PPML cross section data and for not gravity model

[FernandaM Almeida]

Dear,

i) Would be adequate to use ppml estimation for regressions that aren't gravity models?
ii) And for cross section data?
My model is cross section with cound data. There is unobserved heterogeneity and a large number of zeros.

Thanks.

[Andrew Musau]

ppml is from SSC (FAQ Advice #12).

i) Would be adequate to use ppml estimation for regressions that aren't gravity models?
ii) And for cross section data?

From the documentation, you have that

ppml estimates Poisson regression by pseudo maximum likelihood. It differs from [R]
poisson because it uses the method of Santos Silva and Tenreyro (2010) to identify and
drop regressors that may cause the non-existence of the (pseudo) maximum likelihood
estimates.

So you may use the command same as poisson.

My model is cross section with count data. There is unobserved heterogeneity and a large number of zeros.

With cross-sectional data, all you can do is to find control variables that will help explain the unobserved heterogeneity. Only with panel data can you control for it. The estimator plays no role here.

[FernandaM Almeida]

Thank you very much, Andrew Musau.

I tried the poisson and nbreg options, both with independent covariates and control dummies. But, the model isn't converging. My dataset has 87% of zeros.

[Joao Santos Silva]

Dear FernandaM Almeida,

Please try ppml and do let me now if it does not converge.

Best wishes,

Joao

[FernandaM Almeida]

Dear Joao Santos Silva ,

Thanks for your response. Yes, definitely it converged. Does your PPML appropriate for my case?

All the best,

Fernanda.

[Joao Santos Silva]

Dear FernandaM Almeida,

The ppml command just estimates a Poisson regression, so it is likely to be appropriate for your case, unless you want to compute probabilities of certain events. The command also ensures that the estimates exist; please check the help file for the relevant references.

Best wishes,

Joao

[FernandaM Almeida]

Dear Joao Santos Silva, great!
My interest is just in the signals of the relations between the variables.
I really appreciate your help.

All the best, Fernanda.