Exploring a Generation's Bias: A Multi-State Evaluation of the LSI-R

• In Event: Influencing Policy through Research: Investigations of Correctional Programming and Assessments

Abstract

The Level of Service Inventory-Revised (LSI-R) is one of the most widely used risk-need assessments (RNAs) in North America. Discordantly, research shows that implementing a tool across diverse jurisdictions without making substantial adjustments diminishes its performance. Further, the LSI-R's heavy reliance on criminal history indicators to maintain predictive accuracy which can contribute to racial/ethnic and gender bias in risk prediction. These findings have yet to be thoroughly examined, as most studies rely on samples of a single demographic group or state (e.g., Jimenez et al., 2018) or are meta-analyses that produce a broad range of effect sizes (e.g., Olver et al., 2014). To address this gap in knowledge, the current study examines the predictive accuracy of the LSI-R across jurisdictions from four states: Pennsylvania, North Dakota, Minnesota, and Maryland. We compare its performance to a localized RNA, the Static Risk Offender Needs Guide – Revised (STRONG-R), to understand how instrument characteristics affect predictive validity across geographic and demographic subgroups.

Authors

• John Michael Ursino, University of Nebraska Omaha

• Zachary Hamilton, University of Nebraska Omaha