Pretrial risk assessment tools found to be subjective and biased

In their attempts to reform the cash bail system, jurisdictions across the country are turning to automated pretrial risk assessment tools that ‘predict’ if a defendant will be arrested for a new crime while waiting for trial or will fail to appear in court as scheduled.

The Science, Technology, and Public Policy (STPP) program at the University of Michigan’s Ford School of Public Policy has analyzed pretrial risk assessment tools and finds serious concerns with their validity and bias, concluding that they should play no role in pretrial release decisions.

While some policymakers believe that the tools are a more equitable alternative to cash bail, according to a new policy memo, issued as a part of STPP’s Community Partnerships Initiative, there is “substantial evidence that pretrial risk assessment tools replicate the racial and socioeconomic disparities that bail reform seeks to address.”

Almost every state has enacted reforms related to pretrial risk and detention, with Michigan legislators being the latest to consider changes. The report notes, “Today, defendants awaiting trial make up almost half of the jail population in the state. Many of these people do not pose a threat to public safety or a high likelihood of flight risk.”

In the past, decision-makers such as judges or parole officers were responsible for evaluating whether a person was likely to reoffend or fail to appear in court, a system that was vulnerable to individual biases. More recently jurisdictions turned to calculating risk scores or using a standardized framework to estimate risk.

Newer automated tools use inputs like age, prior convictions, employment status, zip code, and neighborhood crime, to compare a person’s individual data to past cases of people with similar data to calculate a risk score, which claims to measure a person’s likelihood of reoffending or failing to return to court.

Because data and calculations are viewed as objective, pretrial risk assessment tools are often applauded as scientific and unbiased. However, the memo raises major concerns about the tools’ impartiality and accuracy.

First, the data these tools use comes from the same biased processes they are meant to address. Prior arrests, convictions, or sentences are often a result of decisions made by the people these tools are supposed to replace. They often reflect outdated policing and sentencing practices, as well as racial and socioeconomic disparities in profiling, policing, and social policy. This can result in discriminatory risk scores for defendants of color and low-income defendants.

In addition, the tools dramatically overestimate the likelihood of missing a court date, and they ignore factors contributing to why people fail to appear in court. People rarely miss court for intentional reasons; instead, they fail to appear due to inconveniences like missing a bus, not finding childcare, or not getting time off work. The memo argues that jurisdictions should invest in addressing these root causes rather than holding people before trial due to faulty risk assessment.

The Michigan State Legislature will soon be considering a bill package to reform pretrial detention. Previous drafts have included a requirement or suggestion for pretrial risk assessment tools, signaling the need for policymakers to be aware of their effects. The goal of the memo is to “help Michigan state and local legislators better understand these concerns, discourage or prevent their use, and especially prevent legislators from codifying tool use into law,” said McCoy.

“By relying on biased data and focusing on negative outcomes, pretrial risk assessment tools do not accurately predict pretrial violence or flight risk,” the memo concludes. “They also do not answer the questions that would most likely assist administrators seeking to reduce pretrial detention.”

Read the full research brief here.

More news from the Ford School