Optimizing Risk Assessments Across State Lines

• In Event: The Latest Developments in Risk Assessments

Abstract

Risk assessments offer criminal justice practitioners a scientific approach to gauge individuals’ likelihood of recidivating. Many of the contemporary tools currently used in the field assume that the items, weights, and risk levels translate when applied to another jurisdiction’s population. Recent research has identified substantial jurisdictional variations, with regard to demographics, local statues, an item response prevalence, which influence an assessment’s accuracy. However, these concepts have not been explored thoroughly; likely due to the size and scope of the study needed to answer basic research questions. Funded by OJJDP, we collected assessment and recidivism data for over 500,000 youth, across 11 states. The current study sought to examine how the predicative accuracy of an assessment varies with regard to jurisdiction, gender, race/ethnicity, outcome type (violent, property, drug, sex), definition (charge vs. adjudication), and duration of prediction (i.e. 12, 24, 36 months). To complete this process we compare current assessment scoring with models created using regularized regression techniques and M-fold validation. These examinations allow for a more robust procedure of risk assessment development and optimization, providing customization options for implementing jurisdictions. A further discussion is offered for practitioners to create stronger tools to better assess reoffending risk, needs, and more effectively allocate resources.

Authors

• Zachary Hamilton, Washington State University

• Eunsuhk Choi, Washington State University