Expert Commentary

The effectiveness of predictive policing: Lessons from a randomized controlled trial

2014 study from the RAND Corporation analyzes the results of a real-world trial of data-driven, predictive policing, assessing its budget impacts and ability to reduce property crime.

A preliminary report from the FBI indicates that during the first half of 2013, violent crime dropped 5.4% in the United States relative to the same period in 2012. The drop was consistent for cities both small and large — from under 10,000 inhabitants to more than 1 million — and across the four U.S. geographic regions. The reduction was smaller than those for 2008-2009 (-6.1%), 2009-2010 (-6.2%) and 2010-2011 (-6.4%), but it is a substantial improvement over the 1.9% increase in violent crimes from 2011 to 2012.

The overall decrease in violent crime is reflective of a long-term trend: Since 1991, the rates of many crimes in the United States have fallen significantly — with the lamentable exception of mass shootings — and just what the driving factors are is the subject of considerable debate. Everything from order-maintenance policing and tough sentencing regimes to community-based policing and even the phasing out of leaded gasoline have been suggested, but the research hasn’t pointed to a single definitive cause.

Whatever the roots of the decline, the challenge of creating effective and accountable police departments remains, particularly after the 2014 events in Ferguson, Missouri. Given recent improvements in computer power and data gathering, law-enforcement agencies are exploring the possibility of “predictive policing.” According to the National Institute of Justice, this method “harnesses the power of information, geospatial technologies and evidence-based interventions” to enable a proactive and anticipatory approach to policing. It has some similarities to the “hot spots” strategy, and leverages information technology in an attempt to anticipate where crimes might be likely to occur.

A 2014 study published by the RAND Corporation, “Evaluation of the Shreveport Predictive Policing Experiment,” looks at the potential impact of predictive policing on property crime and its budget effects. The researchers, Priscillia Hunt, Jessica Saunders and John Hollywood, studied the implementation of a 2012 pilot project in six districts of the police department in Shreveport, La. The study was structured so that it had a high level of internal validity: It is a randomized control trial — three districts used the new method, while three continued with the status quo — and also used a wide array of information sources, including the location of previous crimes, 911 calls, field interviewers and previously targeted hot spots.

The Shreveport program, called Predictive Intelligence-Led Operational Targeting (PILOT), focused on specific locations that were likely to see property crimes: “There are two overarching activities of PILOT: (1) using a predictive model to identify small areas at increased risk of property crime, and (2) implementing a prevention model to conduct policing interventions in the areas assessed to be at increased risk.”

Key study results include:

  • There is no statistical evidence to support the claim that crime was reduced more in the experimental districts than in the control districts. This is attributed to (a) the low statistical power of the test resulting from the duration of the pilot in the participating districts; and (b) the heterogeneity in the implementation of predictive policing across treatment districts.
  • Districts that implemented predictive policing spent between 6% and 10% less than control districts: “Results show that PILOT cost approximately $13,000 to $20,000 less than the status quo … mainly due to the number of hours that patrol officers in particular worked on property crime-related special operations.”
  • Relative to the control groups, reduced officer overtime allowed the police department to offset the cost of the PILOT program.
  • Community relations improved overall in the treatment districts.

Despite the inconclusive results on crime reduction, the authors state that predictive policing may still be a helpful tool, in combination with other policing strategies: “Predictive policing is not an end-all solution, but rather a tool that must be used in concert with other policing resources as part of a broader anti-crime effort.”

Related research: A 2013 study in Urban Affairs Review, “The Postindustrial City Thesis and Rival Explanations of Heightened Order Maintenance Policing,” examines the relationship between an area’s economy and its policing style. The author, Elaine B. Sharpe of the University of Kansas, looks at 180 cities with a population of 100,000 or more; she analyzes arrest rates and charges, governing institutions, policing demands and constraints, and variables representing the “racial threat” thesis — that the increased presence of minorities can trigger intolerance and greater attempts at social control.

Keywords: police, law, crime, big data, crime and violence prevention, law enforcement

About The Author