search

grantmcdermott / etwfe

Extended two-way fixed effects
https://grantmcdermott.com/etwfe/
Other
51 stars 11 forks source link

Adding max period over which att is to be calculated #46

Open fhollenbach opened 7 months ago

fhollenbach commented 7 months ago

I added a simple option to only include t post-treatment periods in the calculation of aggregate effects (this option also exists in did). I have tested it a little but not 100% yet. Wanted to provide option to add this to general package.

grantmcdermott commented 3 weeks ago

Thanks @fhollenbach and super sorry for the slow response. I had to ignore this repo for a while and dropped the ball on your PR.

I'm not adverse to adding extra arguments, but I also think it's worth weighing the cost of extra complexity from the user side. Is the main benefit here that this leads to quicker emfx computation times (since there's less data to cover)? Or, is it mostly about being more convenient than subsetting the emfx object after it's been created?

fhollenbach commented 1 week ago

Hi @grantmcdermott, now I have to apologize for the slow reply. The idea was that sometimes we might want to include all the data to estimate effects, but it may only make sense to calculate an average effect over t post-treatment periods (e.g., 12 month). At the moment you could use emfx( type = "event") and then subset for that period to get the eventtime ATTs for the t-periods. But one would then have to calculate the overall average from that (including the SE). My idea was to be able to restrict the post-treatment time period in the emfx() command so that one could get the overall average effect calculated only over t-post-treatment periods. Does that seem reasonable or did I misunderstand something?