s3alfisc
commented
1 year ago Hi @aeturrell , thanks for the nice words and for always promoting the package!
As a matter of fact, the algorithm implemented in the Fepois method directly follows the algorithm sketched out in pplmhdfe - so it's already implemented and I think your colleague can start estimating his/her regression models already today =) I even (mostly) follow the naming conventions of the paper, so it should be relatively easy to map from paper to the implementation. I guess that I should note in the docs that its a 1:1 implementation of the pplmhdfe algo 😄
The only caveat is that I do not yet allow for "iid" errors (see this issue, not sure what I am doing wrong), but HC1 and CRV1 errors are well tested and match fixest output. One other difference to ppmlhdfe is that they provide a lot of options to check for separation, and fepois currently only provides the most basic check.
Hope this helps! =)
aeturrell
First, thanks for the amazing work on Pyfixest.
I have a colleague who needs to run many, many regressions using Poisson pseudo-likelihood regression with multiple levels of fixed effects (for gravity model estimation), as in this Stata package.
I'm afraid I'm not too familiar with PPML and how different it is from regular Poisson regression, so I wondered a) whether it's possible to do this in Pyfixest as is, and b) whether adding it is on the development plan?
Thank you!