Big Data Tries to Shrink the Prison Population

President Obama has made reducing America’s enormous prison population a centerpiece of his second term, supporting reforms that could roll back jail time for up to 46,000 federal inmates and pushing new rules for sentencing. His chief lieutenant in this campaign, Attorney General Eric Holder, has been outspoken about the relationship between criminal justice and civil rights. But now Holder is questioning a method of thinning the prison population that has been embraced across the country.

Over the past decade, dozens of states have begun using massive databases of crime statistics to determine the odds that an offender will commit another offense. These risk-assessment tools take into account factors like antisocial behavior, prior convictions and age of first arrest. Judges, corrections workers and parole officers have in turn used these assessments to cut sentences for low-risk offenders and reduce prison time for others. Reliance on the sort of statistical analysis popularized by Michael Lewis’ 2003 book Moneyball has increased everywhere from conservative Texas to liberal Vermont as a way to smartly reduce the prison population while lowering costs for cash-strapped states.

But just as bipartisan legislation moving through Congress is set to mandate the use of risk-assessment tools in federal sentencing and corrections, Holder is slamming the brakes. His worry: that using factors that inmates can’t amend, like geography or criminal history, as part of the formula will codify existing biases that already disproportionately affect African Americans, resulting in 20% longer prison terms for young black men than for other offenders.

“I’m really concerned that this could lead us back to a place we don’t want to go,” Holder says. The stakes are high. The number of U.S. prisoners has risen 500% since 1980, to more than 2.2 million in 2012; 95% of them will be released at some point.

Holder isn’t completely against Moneyball-style data analysis. He supports risk-assessment tools built around inmates’ behavior, measuring things like personality patterns and known associates. And he’s fine with statistical analysis of criminals already in jail–so-called back-end assessments–as long as all offenders, including high-risk ones, can try to earn early release.

Supporters of the broad use of data-driven criminal justice–and there are many–say Holder’s narrow approach won’t work. “If you wait until the back end, it becomes exponentially harder to solve the problem,” says former New Jersey attorney general Anne Milgram, who is now at the Arnold Foundation, where she is building risk-assessment tools for law enforcement. Prior convictions and the age of first arrest are powerful risk factors for reoffending and should inform whether a convict goes to prison or not, experts say.

The states are unlikely to be persuaded by Holder, not least because data-driven corrections have been good for the bottom line. Arkansas’ 2011 Public Safety Improvement Act, which requires risk assessments in corrections, is projected to help save the state $875 million through 2020, while similar reforms in Kentucky are projected to save it $422 million over 10 years, according to the Pew Center on the States. Rhode Island has seen its prison population drop 19% in the past five years, thanks in part to risk-assessment programs, according to the state’s director of corrections, A.T. Wall.

Holder’s best chance of success is at the federal level, where more than 200,000 prisoners are incarcerated. Time is of the essence: bills from the House and Senate that reform sentencing and corrections are expected to be reconciled quickly and sent to Obama for his signature before the end of the year.