Scott Cunningham <[email protected]> followed up with a couple more
questions:
> > If you are concerned about overdispersion, there is also -xtnbreg-.
>
> Now, I may expose my ignorance here, but aren't I correcting the
> overdispersion problem in Poisson with FE by using the vce(boot)
> suffix? My impression was that this was dealing with that problem.
To be honest, mentioning -xtnbreg- in this case was a red herring. My
knee-jerk reaction whenever I hear 'overdispersion' in a poisson model context
is to mention the negative-binomial model. Unfortunatly (for me, since I'm
the one who mentioned it), -xtnbreg- models fixed and random effects in the
dispersion parameter. Assuming the panel effects are truly in the mean,
-xtnbreg- is not the model to use; -xtpoisson- is.
> As for bootstrapping and overdispersion, am I correct then that vce(boot) is
> going to control for within group serial correlation via clustering AND
> overdispersion?
I believe so. I would say that using -vce(bootstrap)- will account for
departures from the within-panel iid assumption, which includes serial
correlation and overdispersion. This all hinges on the assumption that you
are consistently modelling the mean.
--Jeff --David Drukker
[email protected] [email protected]
*
* 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/
