st: IIA violations and subsequent steps

[Douglas Garrett <d.garrett@utoronto.ca>]

Hello!
  I am currently running a multinomial logistic regression model and  
have found that it violates the IIA assumption.  Of course, one option  
in this case is to fit a nested logit model to these data to allow  
error terms to correlate within parent-levels of the nested structure.  
 However, all of my predictors describe characteristics of  
participants rather than of choices.  To my understanding, this  
person-level type of predictor is not appropriate for a nested logit  
model, as predictors are required typically that vary along with the  
choices themselves.  If that is correct then, how does one handle IIA  
violations in a multinomial logistic model with predictors that are  
choice-invariant?  Might there be a work-around to allow error terms  
of certain choices to correlate prior to running a multinomial model  
such as this?

Many thanks in advance.



Douglas Garrett
Grady Lab
Rotman Research Institute
Baycrest
Department of Psychology
University of Toronto






*
*   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/