xtpoisson first difference

[Stefan Pichler]

Dear Statalisters,

I have panel data and I would like to estimate a poisson first differenced model.

Is there any command that performs this operation? For FE xtpoisson works quite nicely, but for FD I did not find anything similar.

A short explanation on why FD and Poisson: The dependent variable is autocorrelated, thus I think first difference should be preferred to fixed effects. Moreover, I have many zeros in my dependent variable and thus I am following Santos Silva and Tenreyro (2006) and estimating a Poisson model.

Thanks for any help

Best

Stefan

[Joao Santos Silva]

Hi Stefan,

Doing FD with Poisson kills one of the attracting features of Poisson regression: the exponential functional form. Differences of non-negative variables can obviously be negative and therefore the exponential model is not interesting in this case. I think I would stick to the FE Poisson because that is valid under very mild assumption.

All the best,

Joao

[Stefan Pichler]

Hi Joao,

thank you very much for your answer.

This problem came to my mind and thus I wanted to apply a quasi difference, i.e. (Li,t-1/Lit) Yit-Yi,t-1 with Lit=exp(Xit Beta), as for instance described in Microeconomics using Stata - Revised Edition (2010) by Cameron Trivedi on page 639.

But if you recommend using FE instead, I will follow your recommendation.

Best

Stefan

[Joao Santos Silva]

Dear Stefan,

I did not fully understand what you describe and I do not have the Cameron & Trivedi book at hand. Anyway, if they refer to this procedure it must be a sensible thing to do, so please consider it. Anyway, it does not look like a simple first-difference estimator, which is what I do not recommend.

All the best,

Joao

[Sher Afghan Asad]

Hi Joao,

thank you very much for your answer.

This problem came to my mind and thus I wanted to apply a quasi difference, i.e. (Li,t-1/Lit) Yit-Yi,t-1 with Lit=exp(Xit Beta), as for instance described in Microeconomics using Stata - Revised Edition (2010) by Cameron Trivedi on page 639.

But if you recommend using FE instead, I will follow your recommendation.

Best

Stefan

Hi Stefan

I am trying to implement FD as you talked about here and I have a question. How did you calculate L_it in Stata?

Shery