IE 11 Not Supported

For optimal browsing, we recommend Chrome, Firefox or Safari browsers.

Predictive Policing: The Promise and Perils

As analytical tools have become more sophisticated and data sets much larger, the ability to forecast crime has grown more nuanced.

It should come as no surprise that the man who introduced computer-driven performance management — known originally as CompStat — to policing, is generally credited with envisioning how predictive policing could help fight crime. New York City’s once and current Police Chief William Bratton saw predictive analytics as a way to anticipate gang violence, burglaries and thefts when he was chief of police in Los Angeles back in 2008. In 2011, the L.A. police used predictive analysis to cut property crime by 12 percent in one neighborhood. Bratton suggested that predictive policing could have a real impact when used in conjunction with existing policing techniques, such as community policing.

Cops have used statistical data and geospatial analysis to forecast crime levels for years. But as analytical tools have become more sophisticated and data sets much larger, the ability to forecast crime has grown more nuanced. Not everyone believes that technology can accurately forecast crime — some argue that humans who commit crimes are too random — but a growing body of evidence points to patterns in how and when crimes are committed. Eliminate some of the guesswork and police agencies can direct their manpower where it’s needed.

“Predictive policing is another extremely popular technology when you have limited resources,” said Jim Bueermann, president of the Police Foundation. “It becomes extremely valuable when you can predict where to put your resources to be the most valuable and effective.”

Predictive Policing: A Silver Bullet?

According to a 2013 RAND report, predictive policing has been so hyped that unrealistic expectations have created problems around its use. Some of the biggest myths include:

1. The computer actually knows the future. Computers can simplify the search for patterns, but all of these techniques are extrapolations from the past, making “predictions only as good as the underlying data used to make them.”

2. The computer will do everything for you. Although some predictive policing solutions appear quite comprehensive, humans still must collect relevant data, preprocess the information so it’s suitable for analysis, and then review and interpret it in response to ever-changing crime conditions.

3. Police need a high-powered (expensive) solution to get the job done. In fact, productivity tools, such as Microsoft Office and a geographic information system, can support many predictive methods. RAND says increases in predictive power tend to show diminishing returns.

4. Accurate predictions automatically lead to major crime reductions. Studies show that when the focus is on the analyses and software, the results tend to obscure the fact that predictions are just that: predictions. Actual decreases in crime require taking action on those predictions.

The RAND Corp., a nonprofit research organization, released an exhaustive study on predictive policing in 2013 and found that the technology can be used to: predict the place and times of crimes; predict and identify which individuals are likely to commit a crime; predict the profile that accurately matches likely offenders with specific past crimes; and predict victims of crimes.

But the effectiveness of predictive policing also can be hindered by emphasizing data accuracy over tactical utility, relying on poor-quality data, misunderstanding the factors behind a prediction, inadequate assessment and evaluation of predictions, and overlooking civil and privacy rights when using the software to label areas and people as at-risk. 

Bueermann said predictive policing shows promise, but that more research needs to be conducted on its effectiveness. “The idea that you can forecast where the highest probability of crime will occur — it’s never going to be an exact prediction, only a target for a future hot spot, that’s all.”

One technology tool that, on first glance, may seem to have nothing to do with predictive policing has become increasingly effective at just that. Automated license plate readers have been around since the 1990s, when the British government used cameras and readers to track vehicles used by the Provisional Irish Republican Army. More than 20 years later, the technology is extremely popular, primarily for detecting stolen vehicles. 

But now some police departments are mining the data captured by the readers to identify vehicles near a crime scene. Records of license plate scans are stored in a database for as long as two years, so officers can use the information to solve crimes. Plate readers also provide geographic and time information, making the technology useful for cities that are battling drugs, homicides, burglaries and gang activities, according to RAND.

When matched with hotlists of vehicles with outstanding citations or expired registrations, license plate readers also can boost municipal revenue. But the technology isn’t cheap. Cameras can cost as much as $25,000 per unit. Readers at fixed sites can cost even more — as much as $100,000 — although they can operate 24 hours a day in places where traffic choke points occur, such as bridges or busy intersections. New York City has used readers since 2006 and has increased arrests for grand larceny by 31 percent and recovered more than 3,600 vehicles. Sacramento, Calif., said the technology helped it drop from 6 to 13 in a national ranking for per capita auto thefts.

Many cities acquired readers with federal or state grants. But grants don’t cover ongoing maintenance and data storage costs, which add up over time. Other issues like accuracy in reading plates and the frequency with which hotlist databases are refreshed (the more frequently the better) also can impact the benefit of plate readers.

But as more police departments adopt license plate reading technology, privacy concerns are mounting. Because the systems retain information about every license plate read, privacy advocates say law enforcement agencies could use the information to track the movement of individuals, even if they’re not suspects in a crime. 

In 2013, the ACLU reported that long data retention periods and more information-sharing among law enforcement agencies could degrade personal privacy. Law enforcement agencies need to set policies around the data’s retention period and who controls the data. Done correctly, license plate readers can be an effective tool for crime analysis as well as generating revenue. But if the policies behind the technology are flawed, community response can become swiftly negative, say the experts.

Go back to the full Digital Communities report