Negative Binomial Regression and FE with unbalanced panel data

[Anna Belli]

Dear all,

I am running an analysis on conflicts and in-migration. I have an unbalanced panel dataset and a dispersed count dependent variable thus I was thinking to use a negative binomial regression. I would like to include in my model a location fixed effect. I decided to include a regional fixed effect (the largest spatial unit) but I wanted also to run the regression using district fe (my panel is indeed a district-level panel).
I was wondering how to include these fixed effects. Should I use the xtnbreg command or is it better to use dummy variables?

I hope you may help me with these doubts.

Thanks,
Anna

[Joao Santos Silva]

Dear Anna Belli,

My advice is that you should simply use Poisson regression with FE and avoid the NB regression with FE. There are several posts on this forum about this, so please search for those to see more details.

Best wishes,

Joao

[Anna Belli]

Thank you so much for your answer, I will try to use this approach!

All the best,

Anna

[Jeff Wooldridge]

Anna: Be sure to use the vce(robust) option with xtpoisson, fe in order to obtain valid standard errors.