The use of big data to track and analyze human behavior has crept into perhaps the most consequential and sensitive of contexts in society — criminal sentencing.

While predictions of future criminality have thus far been made primarily through clinical assessments, often involving in-person interviews, the criminal justice system is turning to analytics to make these predictions. Under this actuarial approach, an individual's risk of recidivism is computed by aggregating the rates of recidivism for individuals possessing the same group characteristics. For example, the risk profile of a poor high-school educated white man will be calculated according to the rates of recidivism for individuals with the same socio-economic, educational, race and gender characteristics.


At least 20 states have adopted this "Moneyball" approach in sentencing. Influential legal organizations have endorsed these risk-assessment tools as well. Moreover, members of Congress have proposed bipartisan legislation that would require the Department of Justice to develop a risk-assessment instrument that could help identify low-risk prisoners who could earn early release into community-based or home confinement.

This statistical method is attractive because it is said to distinguish with greater accuracy the high-risk offender from the low-risk offender and thereby promotes a more efficient allocation of strained and costly sentencing resources, reserving incarceration for the most dangerous. Further, empirically-based risk profiles are objective and consistent regardless of who crunches the numbers and thus can promote more uniform sentencing decisions.

U.S. Attorney General Eric Holder has responded to the increased interest in risk-assessment tools by acknowledging that the "'Big Data' movement" has the potential to "make our system far more effective than it is today." But he also expressed concern that these tools "may exacerbate unwarranted and unjust disparities." Indeed, those who benefit from "low risk" diversion may be disproportionately white, white-collar criminals.

Mr. Holder is right to be cautious. First, some factors that can contribute to an individual's risk profile, such as race and sex, may be constitutionally off-limits. Moreover, Congress has declared by statute that the federal sentencing guidelines are to be "entirely neutral as to the race, sex, national origin, creed and socioeconomic status of offenders." While no current risk-assessment tool expressly uses race as part of the composite risk score, scholars have stated that race at least in theory could be so used because of the purportedly high correlation between race and criminality. Even accepting this correlation, constitutional and statutory limits should nonetheless preclude race and other immutable characteristics from inclusion in risk-assessment instruments.

Second, risk-assessment tools may spit out a higher risk score because of factors an individual cannot control, such as race and age, and factors the individual cannot functionally control, such as socio-economic status and characteristics that relate to socio-economic status, including educational achievement and employment history. The poor often lack the meaningful mobility to escape their marginalized positions and circumstances. To assign a higher risk profile on these grounds would be to sever the connection between punishment and individual responsibility, and to penalize someone for being economically and physically stuck.

Third, risk-assessments closely venture into "pre-crime" territory, subjecting individuals to enhanced punishment because of a propensity of individuals with similar characteristics to commit crimes. These tools fail to recognize that the individual may defy his or her group's average. Moreover, people can and do change. The criminal justice system should assist the ability of the individual to change and beat the odds, rather than presume future criminality is a foregone conclusion. Indeed, actuarial assessments could be deployed to individualize and make most effective programmatic support for defendants.

Risk-assessments could focus on an individual's adult criminal history, which may not be as problematic as other factors because the convictions are based on the person's past conduct as adjudicated in court. That said, the relevance of criminal history fades with time, meaning that stale criminal convictions may be weaker indicators of future criminal conduct. Even if adult criminal history is the cleanest of factors, it may be captured anyway in sentencing determinations that already account for an individual's prior criminal acts.

In the end, the advantages of risk-assessment tools and growing concerns about mass incarceration should not blind the federal government to the principled outer bounds of a fair and just sentencing scheme. These competing considerations suggest a skeptical approach to using risk-assessment tools in sentencing.

Dawinder Sidhu is an associate professor of law at the University of New Mexico School of Law. He was born and raised in Maryland and is a member of the Maryland bar. His email is sidhu@law.unm.edu; Twitter: @profsidhu.