Structured Decisionmaking

One of the most powerful system reforms for reducing and preventing DMC is the implementation of structured decisionmaking using statistical risk classification (i.e., a risk assessment instrument). A risk assessment instrument is an empirically based, standardized, objective instrument for use in evaluating a youth’s background and current situation and estimating the likelihood that the youth will continue to be involved in delinquent behavior. Community corrections may use the results of risk assessment to specify the level or intensity of supervision needed; in residential settings, risk assessment results may determine the security level and living unit (Clear and Gallagher, 1983; Wiebush et al., 1995).

In many juvenile justice systems, practitioners make decisions based on their experience and knowledge of a youth’s background, without using research-based tools. However well intentioned, such clinical predictions can be rife with unintentional racial bias that results in DMC. The absence of structured decisionmaking at any point in the juvenile justice process allows practitioners to base decisions on subjective criteria that may be related to race. For example, Iowa’s assessment research indicated that some officers equated the wearing of certain sports apparel with gang membership, so youth wearing such apparel were more likely to be referred to juvenile court instead of diverted ( Leiber, 1994).

Structured decisionmaking holds the promise of enabling practitioners to objectively classify delinquent youth according to level of risk and to reassess level of risk at different stages in the juvenile justice process. Accurate information about level of risk, in turn, can improve decisionmaking regarding treatment, placement, and court disposition. Statistical risk assessment entails having youth complete a standardized risk assessment instrument, which usually consists of questions related to a small number of factors that research has shown can predict future offending. The screener determines the risk level from the numerical scores assigned to the responses and often divides the summated risk score into categories of low, medium, and high risk. (Some instruments allow the screener to override the instrument’s determination or offer opportunities to mitigate or aggravate the score based on favorable or unfavorable characteristics. Screeners must apply overrides and mitigating and aggravating factors with caution, however, because the value of the screening instrument lies in its objectivity.) Statistical risk assessments are a valuable tool for reducing recidivism because they allow practitioners to accurately reassess the level of risk at various decision points and thus respond more efficiently and effectively to youth in the system.

However, even structured decisionmaking instruments can contribute to minority over-representation unless practitioners take proper care at each step of the development process. An obvious example of pitfalls to avoid is the use of race (and/or ethnicity) as a predictor of recidivism. While recidivism differences may be correlated with race, they are not caused by race. Attributes that are themselves correlated with race (e.g., poverty, school failure, a high proportion of unsupervised time in a day, levels of community disruption, amount of police surveillance in the community) cause the differences in recidivism (Gottfredson and Snyder, 2005). When information on these attributes is unavailable (as is often the case), their predictive power is partially captured by race and invites the erroneous interpretation that race is causally related to recidivism.

To address this problem, Gottfredson and Snyder (2005) suggest that risk scale developers add a race variable in the early stages of risk scale development but omit this variable from the published instrument. The authors argue that some statistical methods used in the development of risk scales remove the unique (i.e., independent) effect of race from the prediction process. They maintain that unless race is mathematically included in the initial steps of risk scale development, when race correlates with the criterion measure, one cannot remove racial bias from the resulting risk scale; it remains unobtrusively present beneath the surface, influencing each risk scale score.