It is quite common that the tests give differing results. The SH test
*randomly* divides the data into subsets. You can probably reproduce this by
using different random seeds (-set seed ...-), which might give you
different results (which does not mean you should look for a seed until IIA
holds...). Inconsistencies do happen quite often (see also the book of Long
& Freese). They also note that McFadden suggests that MNL should only be
used when you ccan plausibly assume that the categories are distinct, i.e.
alternatives are dissimilar. That might not be the case with your data.
Greetings,
Daniel
> Dear Stata users
>
> I am testing for IIA after mlogit using the mlogtest command
> (Some output below).
>
> Is there any other testing technique/command available?
>
> Also, base on the Hausman test my interpretation is that IIA
> is OK with my empirical model, but based on the Small-Hsiao
> test my interpretation is the opposite. Am I interpreting
> the results in the right way? Is it common to get opposite
> results with these two tests?
>
> Thanks in advanced for any help
>
> Adrian Gonzalez
> Ph.D Candidate
> Agricultural, Environmental and Development Economics The
> Ohio State University
>
> **** Hausman test of IIA assumption
>
> Test: Ho: difference in coefficients not systematic
>
> *** Summary of results
>
> Omitted | chi2 df P>chi2 evidence
> ---------+------------------------------------
> Reschedu | -0.790 14 --- for Ho
> Partiall | -0.253 14 --- for Ho
> 0_None | -0.198 28 --- for Ho
> 0_Work | -0.360 28 --- for Ho
> 0_Savi | 0.065 28 1.000 for Ho
> 0_W+Sav | 0.404 28 1.000 for Ho
> -_None | 0.076 28 1.000 for Ho
> -_Work | -0.002 28 --- for Ho
> -_W+Sav | 3.098 28 1.000 for Ho
> 0_W+Sav+ | 0.106 28 1.000 for Ho
> -_W+Sav+ | 0.479 28 1.000 for Ho
> More_30 | -2.629 28 --- for Ho
> ----------------------------------------------
> Note: If chi2<0, the estimated model does not meet
> asymptotic assumptions of the test.
>
>
> **** Small-Hsiao tests of IIA assumption
>
> Ho: Odds(Outcome-J vs Outcome-K) are independent of other
> alternatives.
>
> Omitted | lnL(full) lnL(omit) chi2 df P>chi2 evidence
> ---------+---------------------------------------------------------
> Reschedu | -1373.820 -1033.875 679.889 14 0.000 against Ho
> Partiall | -1370.112 -1050.340 639.542 14 0.000 against Ho
> 0_None | -1280.828 -901.153 759.350 14 0.000 against Ho
> 0_Work | -1337.598 -952.108 770.981 14 0.000 against Ho
> 0_Savi | -1374.106 -1029.493 689.227 14 0.000 against Ho
> 0_W+Sav | -1334.401 -948.032 772.738 14 0.000 against Ho
> -_None | -1371.393 -983.072 776.642 14 0.000 against Ho
> -_Work | -1417.680 -1028.537 778.287 14 0.000 against Ho
> -_W+Sav | -1327.989 -944.182 767.614 14 0.000 against Ho
> 0_W+Sav+ | -1423.891 -1037.675 772.432 14 0.000 against Ho
> -_W+Sav+ | -1233.047 -1005.512 455.069 14 0.000 against Ho
> More_30 | -1428.613 -1017.914 821.399 14 0.000 against Ho
> -------------------------------------------------------------------
>
> *
> * For searches and help try:
> * http://www.stata.com/support/faqs/res/findit.html
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
