Ed,
Paul Allison has argued that it is acceptable to estimate a unconditional
fixed-effects model and correct the standard errors. See "Fixed-Effects
Negative Binomial Regression Models"
(http://www.ssc.upenn.edu/~allison/FENB.pdf)
"ABSTRACT
This paper demonstrates that the conditional negative binomial model for panel
data, proposed by Hausman, Hall and Griliches (1984), is not a true
fixed-effects method. This
method-which has been implemented in both Stata and LIMDEP-does not, in fact,
control for
all stable covariates. Three alternative methods are explored. A negative
multinomial model
yields the same estimator as the conditional Poisson estimator and, hence, does
not provide any
additional leverage for dealing with overdispersion. On the other hand, a
simulation study yields
good results from applying an unconditional negative binomial regression
estimator with dummy
variables to represent the fixed effects. There is no evidence for any
incidental parameters bias
in the coefficients, and downward bias in the standard error estimates can be
easily and
effectively corrected using the deviance statistic. Finally, an approximate
conditional method is
found to perform at about the same level as the unconditional estimator."
Also, it may be sufficient to account for the unobserved heterogeneity using a
unconditional fixed-effects Poisson model. At least for the Poisson fixed
effects model, there is no incidental parameter problem (see Cameron and
Trivedi, "Regression Analysis of Count Data" page 280-282).
Scott
----- Original Message -----
From: "Ed Levitas" <[email protected]>
To: <[email protected]>
Sent: Monday, February 09, 2004 1:30 PM
Subject: st: Negative binomial regression question
> Statalisters,
>
> I have a question regarding a negative binomial regression model I want to
> estimate with panel data. I would like to estimate a conditional fixed
> effects model with robust standard errors. XTNBREG does not allow this
> option (perhaps this does not make "statistical sense"). Is there a way to
> estimate such a model with GNBREG (e.g., using GBGREG I've tried to include
> cross-section specific fixed effects in the estimation of the dispersion
> parameter via the LN(ALPHA) command. However, I don't seem to achieve
> convergence doing this.)
>
> Any suggestions would be greatly appreciated.
>
> Thanks
> Ed
>
>
>
> ****************************************
> Edward Levitas, PhD
> Assistant Professor
> School of Business Administration
> University of Wisconsin-Milwaukee
> 3202 N. Maryland Ave.
> Milwaukee, WI 53211
> ph: (414) 229-6825
> fx: (414) 229-6957
> [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/
