st: MI estimate, no observations in some imputations

["Lucy Bilaver" <lbilaver@niu.edu>]

Dear Stata listers,
I am trying to understand how to address the problem of having no observations in some imputations when
trying to use mi estimate.  The problem variable is a 4-level categorical variable with some very low frequency
levels.  Apparently one of the 4 levels has zero responses in some imputations.  I tried to use mi estimate (Stata 13) with the
following command:

mi estimate, esampvaryok: proportion behav_cat

"behav_cat" takes the values 1-4 and was created using mi passive.  The passive variable is the combination of two imputed variables.
Stata reports "mi estimate:  no observations in some imputations.   This is not allowed".

I am wondering whether I should have imputed the 4 level categorical variable instead of separately computing the two categorical
variables that make it up.  Based on some other posts I read, perhaps I should think of this situation as comparable to the issue
of imputing an interaction term versus creating it from two imputed variables.  I am concerned though that this would still lead to the
situation where some of the levels of the imputed categorical variable have "no observations".  I would also like to understand why
this situation is so problematic.  Thanks for the advice.
-Lucy Bilaver





*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/