Panel Geographically Weighted Regression (Panel GWR)
Panel Geographically Weighted Regression · Also known as: Panel GWR, PGWR, spatiotemporal GWR, geographically weighted panel regression
Panel Geographically Weighted Regression (Panel GWR) extends the standard GWR framework to panel data, allowing regression coefficients to vary both across geographic locations and over time. It captures spatially non-stationary relationships in longitudinal or repeated-measures spatial datasets, combining local spatial estimation with panel-data controls for unit-specific heterogeneity.
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 Panel GWR when you have georeferenced data observed over multiple time periods and you suspect that the relationship between predictors and the outcome is not spatially uniform. It is well-suited to problems in regional economics, public health, environmental science, or urban studies where local context drives heterogeneous effects. It is appropriate when the sample includes a reasonable number of spatial units (typically 50+) with repeated observations. Do not use Panel GWR when the assumption of spatial non-stationarity is not supported — if relationships are globally constant, standard panel regression is more efficient and easier to interpret. Also avoid it with very sparse spatial coverage, collinear predictors, or when the number of time periods is very small, as local estimation requires sufficient local data density for stable coefficient estimates.
Strengths & limitations
- Captures spatially varying coefficients, revealing where relationships are stronger or weaker across the study area.
- Handles panel data structure, controlling for unobserved unit-level heterogeneity through fixed or random effects.
- Flexible bandwidth selection adapts to data density, making the method applicable to unevenly distributed observations.
- Produces interpretable maps of locally estimated coefficients, facilitating spatial diagnosis and policy targeting.
- Extends the standard GWR framework without discarding the temporal dimension of longitudinal spatial data.
- Computationally demanding: local estimation is repeated at every spatial unit across all time periods, requiring substantial resources for large datasets.
- Bandwidth selection via cross-validation or AICc can be sensitive to the choice of kernel and may not converge in poorly structured data.
- Locally estimated coefficients can be unstable when local sample sizes are small or when predictors are multicollinear.
- Interpretation is more complex than global panel models; spatially varying coefficients must be mapped and examined rather than reported as single summary statistics.
- No universally agreed-upon formal test for spatial non-stationarity in the panel setting makes it difficult to justify the added complexity over a standard panel model.
Frequently asked
How is Panel GWR different from standard GWR?
Standard GWR is typically applied to a single cross-section of spatial data. Panel GWR extends this to multiple time periods, adding a panel dimension. This allows the model to control for time-invariant unit heterogeneity via fixed or random effects, and it can capture how locally varying coefficients evolve over time, which standard cross-sectional GWR cannot.
How do I choose between a fixed-bandwidth and an adaptive bandwidth?
Adaptive bandwidths are generally preferred when spatial units are unevenly distributed across the study area, as they automatically include more neighbours in sparse regions and fewer in dense ones. Fixed bandwidths assume uniform data density. Use cross-validation or AICc to compare alternatives and select the bandwidth that minimises prediction error.
Can Panel GWR handle both spatial and temporal autocorrelation?
The spatial dimension is handled through the kernel weighting scheme. Temporal autocorrelation within units is partially addressed through the panel fixed-effects transformation, but residuals should still be tested for serial correlation. If strong temporal autocorrelation remains, consider a space-time GWR extension or add lagged terms.
What software supports Panel GWR?
GWR4 and the Python GWpy/mgwr packages implement variants of panel GWR. The R package GWmodel supports geographically weighted models including panel extensions. Some researchers implement custom routines in R or Python using the within-transformation followed by local WLS loops.
How many spatial units do I need for reliable Panel GWR estimates?
There is no universal minimum, but stable local coefficient estimates generally require at least 50 to 100 spatial units. With fewer units, local sample sizes after kernel weighting become very small, and estimates become unreliable. More time periods per unit can partially compensate, but spatial coverage remains the primary constraint.
Sources
- Fotheringham, A. S., Brunsdon, C., & Charlton, M. (2002). Geographically Weighted Regression: The Analysis of Spatially Varying Relationships. Wiley. ISBN: 978-0471496168
- Yu, H., Fotheringham, A. S., Li, Z., Oshan, T., Kang, W., & Wolf, L. J. (2020). Inference in Multiscale Geographically Weighted Regression. Geographical Analysis, 52(1), 87–106. DOI: 10.1111/gean.12189 ↗
How to cite this page
ScholarGate. (2026, June 3). Panel Geographically Weighted Regression. ScholarGate. https://scholargate.app/en/spatial-analysis/panel-geographically-weighted-regression
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
- Local Geographically Weighted RegressionSpatial analysis↔ compare
- Multiscale Geographically Weighted RegressionSpatial analysis↔ compare
- Panel Spatial Error ModelSpatial analysis↔ compare