Jan Kabatek ๐๐ @jankabatek.com ยท Dec 3 Note that the relative performance depends *๐๐ซ๐ข๐ญ๐ข๐๐๐ฅ๐ฅ๐ฒ* on the size of your dataset (see the pictures below). ๐ ๐ซ๐๐ ๐ก๐๐๐๐ฃ๐ฅ performs particularly well in big datasets (10M+), but badly in small ones. (FWIW, I don't know why the factorized regressions are subject to that structural break ๐)
0 likes 2 replies
Replies
3 ) Merging does not have to take forever! Be aware that the 'merge' command sorts both 'master' and 'using' data on matching variables. ๐ you can save crazy amounts of time by running the merge command on datasets that are already sorted! ๐ make sure to store your raw & intermediate data sorted!
Seems crazy to ask you to do more work (this is incredible!) but can you test out some large number of FEs? Like 20million? Would mimic something like a household fixed effect regression using full count census