I have a question regarding predicted probabilities using multiply imputed data.
I know that mim doesn't work with prvalue. As a workaround I was going to do the following using the delta method after consulting the Stata FAQ:
http://www.stata.com/support/faqs/stat/delta.html
xi: mim:logit ixhi12re diimp_depriskgroup $int_catbiophysical, or cluster (wcrsid) robust
mim: predict yesoutpatient
mim: predict stdp, stdp
gen se = yesoutpatient*(1-yesoutpatient)*stdp
But then it is unclear to me how this could be adapted to subgroups of interest (i.e. when my primary explanatory variable of interest is held at certain values (i.e. when diimp_depriskgroup= 1 and then again diimp_depriskgroup=0).
Would it be the following?
mim: mean yesoutpatient if diimp_depriskgroup==1
mim: mean se if diimp_depriskgroup == 1
The standard errors with this approach are consistent with the standard errors I get when I use prvalue (the delta method), running a logistic regression model on just one of the multiply imputed datasets, but I wasn't sure if others have run into this issue.
However, I am also running multinomial logistic regression models and I'm not sure how exactly one could get around running separate models on each of the datasets and perhaps reporting the mean predicted probability from the 5 datsets with the largest standard error?
Thank you for your time.
Jessica
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
