Our latest paper is now available online ahead of print in Criminology & Public Policy. 

Nearly 500,000 unconvicted people sit in U.S. jails awaiting trial. Many of them unnecessarily. Risk assessment tools exist but in the federal system, judges rarely see them. Only ~15% of districts include risk scores in pretrial reports. 

We ran a policy simulation using causal machine learning on ~147,000 federal cases. We asked: what would happen if, instead of relying on magistrate judges’ discretion, a policy presumptively released defendants classified as relatively low risk by a risk assessment instrument? 

The results were clear. A risk-based presumptive release policy could reduce detention by 34%, increase successful releases by 32%, and save $3.5 billion—with only a 1.6% increase in pretrial failures. Black defendants would benefit disproportionately. 

Structuring, not replacing, judicial judgment around validated risk assessment tools can make pretrial decisions fairer, less costly, and safer. 

Access the article here: