Re: Re: st: Hausman test after xtnbreg and alternatives?

["JVerkuilen (Gmail)" <jvverkuilen@gmail.com>]

On Tue, Feb 12, 2013 at 9:05 AM, Christopher Baum <kit.baum@bc.edu> wrote:
> <>

> Note that in xtnbreg, the fixed vs random effect does not enter the 'xb' part of the model, as it does in linear
> regression, but refers to the overdispersion parameter varying across panel units.

I thought there were two sources, negative binomial overdispersion and
the usual panel structure inside the link that would be a random or
fixed intercept shift analogous to the term in, say, xtpoisson? From
checking the manual I see that you're 100% correct. You learn
something every day....


 Whether there is any way of testing the additional
> overidentifying restrictions a la -xtoverid- is not clear to me.

I'd suggest AIC or something like that but I'm really not convinced
those are comparable at all. I have to say probably the best thing is
to focus on the component one wants to study and run two parallel
analyses, hoping that the conclusions are essentially invariant.
Basically this is a sensitivity to assumptions analysis. It's not
ideal, but it works.
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/