Abstract
As efforts to reverse mass incarceration increase, so does the need to supervise more individuals in the community. Faced with heightened demand, community corrections agencies increasingly use risk assessment to allocate resources efficiently and improve public safety. While both static, historical factors as well as dynamic, changeable factors have been incorporated into risk assessment instruments, one factor notably absent is the amount of time an individual remains in the community recidivism-free. Using parametric and discrete hazard models, we examine the relationship between recidivism-free time and observed recidivism among individuals on parole supervision in Pennsylvania where dynamic risk assessment is used. Specifically, we assess whether recidivism-free time predicts recidivism independent of these risk scores and the extent to which single and repeated risk scores accurately predict recidivism. Findings support the use of dynamic risk instruments but suggest that recidivism prediction may benefit from considering recidivism-free time. Implications for community corrections policy are discussed.
Disclosure Statement
No potential conflict of interest was reported by the authors.
Notes
1 Corrections and paroling authorities balance a multitude of factors when making release decisions, including harm of prolonged imprisonment as well as individual risk and potential threats to public safety. While there is evidence that changes in the number of people returning from prison can affect neighborhood crime rates (Hipp & Yates, Citation2009), recent evidence from California on the impact of downsizing of prison system suggests that large reductions in formerly prison-bound individuals did not increase violent crime (Sundt et al., Citation2016).
2 While the decision making aided with risk assessment can lead to desirable outcomes (Kleinberg et al., Citation2018), persistent criticisms, especially regarding fairness and potential bias, exist (Eckhouse et al., Citation2019; Hannah-Moffat, Citation2013).
3 In addition to literature suggesting two paths by which individuals may desist, some studies suggest that there is more heterogeneity in the way risk changes over time. For example, Baglivio and colleagues (Citation2017) find that among risk declines over time for some juvenile offenders but increases over time for others. If this is the case, recidivism-free time would still indicate changes in risk and would be valuable in predicting recidivism.
4 The sample does not include individuals who complete their sentence in prison (“max-outs”) and are thus released without parole supervision. This excluded group of people likely poses higher risk of recidivism and are distinct from parolees as they are not at risk of reincarceration due to technical violations (Ostermann, Citation2009).
5 As is common in recidivism research, our dataset does not include the offense date for recidivism events. Only the arrest date was available for analysis. Though it is possible that the time between an offense and the subsequent arrest obfuscates our recidivism-free time measure, data from the state’s administrative office of the courts suggest that the typical time between arrest and offense is rather short, with a median value of 6 days. There is also no evidence of systematic differences in this length of time for different crime types.
6 Recidivism-free time is defined by re-incarceration (rather than re-arrest) in 24% of the sample. The Pennsylvania Parole Board utilizes a violations and sanctions matrix to standardize the conditions under which individuals on parole are sanctioned for violations of parole conditions. Decisions to re-incarcerate a person under parole supervision by revoking his or her parole are made by parole board members collectively, rather than individual parole officers or supervisors. This process minimizes the amount of the influence of parole officer discretion and unwarranted disparity in the re-incarceration outcome in this sample.
7 It is possible for parole officers to request more frequent assessment of parolee risk. In these data the average time between risk assessments is 12 months, suggesting that parole officers are largely compliant with agency policy.
8 The LSI-R measures criminal history information including prior convictions and incarcerations (among other things). In the present sample there was only a modest correlation between Prior Arrests and LSI-R scores (r = 0.217 for the initial LSI-R score, and r = 0.224 for the annually repeated LSI-R score). There was no evidence that the presence of the prior arrests variable induced multicollinearity in our analyses.
9 Due to multicollinearity, we do not include measures of parolee supervision level. Initial and subsequently changing supervision levels are determined in part by the LSI-R score (r = 0.4899). Inclusion of this measure in addition to the LSI-R scores results in inflated standard errors, so we omit this measure from our analyses.
10 We do not estimate a Cox proportional hazard model to assess the relationship between LSI-R scores, recidivism, and risk. Because our research focuses on changes in recidivism over time and specifically the declining nature of risk, a Cox model was deemed unsuitable because it cannot explicitly model duration dependence ( Bennett, Citation1999; Box-Steffensmeier & Jones, Box-Steffensmeier & Jones, Citation1997 ). The assumptions of the Cox model are questioned in these data and the addition of a Cox model does not add to the conclusions drawn from the parametric and discrete hazard models.
11 Studies of halfway houses and recidivism provide mixed evidence of their effectiveness. Lowenkamp and Latessa (Citation2002) suggest that individuals released to halfway houses recidivate more than those released directly to the community, but the work of Wong et al. (Citation2019) suggest the opposite is true. Costanza et al. (Citation2015) finds no relationship. In Table 2 halfway houses are positively but not significantly related to recidivism; however in Table 3, the halfway house coefficient is positive and significantly related to recidivism in models 3-5.
12 Due to data constraints, this research was unable to address whether changes in specific LSI-R domains (e.g., static domains such as criminal history or dynamic domains such as education and employment) relate to recidivism and recidivism-free time. Future research should explore the extent to which changes in individuals LSI-R domains are related to recidivism and/or recidivism-free time.
13 Parole length was top-coded at the 95th percentile to account for skewness.
Additional information
Notes on contributors
Nicole E. Frisch-Scott
Nicole Frisch-Scott is an Assistant Professor in the Department of Criminology and Criminal Justice at Merrimack College. She completed her PhD. in Criminology and Criminal Justice at University of Maryland, College Park. Her research interests broadly include corrections policy, life course criminology, and quantitative methodology.
Kiminori Nakamura
Kiminori Nakamura is an Assistant Research Professor and the Research Director of the Maryland Data Analysis Center in the Department of Criminology and Criminal Justice at the University of Maryland. His research focus on corrections and reentry, life-course and criminal-career issues, including redemption and desistance.