Policy Evaluation Regression Discontinuity Design
Regression Discontinuity Design for Policy Evaluation · Also known as: Policy RDD, RD design in policy evaluation, regression discontinuity policy analysis, RDD policy impact
Policy Evaluation Regression Discontinuity Design (Policy RDD) exploits a known eligibility threshold in a policy rule to estimate the causal effect of that policy on outcomes. Units just below the cutoff serve as a credible comparison group for units just above it, making RDD one of the most transparent quasi-experimental strategies for assessing what a policy actually achieves.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use Policy RDD when a policy assigns treatment based on a continuous score or index crossing a known cutoff, and you cannot randomize. The design is ideal for evaluating eligibility rules in welfare, education, health, tax, or regulatory programmes — for example, income thresholds for benefit receipt, test-score cutoffs for school admission, or age-based eligibility rules. A sufficient number of observations near the bandwidth is essential; thin samples near the cutoff yield noisy estimates. Do not use RDD when the cutoff is arbitrary or unknown, when the running variable is discrete with few mass points, when units can precisely manipulate their score to cross the threshold, or when the local effect at the margin is not the policy-relevant quantity. RDD estimates a local average effect for marginal units, not an average effect across the full population.
Strengths & limitations
- Exploits a known institutional rule, making the identification assumption — continuity of potential outcomes at the cutoff — transparent and verifiable.
- Does not require a randomly assigned control group; the comparison group emerges naturally from the policy design.
- Allows flexible modelling of confounders through the running-variable function, reducing sensitivity to covariate imbalance away from the cutoff.
- Widely accepted by policymakers, journals, and evaluation bodies as a credible quasi-experimental standard.
- Easily diagnosed: manipulation tests, covariate balance, and placebo checks allow systematic validation of the design.
- Identifies only a local average treatment effect for units at the margin of eligibility; results may not generalize to units far from the cutoff.
- Requires adequate density of observations near the bandwidth — sparse data at the threshold inflates variance and forces wider bandwidths.
- Fails entirely if units can precisely manipulate the running variable to sort just above (or below) the cutoff, invalidating continuity.
- The choice of bandwidth and polynomial order for the running variable can influence the estimated effect, requiring careful sensitivity analysis.
- Applicable only to programmes with a pre-specified, continuous eligibility threshold; policies without a clear rule cannot be studied this way.
Frequently asked
What is the continuity assumption, and how do I check it?
The continuity assumption states that the conditional expectation of potential outcomes is a smooth, continuous function of the running variable at the cutoff — so any observed jump is caused solely by the policy. Check it by (1) running a McCrary density test to rule out manipulation, and (2) testing for jumps in pre-determined covariates at the cutoff using balance regressions. If either test shows a discontinuity, the design is invalid.
What is the difference between sharp and fuzzy RDD in policy evaluation?
In a sharp RDD, crossing the threshold perfectly determines treatment status: everyone above receives the policy and no one below does. In a fuzzy RDD, the threshold only changes the probability of receiving the policy — some below the cutoff are treated and some above are not (non-compliance). Fuzzy RDD uses the threshold as an instrumental variable and estimates a local average treatment effect for compliers rather than all marginal units.
How do I choose the bandwidth?
Use a data-driven optimal bandwidth selector such as the Calonico, Cattaneo, and Titiunik (CCT) selector, which minimizes the asymptotic mean squared error of the local linear RDD estimator. Always report sensitivity checks across a range of bandwidths to confirm that the main finding is not driven by a single choice.
Why is a local linear regression preferred over a higher-order polynomial?
Higher-order global polynomials fitted across the full running-variable range can produce extreme, erratic estimates at and near the boundary — a well-documented problem in the RDD literature. Local linear regression within a narrow bandwidth has better bias properties at the cutoff, is more robust to model misspecification, and is the currently recommended default in applied practice.
Can Policy RDD be combined with difference-in-differences?
Yes. A difference-in-discontinuities design stacks RDD with DiD by comparing the jump at the cutoff before and after a policy change. This is useful when multiple cohorts face the same threshold at different times, or when a pre-existing discontinuity at the cutoff needs to be netted out before attributing the jump to the policy of interest.
Sources
- Lee, D. S., & Lemieux, T. (2010). Regression Discontinuity Designs in Economics. Journal of Economic Literature, 48(2), 281-355. DOI: 10.1257/jel.48.2.281 ↗
- Imbens, G. W., & Lemieux, T. (2008). Regression discontinuity designs: A guide to practice. Journal of Econometrics, 142(2), 615-635. DOI: 10.1016/j.jeconom.2007.05.001 ↗
How to cite this page
ScholarGate. (2026, June 3). Regression Discontinuity Design for Policy Evaluation. ScholarGate. https://scholargate.app/en/causal-inference/policy-evaluation-regression-discontinuity-design
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Difference-in-DifferencesEconometrics↔ compare
- Fuzzy Regression DiscontinuityCausal inference↔ compare
- Instrumental Variables in Health ResearchHealth Economics↔ compare
- Policy Evaluation Difference-in-DifferencesCausal inference↔ compare
- Propensity Score MatchingResearch Statistics↔ compare