Hausman tests of IIA after mlogit: "estimated model does not meet asymptotic assumptions"

[Bruno Hoepers]

Hi all. I'm working on a project with data from the Brazilian military. The dependent variable is promotion to general and has 4 categories (colonel = 0, brigadier gen = 1, division gen = 2, army gen = 3).

Originally, I tried to estimate an ologit model, but the model failed to meet the parallel regression assumption (Brant test).

Next, I considered gologit2. However, the model is too difficult to interpret. Thus, I decided to verify the possibility of using mlogit.

I ran the mlogit model and ran the Hausman's IIA assumption test using mlogtest (J. Scott Long & Jeremy Freese. 2014. Regression Models for Categorical Dependent Variables Using Stata, 3rd Edition. College Station, TX: Stata Press).

Here is what I got:

Hausman tests of IIA assumption (N=1094)

// Ho: Odds(Outcome-J vs Outcome-K) are independent of other alternatives
//
// | chi2 df P>chi2
//------------- +----------------------------
// 0 | 4.978 8 0.760
// 1 | 2.463 42 1.000
// 2 | -6.935 44 .
// 3 | -7.232 65 .
//
// Note: A significant test is evidence against Ho.
// Note: If chi2<0, the estimated model does not meet asymptotic assumptions.


Since my model doesn't meet asymptotic assumptions, how can I proceed?

Thanks.

[Bruno Hoepers]

About gologit2 - from Williams, Richard. 2006. "Generalized Ordered Logit/ Partial Proportional Odds Models for Ordinal Dependent Variables." The Stata Journal 6(1):58-82.