Measuring Spatial Context: A Generalized Additive Spatial Smoothing Approach with Applications to Crime and Migration

Abstract

Spatial contextual effects research rests on the premise that one’s surroundings shape experience and outcomes, yet researchers routinely fail to identify the spatial scale at which contextual processes operate, substituting administrative units for the actual extent of those processes without empirical justification. Current practice exhibits two failures: (1) treating administrative units and their characteristics as “context” rather than bespoke surroundings; or (2) using administrative units as an arbitrarily fixed scale for change-of-support. This dissertation argues both stem from a fundamental Change of Support Problem (COSP): when the geographic support of available data misaligns with that of the underlying process, systematic bias arises. The dissertation adopts three principles for measuring spatial context: egocentricity (support constructed outward from the observation, not inherited from an external classification); fuzziness (surrounding values weighted by spatial proximity); and multiscale (each process–outcome relationship has its own operative extent). Current methods seldom apply all three. Generalized Additive Spatial Smoothing (GASS) operationalizes them through data-driven empirical scale estimation. A simulation study demonstrates that conventional change-of-support operations produce systematic coefficient attenuation through scale misspecification and information loss and validates that GASS substantially reduces bias when spatial information is preserved in predictor data. It shows that MAUP scale effects, ecological fallacy, and the mismeasurement of spatial context share common mechanisms arising from support mismatch. Two empirical applications follow. A crime application tests fuzziness and estimates policy-relevant scales of built-environment effects on intimate partner violence, revealing distinct scales across features and suggesting that some prior theoretical discrepancies may be measurement artifacts. A migration application extends the inquiry to spatial interaction models: the Competing Destinations and Intervening Opportunities frameworks invoke fuzzy origin and destination neighborhoods, yet researchers conventionally fix their scales. Using GASS to estimate factor-specific scales, analysis of U.S. county-to-county migration reveals that origin- and destination-based structures operate at empirically distinct scales, validating the multiscale principle and establishing that scale is mechanism-specific. Together, these studies establish that measuring spatial context is a modeling choice with theoretical consequences, advancing spatial methodology with a unifying cross-support framework and tool, informing crime and migration research, and clarifying for GIScience how spatial representation affects validity.