Researchers from Washington State University are conducting a study to examine whether local methodological and jurisdictional variations affect the predictive performance or validity of juvenile risk assessment tools. Many common risk assessment tools have been tested to demonstrate accuracy in predicting risk for future contact with the justice system. However, this does not necessarily translate to other jurisdictions in which the tool has not been locally validated. Because juvenile justice systems, populations, and offending patterns vary across jurisdictions, local validation may be important to ensure the risk assessment tool is accurate for the specific population with which it is being used. In addition, the local validation process often does not adjust common features of risk assessment tools to account for differences between jurisdictions. These adjustments can include changing items and response weights and developing gender-specific models and outcome-specific predictions (i.e. violent, felony, property, or drug offending). These adjustments could improve the tools' predictive performance.
The researchers in this study are analyzing risk assessment tools to customize them to attain an optimal level of predictive performance in 10 study sites. The tools that the study sites are using, whether home-grown or commercially available, are all based on the Washington State Juvenile Court Administrators Risk Assessment (WSJCA-RA) and include similar domains and items.
This project is funded under the OJJDP FY 2017 Field Initiated Research and Evaluation Program, Category 2: Small Studies and Analyses, which supports efforts to advance knowledge building and innovation in both policy and practice in juvenile justice and delinquency prevention that have been understudied or merit closer attention.
Goals and Objectives:
This study is examining how to customize risk assessment tools based on available jurisdiction data to improve their predictive performance. More specifically, the study will:
- Build site-specific datasets and an integrated multi-site dataset based on risk assessment data from 10 study sites.
- Isolate, test, and evaluate the relative impact of seven notable risk assessment features in each of the study sites. These include: (1) item selection technique, (2) weighting, (3) gender-specificity, (4) race-ethnicity neutrality, (5) outcome specificity, (6) prediction duration, and (7) jurisdiction variation.
- Develop optimized models for each study site's risk assessment instrument based on local data to achieve peak performance.
- Synthesize the common findings across the study sites and develop a set of recommendations about building risk assessment instruments that are best situated for local implementation and validation.
The researchers have finalized data sharing agreements in most project sites and are merging data submissions into a comprehensive, integrated data set. They also completed state data sets and developed preliminary prediction models. The team expects to complete data analysis in 2019. OJJDP plans to post a final technical report at the conclusion of the project.
Zachary Hamilton, Associate Professor
email@example.com | 509-358-7961
Washington State University
Benjamin Adams, Social Science Analyst
Benjamin.Adams@usdoj.gov | 202-616-3687