Degree

Doctor of Philosophy (PhD)

Department

Department of Geography & Anthropology

Document Type

Dissertation

Abstract

Agent-based crime simulation research is still at a very early stage. While there were efforts in the past to demonstrate the possibility to combine environmental criminology with agent-based modeling, they were limited to simple test environments. This dissertation develops an agent-based crime simulation from routine activity theory and tests this model in a real urban environment with a complex road network. Its objectives are three-fold: 1) To build an ABM crime simulation with realistic individual-level routine activities in a large-scale urban environment; 2) To identify agent behaviors and place features that have significant impacts on crime hotspot predictions; 3) To demonstrate its use to inform the public of crime forecast.

The result shows the final model can, with the help of CTPP data and time geography, simulate detailed daily routines of thousands of agents in a complex urban environment at reasonable speed and robberies as the result of their interactions under given circumstances. This environment is composed of a high-resolution road network, a large number of network places, and multiple 500-meter resolution grids for a variety of neighborhood conditions. The performance and flexibility of the final model make it a promising tool to simulate crime in other cities.

Validation with the reported robberies in January 2011 shows the vulnerability of agents has the most significant effect on global crime hotspot pattern. Model consistently shows better hotspot prediction ability when non-police agents are considered vulnerable at any activity state. Whether this indicates that neighborhood and work environment do not provide the same level of protection everywhere remains unclear and needs more investigation. Offenders’ motivation change pattern, target-search radius, and risk-taking preference were related to local and temporal distribution of crime. In comparison, the crime history of a neighborhood had a big impact on both robbery count and hotspot prediction. This corroborates the existence of crime places that are more attractive to offenders.

Significant correlation was found between simulated and reported robbery counts at the ZIP code level with 20 out of 21 different parameter sets. Although the raw count from simulation is not particularly meaningful, the final model can still help policy makers and the public to understand the trend of local crime development by either ranking the ZIP code with simulated count or reclassifying raw count into several levels of severity.

Committee Chair

Wang, Fahui

Available for download on Wednesday, January 06, 2027

Share

COinS