1.3.2 | Regression Discontinuity Design | Masters in Health Economics
Why It Matters
RDD provides health economists with a rigorous, policy‑driven method to estimate causal effects where randomized trials are infeasible, directly informing resource allocation and regulatory design.
Key Takeaways
- •Policy thresholds create natural experiments for causal health effects.
- •Sharp RD has deterministic treatment; fuzzy RD uses probabilistic compliance.
- •Continuity assumption and manipulation tests ensure validity of RDD.
- •Bandwidth choice balances bias versus variance in treatment estimates.
- •RD estimates local average treatment effect, not generalizable beyond cutoff.
Summary
The video introduces regression discontinuity design (RDD) as a quasi‑experimental tool for health‑policy evaluation, explaining how statutory cutoffs such as age‑65 Medicare eligibility or birth‑weight thresholds generate locally random assignment of treatment.
It distinguishes sharp RDD, where the probability of treatment jumps from zero to one, from fuzzy RDD, where the cutoff only raises treatment likelihood. The lecture stresses the continuity assumption—outcomes must be smooth at the threshold—and describes the McCrary density test for detecting manipulation of the running variable. Bandwidth selection is presented as a bias‑variance trade‑off, with optimal choices minimizing mean‑squared error.
Real‑world illustrations include the sharp jump in physician visits at age 65 and a landmark NICU study where infants under 1500 g receive intensive care, cutting one‑year mortality and yielding a $550,000 cost per life saved. These cases demonstrate how RDD isolates the local average treatment effect (LATE) at the cutoff.
For policymakers and researchers, RDD offers a credible alternative to randomized trials when ethical or logistical constraints exist, but its estimates apply only to individuals near the threshold. Proper validation, bandwidth tuning, and sensitivity checks are essential to translate findings into actionable health‑policy decisions.
Comments
Want to join the conversation?
Loading comments...