Skip Navigation

Encoded Inequity: Why Recidivism Risk Assessments Should Stay Out Of The Courtroom


As “smart” data-driven approaches have become increasingly common and successful in police departments across the country, a movement to import similar methods into courtrooms has gained traction. Risk assessment scores use data obtained through a questionnaire or scraped from defendants’ criminal records to rate their probability of recidivism—the tendency of an offender to commit another crime. Already widely used at the state level, these metrics inform a variety of decisions within the courtroom: setting bond, granting parole, and—in nine states—sentencing inmates. Risk assessments provide genuine value in the pre-sentencing phases of the criminal justice process; however, concerns about equal protection under the law and the absence of a relationship between longer sentences and lower rates of recidivism should make us wary about their use in sentencing.

In today’s crowded jails and chaotic courthouses, risk assessments are useful metrics: According to the Marshall Project, states spend more than $13 billion annually on pre-trial detention, which accounts for as many as 443,000 defendants on any given day. Yet numerous studies have shown that many of these detentions are unnecessary. The Marshall Project and Stanford professor Sharad Goel suggest that by using risk assessments designed to evaluate the flight risk of defendants, jurisdictions around the country can reduce their pre-trial detention populations by as much as 50 percent with only marginal increases in the number of defendants who fail to appear at trial. Considering that 57 percent of the $80 billion spent annually on incarceration comes from state governments, reforming the pre-trial detention process could lead to a windfall. Risk assessments can also ensure that jurisdictions’ limited resources for rehabilitation, such as re-entry programs for ex-offenders, are more responsibly allocated. Risk scores can provide a valuable way to determine which inmates might benefit from strict detention and which are already at low risk of recidivism.

Despite the potential for risk assessments to improve our justice system, the potential for misuse looms large. An investigation by ProPublica found that the process of scoring the likelihood of recidivism could implicitly encode racial bias. While questionnaires do not directly ask about race, the score-calculating algorithms target factors that are disproportionately common in communities of color, such as lack of education, uncertain job status, and childhood spent in a single-parent household. But the conclusions of the piece are highly disputed: The Federal Probation Journal published an article detailing a laundry list of errors in the article, including the misuse of one sample of inmates from Florida, the unexplained omission of another group, and a failure to discuss standards in the field for testing the existence of bias.

However, even if risk assessments are racially neutral, they are inappropriate for the sentencing process. True to their name, risk assessments evaluate only one factor that might contribute to a defendant’s sentence: risk of recidivism. On the other hand, a sentence is meant to reflect retribution, rehabilitation, and deterrence proportional to the crime being addressed in court, not for a crime an algorithm predicts that the perpetrator may commit in the future. Finally, there is little evidence that suggests a relationship between longer sentences and lower rates of recidivism, leaving no practical reason as to why risk assessments should be used to justify longer sentences.

Fortunately, there is a viable path toward legally challenging the use of risk assessments for sentencing. In 1976, the Supreme Court established a precedent for rejecting statistically supported generalizations based on gender in Craig v. Boren. As part of the ruling, an Oklahoma statute that allowed only women to purchase a limited selection of beer at age 18 was struck down, even though existing statistical evidence concluded that men between 18 and 21 were over 10 times more likely than their female peers to drive while drunk. As a Harvard paper argues, factoring gender into risk assessments could be similarly unconstitutional even if it makes the algorithm stronger—an example of the legal system valuing fairness over accuracy. If the effect of removing defendants’ genders in risk assessment calculations is significant, a legal mandate to remove gender from risk assessments could weaken the calculations enough to encourage states to reduce their reliance on them.

Unfortunately, such an alteration could render risk assessments completely useless. Indeed, rectifying risk assessments’ less palatable aspects could water them down to the point where they can no longer aid the sectors that they have the potential to help. Nevertheless, the criminal justice system’s first priority must be to uphold its values and ensure the equal protection of all defendants. When risk assessments threaten those standards, there should not be room for them in the American justice system at all. Photo

About the Author

Nicholas Lindseth '21 is the Senior Managing Web Editor of the Brown Political Review. Nicholas can be reached at