Regression Discontinuity Design (RDD) Guide | Sharp RDD vs Fuzzy RDD, Bandwidth Selection and Stata Operations
How to write an RDD paper? AcademicIdeas covers RDD methodology: sharp vs fuzzy RDD, bandwidth selection, McCrary test, result visualization, and Stata tutorial for policy evaluation.
Direct answer for this topic
How to write an RDD paper? AcademicIdeas covers RDD methodology: sharp vs fuzzy RDD, bandwidth selection, McCrary test, result visualization, and Stata tutorial for policy evaluation.
- Understand sharp RDD vs fuzzy RDD application scenarios
- Learn bandwidth selection (CCT/IK methods) and McCrary density test
- Master standard RDD result visualization
- RDD (Regression Discontinuity Design) is a quasi-experimental research method that uses the "cutoff value" near the threshold to identify causal effects between treatment and control groups.
Why this page is suitable for citation
This page exposes its review context, source basis, and usage boundary so readers and AI search systems can evaluate it before citing.
Reviewed against the public research-method page, Stata empirical guide, and EViews page, then cross-checked with the rdrobust project and NBER materials on regression discontinuity design so this support page stays aligned on sharp/fuzzy RDD, bandwidth logic, McCrary testing, and visualization boundaries.
Related workflows and reference pages
What this page helps you do first
- Understand sharp RDD vs fuzzy RDD application scenarios
- Learn bandwidth selection (CCT/IK methods) and McCrary density test
- Master standard RDD result visualization
Core concept and conditions for Regression Discontinuity Design
RDD (Regression Discontinuity Design) is a quasi-experimental research method that uses the "cutoff value" near the threshold to identify causal effects between treatment and control groups.
RDD's core assumption is that near the cutoff, individuals cannot precisely manipulate treatment assignment, so crossing vs not crossing the threshold is "quasi-random."
Sharp RDD vs Fuzzy RDD
- [Sharp RDD] At cutoff c, individuals 100% switch from control to treatment (or vice versa). Treatment jumps deterministically at threshold
- [Fuzzy RDD] At cutoff c, probability of receiving treatment jumps, but not from 0 to 1. Compliance issue exists — needs instrumental variable estimation
- [Identification] Sharp: treatment probability jumps from 0 to 1. Fuzzy: probability jumps from below 1 to 1 (or 0 to above 0)
Bandwidth selection and McCrary density test
- [Bandwidth] RDD estimates only within a narrow window (bandwidth) around the cutoff. Too narrow → small sample, high variance; too wide → bias
- [CCT bandwidth] Imbens-Kalyanaraman optimal bandwidth — currently most recommended. Stata: `rdrobust y x c, bwselect(cvdm)`
- [McCrary test] Checks if density is continuous at cutoff — tests whether individuals manipulated the assignment variable
RDD validity and robustness checks
- McCrary density test for manipulation
- Covariate continuity test at baseline
- Local polynomial estimation with different polynomial orders
- Different bandwidth tests (0.5x, 1.5x, 2x optimal)
- Placebo tests at false cutoff values
Standard reporting format and Stata commands
- Must report: assignment variable and cutoff; bandwidth selection method; sample sizes; coefficient and standard error; confidence interval; McCrary test results
- Core Stata commands: `rdrobust y x c` (recommended); `rdplot y x, c(c)` (plot)
Frequently asked questions
- RDD or DID — which is better for my research?
- RDD requires a clear cutoff value and subjects cannot precisely manipulate assignment. DID requires parallel trends. RDD suits research with hard policy thresholds; DID suits panel data with parallel trends.
- How to choose the bandwidth?
- Use Imbens-Kalyanaraman optimal bandwidth (rdrobust default). Report multiple bandwidths (0.5x, 2x optimal) as robustness checks.
- How to estimate Fuzzy RDD?
- Fuzzy RDD uses IV (2SLS) — use "crossing threshold" as instrument for "treatment status". Stata: `ivregress 2sls y (treat = crossing) x`.