Across methodologies, the first step in this process is the development of a statistical model that assesses offenders’ likelihood of recidivism using key risk factors (e.g., logistic regression model). Here, the variables entered in these models are driven by the extensive body of literature on recidivism risk factors, as well as data completeness and accessibility. The final model is subsequently used to either adjust observed recidivism rates or compare observed recidivism rates against predicted/expected recidivism rates. There are multiple methods to approach such adjustments or comparisons (e.g., Smith & Jones, 2008a; 2008b; Ministry of Justice, 2011; Mulvey et al., 2015; Wartna et al., 2011). We provide a several examples below. However, regardless of the methodology adopted, all risk-adjusted recidivism rates provide fairer comparisons across groups, regions/prisons and across time than can be achieved with raw recidivism rates (Bonta et al., 2003; King & Elderboom, 2013; Office of the Inspector of Custodial Services, 2014).
One approach is to calculate the difference between observed and expected recidivism rates for a cohort (e.g., see Figure 2; also see Smith & Jones, 2008a; 2008b; …show more content…
The RQ compares rates of reoffending amongst offenders who receive a particular rehabilitative program or intervention, with the rates recorded amongst offenders who have an equivalent risk of reoffending, but who have no exposure to this intervention (i.e., a matched comparison group; Ax & Fagan, 2007; Department of Corrections, 2015; Lukkien & Johnston, 2013). Here, offenders are matched on a variety of salient risk characteristics such as “age, gender, ethnicity, sentence length, sentence type, ROC*ROI [risk assessment] scores” (Johnston, 2014, p. 8). New Zealand reports the RQ annually across corrections programs (Department of Corrections, 2015). While this is an impressive measure of correctional performance (particularly since RQs are reported annually as core performance measurements), the calculation of the RQ relies on complex statistical procedures, including propensity score matching, and therefore may not be feasible for many jurisdictions. This broad approach is often used to control for risk of reoffending in research that evaluates the effectiveness of programs (e.g., Kim & Clark, 2013), as well as research that compares private and public prisons (e.g., Duwe & Clarke, 2013) or varying release or sentence types (e.g., Bucklen et al.,