San Francisco District Attorney Wants to Turn Prosecution From "Art" to Data "Science"
BY Sarah Lai Stirland | Tuesday, June 4 2013
If justice is blind, it won't stay that way in San Francisco for long.
Right now, all city district attorney George Gascón knows about the defendants on an aggregate basis is that his office prosecutes is that each of his prosecutors handles, on average, 185 felony cases and 700 misdemeanor cases per year. He wants to know far more, and says his office is now building a database to profile defendants by attributes such as age, ethnicity, gender, education, work history, mental health and substance abuse issues. The system will also track "stabilizing forces" in their lives, such as whether they have housing.
Like private-sector executives who use data profiles of consumers to know that they should stock Pop-Tarts and beer before a hurricane or guess if a customer is pregnant, Gascón wants to use data to better manage his office. Only he's not selling beer. He's managing a team of 130 attorneys who make life-altering decisions on a daily basis: Should the city ask for a maximum or minimum sentence? Should a prosecutor take his next case to court at all? Is a given defendant a good candidate for alternative sentencing?
The current system as Gascón describes it paralyzed by fear of recidivism. Right now, he said, this kind of decision making is done based on intuition and a prosecutor's past experience. Under the current system, risk-averse prosecutors seek maximum penalties, leading to a lock-'em-up mentality that may contribute to America's eye-popping rates of imprisonment. In California, the problem is particularly acute because of its precedent-setting three-strikes law, which imposes harsher sentences on repeat offenders, and has sent soaring numbers of people to prisons and jails, a costly and, many say, ineffective exercise. Gascon, San Francisco's former police chief, himself is on the record saying that prison often just hardens inmates.
The DA says things could be different.
"We're trying to move this process away from being an art to being a science," Gascón said in an interview.
San Francisco runs an alternative sentencing program that takes into account defendants' histories and contributing factors to their behaviors. The thrust of the program is to try to lower the chances of recidivism. His team has tools like referrals to rehabilitation programs and even housing after release for the right candidate, and Gascón says he thinks those are in many cases more effective than just throwing someone in jail. But prosecutors can't waste time and resources extending an offer that a defendant won't accept.
"One of the ways we know how we can keep people from re-offending is giving them housing, a job, and substance control counseling," Gascón said. "But it's not enough to just offer it — the person has to accept it, but the person might not accept it."
Enter "DAStat," the current name for the system Gascón's team is trying to build thanks to a $320,000 budget allocation from the city. The system will be managed by Gascón's brand-new chief information officer, Gonzalo Gonzalez, and the DA says his team is consulting a mix of academics and in-house policy experts on what to track and how to track it. Armed with better profiles of individual defendants and more information about how different types of people perform in alternative programs over time, Gascón says his would be able to mount stronger cases — but for sending fewer people to jail, not more. If a prosecutor reading a DAStat profile decides there's still a high risk of violence, the city might ask the courts to give a defendant jail time. But if the numbers make the case that a defendant presents a low risk, that person might get alternative sentencing instead.
"This is no different from medicine or marketing," he said. "There are many metrics that can be used to predict human behavior — you're looking for factors and combining them with other factors to try and figure out what the future may look like."
Matt Gonzalez, chief attorney for the San Francisco Public Defenders office, worries that this system could create dangerous incentives in the courts. While DAStat tracks defendants, it will also track prosecutors, too, monitoring the way they handle their caseloads over time. That could lead prosecutors to pressure defendants to plead to charges to make whatever numbers their management expects of them — allowing pressure to "juke the stats" to creep into the courts from police departments, which have been using computer systems to track their own performance since New York City pioneered the CompStat program in the 1990s.
And, he added, there's no guarantee that a statistical profile of an accused criminal could predict their future behavior.
"I don't want to say that it's not relevant at all," Gonzalez said in an interview. "But I think it's overstating what you can do with that information — it's so individual and based in that one moment in time in somebody's life, and how they see themselves in it."
But DAStat should have a bright side, too. Tracking cases through the system could also create more accountability for prosecutors if that information was available to the public. Elsewhere in the country, it turns out that could be a much-needed change: Investigative journalists at the New York Times analyzed prosecutions in New York City and reported that victims and accused criminals in the Bronx were, in a disturbingly high number of instances, waiting years before cases were even brought to trial as witnesses forgot events, evidence aged, crimes went unpunished and suspects — possibly innocent ones — waited in jail.
And it seems that the criminologists studying and creating such statistical systems are well aware of the flaws and limitations of predictive analytics. A recent paper from the National Institute of Justice on Philadelphia's recidivism prediction tool stressed that the building a system to guess if someone convicted once might be convicted again is an iterative process, with variables and assumptions that should be constantly critiqued.
In fact, the San Francisco DA's office's project is just the latest initiative in a renewed emphasis that criminal justice officials have placed on performance management and data-driven decision making. In recent years, city officials — especially in major cities like New York, Chicago and San Francisco — turned from using metrics to track performance to using predictive analytics to assist in planning. City officials now lean on statistics to make choices like where to direct police resources or which of thousands of potentially unsafe buildings to inspect first.
In her book "Smart on Crime," former San Francisco DA and current state Attorney General Kamala Harris outlined an approach similar to what her successor is putting in motion. Gascón also finds precedent in Bratton's work, having served with him at the Los Angeles Police Department.
And the Manhattan district attorney's office has implemented its own statistical analysis system implemented through its Crime Strategies Unit, which is comprised of five senior prosectors whose sole jobs are to analyze crime trends in Manhattan and to implement the deployment of resources and prosecution strategies stemming from these findings.
For example, Manhattan prosecutors use arrest alerts and geocoding of incidents to determine which defendants are major drivers of crime, and to geographic patterns of crime. The idea is to keep prosecutors and police in different precincts apprised of what's going on in each precinct so that they can discern patterns and better understand how events are playing out in local communities.
"It's not about just prosecuting people," says Karen Friedman Agnifilo, who is chief of the trial division of Manhattan's office of the district attorney. "It's about prosecuting the right people," which means using whatever intelligence the DAs have gathered about suspects to determine what kind of case they should put on against them -- whether they should seek harsh penalties, or to divert them to mental health and drug counseling, for example, she said.
Agnifilo said that its CSU unit has helped to take down 15 different sets of gangs in Manhattan since 2010, and reduced the homicide rates in specific neighborhoods as a result of that. The statistical analysis also led to the creation of the quality-of-life court, where minor, non-violent offenses are tried in an effort to unclog the criminal court system.
In his book SuperCrunchers, Yale Law Professor and economist Ian Ayres argues that many aspects of decision-making in public life, including criminal justice, will be governed with a combination of intuition and statistical analysis instead of the current system of "government by chance," as he puts it.
This can also mean a government more responsive to logic and reason.
Former Philadelphia Police Commissioner John F. Timoney, who had previously worked with former Los Angeles and NYPD Police Chief William Bratton to implement CompStat in New York City, dealt with the potential for misbehavior in Philadelphia by establishing a Quality Assurance Division of statisticians. That division caught at least one police officer misclassifying crimes to make it seem as if the crime rate was declining precipitously in his precinct.
Eric Cadora, director of the Justice Mapping Center in New York City, collaborated with Laura Kurgan, a data visualization expert and associate professor of architecture at Columbia University, to create maps almost a decade ago that showed that the city and state of New York spent a million dollars a year per block in specific areas of public housing in New York City to incarcerate large numbers of people from those blocks for two to three years. Cadora went back to visit those blocks last year as part of a project that National Public Radio did on cities, and found that the city and the state had reacted to those findings. Officials put in place new services to help prisoners returning to society to adjust, to find housing, and jobs.
Tracking arrests to decide where to send police officers is one thing, but prosecutions are entirely another — in this case, probability models would be put to work in the service of deciding the individual fates an untold number of people accused of a crime. Gascón says that process will be conducted in the open, with the information prosecutors use made available to defense attorneys and the public.
"One of the things that I think is important is to have transparency in government, and in order to have that you have to have data that is timely and accurate. So apart from strategic discussions about ongoing cases and victim information, many of the things we do will be available to the public for regular review," he said.
And, he adds, all the information the prosecution uses to mount its cases will also be available to the defense, and to the public.
"It's not like we're creating a secret book," he said.