Count dependent variable with many zero values

[Sang-Bum Park]

Hi

My dependent variable is the number of policies that a company has adopted in the given year. It ranges from 0 to 20 (maximum). In my sample, there are a lot of companies that have never adopted a related policy. Data is structured as an unbalanced panel. Independent variables are mostly continuous but some are dummies.

I am considering a negative binomial regression (using xtnbreg command) or a zero-inflated model. Is it not good to use xtnbreg , fe? Is it better to use xtnbreg , pa? What is the right command to run a zero-inflated model for panel data?

[Joao Santos Silva]

Dear Sang-Bum Park,

Is 20 the maximum in the sample or the maximum the variable can take? Anyway, I would start with a Poisson regression, possibly with fixed effects because it is much more robust than any of the approaches you are considering.

Best wishes,

Joao

[Sang-Bum Park]

Thank you, Joao Santos Silva for your answer. I will use the xtpoisson command with a specific model option.

[Sang-Bum Park]

Dear Joao Santos Silva,

Could you advise me on how I can get the results of Poisson regressions faster? I ran a Poisson regression by using the xtpoisson command and the fe option. But, it is running for more than twenty minutes and still shows Iteration messages like below.

[Joao Santos Silva]

Please use the user-written ppmlhdfe command; check the help file before using it to familiarize yourself with the syntax.

[Sang-Bum Park]

Joao Santos Silva Thank you. Very helpful.