In Michael Lewis’ Moneyball, a data expert does what baseball scouts cannot: predict the future performance of a player with better accuracy. In addition to relying on the instincts of scouts, the Oakland A’s decided to use analytics to predict whether a player would be a top performer, thus gaining notoriety and kicking off an unprecedented winning streak.
Ex-New Jersey Attorney General Anne Milgram (now at the Arnold Foundation) made a similar point in a TED talk where she touted the value of statistics in the criminal justice system. Her argument was simple: Why not use statistics if they could help? Why wouldn’t statistics be more accurate than the individual decisions of judges? Milgram’s message is the promise of big data: a better, clearer future. And who can complain? You can’t argue with the numbers, right? But does that mean they’re fair?
“We need to think strategically about what incarceration does to someone’s level of risk,” rather than assume that risk is a fixed constant.
Attorney General Eric Holder recently announced his opposition to the use of data analytics in sentencing and corrections decisions (often called “evidence-based sentencing”). He argued that the use of empirical data—including such factors as employment and family history and neighborhood of residence—used to determine whether an offender is likely to commit another crime tends to have a disproportionately negative effect on African Americans and other minorities. His office argues that past criminal conduct—not static factors, like where someone was born or whether their parents were incarcerated—should determine sentencing. Holder’s commentary is all the more noteworthy because evidence-based sentencing is slowly being embraced by many in the judiciary and state governments on both the left and right as a way for states to reduce burgeoning prison populations and the associated high costs.
Currently, over 20 states use data-crunching risk-assessment programs for sentencing decisions, usually consisting of proprietary software whose exact methods are unknown, to determine which individuals are most likely to re-offend. The Senate and House are also considering similar tools for federal sentencing. These data programs look at a variety of factors, many of them relatively static, like criminal and employment history, age, gender, education, finances, family background, and residence.
Indiana, for example, uses the LSI-R, the legality of which was upheld by the state’s supreme court in 2010. Other states use a model called COMPAS, which uses many of the same variables as LSI-R and even includes high school grades. Others are currently considering the practice as a way to reduce the number of inmates and ensure public safety. (Many more states use or endorse similar assessments when sentencing sex offenders, and the programs have been used in parole hearings for years.) Even the American Law Institute has embraced the practice, adding it to the Model Penal Code, attesting to the tool’s legitimacy.
The use of personal information to determine sentencing is a long tradition and firmly embedded in the legal system. Judges have always been able to consider an individual’s situation, family, and history when deciding how long a prison sentence should be. But there’s no requirement that judges assess information in the same way that a computer program might—some judges may consider an individual’s history of poverty as a mitigating factor, making him more deserving of mercy, than an indicator of future crime.
Correctional departments use data-driven analyses because it’s easier and cheaper than individual assessments. It’s hard to ignore the bottom line in most states—lower recidivism and reduced costs. Plus, data on its face seems neutral. It gives a scientific basis for a system that is otherwise based on gut judgments. Given the nation’s incarceration crisis, which has both humanitarian and financial implications, the stakes are high for correctional departments to reduce their inmate population in a way that does not appear to endanger communities.
Attorney General Holder does not oppose the use of risk-assessment data such as drug addiction or criminal history; nor does he oppose the use of data more generally to track trends or to reduce incarceration overall. But Holder argues that the use of this data has the potential to disparately impact minorities and further widen the well-established link between socio-economic status, race, and incarceration. Currently, black men are six times more likely to be incarcerated than white men, a statistic that risk-assessment software is likely to exacerbate because it weighs the very factors that create de facto segregation in America.
The fear is that risk-assessment will widen the gap between who is incarcerated—racial minorities and the poor—and who is not. Public perception is already biased against the notion of the “superpredator”—a career criminal who will continue to break the law no matter what. Crime is widely perceived as an identity, rather than a bad act deserving of just punishment.
Evidence-based sentencing runs the risk of equating people with their bad acts. Judges may use the scores to punish those deemed “high risk,” sentencing them to long prison terms based on their education or marital status. Sonja Starr, a law professor at the University of Michigan, told me that evidence-based sentencing uses many factors that people cannot change and, thus, is tantamount to punishing people for being poor or living in a bad neighborhood. One example she points to is Pennsylvania, which automatically adds risk assessment points to an individual’s “score” for living in urban areas like Pittsburgh or Philadelphia and detracts points for living in rural areas; it just so happens that those metropolitan areas have more minorities than other regions of the state. Even further, these scores take into account things like family history of incarceration, which is a bigger problem for minorities—black children are more than seven times more likely to have a parent in prison than white children.
Data always relies on averages. As a result, some people are bound to behave differently than the data predicts (something that data proponents don’t dispute). The predictions may not be accurate, or at least may be no more accurate than current measures because they look only at past behavior and fixed traits, rather than looking forward to the effect of rehabilitative programs, education, or long prison sentences. Even under a purely utilitarian analysis that sought to reduce crime without regard to moral factors, Starr’s study indicates that evidence-based sentencing is no more likely to prevent crime than relying on the personal assessment of judges.
The fear is that risk-assessment will widen the gap between who is incarcerated—racial minorities and the poor—and who is not.
It may even be unconstitutional. Discrimination is still illegal even if supported by data. In Bearden v. Georgia, the Supreme Court held that a state couldn’t revoke parole because an individual lost his job and was unable to pay a fine; the broader implication of the holding is that someone cannot be held criminally culpable for being poor. Many evidence-based sentencing programs also assign value to gender (men will always be a higher risk than females), which may violate gender discrimination laws. Because there’s no real transparency to the risk-assessment models, there’s no real way for a defendant to appeal a high score, which violates due process rights.
Opponents of evidence-based sentencing agree that analytics have a place; Holder’s office concurs with its use to reduce recidivism and provide better re-entry services for inmates. Even Starr agrees that data collection has a place in the justice system. She argues that such information is better used to prevent crime, and would like to shift the conversation to how data can be used to decide what sentences will actually reduce an individual’s likelihood to recidivate. “We need to think strategically about what incarceration does to someone’s level of risk,” she says, rather than assume that risk is a fixed constant. Even further, perhaps data should be focused on helping people re-enter communities in a way that is less conducive to criminal activities.
Most importantly, labeling something as “science” tends to give it a patina of impenetrability. Starr argues that the “veneer of scientific research” may cause judges to give risk-assessment scores more validity than they merit in real life. Once society believes that a practice is unbiased and “scientific,” people are more likely to assume that it has a rational basis. Examining Anne Milgram’s proposed ranking system, where “dangerous” individuals are given a “red zone” ranking, it’s easy to see how someone looking at this information might be tempted to give it more weight than other, more subjective, factors. What judge would release someone with a high “risk” score and tempt public censure for letting someone out “too soon”?
Data analytics is probably here to say, and the quick adoption of the method attests to its popularity. While Holder’s announcement probably won’t have immediate implications for states already using the programs, it will hopefully cause decision-makers to pause and ask the value of the data they are commissioning. For those who might argue that judges discriminate anyway, they should consider that codifying such discrimination has larger, and more insidious, implications for everyone.