Many RATs aim to predict the likelihood of rearrest and use prior arrest data to predict risk. However, arrest data is an inaccurate proxy for criminal activity or dangerousness, and it is racially biased.
25% of the tools in our database use questions about arrest to determine predicted risk.
Many tools claim they predict “recidivism” or “new criminal activity” through their tools, but because it is impossible to truly measure all criminal activity, the tools are actually predicting whether or not someone will be arrested.1Brandon Buskey and Andrea Woods: Making Sense of Pretrial Risk Assessments, The Champion
Arrest data does not predict the behavior of the person arrested, but rather the behavior of police.
Many factors, far beyond actual criminal activity, influence who gets arrested. Surveillance and over-policing as well as the criminalization of survival behavior, such as homelessness, mental illness, and substance use, all lead to disproportionate arrests in marginalized communities.2Beth E Richie and Andrea J Ritchie: The Crisis of Criminalization: A call for a comprehensive philanthropic response, The Barnard Center for Research on Women
Sandy Mason and Megan Stevenson conducted a comprehensive analysis of misdemeanor cases across the United States and found that there is “profound racial disparity” in the arrest rate for most misdemeanors. Black individuals are disproportionately over-arrested – for some charges, the arrest rate of Black individuals is almost five times as high as that for white individuals. Furthermore, this racial disparity had remained consistent over the past 37 years since the 1980s.3Megan Stevenson and Sandra Mayson: Contributions: The Scale of Misdemeanor Justice, Boston University Law Review
It has been well documented that Black and Brown communities and poor communities are more heavily policed, more often arrested, and receive longer sentences than other communities, even for the same behavior.4Marc Mauer: Addressing Racial Disparities in Incarceration, The Sentencing Project This is especially true in the case of the war on drugs, as Black and Latinx individuals are arrested more often than others for charges such as marijuana use.5ACLU: Marijuana Arrests by the Numbers
This inequality in the data gathered to create a RAT has direct implications on how it generates scores.
As Karen Hao and Jonathan Stray argue: “If Black defendants are arrested at a higher rate than white defendants in the real world, they will have a higher rate of predicted arrest as well. This means they will also have higher risk scores on average, and a larger percentage of them will be labeled high-risk.”6Karen Hao and Jonathan Stray: Can you make AI fairer than a judge? Play our courtroom algorithm game, MIT Technology Review
Predicting criminal behavior, especially violence, is not straightforward. One research team argued that no predictive tool can “say with confidence whether or not a particular individual will commit a future violent act.”7Jan Chaiken, Marcia Chaiken, and William Rhodes: Predicting Violent Behavior and Classifying Violent Offenders, Understanding and Preventing Violence, Volume 4: Consequences and Control
MIT researchers assert that “neither judges nor software can know in advance who will and who won’t commit violent crime,” but because using a RAT makes predictions of violence seem much more certain than the data actually support, RATs can then “lead judges to overestimate the risk of pretrial violence and detain far more people than is justified.”8Chelsea Barabas, Karthik Dinakar, and Colin Doyle: The Problem with Risk Assessment Tools, The New York Times
And in fact, most arrests are for non-violent charges.9Alice Speri: Police make more than 10 million arrests a year, but that doesn’t mean they’re solving crimes, The Intercept
However, many tools do not separate between new arrest and new arrest for a violent charge. Even when they create a separate score for violent crime, they vastly over-predict how many people who are flagged as likely to commit a violent crime actually go on to do so.10Chelsea Barabas, Karthik Dinakar, and Colin Doyle: The Problem with Risk Assessment Tools, The New York Times
The chances of rearrest, especially for a violent charge, are very low. A national study found that overall rearrest rates for those released were only 16%, and violent felony rearrests were only 1.9%.11Shima Baradaran and Frank McIntyre: Predicting Violence, Texas Law Review