xuyiqing
commented
3 years ago Dear SVK,
Great question. The units are considered in the estimation process. They may contribute to the estimated dynamic ATT due to rearrangement.
The negative counterfactuals can be due to a lack of overlap in the loadings. If you like, you can send me and Ziyi an email with your sample data and code.
@. @.
Best, Yiqing
On Tue, Aug 17, 2021 at 4:46 PM svknair @.***> wrote:
Hello Yiqing,
I have a setting similar to the Example 2 (voter turnout) given here: https://yiqingxu.org/packages/gsynth/gsynth_examples.html http://url. And my question is specifically related to different treatment times for various treated units.
After estimation, I see that some rearrangement based on the timing of treatment is done to calculate the ATT. When I examine the difference between actual outcome and predicted Y(0) for individual treated units over time (from the values in out$eff.cnt), it seems all treated units are rearranged in such a way that the treatment time coincides for all treated units. So, it seems the difference between actual outcome and predicted Y(0) is not calculated for some time periods for units that are treated later. Does this mean that some of the pretreatment period observations of units that get treated later are not considered in the estimation process? The rearrangement also means that some treated units do not have an estimated treatment effect in the last few time periods. And the ATT in the last few time periods is based on some of the units that got treated first. I just want to make sure that I am reading the results correctly.
I also see that some of the estimated counterfactuals for some treated units are negative. That looks problematic to me. Could I be doing anything wrong? Appreciate your help.
Thanks, SVK
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/xuyiqing/gsynth/issues/64, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB2PKGADZXRZLEK2ZBE4FATT5LYDXANCNFSM5CK4AQJA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&utm_campaign=notification-email .
-- Yiqing Xu
Assistant Professor Department of Political Science Stanford University https://yiqingxu.org/
svknair
Hello Yiqing,
I have a setting similar to the Example 2 (voter turnout) given here: https://yiqingxu.org/packages/gsynth/gsynth_examples.html. And my question is specifically related to different treatment times for various treated units.
After estimation, I see that some rearrangement based on the timing of treatment is done to calculate the ATT. When I examine the difference between actual outcome and predicted Y(0) for individual treated units over time (from the values in out$eff.cnt), it seems all treated units are rearranged in such a way that the treatment time coincides for all treated units. So, it seems the difference between actual outcome and predicted Y(0) is not calculated for some time periods for units that are treated later. Does this mean that some of the pretreatment period observations of units that get treated later are not considered in the estimation process? The rearrangement also means that some treated units do not have an estimated treatment effect in the last few time periods. And the ATT in the last few time periods is based on some of the units that got treated first. I just want to make sure that I am reading the results correctly.
I also see that some of the estimated counterfactuals for some treated units are negative. That looks problematic to me. Could I be doing anything wrong? Appreciate your help.
Thanks, SVK