search

kylebutts / did2s

Two-stage Difference-in-Differences package following Gardner (2021)
http://kylebutts.github.io/did2s
Other
96 stars 22 forks source link

Event Study Plot with not-yet-treated #13

Closed frederickluser closed 2 years ago

frederickluser commented 2 years ago

Hey Kyle,

thanks for your great work. Very helpful. A quick question: Will it be possible in the nearer future to use the event-study function with only not-yet-treated units? That would be very helpful.

Best, Frederic

kylebutts commented 2 years ago

Hi Frederick -- I thought about doing this, but some of these estimators do not work with not-yet-treated units (e.g. Sun and Abraham) and others require choices about estimation. Therefore, I'm not comfortable setting defaults for these. I could be persuaded to, but at the moment there's no option in the near future planned

frederickluser commented 2 years ago

I see, thanks for the feedback.

frederickluser commented 2 years ago

Dear Kyle, can I ask a quick follow-up question? I'm confused, which estimators can be used with not-yet-treated units only. I'm asking because the fact that the code runs and produces results, doesn't necessarily mean that doing so is a valid approach, I guess… For sure, Roth and Sant'Anna as well as Callaway and Sant'Anna should work properly, right? Gardner and Borusyak, Jaravel, and Spiess should also be valid, right?

Thanks for any input.

kylebutts commented 2 years ago

Callaway and Sant'Anna is the only package that allows this explicitly. If you run an estimator on a sample where everyone eventually gets treated, it depends on the code what happens.

For example, in {did2s}, it would drop the time periods where all units are treated (since these periods would not have time fixed-effects estimated). If you have never-treated units in your sample, those will be included, so you're better off manually dropping them.

frederickluser commented 2 years ago

Thanks!