Can data-analysis innovations help identify bad-seed cops before they act out?
The police world has cast an eye toward that possibility as teams of computer scientists from outside the insular law enforcement bubble work to improve on early-intervention systems designed to red-flag officers with behavioral problems who might engage in “adverse interactions” with citizens.
The goal is to reduce the human casualties and financial damages from the cortege of troubling police-involved deaths of citizens (most of them black) that have dominated police news over the past two years—Eric Garner, Michael Brown, Tamir Rice, Walter Scott, Freddie Gray, Sandra Bland, Laquan McDonald and many others.
But there might be a more direct route to identifying rogues: simply ask police chiefs for the names.
For generations, conventional wisdom has held that every cop—from prowl-car partners to sector sergeants to top-floor executives at headquarters—is acutely aware of which officers are dangerous powder kegs.
Herman Goldstein, an academic pioneer of best-practices police research, once wrote that problem cops “are well known to their supervisors, to the top administrators, to their peers, and to the residents of the areas in which they work.” Yet, Goldstein noted, “little is done to alter their conduct.”
He made those observations in his book “Policing a Free Society,” published in 1977, nearly 40 years ago.
That paradigm hasn’t changed, experts say.
“The police command knows who the high-risk officers are,” says Gregory Gilbertson, a former Atlanta cop who teaches criminal justice in the Seattle area and testifies frequently in police misconduct cases. “They absolutely know.”
“They always have, and the data has always been there,” adds Samuel Walker, an Omaha-based police accountability watchdog.
Most large police departments already have early-intervention systems, and many of them use software that tabulates data on individual officers. But what happens to the data?
“The missing element–call it administrative, if you will—has been the mindset of accountability, one that says we are serious about dealing with the bad officers, and also helping officers who appear to be going off track and need help getting back on track,” Walker says. “An EIS (early-intervention system) is simply a tool that will facilitate that effort. But first you have to have the commitment.”
The bad-seed cop phenomenon has been freshly quantified over the past six months by several data-driven newspaper investigations.
The Chicago Tribune reported in January that just 124 of the city’s 12,000 cops were named in about one-third of all police misconduct lawsuits settled there since 2009. The Orlando Sentinel’s Focus on Force project last November found that five percent of that city’s police officers accounted for a quarter of the cases where force was used against citizens. And in December, The Guardian showed that a handful of quick-draw law enforcers in the Bakersfield, Calif., area were repeatedly involved in shootings, including one cop who killed three people in two months.
Police agencies have done a lousy job of getting rid of violence-prone officers—in part because their culture treats bad cops with the sort of occupational omerta that the Catholic Church used for generations to conceal and enable pedophile priests.
“It goes to the issue that police departments do not like looking for or revealing bad things, negative things,” says Jeff Asher, a former crime analyst for the city of New Orleans who writes about crime data trends for FiveThirtyEight and the New Orleans Advocate.
A stunning illustration of how institutional blinders can inhibit self-assessment was revealed during NYPD corruption hearings in New York in 1993. Daniel Sullivan, the department’s longtime internal affairs chief, testified that he had been intimidated because police commissioners didn’t like “bad news.”
“When I went up with the bad news that two cops would be arrested in the morning or that three cops would be indicted, I felt like they wanted to shoot me,” Sullivan said.
The Politics of Police Discipline
In a sense, the outsourcing of analytics development to identify problem cops might alleviate some those internal tensions. Data scientists at the University of Chicago are working with police in Charlotte, Los Angeles and Knoxville, Tenn., to develop new early-intervention systems.
But will the systems survive the complicated politics of policing?
In the 1990s, the endemically corrupt and sharp-elbowed New Orleans Police Department instituted the Professional Performance Enhancement Program, which included protocols for officer interventions based upon such data points as frequent use of force and citizen complaints.
The system was seen as a model—briefly. It soon fell out of use. Today, 20 years later, the NOPD is under pressure as part of a federal consent decree to institute a new early-intervention system.
Gilbertson says such failures often can be traced to powerful police unions.
“You’ve got a problematic officer, and everyone knows it. So a commander tries to fire him,” Gilbertson says. “Then the union gets involved, pitching a fit and threatening the chief with a no-confidence vote if he fires this officer…And then the mayor gets dragged in. He’s a political animal and needs police union support to get reelected.
“Mayors hate this kind of shit. So even if a chief wants to get rid of somebody, the political climate often will prevent them from doing that,” he says.
Gilbertson predicts that unions will continue to be a bulwark against termination of bad-seed cops, even if innovative early-intervention systems take root. I reached out to two nationally prominent police union officials to ask whether they supported analytics-based intervention for troubled officers; neither replied.
But Tyler Izen, president of the union representing LAPD officers, told the Associated Press that cops fear “Big Brother.” He explained, “If you’re watching over me and there’s a setup matrix that is going to tell you that I’m bad, people are always skeptical of things like that.”
Police departments have effectively used sophisticated crime data technology for two decades now, allowing them to pivot enforcement strategies quickly to react to emerging threats. But most have been slow to use that same nimble technology to evaluate officer behavior.
Nick Selby, a Dallas-area police officer and CEO of StreetCred, a law enforcement data analytics firm, says most police IT operations are handcuffed by antiquated “stove-pipe” computer systems.
“All of the technology that is in current use in police departments today was designed and built and sold under a paradigm that has completely changed in the rest of the world,” Selby says. “The average incumbency (of police computers) is about 17 years. It’s Clinton-era front ends on Reagan-era back ends connected to Nixon-era networks. That’s a legitimate description of law enforcement technology today.”
Selby’s is one of a small number of data-focused firms that sells police agencies software that collects, organizes and analyzes the agency’s own information, freeing officers from tedious, time-consuming work—for example, an automated compilation of exhaustive contact information and criminal backgrounds for friends and family members of the subject of an arrest warrant.
The big dog in this emerging law enforcement data field is Palantir, a Silicon Valley firm whose principals founded PayPal. Palantir is a data-analysis vendor for the LAPD, a consortium of Salt Lake City-area police departments, and a number of federal law enforcement agencies, among many others.
One of Selby’s software products, StreetCred Clarity, analyzes patrol officers’ behavior and performance based upon numerous data points. The information might be used to red-flag a cop who stops an inordinate percentage of minorities—or, more pertinently, whose traffic citations written against African-American motorists are frequently dismissed in court.
“You see patterns in the data, and then you look for mitigation or amplifications,” Selby says. There might be a reasonable explanation–if the officer works in a predominantly black neighborhood, for example.
“Or,” Selby adds, “it could just be that the cop’s a racist.”
Good Cop, Bad Cop
What traits make a person a good cop? Some research suggests that an agreeable, extroverted, emotionally stable, conscientious and open-minded man or woman are best-suited to the job. (Some disagree, of course.)
Conversely, disagreeable introverts might not be. The predictive factors used to identify troubled officers have not changed much over time. They include:
· An anomalous number of uses of some type of force, typically more than one every two months
· An abnormal number of citizen complaints
· Abuse of sick leave
· A pattern of automobile accidents
· Relationship issues, including divorce
· Financial problems
Gilbertson has identified two other predictive factors that he says crop up frequently in his expert witness casework. He says an inordinate number of Washington State officers cited in use-of-force complaints participate in mixed martial arts or Muay Thai, a similar combat sport. And he sees a number of post-9/11 military veterans, particular those who served in special forces units such as the Army Rangers or Navy Seals.
“While these individuals are superior soldiers, their specialized skill sets don’t necessarily translate well to civilian law enforcement,” Gilbertson says.
Sam Sinyangwe, an Orlando-based policy analyst and data scientist with the advocacy group Campaign Zero, suggests that a deeper understanding of police violence might be found in data that looks beyond predicting episodes by individual cops.
Campaign Zero has aggregated police use-of-force reports into a revealing database. Last year, law enforcers in Bakersfield, Calif., killed more people per-capita than any other American locale, at nearly four times the national average. Meanwhile, Riverside, Calif., three hours from Bakersfield, had no officer-involved fatalities.
Oklahoma City, Oakland, Indianapolis, Long Beach, Calif., and New Orleans all had very high rates of police-involved killings. Milwaukee, Denver, Detroit, New York, Philadelphia and Charlotte joined Riverside with very low rates.
“So we’d love to be able to get to the point of understanding that broader view of why some departments shoot so many, and–just as importantly–why some departments shoot so few,” says Sinyangwe.
“There are important local and cultural questions that have to be answered,” adds Asher, the New Orleans data expert. “Like what makes a New Orleans cop bad, and is it the same thing that makes a Chicago cop bad?”
Action vs. Intransigence
Scientists, cops and advocates agree that data-gazing will lead to increased restraint and accountability only if law enforcement intransigence can be overcome.
“I don’t buy it,” Gilbertson tells me. “They can’t even stop crazy cops when everyone in the department knows that a certain cop is crazy—and I mean card-carrying crazy. This is not about a lack of information. This is about what a police agency does or does not do with the information it has.”
“Police departments need to be willing to seriously take a look at themselves critically, and it’s all too rare that they have been willing to do that,” says Asher. “The pattern lately has been that these things happen only after a city has signed a consent decree because their police department has been involved in some horrible civil rights violations.”
Selby, the Texas cop and software CEO, argues that deeper data analytics can add nuance and knowledge to the messy public spectacles that often accompany a controversial police-involved death.
“This is where we suck so badly sometimes,” Selby says. “Police are so bad at articulating a rational position in a situation where an officer has been involved in the use of deadly force…Sometimes, yes, it’s a failure on the part of internal affairs and the police. And sometimes you’ll have a cop who is simply doing exactly what you want them to do. We ought to ask that people trust the integrity of the investigative process. But unfortunately at this point, we have shot ourselves in the foot so many times that it’s very hard to trust the process.”