Dear Statalist users,


I am trying to combine a difference-in-difference design with kernel propensity score matching using the user-written command diff from Villa(2016) - The Stata Journal (2016) 16, Number 1, pp. 52–71

In terms of the dataset, I have two types of firms: they all start in stage A, but after meeting certain requirements they can be promoted to stage B with additional tax benefits.
The research question is whether the transition implies positive or negative impact on efficiency.

I have an unbalanced panel dataset, where I observe firms that never experienced promotion (control group) and firms that stay 2 or 3 years in stage A and then move to stage B where they can be observed for 1 or 2 years (treated group).

Since promotion from A to B can occur in different years, I'm struggling to define what exactly should I code as treated units and the multi period (before and after).

Is the command diff suitable for this, or do you suggest a different approach in order to get semi-parametric matching with fixed effects?


Thank you in advance