How Stockton Is Using Predictive Policing to Thwart Violent Gun Crimes

High crime, low staffing and mounting community frustration drove Stockton, California’s police department to start exploring evidence-based practices in 2012.

Publications like Forbes were calling Stockton the “eighth-most dangerous city in America,” based on 2011 FBI crime data, as the force mulled vendor and in-house data-driven strategic approaches.

Forecasting violent gun crimes using cutting-edge predictive analysis soon emerged as Police Chief Eric Jones’ top priority, and by 2013 the department had bolstered its crime analysis section with an embedded analyst courtesy of Bair Analytics.

“That analyst has been integral in working with the chief there and their captain and Project Forebode,” Tom Sizer, LexisNexis Risk Solutions law enforcement sector senior director, told Route Fifty in an interview. “But it takes good old-fashioned police work in combination to make it all work.”

In 2014, Stockton's police launched its model, which analyzes buckets of data on non-domestic violence-related gun crimes to identify active trends and flag “forecast zones,” where incidents are likely to occur.

Alpharetta, Georgia-based identity intelligence company LexisNexis Risk Solutions acquired Bair just before Jones’ officers began operationalizing forecast zones in March 2015. By creating a call for service the department moves squad cars into projected hotspots at anticipated dates and times.

The same time the following year, the Accurint Crime Analysis forecast and 937F call were added to the force’s monthly Integrated Criminal Apprehension Program meeting. The foundational intelligence layer triggered phase three of Stockton’s efforts: Project Forebode—short for forecast-based deployment.

“We’re not just dispatching cars to the zone but looking at other aspects to flood into the zone. Thirty percent of all gun crimes occur in these small zones,” Jones said. “If we’re able to put resources into zones, when we believe gun crimes will be hot, we are seeing reductions of gun crimes. To be successful to have reductions, we need to ensure 15 percent of the time we get them into the zones.”

Preliminary numbers for March through May 2016 show the department has witnessed 40 to 60 percent month-to-month decreases in non-domestic gun violence-related crimes in forecast zones. Property crimes in those zones decreased 20 to 30 percent at the same time, said police Capt. Antonio Sajor, Jr. in an email. Those statistics have the added benefit of engendering officer buy-in to the new way of doing things.

Next on Jones’ to-do list is honing the forecast model with additional data because gang and gun violence is such an acute challenge for the city. Data from ShotSpotter gunshot technology deployed around Stockton may eventually be incorporated into the system, and the city also runs Operation Ceasefire, a gun violence reduction communications program targeting youths at highest risk.

“Part of the evidence-based approach is, if we’re not getting the reductions we want, we can pivot,” Jones said. “Internally the officers really like it; they want to see this. This type of tech is starting to get recruits, who want to come to a department like this.”

Situationally in the field, predictive policing makes officers more prepared. A seasoned officer might know that crime picks up in a particular 10 to 12 square block area in the summer, but forecast zones are even more precise—broadly validating police intuition while homing it in. Dates and times don’t always align with when cops expect gun crimes to occur.

Currently the forecast algorithm relies on more than a decade of violent crime trends and active knowledge of gang activity. While an incident won’t always occur when predicted, Jones likens crime forecasts to weather forecasts, which give people a pretty good idea of when they’ll need an umbrella.

Accurint Crime Analysis is now in the hands of hundreds of law enforcement agencies across the U.S.

“Toward the beginning predictive policing was very new,” Sizer said. “Now it’s not a fad; it’s really been refined.”

Police departments can always test predictions before implementing them by using historical data to “predict” a violent crime and then see if it actually happened—reducing the likelihood of bad predictions.

LexisNexis is working on a new product that will add to predictions by adding into the model data on where repeat offenders, responsible for most crimes, live.

“We’re bringing the ‘who’ into the equation,” Sizer said.

The company also wants to assist departments looking to bring together crime analysts—working on the “what”, “when” and “where”—and intelligence analysts tracking top offenders. Five or six large police departments across the country are actively trying to monitor top offender lists by merging crime and intelligence analysis.

Stockton might not be there quite yet, but predictive analysis has taken root.

“My end goal here is to make this part of the way we conduct business in policing,” Jones said.