Race not a key factor in test to predict who will commit future crimes

Efforts to relieve the human and financial toll of prison overcrowding often rely on predictive tools that estimate how likely an inmate is to reoffend based on such risk factors as criminal history, education or employment problems and substance abuse.

One way to relieve overcrowding in prisons and jails is to assess offenders on whether they will commit future crimes if released. (Photo by California Watch)

One way to relieve overcrowding in prisons and jails is to assess offenders on whether they will commit future crimes if released. (Photo by California Watch)

Critics of this approach speculate that the tools unfairly target racial minorities and that “risk assessment” is little more than a race assessment that will expose minorities to harsher penalties.

But a new UC Berkeley study comparing recidivism risk scores among blacks and whites challenges this assumption.

“We are recommending that jurisdictions not abandon risk assessment based on an untested fear that the use of these instruments will exacerbate racial disparities,” said study co-author Jennifer Skeem, a social welfare professor at UC Berkeley. “But we strongly recommend that jurisdictions directly test whether the instruments they are using are subject to predictive bias by race.”

In the largest examination yet of whether risk assessments are racially biased in predicting future crimes, Skeem and Christopher Lowenkamp, a criminal justice researcher with the Administrative Office of the U.S. Courts, used data on nearly 35,000 federal prisoners to test for potential bias in the federal Post Conviction Risk Assessment (PCRA), a statistical model aimed at determining the level of supervision and kinds of services an inmate may need upon release.

Their findings, published in the November issue of the journal Criminology, found that the PCRA system strongly and similarly predicted recidivism regardless of whether the offender was black or white.

Using PCRA scores to predict post-release arrests based on FBI records, Skeem and Lowenkamp found that a given score on the PCRA meant basically the same probability of reoffending, across racial groups. For example, they found someone classified as a “moderate risk” had a 48 percent likelihood of re-arrest regardless of whether they were black or white.

“In efforts to unwind mass incarceration without increasing the crime rate, it is critical to take racial fairness into account because, in the U.S., young black men are six times more likely to be imprisoned than young white men,” Skeem said. “But rejecting risk assessment measures will not necessarily make the system more fair.”

The study found that, on average, black offenders scored moderately higher than whites on the recidivism risk scale, mostly because they were more likely to have a criminal history. But that difference was expected given that criminal history is among the most heavily weighted factors in risk assessment tools.

Compared to, say, sentencing guidelines, risk assessment tools can be viewed as more equitable because they balance criminal history, which skews higher among blacks, against other factors, Skeem said.

“Criminal history is already routinely considered by judges and sentencing guidelines,” Skeem said.  “So any racial differences in criminal history are often embedded in the existing sanctioning process.  It is not at all clear that replacing those considerations with a broader-based risk assessment that includes factors other than criminal history would make racial disparities in incarceration worse than they are now.”

Recent decades have seen a rise in the use of data-driven risk assessments across U.S jurisdictions that consider not only crimes committed, but crimes deemed likely to be committed in the future. Under the Sentencing Reform and Corrections Act, currently awaiting a vote in the U.S. Senate, all federal prisoners are to be rated according to their risk of committing future crimes.

But the issue of ranking offenders according to their risk scores is a controversial one. In 2014, U.S. Attorney General Eric Holder called for scrutiny of data-driven criminal justice programs, cautioning that they may adversely affect low-income and minority groups:

“By basing sentencing decisions on static factors and immutable characteristics, like the defendant’s education level, socioeconomic background or neighborhood, they may exacerbate unwarranted and unjust disparities that are already far too common in our criminal justice system and in our society,” he said.

Earlier this year, the news outlet ProPublica published an article that suggests the risk assessment methodology, including its computer algorithms, is biased against blacks. But a peer-reviewed analysis of the same data published in the September issue of the Federal Probation Journal indicates that those conclusions were erroneous, and this latest UC Berkeley study reinforces those results.

“Our study raises the possibility that, if applied objectively, risk assessment could actually reduce rather than increase racial biases in making decisions about criminal justice sanctions,” Skeem said.