Hello Statalisters,
I have a question regarding the use of XTPOIS and XTNBREG. Using panel
data, I've estimated a fixed effect Poisson model (with XTPOIS), and a fixed
effect negative binomial Poisson (using XTNBREG). The standard errors of
the coefficients vary across the models (which, if my understanding is
correct, is to be expected given the modeling of a dispersion parameter in
XTBREG). However, the coefficients across my models vary both in magnitude
and in direction (i.e., coefficients for some variables "switch" signs
across the two types of estimations) as well. If my understanding is
correct, such coefficient variation should not be witnessed in the case of
purely cross-sectional data if the negative bionomial effectively models any
overdispersion (per Long 1997, the expected value of the DV for the negative
binomial distribution is the same as it is for the Poisson distribution).
However, in the panel data setting, is such coefficient variation more
"legitimate"? For example, might this be due to the fact that fixed effects
in the Poisson model are modeled as regressors whereas fixed effects in the
Negative Binomial model are modeled in the dispersion term?
Any help/clarification/citations would be extremely appreciated.
Thanks in advance.
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/
