DiD placebo test

[Adrienne Wold]

I am a running a two-period DiD, with several periods of pre-treatment data and one period of post-treatment. I'd like to understand the correct way to run a placebo test where I set my treatment period to t-1 instead of t. When I run the DiD with the placebo treatment, do I still include the observations from the actual post-treatment period? My results are different when I include vs exclude these observations (significant effects of the fake treatment when including the true post-period, insignificant when excluding it).

[Clyde Schechter]

Yes, when you run the "placebo" test, you still include the observations from the actual post-treatment period. The idea of doing a placebo test is to demonstrate that your supposed treatment effect is actually associated with the treatment and is not something you might find using any arbitrary time distinguishing the pre- and post- treatment periods. So the placebo test is an analysis done exactly as you would do it if the t-1 cutoff for the start of treatment really were the real time when treatment started.

[Carlo Lazzaro]

Adrienne:
as as aside to Clyde's helpful reply, you may want to take a look at:
1) https://mixtape.scunning.com/09-diff...n_differences;
2 Comparative Politics and the Synthetic Control Method. Pages 504-505 describes in-time-placebo studies. Even if the paper focuses on SCM, the aim if in-time-placebo studies holds for DID too.