Robust standard errors for Panel Count Data

[Marianna Sebo]

Hi Statalisters,

I have panel count data and it appears to me that it has rather a negative binomial distribution and not poisson. To make regression I was using nbreg and glm both with vce (robust) and time effects. My question is why is not there a way to use xtnbreg with robust standard errors? Or should I go for the xtpoisson? Since I have panel data it seems to me that the xt estimations are those I should use.

[Andrew Musau]

See Jeff Wooldridge's detailed list on why it is almost never a good idea to estimate a FE negative binomial model: https://www.statalist.org/forums/for...-poisson-model (#3).

[Marianna Sebo]

Thanks for your answer. I forgot to mention that my variable of interest is a dummy I wouldnt use FE but RE.

[Jeff Wooldridge]

Marianna: That your variable of interest is binary does not preclude you from use FE. The key is whether it changes sufficiently across time for at least some units.

[Marianna Sebo]

Hi Jeff, no, it does not at all. So when I run a FE in Stata it doesnt even appear. Therefor I am not so sure how to proceed. Thanks!!

[Jeff Wooldridge]

Marianna: Do you have control variables that change over time? Other than time period dummy variables?

[Marianna Sebo]

Jeff Wooldridge I do, so overall I have monthly data for 4 years for 73 neighborhoods that belong to 4 districts. Apart from my variable of interest that doesnt change I have control variables that change over years like population, income leve ..etc. Thank you for your help!!