Influences on Parole Officers' Decision making

Concerns over the disparate treatment of suspects or offender groups has generated a considerable amount of research on the exercise and control of discretion in the justice system (e.g., Reference AlbonettiAlbonetti 1991; Reference Gottfredson and GottfredsonGottfredson & Gottfredson 1988; Reference SmithSmith et al. 1984; Reference Spohn and HolleranSpohn & Holleran 2000; Reference SteffensmeierSteffensmeier et al. 1998; Reference WalkerWalker 1993; Reference WooldredgeWooldredge et al. 2005; Reference ZatzZatz 2000). Only a modest amount of research, however, has focused on correctional decision making, and most of these studies have examined the decisions of parole boards (e.g., Reference BonhamBonham et al. 1986; Reference Cavender and KnepperCavender & Knepper 1992; Reference Conley and ZimmermanConley & Zimmerman 1982; Reference GottfredsonGottfredson 1979; Reference GrattetGrattet et al. 2009; Reference Huebner and BynumHuebner & Bynum 2006; Reference Morgan and SmithMorgan & Smith 2008; Reference PogrebinPogrebin et al. 1986; Reference Winfree and WooldredgeWinfree, Sellers, et al. 1990; Reference Winfree and SellersWinfree, Wooldredge, et al. 1990). Few studies have examined the decisions of line-level probation/parole officers (see, e.g., Reference ErezErez 1992; Reference HarrisHarris et al. 2001; Reference SimonSimon 1993). In many respects, however, parole officers' sanctioning decisions are comparable to sanctioning decisions made by other justice system actors. When responding to violations, parole officers consider various sanctioning options, some of which impact offenders' liberty (e.g., revocation). Similar to those of other system actors, parole officers' decisions can be influenced by legally relevant criteria such as the severity of violations or other factors that may be considered extralegal (e.g., gender). Given these parallels, it seems reasonable to inform expectations regarding the influences on parole officers' sanctioning decisions by drawing, in part, from the more extensive research conducted on decision making by other justice system actors. These explanations regarding the influences on sanctioning decisions contribute to a more reasonable examination of whether changes in parole officers' decision making can be expected once more formal constraints are placed on their discretion.

In studies of justice system actors' decision making, researchers have found that legal factors account for most of the variation in sanctioning decisions (e.g., Reference Huebner and BynumHuebner & Bynum 2006; Reference JohnsonJohnson 2006; Reference SmithSmith et al. 1984; Reference SpohnSpohn 2000; Reference Spohn and HolleranSpohn & Holleran 2000; Reference Steffensmeier and DemuthSteffensmeier & Demuth 2001; Reference SteffensmeierSteffensmeier et al. 1998; Reference Ulmer and JohnsonUlmer & Johnson 2004; Reference WooldredgeWooldredge et al. 2005). For parole officers, legally relevant criteria include factors such as the severity of violations and offenders' prior history both before (e.g., static risk classification) and after their release (e.g., number of prior sanctions).

In addition to legal factors, extralegal factors have also been found to influence the decisions of justice system actors, prompting consideration of these variables in related studies (see, e.g., Reference JohnsonJohnson 2006; Reference SpohnSpohn 2000; Reference Spohn and HolleranSpohn & Holleran 2000; Reference Steffensmeier and DemuthSteffensmeier & Demuth 2001; Reference SteffensmeierSteffensmeier et al. 1998; Reference Ulmer and JohnsonUlmer & Johnson 2004; Reference Ulmer and KramerUlmer and Kramer 1996; Reference WooldredgeWooldredge et al. 2005; Reference ZatzZatz 2000). Explanations regarding why extralegal factors may affect decision making have been framed within perspectives such as causal attribution (e.g., Reference AlbonettiAlbonetti 1991; Reference Farrell and HolmesFarrell & Holmes 1991; Reference HawkinsHawkins 1987) or focal concerns (Reference SteffensmeierSteffensmeier et al. 1998). For instance, scholars have theorized that justice system actors' legal decisions are based on their beliefs regarding individuals' potential for reform (Reference Bridges and SteenBridges & Steen 1998). Yet system actors rarely have all the relevant information regarding individuals' prospects for future criminality. In order to reduce the uncertainty involved in these decisions, system actors may be influenced by their preconceptions regarding higher risk offenders (Reference AlbonettiAlbonetti 1991). The foundation for these beliefs is often derived from system actors' attributions regarding the personal and environmental causes of criminal behavior (Reference AlbonettiAlbonetti 1991; Reference Bridges and SteenBridges & Steen 1998; Reference HawkinsHawkins 1987). System actors may hold individuals more responsible if they perceive that the acts perpetrated by those individuals were influenced more by personal factors than by environmental factors; this is because these individuals are considered more culpable and therefore thought to pose a greater threat to society (Reference Bridges and SteenBridges & Steen 1998). Due to their overrepresentation in offender populations, minorities, males, and younger individuals may be perceived as more culpable; this could cause system actors to impose harsher sanctions in cases involving individuals with these characteristics (Reference AlbonettiAlbonetti 1991; Reference WooldredgeWooldredge et al. 2005).

Researchers have also suggested that justice system actors make legal decisions guided by three focal concerns, which include their assessments of individuals' blameworthiness, their interests in protecting the community, and the practical constraints surrounding individuals and organizational resources (Reference Huebner and BynumHuebner & Bynum 2006; Reference SteffensmeierSteffensmeier et al. 1998). Assessments of blameworthiness are primarily influenced by legal factors. Individuals with lengthier criminal histories, those who have committed more serious acts, and those who were guilty of playing a major role in the commission of the act are viewed as more culpable and sanctioned more harshly. System actors have an interest in protecting the community, and this desire influences them to make predictions about individuals' odds of reoffending. Because system actors generally perceive that individuals who commit more serious acts and have more extensive criminal histories are a greater risk to reoffend, they frequently impose harsher sanctions in cases involving individuals who posses those characteristics. Finally, justice system actors are also constrained by their desire to maintain relationships with other members of the justice system, their perceptions regarding certain types of individuals' ability to cope with sanctions, or the social costs of imposing sanctions on someone who has certain responsibilities (e.g., dependent children) (Reference Huebner and BynumHuebner & Bynum 2006; Reference JohnsonJohnson 2006; Reference SteffensmeierSteffensmeier et al. 1998).

Similar to attribution theory, the focal concerns perspective also recognizes that justice system actors typically make sanctioning decisions with limited information regarding individuals' prospects for reform, culpability, and so forth. In order to reduce uncertainty when making decisions, system actors develop perceptual shorthand. Stereotypes derived from individuals' characteristics often provide the foundation for patterned responses that link their interpretations of different focal concerns to the characteristics of individuals and situations (Reference JohnsonJohnson 2006; Reference Spohn and HolleranSpohn & Holleran 2000; Reference SteffensmeierSteffensmeier et al. 1998). As a result, individuals who are younger, male, or racial/ethnic minorities may be sanctioned more severely because these individuals are often perceived to be a greater threat to society, with limited prospects for reform. Minorities, males, and younger individuals are also overrepresented in offender populations, so system actors may perceive them as being better able to cope with the sanctions they impose (Reference Spohn and HolleranSpohn & Holleran 2000; Reference SteffensmeierSteffensmeier et al. 1998).

The context surrounding parole officers' work is also filled with uncertainty (Reference Huebner and BynumHuebner & Bynum 2006; Reference McClearyMcCleary 1978; Reference SimonSimon 1993). Officers typically manage large caseloads with few resources, and contact between officers and the offenders under their supervision is infrequent (Reference PetersiliaPetersilia 2003; Reference SimonSimon 1993). As a result, violations of release conditions are frequent and sanction or revocation decisions are often made with limited information about the offenders' day-to-day adjustment in the community, let alone their prospects for rehabilitation. Even though parole officers are often equipped with assessment tools that aid them in determining offenders' risk to reoffend (Reference Beck and HoffmanBeck & Hoffman 1985), they are also aware of the limitations of those tools (Reference Huebner and BynumHuebner & Bynum 2006; Reference Silver and MillerSilver & Miller 2002; Reference SimonSimon 1993; Reference WrightWright et al. 1984). It seems reasonable, therefore, that parole officers might consider legal factors but, in the same way as other justice system actors, also rely on cues derived from stereotypes of offender groups that they perceive to be higher risk offenders. Consistent with the perspectives discussed above, extralegal characteristics of individuals who are overrepresented in the offender population relative to their distribution in the general population can provide the foundation for these stereotypes (Reference Spohn and HolleranSpohn & Holleran 2000; Reference SteffensmeierSteffensmeier et al. 1998; Reference WooldredgeWooldredge et al. 2005). Offenders who are younger, minorities, male, or from lower socioeconomic strata may be perceived as higher risk or more culpable than other persons because they often symbolize the “dangerous class,” which is thought to pose the greatest threat to communities (Reference SteffensmeierSteffensmeier et al. 1998). Due to parole officers' interest in controlling and reducing recidivism (community protection), these factors may be associated with harsher sanctions. Empirical evidence derived from studies of the decisions of other justice system actors' supports the inclusion of these variables in related models (e.g., Reference Conley and ZimmermanConley & Zimmerman 1982; Reference Huebner and BynumHuebner & Bynum 2006; Reference JohnsonJohnson 2006; Reference SmithSmith et al. 1984; Reference SpohnSpohn 2000; Reference Spohn and HolleranSpohn & Holleran 2000; Reference Steffensmeier and DemuthSteffensmeier & Demuth 2001; Reference SteffensmeierSteffensmeier et al. 1998; Reference Ulmer and JohnsonUlmer & Johnson 2004; Reference Winfree and WooldredgeWinfree, Wooldredge, et al. 1990; Reference WooldredgeWooldredge et al. 2005; Reference ZatzZatz 2000).

Although parole officers work in an atmosphere with a high degree of uncertainty, they are typically more involved with the offenders under their supervision than other justice system actors (Reference LynchLynch 1998; Reference SimonSimon 1993). Parole officers' level of involvement with offenders may provide them with additional information that could influence their decision making. Parole officers are generally required to monitor offenders' conditions of release, such as those governing offenders' residential situations, employment, and rehabilitative treatment (Reference MacKenzieMacKenzie et al. 1999; Reference PetersiliaPetersilia 2003). In making sanctioning decisions, officers may consider these situational attributes of offenders' community adjustment because they may reflect in part offenders' commitment to conventional behaviors. If officers perceive that offenders are committed to conventional behaviors, they may view them as less culpable and attribute less blame when those offenders violate the conditions of their release. In contrast, if officers perceive that offenders are not committed to conventional behaviors, they may consider them higher risk and attribute their behaviors to personal factors rather than environmental influences. Because parole officers have an interest in reducing recidivism, offenders whom they perceive as higher risk may be sanctioned more harshly when they violate the conditions of their release. For example, officers may be less likely to initiate a revocation hearing for offenders who are employed or living in a situation that provides some control over their behavior (such as a spouse or parent). On the other hand, officers may be more likely to pursue a revocation hearing in cases where offenders are homeless or at large. Similarly, officers may be more likely to pursue the revocation of offenders who have been afforded more treatment opportunities, such as community-based services or residential programs. Researchers have observed that other justice system actors consider factors such as employment, family situations, and use of rehabilitative programs (Reference BonhamBonham et al. 1986; Reference Koons-WittKoons-Witt 2002; Reference Moore and MietheMoore & Miethe 1986; Reference PogrebinPogrebin et al. 1986; Reference SimonSimon 1993; Reference Spohn and HolleranSpohn & Holleran 2000; Reference WooldredgeWooldredge et al. 2005).

Effects of Policy Changes on the Sanctioning Process

Despite the accumulation of research on justice system actors' decision making, only a limited number of studies have examined the effects of changes in policies governing responses to crime and deviance on the decisions of justice system actors (e.g., Reference Koons-WittKoons-Witt 2002; Reference MartinMartin et al. 2009; Reference MietheMiethe 1987; Reference Moore and MietheMoore & Miethe 1986; Reference Winfree and WooldredgeWinfree, Wooldredge, et al. 1990; Reference WooldredgeWooldredge 2009; Reference WooldredgeWooldredge et al. 2005). Policies designed to restrict discretion emphasize formal rationality and consistency in sanctioning. Discretion, on the other hand, permits system actors the flexibility to base sanctioning decisions on substantive rationality (Reference SavelsbergSavelsberg 1992; Reference Ulmer and KramerUlmer & Kramer 1996). Substantive rationality in parole officers' sanctioning decisions may involve considering offenders' individual needs, situations, or attributes, but such considerations may also produce sanctioning disparities. Thus, the imposition of policies designed to structure parole officers' sanctioning based on formal rationality presents officers with a fundamental dilemma in social control decision making—uniformity versus individualized treatment (Reference Ulmer and KramerUlmer & Kramer 1996).

Some of the existing studies of policies designed to restrict the discretion of justice system actors have revealed moderate increases in uniformity after the respective policies were put into place (Reference MartinMartin et al. 2009; Reference Moore and MietheMoore & Miethe 1986; Reference WooldredgeWooldredge 2009; Reference WooldredgeWooldredge et al. 2005). Increases in uniformity could be viewed as support for these policies. However, justice system actors are used to enjoying considerable discretion in making sanctioning decisions, and substantively rational sanctioning considerations are often ingrained within elements of the justice system (Reference SimonSimon 1993; Reference Ulmer and KramerUlmer & Kramer 1996). Therefore, it is not surprising that researchers have also revealed that some extralegal factors continue to impact system actors' decision making even after the policies designed to encourage uniformity have been implemented (Reference Koons-WittKoons-Witt 2002; Reference MietheMiethe 1987; Reference Moore and MietheMoore & Miethe 1986; Reference SteffensmeierSteffensmeier et al. 1998; Reference Ulmer and KramerUlmer & Kramer 1996; Reference WooldredgeWooldredge 2009; Reference WooldredgeWooldredge et al. 2005).

In this study, we evaluate the impact of the state of Ohio's implementation of an administrative violation response policy on parole officers' decision making, thus extending this line of research to the context of parole officers' sanctioning decisions. We examine whether the policy achieved its primary procedural objectives of (1) promoting a more uniform response to violations of release conditions, and (2) reducing officers' reliance on revocation hearings to manage offender noncompliance.

Parole and Structured Sanctioning Guidelines in Ohio

The state of Ohio has been a determinate sentencing state since 1996. The Ohio sentencing guidelines are presumptive guidelines where sentencing is structured according to the type of offense, the felony level of the offense (range=1–5, where 1 is the most serious), and the defendant's prior record. Judges can, however, depart from the presumption so long as they provide adequate justification (see Reference WooldredgeWooldredge 2009; Reference WooldredgeWooldredge et al. 2005 for a discussion of Ohio's sentencing guidelines). Although the implementation of sentencing guidelines abolished discretionary parole release, the guidelines still provide for post-release control (PRC) supervision for those offenders who would have previously received parole and discretionary PRC placement for nonviolent offenders.Footnote ¹ The Ohio Department of Rehabilitation and Correction (ODRC) is responsible for supervising all adult felony offenders in the state of Ohio. The Ohio Adult Parole Authority (APA), which is contained within the ODRC, is responsible for the release and supervision of adult felony inmates returning to communities from prison.

Offenders under post-release supervision in Ohio are monitored by APA parole officers, who are responsible for aiding offenders in their transition to the community (e.g., making treatment and employment referrals) as well as monitoring and enforcing the conditions of their release (e.g., collecting urinalysis). If offenders do not comply with the conditions of their release, APA officers may pursue violation hearings, which can result in offenders returning to prison. Ultimately, decisions to revoke or return offenders to prison are made by APA hearing officers during violation hearings; however, the majority of all hearings generally result in revocation.Footnote ² It is presumed that APA officers' decisions to pursue revocation hearings are the key decision in Ohio's revocation process (Reference Martin and Van DineMartin & Van Dine 2008). These decisions are typically made within a short period of time after officers become aware of violations (1–2 weeks). Prior to the implementation of the policy discussed below, decisions whether to pursue revocation hearings were made at the discretion of the supervising APA officers, although supervisors were often consulted.

In consultation with the National Institute of Corrections and the ODRC Bureau of Research, the APA developed the Ohio Adult Parole Authority Progressive Sanction Grid. The sanction grid, which is contained in the Appendix, was implemented in July 2005 as part of a larger policy addressing the sanctioning of offenders who violate the conditions of parole or PRC supervision. The policy revision was issued in response to perceived disparities in the treatment of offenders who violated conditions of supervision. Specifically, the policy was designed to:

The policy was also consistent with the ODRC's larger reentry initiative that was published in 2002 and contained recommendations to develop a violation policy that was supportive of the ODRC's reentry goals (e.g., to reduce recidivism) and that structured and provided statewide consistency in the use of progressive sanctioning (ODRC 2002).

More specifically, the policy directs officers to consider both public safety and proportionality when addressing violation behavior, and it refers officers to the sanction grid in order to determine the most appropriate response to violations. Similar to state sentencing guidelines, the sanction grid groups offenders by risk and violation severity. Offender risk scores are based on a static risk assessment developed in Ohio that is used to determine offenders' initial level of supervision. The assessment tool primarily consists of indicators of offenders' criminal history (e.g., prior commitments) and classifies individuals as high, medium, or low risk. Violation behaviors are grouped into three categories: major violations, high-severity violations, and low-severity violations. Major violations, which include sex offender violations, new felony offenses, weapons offenses, and threat behaviors, are not required to be addressed by grid. High-severity violations include behaviors such as absconding, certain misdemeanor offenses, and programming violations, while low-severity violations include reporting, employment, substance abuse (e.g., positive drug tests), and other minor infractions.Footnote ³

Responses to violations are determined by cross-classifying the risk and violation severity categories by the number of sanctioned violation incidents that have occurred since the offender's release. Sanction options are then provided in terms of a level of organizational action, each of which includes multiple sanction alternatives that are detailed in the policy. Officers maintain some discretion concerning the sanction alternative that is imposed within each level of action, although the sanction grid does presume that sanctions will be imposed for all violations. The sanction grid also presumes multiple opportunities to impose unit-level sanctions before proceeding to a violation hearing. As such, the policy was implicitly designed to reduce officers' reliance on revocation hearings and promote the management of offender noncompliance in the community. Generally, the sanction grid is consistent with administrative graduated sanctioning policies that have been proposed or implemented in other jurisdictions across the United States (see, e.g., Reference BurkeBurke 1997; Reference TaxmanTaxman et al. 1999).

Method

The analyses described here focus on parole officers' decisions to pursue revocation hearings before and after the implementation of the administrative violation response policy. We focused on decisions to pursue violation hearings as opposed to other sanctioning outcomes because hearing decisions were to be made in accordance with the legally relevant criteria contained within the sanction grid. Ultimately, the policy was designed to ensure that sanctions were applied in a progressive manner, eventually graduating to (but ideally preventing) a revocation hearing.

The target populations for the study included all the violations committed by offenders who were released on discretionary parole or PRC in Ohio during a three-month period before (October–December 2003) and after (August–October 2005) the policy was implemented statewide. Each offender was followed for a full year after release, permitting an examination of the effects of the policy for approximately one year after it was implemented. Using the procedures described below, we collected data regarding officers' responses to 3,291 violation incidents from the case files of 1,222 offenders.

Data and Measures

The study required information regarding offenders' violations of their release conditions and parole officers' responses to those violations. Because it was unclear which offenders had violated the terms of their release prior to examination of the offenders' case files, the offender samples had to be initially representative of all offenders released on post-release supervision during the time frames described above. For each sample, offenders were selected from a list of all the offenders released under post-release supervision in Ohio for the first time in their current case during the periods mentioned above. Female offenders were selected with certainty and males were selected randomly, with the goal of 95 percent confidence intervals for parameter estimates. The male sample also included an oversample of 20 percent to account for unusable cases (e.g., interstate compacts), cases with missing data, and so forth. These procedures resulted in 1,040 and 1,012 offenders for the two samples, respectively, making 2,052 offenders altogether. Sample weights were derived to adjust for the oversampling of female offenders. These weights were normalized for the multivariate analyses.

Information regarding each offender was collected from a number of official sources, which were cross-referenced against each other in order to increase the reliability of the data. The majority of the information used for this study was obtained from offender case files. Parole officers' field notes were examined for the timing of a number of the offenders' post-release release activities (e.g., violations, sanctions, employment), and this information was cross-referenced with a number of supporting documents. Offenders were followed for a full year after their release or, if applicable, until the date they were returned to prison. Following offenders over time permitted any changes in offenders' situations (e.g., employment) between violation incidents to be recorded. It is important to note, however, that the one-year follow-up period did restrict the generalizability of the results to short-term effects of related policies. The data were collected by two researchers, and an interrater reliability analysis conducted using a random sample of 10 cases revealed an internal consistency between the two researchers of .93.

As discussed above, the study focused on parole officers' responses to offenders' violations of the conditions of their release, so only those offenders who had violation incidents recorded within the first year after their release were ultimately included in the study. Of the 2,052 offenders initially sampled for the study, 1,266 had a recorded violation incident. Offenders were also removed if they were released to outstanding warrants (N=21) or had missing data on any of the variables of interest (N=23). These procedures reduced the sample used here to 1,222 offenders, who had 3,291 violation incidents recorded within the first year after release.

All the measures included in the final models are described in Table 1. As discussed above, the structure of the data was hierarchical, with violation incidents nested within offenders, and variables were measured at both levels of analysis.Footnote ⁴ The outcome variable was a dichotomous measure of whether an officer pursued a hearing for a violation incident. In addition to the outcome variable, several of the predictor variables were also measured at the incident level (within individuals) in order to allow their effects to vary over time (i.e., offenders' situations and the characteristics of their violations may vary between incidents). We included dichotomous variables measuring the type of violation: a major violation or a high-severity violation. Low-severity violations were treated as the reference category. Coding procedures for these categories were generally consistent with the language of the policy described above. If multiple violations were recorded as a part of the same incident for which the most serious violation was selected, however, the models also included a dichotomous variable indicating whether multiple violations were recorded. Also consistent with the policy language, we included the number of prior sanctions that had been imposed prior to each incident. Characteristics of offenders' situations at the time sanctions were applied were also examined. Specifically, we included measures of the number of service referrals that parole officers had made on an offender's behalf, measures of whether the offender was employed, and measures of different living situations, including whether an offender was married and cohabitating, cohabitating with a parent, placed in a residential program, or was homeless or at large. In order to adjust for time at risk, a measure of the number of days under supervision was also included as a statistical control.Footnote ⁵

Table 1. Sample Means and Standard Deviations (Unweighted)

Note: All measures are dummy-coded except number of prior sanctions, number of service referrals, number of days under supervision, age, and felony level of committing offense.

Offender-level (between individual) predictors were also considered. The primary variable of interest was whether the offender was a post-policy case, which indicated they had been released in the period after the violation response policy was implemented. Other offender-level variables included offenders' age at the start of supervision, whether they were female or African American, the felony level of committing offense (range=1–5), and whether the offenders were classified as a sex offender, high-, or low-risk.Footnote ⁶ Sex offenders were defined as a separate risk category regardless of their risk score because responses to violations committed by sex offenders were decided (per policy) on an individual basis. The measures of high and low risk were taken from the ODRC's additive static risk assessment, which is considered within the sanction grid described in the Appendix. Finally, because some offenders in the sample were sentenced prior to the implementation of the Ohio sentencing guidelines (1996), we included a measure of whether offenders were released on discretionary parole as opposed to PRC as a statistical control.

Statistical Analysis

Examination of parole officers' decision making differs from studies of most other justice system actors because officers frequently encounter offenders multiple times during a study period (i.e., offenders often violate release conditions more than once). For example, the data examined in this study contained 3,291 violation incidents committed by 1,222 offenders. This situation creates problems for conventional analytical techniques such as pooled ordinary least-squares regression (e.g., violation incidents are not truly independent of the offenders who commit them) (Reference Raudenbush and BrykRaudenbush & Bryk 2002).

The hierarchical data structure (violation incidents nested within offenders) required the creation of bilevel data sets with violation incidents at level-1 and offenders at level-2. Creating the bilevel datasets with violation incidents at level-1 allowed us to (1) adjust for non-independence among incidents “nested” within the same offender, (2) base the hypothesis tests on the appropriate sample sizes (for incidents versus offenders), and (3) remove (through group mean-centering) between-offender variation in situational characteristics that might have corresponded with differences in hearing rates across offenders (e.g., higher risk offenders may have been less likely to be employed).

The evaluation of the effects of the sanction grid involved estimating a pooled model that included all the cases and the offender-level measure indicating whether the case was a post-policy case. Separate models were then estimated for each sample (before and after), permitting the comparison of coefficient estimates derived from the two samples (i.e., did the effects of particular measures become stronger after the policy was implemented?).

The dichotomous outcome measures were examined using hierarchical Bernoulli regression. First, an unconditional model revealed significant variance (p<.05) in each outcome at level-1 (incidents) and level-2 (offenders). Next, fixed-effects models were estimated that included each of the level-1 predictors variables. In these models, the level-1 model intercepts were allowed to vary randomly across offenders, permitting an analysis of the main effects of the level-2 predictors on the level-1 model intercepts. Reliability indexese for each of the model intercepts exceeded .50, indicating adequate within-group sample sizes for estimation of hierarchical models (Reference Raudenbush and BrykRaudenbush & Bryk 2002). Consistent with objective (3) from above, the measures of offenders' situations were centered on their means for each offender. By contrast, the level-1 legally relevant variables (e.g., number of prior sanctions) and control variables were grand mean–centered in order to control for their effects at level-1 as well as to adjust the level-1 intercepts for their effects (see Reference Raudenbush and BrykRaudenbush & Bryk 2002). Last, the level-2 predictors were entered, permitting the examination of the main effects of the level-2 predictors on the level-1 intercepts. All the level-2 predictors were left uncentered because there was not a higher level of aggregation. The final level-2 models also included a level-2 control for the number of violation incidents recorded for each offender.Footnote ⁷

Once the final models were estimated for both the pre- and post-policy samples, the coefficient estimates were compared using the equality of coefficients test developed by Reference CloggClogg et al. (1995). Reference BrameBrame et al. (1998) demonstrated the applicability of this particular test to a comparison of maximum likelihood coefficients between two independent samples.

Findings

Table 2 contains the incident-level estimates derived from the model of the pooled sample as well as the results generated from analyses of the before and after samples. Table 3 displays the offender-level main effects for all three models. Tables 2 and 3 also contain the differences between the before and after coefficient estimates denoted by the z-scores that resulted from the equality of coefficient tests. Only the z-scores that resulted from tests indicating a significant difference in the magnitude of the effects are reported.

Table 2. Bernoulli Models Predicting Officers' Decision to Pursue a Revocation Hearing (Maximum Likelihood Coefficients Reported With Standard Errors in Parentheses)

^**p<.01; ^*p<.05.

Table 3. Level-2 Main Effects on Officers' Decisions to Pursue Revocation Hearings (Level-1 Intercepts as Outcomes)

Note: Models include a control for the number of violations.

^**p<.01; ^*p<.05.

General Effects on Parole Officers' Decisions to Pursue Revocation Hearings

The results from the analysis of the pooled sample contained in Table 2 show that officers were more likely to pursue a hearing for incidents that involved a major violation, a high-severity violation, or multiple violations. Officers were also more likely to pursue a hearing for incidents that involved offenders who had been sanctioned more times during their term of supervision. It was also more likely for officers to initiate hearings if the offenders involved were homeless, at large, or living in residential programs. In contrast, officers were less likely to pursue hearings if offenders were employed or had been under supervision for longer periods of time. Neither the number of service referrals nor whether an offender was cohabitating with a parent or spouse had an effect on officers' hearing decisions. The significant predictors accounted for a notable percentage (34 percent) of the incident-level variation in officers' hearing decisions.Footnote ⁸

Turning to the offender-level results (Table 3), the analyses revealed that officers were less likely to pursue hearings in cases involving offenders who were released in the period after the administrative violation response policy was implemented. Officers also pursued fewer hearings for violations involving African American offenders or offenders classified as low-risk. On the other hand, officers were more likely to initiate hearings for violations involving offenders who were incarcerated for more serious felonies or were designated as sex offenders. Offenders' age, sex, whether they were high-risk, or released on discretionary parole had no effect on officers' hearing decisions. The relevant predictors in the model explained 27 percent of the between-offender variation in hearing decision rates.

Comparing Officers' Decisions Before and After the Implementation of the Policy

Confirming the results derived from the pooled sample, the incident-level analyses (Table 2) revealed that officers pursued fewer hearings after the administrative sanction policy was implemented. The number of prior sanctions affected officers' decisions in both time periods; however, the effect was stronger in the period after the policy had been implemented. The effects of number of service referrals, employment status, and living in a residential program also differed between the two time periods. Officers were more likely to pursue a hearing if they had made more service referrals in the period before the policy was put into practice but less likely to pursue a hearing in those cases in which they had made more service referrals in the period after the policy was implemented. Offenders' employment status had no effect on officers' decisions to initiate a hearing before the policy was implemented. However, after the policy was implemented, officers were less likely to initiate hearings if offenders were employed. In the time period before the policy was implemented, officers were more likely to proceed to a hearing if the offender lived in a residential program, but whether an offender lived in a residential program had no effect on officers' decisions in the period after the policy was put in place. All the other incident-level effects were either consistent with those derived from the analysis of the pooled sample or not significantly different between the two time periods. The two models explained a comparable amount of the incident-level variation in officers' hearing decisions, 38 percent and 42 percent, respectively.

Table 3 shows that, compared to the time period before the policy was implemented, officers were more likely to pursue hearings for violations that involved sex offenders after the policy was put in place. Otherwise, all the offender-level effects were consistent with those generated from the analysis of the pooled sample or not significantly different between the two time periods. The post-grid model explained 32 percent of the variation in hearing decision rates, compared to only 21 percent in the pre-grid model.

Discussion and Conclusions

Administrative graduated sanction models emphasize formal rationality by structuring the sanctioning of parole violations according to legal criteria such as the severity of violations, the number of previous sanctions imposed, and the level of risk offenders pose to the community. The discretion afforded parole officials is restricted. Sanctions are presumed to be certain, and they are applied in a progressive, incremental manner. By ensuring consistent responses to violations, potential disparities in the treatment of offender groups are purportedly reduced and uniform responses to violations are encouraged (Reference BurkeBurke 1997; Reference TaxmanTaxman et al. 1999). Graduated sanctioning models also hold the promise of proactively preventing revocation hearings by compelling parole agencies to manage offender noncompliance in the community (Reference BurkeBurke 1997; Reference TaxmanTaxman et al. 1999). This study contributed to the limited knowledge regarding these reforms by examining the short-term effects of an administrative violation response policy on parole officers' decisions to pursue revocation hearings.

In general, we found that the policy change was effective in reducing the number of hearings that parole officers initiated. Officers only pursued hearings for 12 percent of violation incidents after the policy was implemented, compared to 17 percent before the policy was implemented. The multivariate analysis confirmed these findings by revealing that, once relevant factors were controlled for, officers pursued fewer hearings for violations that occurred during the time period after the policy was put in place. Because the policy was implicitly designed to compel officers to manage offender noncompliance in the community (as opposed to initiating revocation proceedings), it seems that the imposition of the policy was effective in achieving this goal.

The findings from this study also reveal several differences in how offenders who violated the conditions of their release were treated before and after the implementation of the policy. In the period after the policy was put in place, officers' hearing decisions were influenced more strongly by the number of sanctions previously imposed during the offenders' period of supervision. This finding also suggests that officers were imposing more community-based sanctions prior to initiating a hearing in the post-policy time period.

Whether offenders had received treatment services (i.e., number of service referrals, living in a residential program) increased the likelihood that officers pursued hearings in the period before the policy was implemented, but it had little effect on hearing decisions in the period after the policy was implemented. Officers may have been more likely to initiate revocation proceedings for offenders who had received treatment in the pre-policy period because they felt that those offenders had been given an opportunity to rehabilitate themselves. In the period after the policy was implemented, however, officers were constrained by the policy, which structured sanctioning according to formally rational criteria. In the period after the policy was put in place, officers were also less likely to pursue hearings for violations committed by offenders who were employed. On the one hand, this finding suggests that officers considered this seemingly extralegal characteristic of offenders' situations when making their decisions. In his study of California parole officers, Reference SimonSimon (1993) also found that offenders' employment status influenced officers' violation decisions. However, it could also be that employed offenders committed fewer violations in general, and as a result of the policy did not “graduate” to the point where Ohio officers would pursue a hearing.

The results of the offender-level analyses revealed very few changes in the factors that influenced parole officers' hearing decisions resulting from the implementation of the policy. For example, the demographic characteristics of offenders had no effects on officers' hearing decisions in either time period (before or after). Consistent with the policy directive to treat sex offenders as a separate risk category, parole officers were more likely to pursue hearings for sex offenders in the time period after the policy was implemented. The decision to treat sex offenders as a separate risk category followed from the perceived risk that sexual offenders pose to public safety (see, e.g., Reference Huebner and BynumHuebner & Bynum 2006). Overall, then, the imposition of the policy resulted in very few increases in uniformity because there were few disparities that resulted from officers' decisions to pursue a hearing in the first place.

The broader implications of the findings regarding the effects of the policy may be that concerns regarding the disparate treatment of offenders that often fuel the implementation of policies designed to restrict parole officials' discretion are unfounded. Parole officials appear to make sanctioning decisions primarily based on legally relevant factors and indicators that may reflect offenders' adjustment in the community (e.g., employment). However, the findings here do suggest that violation response policies such as Ohio's can reduce officers' reliance on revocation hearings and promote the management of offender noncompliance with community-based sanctions. These latter findings are promising because the state did not increase the number of sanctioning options, programs, or other alternatives to incarceration after the policy was implemented. Officers simply used the resources already available to them to manage offender noncompliance within the community. Such a finding could be important to states that are seeking to reduce the influence of paroled offenders (via revocation) on prison populations (see, e.g., Reference Blumstein, Beck, Travis and VisherBlumstein & Beck 2005).

Administrative graduated sanction policies such as the one under study here are also predicted to increase offender compliance by promoting consistency, uniformity, and ultimately fairness in the application of sanctioning (Reference BurkeBurke 1997; Reference TaxmanTaxman et al. 1999). Although not the focus of this study, the findings may also be relevant to states seeking to increase offender compliance by enacting policies that promote uniformity in sanctioning. A separate analysis of the data examined in this study revealed no differences between the pre- and post-policy samples in offenders' odds of recidivism (Reference Martin and Van DineMartin & Van Dine 2008). On the one hand, these findings suggest that Ohio's violation response policy did not achieve an increase in offender compliance; however, such findings must be tempered by the results of this study, which revealed a reduction in officers' use of hearings but no significant increases in uniformity across officers' sanctioning practices. These results imply that officers were applying sanctions fairly prior to the implementation of the policy. Thus, if uniformity in sanctioning contributes to offenders' perceptions of fairness in sanctioning, which in turn increases compliance, then there is no reason to expect that offender compliance (other than compliance measured by a reduction in revocations for technical violations) would have increased as a result of the policy. Of course, all this assumes that uniformity in the application of sanctions influences offenders' perceptions of fairness. It may very well be that what offenders perceive as fair and what appears to be fair from the perspectives of justice system actors is not the same. Future researchers may want to investigate these issues.

The primary focus of this study was on the effects of the administrative sanction policy on parole officers' hearing decisions. However, the findings also add to the limited research on parole officers' decision making. Influences of parole officers' decisions were considered within perspectives on justice system actors' decisionmaking. Consistent with these perspectives and studies of justice system actors' decision making (e.g., Reference Huebner and BynumHuebner & Bynum 2006; Reference JohnsonJohnson 2006; Reference SmithSmith et al. 1984; Reference Spohn and HolleranSpohn 2000; Reference SpohnSpohn & Holleran 2000; Reference Steffensmeier and DemuthSteffensmeier & Demuth 2001; Reference SteffensmeierSteffensmeier et al. 1998; Reference Ulmer and JohnsonUlmer & Johnson 2004; Reference WooldredgeWooldredge et al. 2005; Reference ZatzZatz 2000), we uncovered that legal factors accounted for most of the variation in hearing decisions at both the incident- and offender-level of analysis. The results from the incident-level analysis of the pooled sample indicated that the legal predictors accounted for 88 percent of the explained variation in officers' revocation hearing decisions. The offender-level legally relevant variables accounted for 96 percent of the explained variation in hearing decision rates. No substantive changes in these findings emerged in the analysis of the pre- or post-policy sample. Thus, it seems that parole officers arrive at their decisions to pursue revocation proceedings primarily by considering formal rational criteria that seemingly represent the “risk” that offenders pose to the community.

In contrast to a number of studies of the decisions of other justice system actors (e.g., Reference JohnsonJohnson 2006; Reference SmithSmith et al. 1984; Reference Spohn and HolleranSpohn 2000; Reference SpohnSpohn & Holleran 2000; Reference Steffensmeier and DemuthSteffensmeier & Demuth 2001; Reference SteffensmeierSteffensmeier et al. 1998; Reference Ulmer and JohnsonUlmer & Johnson 2004; Reference ZatzZatz 2000), the results of this study reveal that the demographic features of offenders under supervision had virtually no effect on parole officers' hearing decisions. Although the coefficient for offenders' race reached statistical significance in the pooled model, the magnitude of the effect was weak and not in the expected direction. These findings, along with the results of the comparisons of the findings derived from the pre- and post-policy samples, suggest that there was little or no disparity that resulted from officers' hearing decisions. In general, these findings are consistent with other studies of line-level probation/parole officers' sanctioning decisions (e.g., Reference HarrisHarris et al. 2001; Reference SimonSimon 1993).

It may be that parole officers rely less on the demographic characteristics of offenders than other justice system actors because they are often more involved with offenders' lives (e.g., Reference LynchLynch 1998; Reference SimonSimon 1993). This increased involvement of parole officers in offenders' lives may decrease the uncertainty surrounding their sanctioning decisions. In addition to legal factors, parole officers may instead be influenced by more proximate, albeit still substantively rational concerns such as situational indicators of offenders' risk to reoffend. Indeed, the findings from this study reveal that officers were less likely to initiate revocation hearings in cases where offenders were employed. Officers could be more willing to use community sanctions (as opposed to pursue revocation) to hold offenders accountable if they have demonstrated some indication of conventional behavior (e.g., employment). However, if offenders have not demonstrated some level of commitment to conventional behavior, then officers may be more willing to attribute blame to the personal characteristics of those offenders as opposed to environmental factors. Researchers have observed that justice system actors who attribute blame to the personal characteristics of individuals rather than environmental characteristics often sanction them more harshly (e.g., Reference Bridges and SteenBridges & Steen 1998). In further support of this idea, officers were more likely to pursue hearings if offenders were homeless or at large at the time of their violation. Officers also initiated hearings more often if offenders had been afforded an opportunity to rehabilitate themselves in a residential program but violated their conditions of release while in that program. In contrast, offenders who were cohabitating with conventional others (such as a parent or spouse) were not treated differently than offenders living alone or with friends or relatives. Regardless of these latter findings, however, the mixture of results observed here suggests that future studies should examine the relevance of more direct measures of these concepts (e.g., commitment to conventional behaviors). Future studies may also examine what factors parole officers perceive to be reflective of offenders' commitment to convention. The limitations of our measures aside, the findings from this study do suggest that officers are less likely to pursue revocation hearings for offenders who are successfully reintegrating into the community but more likely to initiate revocation proceedings if offenders do not appear to be adjusting to the community. These findings are consistent with Reference SimonSimon's (1993), and similar findings have also been observed in studies of judicial sentencing practices (e.g., Reference Spohn and HolleranSpohn & Holleran 2000; Reference WooldredgeWooldredge et al. 2005).

Taken together, the results discussed above suggest that when faced with uncertainty, parole officers are much more likely to consider substantively rational factors related to successful reintegration over the demographic features of offenders. In other words, the increased involvement that parole officers have in offenders' lives provides them with more information than other criminal justice actors (Reference LynchLynch 1998; Reference SimonSimon 1993). This information, in turn, provides some of the foundation for officers' determinations regarding offenders' prospects for reform (perceptual shorthand). Although still limited in many respects, the substantively rational information parole officers consider seems to be more proximately linked to successful reentry into the community instead of more limited considerations of gender, age, or race. From parole officers' perspectives, offenders whose behaviors demonstrate some level of commitment to convention may be viewed as those offenders who are less likely to reoffend. Of course, additional research including more direct measures of the relevant concepts is needed to substantiate these processes, and future studies may also examine other characteristics of offenders' situations such as whether they have children or their level of education. Studies of decisions by other justice system actors have found that these factors influence legal decision making (Reference Griffin and WooldredgeGriffin & Wooldredge 2006; Reference Koons-WittKoons-Witt 2002).

In addition to the limitations regarding some of the measures used here, it is worth noting several other limitations to this study. First, a number of the variables were created from information retrieved from official sources. Even though attempts were made to increase the reliability of the measures by cross-referencing the information across multiple sources, the information is still potentially subject to some discretionary recording by parole officials. Second, the findings are only generalizable to short-term effects of the policy because offenders were followed for only one year after their release. Several studies of state sentencing reforms have revealed some differences in short- versus long-term effects (e.g., Reference Koons-WittKoons-Witt 2002), although other studies have found virtually no differences over time (e.g., Reference WooldredgeWooldredge 2009). Absent a longer follow-up period, however, we are unable to discern whether the differences between the pre- and post-policy time periods persisted over time or whether new differences emerged. Finally, the limited amount of data collected here prohibited us from examining the potential relevance of community context on officers, sanctioning decisions. Researchers of courtroom actors' sentencing decisions have uncovered important variation in sanctioning across county contexts (e.g., Reference JohnsonJohnson 2006; Reference Ulmer and JohnsonUlmer & Johnson 2004). Although parole officers report to the state, whereas courtroom actors work within counties, it is certainly possible that the context within which offenders are situated could influence officers' revocation decisions (see Reference SmithSmith et al. 1984 for a related discussion pertaining to police officers' arrest decisions). Future research may investigate this possibility.

Altogether, this study of the effects of Ohio's implementation of an administrative graduated sanctioning policy revealed a decrease in officers' reliance on revocation hearings and some modest increases in uniformity. It appears, however, that the effects of the policy on reducing disparities were minor, simply because the relationships between offenders' extralegal characteristics and officers' revocation decisions were weak to nonexistent in the first place. Changes may also have been less likely because the initial sanctioning procedures were generally perceived as fair by officers (Reference MakariosMakarios et al. 2010; Steiner et al. 2011). Studies of the effects of sentencing guidelines have revealed similar results (e.g., Reference Griffin and WooldredgeGriffin & Wooldredge 2001; Reference WooldredgeWooldredge et al. 2005). Further, many of the factors that were related to officers' hearing decisions could be considered indicators of successful reentry, as opposed to demographic characteristics such as race or gender. Still, the findings from this study are limited to one state, so studies of other jurisdictions may uncover different results. The need to consider parole reform is clear (Reference BurkeBurke 1997; Committee on Community Supervision and Desistance From Crime 2008; Reference SolomonSolomon et al. 2005; Reference Travis and LawrenceTravis & Lawrence 2002; Reference Travis and PetersiliaTravis & Petersilia 2001; Reference Travis, Visher, Travis and VisherTravis & Visher 2005), and it is only through continued evaluation of promising reentry initiatives that legal researchers can better understand what works and what does not.