Funded
by the US National Science Foundation |
||
|
||
Monthly auto theft hot spots in Los Angeles (LAPD jurisdiction) January to August 2003. |
||
 |
||
Funded
by the Human Social Dynamics Program at NSF, the UC MaSC Project centers
on theoretical, methodological and empirical work to develop analytical
and computational models of crime pattern formation. Crime mapping forms
a key feature of current approaches to understanding offender behavior
and is a tool used increasingly by police departments and policy makers
for strategic crime prevention. However, despite the availability of
sophisticated digital mapping and analysis tools there is a substantial
gap in our understanding of how low-level behaviors of offenders lead
to aggregate crime patterns such as crime hot spots. Thus, for example,
we are unable to specify exactly why directed police action at crime
hot spots sometimes leads to displacement of crime in space but, surprisingly,
often can also lead to hot spot dissipation and a real reduction in
crime incidences. Agent-based modeling offers a potential avenue for
developing a quantitative understanding of crime hot spot formation
built from the bottom-up around offender behavior. Agent-based models
are not only more consistent with the scale of decisions that offenders
actually take, but they also open the door to the development of custom
statistics that are designed to answer specific behavioral questions
less tractable in general statistical models. However, there is also
concern that agent-based simulations can lead to erroneous results either
because of poor model design or errors in model implementation that
go undetected. A solution to this problem is to design simulations around
well-studied analytical models where the model behavior can be tested
against sound analytical expectations. Only following such testing should
simulation models be extended into areas that cannot be treated analytically
and, only subsequent to this, into applied contexts. The UC MaSC Project
has four components. 1. Drawing on methods in statistical physics and
the mathematics of swarms, we are developing formal models of offender
movement and target selection in variously structured environments.
2. We plan to extend these baseline models to consider offender behavior
on abstract urban street networks. 3. We will then integrate both model
types with Geographic Information Systems (GIS) by exploring the spatial
properties of simulated crime maps. 4. At each stage of model development,
empirical tests will be conducted against spatial crime data provided
by the Los Angeles, San Diego and Long Beach police departments. We
will concentrate empirical testing on comparing simulated crime prevention
interventions with known changes in urban planning and policing strategies
within these southern Californian cities. Simulated offender movement and abstract mapping of crime locations. (a) Simulated offender following a Lévy mobility strategy. Lévy mobility shows clusters of short distance flights interspersed with longer distance flights. (b) Mapped crime locations assuming that criminal opportunities are uniformly distributed in space and that offenses occur only at the end points of Lévy flights. (c) Inverse distance weighted interpolation of offense locations showing crime hot spots generated by the underlying Lévy mobility strategy.
Simulation of spontaneous burglary hotspot nucleation. Panels on the left show the cumulative distribution of locations visited in a two-dimensional plane. Panels on the right show the cumulative distribution of burgled houses. Simulation code written by Jon Azose, a Harvey Mudd undergraduate Math student participating in the NSF Harvey Mudd-UCLA Research Training Group program run by Andy Bernoff. |
||
