Aims: To assess whether Youth Level of Service/Case Management Inventory Australian Adaptation (YLS/CMI-AA) risk/needs data improve recidivism prediction for young offenders under community supervision, compared to static risk data from the Bureau’s Reoffending Database (ROD).
Method: The analysis included all 1,050 young offenders who commenced a supervised community order (other than bail or parole) in 2014 with a valid YLS/CMI-AA and ROD record. Recidivism was defined as a new proven offence within 12 months of order commencement. Logistic regression assessed the individual and collective relationships of static risk factors and YLS/CMI-AA scores to recidivism. Area Under the Curve (AUC), model fit indices and multiple cross-validation methods were used to evaluate the models.
Results: Interactions between variables in models built with the full sample necessitated that separate models be built for Indigenous and non-Indigenous offenders. For non-Indigenous offenders, the AUC for the combined (ROD with YLS/CMI-AA) model (.767, 95% CI (.728, .807)) was within the acceptable range (0.7-0.8) but did not significantly outperform the ROD-only model (.740, (95% CI .698, .781)). For Indigenous offenders, AUCs were significantly lower than for non-Indigenous offenders, below the acceptable range, and also showed no significant benefit from combining YLS/CMI-AA and ROD data. Compared with AUCs for the combined model, cross-validated AUCs were lower, and corresponding AUCs for the 2013 cohort were inconsistent.
Conclusion: YLS/CMI-AA data did not significantly improve the predictive accuracy of static risk-based models of recidivism for Indigenous or non-Indigenous offenders. Validation methods suggested that the results may not generalise beyond the current cohort.