M‐ SGWR : Multiscale Attribute‐Similarity and Geographically Weighted Regression

ABSTRACT The first law of geography is a cornerstone of spatial analysis, emphasizing that nearby locations tend to be more strongly related and similar. However, defining what constitutes “near” and “related” remains challenging because different phenomena operate across distinct spatial dimensions, including physical, social, and network‐based spaces. Conventional local regression models, such as GWR and MGWR, represent nearness and relatedness primarily through geographic proximity. In an era of globalization and digital connectivity, however, geographic proximity alone may not adequately characterize relationships among locations, as many phenomena are shaped by connections that extend beyond physical distance. To address this limitation, we propose a new model, termed M‐SGWR, that characterizes relationships among locations in two dimensions: geographic space and predictor‐specific attribute space. For each predictor, geographic and attribute‐similarity weight matrices are constructed separately and combined using an optimized parameter, , which determines their relative contributions to local model estimation. Analogous to the predictor‐specific bandwidths used in MGWR, the optimal varies across predictors, allowing M‐SGWR to identify geographically dominant, mixed, or attribute‐dominant relationships, including similarities among geographically distant locations. Results from two simulation experiments and one empirical application demonstrate that M‐SGWR consistently outperforms GWR, SGWR, and MGWR across the evaluated goodness‐of‐fit measures.