Building a Crime Prediction Model: Distinguishing between fast and slow dynamics over various contexts

• In Event: People, Places, and Time: Consequences for Crime

Abstract

Existing crime prediction models typically focus on either slow dynamics (e.g., hot spots) or fast dynamics (e.g., repeat events). We attempt to link these two approaches with a statistical model attempting to predict crime that occurs at particular time points at particular locations. We break the data into 4-hour time periods within blocks over the entire year. We estimate the model for one year, and then use the coefficients to predict crime locations and time periods in the following year. We compare the results of our model to that of seven other prediction models based on the number of crime events actually observed in the highest risk time periods. We use data from two different cities and show that the relative importance of fast versus slow dynamics differs over these two contexts, highlighting that it may not be straightforward for prediction models to generalize to all environments.

Authors

• John R. Hipp, University of California, Irvine

• Christopher Jay Bates, University of California, Irvine