Can Police Use Data Science to Prevent Deadly Encounters?

As part of Obama's Police Data Initiative, researchers and police are studying "predictive analytics" to improve existing officer early warning systems

By | |

Law enforcement and academia are working together to find ways of improving early warning systems meant to flag bad police behavior before it becomes a problem. (Image for illustration purposes only.)

Several high-profile cases of law enforcement officers using deadly force against civilians within the past year have politicians, and researchers looking for ways to prevent such incidents. This search includes a closer look at the computerized that many large police departments have used for years to identify officers who are most likely to overreact violently during stressful situations. The main challenge: it is difficult to say with certainty how well or even if these systems actually work.

Early warning systems debuted in large police departments—those with more than 1,000 officers—decades ago as a way to identify those officers whose unprofessional behavior could cause problems in the communities they served. Departments programmed these systems to flag recurring complaints against officers and notify supervisors when certain thresholds were reached, such as a certain number of use-of-force complaints over a given period of time. Early systems’ predictive abilities were crude, primarily because they were capable of basing their analyses only on individual data sources—such as formal complaints—rather than combining information from various police databases that could provide context for an officer’s behavior. This might include the officer’s level of experience, whether the officer responded to an incident alone as well as the time and location of the event.

Pres. Barack Obama’s recently announced seeks to fill this gap via a research program to study the efficacy of law enforcement early warning systems—also referred to as s—and to determine how they might be improved. The goal is to more effectively apply statistical tools, machine learning and other predictive analytics that take current data and look for trends that might continue into the future. When a system identifies an officer whose performance records and behavior suggest the need for some kind of intervention, supervisors can step in to arrange counseling, reassignment or additional training.

As part of the initiative, departments in California, Texas and other states plan to share with outside data scientists certain statistics related to uses of force, pedestrian and vehicle stops, officer-involved shootings and other information. Most of this information has not previously been available outside of law enforcement. “The motivation for early intervention systems when they were started was to identify and help deal with that small group of officers that accounted for an inordinate proportion of [civilian] complaints and uses of force,” says , executive director of the Major Cities Chiefs association, a professional organization through which U.S. and Canadian police executives share information and strategies. The systems improved over time, tracking larger amounts of data and automatically alerting supervisors when an officer reached some predetermined threshold. “The effectiveness of the technology, however, is a difficult question, and I am not aware of any research that has examined [how well departments use] these systems,” adds Stephens, who served for nearly a decade as police chief of North Carolina’s Charlotte–Mecklenburg Police Department.

The Police Data Initiative grew out of Obama’s convened last December in response to a series of violent civilian–police run-ins resulting in civilian deaths in City, , , Ferguson, Mo., and elsewhere. It is unclear whether improvements in early warning systems would have prevented the deaths of Eric Garner in New York last summer or Freddie Gray in Baltimore earlier this year, Stephens says. “One of the challenges in all of this is the fact that officers are permitted to use the force necessary to take someone into custody,” he adds, “and there are times when they must.”

Charlotte–Mecklenburg is one of the first police departments to formally commit to the initiative by agreeing to work with University of Chicago data scientists—although the decision came long after Stephens’ tenure there. The university will first analyze the department’s early intervention system as part of a summer data science fellowship program, investigating early warning indicators for adverse police interactions with the public. The school’s will continue this research when the fellowship program ends. Within the next few months the university expects to deliver a preliminary report to Charlotte–Mecklenburg with recommendations for upgrading its system, in place since 2005.

The data scientists have a lot of work ahead of them, in particular determining whether historical data about a particular officer’s performance can be used to predict that officer’s future behavior, says Rayid Ghani, director of the Center and fellowship program. Researchers plan to apply so-called algorithms to the Charlotte-Mecklenberg data similar to the ones they have developed for previous projects. Among their earlier work was predicting high school dropout rates and public health threats.

The Charlotte–Mecklenburg Police Department wants an outside perspective on how their system works and how it could be improved, says Maj. Sherie Pearsall, director of Internal Affairs. The early intervention system is not supposed to be punitive but rather one that allows supervisors to watch their employees and intervene when necessary, she adds. If an officer is cited for use of force three times within 90 days, for example, the system automatically alerts the department’s internal affairs case-management system.

Use of force has a and can refer to something as straightforward as a heated conversation between an officer and civilian or something more serious such as a takedown or firearm discharge—actions that prompt different levels of departmental response, Pearsall says. Additional details about where and when the incidents occurred might help Internal Affairs better analyze the factors that led to the confrontation. Some conflicts can be attributed to factors as basic as an inadequate number of officers assigned to patrol a dangerous neighborhood during a particular time of day, she adds.

Despite the department’s enthusiasm over the Police Data Initiative, the Charlotte City Council initially was skeptical of giving outside researchers access to sensitive police data. The council was divided on whether to approve participation in the project until newly appointed assured them during a meeting last month that any information that could be used to identify individual officers would be encrypted.

More departments are expected to follow. “The time is well past for these systems to be improved,” Stephens says. Currently, whenever police try to understand how use-of-force situations escalate into civilian deaths, he adds, the necessary data analytics are not there.

see also: