SKATER Additive Learner (STAKERAL): A Data-driven Method for Delineating Covariate-specific Endogenous Spatial Regimes

Understanding the heterogeneity of spatial processes is essential for spatial analysis, as existing methods are often limited in model complexity, flexibility and interpretability. In this paper, we proposed SKATER Additive Learner (SKATERAL), a method for delineating covariate-specific endogenous spatial regimes. SKATERAL extends the recently introduced SKATER regression framework by employing a back-fitting algorithm to optimise spatial regime delineation for each covariate. By modelling the spatial inequalities in self-rated health in Greater Manchester, we demonstrated the superiority of SKATERAL in multiple aspects, over two classic spatially varying coefficient models, Geographical Gaussian Process Generalised Additive Models and Multiscale Geographically Weighted Regression.