Measuring the Unmeasurable: Models of Geographical Context

Abstract

The issue of whether place significantly affects spatial behavior has long created both a philosophical and an operational schism within geography. Here we show how these schisms can be bridged by identifying how place and behavior can be linked through recognizing and incorporating what we term intrinsic and behavioral contextual effects into models of spatial behavior. We argue that spatial modeling frameworks that attempt to relate spatial behavior to aspects of people and places might be seriously misspecified if they do not incorporate both types of contextual effects. We compare three popular statistical modeling frameworks that encompass placed-based contextual effects: Spatial error models, multilevel models, and multiscale geographically weighted regression (MGWR). Based on Monte Carlo simulation and empirical analysis, we demonstrate the reassuring similarity of the results from the three frameworks but also the superiority of MGWR. The inclusion of essentially unmeasurable effects within a nomothetic framework provides an important bridge between two previously distinct philosophies within geography and acts as a binding force within the discipline.