Predictive Policing Is Here Now

“Predictive policing used to be the future,” said career cop William Bratton, “and now it is the present.”

In mid-May 2015, this visionary former chief of police in Boston, New York City and Los Angeles — and currently again top cop in New York — was talking about his early days as an officer in Boston, about what worked and what didn’t, and about what can work better in the future. That future will involve predictive policing, which Bratton is bringing to New York.

Bratton helped develop predictive policing when he headed the Los Angeles Police Department during the 2000s. It seeks not just to fight crime, but to anticipate and prevent it. It uses cutting-edge technology and Big Data — some from past analysis and some that stream in real time to an on-board computer in a patrol car — to identify high-risk areas, which can then be flooded with police. The aim is not to make arrests but to deter crime before it occurs.

Predictive policing relies on community engagement — it can work only when the police are seen as part of the neighborhood, rather than as an occupying presence. At a time when police-community relations are frayed and many cities face rising violent-crime rates as well as renewed concern about terrorist threats, the approach may provide a better way forward.

Currently, about a dozen American cities — including Los Angeles as well as Santa Cruz, Calif., Atlanta, and Tacoma, Wash. — are using PredPol, a leading predictive-policing software and analytics program. Many of these cities rolled out the system over the past few years, and they are seeing positive, and sometimes dramatic, results. In Los Angeles, a nearly two-year study by UCLA crime scholars and law-enforcement officials, released this past fall, found that PredPol successfully predicted — and prevented — twice as much crime as human crime analysts did. The LAPD is now using PredPol in 14 of its 21 divisions.

“Every police department in cities of 100,000 people and up,” says criminologist Craig Uchida, “will be using some form of predictive policing in the next few years.”

Like every innovation, the method has advocates and critics. Predictive policing strikes its detractors as potentially Orwellian law enforcement, but at its best, it aspires to something quite the opposite — a return, albeit a high-tech one, to the days of police on the beat who knew their constituents and worked with them to keep neighborhoods safe.

“The use of information to respond to crime has always been part of the history of policing, an essential part to solving crime after the fact,” Bratton explains. “When Sir Robert Peel [home secretary in early 19th-century England] created the British Metropolitan Police Force, he had nine principles of policing, which were focused on the prevention of crime.”

Before Peel created the Met in 1838, London was policed by the Bow Street Runners, six officers who constituted the city’s first professional police force and solved crimes as a civic service — not, as professional “thief-takers” previously did, for a fee. The force was founded in 1749 by Henry Fielding, author of Tom Jones. It disbanded in 1839, the year after Peel created the Met as a citizen police presence and an alternative to a military force. The Met’s authority depended on public approval. To Peel, this meant that the police — nicknamed “Bobbies,” for Peel’s first name — had to behave respectfully, succeeding not through compulsion but through the willing cooperation of citizens.

Bobbies traditionally didn’t carry firearms; when force was necessary, it was to be minimal. And the police were not the judiciary. Bobbies did not judge guilt or innocence, did not punish or avenge. “The police are the public,” Peel said, “and the public are the police.” Peel believed that the Met proved itself not by the number of criminals it caught but by the absence of crime.

In the United States, policing took shape along Peelian principles: Cops walking the city beat deterred crime by their presence. They got to know the neighbors and the neighborhoods, which were more stable than they became after World War II. For years, many cops claimed that the most realistic police show on TV was Barney Miller because it showed the everyday neighborhood problems that most officers dealt with: homicides were rare, and few cops ever had cause to pull a gun. In New York City, you drew your weapon only in life-threatening situations, and, when you did, you shot to kill. The most common response to shooting a dangerous criminal was to vomit. But Barney Miller was an exercise in nostalgia. It was produced just as the system was changing.

Bratton began his policing career in 1970. “It was a time of great transition,” he says. “We were coming out of the 1960s, that time of social turbulence.” Police had become reactive. Instead of a benign presence, designed to prevent crime, they spent more and more time in their squad cars, responding to calls. At a time when people were attacking police, throwing bricks off roofs, using guns and Molotov cocktails, and fomenting revolution in the streets, patrol cars were seen as safer — and more efficient, as budgets became stretched. Officers could cover a greater area more quickly in a car than on foot. Random patrols and responses to calls — counterpunching — supplanted an assigned beat. It was “an ultimately damaging refocusing of the police mission and responsibility,” as Bratton puts it.

The introduction of the 911 calling system, starting in the late 1960s, reinforced the trend to reactive policing, as did a technological revolution in everything from ballistics to serology. The lab replaced the street. Suddenly, cops had a new toolbox of crime-solving techniques. “Policing was professionalized,” Bratton said. There was “better training, better education and more focus on solving crime than preventing it.”

The new methods generated new problems: Police departments found themselves swamped with the surge in calls coming in. They couldn’t keep up with the demand for their services. Making matters worse: To save money, cities began cutting back police forces around the country.Boston’s police department went from 2,800 officers to 2,000. New York City’s force shrunk from 31,698 in 1970 to 22,588 in 1980, about two-thirds the size of the force today.

The police who remained faced many new frustrations, especially with the growing legal emphasis on criminals’ civil rights. The Supreme Court’s 1964 ruling in Escobedo vs. Illinois (establishing the right of suspects to have a lawyer present during police questioning) and the 1966 ruling in Miranda vs. Arizona (requiring police to read suspects their rights before they can be questioned) transformed policing. Cops were overwhelmed, too, by the growth of trafficking in hallucinogens and the various forms of cocaine and by the increasing numbers of young people drawn into the drug culture. Things worsened during the 1980s. “We were being briefed [in Boston] on this new type of cocaine, crack, that had decimated New York,” Bratton said, “and … within a short period of time: whoosh, it was like one of these tsunamis.” Crime numbers climbed to frightening new highs.

Deinstitutionalization compounded the problem. A flood of homeless mentally ill people ended up on the streets, degrading neighborhoods further. Every day, people were confronted with evidence of a crumbling social contract. By the end of the 1980s, Bratton observes, the reactive approach dominated police methods, which focused on improving performance in two areas: responding to 911 calls and investigating major crimes. Police neglected a third: preventing disorder.

In March 1982 came George L. Kelling and James Q. Wilson’s soon-to-be-famous article “Broken Windows,” about policing that focused on quality-of-life crimes. The Broken Windows theory was reinforced by CompStat, the crime analysis and accountability system that Jack Maple created and instituted in the New York Police Department in 1994, when Bratton began his first tenure as the city’s police commissioner. CompStat, the forerunner of predictive policing, helped Bratton and the NYPD transform a city that had been overrun for decades by crime and disorder.

“We began to use intelligence-led policing to focus on the prevention of crime,” Bratton said. “And, increasingly, as we got more capacity to get that information in a timely fashion [real time], we were able to stop patterns and trends after the second, third or fourth incident rather than catching them when they’ve committed 20 or 30 or 40. We had found a way to prevent crime by stopping perpetrators after the second or third act.”

The result of the NYPD’s innovations is well-known: between 1993 and 2000, a drop of 57.26 percent took place in the seven major crime categories, from murder and rape to burglary and auto theft. Two decades after these practices first became systematized, the building blocks were in place for the next phase of the policing revolution: predictive policing.

The PredPol system works by looking at a specific geographical area. A precinct map gets divided up into 500-by-500-foot “boxes,” each about the size of a city block. A box thought likely to be a locus of crime — because of its history, location, current activity and other factors — becomes a “hot box.” A computer analyzes what crimes have occurred there in the past, what is happening in real time and other inputs. Even the weather can be a factor: Fewer potential criminals come out when it’s below freezing, for example.

What sites in the area — schools, strip clubs, bars, malls — might serve as magnets for potential burglars? The cops look at an “intersection of risk factors,” said Megan Yerxa, a crime analyst at the Tacoma police department, which fully deployed PredPol in October 2013. They have to look at “a terrain that shows hot areas,” such as places where two cultures rub up against each other. Once the hot spots are identified, the police visit them two or three times during a tour, making their presence felt and, by doing so, deterring crime.

“PredPol doesn’t take into [account] socioeconomic factors,” Yerxa says. It avoids targeting a particular race or income group, and the statistics tend to hold true for all kinds of neighborhoods.

The evidence suggests that crime is predictable, at least statistically, because criminals follow patterns. They tend to go back to work in the same area, the same house or business, because they’re familiar with the setup — and they got away with it the first time. The nearby risk of one burglary, for instance, tends to be 600 feet from the first crime over a period of 14 to 28 days. “An offender hits a neighborhood three times,” Yerxa said — and then tends to go to the next neighborhood over. “They travel to avoid detection, but they don’t travel too far.” The questions she asks: Do the last 28 days predict the next seven days? How do they predict it? And is there a way to intervene to interrupt the repeating pattern of criminal behavior without violating someone’s civil rights?

UCLA anthropology professor Jeff Brantingham and his PredPol co-founder George Mohler of Santa Clara University thought it was possible to predict certain crimes, using a system similar to the one that forecasts earthquake aftershocks. Initial crimes are often followed by an “aftercrime” — for instance, a car theft can be followed by a robbery in which the stolen car is used. They developed an algorithm to predict future crime locations.

Departments using PredPol look at what has happened recently in a hot box — say, a series of house break-ins before dawn. The local precinct may send patrol cars and beat cops into that box. If the cops spot a known offender — someone with a record of that particular crime at that time of day — who is casing a house, that gives the officer probable cause to stop and engage the suspect and, possibly, to make an arrest.

“I’ll stop and talk to a guy,” said Los Angeles Foothill Division assistant watch commander Sgt. Scott Kennedy, “and engage him.” Kennedy tries to create a consensual, not a confrontational, encounter. This is not a dragnet, scooping up anyone who looks suspicious. It is targeted — and it’s used only when there is clear probable cause.

The police can make cities safe, but there is always a cost. Freedom or security: It’s a balance that we weigh every day. At what point does protecting society run the danger of infringing excessively on individual rights, or, more broadly, changing the nature of our democratic republic? And are there reasonable alternatives to these choices?

There are “three potential problems with predictive policing,” according to Chris Calabrese of the American Civil Liberties Union. “First, garbage in, garbage out. If you put bad data in, you get bad results. By bad data, I mean historical inequities.” Calabrese focuses on stop-and-frisk, which he seems to equate with predictive policing, though most advocates of the latter are careful to distinguish between these approaches. The second problem, Calabrese said, “is transparency: How are these algorithms created, can you audit them, or are they a black box where the computer tells you where you focus, and there’s no accountability? Who is vetting the algorithm?”

“Third,” says Calabrese, “what types of enforcement are being used?” Are the police looking at “where crimes were reported in the last few months, [or] are they looking on Facebook pages? Are they looking at friends of people who have been targeted on Facebook?” Would the ACLU be willing to work with PredPol? I ask. Calabrese hesitates: “Our work [is] to monitor civil rights.” But “our door is always open.”

Hanni Fakhoury, an Electronic Frontier Foundation lawyer, says his fear “is that these programs are creating an environment where the police can show up at anyone’s door at any time for any reason.” He cites an incident in which the Chicago police warned Robert McDaniel, 22, that he was on a “heat” list generated by an algorithm developed at the Illinois Institute of Technology to identify those most prone to violence and that the cops were watching him — though he had no violent criminal record.

“This is what civil rights activists are concerned about,” concedes Sean Malinowski, project manager for the LAPD’s predictive-policing initiative. The worry is that police will create, as one headline suggested, a “real Minority Report computer to predict crimes,” referring to the dystopian film and its concept of “pre-crime,” in which individuals are arrested based on offenses that they supposedly would commit in the future. A number of predictive-policing software and analytics systems have been developed — IBM Modeler, BairAnalytics and ESRI Mapping, among others — that do tend to focus on individuals. That could lead to constitutional or civil rights challenges in the future.

The PredPol police departments are different. None would generate a “heat” list such as the one used in Chicago. Again, PredPol focuses on geography — “place-based policing,” as Tacoma assistant police chief Peter Crimmins describes it. “The Constitution doesn’t go away,” adds Uchida. “We made a conscious decision that we weren’t going to target individuals. PredPol has stayed away from individuals and will continue to do that.”

To prevent the new era of Big Data from becoming authoritarian — to balance security with freedom — Bratton, who is sensitive to the danger, says that “it’s also important to make sure to have good people elected and appointed to office.” He continues: “The trust on which democracy itself rests is: You have standards of oversight and the ability to attract the right people into these positions of great trust. Isn’t that the whole foundation of what democracy is based on? We give up certain freedoms to other individuals, who have the power — as in the case of the police — to control your behavior. The challenge for the police is that it be done constitutionally. Legally. Respectfully. Compassionately. That it be done consistently. By consistently, I mean not policing differently in a poor neighborhood than in a rich neighborhood.”

The best defense against misuse of the police is citizen vigilance. When we give up our duties to police our neighbors to a professional force, we are not surrendering our responsibility to monitor those who act on our behalf.

It seems inevitable that some form of predictive policing is on the way. We can’t put the technology of policing back in the bottle any more than the Luddites could stop power looms. We’ll need to decide what forms the system should take — especially with the mounting danger to cities from terrorism.

The most optimistic scenario is that the technological revolution — through systems like PredPol — re-creates the best of traditional beat-cop policing. Officers will get to know their community and will, as in the PredPol model, incorporate community outreach. Police departments will do more with tighter budgets and fewer resources while also protecting individual rights.

PredPol allows cops to reinvent Robert Peel’s preventive policing — to return to the humane principles of the 1830s with 21st-century tools. Bratton envisions more interaction among the three great police departments in New York, L.A. and London. “We’re testing both PredPol and a new system, HunchLab,” Bratton explains. “We haven’t committed to either. In the meantime, we have developed our own predictive model that we are field-testing against the other two.”

Now 68, with nearly a half-century in law enforcement behind him, Bratton remains as committed as ever. “It’s my heart, soul and passion,” he says. “The importance of policing. No one can do it better than us, but we can’t do it alone.”

This piece originally appeared in The Dallas Morning News, adapted from the Winter 2016 Issue of City Journal

This piece originally appeared in The Dallas Morning News