EvalCompassEvalCompass
Causal inference & experimentsCausal inferenceThresholdNatural experimentQuasi-experimentLocal effects

Regression Discontinuity Design

Regression Discontinuity Design
Also known as: RDD、Regression discontinuity design、Threshold analysis

Uses schemes where assignment changes at a continuous indicator's threshold to compare nearby outcomes. Estimates a local threshold difference, not automatically generalizable to distant targets. [1]

RDD design and diagnostics checklist — openReferences and checked sections
Difficulty
Advanced
Cost
Medium
Duration
Illustrative planning durations, not common standards: preparation 1–3 weeks, analysis/diagnostics 1–2 weeks
Content updated: 2026-09-13Source checked: 2026-09-13 (includes AI cross-checking)No documented human reviewIncludes fictional examples

How it works at a glance

Check the running variable, cutoff and actual rules. Examine whether outcomes would be continuous without intervention and whether manipulation or other schemes create jumps. [1]

Distinguish sharp assignment changes from fuzzy probability changes. Check robustness across bandwidths and specifications. [1]

ResultsAssignment variable (e.g. score)Threshold (cutoff)Discontinuity = effectNot eligibleEligible for intervention
Treat the size of the jump in outcomes around the assignment threshold as the effect. This is a local causal effect near the cutoff.

When to use it and when it is unsuitable

When to use it

  • Eligibility determined by available numerical score, income or age cutoffs
  • Retrospective education selection, income-based social benefits or subsidy programs with clear cutoffs
  • Receipt or probability changes at a usable threshold
  • Local cutoff effects suffice for policy decisions

When it is unsuitable

  • Too few observations near the boundary for sufficient precision
  • Running variables can be manipulated deliberately near the cutoff
  • Need whole-program or all-target average effects rather than local effects

Participants, roles, and deliverables

Participants and roles

1–2 econometric/statistical specialists collaborating with authorized administrative/score-data managers

Main deliverables

  • RDD plot: running variable/outcome scatter and local polynomial fits
  • Discontinuity estimates, standard errors and confidence intervals
  • Robustness: bandwidths, placebo cutoffs, etc.
  • McCrary density test results for manipulation

Implementation steps

  1. Identify running variable and cutoff
    Record assignment variable, cutoff and actual receipt. Examine sharp versus fuzzy changes and exceptions.
  2. Collect data and check distributions
    Collect running variables, outcomes and pre-traits; examine density, distributions and implementation. Density tests alone do not prove no manipulation; consider discrete scores and score generation.
  3. Visualize near the cutoff
    Plot outcomes against running variables with local polynomial fits to inspect jumps. Try multiple bin widths for stability.
  4. Select bandwidth
    Set local data range and method and explain selection. Report multiple bandwidth analyses, not only favorable ones.
  5. Calculate RDD
    Estimate conditional outcomes on each side and their cutoff difference. Sharp designs estimate mean cutoff effects; fuzzy designs consider outcome jumps divided by receipt-probability jumps under additional assumptions. Match intervals to estimation.
  6. Robustness tests
    Report bandwidth/model changes, pre-traits and other cutoffs. Check other interventions changing at the same cutoff. No detected discontinuity does not prove continuity.

Strengths and limitations

Strengths

  • Valid threshold comparisons estimate nearby effects
  • Uses existing administrative or eligibility rules without new experiments
  • Transparent running-variable/cutoff settings and many established diagnostic tests
  • Allows specific examination of assignment and continuity grounds

Limitations

  • Sharp/fuzzy estimands differ; both usually concern near-cutoff effects, not all-target averages.
  • Deliberate running-variable manipulation undermines identification
  • Sparse near-cutoff samples increase uncertainty and may lack practical precision
  • Counterfactual continuity of potential outcomes near the cutoff cannot be proven

Common pitfalls

  • Check manipulation and other cutoff-linked schemes using institutional documents, implementation, distributions and pre-traits.
  • Do not choose bandwidth to fit findings. Show rationale, multiple settings and uncertainty.
  • Do not generalize local effects as whole-program effects.
  • Do not merely fit complex curves over broad ranges. Check local specifications and sensitivity.

Conditions of use and evidence

  • Check threshold assignment mechanisms and interpret effects locally. [1]
Fictional example; not actual results

Assign training at a score cutoff

A fictional scheme assigns training at scores of 60 or above to illustrate sharp RDD estimates.

EstimandFictional estimate at cutoff 60
Limit of use rate from below50%
Limit of use rate from above62%
Cutoff difference12 percentage points

Common errors in judgment

Common mistake

Simply compare everyone scoring 60 or above with everyone scoring below 60

How to revise

Estimate functions locally on both sides

ReasonDistant scores may have very different preconditions.

Common mistake

Ignore exceptions and report sharp RDD

How to revise

Check actual receipt-probability jumps before selecting design

ReasonMeaning depends on correspondence between assignment and receipt.

Organize your thinking by writing

RDD design and diagnostics checklist

A practical template is available to fill in and print on A4 paper.

Open worksheet

Related practical guides

What is impact evaluation?

Related methods

Randomized Controlled TrialDifference-in-DifferencesInterrupted Time Series AnalysisPropensity Score MatchingContribution Analysis

References and checked sections

  1. Development Research in Practice: Research design for impact evaluation
    World Bank DIME Analytics
    Section consulted
    Regression discontinuity
    How it relates to the text
    Check threshold assignment mechanisms and interpret effects locally.
    Source checked: (includes AI cross-checking)

Editorial and checking information

Content updated: Source checked: (includes AI cross-checking)No documented human reviewIncludes fictional examplesEditorial policy and scope of checkingCite this pageReport a correction

How to cite this page

EvalCompass Editorial Team. “Regression Discontinuity Design”. EvalCompass, updated 2026-09-13. https://evalcompass.jun-nakatani.com/en/methods/rdd/
Add the date you accessed the page. Text created by this site is available under CC BY 4.0. Quotations, external materials, and bundled fonts have their own terms of use. When reusing content, credit the source, link to the license, and indicate any changes.

Suggest a correction or improvement to this page

Change history
  • — Checked sources for definitions and conditions and added fictional examples, judgment rationales and completed entries for every field. Revised unsupported historical or uptake claims. Reviewed procedures and cautions, adjusting fixed counts, thresholds and institutional generalizations. Preserved field positions, types and counts while revising fixed conditions and thresholds.