Negative Binomial Regression model with respect to fixed and random effect models

[Gilles De Groot]

Hi Statalists,

I am currently writing my thesis about the relationship between terrorism and remittances. Terrorism is the dependent variable and measured as the amount of kills per year per country and is thus a count variable. Previous papers concluded that I have to use a standard Negative Binomial regression model but the I am puzzled about its use: The Hausman test states that I have to utilize the fixed effects model. My questions yield:
1. Should use the negative binomial regression model and the fixed effects model seperately or should I combine the two in one model?
2. What is the relationship between fixed and random effects models w.r.t. a binomial regression model?

[Joao Santos Silva]

Dear Gilles De Groot,

My general advice in these cases is to recommend Poisson with fixed effects because it is the only reasonably robust approach. If you search this forum, you will find several posts related to this including a reference to justify this approach.

Best wishes,

Joao