2026/07/01 by Martin A. Andresen
Social Sciences · Economics, Econometrics and Finance · #Crime Patterns and Interventions #Spatial and Panel Data Analysis #Geographic Information Systems Studies
paper · doi:10.1016/j.jcrimjus.2026.102703
Averaging crime data across multiple years is a common practice in spatial criminology, typically motivated by the desire to smooth aberrant annual observations and recover a more stable estimate of the underlying spatial distribution of crime. This practice is theoretically justified when the spatial data generating process is stable across the years being averaged, but may lead to errors in measurement. Specifically, under such conditions, averaging reduces variance around the true underlying spatial pattern. This study examines whether that assumption holds in practice and, if so, whether the predicted benefits of averaging materialise. Spatial stability is assessed pairwise across 23 years of crime data (2003–2025) in Vancouver, British Columbia with predictive consequences estimated through a bootstrap simulation across eight crime types, two spatial scales (census tracts and dissemination areas, common units of analysis, particularly in spatial crime analysis), seven training years (2018–2024), using six regression models. Where spatial patterns are stable, averaging produces near-zero average effects on prediction error with large unpredictable variability across years. As such, the theoretical variance reduction does not materialise in practice. Where spatial patterns are unstable, the theoretical justification for averaging does not hold, and the effects on prediction error are large and unpredictable in direction. Consequently, averaging may substantially improve or substantially harm prediction depending on crime type and year.