Dear Statalists,

I am currently using a multinomial logit model, where the dependent variable has four categories, 0, 1, 2, and 3. The base case is category 0. Before running the multinomial logit regressions, I test the independence of irrelevant alternatives (IIA) assumption. I use two different tests: Hausman test and Small-Hsiao test.

The result of Hausman test shows below:
Code:
. mlogtest, haus base 

**** Hausman tests of IIA assumption (N=3106)

 Ho: Odds(Outcome-J vs Outcome-K) are independent of other alternatives.

 Omitted |      chi2   df   P>chi2   evidence
---------+------------------------------------
       1 |   257.499   28    0.000   against Ho
       2 |     0.000    1    1.000   for Ho    
       3 |     0.000    1    1.000   for Ho    
       0 |     0.000    2    1.000   for Ho    
----------------------------------------------

.
From the above results, I guess the IIA assumption is violated.

Then I use the Small-Hsiao test:
Code:
. mlogtest, smhsiao base 

**** Small-Hsiao tests of IIA assumption (N=3106)

 Ho: Odds(Outcome-J vs Outcome-K) are independent of other alternatives.

 Omitted |  lnL(full)  lnL(omit)    chi2   df   P>chi2   evidence
---------+---------------------------------------------------------
       1 |   -488.683   -476.399  24.568   30    0.746   for Ho    
       2 |  -1213.261  -1197.199  32.123   30    0.362   for Ho    
       3 |  -1051.023  -1032.170  37.705   30    0.157   for Ho    
       0 |   -376.918   -355.686  42.464   30    0.065   for Ho    
-------------------------------------------------------------------


.
Small-Hsiao shows that IIA is not violated.

In my case, Hausman and Small-Hsiao give different results. Could anyone help with which test I should follow and can I still use multinomial logit regression? Many thanks.