[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: ml for a missing data problem
Has anyone written a program to do FIML (full likelihood maximum likelihood)
estimation (in the presence of missing data)?
To be concrete, suppose we have a MVN distribution of (Y, X1, X2, ... , Xp),
and we want to estimate the regression of Y on the X vector. We would like
to use all the available information, rather then list-wise deletion.
The data will have to first be inspected for missing values patterns, and
contribution to the likelihood will have to be computed separately within
each pattern, and then combined. I think it will be necessary to estimate
the mu vector of expectation and the sigma matrix of var-cov of the vector
(Y, X1, X2, ... , Xp), and then use a transformation to get the vector of
betas for the regression coefficients and its var-cov matrix.
I know that there are commercial programs out there that can do this. AMOS
is one such example.
Has anyone done this in Stata?
Thanks for your help,
Estie Sid Hudes, PhD MPH
University of California
Prevention Sciences Group &
Center for AIDS Prevention Studies
74 New Montgomery Street, Suite 600
S.F. CA 94105
* For searches and help try: