Abstract
Risk assessments are often mandated within the juvenile justice system (JJS). Yet, it is unclear whether these protocols reflect equitable clinical tools, and little is known about the community's perspectives on commonly assessed risk domains. In response, we introduced, and subsequently tested, a multifaceted definition for fairness in risk assessment. An embedded mixed-method study was conducted, such that Study 1 informed Study 2's methods, and the studies were subsequently integrated. In Study 1, caregivers (N = 22) and adolescents (N = 21; Mage = 14.28; 42.9% identified as Black, 42.6% White; 66.7% Male) involved with a JJS-diversion or probation program completed qualitative interviews on risk domains for offending behavior. Next, we examined the statistical fairness of salient risk domains from Study 1 in a sample of JJS-involved adolescents (N = 1,354; Mage = 16.04; 41.4% Black, 33.5% Hispanic, 20.2% White; 86.4% as male). An evidence-based medicine analytic approach, which was compared to artificial neural networks, tested subpopulation differences across performance indices. Overall, temperament, peer relations, cognitive styles, and school functioning emerged as salient risk domain themes across identities and informants. Subsequently, moral disengagement and delinquent peers emerged as equitable predictors of prospective violent and nonviolent rule-breaking behavior. A model comprised of these predictors was acceptable (i.e., areas under the curves ≥ 0.70; diagnostic likelihood ratios ≥ 2.0) and equitable. Artificial neural network models did not improve prediction. Risk assessments focused on moral disengagement and peer delinquency may lead to community-aligned and statistically fair assessment processes. These findings can lead to more equitable and engaging JJS risk-management approach. (PsycInfo Database Record (c) 2026 APA, all rights reserved).