RE: st: Confusion specifications GLM

["Verkuilen, Jay" <JVerkuilen@gc.cuny.edu>]

Peter F. Schmidt wrote:

>>thanks for the hint. that seems to the right model. i tried to use the
ztnb but stata gives the error message that convergence could not be
achieved. on the ucla stata page is a warning that the method might be
problematic with a small sample, but it is unclear what is a small
sample. i have 180 observations. do you have any idea how to get the
regression running?<<

Run a simplified version of your model with no or only a few independent
variables. Then *slowly* add variables and see where/when things break.
You could easily have multicollinearity problems. You may need to give
the model good starting values taken from another procedure. Stata does
its darndest to give you good starting values but that may not be
enough.

Maarten also suggested using ztp rather than ztnb. The overdispersion
may be sufficiently accounted for by the regressors so that each
observation is conditionally truncated Poisson. What the *marginal*
distribution is can vary greatly; it does not have to be Poisson for
Poisson regression to be appropriate and ignoring regressors generally
induces overdispersion. 

You may want to check out J. S. Long and J. Freese (2005). Regression
with Categorical and Limited Dependent Variables Using Stata. Stata
Press. 

Jay




*
*   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/