st: Comparing multinomial models in survey data

["Polis, Chelsea B." <cpolis@jhsph.edu>]

Dear Statalist,

I am doing multinomial regression using data that has been svyset.  Have I correctly interpreted 
Hosmer and Lemeshow's "Applied Logistic Regression" (p 219-20) in understanding that a legitimate 
approach is to ignore the survey design when comparing models (i.e., when using fitstat in SPost)
by running the models without the svy command, use the commands such as fitstat to aid in selection 
of a final model, and then run the final model with the svy command to obtain more accurate standard 
errors?

If so, I have a second question.  If my final model(s) under svy will also use the subpop command 
(I need separate models for males and females), is it appropriate when doing the process outlined 
above (of ignoring survey weights for the purpose of comparing models) to restrict by using an if 
command (i.e., xi: mlogit depvar i.indepvar i.indepvar2 i.indepvar3 if gender==1)?

Many thanks,
Chelsea

*
*   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/