Space-Time Spatial Error Model
Also known as: SEM panel, spatial error panel model, space-time SEM, spatiotemporal error model
The Space-Time Spatial Error Model (space-time SEM) is a spatial panel regression technique that accounts for spatial dependence confined to the error term across geographic units and time periods. It corrects biased inference caused by spatially correlated disturbances while estimating covariate effects on a panel of spatial observations.
Key highlights
- Corrects biased standard errors and invalid inference caused by spatially correlated disturbances in panel data.
- Separates spatial error autocorrelation from time-invariant unit heterogeneity when unit fixed effects are included.
- Covariate coefficients (beta) retain a direct, interpretable meaning — the spatial error term is a nuisance correction, not a substantive spillover.
- Well-supported by established ML and GMM estimators in software packages (R spdep/spatialreg, Stata xsmle, Python PySAL).
- Can be extended with two-way fixed effects to additionally control for common temporal shocks.
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 the space-time SEM when you have panel data spanning multiple regions and time periods, and Lagrange Multiplier tests or Moran's I on OLS residuals indicate spatial error dependence — that is, neighboring units share omitted influences rather than directly affecting each other's outcomes. It is appropriate when the theoretical mechanism is contamination of errors (e.g., common local shocks, spillovers in unobservables) rather than peer effects on the dependent variable itself. Do not use it when the substantive interest is the magnitude of cross-unit spillovers in the outcome — prefer the spatial lag or spatial Durbin panel model in that case. It is also less suitable when neighborhood relationships change substantially over time, when the weights matrix is poorly defined, or when the panel is very short (few time periods).
Strengths & limitations
- Corrects biased standard errors and invalid inference caused by spatially correlated disturbances in panel data.
- Separates spatial error autocorrelation from time-invariant unit heterogeneity when unit fixed effects are included.
- Covariate coefficients (beta) retain a direct, interpretable meaning — the spatial error term is a nuisance correction, not a substantive spillover.
- Well-supported by established ML and GMM estimators in software packages (R spdep/spatialreg, Stata xsmle, Python PySAL).
- Can be extended with two-way fixed effects to additionally control for common temporal shocks.
- Choice of spatial weights matrix W is subjective and results can be sensitive to this specification.
- Does not model substantive spatial spillovers in the dependent variable — if peer effects are the interest, the spatial lag model is more appropriate.
- Maximum likelihood estimation can be computationally intensive for large N or T.
- Requires stationarity assumptions in the spatial error process; non-stationarity in space or time complicates inference.
- The lambda parameter captures aggregate spatial error dependence but does not vary across units or time.
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 is the space-time SEM different from the space-time spatial lag model?
In the spatial lag panel model the spatially lagged dependent variable W*y enters the right-hand side, capturing direct spillovers in the outcome. In the space-time SEM the spatial dependence is entirely in the error term, meaning neighboring units share unobserved influences rather than directly affecting each other's outcomes. The Lagrange Multiplier test helps select between them.
How do I choose the spatial weights matrix W?
Common choices are queen or rook contiguity (binary adjacency), inverse-distance weighting, or k-nearest neighbors. Theory should guide the choice — e.g., contiguity for administrative spillovers, distance for economic or environmental diffusion. Always row-standardize W and compare results under alternative specifications as a robustness check.
Can I include both unit and time fixed effects in the space-time SEM?
Yes. Two-way fixed effects (individual and time dummies) control for time-invariant unobserved heterogeneity and common temporal shocks, respectively, before estimating the spatial error parameter lambda. This is typically recommended when both sources of omitted variation are plausible.
What does a significant lambda mean?
A significant and positive lambda indicates that the unobserved factors driving the outcome are spatially clustered — neighboring units tend to have correlated errors. It does not imply that unit i's outcome directly depends on unit j's outcome; that would require the spatial lag model.
Which software packages support the space-time SEM?
In R the spatialreg and plm packages (and the SDPDmod package for spatial panel) support the space-time SEM. In Stata, xsmle provides ML estimation. Python users can use PySAL's spreg module. All require pre-computing the spatial weights matrix W.
Sources
- 1.Anselin, L. (1988). Spatial Econometrics: Methods and Models. Kluwer Academic Publishers.ISBN 978-9024737247
- 2.Elhorst, J. P. (2014). Spatial Econometrics: From Cross-Sectional Data to Spatial Panels. Springer.ISBN 978-3642403392
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Space-Time Spatial Error Model. ScholarGate. https://scholargate.app/spatial-analysis/space-time-spatial-error-model