Generalized Local Additive Spatial Smoothing (GLASS): A New Approach to Mitigating the Modifiable Areal Unit Problem in Local Regression Modeling

Addressing spatial non-stationarity and the Modifiable Areal Unit Problem (MAUP) in regression models is essential for accurate spatial analysis. This study extends the Generalized Local Additive Spatial Smoothing (GLASS) framework, a technique that leverages spatial smoothing to adjust the spatial support of covariates to a specified target within local modeling. By doing so, GLASS mitigates MAUP-related biases that arise from data aggregation that could distort the inference of local processes. Unlike traditional local regression approaches, which assume consistent spatial support across covariates and often require prior aggregation before model calibration, GLASS endogenously aligns differing supports during estimation. A multivariate extension is introduced that integrates adaptive smoothing and variable-specific bandwidth selection through a nested iterative backfitting procedure. Simulation experiments and an empirical case study modeling diabetes prevalence in the Washington D.C.– Maryland–Virginia region demonstrates that GLASS outperforms Geographically Weighted Regression (GWR) and Multiscale GWR, offering a promising framework for analysing spatially heterogeneous relationships across misaligned supports.