Association for Public Policy Analysis & Management

To effectively direct criminal justice services and treatment, it is necessary to understand the factors that have demonstrated a relationship with offending behavior as well as identify where a particular offender falls along this continuum.  This is commonly done by assessing the risk an individual poses and identifying the individual’s treatment needs.  In juvenile offenders, research shows they engage in delinquency as a combination of psychological pathology and/or unfavorable environmental factors (Cullen & Agnew, 2006).  Identifying juvenile recidivism risk involves understanding the impact of these factors, and is made easier using diagnostic tools such as the Youth Level of Service/Case Management Inventory (YLS/CMI).  According to Onifade, Davidson, Livsey, Turke, Horton, Malinowski, Atkinson, and Wimberly (2008), the YLS/CMI is one of the most popular juvenile assessment tools. Although previous researchers have established the reliability and construct validity of the YLS/CMI (Rowe, 2002; Schmidt, Hoge, & Robertson, 2002) and have examined the criterion and predictive validity of the instrument (Jung, 1996; Jung & Rawana, 1999), but few have assessed the instrument at the individual item level. 

Using Rasch analysis, this study examines the YLS/CMI at the item level among a sample of high risk juveniles receiving Intensive Aftercare (IA) services from the Kentucky Department of Juvenile Justice. The analysis was conducted using the 42 standard YLS/CMI items and conducted in Winsteps. Data from 52 IA program participants were used. Juveniles had to receive a score of 23 or above to be initially included in IA. The Rasch results suggest the YLS/CMI generally functioned well as a diagnostic instrument, but had a reliability issue. The item-fit scores were appropriate, with item-infit statistics ranging from -1.3 to 1.0 and item-outfit statistics ranging from -1.2 to 1.7. Item difficulty levels fell between -3.30 and 2.51. These item difficulties suggest there was a range of easy-to-difficult items. There was some repetitive clustering of items around the 0.0 logit point, suggesting the model items could be reduced. The reliability results were mixed. The item-reliability was at .91, suggesting the difficulty order would not alter greatly with another sample (Bond & Fox, 2007). However, the person-reliability was at .37, suggesting that other sample subjects with similar risk levels would receive varying scores and that responses of the subjects may not remain stable longitudinally. This could indicate a problem with the YLS/CMI instrument or may be linked to rater scoring practices.

Understanding the performance of the YLS/CMI as a diagnostic instrument is vital, given its usages as a service screening tool and risk of recidivism identifier. Basing policy decisions or enacting policies using a diagnostic tool with validity issues is problematic. First, it can lead to policy decisions which misuse resources and place individuals in the broader community at increased risk.  Secondly, it can lead to inadequately meeting the needs of the participants being screened by the instrument.  For these reasons, understanding the item level performance of the YLS/CMI and similar diagnostic instruments is vital for effective policy decisions and resource allocation as well as increased public safety.