Dear Stata users,
I am estimating count data models with instrumental variables using -qvf-
(Hardin et al., 2003). I'm trying to choose between poisson and negative
binomial. Actually, I'm pretty sure there is overdispersion (hence I'll use
a negative binomial), but I would like to report on this. There are various
ways to test for overdispersion. The LR test uses the log likelihood, while
a Wald test uses the reported t-statistic for the estimated alpha (Cameron &
Trivedi, 1998). But -qvf- does not use maximum likelihood and hence does not
report the log likelihood. It also does not seem to report the estimated
alpha or its t-statistic. Is there a way to test for overdispersion
in -qvf-? I don't suppose I can cheat and do the 2-stages by hand (-reg-
followed by -nbreg-) just to get a likelihood statistic?
Your advice would be appreciated.
Thanks,
Peter.
References:
Cameron & Trivedi (1998) Regression Analysis of Count Data, Cambridge.
Hardin, Schmiediche & Carroll (2003) 'Instrumental variables, bootstrapping,
and Generalized Linear Models' Stata Journal, 3(4): 351�360.
Peter Siminski
PhD Student
School of Economics / Social Policy Research Centre (SPRC)
University of New South Wales
Ph: 0425223257
[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/
