Count data - Negative Binomial regression (QML)

[thesis]

Hi!

Since my count data is overdispersed I used the negative binomial regression method.( With a GLM estimator) For a robustness check I want to split my data in 2 parts and run the regression again, however in that case, 36 observations are included, is this sample large enough for negative binomial regression or do I need to use another method and if so what method?

Thank you!

[EViews Glenn]

What is large enough is a question that doesn't have a single answer, but 36 obs is probably pretty small...

[thesis]

Ok, thanks. Is another method more suitable?

[EViews Glenn]

That's a very general question for which I don't think there is a right answer. Perhaps others would care to join in.