LAPD Taps Predictive Policing To Focus Patrols
It’s a little after two in the afternoon at the LAPD’s Foothill Division — prime time for the administrative work that makes boots-on-the-ground service possible.
But the division’s a bit handicapped today. Its servers are down.
“I just wonder if I’m getting this wrong, ’cause it’s not letting me in.” Capt. Sean Malinowski stares at his monitor, and for a moment, the former Fulbright scholar with nearly 20 years on the force is your befuddled father trying in vain to unlock the family computer.
He wants to pull up the analytic research backing the department’s new “predictive policing” initiative. But despite what the name might suggest, this is hardly "Minority Report."
Less law enforcement tracking the thoughts of the public than it is academics curating crime stats and patterns, the evolution of crime mapping is well on its way to changing how police patrol the streets of Los Angeles. And it looks like it’s working: In the last year, areas surrounding Foothill Division have seen a 12 percent decrease in crime, 26 percent in burglaries alone. Using calculations from a Rand Corporation study, the LAPD estimated efficient patrolling has saved the community some $4 million in costs related to police response and prosecution.
Southwest, Olympic, and North Hollywood divisions have joined Foothill in a trial phase of the new program, PredPol. Variations of the system already exist in cities like Chicago, Minneapolis, and Dallas. The LAPD is expected to roll out the program across several more divisions by this summer including parts of the San Fernando Valley, and possibly Topanga Canyon according to Malinowski.
In 2005, the LAPD tapped a team of Ph.D. scientists to help them preempt and discourage crime around the city. Those scientists — two mathematicians, an anthropologist, and one criminologist — founded PredPol and began collecting three years’ worth of crime data, giving more weight to recent incidents to observe patterns in neighborhoods.
The team tackles each area by breaking it up into 500-by-500 foot boxes — a little bigger than a major intersection, corner to corner. Focusing primarily on property crime — including grand theft auto, burglary, and burglary from a car — the predictive model considers crime type, date, time, and location of past incidents to forecast high-crime “hot spots” based on human behavior.
UCLA anthropology professor and PredPol cofounder Jeff Brantingham was clear that no information about individuals goes into his algorithm. “We don’t need it,” he said by phone, minutes before boarding a plane in Chicago. “The where and when of crimes tell us all we need to know about why they’re occurring there.”
The PredPol team emails a PDF to Malinowski every morning, with new data from the previous day included. “It’s locations within that neighborhood that are high-crime,” Brantingham said, “and today’s might not be tomorrow’s.”
The captain hands out copies to officers at roll call and tells them which of the boxes they’ll be assigned to that day, an extension of their regular missions. Malinowski cites research that showed spending just 15 minutes every two hours policing these boxes can have an impact on crime rates.
Police presence is meant to serve as a deterrent. “Knucklehead may be showing up to commit the crime,” Malinowksi says. “But now you’re there, and it won’t happen.” Brantingham compared it to a grandmother sitting out on a porch. “She knows all the teenagers on the block, so she can say, ‘Don’t you do that.’ The same thing goes for police even though they do more. They’re changing the opportunity structure for crime to prevent it.”
But, Malinowski clarifies, “If we make an arrest in a box, it’s incidental. That’s not the whole point.”
Back at his Foothill office computer, Malinowksi eventually pulls up a presentation explaining the program. One slide features the kind of equation you might see at a time management or job skills seminar: “Risk-based deployment” plus “removal of impact players” and “analysis-based goal setting” adds up to “incremental crime reduction.”
Malinowski winces over the second factor, which involves tracking suspects through data and investigating crimes. “I don’t want people to misinterpret ‘removal,’ but that’s what it is,” he says. “They’re predators. And everybody who lives there wants them out.”
Malinowski says that isn’t easy to pull off in most neighborhoods. “People won’t testify against [criminals] because they're so afraid of them. It’s a problem.”
Andrew Ferguson, a law professor at the University of the District of Columbia, says low reporting rates are part of an inherent flaw in the system. “Predictive policing is only as good as the data input,” Ferguson said. “If you have some imperfect data inputted, you’re gonna have an imperfect prediction, and if everything is based on that prediction — well, it’s not terribly reliable.”
Emory Law Journal is set to publish Ferguson’s third article examining predictive policing methods and their impact on the Fourth Amendment, which requires probable cause for officers to conduct searches and reasonable suspicion to make arrests. In “Predictive Policing and the Future of Reasonable Suspicion,” he urges the legal community and court system to start thinking seriously about how to approach cases concerning predictive policing.
“If a law enforcement computer algorithm can change Fourth Amendment freedoms,” Ferguson writes, “then courts have an extra responsibility to ensure that the technology meets reasonable standards of reliability and accuracy.”
He says he was skeptical upon hearing first word of the new practice. “It’s not just looking at crime data and thinking retrospectively. It’s saying, ‘We can now take that information and use it prospectively, and in fact we’re confident enough in our data that we can predict where a crime would be,’ ” Ferguson says. “And I thought, ‘Really? This doesn’t make any sense.’ ”
Ferguson started digging, and eventually reached out directly to Malinowski, emailing a draft of his article along with a litany of concerns. The LAPD and the team at PredPol seem genuinely grateful for his input — Malinowski and Brantingham both dropped Ferguson’s name and said the department was being careful to avoid Fourth Amendment infringements.
“This is the opposite of a dragnet,” Malinowski says. “People who rightfully are concerned that the police are going to go out and drag everybody in — they usually like this approach, because it’s not offender-based or arrest-based. It’s denying the criminal the opportunity to commit the crime.”
Ferguson says his contact when working with the LAPD was limited and that he never actually met the captain or officers who would be carrying out these predictive orders. “I don’t really have a stake in whether it’s good or bad,” he says. “In terms of how police choose to allocate their own resources, they could do it by throwing darts at a wall or they could do it by using a sophisticated computer program. I would hope they’d use the smarter policing method, and I think they are.”
In his article, he notes the benefits of the program — that it was cost-effective for a department facing budget cuts and could bring good publicity with its “promise of a high-tech, progressive-sounding plan to stop crime.” Ferguson writes the boxes were “more precise” than patrolling generalized high-crime areas, and that the method “appears to reduce crime with minimal disruption to regular policing responsibilities.”
But he voiced concerns about the types of crimes targeted. Southwest Division has started using predictive policing to address violent crime, and the Charleston Police Department in South Carolina is using a similar program developed by IBM to track patterns of armed robberies. “That might be right, but it also might be wrong. We don’t know enough,” Ferguson says. “We may be doing something that needs real rethinking.”
He also says there might be reason to consider PredPol’s stake in the partnership. “Any for-profit enterprise, I think, has to be looked at with a bit more questioning than others doing it in a pure, academic way,” Ferguson says. “I’m not saying that in any way to malign the intention of the people who created it, but the questions should be raised because there will now be financial incentives to show that this does work.”
That responsibility of skepticism falls to the public. “I hope when the LAPD releases all its findings, people actually start delving into the numbers to see if it works,” Ferguson says. “It’s too easy to manipulate data. You can always lie with statistics and numbers.”
Ferguson said he doesn’t think that’s the case here necessarily, and gave the LAPD credit for being among the law enforcement agencies transparent enough to regularly publish their crime statistics. “That’s a good thing — if it’s being accurate.”
All parties — police, researchers and critics alike — acknowledged this was a work in progress. “It’s important to take endeavors like this step by step,” Brantingham said. “You can’t just jump from nothing to a fully solved problem or a fully deployed solution. The partnership with the LAPD has really made that possible. I can’t say enough about them for having that vision to see that through.”
Looking ahead, Malinowski says he wants to continue tweaking the “dosage,” or how many minutes officers need to spend patrolling the boxes during a watch. He’d also like to see more streamlined delivery of the PDFs; he wants them sent straight to dashboard computers. “The officers actually complain about [the printouts], Malinowski says, grinning. “They’re like, ‘You know, we’re killing a lot of trees.’ ”
He wants to do away with the self-reporting officers do to track time in the boxes, which he called “a little Mickey Mouse for what we’re trying to do.” Instead, patrol cars would be equipped with GPS systems.
If anything, the police are facing more heightened surveillance than the communities they patrol. That may come as some reassurance to Ferguson.
“Like all things, I think you need to have protocols and policies that encourage accountability,” he said. “It’s always useful to know someone’s watching.”
Read more of Neon Tommy's LAPD coverage here.