Question about covariates

[mingyuqi0914]

Hi Liyang,

Sorry to bother you again! I found that the estimates I generated using eventstudyinteract have little difference with or without adjusting for covariates. If I took out covariates and always treated units, the estimates I got were close to what I got from CSDID. If I added in covariates, the estimates obtained from CSDID changed a lot in terms of magnitude, but the ones obtained from eventstudyinteract only had minor changes. I don't know if this is just a single case, or if anyone has brought this to you. I wonder if you could give me some advices on this when you have a free moment.

Thanks,
Mingyu

[lsun20]

Dear Mingyu,

eventstudyinteract controls for covariates linearly, which would be correct specification if the counterfactual trends are linear in the covariates. To my knowledge, CSDID from Callaway and Sant’Anna (2021) controls for covariates flexibly and is not restricted to linearity.

Hope the above helps,
Liyang

[mingyuqi0914]

Thank you Liyang!