Hi Bob,
The textbook by J. Scott Long (Regression Models for Categorical and Limited
Dependent Variables; Sage Publications, 1997) discusses the Brant test. On
page 144, there is a table with the omnibus test as well as the tests for
the individual variables, showing (as in your case) that "there is strong
evidence for the violation of the assumption for some variables but not for
others". Long does not include any further recommendation.
One thing you could do is to drop the variables (ie, those for which the
assumption of proportional odds is violated) from your model -- that is, if
you don't have strong a priori reasons for including them in the model in
the first place.
The other thing to do, which seems to me to be more appropriate, would be to
fit a model that does not require the assumption of proportional odds.
Generalized ordered logit would be one such model.
Cheers,
Alex
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of [email protected]
Sent: Wednesday, August 04, 2004 4:36 PM
To: [email protected]
Subject: st: Conflicting Brant Test Results
I estimated some ordered logit models and checked whether the proportional
odds assumption was violated using the Brant test. In some cases the
omnibus Brant test suggests the assumption is violated, but Brant tests of
the individual regressors in the same models suggest the assumption is not
violated.
What should one conclude in this case? Is it best to be cautious and use,
say, generalized ordered logit, or is it reasonable to stick with the
ordered logit results? Thanks in advance for any advice.
Bob Kaminski
*
* 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/
