Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Re: hausman test after mlogit - negative chi-square


From   "Clive Nicholas" <Clive.Nicholas@newcastle.ac.uk>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Re: hausman test after mlogit - negative chi-square
Date   Wed, 22 Mar 2006 03:07:50 -0000 (GMT)

Marilyn Ibara wrote:

> I estimated a Multinominal logit model with 7 choices.  I
> have 17 independent variables.  I specified mlogtest after I
> ran the mlogit command.

It would have been worth mentioning the _full_ command (I suspect) you
used to get the post-estimation results shown in your post: -mlogtest,
all-.

> I am not sure about my results.  Does anyone know how it is
> possible to get a negative Chi-square for the hausman test
> for IIA as I did in my results?  Also, how are the degrees
> of freedom calculated, it appears that I should have 17 and
> not what is reported.

This question is answered by Long and Freese (2006: 244-5) in the second
edition of their excellent textbook, cited below my signature.
Essentially, this is very common, and it is often evidence that the
independence of irrelevant alternatives (IIA) has _not_ been violated.
They also suggest a good tip to confirm your results here: choose a
different base category and run the same -mlogtest- command.

> Furthermore, my results of the hausman test differ from the
> small-hsiao test, they contradict each other, I read online
> this is common, does anyone know how one determines which
> one is appropriate?

Long and Freese (2006: 243), quoting Monte Carlo research conducted by
Cheng and Long (2005), report that neither the Hausman-McFadden (HM) nor
the Small-Hsiao (SH) test is especially useful for assessing violations of
IIA. This is largely because both have poor size properties even with some
kinds of data, even N > 1000 in the case of HM but also sometimes for SH
when N > 500. The best advice they give is only to use -mlogit- when you
can _clearly_ distinguish between the outcome categories in your dataset.
Quite frankly, I couldn't agree more.

In sum, buy Long and Freese!

CLIVE NICHOLAS        |t: 0(044)7903 397793
Politics              |e: clive.nicholas@ncl.ac.uk
Newcastle University  |http://www.ncl.ac.uk/geps

References:

Cheng S and Long JS (2005) "Testing For IIA in the Multinomial Logit
Model",  University of Connecticut, Working Paper.

Long JS and Freese J (2nd ed, 2006) REGRESSION MODELS FOR CATEGORICAL
DEPENDENT VARIABLES USING STATA, College Station, TX: Stata Press.

*
*   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/



© Copyright 1996–2020 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index