Spatial association effects, perhaps the most important concern in the analysis of spatial data, have been amply studied from a global perspective in the exploratory and modeling domains, and more recently also from a local perspective in the realm of exploratory data analysis. In a local modeling framework, however, the issue of how to detect and model spatial association by using geographically weighted regression (GWR) remains largely unresolved. In this paper we exploit a recent development that casts GWR as a model of locational heterogeneity, to formulate a general model of spatial effects that includes as special cases GWR with a spatially lagged objective variable and GWR with spatial error autocorrelation. The approach also permits the derivation of formal tests against several forms of model misspecification, including locational heterogeneity in global models, and spatial error autocorrelation in GWR models. Application of these results is exemplified with a case study.