Summary
Aim
To assess whether it is possible to screen juvenile offenders for recidivism risk from information readily available at the time of cautioning a young offender.
Method
Data on all 8,537 juveniles cautioned by police or courts in 2006 were analysed using logistic regression. The dependent variable in the logistic regression model was a binary variable measuring reoffending. The potential predictors included number of previous cautions, conferences or court appearances, jurisdiction issuing the caution (court vs. police), Indigenous status, gender, age at index caution, offence type, offence seriousness, prior violence, remoteness (ARIA) and social and economic disadvantage.
Results
The final model included prior contacts, jurisdiction issuing the index caution, Indigenous status and gender as predictors. The c-statistic (area under the ROC curve) when comparing predicted with observed values was 0.767, 95% confidence interval (0.757, 0.777). The model fit was confirmed by comparing predicted values from half the dataset with observed values from the other half.
Conclusion
It is possible to screen juveniles for future risk of reoffending from data readily available at the time they are cautioned.