Policing by the Odds
When a full moon rises in Richmond, Virginia, police officers may feel their hackles rise, too. The presence of a full moon statistically correlates with...
When a full moon rises in Richmond, Virginia, police officers may feel their hackles rise, too. The presence of a full moon statistically correlates with upticks in reports of crime. Every cop has his or her own theory as to why this is so, even if some scientists believe the "lunar effect" is a myth. Nevertheless, when Richmond's police began using a computer system to predict when and where crime would occur, they made sure to program the dates of different phases of the moon into the system.
Other factors go into Richmond's crime modeling, as well. Pay days are in the equation, because robberies surge then around check-cashing businesses. The weather is programmed in, too, so cops can see if crime spikes on hot days or plummets in the rain. Even Super Bowl Sunday is plugged in, since that usually is the slowest crime day of the year. In all, five years' worth of historical crime data sits in the system, where it mixes with real-time data on crime as it happens. By analyzing patterns, commanders believe they can predict where crime will happen, and mobilize patrol officers to anticipated hotspots.
Richmond isn't the only city using predictive analytics to manage police work. Houston, Las Vegas, Tallahassee and Miami, among others, also are counting on data-crunching software to move them beyond simply responding to crimes after the fact. The science isn't quite up to the level of "Minority Report," the movie in which officers from the "Department of Pre-crimes" know exactly where to be, and at what time, in order to knock the knife out of a stabber's hand. But it is advanced enough to help police departments prevent crime by stationing cops where it is most likely to occur. "It's not about catching him in the act," says Stephen Hollifield, the Richmond department's information services manager. "It's about deterring."
Does crime prediction work? That's hard to prove, if only because it's impossible to tally offenses that never happened. Still, Rodney Monroe, the police chief who put the system in place a few years ago, is a believer. In 2004, Richmond was tarred as the fifth-most dangerous city in the United States. By 2006, its ranking had dropped to number 38. Monroe, who has since become police chief in Charlotte, credits the predictive tool for much of the improvement in Richmond's safety. "We were seeing significant reductions across the board on crime," he says. "That was the true measure."
At 2nd Precinct headquarters, on the south side of Richmond, there's a conference room known as the "dugout." Maps cover the walls, and there are numerous computer screens, a 30-inch television tuned to the news, and usually, loads of Mountain Dew on the table. Captain Steve Drew spends most days here, poring over crime reports with an executive lieutenant, an executive sergeant and a crime analyst.
The dugout is the nerve center of the 2nd Precinct's predictive capabilities. Drew and his crew monitor crime data as it comes in, querying the computer system for historical and real time patterns, and trading tips with officers who pop in and out of the room. When a trend worth acting on emerges -- a lot of break-ins are happening on Sunday mornings when people are at church, for example -- the analyst drills down deeper into the data. She'll look at what neighborhoods got hit and when, digging for clues to help commanders decide where officers should patrol the next Sunday.
The idea of predicting where crime will happen can sound a bit like science fiction. But police have always done their own version of what goes on in the dugout room. They just did the analysis and guesswork in their heads. Any good cop knows the liquor store where customers have a way of getting robbed, the corner where the drug dealers hang out, or other hotspots that are worth their patrol time. Sponging up knowledge from the street and acting on hunches: That's always been the nuts-and-bolts of police work.
What the new predictive tools allow is to do this gut-level analysis more systematically, thoroughly and in real-time. By matching new crime trends with old patterns, and mining the data for commonalities and clues, the software connects dots that street cops might never see. "The guy on the beat 10 years has a general idea of where crime is going to be committed," says Jeff Vining, a vice president at the IT research firm Gartner, who has studied Richmond's strategy. But predictive software adds hard data to that knowledge, and quickly supplements it when, say, a housing development is built and there is a large influx of new residents into an area. "It adds more information than the human mind" can process, Vining says.
More important, the technology doesn't get sick, take vacations or get moved to a different precinct where long-cultivated street smarts are lost. The software also helps to connect crimes that occur during different patrol shifts. For instance, if an officer were to learn that an iPod was stolen during his shift, it wouldn't necessarily raise a red flag. Neither would another iPod getting stolen on each of the next two shifts. When four such incidents occur, however, the software spots what may be a small wave of iPod theft. For serious crimes, it can be set to automatically broadcast alerts to officers in the field, via e-mail and text messages. "I can take action after four," Monroe says, "rather than I come in the next day and there are 10 of them."
Not every police department that has taken to predictive tools is using them the same way. In Houston, the focus is on helping officers who are en route to a crime scene know what they might encounter when they arrive. For example, if a disturbance is linked to a mentally ill resident, officers can quickly sift through 15 years' worth of data and check whether the person has a history of not taking medications, or of being confrontational or violent with police. The data can be as fresh as 30 seconds old. This new tool is "law enforcement analytics on steroids," says Captain Mark Eisenman, with the Houston Police Department. "It's big and we're improving the heck out of it."
Some small-town police departments also are finding uses. The police department in Erlanger, Kentucky, implemented a predictive system along with 10 nearby law enforcement agencies. These small Kentucky municipalities don't see as much violent crime as cities such as Houston and Richmond do. But data analysis can be useful for preventing other problems. For instance, if the software notices a rise in injuries related to auto accidents on a certain section of roadway, during a certain set of hours, cops can check to see if there's a hazard to blame, or deploy radar enforcement to try to slow down traffic. "Now we're able to get a little bit ahead of the curve and be a little proactive," says Steve Castor, the public safety communications manager in Erlanger. "Even if it's auto accidents."
Not a Cure-All
If the top brass in some departments are fans of predictive technologies, the view isn't universal. Some longtime cops simply feel fatigued by the proliferation of new tools being pushed out to the squad car these days. Others scoff at the idea that a computer could know more about what's happening on their beat than they do -- or that they would need some slick software to tell them what they've always known about a full moon. In Richmond, the IT director himself was among the skeptics when Chief Monroe proposed it. "They said, 'We're going to predict crime,'" Hollifield recalls. "We thought it was a joke."
But the chief was determined to give it a try. Too often, Monroe says, police departments put officers on the street "without knowing why." He wanted the department to focus its limited resources better. Richmond police were in a reactive mode when it came to fighting crime. Monroe thought that by analyzing more data, commanders could deploy patrol units more tactically -- and just maybe beat crime to the punch.
The tool is pretty powerful. If commercial robberies were high in December 2007, the software will predict another spike in December 2008. Police then can look at what type of businesses were hit, their location and what time of day the crimes occurred. The system can even analyze a robber's modus operandi, down to what he said to his victims -- "Give it up," for example, versus, "Hand over the cash."
Then police can start looking into suspects, questioning informants and figuring out whom they want to watch and where. "I might tell my officers to be on this corridor between 8 and 12," says Captain Drew. "Be aware of black Yukons with two people in them." It's not a crystal ball, however, as Drew is quick to point out. "I don't have a tool that says, 'Tomorrow at 7 a.m. there's going to be a robbery.'"
Monroe, too, is careful not to oversell the technology. It's not a cure-all for fighting crime. If it were, he jokes, he would have quit policing and gone around the country making millions with it. Rather, Monroe sees predictive analysis as one more weapon for the good guys -- like digital mapping, records management and computer-aided dispatch -- in the arsenal of technologies that has made modern policing dramatically more effective. "It led you in the direction you needed to go," he says of the system in Richmond. "You've got to trust it. You've got to believe in it. You've got to use it."