Measuring the Effect of the Experience of Incarceration on Reoffending

• Activity Number: 214
• Type: Topic Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Daniel Nagin*+ and Jose Zubizarreta
• Companies: Carnegie Mellon and Columbia University
• Keywords: causal inference ; matching ; instrumental variables

Abstract

Every year more than 1 million individuals are released from US prisons and jails yet good evidence on the effect of the experience of incarceration on reoffending is scarce. The best available evidence exploits the random assignment of cases to judges and differences in punitiveness across judges to identify this effect. We improve upon existing evidence based on the judge-based instrumental variable (IV) by using methods designed to increase the power of the IV to detect an effect. Specifically, we use recent matching methods based on integer programming for precisely adjusting for observed covariates and augmenting the strength of the instrument. These improved methods are applied to a large data set of over 100,000 individuals sentenced in Pennsylvania from 1998 to 2000.