[R] Seeking Advice on Diff-in-Diff Model When Treatment Effect Disappears After Intervention

[David Studer]

Hi everyone,
I�m working on a Difference-in-Differences (Diff-in-Diff) analysis at the school district level to study the impact of switching from a dual college admission algorithm system (Immediate Acceptance (IA) and Deferred Acceptance (DA)) to a uniform DA system implemented in England after 2007.
Usually, Diff-in-Diff models are used when an intervention creates a difference between treated and control groups after the policy change. However, in my case, the opposite is expected:

  *
Before 2007, districts using IA had different rates of successful appeals against college admissions compared to those using DA.
  *
After 2007, with DA applied everywhere, these differences should disappear or converge.

This means that the "treatment effect" I�m estimating is actually a reduction or elimination of pre-existing differences, rather than the emergence of new differences after the intervention.
Has anyone encountered this reversed Diff-in-Diff setting before? How did you model or interpret the interaction term when the expected effect is convergence rather than divergence? Are there any specific methods, robustness checks, or papers you could recommend for such a scenario?
Thanks in advance!


	[[alternative HTML version deleted]]