Regression Discontinuity Design (RDD)

Regression discontinuity design (RDD) is a quasi-experimental method that estimates a causal effect by comparing outcomes just above and just below a cutoff on a running variable that determines treatment.

회귀불연속설계(RDD)는 처치를 결정하는 배정변수의 기준점 바로 위와 아래의 결과를 비교하여 인과효과를 추정하는 준실험 방법이다.

How do I estimate a causal effect when treatment is assigned by a score cutoff?

점수 기준선으로 처치가 배정될 때 인과효과를 어떻게 추정하나요?

Regression discontinuity design compares units just on either side of a threshold, such as a test score or age limit, where treatment assignment changes abruptly. It fits settings with a known cutoff rule, and in the fuzzy version the cutoff changes the probability of treatment rather than determining it fully.

When to use Regression Discontinuity Design (RDD)

• Eligibility determined by a cutoff on a test score, income, age or vote share
• Many observations near the threshold
• Evaluating programs where randomization was not possible

When to use something else

• Treatment starts at a point in time rather than at a score threshold: use interrupted time series
• No assignment rule exists but a valid instrument does: use two-stage least squares

Assumptions and requirements

• Units cannot precisely manipulate the running variable around the cutoff
• Other covariates are continuous at the cutoff
• Correct functional form within the bandwidth (local linear fitting)
• Sufficient data near the threshold

What the analysis reports

• Local treatment effect at the cutoff with robust confidence interval
• Optimal bandwidth
• RD plot with binned means
• Density (manipulation) test
• Covariate balance and placebo cutoff tests

How to report it

Using local linear regression with a bandwidth of [value], crossing the cutoff increased [outcome] by [value] (robust 95% CI [[lower], [upper]], p = [value]).

Related methods

• Interrupted Time Series (ITS)
• Two Stage Least Squares (2SLS)
• Difference in Differences (DiD)

Run Regression Discontinuity Design (RDD) on Quorant