I honestly have no immediate opinion on which FE to use. This is my issue with economists more generally, they tend to use FE of all kinds as this catch all aspect of estimation. You could have firm, county, state, year fixed effects, but that doesn't mean it's a good idea to do so.
This is why I suggested SCM, which actually (when used correctly) accounts for unobserved confounding, given a set of covariates, unlike DD. It takes away the need for "do i need fixed effects or not" and focuses, in my opinion, on more important issues of design.
I don't know how to answer your last question since you didn't provide any data or say what the problem was. What do you mean by
? I'm also confused by
Why not just do
This is why I suggested SCM, which actually (when used correctly) accounts for unobserved confounding, given a set of covariates, unlike DD. It takes away the need for "do i need fixed effects or not" and focuses, in my opinion, on more important issues of design.
I don't know how to answer your last question since you didn't provide any data or say what the problem was. What do you mean by
doesn't seem to be exactly correct
Code:
|(treated==0)
Code:
qui reg illness b22233.date##treated, robust estimates store Untreated
Comment