search

lsun20 / EventStudyInteract

MIT License
34 stars 12 forks source link

Issue of plot interpretation in Roth 2024 #11

Closed Alalalalaki closed 5 months ago

Alalalalaki commented 5 months ago

Hi,

I recently learned from Roth 2024 that the default plots by recent methods (de Chaisemartin and D’Haultfuille, 2020; Callaway and SantAnna, 2021; Borusyak, Jaravel and Spiess, 2024) do not match those of traditional (TWFE) event-studies. And it is suggested adding long2 when using CS/dCDH packages.

I just wonder that is the estimator here also subject to similar issues? And does the default setting of EventStudyInteract provide event study plots comparable to TWFE?

Thanks!

lsun20 commented 5 months ago

Thank you for sharing the note by Roth (2024)! This estimator collapses to the TWFE in the non-staggered case, which was illustrated in the replication code of Roth (2024). For the Stata version eventstudyinteract, you need to manually specify relative time indicators as you would have included in the TWFE, and usually that means leaving out the -1 relative period in the non-staggered case.