Dear all,
I would like to determine whether I am better off using -xtnbreg- or
-xtpoisson- to estimate a odel using time series cross section (TSCS)
data. The version 8 manuals discuss -poisgof- as a means of comparing
the two. My stata 9 says that -poisgof- has been changed to -estat
gof-. Neither works after -xt...- estimation.
Could someone suggest a way to conduct a -poisgof- type of test to help
determine whether I should be using a poisson or a negative binomial
model?
I don't have a direct answer to this question, but I think you would benefit from
referring to Chapter 9 of Cameron and Trivedi, "Regression Analysis of Count Data." On
page 280 they write: "It should be kept in mind, however, that a common reasons for such
extensions [from Poisson to Negative Binomial] in using cross-section data is to control
for unobserved heterogeneity. The longitudinal data methods already control for
heterogeneity, and Poisson longitudinal models may be sufficient."
