Recidivism: Prediction with Bayesian Models

• Activity Number: 88
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Gail Blattenberger*+ and Richard Fowles
• Companies: University of Utah and Department of Economics
• Keywords: BART ; Bayesian Model Averaging ; Survival Analysis ; Recidivism,

Abstract

The United States has the highest prison incarceration rate in the developed countries in the world. The shares of state budgets devoted to departments of corrections have grown substantially. About half of all persons in the United States who are released from prison return within three years. This high recidivism rate contributes the budgetary difficulties faced by many states in recent years. Because of high incarceration costs, lower cost alternatives to prison are becoming increasing important. This study looks at a large set of probationers in Utah to predict a failure in probation sentencing (such as incarceration). The new data complemented by a unique survey of released prisoners allows the exploration of many factors relating to recidivism not examined in earlier studies and results are relevant to policy decisions. The study employs Bayesian Additive Regression Trees (BART) and Bayesian Model Averaging to forecast recidivism and compares the results to Survival Analysis.