Santa Cruz PD spokesman Zach Friend says predictive policing reduces crime at no cost. (Chip Scheuer)

Santa Cruz PD spokesman Zach Friend says predictive policing reduces crime at no cost. (Chip Scheuer)

On top of a parking garage in downtown Santa Cruz a new high-concept weapon is being tested. It’s not a gun or a chemical, but police believe that, paired with officer intelligence, it is powerful enough to stop a crime before it even occurs. Deputy Chief of Police Steve Clark demonstrates.

An affable man prone to wide grinning, Clark eases the cruiser up the ramp and onto the top level of the lot with its view of the town’s rooftops. “Anytime I go into any area I want to get the heartbeat of what’s going on,” Clark says. “If I was working a beat, I’d go out there and drive around the beat—just kind of set the stage for ‘normal.’ Now I can start hunting. Now I can look for the anomalies.”

Hunting is exactly what riding shotgun on a predictive policing patrol with Clark feels like. We have a suspicious vehicle directly in our line of sight, but we are idling one row over. The cars lined up on either side of us have the effect of tall grass obscuring the cruiser, a predator lying in wait. He takes down the vehicle’s license plate number to run it through the database. 

Clark is taking the heartbeat of the top level of Front Street’s three story-parking garage. When we first rolled up to this level, we spotted a man in his mid-twenties standing between a wall and a large white SUV.

“What’s that guy doing?” Clark asked. “He’s hanging out on the side of that car. What am I going to do? I’m not going to stop and talk to him. I’m going to cruise around, pretend like I didn’t see him. How does he react to me? Does he give us the oh-crap-there-is-a-cop look? Does he wave and say ‘Hey officer!’”

Clark mutters to himself while reviewing the information he’s summoned. “The car doesn’t come up as stolen…” The man is now inside the car, along with a female driver, but they’re not budging from the spot. “Nothing really pings. Not terribly from out of the area. They are probably just hanging out in their car,” he concludes.

On July 1, 2011, the Santa Cruz Police Department embarked on a six-month experiment.html to see if it could reduce crime by proactively deploying officers to patrol areas deemed by a statistical algorithm to be the most likely spots for a crime to occur.

When Santa Cruz became the first department in the country to put this theory into practice, the move attracted an avalanche of attention. It was written up in The New York Times, landed on the cover of Popular Scienceand was named one of the 50 best innovations of 2011 by TimeMagazine—all before the program had any real results to provide.

It is a sexy concept, the kind journalists eat right up. It features, front and center, a futuristic technological advance that improves public safety. It costs the department nothing while maximizing the precious few resources left intact after severe budget cuts. It can’t be accused of unfairly profiling by race or socioeconomic standing, because the only things the computer cares about are the crime’s location, time and type.

The cherry on top is a tidy little parallel to Minority Report, the Philip K. Dick sci-fi novel-turned-Tom Cruise blockbuster in which police use psychics to make arrests before a crime is committed.

It’s almost too good to be true.



Zach Friend, crime analyst and public information officer for the SCPD, pulls up a GoogleMap with predictions of where property crimes in the city of Santa Cruz will occur today.

“This is today’s map. It provides the 10 hotspot locations within the city, with this first one being the theoretical number one: the triple-decker garage.”

On the map are 10 red boxes; today most are clustered in the downtown area, which the police call the “central theater” in the fight against crime.

Crime forecasts are distributed at roll call at the beginning of each shift. For each of the 10 locations pinpointed, the algorithm provides a probability of a crime occurring (10.36 percent for the parking garage) and the probability that it will be a vehicle crime (100 percent in this case, naturally) or a residential one. It also lists two times of day at which the crime has the highest likelihood of occurring—6am to 7pm and 5pm to 6pm.

At a time when cities everywhere are struggling under the pressure of budget shortfalls and being forced to cut essential services, the promise of a technology with the ability to improve officer efficiency without threatening jobs is so attractive it almost borders on intoxicating. 

In January, the Santa Cruz City Council approved cuts of between $100,000-$200,000 to the police department. That news came at a time when SCPD is already operating at reduced capacity. The department is budgeted for 94 positions, Clark says (down from 104 in its heyday), but at the moment the department has only 86 officers.

Officers have quit or retired, and there are staffing decisions that are being held up because they would require approval by the city council, leaving eight vacancies. “Eight positions is a lot of positions. I could do a lot with that. That’s a whole team,” Clark says.

What the use of predictive policing has allowed the force to do, Friend says, is help officers use their patrol time more effectively.

The best news is that the algorithm promises to improve the force’s efficiency for free. As the beta agency, the total cost to SCPD for this cutting-edge technology is nothing at all. Friend is a salaried employee, so even though he’s coming in seven days a week now to run the algorithm that produces the crime forecasts, he’s not accruing overtime pay while he does it.

Don’t feel too bad for Friend, though. Since the program’s inception, by his own count he has done “well over a hundred” interviews or media appearances, many with national outlets—and you can’t put a price on exposure like that for someone running for public office. Friend, a former deputy director of special projects for the Democratic National Committee and 2008 Obama campaign spokesman in Philadelphia, kicks off his campaign for the Second District’s seat on the Santa Cruz County Board of Supervisors Feb. 15 at Bittersweet Bistro in Aptos.

In several of the interviews touting the program’s success, Friend has said that in its first six months, predictive policing has saved 40 Santa Cruzans from being the victims of crime. When asked where he got that number, he explains he subtracted the number of crime victims in the last six months of the year from the number of crime victims in the first six months—a positive trend to be sure, but a textbook example of fuzzy math, and one that invites some scrutiny.

In Friend’s defense, it is difficult to measure the success of predictive policing. Sending officers to vulnerable locations is meant to act as a deterrent—if it works, nothing happens. There have been arrests associated with the program, though: 13 suspects have been cuffed in a designated hotspot in the program’s first six months.



The model being tested in Santa Cruz is an outgrowth of a research project by George Mohler, now an assistant math professor at Santa Clara University. “Self-Exciting Point Processes Explain Spatial-Temporal Patterns in Crime” was the name of his paper, published last March in the Journal of the American Statistical Association. Depending on whom you ask, it was either Friend or Clark who first saw a newspaper article about the research Mohler conducted at UCLA, where he was working as a postdoctoral researcher.

The Mathematical and Simulation Modeling of Crime research project at UCLA is funded by federal money from the National Science Foundation and headed by anthropologist Jeff Brantingham. Last spring Mohler was working on the project, which uses math to model crime statistics in space and time. Time-lapse versions of these maps show crime moving across a map like weather patterns across a blue screen.

“We all believe that we have lots of deterministic control over what we do and where we do it and how we do it and why,” Brantingham says. “But at the same time there are also a lot of aspects of human behavior that are actually very general and can be described very effectively in simple ways.”

When observed from a distance, Brantingham says, there is not much difference between the way people behave and the way fluids or gas molecules behave. With enough data, he says, it is possible to create models that will predict criminal behavior with a fair degree of accuracy. 

In Santa Cruz, the model is only being used for property crimes: burglary, car theft and theft from vehicles. The basic idea is that criminals are predictable creatures. If there is an opportunity—a streetlight out on a certain corner, say—a criminal will exploit that opportunity as long as it exists.  A burglar sees a dark street corner as an opportunity to break into a house, and when that crime is reported, the relevant information (where, what time, what kind of crime) feeds into the predictive policing algorithm.

Maybe that burglar decides to make another trip to the same house, or perhaps a second burglar sees the same streetlight as an opportunity to burglarize the house next door—these are the ways that one crime can foreshadow another.

“We’re not nearly as complex as we think we are,” Brantingham says.



What does predictive policing look like in practice? It’s actually pretty boring. After running the plates at the parking garage, we spend the next two hours driving around to different cells highlighted on the map. Several of the 500-foot by 500-foot hotspots are located along the San Lorenzo River levee, a couple are in Beach Flats, a couple are off Laurel near Santa Cruz High and one is way up on Water and Morrissey.

As we cruise the streets, Clark points out the kind of elements he would be looking for on a check—a bent window screen on a house or someone eating alone in a car (it suggests that person does not live the neighborhood).

On one residential street, the deputy chief notices two teenagers in a parked sedan. They see the police car and quickly exit their vehicle. He notes that the driver has the car keys in her hand, a fact that reassures him. “Probably smokin’ some pot. Don’t want mom and dad to catch them sparking a spliff, man!” he jokes, assuming a stoned surfer drawl.

Predictive policing, as implemented in Santa Cruz, doesn’t usually take the form of driving around scoping the scene for hours on end. Typically, an officer will devote a portion of whatever discretionary time is left in a shift—an extra 15 or 20 minutes—to patrolling one of the day’s hotspots.

Clark points to the computer screen in the cruiser’s dashboard that lists each police car in the city and its location. If the car is on a call the entry will be colored yellow, he explains; if it is free it will be green. Right now, all except one are green. This is the time when an officer will go on a predictive policing check, or eat lunch—or both.

A few months earlier a sergeant on the graveyard shift was taking his middle-of-the-night meal on the top of the triple-decker parking garage. “He’s up there with his lunch pail, and as he’s sitting in the corner, this knucklehead comes up and starts breaking into cars,” Clark chuckles. “And this poor guy is trying to eat his peanut butter sandwich. So we wound up catching that guy and arresting him.”

That arrest was one of the 13 that Friend says are directly attributable to a predictive policing check.



“The arrests were routine,” begins the article, which ran in The New York Times on Aug. 15. “Two women were taken into custody after they were discovered peering into cars in a downtown parking garage in Santa Cruz, Calif. One woman was found to have outstanding warrants; the other was carrying illegal drugs.”

The arrests were not routine, though. The police officers made them while patrolling the area on a predictive policing check. The anecdote was fed to the Times by Zach Friend, working in his capacity as police department spokesman, to a writer who composed the piece entirely from New York City. It ran with a photo depicting an unrelated arrest.

The article interested Andrew Ferguson, a constitutional law professor at the University of District Columbia’s David A. Clarke School of Law, because under normal circumstances, he says, the police could not necessarily have stopped these women. 

“In a pre-prediction land, if the officer is patrolling that same area—that garage—and sees two women looking into windows, that’s not enough to search them,” Ferguson says. “And it’s certainly not enough to arrest them.”

At the heart of Ferguson’s doubts about predictive policing is the concern that it could endanger fourth amendment liberties. The fourth amendment protects citizens’ privacy. It says that police need to establish reasonable suspicion of a crime—probable cause—in order to conduct a search of a person or that person’s property.

Before he went into academia, Ferguson was a public defender in Washington, D.C. In D.C., he spent a lot of time contesting police stops in so-called “high crime” areas. The Supreme Court has said running away from an officer in a “high crime” area is enough reasonable suspicion for an officer to stop that person. Ferguson, who has been studying the issue, wonders whether a statistical prediction by police will similarly influence what constitutes probable cause for a stop or search.

As far are the courts are concerned, Ferguson says, predictive policing is unprecedented. He warns that Santa Cruzans, as the guinea pigs in this experiment, should be considering the implications the program could have on their rights to privacy and freedom from unreasonable searches and seizures.

“You can imagine in every block that has been so designated with that nice red circle around it, there are lots of people who live there and there are lots of people who go about their daily business,” he says, “and some of those people are going to be doing innocent things that are going to correspond with what you might think ‘theft from auto’ might look like. If you’re carrying a screwdriver, are you going to be searched? If you’re carrying a bag?”

“These are the questions that get raised.”



“There is an innovation here, and it’s not something that should be thrown out, because no one wants crime in their neighborhoods, everyone wants the police to be more efficient,” Ferguson says.  “You want data to do things, but you also want controls and checks on that data.”

“The officers show up to their morning shifts are told ‘Look, here are the 10 hot spots to go look at.’ They are not being asked to analyze the data,” Ferguson says. “They’re doing what they’re told, which is what they should do, right? So you’re putting a lot of power in the crime analysts.”

Zach Friend says that the two women from The New York Times article were first stopped because they were in violation of a municipal code called the parking lot trespass law. “You can’t just chill in a city parking lot for years and years and years,” Friend says. “We have a 15-minute limit because people were, in essence, using it as daytime camping.”

“But,” Friend continues, “it’s not unreasonable for a policeman to ask why you are looking into multiple vehicles—it’s not like that is your car. That is a reasonable point of contact for a suspicious activity.”

Santa Cruz Public Defender Larry Biggam sees the practicality of predictive policing, but he’s skeptical. “I understand why they are trying to do this on a limited budget to try and maximize their resources, but they have got to have facts. You just can’t stop people on a hunch or suspicion or computer printouts,” Biggam says, “There has to be a factual nexus between the stop and criminal activity.”

Getting smart on crime, Biggam says, is necessary across the justice system—from police to courts to corrections. “To the extent that these policies are transparent they are probably healthy, but we’re going to contest them if we see a breach of the fourth amendment—people’s right to privacy—and we’re going to litigate that.

“In other words,” Biggam says, “the Bill of Rights is over 200 years old, and it still applies to today’s technology.”




Now that six months’ worth of data has been aggregated and the trial period for Santa Cruz PD’s experiment has elapsed, one question remains: did it work?

So far, there isn’t a conclusive answer. If you ask Friend, he’ll say that during the first six months of the program, crime in Santa Cruz dropped by either 4 percent or 11 percent, depending on the baseline you use.

During the first half of 2011, Friend says, Santa Cruz experienced the biggest spike in crime in the city’s history. If the crime statistics from July through December of last year are compared to the statistics from January through June, the drop is 11 percent. When compared to historical crime averages, the numbers are slightly less remarkable—4 percent.

That discrepancy may not be high enough to be statistically significant, and while Friend insists that the introduction of predictive policing was the only thing SCPD did differently during these six months, any number of other factors could have conspired to move the needle just that much.

Jeff Brantingham seems to want data that is a bit more conclusive. The experiment his team is currently running in Los Angeles is, essentially, a double-blind randomized control study like those pharmaceutical companies use to test the effectiveness of drugs. In this case, a placebo map of crime hotspots created by a knowledgeable and experienced officer is tested against a map created by the computer. Both are distributed to officers. That trial is expected to wrap up in May.