Fuzzy Regression Discontinuity Design in Education Research
Also known as: Fuzzy RDD, Fuzzy RD, Imperfect RDD, Non-sharp RD
Fuzzy Regression Discontinuity Design (Fuzzy RDD) is a quasi-experimental causal method that exploits a known score threshold — such as a test cutoff — to estimate the effect of a program or intervention when assignment is imperfect. Widely used in education research to evaluate summer school, remedial programs, scholarships, and class-size rules, it uses two-stage least squares to recover a local average treatment effect for students near the threshold.
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 Fuzzy RDD in education research when there is a known score or eligibility threshold that strongly — but imperfectly — determines who receives an intervention, such as a remedial program, scholarship, grade retention policy, or class-size cap. The running variable must be continuous, and students should not be able to precisely manipulate their score to just above or below the cutoff. The method estimates a LATE for students near the threshold and should not be generalised to students far from the cutoff. Do not use it when compliance is essentially perfect (use Sharp RDD), when the running variable is discrete without enough mass at the threshold, or when the threshold is unknown or endogenously set.
Strengths & limitations
- Provides a credible causal estimate in observational education data without requiring random assignment, exploiting a naturally occurring cutoff.
- Handles imperfect compliance using instrumental variables logic, making it applicable to the common real-world case where assignment rules are not perfectly enforced.
- Requires only local identification near the threshold, so global modeling assumptions across the full score range are not needed.
- Transparent and falsifiable: density tests and placebo cutoff checks allow rigorous validation of the design assumptions.
- Widely understood by education policy audiences and accepted by high-quality journals as a credible identification strategy.
- Estimates apply only to compliers near the cutoff (LATE), limiting external validity for students far from the threshold or in other contexts.
- Requires a sufficiently large sample in the bandwidth window — very small samples near the cutoff produce imprecise estimates.
- Vulnerable to manipulation of the running variable: if students or institutions can sort just above or below the cutoff, the design is invalidated.
- Bandwidth and polynomial order choices are consequential and must be reported transparently; results can be sensitive to these decisions.
- A weak first stage (small jump in treatment probability at the threshold) causes the 2SLS estimator to be biased and unreliable.
Frequently asked
What makes an RDD 'fuzzy' rather than 'sharp'?
In a sharp RDD, crossing the cutoff deterministically assigns treatment — everyone above receives it and no one below does. In a fuzzy RDD, crossing the cutoff changes the probability of treatment but does not guarantee it, because some eligible students do not take up the program and some ineligible ones find their way in. Fuzzy RDD uses the jump in assignment probability as an instrument for actual treatment receipt.
What does the fuzzy RDD estimate — and for whom?
Fuzzy RDD estimates the Local Average Treatment Effect (LATE) — the causal effect of treatment for compliers near the cutoff. Compliers are students who receive treatment because they crossed the threshold and would not have otherwise. The estimate does not apply to always-takers (who would receive treatment regardless) or never-takers (who would not regardless).
How do I choose the bandwidth in a fuzzy RDD?
Use a data-driven bandwidth selector such as the Imbens-Kalyanaraman (2012) or Calonico-Cattaneo-Titiunik (2014) optimal bandwidth, which balances bias and variance. Always report results for multiple bandwidths (e.g., half and double the optimal) to demonstrate robustness, since results that depend heavily on a single bandwidth choice are fragile.
What validation tests should I run?
At minimum: (1) a McCrary density test to check there is no discontinuous bunching of students just above the cutoff (manipulation test); (2) placebo cutoff tests at other score values where no discontinuity should exist; (3) baseline covariate smoothness checks — pre-treatment characteristics should not jump at the cutoff; and (4) report the first-stage F-statistic to confirm the instrument is not weak.
Can I use fuzzy RDD with a small sample?
A small sample near the cutoff produces imprecise estimates and inflates standard errors. If fewer than roughly 50-100 observations fall within your chosen bandwidth, the estimates are unlikely to be reliable. Consider widening the bandwidth (accepting more bias) or, if the sample is fundamentally too small, acknowledge that the design is underpowered and the estimates should be interpreted cautiously.
Sources
- 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 ↗
- Jacob, B. A., & Lefgren, L. (2004). Remedial education and student achievement: A regression-discontinuity analysis. Review of Economics and Statistics, 86(1), 226-244. DOI: 10.1162/003465304323023778 ↗
How to cite this page
ScholarGate. (2026, June 3). Fuzzy Regression Discontinuity Design in Education Research. ScholarGate. https://scholargate.app/en/causal-inference/fuzzy-regression-discontinuity-in-education-research
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
- Instrumental Variables in Health ResearchHealth Economics↔ compare
- Local Average Treatment EffectCausal inference↔ compare
- Propensity Score MatchingResearch Statistics↔ compare