Big data can help states decide whom to release from prison
IN MONTY PYTHON'S "Life of Brian", the Jerusalem crowd picks wrongdoers for Pontius Pilate to release according to whether their names begin with "r", since they find it amusing that the Roman governor cannot pronounce that letter. ("Welease Woger!"). Two thousand years later, America aims to select prisoners for parole by more rational criteria, such as "Are they likely to re-offend?"
It turns out, however, that granting parole wisely is hard. Parole boards may be biased, perhaps without realising it. In general, they tend to overestimate the likelihood that a prisoner will re-offend, says Lance Lowry of AFSCME Texas Correctional Employees, a warders' union. Many fear that if they free a thug who then commits an atrocity, their reputation will be ruined.
This makes them err on the side of severity. In Ohio, for example, a paroled murderer was arrested last year for allegedly murdering a 13 year old girl. (He later died in custody.) The parole board took a beating in the press. Being granted parole in Ohio is now only slightly more likely than winning the lotto, says Barry Wilford, a local lawyer. Among applicants given hearings, in some months less than 1% are released. (In neighbouring West Virginia the average is 48%.)
Help may be at hand, in the form of "risk-assessment" software, which crunches data to estimate the likelihood a prisoner will re-offend. Such software tends to increase the proportion of applicants who are granted parole while also reducing the proportion who re-offend. Two such programmes, LSI-R and LS/CMI, appear to reduce parolee recidivism by about 15%. Developed by Multi-Health Systems, a Canadian firm, they were used to assess 775,000 parole applications in America in 2012. Four-fifths of parole boards now use similar technology, says Joan Petersilia of Stanford University.
The data that matter include the prisoner's age at first arrest, his education, the nature of his crime, his behaviour in prison, his friends' criminal records, the results of psychometric tests and even the sobriety of his mother while he was in the womb. The software estimates the probability that an inmate will relapse by comparing his profile with many others. The American version of LS/CMI, for example, holds data on 135,000 (and counting) parolees.
It is better to be guided by software than one's gut, says Olivia Craven, head of the Idaho Commission of Pardons and Parole. Donna Sytek of the New Hampshire Parole Board agrees. Unaided, parole board members rely too much on their personal experiences and make inconsistent decisions, she says.
Software can be used to help make better decisions not only about whom to release, but also about how to manage parolees on the outside. South Carolina uses a programme called COMPAS, developed by Northpointe Inc, an American firm, to help with rehabilitation. It tells officials which factors are likely to be "feeding into" a person's criminal behaviour, says Shaunita Grase, the official in charge of making it work. If COMPAS concludes that a parole applicant's delinquency was fuelled more by lack of education than, say, household strife, it might make sense to let him live at home as long as he attends evening classes. If his education is irrelevant, limited resources might be better spent on, say, anger-management training.
In some cases, software may actually reduce crime. ORAS, a programme designed for the Ohio Department of Rehabilitation and Correction, flags the low-risk criminals whose likelihood of reoffending actually increases the longer they are locked up alongside violent criminals. So says Jennifer Lux, an ORAS designer at the University of Cincinnati's Centre for Criminal Justice Research.
Some officials dislike all this new technology, notes Mr Lowry. Some resent having their gut feelings overruled. Others are sorry that software makes it harder to disguise political favours, such as respecting the wish of a state senator to see the man who burgled his house denied parole.
However, it meets a need. America's prisons are overcrowded--with less than a 20th of the world's population, it locks up nearly a quarter of prisoners--and ruinously expensive. If software helps cut the cost of incarceration without endangering the public, states will surely use it.
Click here to subscribe to The Economist
