Agent-Based Simulation of Near Repeat Patterns

Abstract

In space-time-windows around a previous victimization, the risk for a new victimization will quite often be larger than it would have been when no previous victimization had taken place close by in space and time. While often demonstrated empirically, the underlying explanation of such Near Repeat patterns is unclear. We discuss several mechanisms that might be causing the pattern. Testing these mechanisms is difficult, however, as near repeat analysis is usually done on police victimization data. Many such crimes will never be solved, and therefore no offender(s) become known, which makes it impossible to decide whether the same offender has actually committed a pair of crimes. We therefore use an Agent-Based Model to operationalize the mechanisms and test whether they are sufficient to produce Near Repeat patterns. We simulate a world according to main propositions of environmental criminology, i.e. inhabited by potential offenders who have dynamic awareness spaces and targets with heterogeneous suitability and various spatial patterns. Finally, to compare the outputs of the simulations to Near Repeat patterns discussed in empirical studies (often in the form of a Knox table), we propose a new way to quantify ‘near repeat’.

Wouter Steenbeek

Senior Researcher

My research interests include spatio-temporal patterns of crime, offender decision-making, neighborhoods, machine learning, and R.