Predicting police misconduct before it happens

Every day there are thousands of interactions between police officers and citizens across the country. While most are uneventful, a small number leave a member of the public disrespected, unprotected, harassed or — in all too many cases seen recently — hurt or even killed.

This summer, fellows with Data Science for Social Good — a program at the University of Chicago that connects data scientists with governments and nonprofits — are working to predict when officers are at risk of misconduct, the goal being to prevent problems before they happen.

The effort’s part of the White House Police Data Initiative, which aims to increase transparency and community trust, while decreasing inappropriate uses of force. (That DSSG was approached by the White House wasn’t surprising; its program director, Rayid Ghani, was the Chief Data Scientist for Obama for America in 2012.)

Police departments around the country — 21 in all — are participating in the national effort. (Chicago police were not one of the departments picked to participate.) The White House matched DSSG with the Charlotte-Mecklenburg Police Department. Like many agencies, CMPD has early intervention systems. The challenge for DSSG was to find ways to improve them and avoid misconduct.

“So we’re trying to identify these opportunities to give them the information and training they need to avoid these negative interactions,” said Joe Walsh, a mentor with DSSG overseeing the project.

CMPD currently looks at measures such as use of force, accidents and injuries, and sets a number of incidents that should trigger a response from the department. Officers who are flagged by the system will meet with a supervisor to review an incident, receive counseling or additional training.

“It’s not the most effective system,” said CMPD Capt. Stella Patterson. “We realize there’s some enhancements that need to be made to it.”

Working through the partnership was sometimes intense. The Charlotte City Council had to approve an ordinance to share the data with DSSG (fellowship staff traveled to the city to make that happen), and some officers — including Patterson — were anxious sharing so much information with people outside the department. Still she feels that the project will help in the long run.

“As a police officer, I’m going to tell you personally, it was a little uncomfortable, because now you’re exposing yourself really to the world,” Patterson said. “People will look at this project as a model for the rest of law enforcement. But the benefit we’re going to get from it is going to be great. While some of us may feel like we’re opening up ourselves, I feel like law enforcement today and moving forward is going to require that.”

To find common patterns, the DSSG team analyzed incidents and anonymized data of the officers involved. They considered things like: when and where an arrest or traffic stop occurred; had the officer worked extra shifts; how long had they been on the force; even what the weather was like at the time.

“Because it’s sort of a new problem, we spent a lot of time trying to grasp what was important and what wasn’t, and that’s something we’re still working on,” said fellow Kenny Joseph, a computer science student at Carnegie Mellon.

That explains why the group of data analysts got face-time with CMPD, meeting department top brass and even going on ride-alongs with officers.

“We would not be able to do a good job had we not gone down,” said fellow Ayesha Mahmud, a demography student at Princeton University. “None of us had any idea coming in what the everyday life of a police officer was like.”

Mahmud said she was struck by how much time a police officer spends during each shift just speaking with residents to gather information and diffuse problems.

“I think we all came to the realization that the data can only capture a very small part of that story,” she said. “I think that really helped us think about this problem.”

With the fellowship finishing up next week, DSSG has identified a few indicators they hope can identify possible problem officers — such as previous uses of force, working extra shifts, or responding to other stressful calls — all before they create problems.

Still, each fellow was careful to point out they haven’t tested and refined the model enough to draw any causal conclusions just yet.

“There’s just so much that could be at play here and we only have three months,” Walsh said. “So while we may be able to improve the system that they have, there’s still a long way to go.”

Patterson said CMPD plans to review the proposed model before they update their current system, but is open adding the findings to their discussions.

“We may realize, looking at all the data and the research, that the thresholds we have now are inadequate,” Patterson said. “That piece of it is still to be determined, and we are certainly going to work with University of Chicago as well as our other partners, other agencies, to see what the best practices are.”

While they remain cautious, the fellows believe the model they’ve created can help the department do a better job identifying problems before they happen.

“It can’t solve everything, but I do think our data can help CMPD do a better job targeting their interventions,” Mahmud said. “Even if we can help prevent 25 more adverse events in a year, that’s better than their current system.”

Walsh said that DSSG plans to continue the project next year, and he’s hopeful they can get data from more police departments. The next up is Knoxville, Tennessee.

Chris Hagan is a web producer and data reporter with WBEZ. Follow him at @chrishagan.

Source: Predicting police misconduct before it happens

Via: Google Alert for Data Science