ScholarGate
Get started
Regression modelCausal inferenceSpatial causal inferenceModel

Spatial Difference-in-Differences

Also known as: Spatial DiD, Geo-DiD, Difference-in-Differences with Spatial Autocorrelation, Mekansal Fark-içinde-Farklar

OriginatorDelgado & FloraxYear2015Sources1Related methods10

Spatial Difference-in-Differences (Spatial DiD) extends the classical DiD estimator to settings where observations are geo-referenced and outcomes may be spatially autocorrelated or subject to spillover effects. Introduced by Delgado and Florax (2015), the method augments the standard two-way fixed-effects DiD regression with a spatial lag or spatial error term, yielding unbiased treatment-effect estimates even when policy shocks propagate across geographic units. It is used by economists, regional scientists, and urban planners evaluating place-based interventions such as infrastructure investment, environmental regulations, or zoning reforms.

Key highlights

  • Accounts for spatial autocorrelation that would otherwise inflate standard errors or bias the treatment estimate in conventional DiD
  • Explicitly separates direct local treatment effects from indirect geographic spillovers, yielding more policy-relevant decompositions
  • Compatible with the standard two-way fixed-effects framework, so existing DiD software extensions can accommodate spatial terms
  • LM diagnostic tests provide a principled, data-driven way to choose between spatial-lag and spatial-error specifications

Intuition

This section is available to Pro members. Upgrade to Pro

How it works

This section is available to Pro members. Upgrade to Pro

When to use it

Use Spatial DiD when you have a panel of geo-referenced units, a clearly defined treatment group exposed to a policy shock, and theoretical or empirical reason to suspect spillovers or spatial autocorrelation in outcomes. The method requires that the parallel-trends assumption hold after controlling for spatial dependence, and that the spatial weight matrix be specified ex ante on substantive grounds. It is inappropriate when treatment assignment is purely random without spatial structure, or when the study area has too few units for reliable spatial-parameter estimation. Synthetic control or matching DiD are preferable when pre-treatment trends diverge sharply.

Strengths & limitations

Strengths
  • Accounts for spatial autocorrelation that would otherwise inflate standard errors or bias the treatment estimate in conventional DiD
  • Explicitly separates direct local treatment effects from indirect geographic spillovers, yielding more policy-relevant decompositions
  • Compatible with the standard two-way fixed-effects framework, so existing DiD software extensions can accommodate spatial terms
  • LM diagnostic tests provide a principled, data-driven way to choose between spatial-lag and spatial-error specifications
Limitations
  • Requires the researcher to pre-specify the spatial weight matrix W; results can be sensitive to its construction
  • Consistent estimation demands maximum likelihood or IV/GMM, which are more computationally demanding and less transparent than OLS
  • The parallel-trends assumption must hold conditionally on spatial dependence, which is harder to verify and defend than in the aspatial case
  • With small numbers of geographic units, spatial-parameter estimates are imprecise and diagnostic tests have low power

Common pitfalls

This section is available to Pro members. Upgrade to Pro

Applications

This section is available to Pro members. Upgrade to Pro

Frequently asked

How do I choose the spatial weight matrix?

The weight matrix should reflect the hypothesized mechanism by which outcomes in one unit influence outcomes in another. Queen contiguity is standard for administrative areas sharing borders; inverse-distance weights suit continuous geographic processes. The choice must be justified on substantive grounds before estimation. Sensitivity analyses using alternative matrices are strongly recommended to assess robustness.

Can Spatial DiD handle staggered treatment timing?

Yes, but with added complexity. Recent advances in staggered DiD — such as Callaway-Sant'Anna or Sun-Abraham estimators — do not yet fully integrate spatial autocorrelation corrections. Researchers currently combine spatial lag or error terms with heterogeneous-treatment-timing designs manually, though this remains an active area of methodological development.

What software implements Spatial DiD?

R packages spdep and spatialreg provide spatial-lag and spatial-error MLE alongside standard DiD regression infrastructure. Stata users typically combine spatreg or spreg with manual DiD interaction terms. Python users can employ PySAL for weight-matrix construction and spreg for estimation, then add the DiD interaction within standard panel-data frameworks.

Sources

  1. 1.
    Delgado, M. S., & Florax, R. J. G. M. (2015). Difference-in-differences techniques for spatial data: Local autocorrelation and spatial interaction. Economics Letters, 126, 35–40.
    DOI 10.1016/j.econlet.2015.10.035

You have read it. What now?

Ask about this methodSimilar methods

Cite this page

ScholarGate. (2026, June 2). Spatial Difference-in-Differences. ScholarGate. https://scholargate.app/causal-inference/spatial-difference-in-differences

Spatial Difference-in-Differences | ScholarGate