Association for Public Policy Analysis & Management

Concern about the “revolving door” of prison has led to increased interest among researchers and policymakers to understand the phenomenon of prison recidivism. Existing studies of recidivism often note that a large proportion of people released from prison return. For example, Durose et al (2014) find that 50 percent of those released from prison in 2005 returned within three years. However, existing studies often do not discuss the risk of recurring prison recidivism, nor do studies adequately explore the heterogeneity around the circumstances of release from prison. Yet, evidence suggests that the number of prior prison terms is related to recidivism (Kernan et al 2016). Thus, this current project is an effort to address this gap in research by using a large national longitudinal administrative data set to examine the patterns of individuals’ repeated prison returns over time.

Specifically, this paper will use restricted-use data from the National Corrections Reporting Program (NCRP) at the Bureau of Justice Statistics (BJS), which contains individual-level information on all state prison admissions and releases since 1983. I received restricted data very recently (March 2017); thus, initial analyses will use a subset of the data. This subsample includes records linking release and admission data across 3.1 million unique individuals in 17 states from 2000 to 2014. During this time period, over 550,000 of these individuals had three or more prison stays. Using these data, I conduct a recurrent event history analysis to answer the questions: what factors increase the risk of multiple returns to prison, and what factors shorten the time to prison return for repeat recidivists?

I utilize several recurrent event history approaches to construct models of prison recidivism: multi-state models; frailty models; and Prentice, Williams, and Peterson models. The key explanatory variables I examine are state of jurisdiction, offense type, number of counts for conviction offense, whether the return is due to a new conviction or technical violation of parole, as well as demographic characteristics. These unique data permit analyses to identify multiple instances of recidivism for up to a 15-year period, which will provide more accurate estimates and standard errors than has been the case in studies treating each return as committed by a unique individual, or studies that only focus on an individual’s first prison return (Amorim and Cai 2015, Yang 2017). Furthermore, these approaches allow me to analyze the time that an individual spends in prison after each prison return, not just the time during which an individual is at risk of recidivating or the number of total repeated prison returns.

This project responds to calls for evidence-based policymaking in reducing recidivism and the prison population as a whole. Specifically, it seeks to push both researchers and policymakers to consider the long-term effects of the ever-revolving door of prison and the factors that may be contributing to this cycle. Furthermore, this project is a call for social scientists to apply recurrent event history methods to more accurately estimate phenomena such as repeated spells of imprisonment.