Spatial Sensitivity Analysis for Causality
Spatial Sensitivity Analysis for Causal Inference · Also known as: spatial causal sensitivity, spatial robustness checks, SSAC, spatial confounding sensitivity
Spatial sensitivity analysis for causality systematically tests whether a causal estimate derived from georeferenced data holds up as spatial structure, spillovers, and the choice of spatial weights matrix are varied. Because nearby units often share unmeasured confounders — soil quality, local infrastructure, neighbourhood norms — a naive regression may yield biased causal estimates. This method reveals how fragile or robust a claimed causal effect is to alternative spatial specifications.
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 spatial sensitivity analysis whenever you are drawing a causal conclusion from georeferenced data and there is a plausible reason to believe that spatial spillovers or spatially structured confounders are present — for example, in environmental epidemiology, regional economics, urban policy evaluation, or agricultural trials. It is particularly important when the treatment itself is spatially assigned (e.g., a policy rolled out in selected districts) because spillover and SUTVA violations are then likely. Do not apply it as a substitute for a valid identification strategy; it supplements, not replaces, methods like DiD or IV. It is not appropriate for truly independent random samples where spatial location carries no information.
Strengths & limitations
- Directly addresses the SUTVA (Stable Unit Treatment Value Assumption) violation risk in spatial data by making spillover and neighbourhood assumptions explicit.
- Provides a transparent audit of how sensitive the causal estimate is to the choice of spatial weights matrix, which is inherently researcher-driven.
- Integrates naturally with standard spatial econometric software and requires no special randomisation or instrumental variable.
- Can quantify the magnitude of spatial confounding needed to nullify the estimated effect, giving a principled bound on credibility.
- Applicable across a wide range of settings: areal data (regions, census tracts), point-referenced environmental measurements, and panel spatial data.
- No single sensitivity analysis can rule out all forms of spatial confounding; the approach depends on the candidate weight matrices actually considered.
- Spatial weights matrices are inherently arbitrary — there is no gold-standard neighbourhood definition, and the results can be framed to appear robust by choosing a favourable set.
- The method does not provide a causal estimate on its own; it interrogates an estimate from a primary identification strategy that must already be defensible.
- Spillover effects (indirect treatment effects) are difficult to disentangle from direct effects without additional structural assumptions.
- Computationally intensive when the number of candidate spatial specifications is large or when data sets are very large.
Frequently asked
Is spatial sensitivity analysis the same as a spatial econometric model?
No. Spatial econometric models (lag, error, Durbin) are estimation tools. Spatial sensitivity analysis is a diagnostic practice: you run multiple such models under varying assumptions and compare how the causal estimate changes. The analysis uses spatial models as instruments of interrogation, not as an end in themselves.
How do I choose which spatial weights matrices to vary?
A principled approach is to consider at least three structurally different definitions: contiguity (queen or rook), k-nearest neighbours for several values of k, and a distance-decay kernel. If results are stable across all of these, the evidence for robustness is stronger than if only one type is tested.
What does it mean if my estimate changes a lot across spatial specifications?
It signals that the causal claim is sensitive to how spatial dependence is modelled, which usually indicates the presence of spatially structured confounding or spillovers that the original identification strategy did not account for. The honest response is to widen your uncertainty statements and seek a stronger identification strategy.
Can I use this method with panel data?
Yes. Spatial-temporal panel models extend the lag and error frameworks to multiple periods. Fixed or random effects can be combined with spatial weights, and the sensitivity analysis proceeds by varying the spatial specification while holding the temporal structure constant.
Does spatial sensitivity analysis prove causality?
No. Like any sensitivity analysis, it can strengthen or weaken confidence in a causal claim, but it cannot establish causality by itself. A credible identification strategy (natural experiment, instrumental variable, randomisation) remains the primary requirement.
Sources
- Anselin, L. (1988). Spatial Econometrics: Methods and Models. Kluwer Academic Publishers, Dordrecht. ISBN: 978-9024737322
- Reich, B. J., Yang, S., Guan, Y., Giffin, A. B., Miller, M. J., & Rappold, A. G. (2021). A review of spatial causal inference methods for environmental and epidemiological applications. International Statistical Review, 89(3), 605-634. DOI: 10.1111/insr.12452 ↗
How to cite this page
ScholarGate. (2026, June 3). Spatial Sensitivity Analysis for Causal Inference. ScholarGate. https://scholargate.app/en/causal-inference/spatial-sensitivity-analysis-for-causality
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
- Geographically Weighted RegressionSpatial analysis↔ compare
- Instrumental Variables in Health ResearchHealth Economics↔ compare
- Propensity Score MatchingResearch Statistics↔ compare
- Spatial Error ModelSpatial analysis↔ compare
- Spatial Lag ModelSpatial analysis↔ compare