Panel Multiscale Geographically Weighted Regression (Panel MGWR)
Panel Multiscale Geographically Weighted Regression · Also known as: Panel MGWR, MGWR panel data, multiscale GWR panel, panel spatially varying coefficient model
Panel MGWR extends Multiscale Geographically Weighted Regression to repeated-observations (panel) data, allowing each predictor to operate at its own spatial bandwidth while controlling for unit-specific or time-specific fixed effects. It is used when both spatial heterogeneity and temporal structure matter simultaneously.
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 MGWR when you have georeferenced observations measured over multiple time periods and you expect the effect of at least one predictor to vary across space at potentially different scales. It is appropriate for regional science, urban economics, epidemiology, and environmental studies where both where and when matter. Do not use it when your data are cross-sectional only (use standard MGWR instead), when sample sizes per location and time period are very small (local estimation breaks down), when relationships are unlikely to be spatially non-stationary (a global panel model is simpler and more efficient), or when computation time is prohibitive given the iterative back-fitting procedure.
Strengths & limitations
- Allows each predictor to operate at its own spatial scale, revealing which effects are global versus locally heterogeneous.
- Controls for unit- and time-specific fixed effects, preventing stable omitted variables from confounding the spatially varying estimates.
- Produces a rich set of local coefficient maps that make spatial heterogeneity visually and analytically accessible.
- More parsimonious than standard GWR when some predictors are truly global, as their bandwidth converges to the full study area.
- Combines the interpretability of panel regression with the spatial flexibility of geographically weighted methods.
- Computationally intensive due to iterative back-fitting across multiple bandwidths; large panels with many predictors can be very slow.
- Requires sufficient spatial replication — if observations are sparse or unevenly distributed, local estimates in data-thin areas are unreliable.
- Inference methods are still maturing; standard errors and significance tests depend on asymptotic approximations that may be poor in small samples.
- Assumes that relationships are stationary in time within the panel window, or that temporal change is fully captured by time fixed effects.
- Results are sensitive to the choice of kernel function and bandwidth selection criterion.
Frequently asked
How does Panel MGWR differ from standard panel GWR?
Standard panel GWR uses a single bandwidth for all predictors. MGWR extends this by estimating a separate optimal bandwidth per predictor via back-fitting, so some effects can be nearly global while others are hyper-local. The panel component (fixed effects) is the same in both.
How do I choose between unit fixed effects and time fixed effects?
Unit fixed effects control for time-invariant spatial confounders (e.g., geography, culture). Time fixed effects absorb common shocks affecting all units in a given period. Including both (two-way fixed effects) is safest when you have both concerns, though it requires sufficient within-unit and within-period variation.
What software can run Panel MGWR?
The MGWR 2.x Python package (pysal/mgwr) implements standard MGWR with adaptive bandwidths. Full panel extensions with fixed effects are available in custom R and Python scripts drawing on the spdep and plm ecosystems; no single turnkey package covers all panel MGWR variants as of 2024.
What sample size is recommended?
As a rule of thumb, you need enough observations per bandwidth kernel to fit the local model stably — at least 30-50 spatially proximate observations per unit. For panel data, more time periods increase precision of the fixed effects but do not substitute for spatial density.
Should I test for spatial autocorrelation before using Panel MGWR?
Yes. A global Moran's I test on OLS or panel-OLS residuals confirms whether spatial structure exists and motivates the spatially varying approach. Post-estimation, testing residuals from Panel MGWR with Moran's I checks whether the model has adequately absorbed spatial non-stationarity.
Sources
- Fotheringham, A. S., Yang, W., & Kang, W. (2017). Multiscale Geographically Weighted Regression (MGWR). Annals of the American Association of Geographers, 107(6), 1247-1265. DOI: 10.1080/24694452.2017.1352480 ↗
- 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 Multiscale Geographically Weighted Regression. ScholarGate. https://scholargate.app/en/spatial-analysis/panel-multiscale-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 Durbin ModelSpatial analysis↔ compare
- Panel Spatial Error ModelSpatial analysis↔ compare