Staggered difference-in-differences

[Waterdesk]

I would like to do a difference-in-differences analysis to compare units who received a treatment vs units who did not receive treatment. I found a good article that described how to do difference-in-differences in SAS; article can be found here

The problem I have is that each unit received treatment at different dates; therefore, I have a staggered treatment design. The article describes how to do difference-in-differences if all units receive treatment at the same time, but I can't find how to run such an analysis when the treatment is staggered. I want to point out that creating the model in SAS isn't much of an issue for me. What I need assistance in is how to incorporate some of the newer methods that are described in the following article: https://cran.r-project.org/web/packages/did/vignettes/did-basics.html. 

TLDR: Is there a guideline on how to run staggered DiD in SAS that uses the newer methods that were proposed by Callaway and Sant'Anna?

[PaigeMiller]

Can't you just set the start date for each unit to zero, and then work with the number of elapsed days?

[Waterdesk]

Yes, I can do that but I believe that the later treated units have a different treatment effect than the earlier treated units; there are heterogenous treatment effects.

[SteveDenham]

Perhaps you could use the EFFECT statement to fit a spline to the start dates to deal with the suspected change in the treatment effect, and use @PaigeMiller 's suggestion of fitting elapsed days as the independent variable. You would probably need an interaction term. This would fit a response surface for your DiD data. Does that make any sense?

SteveDenham

[PaigeMiller]

---

@SteveDenham wrote:

Perhaps you could use the EFFECT statement to fit a spline to the start dates to deal with the suspected change in the treatment effect, and use @PaigeMiller 's suggestion of fitting elapsed days as the independent variable. You would probably need an interaction term. This would fit a response surface for your DiD data. Does that make any sense?

SteveDenham

---

@SteveDenham I was thinking a similar thing, somehow add the effect of the treatment by actual calendar date into the model somehow. But you beat me to it!

[StatDave]

It doesn't really matter that the treatment isn't applied at the same time for each subject. The model just needs an indicator of pre- vs. post-treatment, whenever it actually occurs. See the repeated measures sections of this note.