Standard errors and P values not calculated and very low number of observations in xtnbreg, fe

[Adnan Habib]

Hi,
I am running a xtnbreg, fe regression where the dependent variable has a lot of zero values. I read somewhere that just having lots of zeroes does not necessarily mean I have to use zinb. And since I needed the fixed effect and zinb does not allow fixed effect so I used xtnbreg. However in the result a lot of the regressors do not have any standard errors or p-values. Also the number of observations is very low at 798 only, whereas when I look at the summary statistics of all the variables there are none with number of observations lower than 63,000. How can I get standard errors and p-values for all my regressors?

[Joao Santos Silva]

Dear Adnan Habib,

My suggestion is that you avoid xtnbreg fe and use xtpoisson fe, which estimates a proper fe model and is much more robust.

Best wishes

Joao

[Adnan Habib]

Dear Joao Santos Silva

Thanks for the suggestion. I know that the requirement for using xtpoisson is that the mean and variance of the data are equal. But in the data I am using the variance is higher than the mean, hence I chose to use xtnbreg.

[Joao Santos Silva]

Dear Adnan Habib,

I am afraid that is not correct. You would need the conditional variance to be equal to the conditional mean if you wanted to compute probabilities, but:

a) if you are using fixed effects, you cannot compute probabilities, so I assume that all you want to do is to estimate parameters of the conditional mean;

b) because the equality is between the conditional mean and variance, you cannot know if the equality holds without estimating the model (including the fixed effects), so at this stage you do not know if the equality holds.

Best wishes,

Joao