The Center for Court Innovation, a New York-based criminal justice reform agency, recently released a study on supposedly race-neutral risk assessment tools. These tools are widely used by courts to determine whether defendants should be granted bail or held in jail until trial.
The tools parse each defendant’s demographics and criminal history, the seriousness of the current charge, and any past interactions with the justice system, such as arrests without charges. They then spit out a number corresponding to the theoretical risk that the defendant will commit another crime if released on bail.
Risk-assessment tools are sold as objective, data-based decision-making algorithms that replace subjective human judgment with cold facts. The idea was to prevent racial bias in decisions about bail and pretrial release. But do they simply reinforce existing stereotypes?
There’s reason to believe they do. In 2016, the independent newsroom ProPublica published a study of Broward County, Florida’s COMPAS tool.
When COMPAS was used, African-American defendants were nearly twice as likely as whites to be labeled high-risk but not commit additional crimes — the “false positive” effect. White defendants were more likely to be labeled low-risk but commit new crimes.
The Center for Court Innovation created its own risk-assessment algorithm and used it to predict who would commit further crimes if let out on bail. No real-world bail decisions were made using the tool. They simply wanted to see how accurate its predictions would be.
The results were the same as ProPublica’s. African-Americans and Hispanics were falsely labeled high-risk much more often than whites.
Why? The thought is that some things considered by the tools, such as arrests that didn’t lead to charges, are the result of a history of unduly aggressive law enforcement in African-American neighborhoods. Moreover, African-Americans have historically been more likely to be arrested, charged and convicted than white peers. Considering past run-ins with law enforcement in the risk assessment may carry these past injustices forward.
Outcomes based on reason, not bias, may require a different approach
The Center for Court Innovation found that it could address the racial disparity by changing the scenario to reduce pre-trial detention overall. Pre-trial detention would only be ordered for those charged with violent felonies or domestic violence. People charged with nonviolent crimes would simply be released without bail.
After that, the risk-assessment tool would be used to predict who among those charged with violent crimes was likely to commit additional crimes if released.
This two-step process reduced pre-trial detention by over half and eliminated the racial disparities in false-positives.
It may indicate how racially biased the bail system is when an effective way to reduce the bias is to eliminate bail for the majority of defendants.
