xuyiqing
commented
10 months ago T = -1 is two periods before the treatment's onset. T = 0 is the last period before the treatment kicks in, different from BJS's notation. The way the imputation/counterfactual estimator works means that we use the average of all pretreatment periods as the benchmark. Hope this helps.
On Fri, Oct 6, 2023 at 6:24 AM zhizhongpu @.***> wrote:
In most existing ES estimators, period-specific coefficients are normalized such that the coefficient at T=-1 equals 0. However, this seems not the case using your method (see figure). My collaborator and I are not sure about how to interpret this non-0 coefficient.
I am guessing that the coefficient reflects the absolute gap between treated units' factual and imputed counterfactual, and thus one can't just shift the coefficients so that $\delta_{T=-1}$ equals 0. My collaborator is instead guessing, however, based on Borusyak Jaravel and Spiess (2021), that you still use some pre-treatment period to normalize all coefficients.
Could you explain the non-0 $\delta_{T=-1}$?
[image: image] https://user-images.githubusercontent.com/84325421/273219070-b6507bd1-e82c-4b94-a07c-cb1ce3593eff.png
— Reply to this email directly, view it on GitHub https://github.com/xuyiqing/fect/issues/33, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB2PKGFLIBDOWEU57ITIG4TX6ABAJAVCNFSM6AAAAAA5V2L7PWVHI2DSMVQWIX3LMV43ASLTON2WKOZRHEZTAMJYGEZDCOI . You are receiving this because you are subscribed to this thread.Message ID: @.***>
-- Yiqing Xu
Assistant Professor Department of Political Science Stanford University https://yiqingxu.org/
In most existing ES estimators, period-specific coefficients are normalized such that the coefficient at T=-1 equals 0. However, this seems not the case using your method (see figure). My collaborator and I are not sure about how to interpret this non-0 coefficient.
I am guessing that the coefficient reflects the absolute gap between treated units' factual and imputed counterfactual, and thus one can't just shift the coefficients so that $\delta_{T=-1}$ equals 0. My collaborator is instead guessing, however, based on Borusyak Jaravel and Spiess (2021), that you still use some pre-treatment period to normalize all coefficients.
Could you explain the non-0 $\delta_{T=-1}$?