Validation of the Custody Rating Scale for Black Men Offenders

What we did

The predictive validity of the CRS for Black men offenders was assessed using a dataset of admissions to federal custody between 2013/2014 and 2018/2019. The dataset included Black men offenders admitted on a Warrant of Committal (WOC, i.e., a new sentence; N = 1,813), or following a revocation of conditional release (N = 428).

What we found

A higher CRS designation was associated with a greater likelihood of involvement in a negative institutional event (see Table 1). This relationship was stronger for new admissions relative to revocation admissions. OSL was similar to the CRS with respect to associations with outcomes. Area Under the Curve (AUC) values reached the threshold of ‘good’ predictive accuracy for the CRS in relation to the outcomes of receipt of a serious institutional charge and receipt of any institutional charge, and ‘acceptable’ predictive accuracy in the case receipt of a minor institutional charge and involvement in an institutional incident. Predictive accuracy was similar for OSL relative to the CRS. Predictive accuracy was lower for the revocation group in the case of both the CRS and OSL (see Table 2).

What it means

Results indicate that the CRS holds predictive validity for Black men offenders at intake, and suggest efficacy in the use of professional judgement as evidenced by the similar predictive ability of the CRS and OSL. However, predictive accuracy was found to be weaker in the case of revocation admissions. 

^a Refers to involvement in at least one security or behavioural incident during the sentence for which the CRS was applied and following CRS administration.

^b Refers to an offender’s receipt of any serious or minor disciplinary charge during the sentence for which the CRS was applied and following CRS administration. Only charges resulting in an outcome of ‘guilty’ were included.

* The Area Under the Curve (AUC) is used to examine the predictive accuracy of a measure in ROC analyses; a value of 0.5 represents chance prediction, while a value of 1.0 represents perfect prediction.

Note. CI = Confidence Interval