I have the following questions on the EM and multiple imputation.
1. Assuming my response is fully observed and I have missingness in my
continuous covariates, is there a way to compute fitted values after using
the EM algorithm to get estimates for the parameters?
2. I recently read a paper by Ibrahim, Lipsitz and Chen (1999) in
Biometrics, where they used the Monte Carlo EM to maximise the likelihood
when there were missing continuous covariates, is this similar to the use of
data augmentation after EM (as used by Schafer)? Infact the paper by Wei and
Tanner (JASA, 1990), proposing the Monte Carlo EM states that the
imputations in the E-step are called multiple imputations by Rubin.
Thanks.
Vumani Dlamini
Central Statistical Office
Ministry of Economic Planning
Swaziland
_________________________________________________________________
STOP MORE SPAM with the new MSN 8 and get 2 months FREE*
http://join.msn.com/?page=features/junkmail
