This paper proposes a method for more efficient identification of those at higher risk of re-offending.
Aims: The objectives of this study were (1) to explore the effects of applying a screening tool to determine who is administered the Level of Service Inventory – Revised (LSI-R); and (2) to examine the predictive utility of including LSI-R subscale scores along with standard risk factors in a model of recidivism.
Method: Aim (1) was addressed by developing a screening tool using routinely collected data. Predicted probabilities of re-offending were obtained from this tool. Alternative thresholds of predicted probabilities required for an LSI-R assessment were then applied. The effect of screening was examined in terms of whether those who went on to re-offend were predicted to do so, having met the applied screening tool and LSI-R risk category criteria. Aim (2) was addressed by constructing and comparing logistic regression models with and without LSI-R subscale scores to assess whether models which included LSI-R subscale scores in addition to routinely collected data were better at discriminating those who re-offended within 12 months from those who did not. Analyses were conducted separately for males and females.
Results: Aim (1): By administering the LSI-R to those with a predicted probability of re-offending of at least .15, 80 per cent of male and 71 per cent of female recidivists would have been identified as being likely to re-offend, using LSI-R risk level criteria of at least low-medium. Aim (2): For males and females, after controlling for standard risk factors, the LSI-R subscales education/employment and attitudes/orientation were associated with re-offending. Further, criminal history, alcohol/drugs and accommodation subscales were associated with re-offending in males, and the companions subscale was associated with re-offending in females.
Conclusion: More efficient identification of those at higher risk of re-offending could be achieved by using a screening tool based on routinely collected data to determine who the LSI-R is administered to. Further, the inclusion of LSI-R subscale scores in models of recidivism could improve the predictive accuracy of models developed for evaluation purposes.