Obtaining sigma squared after -xtnbreg, fe-

[Anton Ivanov]

Hello dear forum members,

Could you please suggest how I can obtain sigma squared after -xtnbreg, fe-. I want to check the assumption that sigma sq > 1 in order to use negative binomial model.

I did try -nbvargr depvar-, wihch indicated over-dispersion = 1.674, yet I am not sure this is the correct way to check the assumption.

Thankfully,
Anton

[Jeff Wooldridge]

The NegBin FE model forces there to be overdispersion. In fact, it requires overdispersion of a very special type for each cross-sectional unit. The NB FE estimator also requires serialy independence over time. I much prefer the Poisson FE estimator, which is fully robust (to any kind of overdispersion and serial correlation).

In any case, the assumption cannot be "tested" because it is imposed.

[Anton Ivanov]

Thank you sincerely for reply, Professor Wooldridge.

Following your prior advise I am using Poisson, FE estimator as a main one. Yet, following your elaborations in the "Econometric Analysis of CS and Panel Data" I wanted to run some robustness tests with NegBin I. Additionally, serial correlation is present in the data and to check the robustness of the estimates I run an AR(1) model with a log-transformed dv.

Respectfully,
Anton