Randomized Controlled Field Trials of Predictive Policing

Abstract

The concentration of police resources in stable crime hotspots has proven effective in reducing crime, but the extent to which police can disrupt dynamically changing crime hotspots is unknown. Police must be able to anticipate the future location of dynamic hotspots to disrupt them. Here we report results of two randomized controlled trials of near real-time Epidemic Type Aftershock Sequence (ETAS) crime forecasting, one trial within three divisions of the Los Angeles Police Department and the other trial within two divisions of the Kent Police Department (UK). We investigate the extent to which i) ETAS models of short term crime risk outperform existing best practice of hotspot maps produced by dedicated crime analysts, ii) police officers in the field can dynamically patrol predicted hotspots given limited resources, and iii) crime can be reduced by predictive policing algorithms under realistic law enforcement resource constraints. While previous hotspot policing experiments fix treatment and control hotspots throughout the experimental period, we use a novel experimental design to * Department of Mathematics and Computer Science, Santa Clara University † School of Mathematics, Georgia Institute of Technology ‡ Los Angeles Police Department § Kent Police Service ¶ Department of Criminology, University of California, Irvine Department of Mathematics, University of California, Los Angeles * * Department of Anthropology, University of California, Los Angeles 1 allow treatment and control hotspots to change dynamically over the course of the experiment. Our results show that ETAS models predict 1.4-2.2 times as much crime compared to a dedicated crime analyst using existing criminal intelligence and hotspot mapping practice. Police patrols using ETAS forecasts led to a average 7.4% reduction in crime volume as a function of patrol time, whereas patrols based upon analyst predictions showed no significant effect. Dynamic police patrol in response to ETAS crime forecasts can disrupt opportunities for crime and lead to real crime reductions.

Unfalsifiability of security claims

“…A further justification for treating attacks probabilistically is that attacker adaptation, which complicates the question of assigning probabilities to attacks, is seldom cost-free. Although attackers with perfect knowledge and zero switching costs are hard to model, assuming realistic limitations on their abilities, knowledge, and costs makes probabilistic approaches very useful in practice (14,15).…”

Crisis analytics: big data-driven crisis response

“…Mohler et al provide an example of it in which they relate crime to earthquakes in nature, which are recurrent, and model it with Epidemic Type Aftershock Sequence (ETAS) point process. It has been shown to perform better than the traditional methods used by criminal analysts by 1.4-2.2 times (Mohler et al 2015). Similarly, concepts specially from statistical physics and network science (Barabási 2013) can be used to model the behavior of epidemics and methods of predictive policing can be deployed to fight the spread of such diseases.…”

Policing in the Era of Big Data

“…Pushing data to police can make them more effective if they use it (Casady et al 2015) but not if they never access it (Taniguchi & Gill 2013). Predictive models can produce better predictions than human crime analysts but have not yet demonstrated a crime prevention benefit (Mohler et al 2015).…”

“…The ETAS model outperformed the crime analyst by a factor of 1.4 to 2.2. Mohler et al (2015) also conducted an RCT and found that police spending 1,000 minutes in ETAS-predicted hot spots eliminated one crime on average, whereas police would need to spend 2,000 minutes in hot spots generated from LAPD's standard practice to have the same one-crime reduction. This difference is not statistically significant, but signals that some gains, although modest, are possible with predictive policing.…”

“…Mohler et al (2015) put a crime analyst and an epidemic-type aftershock sequence (ETAS) crime forecasting model in a head-to-head prediction competition. During a silent test period in which officers in the field received no predictions, the crime analyst and the ETAS model would select areas of fixed size (typically representing less than 1% of the entire patrol area) forecasted to be likely to contain a large share of future crime.…”