Glossary Entry

Regression Discontinuity Design

A design that reads the causal effect off the outcome's jump at a threshold rule, since units just above and below the cutoff are effectively randomized.

Statistics Decision Making

Also called: regression discontinuity, RDD, running variable

Seed source: Causal Inference - The Mixtape, ch. 6

When a rule assigns treatment by thresholding a continuous running variable (spend over 500 earns gold status, a score above the bar approves the loan), units immediately on either side of the cutoff are comparable in everything except treatment. Under the continuity assumption, that nothing else jumps at the threshold, the discontinuity in the outcome is the causal effect at the cutoff.

Estimation is local: fit regressions on each side within a bandwidth and difference them at the threshold, preferring local linear fits over high-order global polynomials. The design dies if units can manipulate the running variable to clear the bar, which the McCrary density test checks for, and its estimate is honest only for units near the cutoff.