Space-Time Spatial Regression
Also known as: spatio-temporal regression, spatial panel regression, space-time regression, ST spatial regression
Space-Time Spatial Regression extends classical spatial regression to panel settings where georeferenced units are observed across multiple time periods. By embedding a spatial weights matrix into a panel regression framework, it simultaneously controls for spatial dependence among cross-sectional units and temporal dynamics, yielding unbiased and consistent estimates in spatio-temporal data.
Key highlights
- Jointly controls for spatial spillovers and time dynamics, preventing omitted-variable bias from geographic contagion.
- Flexible model family — spatial lag, spatial error, and Spatial Durbin variants address different spillover mechanisms.
- Direct interpretation of spatial multipliers: the total effect of a covariate includes both direct and indirect (neighbor) impacts.
- Compatible with standard panel techniques such as fixed and random effects, allowing control for unobserved unit heterogeneity.
- Robust ML and GMM estimators available in widely used software (R: spdep, spatialreg; Python: PySAL; Stata: spxtregress).
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 Space-Time Spatial Regression when your panel data exhibit spatial dependence — for example, disease rates, housing prices, or pollution levels that cluster geographically — and you need unbiased coefficient estimates for policy analysis. It is appropriate when N (cross-sectional units) is at least moderate (≥ 30) and T is at least a few periods. Do not use it when spatial dependence is absent (verify with Moran's I on OLS residuals), when the spatial structure is unknown or arbitrary, when N is very small (< 15), or when the outcome is categorical — logit or count models with spatial extensions are more suitable in those cases.
Strengths & limitations
- Jointly controls for spatial spillovers and time dynamics, preventing omitted-variable bias from geographic contagion.
- Flexible model family — spatial lag, spatial error, and Spatial Durbin variants address different spillover mechanisms.
- Direct interpretation of spatial multipliers: the total effect of a covariate includes both direct and indirect (neighbor) impacts.
- Compatible with standard panel techniques such as fixed and random effects, allowing control for unobserved unit heterogeneity.
- Robust ML and GMM estimators available in widely used software (R: spdep, spatialreg; Python: PySAL; Stata: spxtregress).
- Requires a correctly specified spatial weights matrix W; results can be sensitive to the choice of contiguity rule or distance threshold.
- Estimation complexity grows with N: full ML requires inverting or approximating an N×N matrix at each iteration, which can be slow for large grids.
- Interpretation of spatial multipliers (direct vs. indirect effects) is non-trivial and often overlooked in applied work.
- Assumes a single, time-invariant weight matrix, which may misrepresent networks that change over time.
- With short panels (small T), variance estimation for spatial parameters may be imprecise.
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 between the spatial lag and spatial error model?
Run OLS on the panel (with fixed effects) and apply the Lagrange Multiplier (LM) tests for spatial lag and spatial error dependence to the residuals. If LM-lag is more significant, prefer the spatial lag model; if LM-error dominates, use the spatial error model. When both are significant, the Robust LM variants help discriminate.
What is the difference between direct and indirect effects?
In a spatial lag model, a change in covariate X_i affects y_i directly (direct effect) and also affects neighboring y_j through the spatial multiplier (indirect effect). The total effect is the sum of both. Reporting only the raw coefficient β misses the indirect spillovers, which can be larger than the direct effect.
Can I use time-fixed effects together with spatial dependence?
Yes. Time-fixed effects absorb common temporal shocks (e.g., a global recession) and are routinely included alongside the spatial weight structure. Unit fixed effects absorb time-invariant unobserved heterogeneity. Both can be combined, though the spatial parameter must still be estimated via ML or GMM.
How sensitive are results to the choice of spatial weights matrix W?
Results can vary with the W specification. Best practice is to re-estimate the model under two or three alternative W definitions (e.g., queen contiguity, distance bands, k-nearest neighbors) and report whether the main substantive conclusions hold. Large changes suggest the spatial structure is poorly identified.
What software implements space-time spatial regression?
In R, the spatialreg and plm packages together support spatial panel models; the splm package provides a unified interface. Python's PySAL includes spreg for spatial econometrics. Stata 15+ offers the spxtregress command for spatial panel data.
Sources
- 1.LeSage, J. P., & Pace, R. K. (2009). Introduction to Spatial Econometrics. CRC Press / Taylor & Francis.ISBN 978-1420064247
- 2.Anselin, L., Le Gallo, J., & Jayet, H. (2008). Spatial Panel Econometrics. In L. Matyas & P. Sevestre (Eds.), The Econometrics of Panel Data (pp. 625-660). Springer.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Space-Time Spatial Regression. ScholarGate. https://scholargate.app/spatial-analysis/space-time-spatial-regression