Re: st: Panel Multinomial Logistic Model

[David Jacobs <jacobs.184@sociology.osu.edu>]

First a random-effects multinomial logit routine may be in 
Limdep.  Although that program is difficult to use, employing it may 
be less trouble than what you propose.

Second, an alternative solution often recommended on the list when no 
pooled time series estimator is available in Stata is to simply 
cluster on cases.  While there's no adjustment for separate 
case-specific intercepts, at least the standard errors may be OK.

I can't evaluate your proposal, however.  Let's see if you get any 
response to this interesting idea.

Dave Jacobs

At 01:38 PM 5/24/2010, you wrote:
> Hi,
>
> I understand that there is not a stata command for multinomial 
> logistic model for panel data estimation.  I wonder if the following 
> can be done for a three-outcome categorical dependent variable (say, 0, 1, 2):
>
> 1.  Estimate a panel logit for outcome=1, and predict the 
> exp(xb+random effects),
> 2.  Estimate a panel logit for outcome=2, and predict the 
> exp(xg+random effects), and
>
> 3.  Hand calculate a panel multinomial logit for the prob(outcome=1) 
> by scrambling the predictions of the two panel logits, namely, 
> exp(xb+random effects)*(1+exp(xb+random effects)+exp(xg+random effects))^-1
>
> I believe that this equivalence exists between multinomial logit and 
> a series of binary logit regressions in a non-panel setting.  I 
> wonder if the same logic extends to panel data.
>
> Thanks in advance,
>
> Albert Lee
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/