Regression Discontinuity

Regression discontinuity (RD) exploits a sharp cutoff rule that decides who gets a treatment — a test score, an age, an income threshold. Units just below and just above the cutoff are nearly identical except that one side got treated, so comparing them estimates the causal effect right at the boundary.

The logic is that around the threshold, which side you land on is almost as good as random. Its strength is credibility with weak assumptions; its limit is generalizability — the estimate only speaks to units near the cutoff, and it requires that nobody can precisely manipulate their position to jump the line.