st: Hausman, smhsiao or suest?

["Pungkas B. Ali" <pungkas.ali@gmail.com>]

Dear Statalisters

I am developing discrete models of health provider choices (with 5
alternatives).

I use Multinomial logit (-mlogit-) and test IIA assumptions using
-mlogtest, hausman- and -mlogtest, smhsiao- . Both gave different conclusion
with regard to Ho rejection (Smhsiao supports the IIA assumption).  

I further tried -hausman full reduced- after running full and reduced form
but it fails to meet the asymptotic assumptions.  Finally I used -mlogtest,
suest- and it gave different result 

Question : For IIA test choices, some suggest me to choose the safest (no
further model modification) which is smhsiao. Is this suggestion
justifiable? Why did those 4 tests come to different conclusions?

Notes: I try to avoid -nlogit- for I will include alternative specific
characteristics in my next full model (using -asclogit- or -mixlogit-)

Thanks for help
Pungkas ali





*
*   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/