Autocorrelation control for nbreg

[Emily Rollins]

Hello,

I am using a negative binomial model for my time series count data, but I haven't been able to find a way to correct for autocorrelation with this model. Does anyone know of a post estimation command for autocorrelation with a negative binomial regression?

Thank you,
Emily Rollins

[Jeff Wooldridge]

If you are estimating a static model, or even a distributed lag model (lagged exogenous variables), then you might be better off using the Newey-West standard errors. Having said that, I don't think Stata allows that option for nbreg, but maybe others know.

Modeling the serial correlation is much more challenging.

[Andrew Lover]

While NW errors aren't allow for -nbreg- itself, you can run identical models using -glm- with family(nbinom) and vce(hac nwest).

[Jeff Wooldridge]

Thanks, Andrew. I should've remembered this, as I use glm often (although I didn't know it had an nbinom option). Along these lines, since the Poisson estimator is fully robust, I would try that, too, with Newey-West standard errors.

[Andrew Lover]

Hi Jeff, yes it's always funny and/or surprising what options are implemented under the stand-alone models vs -glm-. Makes for fun explorations!

[muly san]

While NW errors aren't allow for -nbreg- itself, you can run identical models using -glm- with family(nbinom) and vce(hac nwest).

Hi Andrew,

Can this solution work also with panel data?

My code is:

glm y x i.group i.year, family(poisson) vce(hac nwest 3 ) t(year)

And I get the error message:

repeated time values in sample.

I think that it is the same problem that happens when you try to run -newey- on panel data. But in this case, I cannot solve the problem using the force option.

Thanks!