TIME
DELIVERING QUALITY SUPERVISION TAKES TIME
1. Forecasting
With the nation’ s highest caseloads, DCS must analyze statistical trends to determine where to focus its limited resources. Following the Risk-Need-Responsivity( RNR) framework, the first step of productive supervision involves matching an individual’ s intervention intensity with their likelihood of reoffending( i. e., the Risk Principle). We are advancing knowledge in this area in several ways:
Integrated Dynamic Risk Assessment for Community Supervision( IDRACS)
With support from the National Institute of Justice, we are developing a new risk assessment instrument to predict the probability of felony or violent misdemeanor arrests for individuals on probation or parole. The IDRACS project has multiple objectives:( a) testing artificial intelligence / machine learning algorithms against traditional statistical models,( b) incorporating dynamic variables,( c) identifying protective factors, and( d) integrating uncertainty into predictions. In short, IDRACS will assess risk factors to inform real-time supervision strategies.
Research Partners: John Speir, Applied Research Services( ARS); Chris Inkpen, Research Triangle Institute( RTI)
Needs and Responsivity Assessment Component
In addition to risk, the RNR model emphasizes tailoring services to individuals’ criminogenic needs and their responsiveness to interventions. In this vein, the Needs and Responsivity Assessment Component will upgrade IDRACS in 2 ways:( 1) systematically capture individuals ' criminogenic and responsivity factors and( 2) automatically generate suggested services and providers matching individuals ' needs, risk levels, and responsivity factors.
Research Partner: Research Triangle Institute( RTI)
Strike a Balance
Regardless of academic findings or automated algorithms, the responsibility for implementing Person-Centered Supervision rests with officers. For instance, officers often exercise discretion when applying risk algorithms, but the impact of deviating from a risk tool needs to be clarified. Therefore, this project examines the interplay of officer risk-score overrides and recidivism.
Research Partner: James McCafferty, Kennesaw State University( KSU)
Human-Machine Synergy: Machine Learning Engagement and Nurturing Peer Mentorship for Dignified Recovery in Community Reintegration( MEND-R)
We are utilizing machine learning to predict substance use treatment outcomes in probation and parole. In addition, we analyze administrative data to examine the features that predict substance use treatment outcomes among Matrix and DRC participants. Substance use disorders among justice-involved individuals result in high rates of relapse, overdose, and incarceration. Yet, probation and parole programs lack structured tools to personalize interventions and address urgent unmet needs. This project develops and evaluates“ MEND-R”, a novel decision support tool that integrates machine learning and peer mentorship to enhance treatment engagement in Georgia’ s Day Reporting Centers. By complementing experience-based decision-making with data-driven strategies, MEND-R aims to optimize resource allocation, enhance treatment engagement, reduce relapse, and support long-term recovery for individuals at high risk of suspension or dropout from community supervision programs.
Research Partner: Frances Chen, Georgia State University( GSU)
DCS FY25 Research Agenda Page 3