DataTerminology.comNEW
Also known as: RD Design
Estimating causal effects at a threshold-based assignment rule.
DESCRIPTION
Regression Discontinuity (RD) exploits a cutoff in an assignment variable to estimate local treatment effects. Units just above and below the threshold are assumed comparable, allowing causal inference under continuity assumptions. Sharp RD assigns deterministically at the cutoff; fuzzy RD assigns probabilistically and uses instrumental-variable techniques. RD is widely used in policy evaluation, education, and economics when randomization is infeasible. Proper design requires careful bandwidth choice, functional form checks, and robustness analyses.
ALTERNATIVE DEFINITIONS
Wikipedia
The regression discontinuity design is a quasi-experimental pretest-posttest design that assigns a cutoff or threshold above or below which an intervention is assigned.
Emphasizes threshold-based assignment in quasi-experimental settings.
NBER (Imbens & Lemieux)
Regression discontinuity designs exploit a known rule that assigns treatment based on a running variable crossing a threshold, allowing credible causal inference near the cutoff.
Highlights credibility and local causal interpretation near the threshold.
Stata
The regression-discontinuity design (RD) estimates treatment effects by comparing outcomes for units with running-variable values just below and just above a known cutoff.
Focuses on local comparison around the cutoff and estimation tooling.
RELATED TERMS
Instrumental Variable Causal Inference Running Variable Sharp Design Fuzzy Design Local Polynomial Regression Bandwidth Selection A/B Testing
COMPETING TERMS
Difference-in-Differences Propensity Score Matching
KNOWLEDGE AREAS
Data Science Mathematics Testing & Validation
TARGET AUDIENCE
Data Scientist Data Analyst Information Analyst AI Product Manager
EXTERNAL SOURCES
- Wikipedia - Regression discontinuity design
- Cunninghamâs Causal Inference: The Mixtape - Accessible RD tutorial
- rdrobust authors - RD software and papers
- NBER - Guide to RD practice
TOP QUESTIONS
What is Regression Discontinuity in causal inference?
It estimates local treatment effects by exploiting a threshold-based assignment rule and comparing units near the cutoff.
Category: Definition
When is Regression Discontinuity preferred?
When randomization isnât possible but a deterministic or probabilistic threshold governs treatment assignment, common in policy and education.
Category: Relevancy
What is the difference between sharp and fuzzy RD?
Sharp RD assigns exactly at the cutoff; fuzzy RD has imperfect compliance and uses instrumental-variable methods.
Category: Misunderstanding
How do you select bandwidth in RD?
Use data-driven methods (e.g., cross-validation) to balance bias and variance; test robustness across bandwidths.
Category: Requirements
What assumptions underpin RD validity?
Continuity of potential outcomes at the cutoff, no precise manipulation of the running variable, and correct functional form near the threshold.
Category: Requirements
What are typical RD pitfalls?
Overfitting polynomial trends, poor bandwidth choice, failing to test for sorting around the threshold, and ignoring heterogeneity.
Category: Mistakes
What risks arise if RD assumptions fail?
Biased local effects and misleading inference; manipulation at the cutoff undermines identification.
Category: Risks
What methods compete with Regression Discontinuity?
Difference-in-Differences and Propensity Score Matching provide alternatives under different identifying assumptions.
Category: Alternatives
Where to learn RD method and practice?
Read Imbens & Lemieux (NBER), use rdrobust documentation, and explore accessible tutorials like Cunninghamâs Mixtape.
Category: HowTo