Spatial Propensity Score Matching
Spatial Propensity Score Matching Estimator · Also known as: Spatial PSM, Geospatial PSM, Spatially-adjusted propensity score matching, Geographic propensity score matching
Spatial Propensity Score Matching (Spatial PSM) extends the classic propensity score matching framework to settings where units are embedded in geographic space and treatment assignment or outcomes may be spatially correlated. By incorporating spatial covariates and adjacency structure into the propensity model and matching procedure, it produces causal estimates that account for geographic confounding and spillover effects.
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.
+2 more
When to use it
Use Spatial PSM when treatment assignment is geographically structured — for example, a policy rolled out in certain regions, environmental exposure driven by proximity to a source, or an infrastructure intervention affecting particular areas — and when control units can be found within reasonable geographic proximity. It is appropriate for cross-sectional or repeated cross-sectional data with geographic identifiers, moderately large samples (ideally n > 100 per group), and continuous or binary outcomes. Do not use it when geographic coverage is too sparse to find proximate controls, when spatial spillovers are so pervasive that no clean control group exists, or when the treatment boundary itself is endogenous to local conditions (in which case a spatial regression discontinuity design may be preferable).
Strengths & limitations
- Removes bias from spatial confounders by explicitly incorporating geographic context into the propensity model.
- Produces a geographically comparable control group, reducing the risk that matched pairs differ on unmeasured local conditions.
- Flexible: spatial covariates can be added to any standard PSM workflow without requiring specialised spatial regression software.
- Compatible with a range of matching algorithms (nearest-neighbour, kernel, radius, stratification) and distance metrics.
- Helps diagnose and partially address spatial spillover by identifying treated-control pairs that are geographically too close.
- Requires reliable geographic identifiers and the construction of spatial covariates, which may be data-intensive.
- Matching on both propensity score and geography simultaneously can make it difficult to find close matches, reducing effective sample size.
- Does not fully solve the problem of spatial interference (SUTVA violations); if spillovers are strong, even spatially-matched estimates may be biased.
- Balance on measured spatial covariates does not guarantee balance on unmeasured local factors.
- Performance depends on the quality of the spatial covariate specification; poorly chosen neighbours or distance metrics can introduce new biases.
Frequently asked
How does Spatial PSM differ from standard PSM?
Standard PSM matches on observed covariates but ignores geographic location. Spatial PSM additionally includes geographic covariates in the propensity model and imposes a geographic constraint on matching — preferring or requiring control units to be spatially close. This reduces bias from local unobserved factors that standard PSM cannot address.
What spatial covariates should I include in the propensity model?
Useful candidates include geographic coordinates (latitude/longitude), spatial lag of key covariates (average values among neighbouring units), distance to treatment boundary, local area fixed effects (region or district dummies), and any contextual variables capturing local economic, demographic, or environmental conditions. The goal is to absorb spatial confounding before matching.
How do I handle spatial spillovers?
First, assess whether control units near treated areas may be partially treated themselves. If so, exclude them from the control pool using a buffer zone around treatment boundaries. Alternatively, estimate separate effects for near-boundary versus distant control units, or use spillover-robust estimators that model interference explicitly.
Does Spatial PSM require specialised software?
Not necessarily. In R or Python you can compute spatial covariates (e.g., with sf, spdep, or PySAL) and then use standard PSM packages (MatchIt in R, causalml or zEpid in Python), adding the spatial variables and a geographic caliper. Some applied papers implement the geographic caliper via maximum acceptable distance between matched units.
When should I prefer a spatial regression discontinuity design instead?
Prefer spatial RDD when treatment assignment is determined by crossing a sharp geographic boundary (e.g., administrative border, jurisdiction line) and units on either side of that boundary are otherwise comparable. Spatial PSM is more appropriate when treatment is broadly geographically clustered but does not follow a sharp boundary, and when a broader set of control units is available.
Sources
- Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in observational studies for causal effects. Biometrika, 70(1), 41-55. DOI: 10.1093/biomet/70.1.41 ↗
- Kelejian, H. H., & Prucha, I. R. (2004). Estimation of simultaneous systems of spatially interrelated cross sectional equations. Journal of Econometrics, 118(1-2), 27-50. DOI: 10.1016/S0304-4076(03)00133-7 ↗
How to cite this page
ScholarGate. (2026, June 3). Spatial Propensity Score Matching Estimator. ScholarGate. https://scholargate.app/en/causal-inference/spatial-propensity-score-matching
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.
- Coarsened Exact MatchingCausal inference↔ compare
- Propensity Score MatchingResearch Statistics↔ compare
- Spatial Doubly Robust EstimationCausal inference↔ compare
- Spatial Instrumental VariablesCausal inference↔ compare
- Spatial Regression Discontinuity DesignCausal inference↔ compare
- Spatial Synthetic Control MethodCausal inference↔ compare