search

sergiocorreia / reghdfe

Linear, IV and GMM Regressions With Any Number of Fixed Effects
http://scorreia.com/software/reghdfe/
MIT License
214 stars 56 forks source link

reghdfe postestimation influence #195

Open laszlosandor opened 4 years ago

laszlosandor commented 4 years ago

I was wondering how much of this could work with reghdfe: https://www.stata.com/features/overview/linear-regression-and-influence/

Currently (v 5.7.2) dfbeta surely does not.

FWIW, I was hoping to learn where my identification is coming from. Not just on the conceptual level but observation-by-observation.

sergiocorreia commented 4 years ago

Dfbeta appears to require one regression per observation (kinda like leave one out regression). That's a bit insane from a computational perspective and for the usual dataset size of reghdfe, so I don't think it's practical.

There might be a few ways to speed up the leave-one-out computations, but they would involve a lot of code.

laszlosandor commented 4 years ago

Yeah, I'm not sure what I was hoping for. They surely have no other general-purpose solution other than looping through all observations and if them out one by one.

OK, other than that, is there a reasonable way to detect which observations influence reghdfe estimates most? If not for each coefficient separately, at least for overall leverage or something?