### Friday, February 24, 2006

## Re: st: Nested Logit with Panel Data

The terms "fixed effects" and "random effects" are really unfortunate in the panel data. Those are in fact "conditional estimates" and... well, "random effects" is probably proper. With the "fixed effect" linear regression estimator, you are in fact conditioning on the fixed effects; and when you do the same in the logit regression, you get -clogit-, which is far from being equivalent to inserting a bunch of year dummies. If you are lucky enough to find sufficient statistics, the fixed effects estimator is feasible; this limits the class of models to the exponential family. With the random effects, you integrate the random effect out, which in the linear case magically coincides with the GLS regression due to underlying normality that nobody cares to remember about...

Think now about where the random and fixed effect will go into your nonlinear nested logit model, and how you will either condition on those, to call this a fixed effect estimator, or integrate those out, to call this a random effect estimator. You would probably need one effect per each branching of your nested logit; I wouldn't even try to think of getting it right for fixed effects. And for the random effects, you would need to specify a rather big covariance matrix.

I think -gllamm- has -mlogit- as one of the links; that might be the only way to specify panel multiple outcomes in Stata thus far.

On 2/22/06, Matthias Grueninger <m.grueninger@unibas.ch> wrote: > > > Dear all, > > In the statalist archive I read that the command nlogit cannot be used > to estimate a nested logit model with panel data > (http://www.stata.com/statalist/archive/2005-08/msg00735.html) > > - Is there another possibility to estimate the nested logit model with > panel data, or do I have to estimate pooled? > > - Does it make sense / is it allowed to use year dummies if I use the > nlogit command (if I have a panel)? > > - If I have to estimate a pooled model, is there at least a possibility to > correct for the fact that it is a panel (something like the option 'cluster')? > > - If I include year dummies and also dummies for every entity and then use > the nlogit command, don't I estimate > a (fixed effect) nested logit model with panel data? > > > Thanks a lot for any help! > > Matthias > > > > * > * 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/ >

-- Stas Kolenikov http://stas.kolenikov.name

* * 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/

Tag: statalist