Why It Matters
The Public Safety Assessment (PSA) was developed as a generalized tool using data from approximately 750,000 cases across 300 jurisdictions nationwide. A pretrial assessment with good generalizability means its results are valid for multiple, diverse populations. RTI International (RTI) will study how generalization affects the PSA’s predictive validity and statistical bias when applied to a single jurisdiction. Localizing the PSA has the potential to improve the predictive validity and lessen statistical bias on a site-specific basis.
For questions about PSA performance or validation, contact Stephen Tueller. For questions about PSA localization, contact Anna Godwin.
Research products resulting from studying PSA performance will include:
- Validation and bias reports for each Advancing Pretrial Policy and Research (APPR) Research-Action Site
- Instructions for scoring the localized PSA should a localized PSA improve upon the original PSA
- The code for the statistical models used to localize the PSA, which will be archived on the Open Science Framework website to ensure transparency and reproducibility
- A freely available, open-source R software package that can be used to produce validity and bias reports for data that has been prepared according to definitions in PSA Implementation Guides
Research Methods
RTI will use the following research methods to study PSA performance:
- Predictive validity: Measure how accurately a tool estimates the likelihood of appearing in court and remaining arrest-free during the pretrial period
- Statistical bias: Measure how the tool performs differently across gender, race/ethnicity, and other protected groups of people
- Localization: Use local data to adapt the PSA for use in that jurisdiction, then measure how a localized PSA performs relative to the original PSA
Read the technical details of RTI’s analysis plan.
Research
Research products will be posted here when they become available.
