Strengthening data-driven pretrial release in New Jersey

In collaboration with the New Jersey Administrative Office of the Courts, the applicant will examine and revise use of the Public Safety Assessment (PSA). PSA is a pretrial risk assessment tool designed to yield objective pretrial release recommendations but criticized for its inherent bias. Concerns include exacerbating racial disparities in the criminal justice system and failure to identify risk of serious offenses, particularly gun violence or domestic violence. The goal is to refine the tool to support the state’s pretrial reform objectives of limiting the use of pretrial detention, and reducing racial disparities in the system, while maintaining public safety. Researchers at the University of Chicago Crime Lab will collaborate with the New Jersey Courts Information Technology Office to access administrative data. They will use machine learning and other data analysis to identify and implement tool modifications reflecting stakeholder policy choices commensurate with pretrial reform goals. Immediately following implementation, they will assess the impact of the modified tool on pretrial release decisions using quasi-experimental methods. Twelve months after implementation, the research team will conduct a re-validation of the modified tool’s performance. The researchers will identify PSA concerns and deficiencies through discussions with New Jersey Courts and other stakeholders, including the New Jersey Office of the Attorney General, the Public Defender’s Office, the New Jersey Coalition to End Domestic Violence, and the ACLU of New Jersey. They will diagnose policy and analysis choices in the PSA’s development and implementation contributing to observed performance. They propose to build alternative statistical models to test effects of alternative analysis choices on tool performance. Modified tool performance will be assessed on three measures: (1) statistical accuracy in predicting failure to appear and new criminal arrests (violent, domestic violence, and other); (2) changes to release recommendations (percentage of defendants recommended for restrictive release conditions or detention); and (3) racial disparities in statistical accuracy or release recommendations. Finally, a quasi experimental strategy will be used to assess modified tool performance in relation to judicial decisions (for increased agreement with tool recommendations), apart from other concurrent shifts in the criminal justice
environment. A differences-in-differences approach will test whether pretrial decisions have a stronger correlation after the implementation of the modified tool. The team includes experts in statistics and courts research at Carnegie Mellon University and University of Maryland. In addition to the modified pretrial risk assessment tool, they will develop a framework of statistical methods to characterize how altering these choices in the development of risk assessment tools can impact important policy objectives. This will result in a ‘playbook’ for policymakers as procurers of these tools, and a ‘playbook’ for data scientists as partners in their development. Dissemination includes policy papers on the Crime Lab’s website and research journal articles. "Note: This project contains a research and/or development component, as defined in applicable law," and complies with Part 200 Uniform Requirements - 2 CFR  200.210(a)(14). CA/NCF