Panel Spatial Durbin Model
Panel Data Spatial Durbin Model · Also known as: SDM panel, spatial Durbin panel model, panel SDM, PSDM
The Panel Spatial Durbin Model (PSDM) extends the cross-sectional Spatial Durbin Model to panel data, capturing both spatial lag dependence in the outcome and spatial spillovers from neighbouring units' explanatory variables across multiple time periods. It simultaneously accounts for unobserved unit-specific and time-specific heterogeneity, making it one of the most comprehensive spatial panel specifications available.
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 the Panel Spatial Durbin Model when your panel data exhibit both outcome-level spatial dependence (neighbours' output affects your output) and predictor-level spillovers (neighbours' covariates affect your outcome), and when ignoring either channel would bias estimates. It is the preferred starting point in exploratory spatial panel work because it nests the Spatial Lag and Spatial Error models as special cases, enabling model selection via standard tests. Prefer it when theory suggests cross-unit externalities — e.g., regional economics, epidemiology, environmental policy. Avoid it when the spatial weights matrix W is poorly justified, when n (number of units) is very small (fewer than 30), or when the panel is very short (fewer than 3 time periods), as parameter identification becomes fragile.
Strengths & limitations
- Nests both the Spatial Lag Model and the Spatial Error Model, allowing formal LR/Wald tests to select the appropriate specification.
- Explicitly decomposes total effects into direct (own-unit) and indirect (spillover) components, which is essential for policy evaluation.
- Simultaneously controls for unit-specific and time-period heterogeneity via fixed or random effects, reducing omitted-variable bias.
- Handles two-way spatial dependence: in the outcome variable (rho) and in the predictors (theta), giving a richer picture of spatial dynamics.
- Applicable to a wide range of georeferenced panel datasets in economics, public health, environmental science, and regional planning.
- Requires a pre-specified spatial weights matrix W; results can be sensitive to the chosen neighbourhood definition.
- Computationally intensive for large N, as ML estimation involves repeated computation of the determinant |I - rho*W|.
- With many predictors, the spatially lagged counterparts W*X_it double the covariate set, raising multicollinearity concerns.
- Fixed-effects transformation (within estimator) removes time-invariant variables, preventing estimation of their effects.
- Interpretation of indirect effects is non-trivial and often overlooked in applied work, leading to under-reporting of spillovers.
Frequently asked
How does the Panel SDM differ from the Panel Spatial Lag Model?
The Panel Spatial Lag Model includes only the spatially lagged dependent variable (W*y), while the Panel SDM additionally includes spatially lagged predictors (W*X). The SDM is therefore more general and can capture spillovers operating through covariates rather than only through the outcome itself.
Should I use fixed effects or random effects in a Panel SDM?
A Hausman test adapted for spatial panels helps decide. Fixed effects are preferred when unit-specific heterogeneity is correlated with the regressors — which is common in regional economics. Random effects are more efficient if that correlation is absent but impose stronger distributional assumptions.
What are direct and indirect effects, and why do they matter?
Because spatial lags create feedback loops, a change in one unit's predictor ripples through the system. The direct effect is the average impact on the same unit (including own feedback), and the indirect effect is the average spillover to all other units. Summing them gives the total effect. Reporting only raw coefficients misses these spatial multiplier dynamics.
How do I choose the spatial weights matrix W?
W should reflect the mechanism driving spatial dependence: a contiguity matrix suits border-sharing effects; an inverse-distance matrix suits diffusion processes; a k-nearest-neighbours matrix works when geographic contiguity is unclear. The choice should be motivated by theory before looking at the data.
Can I include time-invariant variables in a Panel SDM with fixed effects?
No. The within (demeaning) transformation used to eliminate unit fixed effects also removes any variable that does not vary over time. If time-invariant predictors are theoretically important, use random effects or a correlated random effects specification instead.
Sources
- Elhorst, J. P. (2014). Spatial Econometrics: From Cross-Sectional Data to Spatial Panels. Springer. ISBN: 978-3642403408
- LeSage, J. P., & Pace, R. K. (2009). Introduction to Spatial Econometrics. CRC Press/Taylor & Francis. ISBN: 978-1420064247
How to cite this page
ScholarGate. (2026, June 3). Panel Data Spatial Durbin Model. ScholarGate. https://scholargate.app/en/spatial-analysis/panel-spatial-durbin-model
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.
- Geographically Weighted RegressionSpatial analysis↔ compare
- Panel Spatial Error ModelSpatial analysis↔ compare
- Panel Spatial RegressionSpatial analysis↔ compare
- Spatial Durbin ModelSpatial analysis↔ compare
- Spatial Lag ModelSpatial analysis↔ compare