Propensity Score Matching for repeat treatment difference in difference


New Member

I want to make a difference in difference analysis with analysis objects that sometimes have repeat treatments over time (for example, in 2010, 2013). As I do not expect the repeat treatments to magnify the effect, I would just stack the objects according to the timing and restructure my dataset to reflect t-3, …, t0, … t+3 with some observations appearing multiple times.

My treatment sample has around 400 objects and the population has around 11,000. So I have quite some space to maneuver to find a suitable control group that ideally matches the pre-treatment development of the treatment group as close as possible. However, as I move the variables in time from fixed year (e.g., 2013) to relative timing (e.g., t+3), with some objects receiving treatments in multiple years, how would I treat these issues in Propensity Score Matching?

Would it be a problem to essentially have repeated objects in PSM at different time points? If yes, are there any solutions to it?

Thank you in advance!


New Member
The treatment group is based on applications, not random and the average profile does not match the population averages. However, the population consists of very different profiles but also has many
objects that are similar to the treatment group.

Therefore, by matching, I hope to have a control group that is as close to the treatment group before the treatments occur.