Using Cluster Analysis to Predict Recidivism among a Population of Chronically Homeless ‘Frequent Users’ of Jail and Public Mental Health Systems

• In Event: Health Matters and Prisoner Reentry

Abstract

Chicago’s Frequent Users of Jail and Mental Health Services Initiative (“Chicago FUSE”), implemented by the Corporation for Supportive Housing, targeted individuals with histories of chronic homelessness and mental illness who were ‘frequent users’ of jail. Data were collected from 161 eligible individuals who would be leaving the Cook County Department of Corrections (jail) between 2009 and 2010, some of whom were then randomly selected to obtain permanent supportive housing (PSH) with enhanced services upon release. This paper uses K-means cluster analysis on the full sample (n=161) to identify discrete sub-groups of ‘frequent users’ based upon constructs in the domains of mental health, physical health, homelessness and criminal justice involvement. Cluster membership is then used to predict re-arrest and re-incarceration. Findings will attempt to uncover whether particular characteristics can be used to reliably differentiate members of this small, but costly, high-risk population; and if clusters can be used to predict recidivism. The clustering methodology will be discussed at length since a number of novel, statistical precautions were taken to avoid the common criticisms of cluster analysis. Implications for service provision, policy and future research will also be considered.

Authors

• Courtney S. Harding, Temple University

• Caterina Roman, Temple University