute
which specifies the number of imputations. See the specific
function called for a list of all arguments.
mi
(multiple imputations) object is returned.
See the help files for
,
,
,
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,
impGauss
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.
#Generate list of starting values using a bootstrap start <- list() for(i in 1:5) start[[i]] <- paramIter(emGauss(cholesterol, subset=sample(1:28,14,T), prior="ml")) # draw 5 imputations from 5 parallel chains, each run for 50 iterations cholesterol.imp <- impGauss(cholesterol, prior="non", start=start, control=list(niter=50))