A bootstrap based space–time surveillance model with an application to crime occurrences

This study proposes a bootstrap-based space–time surveillance model. Designed to find emerging hotspots in near-real time, the bootstrap based model is characterized by its use of past occurrence information and bootstrap permutations. Many existing space–time surveillance methods, using population at risk data to generate expected values, have resulting hotspots bounded by administrative area units and are of limited use for near-real time applications because of the population data needed. However, this study generates expected values for local hotspots from past occurrences rather than population at risk. Also, bootstrap permutations of previous occurrences are used for significant tests. Consequently, the bootstrap-based model, without the requirement of population at risk data, (1) is free from administrative area restriction, (2) enables more frequent surveillance for continuously updated registry database, and (3) is readily applicable to criminology and epidemiology surveillance. The bootstrap-based model performs better for space–time surveillance than the space–time scan statistic. This is shown by means of simulations and an application to residential crime occurrences in Columbus, OH, year 2000.