Spatial Instrumental Variables (Spatial IV / Spatial 2SLS)
Spatial Instrumental Variables Estimation · Also known as: Spatial IV, Spatial 2SLS, Spatial Two-Stage Least Squares, S-IV
Spatial Instrumental Variables (Spatial IV) is a causal inference method for settings where units — regions, firms, neighborhoods — are spatially interdependent, creating endogeneity that standard IV approaches ignore. It constructs instruments from the spatially lagged values of exogenous characteristics of neighboring units, then applies two-stage least squares to recover unbiased causal estimates in the presence of both endogenous regressors and spatial autocorrelation.
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 IV when you have geo-referenced observational data, an endogenous treatment or regressor, and credible exogenous variation that operates through spatial proximity — for example, evaluating how a neighboring region's policy affects local economic outcomes, or how market access driven by neighbors' infrastructure affects local firms. It is appropriate when units are clearly embedded in a geographic or network structure and spillovers are substantively important. Do not use it when the neighborhood structure is arbitrary or the exclusion restriction cannot be defended, when all spatial variation in the instrument also directly affects the outcome, or when observations are independent (standard IV is sufficient). Avoid if the spatial weights matrix is misspecified — results are sensitive to the choice of W.
Strengths & limitations
- Addresses endogeneity in spatially dependent data where conventional IV instruments are themselves contaminated by spatial autocorrelation.
- Exploits natural geographic variation — neighbors' exogenous characteristics — that is often plausibly exogenous and easily measured.
- The spatial weights matrix makes the identification assumption explicit and transparent, enabling robustness checks by varying W.
- GS2SLS simultaneously handles endogenous regressors and spatially autocorrelated errors, providing consistent and asymptotically efficient estimates.
- Applicable to both cross-sectional and panel data with spatial structure, covering a wide range of empirical settings.
- The exclusion restriction — that neighbors' instrument values affect the outcome only through the endogenous variable — is often difficult to defend when outcomes themselves spill over spatially.
- Results are sensitive to the choice and specification of the spatial weights matrix W; no single 'correct' W exists in most applications.
- Weak instruments remain a risk: spatially lagged exogenous variables may be only weakly correlated with the endogenous regressor, especially when spillovers are limited.
- Requires sufficient spatial density and variation; sparsely observed geographic regions or arbitrarily defined neighborhood boundaries reduce credibility.
- Software implementation (spatialreg in R, spreg in Python's PySAL) is more complex than standard 2SLS, raising the risk of specification errors.
Frequently asked
How do I choose the spatial weights matrix W?
W should reflect the substantive mechanism of spatial interdependence: contiguity (shared borders) for administrative units, inverse distance for market interactions, k-nearest neighbors when density varies across the sample. Justify the choice on theoretical grounds and test robustness by repeating the analysis with alternative W specifications. Avoid choosing W based solely on which yields the strongest first stage.
What makes a valid spatial instrument?
A valid spatial instrument must be (1) relevant: neighbors' exogenous characteristic must predict your endogenous regressor with sufficient strength (first-stage F > 10 is a common rule of thumb); and (2) excludable: it must affect the outcome only through the endogenous variable. The exclusion restriction is violated if outcomes themselves spill over to neighbors, which is common — defend it carefully with contextual arguments and sensitivity analyses.
Do I need to correct for spatial autocorrelation in the residuals?
Yes, in most geographic applications. Unobservable confounders tend to cluster spatially, producing correlated residuals that inflate standard errors if ignored. Use Moran's I to test residuals after estimation; if spatial autocorrelation is detected, use GS2SLS (Kelejian-Prucha 1998) or HAC-robust standard errors.
When does Spatial IV reduce to standard IV?
When the spatial weights matrix encodes no economically meaningful neighborhood structure, or when there are no spatial spillovers, Spatial IV collapses to standard 2SLS. In practice, always test whether spatial structure matters before adopting the more complex Spatial IV procedure.
Can I use Spatial IV with panel data?
Yes. Panel Spatial IV extends the framework by adding unit fixed effects to remove time-invariant unobservables and using within-unit variation over time alongside cross-sectional spatial variation. This requires care in constructing the spatial weights to be consistent across time periods and in specifying whether spillovers are contemporaneous or lagged.
Sources
- Kelejian, H. H., & Prucha, I. R. (1998). A Generalized Spatial Two-Stage Least Squares Procedure for Estimating a Spatial Autoregressive Model with Autoregressive Disturbances. Journal of Real Estate Finance and Economics, 17(1), 99-121. DOI: 10.1023/A:1007707430416 ↗
- Anselin, L. (1988). Spatial Econometrics: Methods and Models. Kluwer Academic Publishers, Dordrecht. ISBN: 978-9024737208
How to cite this page
ScholarGate. (2026, June 3). Spatial Instrumental Variables Estimation. ScholarGate. https://scholargate.app/en/causal-inference/spatial-instrumental-variables
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.
- Instrumental Variables in Health ResearchHealth Economics↔ compare
- Panel Data Instrumental VariablesCausal inference↔ compare
- Spatial Causal Impact AnalysisCausal inference↔ compare
- Spatial Doubly Robust EstimationCausal inference↔ compare
- Spatial Propensity Score MatchingCausal inference↔ compare
- Spatial Regression Discontinuity DesignCausal inference↔ compare