A surveillance modeling and ecological analysis of urban residential crimes in Columbus, Ohio, using Bayesian hierarchical data analysis and new space-time surveillance methodology

This dissertation analyzes crime in both ecological and surveillance perspectives. In ecological perspective, many studies ignore spatial effects in the models, leading to inefficient and biased results. This dissertation, by applying Bayesian hierarchical analysis, accounts for spatial effect in the model and presents correct socio-demographic factors related to residential crime occurrences. In surveillance perspective, literature to date has limitations in presenting exact locations of crime hotspots and implementing continuous analysis over time. Use of population information is the main reason of the limitations in the literature. Because the population information is based on census administrative area unit and is only updated in decennial bases, corresponding hotspots involve approximations in both population size and their locations. However, this study handles the problem by applying a newly devised surveillance method, which uses only crime accounts over time without the use of population information. The goal of this dissertation is providing significant demographic factors of crime and crime hotspots in near real time base, which will contribute to crime control. This goal is achieved by (1) handling spatial autocorrelation and heterogeneity in the analysis, (2) visualizing spatial effects on a map, (3) enabling continuous surveillance over time, (4) providing precise crime hotspot locations, and (5) presenting local changes in clusters over time. The models presented in this dissertation is applied to residential crimes occurred in Columbus, Ohio for the year 2000. Empirical results present significant demographic factors of residential crimes and locations of crime hotspots over time in near real-time framework.; Keyword. hierarchical Bayesian data analysis, surveillance, space-time surveillance, crime.