Re: st: RE: poisson - bootstrapping or clustering?

[Scott Cunningham <scunning@gmail.com>]

Carter,

My main reason for sticking with the Poisson for the time being is  
that the negative binomial model is less robust to distributional  
misspecification than the POisson.  The Poisson is consistent under  
the weak assumption that I've correctly the conditional mean - even  
if the data is not Poisson distributed.  But from my understanding of  
Cameron and Trivedi (2005), chapter 20.2 and 20.4 (p. 666-677), the  
negative binomial will be inconsistent even if the conditional mean  
is correctly specified (except for a special case of the NB2 model).

Also, I'm using a fixed effects, and my understanding again is that  
the Poisson fixed effects has desirable characteristics - like not  
suffering from incidental parameter bias and not needing to specify  
any specific distribution for the unobserved random variable.   
Whereas with the negative binomial, I think I have to buy off on  
certain assumptions about the unobservable - that it is gamma  
distributed.  I'm not an expert on count data and count models, though.

Insofar as I can correct for the overdispersion problem within the  
poisson mle, I wanted to do so.  Historically, I've used  
bootstrapping, but it's been a while since I ran those models, and I  
can't figure out why now the model fails so often.  I'm using  
numerous indicator variables - a six year panel with state fixed  
effects (51 total when you include District of Columbia), year fixed  
effects (6 years of data), a race and race interacted with a  
continuous variable.  There are 2282 observations per year in the  
sample - a total of 13692 observations.  I don't know if this is what  
is causing the problem or not.  When I've successfully used  
bootstrapping to correct for overdispersion, I was not using a race  
dummy, plus only 700 or so observations with a shorter panel.  Maybe  
that is part of the problem.


Richard,

I will experiment with different seeds as you suggest.


scott

On Sep 12, 2006, at 8:20 AM, Carter Rees wrote:


> I missed the previous post you are referring to but why not use - 
> xtnbreg- if
> you know you have overdispersion in your data?
>
> Carter Rees
> School of Criminal Justice
> University at Albany, SUNY
> carterrees@gmail.com
>
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Scott  
> Cunningham
> Sent: Tuesday, September 12, 2006 8:13 AM
> To: statalist@hsphsun2.harvard.edu
> Subject: st: poisson - bootstrapping or clustering?
>
> I'm estimating a model of sexual partners using -xtpoisson- and -
> poisson-.  The data suffers from overdispersion, and so I'm trying to
> correct for that using bootstrapping within -xtpoisson-.  But as I
> posted the other day, I'm having trouble recovering the marginal
> effects in post-estimation.  I have a memory of someone telling me
> that the cluster() option within -poisson- can correct for
> overdispersion.  Does anyone with experience in count data have
> recommendations?  This is micro-level data from the National
> Longitudinal Survey of Youth (1997).  The problem with -mfx, dydx-
> appears to be that bootstrapping within -xtpoisson- had numerous
> failures in calculating the standard errors.  That's at least what I
> think is going wrong.
> *
> *   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/
>
> *
> *   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/
*
*   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/