Violation of IIA assumption while modeling a Multinomial Logit Model

[Jays Dutta]

Is it absolutely mandatory to not violate IIA assumption while modeling a Multinomial Logit Model ? I had studied in 8.4 section of the book by Long and Freese about the inconsistencies of such tests. They started the -mlogtest- after all. But do reviewers strictly want IIA results to be not violated during MNL at all? I did the following -mlogit- using three categories of fuels and on running the -mlogtest- command I get the following results. Note that in the first Hausman test my p-values are all significant and two of he chi2 values are positive and one negative. What does this imply?Every other measure of Fit test etc are coming satisfactory though. Finally,Can I still justify this model and my paper with MNLM to reviewers despite such IIA test results as below? Please please help. Need some guidance during this lockdown far away from university

----------------------------------------------------------------------
Hausman tests of IIA assumption (N=101457)

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

| chi2 df P>chi2
-----------------+-------------------------
Primitiv | -214.024 19 .
Transiti | 4288.460 19 0.000
Advanced | 102.933 19 0.000

Note: A significant test is evidence against Ho.
Note: If chi2<0, the estimated model does not meet asymptotic assumptions.
-----------------------------------------------------------------------------------------------------------
suest-based Hausman tests of IIA assumption (N=101457)

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

| chi2 df P>chi2
-----------------+-------------------------
Primitiv | 569.718 19 0.000
Transiti | 322.106 19 0.000
Advanced | 838.009 19 0.000

Note: A significant test is evidence against Ho.
--------------------------------------------------------------------------------------------------------------
Small-Hsiao tests of IIA assumption (N=101457)

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

| lnL(full) lnL(omit) chi2 df P>chi2
-----------------+-----------------------------------------------
Primitive Fuels | -3978.750 -3951.535 54.431 19 0.000
Transition Fuels | -2.35e+04 -2.35e+04 28.575 19 0.073
Advanced Fuels | -3607.479 -3580.977 53.003 19 0.000

Note: A significant test is evidence against Ho.

[Phil Bromiley]

Welcome to Stata list. You will increase your chances of useful answer by following the FAQ on asking questions-provide Stata code in code delimiters, readable Stata output (which you do), and sample data using dataex.

It looks like you may need an estimator that handles IIA. As I understand it, a nested logit is often used for this.

[Jays Dutta]

My data is a large scale survey data of above 1 lakh households. Is it still possible that IIA is fulfilled in such a model?