A Person-centered Approach to Risk and Need Classification in Drug Court

Abstract

A sample of 265 New York City drug court participants completed the Level of Service Inventory—Revised (LSI-R) and Texas Christian University Drug Screen II (TCUDS). Three participant clusters were identified through a person-centered analysis of their LSI-R and TCUDS responses: low risk (LR), criminogenic risk (CR), and complex behavioral health needs (CBHN). Although CBHN scored higher than CR and LR on the LSI-R and TCUDS, they were no more likely to be re-arrested at 24 months and no higher in their rate of positive drug tests. The CR cluster predicted re-arrest beyond the LSI-R and rate of positive drug tests beyond the LSI-R and TCUDS. CBHN participants placed in a residential (vs. non-residential) setting were disproportionately likely to be re-arrested. Results point to a sub-population of drug court participants not captured in variable-centered summary risk scores, who might require intensive case management or referral to suitable treatment.

Keywords:

• drug courts
• treatment
• risk
• clustering

Notes

¹ The HICLAS algorithm and software are available from the lead author.

² A four-cluster HICLAS solution was not as easily interpretable as the three-cluster solution. We therefore adopted the three-cluster solution shown in the Appendix.

³ Testing the predictive power of clusters beyond scale scores constructed from many of the same items raises the possibility of multicollinearity. It is true that cluster membership was correlated with the LSI-R and TCUDS summary scores (See Table 2). We do not believe multicollinearity is a serious concern, however. First, the clusters were not linear composites of LSI-R and TCUDS scores; rather, they were binary indicators that were derived from different combinations of LSI-R and TCUDS items. Moreover, Tables 4 and 5 show that the coefficients and odds ratios for criminogenic risk and complex behavioral health needs did not substantially change from Model 1 (clusters only) to Model 2 (controlling for LSI-R and TCUDS), although complex behavioral health needs became a non-significant Model 2 predictor of re-arrest. Variance inflation factors and tolerance were also well within acceptable range: maximum VIF = 2.82 minimum tolerance = .36 (both for complex behavioral health needs).

Additional information

Notes on contributors

Warren A. Reich

Warren A. Reich, PhD, is a Principal Research Associate at the Center for Court Innovation. He received a BS in Psychology from the Pennsylvania State University and a PhD in Social-Personality Psychology from Rutgers University in New Brunswick, NJ.

Sarah Picard-Fritsche

Sarah Picard-Fritsche, MA, is an Associate Director of Research with the Center for Court Innovation. She is the project director of an NIJ-funded randomized trial of an evidence-based assessment tool for drug-involved offenders and is currently a doctoral student in criminal justice at CUNY’s graduate center.

Michael Rempel

Michael Rempel, MA, is Research Director at the Center for Court Innovation, overseeing all research conducted at the agency since 2002. He is currently principal investigator of a randomized controlled trial of evidence-based assessment tools. Recent publications include research on domestic violence courts, community courts, and procedural justice.