Counterfactual treatment

[Mansour Mohseni]

Hi

I am working with a time-varying, staggered treatment setting in which different firms receive the treatment at different points in time, alongside a set of comparable firms that remain untreated. In addition to estimating the average treatment effect, I am interested in conducting counterfactual analysis, specifically, I would like to estimate what would have happened if the treatment had occurred earlier.

Is there a methodological framework that allows for this type of “what-if” analysis in staggered treatment settings?

[George Ford]

Is that not what the pre-treatment options tell you?

[Mansour Mohseni]

Is that not what the pre-treatment options tell you?

Dear Dr. Ford
To be honest, I am not sure whether pre-treatment options could act as a counterfactual; in my understanding, it is more like a difference level before actual treatment. I was thinking more from a structural perspective, in a similar way that we can estimate counterfactuals there.

[George Ford]

The counterfactual is constructed to produce the ATE. What you are interested in is a pseudo-treatment effect. As I understand it, the pre-treatment differences (1 period prior) is what you are after. I suppose you could check this by reducing the date used as the treatment by one period. Depending on the algorithm, it may be the same or different than the -1 effect.