Data-driven policing helps predict when, where crimes will occur
Retailers have mastered the art of predicting what their customers are likely to buy. Now police departments across the country are using their own predictive strategies such as algorithms, time/space analysis, and social network analysis to become "data detectives" in an effort to stop crime before it starts, according a new report from the Johns Hopkins University Center for Advanced Governmental Studies.
The report, "Predictive Policing: Preventing Crime with Data and Analytics," was written by Jennifer Bachner, program coordinator and lecturer in the Center for Advanced Governmental Studies.
"Law enforcement agencies are on the frontier of the data revolution," Bachner said. "Predictive policing is a part of intelligence-led, proactive policing that is focused on what is likely to occur rather than what has already happened."
Bachner studied three early adopters of what is known as predictive policing. They include police departments in Baltimore County, Md., Richmond, Va., and Santa Cruz, Calif. Santa Cruz officials credit the predictive police work for the department's 27 percent drop in burglaries from 2010 to 2011.
Law enforcement agencies have been collecting and crunching statistics for years, using it to map and track criminal activity. What's different about predictive policing, Bachner said, is police are using data analysis to prevent crime from happening at all.
The report points to three main ways in which police use predictive policing:
A number of police departments now supplement their investigative work with computer algorithms that can determine high-risk areas by looking at everything from where criminals live to the availability of escape routes to areas where potential victims might gather, such as tourist venues. Santa Cruz, one of the first cities in the nation to embrace predictive policing, partnered with academic researchers to develop software that determined the city's 15 most likely areas to experience crime.
Police departments also use time and space analysis—including factors such as the weather, time of day, and whether it's payday—to forecast crime sequences. For instance, in warmer weather, people generally stay outside and away from home for extended periods, making their homes targets for burglars. Baltimore County used this strategy to catch robbery suspects. In one instance several years ago, police were able to use directional relationship mapping to pinpoint the next likely grocery store robbery after a string of similar robberies.
A third strategy centers on analyzing social networks. Using this tactic, police can identify a suspect's social network and map how information flows between various people. Richmond police forced a suspect to turn himself in after cutting off his social network. After police had been searching for a homicide suspect for a month, analysts built a network of his family and friends. Officers then asked these individuals to notify them if the suspect made contact. The suspect turned himself in to police within hours.
"We can use data from a wide variety of sources to compute estimates about phenomena such as where gun violence is likely to occur, where a serial burglar is likely to commit his next crime, and which individuals a suspect is likely to contact for help," Bachner said. "We can estimate the probability that a car will be stolen, for example, if we know the location of the car, characteristics of the car, and the time of day."
Bachner concluded that predictive policing is a cost-effective way to fight crime, but not a replacement for traditional police work.
"Data-driven data programs complement but do not replace cops on beats," she said. "Still, there's evidence here of real cost savings."