From a computer at Chicago’s police headquarters on the city’s South Side, Deputy Chief Jonathan Lewin pulls up a list of about 1,400 names of people with criminal records. Attached to each name is a mug shot, demographics, prior arrest record, and gang affiliation. There is also a score.
Determined by a computer algorithm based on 11 weighted variables, the score measures the likelihood of a person to be involved in a shooting, either as a victim or an offender. The scores range from 1 to 500, and the higher the score the more likely the person will be a party to violence.
“This is really evidence-based policing,” says Deputy Chief Lewin.
Chicago is at the forefront of a growing trend in police departments across the United States to use big data and algorithms to determine the individuals and places that are most as risk for violence. Some of these risk assessments are astonishingly accurate.
But while the people-centered model pioneered in Chicago is now spreading, predictive policing so far has not succeeded in bringing down the murder rate here. That's led some to question whether big data alone can prevent crimes.
It is certainly possible that other factors – such as a breakdown in trust between police and minority communities, a need to hire more officers, or the overturning of the city's gun laws in federal court – may be driving the increase in the murder rate, which is up 84 percent over this time last year. Supporters of the data program, for their part, say they believe the murder spike would be worse without the predictive tools.
“The idea that one might be able to target and approach the human drivers and do something about that is obviously incredibly attractive to police departments,” says Andrew Ferguson, a law professor at the University of the District of Columbia in Washington. “The difficulty with Chicago as a laboratory is that [predictive policing] is obviously not working. That might not be the fault of the theory, but just empirically we’re seeing the bodies pile up in the city.”
Other cities, however, have seen success in using data to bring down the crime rate: The Los Angeles Police Department (LAPD) expanded a program to predict where to deploy its officers to 14 of its 21 divisions (up from three in 2013), after a 2015 University of California at Los Angeles study showed that the computers prevented twice as much crime as trained analysts.
Chicago uses a different algorithm, and it also may have aggravating factors contributing to its rise in murders. In the wake of shootings of unarmed minorities, the city has been wracked by months of turmoil and protests that led to the firing of the former police chief. The federal government is investigating the shooting of unarmed teenager Laquan McDonald, and a months-long investigation by the police accountability task force concluded in April that the Chicago Police Department has a history of using excessive force against minorities and perpetuating a code of silence.
“Can you imagine where we’d be if everyone wasn’t working as hard as they were now?” says Christopher Mallette, executive director of the Chicago Violence Reduction Strategy. “The reason that we struggle with a solution is that people don’t want to look at the problem in its totality. We want to look at the part that makes sense to us.”
Can math save lives?
Predictive policing has its origins in a program started in 2008 by the LAPD. That early program used analytical techniques to determine locations that were most at risk for criminal activity.
Chicago became the first city to use a people-centered model when it launched the first version of its Strategic Subject List algorithm in 2012. Based on the public health concept that crime is like a contagion that can spread from person to person, the model analyzes social networks to determine who is most likely to be involved in violent crimes. Since then, Kansas City, Los Angeles, and New Orleans have started using big data to determine who among their population is most at risk.
Sixty-six people were killed in the city in May alone, making last month the deadliest May in 21 years. More than 74 percent of shooting victims and 80 percent of individuals arrested for shootings in 2016 so far have been on the SSL.
So why hasn’t the accuracy of the algorithm translated into less crime on the street? Professor Ferguson says that while many initially saw big data as a panacea for reducing crime, realistically it may be better at identifying problems that preventing murders.
“People should recognize that big data does provide real insights and allows you to see things that you might not see without the data,” says Ferguson. “But the solution is not in the data, it’s in what you do with the data. I think what a lot of people have done is take the first step, which is the easier step, to crunch the numbers. And not taken the second step, which is to ask: Now that we’ve identified these risks, how do we remedy this?”
The home visit model
In Chicago, law enforcement officers, community members, and social service agencies all take part in the home visits – called custom notifications – the police department does for people on the Strategic Subject List.
During the home visit, police officers warn the person on the SSL and their family that they are in danger of either being shot, killed, or put in prison. Then a community partner offers help in the form of job-training opportunities, substance abuse counseling, and better housing options.
This carrot-and-stick model comes from a realization on the police department’s part that, as new Superintendent Eddie Johnson said in an interview, “we cannot arrest our way out of this situation.”
But at the same time, Superintendent Johnson says that those who do not accept the offer of help on custom notification visits should not expect leniency.
“What those individuals need to know if they choose to stay in that lifestyle, we’ll come after them with everything that we have,” says Johnson.
Rand Corp. is currently studying the effectiveness of Chicago’s SSL and its data-informed interventions. According to the individuals who go on home visits, results thus far have been mixed.
Roughly 1 in 4 home visits results in someone accepting an offer of help, according to Mr. Mallette, who invites representatives from one of six outreach and support organizations to each house call. He says that the police have connected with between 700 and 750 individuals through custom notification during the past three years.
While it's too early to tell the long-term effect of the house visits, Mallette says that he's seen some changes in those individuals in the short term.
Chicago Police Deputy Chief Dave McNaughton recalls one occasion where he went to do a custom notification for a young man. He says his mother was in complete denial about her son’s risk for gun violence.
“She pretty much told us to leave the home, which we did of course,” says Deputy Chief McNaughton. “I remember three days later there was a person shot right behind her house that was playing with her son.”
On the same day that McNaughton met the mother, he did a custom notification for a person who was deeply involved in the violence in the 8th police district on Chicago’s Southwest side. He met with the subject’s significant other and she accepted their offers of help.
The couple got services to move out of the neighborhood, and McNaughton says that person has not been involved in violence since.
Who should take the lead?
Ferguson says that part of the problem with Chicago’s data-driven interventions might be the very fact that the police are the ones leading them.
Like risk assessment lists in many cities, Chicago’s Strategic Subject List is made available only to law enforcement officers and is not given out to the social service agencies who lead violence prevention programs or to the community groups who may know the individuals most at risk.
“To me, the big mistake of all of this is putting predictive policing in the hands of the police,” says Ferguson. “The identifying of risk within a society should be put in the hands of other social service entities. It should be with the larger city rather than just the police.”
Mallette agrees that the community should play an important role in data-driven interventions.
“Our focus isn’t on locking people up. Our focus is on saving people’s lives,” says Mallette. “We’re trying to take active measures to try and provide assistance from multiple angles. Law enforcement is doing what law enforcement is doing. Our message is that the community loves you, we value you, but we need you in your rightful place.”