Exploring Back-end Sentencing: A Study of Predictors of Parole Revocation through a Focal Concerns Theoretical Framework

Abstract

Parole revocation, the process of returning individuals to incarceration due to non-compliance with conditions of community supervision, contributes to mass incarceration and is tied to the complex process of back-end sentencing. This discretionary decision is typically opaque, and theoretical understanding is limited. Applying the focal concerns framework, this study employs population-level data from an American jurisdiction on individuals who are on parole and have violated their supervision (n = 13,121). We found that practical, extralegal considerations such as the amount of time left on parole, the number of programs geared toward people on parole, and whether they participated in community programs after release from prison significantly influenced back-end sentencing. These results question the assumed relationship between individual-level and systemic factors during revocations while also providing empirical support for broadening the scope of theoretically relevant domains.

INTRODUCTION

In the United States, many states prison systems, as well as the federal system, have experienced sustained population declines over the past several years, with 2014 marking the first year that both systems experienced decreases in their populations since the Bureau of Justice Statistics started tracking corrections population statistics (Kluckow and Zeng 2022). Several factors have likely contributed to the stark reversal of decades-long consistency in the growth of incarceration rates. Some of the differences may be attributed to reforms on the front end of sentencing and correctional policy, driven by both new empirical research and shifts in political sentiment. In many cases, a systematic reliance on “tough on crime” responses to offending has been scaled back. The package of policies that have been recently reformed, many of which were implemented in the late 1970s, includes lengthy sentences for non-violent crimes, most notably for drug offenses (Western 2006), three-strikes laws, and reliance on mandatory minimum sentences (Travis 2005). Other more recent efforts have focused on the diversion of mentally ill and other vulnerable individuals away from prison (Broner et al. 2004; Steadman and Naples 2005). At the same time, attention to reforming the parole process has been more limited, perhaps due to the low levels of visibility of that decision-making process and the embedded nature of non-judicial hearings. While some structural reforms forms have addressed discretionary parole, there has been little movement since the 1990s (Rhine, Mitchell, and Reitz 2019). Many policies in that area that often contribute to prison population growth, such as truth-in-sentencing laws, remain active in many jurisdictions (Petersilia 2003; Stemen, Rengifo, and Wilson 2005; Spohn 2014).

The process of returning people on parole back to prison as the result of supervision revocation is a practice that is commonly known as “back-end” sentencing. Many of the social and political pressures that have been increasingly recognized as important contributors to mass incarceration (Frost and Clear 2009) are also prevalent when people on parole are sanctioned (Travis and Christiansen 2006). Reitz and Rhine (2020) note that these back-end decisions are a significant, and too often overlooked, contributor to mass incarceration. Jeremy Travis (2007) characterizes the rise of the problem of back-end sentencing by highlighting that in 2000 the United States sent over two hundred thousand individuals back to prison for parole violations, which represented a higher number than the total number of people sent to prison for any reason just twenty years prior. Additionally, while the per capita growth in incarceration increased approximately fourfold from 1973 to 2000, imprisonment due to parole revocations grew by about sevenfold. Jeffrey Lin (2010) has demonstrated that parole revocations comprise approximately 22.2 percent of all prison admissions in the United States, a figure that was approximately three times the rate of 7.1 percent recorded in 1980.

Recently, a limited number of scholars have sought to empirically examine how and why back-end sentencers exercise their discretion. For example, Lin, Ryken Grattet, and Joan Petersilia (2010) explored the potential predictors of back-end sentencing. In addition to evaluating the influence of individual-level and demographic variables, their analysis was informed by the broader institutional context, as conceptualized within the focal concerns theoretical framework. By focusing on unobserved factors, such as practical and systematic constraints directly or indirectly placed on decision makers as well as critical dimensions such as a person’s perceived threat to public safety and their blameworthiness, a holistic picture of the decision-making process can emerge (Eisenstein and Jacob 1977; Lipsky 1983; Nardulli, Flemming,and Eisenstein 1985; Albonetti 1991; Steffensmeier, Ulmer, and Kramer 1998).

Revocation decisions made by parole boards are shaped by the complicated interaction of an amalgam of individual-, organizational-, and community-level variables. This suggests that individualized decisions are shaped by institutional and structural macro-sociological influences whose impact is frequently unobserved and unrecognized within the practical and academic literatures and represents a complicated interaction of factors. To that end, the current study contributes to the back-end sentencing literature in two primary ways. First, by distinguishing the back-end sentencing decision process from contemporaneous processes, including sentencing for new criminal conduct, we isolate the rate of parole revocations from other instances resulting in incarceration. To accomplish our analyses, we employed data from a highly populated state on the east coast of the United States. These data cover a population of formerly incarcerated people who were released to parole supervision over a 129-month period beginning in 2005 and subsequently were considered for parole revocation by the state’s parole board (n = 13,121). Of this group, approximately 81 percent (n = 10,683) were revoked and sent back to prison.

Second, we built upon and expanded the foundational literature in this area by increasing the number and range of predictor variables previously examined within the focal concerns framework. Our analytic strategy culminated in the construction of four nested logistic regression models to explore theoretically and practically relevant predictors of affirmative parole revocation decisions. We assessed these factors in light of the focal concerns theoretical approach to judicial decision making (Albonetti 1991), with an emphasis on offender blameworthiness, community protection, and dangerousness (Ulmer and Kramer 1996; Steffensmeier, Ulmer, and Kramer 1998; Kramer and Ulmer 2002; Huebner and Bynum 2006; Lin, Grattet, and Petersilia 2010). We expand the scope of factors previously integrated into the focal concerns framework by including elements relating to a person on parole’s participation in community programming as well as pragmatic factors capturing the nature of the violation event and what we classify as parole investment (for example, time on the street, nature of reentry transition, duration of the remaining sentence). By contextualizing the factors predictive of parole revocation within focal concerns, we seek to lay some additional groundwork for both sentencing policy reform and the advancement of discretionary decision-making theory within a post-adjudication, quasi-judicial context.

The following sections describe the operation of the discretionary parole release and revocation processes, the focal concerns theoretical framework that guides this inquiry, and the current, albeit limited, existing empirical knowledge base linking theory to practice. Taken together, this serves as a detailed contextualization for the back-end sentencing processes examined in this study. We then describe the data employed for the study, detail the variables that comprise our conceptual clusters, and present our results. We conclude by contextualizing the study’s findings and discussing its implications for theory, policy, and future research.

PAROLE AND THE REVOCATION PROCESS

Parole allows for correctional authorities to release incarcerated people from prison prior to the expiration of their sentence (Petersillia 2003; Tonry 2007; Ostermann 2015). Shifting these people into the community can facilitate treatment programming, encourage community reintegration, and, at the same time, decrease the monetary costs associated with incarceration (Petersillia 2003; Tonry 2007). These tangible benefits have led in recent years to an increase in the number of people on parole, especially relative to the size of the incarcerated population (Kaeble, Maruschak, and Bonsczar 2015). The process of determining whether parole supervision should be revoked and an individual returned to full incarceration, often overseen by non-judicial parole board officials, is referred to as “back-end sentencing” (Tonry 2007). Though largely overlooked as a major contributor to prison populations for many years, scholars and policy makers have become increasingly aware of the extent to which these back-end decisions contribute to mass incarceration and hamper the reintegration process (Travis and Lawrence 2002; Solomon, Kachnowski, and Bhati 2005; Tonry 2007; Lin 2010; Rhine, Petersilia, and Reitz 2016; Rhine, Mitchell, and Reitz 2019; Reitz and Rhine 2020).

Parole remains a major component of corrections in the majority of jurisdictions. All states, except for Maine, have people on parole living in the community, though the nature of supervision may vary significantly (Kaeble, Maruschak, and Bonsczar 2015). As many of these individuals remain criminally active or are non-compliant with the administrative conditions of their supervision, returns to prison are not uncommon (Solomon, Kachnowski, and Bhati 2005). Lin (2010) found that 22 percent of new admissions to state corrections systems in 2008 were due only to the back-end sentencing process following a parole revocation, a threefold increase over the 7.1 percent observed in 1980. Revocation rates have fallen slightly since then, driven by disproportionally large decreases within the California state and federal systems, as other estimates for the rates in 2014 suggest that 14 percent of people on parole were reincarcerated that year—8 percent for a technical violation, mirroring Lin’s (2010) findings, and an additional 4 percent for a new criminal sentence (Kaeble, Maruschak, and Bonsczar 2015). Irrespective of the exact rate, the fact that back-end sentencing is a direct, significant, and often overlooked contributor to prison populations remains clear.

Parole revocation is a distinct process from sentencing, despite sharing several common protections and procedures. Parole revocation proceedings are typically administrative, rather than judicial, processes that are often initiated by a person on parole’s supervising officer and are overseen by a parole board hearing member or another independent official. ¹ Therefore, although parole revocation decisions are not the result of an adversarial trial, they still may directly result in incarceration; there are costs, pressures, and philosophical goals that closely mirror the judicial sentencing process. Parole officers, officials, and boards have a significant amount of discretion (Steen et al. 2013). In exercising this discretion, and unlike sentencing judges, they may consider institutional behavior and positive and negative post-release conduct within the community, in addition to the data initially considered at sentencing, to help shape their decisions throughout the revocation process (Gottfredson 1979; Petersilia 2003, 2004). In general, the day-to-day practices of parole boards making back-end sentencing decisions are typically shrouded from the public’s view (Steen and Opsal 2007; Tonry 2007; Lin 2010). This is likely because parole revocations are administrative decisions made in correctional settings rather than legal decisions made in open court (Steen and Opsal 2007; Lin 2010).

Extant research on the predictors of parole revocation remains limited and focused on individual-level characteristics associated with failure (Steen and Opsal 2007; Ostermann 2015). This classification of studies includes nationally representative as well as state-specific explorations into the failure patterns of conditionally versus unconditionally released individuals upon their reintegration back into their communities. For example, in a study of 272,111 individuals released across twelve states in 1994 (Langan and Levin 2002), people on parole with twelve or more prior arrests and those serving sentences for drug, property, or violent crimes were at heightened risks of rearrest within two years of release when compared to those that were unconditionally released from prison (Solomon, Kachnowski, and Bhati 2005).

Less scholarly attention has been directed at the ideological context in which back-end sentencers make these complex, high-stakes decisions (Travis and Christiansen 2006). Michael Gottfredson and Don Gottfredson (1988) draw on prior field work by Victor O’Leary to suggest that parole board members are forced to take on six contemporaneous values-driven frames of reference when considering a release or revocation parole decision: jurist (due process and fairness); sanctioner (equitable and punitive results); treater (rehabilitative); controller (amelioration of risk); citizen (preservation of social order); and regulator (balancing of correctional system needs). As is the case with the parallel purposes of front-end sentencing, these broad frames of reference can be inconsistent, vary by case, and resist the classification of decisions (or deciders) along these parameters. Though not empirically validated at the time, this conceptualization of how discretionary release authorities view their role supports the application of a theoretical framework for the exercise of discretion, which now enjoys some support within the correctional literature.

THE FOCAL CONCERNS THEORETICAL PERSPECTIVE

Traditional judicial sentencing is insulated from many of the pressures that inform other decisions made during the adjudication and punishment processes. Though this discretion is by constitutional design, it renders the landscape of influences on criminal verdicts largely opaque. Sentencing, however, does not take place within a vacuum (Kozinski 1992; Baum 2009; Epstein, Landes, and Posner 2013). Several approaches have been developed over recent decades to understand how, and why, certain people on parole are remanded to prison while other, largely similar, individuals are not (see, for example, Johnson 2003). Of these, the focal concerns approach has emerged as a viable conceptualization of how discretion influences the incarceration decision-making process. Strongly influenced by Walter Miller’s (1958) work with juveniles, there are practical and philosophical elements that are, either directly or by proxy, likely to be considered contextually relevant by a decision maker.

Sentencing in any context is a complex decision, with the decision maker responsible for weighing the defendant’s liberty interest against philosophical goals of punishment, including retribution and incapacitation. Noting that judges make these decisions with incomplete information and limited time, Celesta Albonetti (1991) argues that factors such as race, age, and gender, though not explicitly part of the sentencing process, are relied upon as extralegal shorthand for likely future conduct. They note that this effort to create a “bounded rationality” can encourage a reliance on stereotypes and other assumptions, some of which are then reflected in the perpetuation of disparities in punishment (250). This idea of a bounded rationality, in turn, directly informs the structural assumptions about the decision-making process that form the theoretical foundations of the focal concerns framework. Darrell Steffensmeier (1980), in considering gender disparities in sentencing, finds that judges can employ gender as a proxy for several traits thought to be pragmatically germane to sentencing, including family status and culpability. In light of a series of parallel observations, the range of factors considered in this manner has been expanded to include characteristics of the offense and the overall justice system. These groupings of salient characteristics have served as a foundation for a subsequent reformulation of the focal concerns into theory-driven, categorical factors believed to strongly influence the structure of judicial decision making (Steffensmeier, Kramer, and Streifel 1993; Steffensmeier, Ulmer, and Kramer 1998; Steffensmeier and Britt 2001; Steffensmeier and Demuth 2001). Focal concerns have become, in many ways, the dominant theoretical framework for the consideration of disparities at sentencing, most notably by race (Lynch 2019). ²

More recent research has continued to refine our understanding of the interactions between the dimensions of focal concerns during sentencing. In one of the earliest expansions of the theory, researchers considered the complex role of race, age, and gender within sentencing contexts (Steffensmeier, Ulmer, and Kramer 1998). Relying on statewide sentencing data from Pennsylvania, Steffensmeier, Jeffery Ulmer, and John Kramer (1996) identified significant influences on sentencing outcomes attributable to these factors, observable both independently of, and interacting with, case characteristics such as offense severity. These results were synthesized into the foundation of the modern theoretical model in which three focal concerns are employed to explain judicial discretion: blameworthiness, community protection, and practical constraints and consequences (766–68). Blameworthiness was conceptualized as a measure of how responsible the individual was for the crime for which they were being sentenced. These included factors derived from their criminal history as well as additional offender- and offense-specific traits. Community protection, a focal concern centered on the judicial need to predict how a person would act when released from prison, included sociodemographic variables, such as family status, believed to correlate with recidivism, as well as the (duplicative) application of criminal history. Lastly, practical constraints were included under a focal concern developed to capture organizational capacity, such as correctional resources and space availability as well as individual-level capacity, such as the defendant’s perceived “ability to do time” (see also Steffensmeier, Kramer, and Striefel 1993). For parole boards, these practical considerations could also include contextual factors that relate to the organizational context. For example, board members could consider risk of recidivism as relating to the potential political and institutional repercussions should a released individuals commit a crime so serious that the executive or legislative branch actors could seek to rein in parole generally (see Goldkamp et al. 2010) or limit (or abolish) the authority of the board, a rapid policy shift that, for various reasons, has happened in some jurisdictions (Marvell and Moody 1996; Tonry 2005; Rhine, Petersilia, and Reitz 2016; Rhine, Mitchell, and Reitz 2019). Though practical constraints are described in their conceptualization, Steffensmeier, Ulmer, and Kramer (1998), given the scope of their study, did not test any variables appropriate for inclusion therein, though this was addressed in subsequent studies.

John Kramer and Jeffery Ulmer (2002) find additional empirical support for the integration of criminal history and offense seriousness into the blameworthiness concern. In examining the characteristics of below-guidelines departures, they find that criminal history and crime characteristics interact to decrease the likelihood of a lower sentence; race, gender, and age lower this probability even further. They expand the scope of the practical constraints concerns by recognizing the influence of the “court community” environment and other contextual factors on sentencing (Ulmer and Kramer 1996). They note that a need for efficiency and systematic productivity, extralegal factors, including organizational culture, play a role in the exercise of discretion, especially in the face of large caseloads and limited resources. Parallel effects have been observed for plea deals reflecting a judicial preference for defendants who both accept responsibility and ensure rapid disposal of the matter (Ulmer and Bradley 2006). In sum, the focal concerns perspective provides a framework for the exercise of discretion, synthesizing the imputations about defendants’ characteristics and likely future conduct in light of practical, systematic pressures. Through a unique process, the same tensions and goals observed at the front end of the sentencing process may be mirrored in the parole revocation process.

FOCAL CONCERNS AND BACK-END SENTENCING

In many ways, front- and back-end sentencing are roughly parallel endeavors, making an extension of the focal concerns framework both logical and requiring minimal modification. This does not suggest that the decision or the deciders are identical. The addition of a wider range of individual-level data, including risk assessments and actuarial screenings conducted in prison or by parole authorities, as well as the individual’s post-adjudication behavior, can provide parole boards with more information. This directly expands the scope of factors that can be integrated into the blameworthiness and community protection concerns. Observations of the individual’s behavior, as well as the total amount of their remaining sentence, inform the practical considerations concern in this context. Additionally, as we propose here, the extent to which a person on parole has engaged in treatment and reentry programming—a potential indicator of his or her willingness to reform, which is similar to a plea at trial—may also inform the discretionary decision to revoke parole.

The back-end sentencing decisions made during the parole revocation process remain largely unexplored in this manner, with only a handful of studies examining any aspect of the discretionary parole decision-making process through this conceptual lens. Huebner and Bynum (2006) employ a focal concerns-driven analysis to examine the contribution of race and ethnicity to the decision to grant parole. While a distinct event from the back-end revocations that are the focus of the current analysis, this extends the applicability of the theoretical foundation beyond the traditional sentencing purposes for which it was developed. In this instance, they examined the sources of variation in time until parole among a sample of seventeen to twenty-four-year-old males incarcerated in three state prisons in a single jurisdiction who were eligible for parole (n = 423). Data on individual characteristics (for example, age, race, time served, and so on), legal characteristics (for example, criminal history, offense type, misconducts, and so on), and community contexts (for example, rates of disadvantage and crime in the release area) were employed in a series of multivariate proportional hazard models. Findings demonstrated that, from a focal concerns perspective, parole boards strongly considered factors relating to community protection (here, parole guideline recommendations, institutional misconduct, and crime type) as well as extralegal factors (in this case, race), resulting in significant increases in time to parole release by the board in the jurisdiction.

Lin, Grattet, and Petersilia (2010) have extended this theoretical and analytical framework to back-end sentencing in their analysis of the decision-making processes of parole board members at the California Department of Corrections and Rehabilitation. Using data covering all revocation decisions rendered in 2003 and 2004 (n = 114,820) and a focal concerns theoretical framework, the authors explore the impact of focal concerns on parole board decisions to revoke parole for alleged violations using relevant data at three distinct societal levels: individual, organizational, and community. Importantly, this analysis captures the full spectrum of the back-end sentencing process by including direct (new criminal cases), technical (violations of the rules of supervision), and absconding (failing to report or attend mandatory supervision or court meetings) violation cases. The results of their hierarchical logistic regression models indicated that individuals with certain pivotal classifications (for example, sex offender, history of violence) were at increased risk to have their parole revoked for both criminal and technical violations, and status characteristics (for example, rage, age, gender) were associated with increased odds of being returned to prison in direct violation cases for new offenses. Interestingly, in this context, many of the community-level theoretically informed variables (for example, county-level beliefs in punitiveness, community demographics, interactions between these measures and pivotal categories, race, and others) were, with few exceptions, not predictive of revocation across the criminal, technical, and absconding decisions.

More recently, Sara Steen and colleagues (2013) rely on a focal concerns perspective in their examination of a cohort of three hundred individuals released on parole in Colorado between 2005 and 2006. They conducted a series of independent bivariate regressions predicting decisions at multiple stages during the revocation and back-end sentencing decision-making processes with regard to only technical violations. They found that status characteristics (gender, age, and race) were significantly associated with parole officer decisions to revoke, as were pivotal classifications (for example, sex offender). Interestingly, and unlike Lin, Grattet, and Petersilia (2010), they do not identify any significant predictors of a revocation decision by the parole board among the variables in their study.

Taken together, the literature supporting the theoretical foundations for focal concerns in correctional decision making (Steffensmeier, Ulmer, and Kramer 1998), in conjunction with the findings from the extremely limited number of empirical applications to parole, suggest that this approach can provide important insight into the subtle and wide-ranging influences on the exercise of discretion in back-end sentencing. The three primary dimensions within the focal concern framework are independently supported: (1) blameworthiness is often captured by the disparate impact of both status characteristics and other classifications (for example, Morgan and Smith 2008; Mechoulan and Shauguet 2015); (2) community protection encompasses the role of case-based controls, many of which serve as a proxy for blameworthiness and for the avoidance of harms in the broader community (Caplan 2007); and (3) practical constraints includes diversionary and non-penal sanctions for parole violators, often influenced by capacity and need (White et al. 2011). As Lin, Grattet, and Petersilia (2010) have demonstrated, these factors can have a meaningful influence on the back-end sentencing decision when considered holistically.

In the current study, we support and extend these analyses by adding a fourth focal concern: parole investment. It has become increasingly clear that both the nature and timing of an individual’s engagement in correctional programming offered to them influences the decisions of the parole board (for example, Ostermann and Hyatt 2018). Variables that relate to an individual’s completion of programming may serve as a proxy for how engaged and committed an individual is to reintegration; program failures may communicate the opposite. ³ Additionally, these data also capture the practical and fiscal costs that have been expended for that individual’s community supervision, a different set of outlays (including sunk costs opportunity costs and general programmatic investments) that may be of interest to the parole board during the revocation decision-making process.

STUDY CONTEXT

The current study focuses on identifying factors within the four dissensions of our updated focal concern framework that are associated with the outcomes of the back-end sentencing process. In this context, an affirmative finding by a parole board that an individual’s supervision should be revoked means that they will likely be promptly returned to incarceration. Generally, the back-end sentencing process begins when a parole officer believes they have probable cause that an individual has violated the terms or conditions of their parole supervision, an infraction that may be direct (new crime) or technical (rule violation) (Steen et al. 2013). In the subject jurisdiction, the officer may arrest the individual at this time, thereby entering them into the “revocation stream,” a term that describes the procedural pipeline from that moment to the eventual decision by the state parole board (SPB). Once entered into this revocation stream, individuals on parole are often incarcerated at a local county jail or in a state parole detention facility. At these facilities, hearing officers ⁴ sequentially hold both probable cause and revocation hearings for each violation case. At the first of these hearings, the parole officer provides testimony about the serious or persistent violation-worthy behaviors that the individual under supervision allegedly committed. If the hearing officer finds that there is probable cause to believe that a violation took place, the case can then progress to a full revocation hearing. At this stage, a new hearing officer assesses the evidence of a violation and, using a more rigorous “clear and convincing” standard, determines if there is sufficient evidence that violation-worthy behaviors occurred. They subsequently provide a recommendation to the two-member board panel. ⁵ This panel is subsequently responsible for making the back-end sentencing decision with regard to the parole violation on that case in light of the evidence and the hearing officer’s recommendation.

The board panel must consider the findings of the hearing officer after the revocation hearing as well as administrative data on the individual’s conduct while under supervision and either affirm or deny the parole revocation. Affirmed revocations result in the person’s return to prison for the remainder of their initial court-imposed sentence or until a date where they can potentially be reconsidered for parole release, which is set at the time of the revocation. Day-for-day credit is granted for the time they were successfully residing in the community during their parole supervision. Denied revocations result in the individual’s supervision term being continued (that is, to resume uninterrupted in the community), although the board panel can set additional conditions that the individual must adhere to in the future and to avoid a future revocation (for example, successfully complete a drug treatment program).

According to statistics published by the SPB that provided data for this study, parole revocation proceedings rarely culminate in an individual having their parole continued. For example, in 2011, the SPB conducted 2,408 revocation hearings; only 506 resulted in a continuance (a revocation rate of 78.9 percent). The rates of parole continuances were similar in 2012, 2013, and 2014 with the SPB conducting 2,368, 2,164, and 1,972 parole revocation hearings respectively and 368, 318, and 329 hearings culminating in a finding that parole continuance (that is, that a violation did not take place) was warranted during these respective years.

DATA AND METHODS

Data Sources and Conceptual Clusters

Data were provided to the research team by a SPB located in a highly populated state on the east coast of the United States. The analyses were conducted on a dataset including the complete population of individuals who were released to parole supervision from January 1, 2005, to September 22, 2011, and who subsequently violated the terms and conditions of their supervision and were considered for parole revocation (n = 13,121). The dataset contained information about each individual’s demographic and case profiles, the date of their release from prison to the supervision of the SPB, the date they were considered for parole revocation, and the date on which their court-imposed sentence was set to expire. Additionally, the dataset included information about the timing and use of post-release parole programs at an individual level. We used the information within this dataset to construct groups of variable classifications that we have labelled as blameworthiness, community protection, practical considerations (the three foundational focal concerns), and parole investments (an additional concern added in this study). We explored the contribution of each set of variables to the ability to predict parole revocations through the construction of a set of logistic regression analyses.

Data Collection and Variable Descriptions

Data on the dimensions of blameworthiness and community protection were derived from the SPB’s administrative dataset, including demographic variables (for example, gender, race, and age at the time of the revocation decision); information highlighting relevant case-control information, such as the individual’s commitment offense type (for example, public order, property, drug, sexual, or violent); whether the state statutorily mandated a term of parole supervision for the individual due to their classification as either a “violent” or “sexual” criminal; the risk classification at the time of the revocation decision according to the results of a level of service inventory-revised (LSI-R) assessment; ⁶ and the county to which the person was initially returned upon their release from prison. This latter variable was used to construct a county relative deprivation index by collecting information about the proportion of the jurisdiction that had female-headed households, the proportion that was Black, the proportion living below the poverty level, and the unemployment rate from the 2000 decennial census. These county-level variables were combined through a principal component factor analysis (α = 0.763; range = 7.863–11.658).

A unique state-level identifier was used to construct the final community protection variable by matching each person included in the SPB’s dataset to their records in a criminal history and recidivism data abstracting system maintained by the state’s Department of Criminal Justice (DCJ). This linkage yielded information about the number and types of crimes for which each person was previously convicted (collapsed into categorical indicators for murder, robbery, and other violent convictions; burglary and other property convictions; drug convictions; sexual convictions; and all other prior convictions) relative to the date on which they were released to parole supervision.

In addition to criminal history data, the DCJ’s data-abstracting system was used to gather case-specific outcomes, including rearrest, and information for each individual under supervision. This exercise allowed for the construction of a practical considerations variable: whether the individual was arrested for a new crime between the date of their release from prison and the date on which the SPB’s board panel considered the individual for revocation. The relationships between each individual’s date of release, revocation hearings, and the date of the expiration of their court-imposed sentence were used to construct the remaining practical considerations variables. Namely, we use these data to construct a set of variables that characterize the timing of, and the timing between, key benchmarks in the analysis: the number of months between each individual’s release from prison and the revocation decision date (that is, the revocation decision date minus the release date); the number of months each person had remaining on their parole term on the date of their revocation decision (that is, the court-imposed sentence expiration date minus the revocation decision date); and the percentage of the term of supervision that had been successfully completed on the date their revocation decision was rendered. This latter variable was constructed by subtracting the parole release date from the maximum sentence expiration date to formulate an individualized parole supervision term in months (µ = 19.75; SD = 11.87; range = 0–66 months). We arrived at a percentage by dividing the supervision term’s length by the number of months spent engaged in the community prior to their revocation decision date.

The new parole investments variables communicate the number of, and time spent within, post-release parole-sponsored programmatic resources. These resources are used to transition individuals on parole into a residential treatment facility (typically for between ninety and 180 days). While attending these programs, participants receive various rehabilitative services such as life skills, cognitive behavior group and individual treatments, anger management, employment readiness, and familial programming. ⁷ In this jurisdiction, individuals on parole are each administered the LSI-R (see generally Schlager and Pacheco 2011; Ostermann and Hyatt 2022), and the results of that needs assessment drive the individualized selection of the programming offered, an approach that is in line with the principles of risk-needs responsivity (Taxman 2019). We used information within the SPB’s dataset that indicate the number of programs successfully completed between an individual’s release from prison and the date on which they were considered for revocation to construct dummy variables (0 = no; 1 = yes), communicating whether each person completed zero (46.8 percent), one (41.09 percent), two (9.85 percent), or three or more programs (2.26 percent). Additionally, we compared the dates on which every individual entered and left each program to construct a cumulative number of months spent in post-release parole programs variable (µ = 2.73; SD = 4.95; range = 0–23). Our final parole investments variable communicates whether, at the time of their initial release from prison, the board panel transitioned each person through a residential post-release parole program (0 = no; 1 = yes). ⁸ This finding, in effect, becomes a measure of the amount of the board panel’s resources that have been expended on that particular individual’s reentry and rehabilitation.

Analytic Plan

Our bivariate analyses used chi-square and t-tests to compare predictors of parole revocation between those people on parole who have their parole revoked by a board panel (n = 10,638) and those who are permitted to continue their parole supervision in the community (n = 2,438). Within our multivariate analyses, we regressed various predictor variables upon the dependent variable of whether a revocation decision results in either a continuance (0) or a revocation of parole and a return to prison (1). Due to the binary nature of our independent variable, a series of iterative logistic regression analyses were constructed to communicate each independent variable’s ability to predict parole revocation (Long 1997). Four models were constructed with each model cumulatively adding the predictors from each of our theoretically driven variable groupings: blameworthiness, community protection, practical considerations, and parole investments. We have communicated the results from our logistic regression analyses through the presentation of beta coefficients as well as the exponential form of the beta value (that is, an odds ratio [OR]) for predictor variables that demonstrates statistical significance at the p ≤ 0.05 level.

Finally, we conducted a series of analyses to demonstrate the robustness of these findings. Post hoc regression diagnostics included the computation of several measures of collinearity for the fully specified model indicated that collinearity was not an issue of significant concern. The highest Pearson correlation coefficient value between each of the thirty-four variables within the final model was – 0.517 (the correlation between the zero post-release parole programs attended and months spent in post-release parole programs variables). Further, the mean variance inflation factor (VIF) value for all thirty-four variables contained within the final model was approximately 3.34; the highest VIF value did not exceed five. ⁹ Finally, a post hoc sensitivity power analysis indicated that our analyses would be able to detect small effect sizes given our large sample size (n = 13,121) and assumed alpha and statistical power levels of 0.05 and 0.80 respectively.

RESULTS

Sample Descriptions and Bivariate Analyses

Bivariate analyses comparing the predictor variables between individuals who had their post-release supervision term continued (N = 2,438) or revoked (N = 10,683) are presented in Table 1. Both groups were primarily male (92 percent for continued and 93 percent for revoked; χ² = 5.99; p = 0.014) and were approximately thirty-three years of age at the time of the revocation decision (for continued: 33.18; standard deviation [SD] = 8.92; for revoked: 33.65; SD = 9.06; t = –2.31; p = 0.020). Racial breakdowns of the two groups were practically similar. Approximately 61 percent of those individuals who were continued on parole and 64 percent of those who were revoked were Black (χ² = 9.08; p = 0.004); 25 percent of those continued and 21 percent of those revoked were white (χ² = 15.90; p = 0.000); 14 percent of the individuals who had their parole continued and 15 percent of the individuals who had their parole revoked were Hispanic (χ² = 0.946, non significant [ns]), and less than 1 percent of both groups were classified as other (χ² = 1.61, ns). Both groups experienced their first felony conviction at approximately twenty-one years of age (21.2 for continued; 21.24 for revoked; ns). Criminal histories were also largely similar. Approximately 2 percent of both groups were serving commitment offenses for public order offenses with 17, 46, 36, and less than 1 percent of those who had parole continued serving commitment offenses that were property, drug, violent, and sex crimes (respectively; all ns). Similarly, 18, 45, 33, and 1 percent of individuals who had their parole revoked were serving commitment offenses that were property, drug, violent, and sex crimes (respectively; all ns). Both groups served approximately twenty-nine months in prison prior to being released on parole (28.3 for continued; 29.02 for revoked; ns).

TABLE 1. Bivariate Analyses Of Predictors Of Parole Revocation Outcomes

Notes: *p ≤ 0.05; **p ≤ 0.01; ***p ≤ 0.001. Means of categorical variables are presented as percentages.

The groups differed significantly with regard to their actuarial risk classifications as indicated by their score of an LSI-R screening administered prior to release from custody. While approximately 8, 32, 42, and 18 percent of individuals who had their parole continued were classified as low, moderate, medium, and high risk (respectively); 5, 26, 45, and 24 percent of individuals who had their parole revoked were classified within the same risk bins. These between-group proportional representations significantly differed across each of the risk categories (low: χ² = 25.04; p = 0.000; moderate: χ² = 33.29; p = 0.000; medium: χ² = 5.61; p = 0.018; high-risk: χ² = 36.37; p = 0.000). Additionally, individuals who had their parole revoked were returning to, on average, counties with higher degrees of concentrated disadvantage (t = –4.79; p = 0.000) and were significantly more likely to be statutorily classified as sex offenders (χ² = 10.94; p = 0.001) when compared to individuals who had their parole continued. The two groups had fairly similar criminal history patterns concerning the number of convictions for murder, robbery, other violent, burglary, other property, drug, sex, and all other convictions they had on their rap sheets prior to their release on parole.

Approximately 40 percent of the individuals who had their parole revoked, as compared to 25 percent of individuals who had their parole continued, were arrested for a new crime between their release from prison and the date of their parole revocation decision (χ² = 206.83; p = 0.000). Individuals who had their parole revoked also were, on average, in the community for a significantly greater number of months when compared to those who had their parole continued (t = –4.53; p = 0.000) and completed a significantly higher proportion of their assigned parole term when compared to individuals who had their parole continued (t = –19.09; p = 0.000). Additionally, about 42 percent of individuals who had their parole revoked and 37 percent of individuals who had their parole continued completed one post-release parole program (χ² = 19.93; p = 0.000), and 9 and 13 percent of individuals who had their parole revoked and continued completed two programs, respectively (χ² = 26.96; p = 0.000). A higher proportion of individuals who had their parole revoked were transitioned from prison through a parole program by the board panel making the initial release decision (χ² = 26.82; p = 0.000), and individuals who had their parole revoked spent significantly fewer months, on average, in post-release parole programs when compared to those who had their parole continued (t = 17.68; p = 0.000).

Logistic Regression Analyses

Results from the four logistic regression models are presented in Table 2. The analytic strategy employs an iterative nesting approach for the regression models, starting with a model that only includes blameworthiness variables; community protection variables, practical consideration variables, and parole investment variables are added successively in subsequent models. All four models were significantly good fits for the data (Model 1: LR χ² = 27.02; df = 5; p = 0.000, LL = –6,285.73; Model 2: LR χ² = 144.33; df = 25; p = 0.000; LL = –6,227.07; Model 3: LR χ² = 842.50; df = 29; p = 0.000; LL = –5,877.99; Model 4: LR χ² = 1,354.13; df = 34; p = 0.000; LL = –5,622.17). While the explanatory power of the primary (demographics only) model was quite low (R² = 0.21), the fully specified model explains approximately 11 percent of the variance in an affirmative revocation decision (R² = 10.75).

TABLE 2. Logistic Regression Analyses Predicting Parole Revocation

Notes. *p ≤ 0.05; **p ≤ 0.01; ***p ≤ 0.001. Black, public order commitment offense, low risk classification, and zero post-release parole programs completed prior to revocation decision serve as reference categories for the race, commitment offense, actuarial risk score classification, and number of post-release parole programs completed prior to decision variables (respectively).

In Model 1—blameworthiness—males are at an approximately 20 percent increased odds of revocation when compared to females (OR = 1.197; SE = 0.100; p = 0.032), and whites demonstrate about an 18 percent decrease in the odds of revocation when compared to Blacks (OR = 0.820; SE = 0.044; p = 0.000). However, upon the inclusion of the community protection variables in Model 2, gender becomes a non-significant predictor of revocation, and the effect of being white (relative to the reference to Black) is diminished (OR = 0.838; SE = 0.050; p = 0.003) and becomes non-significant in Model 3 with the addition of the practical consideration variables.

Several of the community protection variables are significant predictors of revocation in Model 2. While commitment offense type and the makeup of previous convictions on an individual’s rap sheet were not predictive of revocation even in the first model, actuarial risk classification, the deprivation level in an individual’s county of return, and if they were statutorily classified as a sex offender were significant predictors of revocation. Namely, individuals at moderate risk of recidivism had increased odds of revocation of approximately 21 percent when compared to those assessed as low risk (OR = 1.209; SE = 0.115; p = 0.046), with the effect sizes increasing to 55 percent of increased odds of revocation for individuals with a medium-risk assessment score (OR = 1.547; SE = 0.145; p = 0.000) and approximately 90 percent increased odds of revocation for individuals at high risk of recidivism (OR = 1.897; SE = 0.195; p = 0.000) when compared to those with a low-risk assessment. Further, a one-point increase in the county deprivation index scale increased the odds of revocation by approximately 9 percent (OR = 1.086; SE = 0.027; p = 0.001) and being statutorily classified as a sex offender increased the odds of revocation by 70 percent (OR = 1.702; SE = 0.356; p = 0.011).

All of the community protection variables were significant predictors of the SPB’s decision to revoke; Model 2 remained significant with the addition of the practical consideration variables in Model 3. The effects of the risk classification variables were somewhat magnified within Model 3. Individuals at a moderate and medium risk had approximately 23 percent (OR = 1.228; SE = 0.121; p = 0.037) and 70 percent (OR = 1.695; SE = 0.165; p = 0.000) increased odds of revocation, respectively, and individuals at a high risk of recidivism had over two times the odds of revocation when compared to those assessed as low risk (OR = 2.309; SE = 0.247; p = 0.000). While the effect of the county deprivation index variable was lessened when moving from Model 2 to Model 3 (OR = 1.072; SE = 0.028; p = 0.007), the effect of the statutory sex offender variable grew, with individuals who were labeled as such experiencing almost three times the odds of revocation when compared to people without this label, while holding all of the other predictors within Model 3 constant (OR = 2.745; SE = 0.776; p = 0.000).

All of the practical consideration variables were significant predictors of revocation when first included in Model 3. Among all individuals under supervision who entered the revocation stream, those who experienced a new arrest between their release from prison to parole and the date on which the revocation decision was rendered had approximately twice the odds of revocation when compared to people that did not experience a new arrest (OR = 2.070; SE = 0.114; p = 0.000). Further, each additional month that an individual spent on parole prior to the date of their revocation decision translated into an approximate 4 percent decrease in their odds of revocation (OR = 0.963; SE = 0.003; p = 0.000), and each additional percentage point increase in the ratio of the percentage of the parole term completed resulted in an approximate 3 percent increase in the odds of revocation (OR = 1.025; SE = 0.001; p = 0.000). While a significant predictor of revocation, the number of days left on the parole term at the time of the revocation decision exhibited a very low impact on the odds of revocation (OR = 1.001; SE = 0.001; p = 0.033).

Within the fully specified Model 4, all of the community protection variables that were previously significant in Model 3 remained so and retained the magnitude of their effect sizes. The lone exception was the comparison between people under supervision who were classified as moderate and those assessed as low risk on the LSI-R. Within Model 4, which included blameworthiness, community protection, practical considerations, and parole investment variables, the odds of revocation between these two risk groups were statistically indistinguishable. As was the case for the practical consideration variables, all of the parole investments variables were significant predictors of revocation. When comparing individuals who did not participate in any post-release programs, individuals who completed one program demonstrated about a 47 percent increase in their odds of revocation (OR = 1.473; SE = 0.110; p = 0.000), those that completed two programs had about a 32 percent increase in their odds of revocation (OR = 1.321; SE = 0.140; p = 0.009), and those that completed three programs had about four times the odds of revocation (OR = 4.001; SE = 0.822; p = 0.000). Additionally, individuals who were transitioned through a program immediately upon their release from prison had an approximate 78 percent increase in the odds of revocation relative to those who were not transitioned through such a program by a board panel (OR = 1.780; SE = 0.131; p = 0.000). Finally, every additional month spent in a post-release parole program translated into an approximate 11 percent decrease in the odds of revocation (OR = 0.887; SE = 0.005; p = 0.000).

DISCUSSION

The current study was developed to provide further empirical evidence for the applicability of the focal concerns theoretical framework within the back-end sentencing process and to expand the scope of factors considered applicable to parole revocation. The research employs a large dataset, reflecting parole revocation decisions rendered for all individuals who were released to parole and subsequently entered the revocation stream in a highly populated state on the east coast of the United States from 2005 to 2011. Our analyses focused on four domains: blameworthiness (largely demographic data), community protection (case-control information), practical consideration (contextual sentence factors), and a new set of parole investment variables (systemic outlays for community supervision). Bivariate findings demonstrated that gender, age, race, actuarial risk classification, the level of deprivation of the county of return, whether the individual was statutorily identified as a sex offender, and the number of prior burglary, drug, and sexual convictions significantly differed between these groups in expected directions. Additionally, practical considerations, such as whether the individual experienced a new arrest between their release and the revocation consideration, the amount of time they were successfully engaged in the community between their release and the decision date, and the percentage of their parole term successfully completed prior to the revocation decision, also differed between these groups. Whether the parole board made programmatic investments by gearing post-release programs toward an individual considered for parole revocation, whether the individual was transitioned through a program at the time of their release from prison, and the number of months spent in these programmatic resources also differed between those that had their supervision term continued and those that had it revoked and were sent back to prison.

Our multivariate regression models demonstrated that the demographic and case-control variables produced relatively low levels of explanatory power for predicting the odds of whether parole supervision would be revoked or continued. Specifically, demographic variables alone explained less than 1 percent of the variance in the revocation decision; the addition of the case-control variables increased this explanatory power by less than 1 percent. However, our fully specified model coalesces to explain over 10 percent of the total variance in the continuation/revocation decision. Hence, much of the explanatory power behind the SPB’s decision-making processes lies in the addition of the practical consideration and parole investment variables; all of the predictors within these conceptual groupings demonstrating statistically significant predictive power at the p ≤ 0.05 level.

Pivotal categories referencing a history of sexual offending and/or violence were both meaningful in this analysis. As was the case in Lin, Grattet, and Petersilia’s (2010) exploration of the parole revocation decision-making practices, we found that the pivotal category of “sex offender” was a potent predictor of parole revocation. Though our studies differ in operationally defining a “sex offender” (Lin, Grattet, and Petersilia focus on registered sex offenders under California’s Megan’s Law, ours uses a (likely) stricter, jurisdiction-specific statutory definition that targets the state’s “most egregious” sex offenders for mandatory supervision), both studies demonstrate that people classified as sex offenders were over twice as likely to have their parole revoked, all else being equal. Interestingly, while this pivotal category was the strongest predictor of revocation in Lin, Grattet, and Petersilia’s (2010) work, the previously unexplored variable of actuarial risk classification demonstrated similar predictive power as a sex offender classification within the current study; scoring as high risk (as opposed to low risk) increased the odds of revocation twofold. Research has shown some overlap between assessments of sexual and general recidivism, and although the instruments and outcomes remain distinct (see, for example, Hanson and Morton-Bourgon 2009), SPB members may employ these categorical classifications in a similar manner within their decision-making process.

Committing, or at least being arrested for the commission of, a new crime, perhaps unsurprisingly, had a significant influence on the outcome of the back-end sentencing process. It is worth reiterating that, in the case of a direct violation, the parole revocation decision would not affect the prosecution of, and eventual sentencing for, the criminal act itself. Individuals on parole who were rearrested for new crimes, unsurprisingly, had much higher odds of revocation than those who were not. Like those individuals who were assigned the pivotal category of sex offender or actuarially assessed as high risk, individuals who are rearrested are likely classified by SPB members in similar ways, presenting a profile that suggests they are more dangerous and less amenable to treatment, pose an increased public safety risk, and are less likely to be successful if given the chance to continue their parole supervision or lack remorse.

Interestingly, the strongest predictor of revocation within the current study was not history, classifications, or new criminal conduct. Programmatic investments, the new dimension included in this analysis and not represented in previous studies, provided the greatest amount of explanatory power regarding the revocation decision. An individual who completed three or more parole programs during their supervision was over four times more likely to be revoked than people who had not completed any programs. Generally, providing programmatic resources to an individual increased the likelihood that they would have their supervision revoked if they were to be subsequently entered into the revocation stream.

We see three possible interpretations for this finding: First, it is possible that the individual’s failure to remain compliant with the conditions of their supervision after being offered these programmatic resources may increase their perceived blameworthiness during the revocation decision. This increased assessment of blameworthiness, in turn, communicates that they have already had the opportunity to be successful, and, since this was unsuccessful, the provision of additional opportunities, programs, and resources is unlikely to meaningfully change their (re)offending trajectory. Incapacitation through incarceration becomes a more reasonable option; the individual is seen as having run out of chances.

Second, and more pragmatically, the parole board decision makers may have observed that significant programmatic resources had already been spent on that particular person. They may believe that there may not be additional, unique resources that are reasonably available or categorically different than what had already been used to attempt to facilitate reintegration. In these cases, SPB members are faced with the decision to either use more of a similar (or the same) resource that has already proven ineffective at markedly changing the violation-worthy behavior patterns or simply to revoke their parole supervision. There is also an opportunity cost to the SPB; reenrolling the violator may mean that there will not be a space for another individual who has neither violated their supervision nor been in that program before. This exercise in generalizable interpretation is further challenged by a lack of normative agreement, both within and between jurisdictions, about how decisions regarding parole release should be made (see Ball 2011).

Finally, there may be a relationship between the assessed and/or perceived risk levels of the individuals in conjunction with their performance on parole that guides the response by SPB members. Under the principles of risk, need, and responsivity (RNR), people at higher risk of failure should receive more services that target their criminogenic needs (Andrews, Bonta, and Hogue 1990; Andrews, Bonta, and Wormith 2006; Lowenkamp, Latessa, and Holsinger 2006). Importantly, this can impact the allocation of programmatic resources during parole supervision (Ostermann and Hyatt 2022). Here, an analysis of the relationship between risk and the number of pre-revocation programs that an individual received suggests that a greater proportion of higher risk individuals did indeed receive programming. ¹⁰ This difference was most evident in the decision to send people on parole to their first program: 46.73 percent of high-risk individuals, as compared to 32.27 percent of low-risk people, received one program while under supervision. The magnitude of these differences, however, shrinks meaningfully when considering additional program engagement: 2.55 percent of high-risk, 2.48 percent of medium-risk, 1.88 percent of moderate-risk, and, lastly, 1.25 percent of low-risk individuals received three or more programs. Overall, higher risk groups received more treatment, both in terms of the proportion participating in programming and the number of programs in which they participated, a result supportive of engagement with the principles of RNR. It may be the case that, while risk, which is a proxy for community protection, is a factor in the decision to revoke, programming also plays an important, independent role. Both the binary decision to assign any programming before entering people into the revocation stream and the extent of programmatic engagement are illustrated in these findings. ¹¹ These relationships warrant further examination, using additional, more detailed treatment data and within other RNR-focused supervision systems.

The findings support reconceptualization of the functioning of the courts and correctional systems, supporting emphases on cost management, actuarial assessments, and systemic efficiency over the goals of individualized rehabilitation (Garland 2012), a call that reflects other efforts to reconceptualize the framing and assessment of focal concerns in front-end sentencing (Lynch 2019; Ulmer 2019). A focus on the costs for the system, rather than on the individual, during the revocation process is in line with the organizational pressures inherent in the framing of the new penological paradigm (Feely and Simon 1992). Given that the back-end sentencing process is largely opaque and that decision-maker performance is not often measured according to the readily observable number of parole terms continued or revoked, parole board members are free to consider risk more diversely. In this context, this consideration includes both actuarial and perceived risk of harm to the community as well as to the system in which community supervision and revocations take place. Contemporary parole (and the broader community corrections and criminal justice systems) decision making generally emphasizes increased efficiency, utility, and managerialism over rehabilitative orientations (Lynch 1998). Our findings, especially that decision makers are apt to revoke parole for people who have already successfully completed post-release parole programs, support this framing.

Future research should consider the theoretical mechanisms that would encourage and support back-end sentencers to rely on the constellation of factors described here when making a revocation decision. Scholars should attempt to better determine the functioning of back-end sentencing workgroups—namely, what actors are involved, what roles are played by each actor, and how the “going rates” for revocation decisions are collectively interpreted (Eisenstein and Jacob 1977; Lipsky 1983; Nardulli, Flemming, and Eisenstein 1985; Albonetti 1991). Additional complementary methodologies including qualitative and ethnographic approaches should be employed to better understand the decision makers and better contextualize the administrative context where revocations are generated and decided (for example, administrative hearings, parole agencies) (Lynch 2019). In this study, revocation decisions were made by two-person board panels based on information gathered by supervising parole officers and administrative hearing officers. Worthy directions for future research include the individual-level characteristics of panel members; the differential back-end sentencing practices of the various possible iterations of two-member panels (for example, there were a total of fourteen panel members in our study context); how different panel compositions potentially impact perceptions about the appropriate thresholds for revocations; “violation worthy” behaviors; and board panel member’s perceptions of their tolerance for risk, community safety, and cost-efficacy of parole programming.

Additionally, the current inquiry could be narrowed to focus on specific subgroups within the community supervision population. This may include lower risk individuals who may face an increased risk of revocation when over-programming takes place (Ostermann 2022) and individuals who have completed multiple programs according to the type of program in which they have participated (for example, skills, anger management, semi-residential, day reporting, and so on) (see Latessa 2011). Finally, the current analysis should be extended to other jurisdictions, especially those outside the American context, thereby supporting the generalization of the extended theoretical framework developed in this study. ¹²

Our findings suggest that there is a significant and positive relationship between the use of programmatic resources and revocation likelihood. This is both unique and counterintuitive and so should be further explored. A particularly fruitful avenue for future exploration may be the conclusion that front-loading high impact services as a part of a step-down process from prison to the community markedly increases the likelihood of being revoked after entering the revocation stream. To better contextualize and interpret these findings, the factors that correlate with successful program completion should be explored in more depth. These studies may be demographic, risk-based, or related to the substantive focus of the treatment (see, for example, Zanis et al. 2009; Zarling, Scheffert, and Russell 2022). Variables relating to the parole context, including both conditions of supervision and officer-specific factors, may also play a role in program assignment and completion and, accordingly, impact the likelihood of revocation (Ostermann and Hyatt 2022). Understanding their individual contribution to the program-related effects observed here would provide useful information about revocation that could be potentially actionable at the policy level.

Relatedly, as the recidivism literature clearly indicates, the first months after release from prison are the most difficult time for formerly incarcerated people, which translates into the period of time when recidivistic behavior patterns are most likely to manifest (Petersilia 2003; Solomon, Kachnowski, and Bahti 2005; Ostermann and Hyatt 2018). Paired with findings from the “what works” literature, as well as the guiding principles of RNR, the targeting of rehabilitative services toward high-risk formerly incarcerated people in the early months after release from prison has become a cornerstone of evidence-based policy in applied community corrections (Andrews, Bonta, and Hogue 1990; Petersilia 2003, 2004; Andrews, Bonta, and Wormith 2006; Lowenkamp, Latessa, and Holsinger 2006).

The goal of enrolling people under parole supervision into life skills classes, cognitive behavior theory, and anger management is clearly not to increase the likelihood of reincarceration, even when the individual does not complete the full needs-based curriculum. We believe that this relationship may be, at least, in part due to the fact that, in this jurisdiction, the same administrative parole board (though likely comprised of different individual actors), is charged with making the discretionary decision to release, assessing for whom and with what programming frontloading can take place and, on the back end, deciding to revoke or continue parole at the culmination of the revocation stream. This arrangement may be uncommon; other jurisdictions allocate the responsibility for release and revocation in other ways (Ruhland et al. 2016). In this jurisdiction, the fact is that the same agency that is responsible for assessing the sanctioning for an individual who has failed to complete the programing they prescribed, and whose resources are being indirectly expended, may influence the revocation decision. This may be by heightening the signaling of blameworthiness and encouraging the parole board to “get it right” the second time around.

More pragmatically, panel members may be highly aware of the promises and limitations of the programming available by virtue of having assigned them once already. They may feel that, since the individual has exhausted the commonly employed programming, incarceration is the only remaining option. Further empirical and narrative research, therefore, should examine the mechanisms and processes at play when frontloaded services could potentially put certain paroled people at a disadvantage if they enter the revocation stream during the course of their supervision. This work can be situated within the context of the interrelated principles of risk, needs, and responsivity (see, for example, Gendreau and Ross 1979; Andrews, Bonta, and Hogue 1990; Cullen and Gendreau 2000; Byrne, Taxman, and Young 2002; Petersilia 2004), as not all programming is equally effective or appropriate for all individuals under parole supervision. These future lines of inquiry should be paired with a critical eye toward the functioning and content of frontloaded services (for example, the relative quality of the services that are delivered) as well as the context in which the services are delivered). For example, if frontloaded services are delivered in a residential facility that is not functionally different from the prison environments from which the individual was recently released, and the content of the programs is homogeneous or poorly implemented, it should not be surprising to parole decision makers that the services are ineffective. In certain situations, the crucial early days of reentry can be spent delivering a level of treatment and control that largely mirrors what the individual experienced in prison and/or a halfway house. This represents a systematic failure for community corrections, not just an individual’s failure to reintegrate successfully.

LIMITATIONS

The results of our study should be consumed with an understanding of their limitations. Like all secondary data analyses, our work is constrained by the administrative data and measures collected by the community-supervision agencies and subsequently made available. Some key data were never gathered, were not retained, or were stored in a manner unsuitable for the current analytical approach. Examples include the initial conditions of parole supervision, specific behaviors demonstrated by individuals that landed them in the revocation stream in the first place, information about institutional infraction histories, the specific programmatic curriculums that were received during the course of supervision, and prior participation in in-prison treatment programming.

CONCLUSION

The back-end sentencing of individuals on parole is likely a major contributor to prison admission rates across the United States (Travis and Christiansen 2006; Travis 2007). In the indeterminate sentencing jurisdictions where they remain active, parole boards wield a unique set of authorities (Rhine, Mitchell, and Reitz 2019). On the one hand, they make the discretionary release decisions that allow for significant numbers of people to serve a portion of their court-imposed sentences in their communities. On the other hand, parole boards provide supervision and rehabilitative oversight to people who have been released from prison as they attempt to reintegrate. Those people on parole who either commit new crimes during their supervision or run afoul of the rules and regulations of parole are at significant risk of having their supervision terms revoked, a decision that often falls under the purview of the same parole board. Despite these myriad responsibilities and the unique position of parole boards to substantially impact prison population sizes through administrative, rather than judicial, decision-making processes, few empirical investigations have endeavored to better understand the factors associated with the back-end sentencing practices of these key stakeholders (Rhine, Petersilia, and Reitz 2016).

The current study explored various predictors of the decision to revoke paarole supervision by state board members. Driven by a focal concerns theoretical framework (Steffensmeier, Ulmer, and Kramer 1998; Lynch 2019), our analytic strategies culminated in the construction of regression models that explained approximately 11 percent of the variance of this dependent variable, and our analyses contributed to our collective understanding about how these decisions are shaped by theoretically and practically relevant conceptual clusters (demographics, case controls, practical considerations, and parole investments). We have demonstrated that parole board members primarily consider a combination of actuarial risk classifications, statutory identification as a sex offender, and previous programming efforts made by the parole board when rendering their revocation decisions. Interestingly, the more programmatic resources that were geared toward people on parole, as well as the frontloading of high-impact services shortly after release from prison, disadvantaged individuals during the back-end sentencing process. This is a highly counterintuitive finding. It is important, therefore, to continue to better our understanding of how these revocation processes function in practice so that evidence-based and effective community-supervision and sentencing policies can be developed and implemented.