Spatial Regression Discontinuity Design (Spatial RDD)
Spatial Regression Discontinuity Design · Also known as: Spatial RDD, Geographic RDD, Border RD Design, Geographic Discontinuity Design
Spatial Regression Discontinuity Design uses a geographic or administrative boundary as the threshold that assigns units to treatment. Observations just inside one side of the boundary are compared with those just outside it, exploiting the near-random variation in treatment status near the cutoff to recover a local causal effect. The approach is widely used in economics, political science, and public health when policies or institutions change sharply at a border.
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.
+5 more
When to use it
Use Spatial RDD when a treatment, policy, or institutional rule is assigned according to which side of a known geographic boundary a unit lies on, and when geo-coded unit-level data are available. The method is most credible when the boundary was drawn for reasons unrelated to the outcome of interest, and when units near the border are plausibly comparable. It is not appropriate when the boundary itself was drawn endogenously to the outcome (for example, gerrymandering based on the outcome variable), when there are multiple conflated policies that change at the same border, or when the density of observations near the boundary is too thin to support local estimation.
Strengths & limitations
- Provides a credible causal estimate from observational data by exploiting sharp, exogenous geographic discontinuities without requiring randomisation.
- Controls implicitly for all smoothly varying geographic confounders — only the sharp jump at the boundary is attributed to treatment.
- Applies to a wide range of policy-relevant questions wherever administrative or political borders create sharp changes in rules, institutions, or programmes.
- Directly testable: covariate continuity and density tests provide falsifiable checks on the identifying assumptions.
- Complements standard RDD theory with a rigorous spatial framework, allowing researchers to select bandwidths and functional forms using established optimal methods.
- Identifies only a Local Average Treatment Effect for units near the boundary, which may not generalise to the full population far from the cutoff.
- Requires a sufficiently dense sample of observations near the boundary; sparse coverage makes local estimation unreliable.
- Multiple policies or institutional changes that coincide at the same geographic border confound identification, making it impossible to attribute the effect to a single cause.
- The assumption that the boundary is exogenous to the outcome can be violated when borders were drawn in response to the same phenomenon being studied.
- Two-dimensional geography can introduce spatial autocorrelation in residuals that standard RDD standard errors do not account for, requiring spatially clustered or HAC-robust inference.
Frequently asked
How does Spatial RDD differ from standard RDD?
Standard RDD uses a one-dimensional running variable (such as a test score) with a scalar cutoff. Spatial RDD uses geographic location as the assignment mechanism, so the cutoff is a line or curve in two-dimensional space rather than a single number. The key additional challenge is projecting two-dimensional location onto a meaningful running variable — typically signed distance to the boundary — and ensuring that this compression does not discard information that confounds the estimate.
What makes a geographic boundary credible for this design?
The boundary must have been drawn for reasons unrelated to the outcome being studied, so that units just inside and just outside are as-good-as-randomly assigned to treatment. Historical borders drawn for administrative convenience, colonial demarcation lines, or natural features used as political dividers are often credible. Borders drawn explicitly in response to the outcome variable — such as gerrymandered districts or boundaries redrawn after the outcome was observed — are not credible.
How do I choose the bandwidth?
The preferred approach is the data-driven mean-squared-error-optimal bandwidth selector of Calonico, Cattaneo, and Titiunik (CCT), which is implemented in standard software packages such as rdrobust in R and Stata. You should also report results for half and double the optimal bandwidth to demonstrate robustness.
What should I do if the boundary is curved rather than straight?
Compute signed distance from each unit's location to the nearest point on the boundary curve. This generalises the running variable construction to non-linear boundaries. Verify that the curvature is not so severe that units at very different parts of the boundary are being pooled into the same local comparison, as this could introduce heterogeneity that biases the estimate.
Do I need spatial data or can I use administrative records with a location code?
You need coordinates (or at minimum a location identifier linkable to coordinates) for every unit so that signed distance to the boundary can be computed. Administrative records with municipality or district identifiers are often sufficient if those units are small enough that their centroid distance to the boundary is a good approximation of true distance.
Sources
- Dell, M. (2010). The Persistent Effects of Peru's Mining Mita. Econometrica, 78(6), 1863-1903. DOI: 10.3982/ECTA8121 ↗
- Keele, L., & Titiunik, R. (2015). Geographic Boundaries as Regression Discontinuities. Political Analysis, 23(1), 127-155. DOI: 10.1093/pan/mpu014 ↗
How to cite this page
ScholarGate. (2026, June 3). Spatial Regression Discontinuity Design. ScholarGate. https://scholargate.app/en/causal-inference/spatial-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
- Propensity Score MatchingResearch Statistics↔ compare