Math for Causal Inference
All simulators

Regression Kink Design

A capped benefit schedule: dividing the outcome's slope change by the schedule's slope change.

Day 78: Regression Kink & RD Extensions
1.41.61.822.2-1-0.500.51earnings (centred at the cap)benefit123-1-0.500.51earnings (centred at the cap)outcome (binned means)
outcome slope change
-0.614
RKD estimate of β
1.23
truth 0.80

Unemployment benefits replace 50% of past earnings up to a cap (the kink at 0, left panel). The outcome's slope changes at the kink by β × (−0.5); dividing the slope change by the schedule's slope change estimates β. Slopes are estimated with local polynomials on each side; curvature in the outcome biases local linear slopes, and local quadratic fits are much noisier.