Active Supervision and Its Impact Upon Parolee Recidivism Rates

Abstract

Studies that compare recidivism rates between parolees and unconditionally released inmates typically attach these statuses upon release, and then follow these groups until they either fail or meet the censor date. However, this method of identifying former inmates as parolees does not comport with how parolees are conceptualized by the agencies that supervise them. Parole boards identify parolees as released inmates whom they actively supervise. This study explores the relative impact of this strategy of attaching the parole status compared with the traditional strategy used throughout the recidivism literature. I use 3 years of postrelease data from all prisoners released from 2005 to 2007 in a highly populated state on the East Coast (N = 29,299). My findings indicate that after 3 years, parolees are predicted to recidivate at a 1% lower rate compared with unconditionally released inmates when the time of active supervision is not considered. However, parolees who are assigned supervision terms of at least 3 years evidence a predicted 8% lower recidivism rate when compared with unconditionally released inmates. These findings demonstrate that parole boards can be successful at isolating those under their active supervision from reengaging in criminal activities when compared with those who are not supervised post-release, but that parole does not have long-lasting rehabilitative effects. This lack of long-term impact is likely associated with a parole board’s focus on offenses that occur solely during the course of active supervision that may create incentive to manage cases in such a way that undermines the pursuit of long-term rehabilitative goals in favor of working toward short-term successful discharges.

Keywords

parole recidivism corrections community supervision discharge

Hundreds of thousands of people are released from our prisons every year and attempt to reintegrate into society (Petersilia, 2003; Travis, 2005). Most are assisted with this transition through parole supervision (West, Sabol, & Greenman, 2010). Despite much of the prison population serving a portion of their sentence in the community, several studies, those that are nationally representative (Solomon, Kachnowski, & Bhati, 2005) as well as those particular to certain states (Jackson, 1983; Ostermann, 2011a; Sacks & Logan, 1979, 1980), have indicated that parole is not overly effective at reducing recidivism relative to those who are not supervised upon their release from prison. However, the currently available studies that have compared reoffense patterns of those released to parole supervision with those who are not, have largely failed to incorporate the time that parolees are actively supervised into their analyses.

Within the current recidivism literature, an individual is generally identified as a parolee at the release event, and this status remains until the individual either fails in the community or the follow-up time expires (Gottfredson, Mitchell-Herzfeld, & Flanagan, 1982; Ostermann, 2011a; Schlager & Robbins, 2008; Solomon et al., 2005; Spivak & Damphousse, 2006; Spivak & Sharp, 2008). This is essentially an ad infinitum status. However, this does not comport with how parolees are ultimately identified and supervised by paroling authorities. Assigning the parole status in such a fashion lacks construct validity (Shadish, Cook, & Campbell, 2002) because after parolees have served the remainder of their sentence under a parole board’s supervision, they are discharged and are no longer considered parolees by the supervising entity. After discharge, the former inmate is not required to report to their parole officer and is no longer offered services and other resources that are made available through the agency.

This study investigates whether operationalizing the construct of parole according to the traditional strategy that has been used throughout the literature versus a construct that incorporates the time that parolees are actively supervised impacts recidivism rates in meaningful ways. Explorations of the effect of parole supervision are of critical importance to contextualizing our collective understanding of whether parole serves mainly as a short-term deterrent or whether it is able to accomplish more long-term rehabilitative goals. Throughout this study, I define recidivism as either an arrest for a new crime or a parole violation that resulted in a return to custody, whichever occurs soonest after release. Twelve logistic regression models are used to predict recidivism over several follow-up times, including 6 months, 1 year, 1.5 years, 2 years, 2.5 years, and 3 years. Models are constructed twice for each follow-up time. One model uses a variable that encapsulates the traditional parole construct and the other uses a variable that incorporates active supervision into its operationalization of parole. To isolate the impact of parole upon recidivism, identical control variables that have been previously found to be predictive of re-offending are entered into the models. Data from all individuals released from prisons within a state on the East Coast from 2005 to 2007 are used for this exploration (N = 29,299).

I use two postestimation strategies to communicate my findings in addition to the analysis of odds ratios (OR) from the logistic regression models. First, I present results from average marginal effect analyses. These analyses compare the predicted probabilities that parolees recidivate relative to unconditional releases within each follow-up time while controlling for the predictor variables entered into the regression models. Second, I simultaneously estimate both regression models within a given follow-up time through seemingly unrelated estimation analyses and conduct post hoc tests of the variables that encapsulated the different parole constructs. This allowed for the investigation of whether the two constructs differentially predicted recidivism within each follow-up time. The literature review presents the structure and theoretical underpinnings of parole, prior studies that have investigated postrelease offending patterns of former inmates, as well as how these patterns differ between supervised and unsupervised groups.

Release Mechanisms, Former Inmates, and Recidivism

Prison releases fall into two major categories: conditional and unconditional. Conditional release involves the supervision of an inmate, typically by a parole board, whereas unconditional release does not require former inmates to be supervised. Conditional releases spend a portion of their court-imposed sentence in the community, whereas unconditional releases leave prison at the end of their sentence without any further requirements. Inmates can be unconditionally released due to parole ineligibility, denial of parole release by a parole board, assignment of a determinate sentence, or they can voluntarily opt out of the parole process (Ostermann, 2011b, 2011c). Inmates can be conditionally released as a mandatory operation of law (Travis, 2005) or they can be released because a parole board, using its discretion, extends them the opportunity to serve a part of their sentence in the community.

Parole supervision involves parolees adhering to the conditions of their parole term during the time that they are actively supervised. These conditions can include meeting with parole officers, participating in community-based programming, submitting to drug testing, obtaining and retaining gainful employment, and remaining crime free in the community. Parole officers monitor the activities of parolees in the community, perform home visits, schedule parolees to attend office visits, and refer those they supervise to social services. If a parolee is not effectively adjusting to life in the community, parole officers can initiate a process of having the parolee’s supervision term revoked. Parolees who have their supervision revoked are returned to prison where they either must serve the remainder of their sentence or await another opportunity to be released to parole.

Parole has been traditionally entrenched within the rehabilitative ideal (Morgan & Smith, 2005). In its early stages in the American criminal justice system, discretionary parole was mainly viewed as a mechanism to allow inmates the opportunity to be released from the punitive portion of their sentence if they showed promise that they were able to be rehabilitated and demonstrated a willingness to conform to society’s rules (Petersilia, 2003). Discretionary release was viewed as a means to engender reform because it encouraged inmates to participate in rehabilitative programs so they would appear as better candidates for early release to the parole board. Furthermore, the supervision portion of parole provides parolees with the opportunity to receive rehabilitative services in the community. This is geared toward greater ensuring that parolees do not recidivate after completing their court-imposed sentence (Travis, 2005).

However, in the late 1970s, discretionary parole release came under attack because it began to be viewed as too lenient, the decision-making practices did not have enough oversight, and there was little scientific evidence that parole release and subsequent supervision decreased postrelease offending (Petersilia, 2003). These views gave rise to mandatory parole mechanisms. By 2002, all but 16 states abandoned their discretionary release mechanisms. Although more than 95% of U.S. prisoners were at one time released via a discretionary mechanism, less than one in four prisoners are currently released under this mechanism (Travis, 2005). In addition, the most currently available Bureau of Justice Statistics (BJS) data demonstrate that prison release rates are steadily increasing and that these increases can be tied to a higher proportion of former inmates being unconditionally released at the end of their sentences (Sabol, West, & Cooper, 2009). To date, only a handful of studies have compared reoffense patterns of former inmates who are conditionally released with those who are unconditionally released.

Prior research addressing recidivism of newly released inmates has largely demonstrated that this group has high rates of recidivism. The BJS has conducted two nationally representative recidivism studies of reintegrating populations. The first of these studies (Beck & Shipley, 1989) analyzed rearrest, reconviction, and return to custody patterns of 108,580 people released from prisons in 11 states in 1983. The second study (Langan & Levin, 2002) analyzed rearrests, reconvictions, resentences to prison, and returns to prison without a new sentence of 272,111 people released in 15 states in 1994. Both studies used a 3-year follow-up time. Results from the study of the 1983 release cohort indicated a rearrest rate of 62.5%, a reconviction rate of 46.8%, and a return to custody rate of 41.4% (Beck & Shipley, 1989). These rates grew to 67.5%, 46.9%, and 51.8%, respectively, in the analysis of the 1994 release cohort (Langan & Levin, 2002).

The Urban Institute (UI) reanalyzed the BJS’s data from the 1994 release cohort in an effort to explore differences in recidivism rates between those released to parole supervision and those released without supervision (Solomon et al., 2005). Those released to parole were further disaggregated into those who were released via a discretionary mechanism and those who were released via statutory mandate. The researchers used a 2-year follow-up period rather than Langan and Levin’s (2002) original 3-year follow-up period in an effort to establish that those who were released to parole were under active supervision. This strategy was reflective of findings from a previous BJS report that highlighted the average time served on parole in 1990 was 23 months and in 1999, it was 26 months (Hughes, Wilson, & Beck, 2001; Solomon et al., 2005). The UI report found that those who were unconditionally released in 1994 were rearrested at a rate of 62%, those who were released to mandatory parole supervision were rearrested at a rate of 61%, and those who were released to discretionary parole were rearrested at a rate of 54%.

After using multivariate regression modeling to control for prerelease differences between the groups, the differences in rearrest rates were much less pronounced. Upon controlling for pertinent predictors of recidivism, the rearrest rate for unconditional and mandatory parole releases were identical at 61%, whereas the rearrest rate for discretionary parolees was 57%. Using a similar strategy through using Cox proportional hazard modeling while controlling for potential interruptions in time at risk due to experiencing technical parole violations, Ostermann (2011a) also found a 4% difference in rearrest rates between supervised and unsupervised groups. In this study, Ostermann investigated 3 years of follow-up data for a cohort of New Jersey offenders released from prisons in 2006.

The work of Sacks and Logan (1979, 1980) provides perhaps the most complete analysis of the impact that active parole supervision has upon recidivism rates. These researchers compared reconviction rates over the course of 3 years for two groups of low-level felony offenders released from Connecticut prisons in 1974. The first group consisted of 115 males who were released from prison without parole supervision while the second group consisted of 57 males who were released with supervision. Although Sacks and Logan’s (1979) first study examined results after 1 year of follow-up time and found that “parole had definite, but modest, effects on recidivism among the subjects” (p. 12), they ultimately concluded that the follow-up time was too short because parolees within their study group typically served periods of supervision longer than a year.

In their follow-up report (1980), the researchers explored an additional 2 years of data and presented recidivism rates as they related to the time that the parole group spent on supervision. Within this follow-up study, Sacks and Logan (1980) ultimately concluded that parole served as a temporary insulator for keeping released inmates from experiencing a new conviction during the period of supervision, and for a short time afterwards, but

an eight-month period of parole had no preventative effects after 2 (or 3) years following release. Parole seems to affect recidivism while the parolee is on parole . . . but these effects soon begin to dissipate and tend to disappear by the time parolees have finished 2 full years in the community. (p. 15)

Prior studies have defined parolees as those who are released to the supervision of a parole board, but have largely not considered the time that parolees spend being actively supervised. Solomon et al. (2005) attempted to address this through using a national average time on parole. However, this approach does not capture the individual-level differences in the parole population regarding the time that they are under supervision. Furthermore, although Sacks and Logan’s (1980) study specifically attempted to address this impact, sample sizes were small and only represented low-level offenders. In addition, both of these previous studies use data that are currently more than a decade old. This study uses contemporary data from the entire population of released inmates within a state from 2005 to 2007 representing more than 19,000 parolees with unique periods of supervision.

Data

Data for this study were collected from a state parole board’s (SPB) parole board information system (PBIS). The SPB is located on the East Coast and makes discretionary parole release decisions and subsequently supervises all inmates who are affirmed parole release. The PBIS allowed for a database to be constructed that reflected all individuals who were released from prison from 2005 to 2007 (N = 29,299) with attendant demographic, instant offense, and release type information attached to cases. The PBIS was used to abstract information from an existing data exchange between the SPB and the state police’s criminal reporting system. This latter system archives all formally recognized criminal justice involvement that is associated with a case from the time the individual is first arrested to the date on which the data are queried. An inmate was considered a recidivist if he or she experienced either a new arrest or a parole infraction that led to his or her supervision term being revoked after their release date.

I used this combined measure of recidivism to facilitate comparisons between the conditionally and unconditionally released groups. Unconditionally released inmates cannot experience parole infractions, and parolees that do experience these types of infractions are often returned to prison. If I only analyzed rearrest data, parolees who were returned to prison for violating the terms of their supervision would appear successful despite being incarcerated, and thus not at risk of experiencing a new arrest in the community. Other studies have controlled for these types of interruptions in time at risk through incorporating the time spent not at risk into their dependent variables within survival regression models (Bales, Bedard, Quinn, Ensley, & Holley, 2005; Ostermann, 2011a); however, we did not have access to data that highlighted the dates on which parolees were released from an incarcerated setting after experiencing these infractions. Not having access to these release dates precluded us from accurately calculating these types of interruptions. For parolees, recidivism dates were compared with the parole discharge date to explore whether the recidivism event occurred when they were being supervised by the SPB or whether it occurred after they were discharged.

Data were gathered for the parole group to reflect the date on which they were released to supervision as well as their discharge date. The discharge date represents the date on which a parolee’s court-imposed sentence expires. The date of discharge was compared with the date of release to construct a measure of the time that a parolee was assigned to active supervision. On average, parolees who were released from prison from 2005 to 2007 were supervised for 505 (SD = 416) days or approximately 1.4 years. The median time on supervision for parolees was 446 days or approximately 1.2 years. Supervision terms varied widely. The shortest supervision time was 1 day and the longest supervision time was about 20 years (7,276 days). Although approximately 12% (n = 2,266) of parolees were assigned supervision terms of at least 3 years, about 21% (n = 4,066) were supervised for less than 6 months. Approximately 79% (n = 15,308) of parolees were assigned terms of supervision of at least 6 months, about 60% (n = 11,563) were to be supervised for at least a year, about a third (n = 6,523) were to be supervised for at least 1.5 years, about 20% (n = 3,912) were assigned supervision periods of at least 2 years, and approximately 14% (n = 2,694) of the parolees were supposed to be supervised by the parole board for at least 2.5 years.

Analytic Plan

Logistic regression analyses were constructed to isolate the impact of parole upon recidivism across several follow-up times. Logistic regression was used rather than survival modeling so that results could be discussed within the context of absolute quantities (e.g., predicted probabilities) rather than relative quantities (e.g., hazard rates) within a given follow-up period (Fox, 2002). Within these models, I controlled for established predictors of re-offending. I predicted recidivism within 6 months, 1 year, 1.5 years, 2 years, 2.5 years, and 3 years of release. Two models were constructed within each follow-up period, each using a different parole construct while using identical control variables. The first construct identifies parolees as inmates released to parole supervision and does not consider whether the individual is actively supervised. Under this construct, recidivism is attached to parole cases even if they are not actively being supervised by the parole board at the time that they recidivate. The second construct identifies parolees as inmates released to parole and who are actively supervised for at least the follow-up time under investigation. Under this construct, recidivism is attached to parole cases only if they are actively being supervised at the time that they recidivate.

For example, using the first definition for a 2-year recidivism analysis, anyone who was released to parole would have 2 years of postrelease follow-up data analyzed. Parolees would be considered recidivists if they were released to parole supervision and they were either rearrested or had their supervision term revoked within 2 years of their release date. Using the second definition for a 2-year recidivism analysis, anyone who was released to parole and who had a parole term equal to or greater than 2 years would have 2 years of postrelease follow-up data analyzed. Parolees would be considered recidivists if their parole term was greater than or equal to 2 years and they were either rearrested or had their parole revoked within that 2-year period. Using this definition, parolees who were rearrested within their 2nd year after release, but who were only supervised for 6 months, would not be considered within the analysis. However, if they were supposed to be supervised for 2 years and were rearrested within 6 months, they would be considered parolee recidivists within the 2-year analysis.

In addition to the parole variable, control variables that were entered into the regression models include age, gender, minority status, marital status, a deprivation index of the county of return, a prerelease Level of Service Inventory–Revised (LSI-R) score, the number of instant offenses for which the individual was incarcerated, the type of instant offense for which the individual was incarcerated, and the number of arrest events the individuals had on their official record prior to their release. Minority status indicated that the individual was non-White. Marital status indicated that the individual was single. The deprivation index variable was adapted from Solomon et al.’s (2005) reanalysis of the BJS’s 1994 recidivism study. The index represents a scalar measure of information gathered from the U.S. Census about the county in which the individual was convicted (serving as a proxy measure of the county of return). Instant offense types indicate the most serious offense for which the individual was incarcerated prior to their release and include property, drug, sex, violent, and other types of crimes.

Two postestimation analyses were conducted. First, average marginal effects were calculated upon the parole variable within each follow-up period using each of the two definitions of parole. This analysis communicates the predicted probability of parolees recidivating within a given follow-up time relative to unconditionally released inmates after controlling for the other covariates within the model. Within these analyses, all cases are assigned the referent category (unconditional release) and then the category of interest (parole) and the predicted probabilities of experiencing the dependent variable for each assignment are presented. Second, the regression models using the two different definitions of parole within each follow-up time were simultaneously estimated. Within these simultaneous estimations, the parole variable was compared between the two models using a seemingly unrelated estimations protocol. This allowed for the analysis of whether the parole variables significantly differed from one another across models predicting recidivism over the same follow-up times in the presence of identical control variables.

Results

Descriptive Statistics

Descriptive statistics for unconditionally released inmates, all inmates released to parole, and parolees according to their time of active supervision are presented in Table 1. Parolees and unconditional releases were, on average, in their mid-30s at the time of their release from prison, about 90% were male, approximately 60% were Black, about 20% were White, and approximately 80% were single. Unconditional releases had, on average, 3 more points on a prerelease LSI-R assessment and had about two more prior arrests when compared with parolees. Unconditionally released inmates as well as parolees were serving prison sentences for about two instant offenses. A slightly lower proportion of parolees were serving time for property offenses and a slightly higher proportion were serving time for drug-related offenses when compared with unconditionally released inmates. Those released to parole experienced lower rates of postrelease recidivism when compared with unconditionally released inmates. About 19% of parolees were either rearrested or had their parole term revoked within 6 months while about 26% of unconditional releases were rearrested within 6 months of being released. Within 3 years, about 58% of parolees were either rearrested or had their parole revoked, whereas about 63% of those who were unconditionally released from prison were rearrested.

Table 1.

Descriptive Statistics: Unconditionally and Conditionally Released Inmates, and Parolees With Various Active Supervision Terms.

	Unconditional releases	Released to parole	Parole ≥ 6 months	Parole ≥ 1 year	Parole ≥ 1.5 years	Parole ≥ 2 years	Parole ≥ 2.5 years	Parole ≥ 3 years
Age	36.29	34.94	34.93	35.03	35.54	35.82	35.81	35.78
% male	92.45	91.69	91.32	90.56	91.45	93.12	93.43	93.25
% race
White	19.11	19.07	19.73	21.22	22.46	22.42	23.53	24.36
Black	57.42	59.33	58.16	56.33	54.27	52.91	52.23	51.81
Hispanic	14.75	18.74	19.51	20.12	20.88	22.34	22.01	21.45
Other	0.37	0.79	0.86	0.99	1.29	1.61	1.71	1.85
% single	83.46	81.97	81.54	80.83	79.55	78.8	78.74	78.7
LSI-R	25.07	22.98	22.67	22.37	22.32	22.02	22	22.02
Instant offenses	2.18	2.09	2.07	2.04	2.15	2.22	2.24	2.2
% instant offense type
Property	19.27	14.76	14.03	13.56	12.34	9.24	6.85	5.72
Drug	43.34	48.44	48.36	46.98	38.43	29.01	20.18	16.2
Sex	1.04	1.56	1.67	1.8	2.34	3.23	4.08	4.57
Violent	32.89	32.57	33.43	34.96	44.34	56.33	67.57	72.64
Other	3.45	2.68	2.52	2.71	2.56	2.2	1.32	0.87
Prior arrests	9.8	7.77	7.34	6.97	6.66	6.13	5.64	5.4
% recidivated
6 months	26.39	18.91	17.78	16.11	15.21	14.03	13.77	13.64
1 year	41.63	34.04	—	30.91	28.41	26	25.24	24.58
1.5 years	50.67	43.81	—	—	37.54	34.1	33.44	32.44
2 years	56.28	50.21	—	—	—	40.41	39.12	38.31
2.5 years	60.25	54.81	—	—	—	—	43.47	42.41
3 years	63.26	58.27	—	—	—	—	—	46.56
n	9,925	19,374	15,308	11,563	6,523	3,912	2,694	2,266

Note: LSI-R = Level of Service Inventory–Revised.

Several additional differences emerge when the parole group is disaggregated according to the time that they were to be supervised in the community. Those who were to be supervised for at least 2 or 3 years had about 2 fewer arrests on their official records at the time of their release when compared with parolees who were to be supervised for at least 6 months to a year. As the time that parolees were to be supervised in the community increased, so did the proportion that were serving instant offenses that were violent in nature. Although approximately 33% of parolees who were supposed to serve at least 6 months on parole were incarcerated for violent offenses, about 73% of those who were supposed to serve 3 years or more of parole supervision were incarcerated for violent offenses. Furthermore, almost half of the parolees who were supposed to be supervised for at least 6 months were incarcerated for drug-related offenses, whereas approximately 16% of parolees who were to be supervised for at least 3 years were incarcerated for drug crimes.

Rates of recidivism substantially differed within the follow-up periods when comparing the patterns of parolees, regardless of their term of supervision, to parolees who were supervised for at least the follow-up period under consideration. These differences become larger when longer follow-up periods are considered. For example, although approximately 19% of all inmates who were released to parole recidivated within 6 months, 18% of parolees who were actively supervised for at least 6 months recidivated. The proportion of all inmates released to parole who recidivated in a year differed by 3% when compared with those released to parole who were actively supervised for at least a year (34% vs. 31%). Approximately half of all inmates who were released to parole supervision recidivated within 2 years of leaving prison, but about 40% of those parolees who were actively supervised for at least 2 years recidivated. After 3 years of follow-up, about 58% of parolees recidivated, whereas approximately 47% of those who were released to parole and who were supposed to be actively supervised for at least 3 years recidivated.

The differences in recidivism rates are more pronounced within each follow-up time when comparing unconditional releases and parolees who have active supervision terms that encompass a given follow-up time. Although 26% of unconditional releases were rearrested within 6 months of release, approximately 18% of parolees who were supposed to be supervised for at least 6 months were either rearrested or had their parole revoked. Within 1 year of release, 42% of unconditional releases were rearrested, whereas 31% of parolees who had supervision terms of at least 1-year recidivated. Within 3 years, this figure grew to approximately 63% for unsupervised former inmates and approximately 47% for parolees who were assigned active supervision periods that encompassed at least 3 years.

Multivariate Logistic Regressions

Table 2 presents OR from the 12 logistic regression models that were constructed to isolate the impact of parole supervision upon recidivism. In Models 1 through 6, I operationalize the parole variable as inmates who are released to parole, but I do not consider whether the parole board actively supervised the inmate within the follow-up period. In Models 7 through 12, I operationalize the parole variable as inmates who are released to parole and who are supervised for at least the follow-up period under exploration. All 12 of the logistic regression models provided statistically significant good fits to the data.

Table 2.

Odds Ratios from Logistic Regression Models Predicting Recidivism at Various Follow-Up Times.

Follow-up/active supervision time	6 months	1 year	1.5 years	2 years	2.5 years	3 years
Models 1-6
Age	0.953***	0.955***	0.954***	0.953***	0.953***	0.952***
Male	1.561***	1.427***	1.421***	1.371***	1.379***	1.328***
Minority	1.119*	1.172***	1.149*	1.151***	1.140***	1.103*
Deprivation	1.101***	1.093***	1.078***	1.089***	1.092***	1.093***
Single	1.089	1.085	1.111*	1.120**	1.135**	1.111*
LSI-R	1.053***	1.055***	1.056***	1.053***	1.053***	1.053***
Prior arrests	1.061***	1.069***	1.077***	1.086***	1.091***	1.095***
Number of offenses	1.023	1.049**	1.053***	1.064***	1.063***	1.069***
Offense type
Property	1.203	1.154	1.160	1.154	1.138	1.077
Drug	0.997	0.991	1.007	1.046	1.055	1.033
Sex	0.653	0.655*	0.582**	0.595**	0.599**	0.576**
Violent	1.060	0.969	0.962	0.968	0.968	0.929
Other (referent)	—	—	—	—	—	—
Parole	0.739***	0.836***	0.919*	0.949	0.975	0.971
Model statistics
χ² (df)	1,603.58 (13)	2,231.16 (13)	2,501.66 (13)	2,663.95 (13)	2,724.45 (13)	2,758.41 (13)
Log likelihood	−10,487.74	−13,082.39	−13,580.07	−13,473.50	−13,219.52	−12,906.75
Pseudo R²	0.0710	0.0786	0.0843	0.0900	0.0934	0.0965
Significance	.000	.000	.000	.000	.000	.000
Models 7-12
Age	0.953***	0.957***	0.956***	0.952***	0.951***	0.949***
Male	1.640***	1.481***	1.386***	1.444***	1.563***	1.540***
Minority	1.158**	1.156**	1.225***	1.243***	1.230**	1.246**
Deprivation	1.099***	1.121***	1.101***	1.113***	1.116***	1.092**
Single	1.064	1.047	1.066	1.081	1.039	1.019
LSI-R	1.055***	1.062***	1.059***	1.051***	1.042***	1.037***
Prior arrests	1.060***	1.066***	1.074***	1.079***	1.077***	1.079***
Number of offenses	1.020	1.066***	1.064**	1.095***	1.107***	1.106***
Offense type
Property	1.189	1.084	1.067	1.003	1.041	1.009
Drug	0.976	0.964	0.990	0.972	1.037	0.988
Sex	0.629	0.749	0.816	0.743	0.656	0.488*
Violent	1.011	0.909	0.901	0.912	0.964	0.902
Other (referent)	—	—	—	—	—	—
Parole	0.716***	0.813***	0.829***	0.733***	0.677***	0.692***
Model statistics
χ² (df)	1,415.36 (13)	1,688.18 (13)	1,331.33 (13)	1,105.09 (13)	871.52 (13)	763.19 (13)
Log likelihood	−8.653.92	−8,784.97	−6,269.18	−4,765.51	−4,061.58	−3,750.75
Pseudo R²	0.0756	0.0877	0.0960	0.1039	0.0969	0.0923
Significance	.000	.000	.000	.000	.000	.000

Note: LSI = Level of Service Inventory–Revised. Contrasts in the parole variable for Models 1 to 6: Inmates identified as parolees at time of release regardless of active supervision period = 1 and unconditionally released inmates = 0. Contrasts in the parole variable for Models 7 to 12: Inmates identified as parolees at time of release who were supervised for at least the follow-up period = 1 and unconditionally released inmates = 0.

p ≤ .05. **p ≤ .01. ***p ≤ .001.

Many of the control variables were statistically significant predictors of recidivism. Age was negatively related to recidivism in all of the models indicating that for every additional year old offenders were on their release date, the likelihood of recidivating in the community decreased. OR across the models indicate that an additional year of age at the time of release is associated with about a 5% decreased likelihood of recidivating. Minority offenders were significantly more likely to recidivate when compared with nonminorities. In the 6-month follow-up analyses, minorities were approximately 56% (R = 1.561, p = .000) more likely to recidivate than nonminorities in the model that used the traditional parole construct. In the model that used the parole construct that incorporates the time of active supervision to predict recidivism over 6 months, minorities were 64% (OR = 1.640, p = .000) more likely to recidivate than nonminorities. Like minority status, being male, the deprivation index of an offender’s county of conviction, the prerelease LSI-R score, prior arrests, and the number of instant offenses for which an individual was incarcerated were also significant predictors of recidivism in positive directions.

When defining parole under the traditional definition used in previous studies, the variable was a statistically significant predictor that an inmate would be less likely to recidivate when compared with those who were not released to parole supervision. However, under this definition, parole was only a significant predictor within half of the models. When predicting recidivism that occurs within 6 months of release, being released to parole supervision was associated with an approximate 26% (OR = 0.739, p = .000) decreased likelihood of experiencing failure in the community when compared with inmates who were unconditionally released. Within 1 year, parolees were approximately 16% (OR = 0.836, p = .000) less likely to recidivate compared with unconditionally released inmates and within 1.5 years, parolees were about 8% (OR = 0.919, p = .012) less likely to recidivate. After 1.5 years, being released to parole was not a significant predictor that an inmate would be either more or less likely to recidivate when controlling for the other variables contained within the regression models.

Parole was a statistically significant predictor of recidivism in Models 7 through 12. Within 6 months of release, those who were released to parole who were assigned to be actively supervised for at least 6 months were approximately 28% (OR = 0.716, p = .000) less likely to recidivate when compared with inmates who were released unconditionally at the culmination of their sentence. Parolees who were to be supervised for at least 1 year were about 19% (OR = 0.813, p = .000) less likely to recidivate within 1 year and those who were to be supervised for at least 1.5 years were about 17% (OR = 0.829, p = .000) less likely to recidivate within 1.5 years when compared with those who were unconditionally released. Within 2 years, parolees who were released to at least a 2 year supervision period were about 27% (OR = 0.733, p = .000) less likely to recidivate when compared with inmates who were released without any supervision. Effect sizes became larger as longer follow-up times were considered. Within 2.5 years, supervised parolees were approximately 32% (OR = 0.677, p = .000) less likely to recidivate and within 3 years supervised parolees were about 31% (OR = 0.692, p = .000) less likely to recidivate when compared with unconditionally released inmates after controlling for the other predictor variables contained within the models.

Average Marginal Effects

Table 3 presents results from average marginal effects that were derived from the 12 logistic regression models. The first row of Table 3 presents results using the parole definition used in Models 1 through 6, whereas the second row of Table 3 presents results using the parole definition used in Models 7 through 12. According to the traditional definition of parole used in previous studies, within 6 months of release, if all of the cases were treated as if they were unconditionally released inmates, the average predicted probability that they would recidivate would be about 26% (Average marginal effects; AME = .2559, 95% confidence intervals [CI] = [.2449, .2668]) compared with about 21% if all of the cases were treated as if they were parolees (AME = .2059, 95% CI = [.1998, .2121]). This represents an approximate 5% difference in the predicted probability of experiencing recidivism in the community within 6 months of release. This difference is similar when considering the second parole definition that incorporates the time of active supervision.

Table 3.

Average Marginal Effects of Group Membership on Recidivism at Various Follow-Up Times: Unconditionally Released Inmates, All Parolees, and Actively Supervised Parolees.

	6 months	1 year	1.5 years	2 years	2.5 years	3 years
Unconditional releases	.2559 [.2449, .2668]	.4042 [.3919, .4166]	.4860 [.4733, .4987]	.5457 [.5330, .5584]	.5851 [.5725, .5978]	.6203 [.6077, .6328]
Released to parole	.2059 [.1998, .2121]	.3640 [.3588, .3731]	.4672 [.4599, .4746]	.5342 [.5269, .5415]	.5797 [.5725, .5868]	.6141 [.6071, .6212]
Unconditional releases	.2527 [.2418, .2635]	.3961 [.3838, .4085]	.4897 [.4770, .5023]	.5625 [.5500, .5750]	.6133 [.6009, .6257]	.6520 [.6399, .6641]
Actively supervised parolees	.1986 [.1917, .2056]	.3526 [.3430, .3621]	.4483 [.4343, .4623]	.4945 [.4742, .5148]	.5288 [.5010, .5565]	.5744 [.5423, .6066]

Note: 95% confidence intervals are presented in brackets. Row 1 presents average marginal effects of the parole variable in Models 1 to 6 in Table 2. Row 2 presents average marginal effects of the parole variable in Models 7 to 12 in Table 2.

Using this definition, within 6 months, about 25% (AME = .2527, 95% CI = [.2418, .2635]) of unconditionally released inmates are predicted to recidivate, whereas approximately 20% (AME = .1986, 95% CI = [.1917, .2056]) of parolees are predicted to recidivate. Similar patterns were estimated within 1 year of follow-up. Within 1 year, 40% (AME = .4042, 95% CI = [.3919, .4166]) of unconditionally released inmates and 36% of parolees (AME = .3640, 95% CI = [.3588, .3731]) were predicted to recidivate according to the traditional definition of parole. According to the parole definition that incorporates the supervision term, about 36% of parolees (AME = .3526, 95% CI = [.3430, .3621]) and about 40% (AME = .3961, 95% CI = [.3838, .4085]) of unconditionally released inmates were predicted to fail within a year of release.

Within 2 years of release, the predicted likelihood of recidivism for inmates released to parole supervision compared with unconditionally released inmates differed by approximately 1% if the time that parolees were actively supervised is not considered (parolees’ AME = .5342, 95% CI = [.5269, .5415] compared with unconditionally released inmates’ AME = .5457, 95% CI = [.5330, .5584]). When only considering parolees who were assigned to be actively supervised for 2 years, however, the predicted difference in recidivism between supervised and supervised former inmates is about 6% (parolees’ AME = .4945, 95% CI = [.4742, .5148] compared with unconditionally released inmates’ AME = .5625, 95% CI = [.5500, .5750]) within 2 years of their release from prison.

The predicted between-group difference in recidivism remains at 1% under the traditional parole definition within 2.5- and 3-year follow-up times. However, when only considering parolees who were assigned active supervision terms of at least 2.5 and 3 years, the predicted difference in recidivism is about 8%. Approximately 53% (AME = .5288, 95% CI = [.5010, .5565]) of parolees who were to be supervised for at least 2.5 years were predicted to recidivate within 2.5 years and approximately 57% (AME = .5744, 95% CI = [.5423, .6066]) of parolees who were to be supervised for at least 3 years were predicted to recidivate within 3 years. Within 2.5 years about 61% (AME = .6133, 95% CI = [.6009, .6257]) of unconditionally released inmates were predicted to recidivate and within 3 years about 65% (AME = .6520, 95% CI = [.6399, .6641]) of this group were predicted to recidivate.

Seemingly Unrelated Estimations

Results from the post hoc seemingly unrelated estimation protocol indicated that the parole variable differentially predicted recidivism within an identical follow-up period across the two definitions of parole that I used. When simultaneously estimating the models that used the two different parole definitions to predict recidivism that occurred within 6 months of release (Models 1 and 7), the predictive value of the parole variable significantly differed between the models (χ² = 5.40, p = .020). These differences were apparent during 1 year when comparing estimations from Models 2 and 8 (χ² = 2.64, p = .014), 1.5 years when comparing estimations from Models 3 and 9 (χ² = 12.87, p = .000), 2 years when comparing estimations from Models 4 and 10 (χ² = 35.07, p = .000), 2.5 years when comparing estimations from Models 5 and 11 (χ² = 37.19, p = .000), and 3 years when comparing estimations from Models 6 and 12 (χ² = 23.39, p = .000).

Discussion

This study explored whether recidivism rates are impacted according to the way we consider the construct of parole. Two different parole constructs were presented: (a) a definition that is traditionally used within the recidivism literature that defines parolees simply as those who are released to parole supervision and (b) a definition that identifies parolees as those who are released to parole and who are assigned to be actively supervised during the time in which their reoffense patterns are being explored. The study was conducted because the extant recidivism literature has largely failed to encapsulate the construct of parole according to the reality in which it functions. Results indicated that the way that parole is defined has significant impacts upon recidivism rates. The seemingly unrelated estimations demonstrated that the two constructs differentially predicted recidivism across all of the follow-up periods that I explored. The logistic regression and average marginal effect analyses demonstrated that the importance of the parole definition manifests after approximately 1 year of follow-up time.

The predicted likelihood of recidivating differed by about 5% between supervised and unsupervised groups in the 6-month and 1-year follow-up periods whether active supervision was included in the parole construct or not. After 1 year, these differences become very small if active supervision is not included in the parole construct. If only parolees who are actively being supervised for 1 year or more are included, the difference in the predicted recidivism rates is substantial. For example, the regression model that was constructed to predict failure over 3 years and incorporated the traditional definition of parole demonstrated a 1% lower predicted recidivism rate for parolees when compared with inmates who were released without supervision. This finding was similar to previous research that has used this definition of parole and defined recidivism as a rearrest (Ostermann, 2011a; Solomon et al., 2005). However, the model that only included parolees who were actively supervised for at least 3 years estimated that parolees would recidivate at an 8% lower rate. This difference in reoffense patterns is approximately twice as large as those found in previous studies that did not incorporate the time of active supervision into their parole constructs (Ostermann, 2011a; Solomon et al., 2005).

These findings are compelling considering that the average parole supervision term was approximately 1.4 years and the median supervision term was 1.2 years. Many of those who were released to parole and were followed beyond 1 year would not be under parole’s supervision by that time. Indeed, only one third of the parole population had active supervision periods that met or exceeded 1.5 years. This is indicative that supervision can insulate offenders from recidivism, but after supervision has expired, parole does not have substantial long-lasting effects. If parole had long-lasting effects, it would be expected that the predicted rate of recidivism of those released to parole’s supervision, without taking into account the timing of active supervision, would substantially differ from those who are unconditionally released at the culmination of their sentence within follow-up times that go beyond the average length of a supervision term.

This lack of a long-term impact is likely due to the relationship between parole’s focus upon keeping those under their supervision crime free and the resultant management of rehabilitative resources and community programming during the course of supervision (Marciniak, 1999; Ostermann, 2009; Taxman, Soule, & Gelb, 1999; Tonry, 1997). Community programs offered through parole can and should be used by supervising authorities to build skills in parolees to better ensure prosocial engagement, most notably for those who demonstrate as at high risk (Andrews, Bonta, & Wormith, 2006; Lowenkamp & Latessa, 2005a, 2005b; Lowenkamp, Latessa, & Holsinger, 2006; Lowenkamp, Lovins, & Latessa, 2009; Lowenkamp, Pealer, Smith, & Latessa, 2006). However, these programs can also be used primarily as diversionary tools that provide pseudoincapacitation in the community. If parole boards are primarily concerned with keeping parolees law-abiding during the course of supervision in the immediate short term, pursuing this goal may encourage a “divert to success” mentality within these agencies.

Under this model, community programs and other parole resources can be used to mitigate time at risk in the community in an effort to keep parolees incapacitated, but still successfully engaged in parole, until they are discharged from supervision. Research has demonstrated that high impact community-based programming should be geared toward high-risk individuals in line with their criminogenic needs (Andrews, 1982; Dowden & Andrews, 1999a; Lowenkamp & Latessa, 2005a, 2005b; Lowenkamp & Latessa et al., 2006; Lowenkamp & Pealer et al., 2006). Nevertheless, successfully adhering to this model assumes that the primary goal of parole boards is to have long-term rehabilitative impacts upon those they supervise rather than simply keeping them out of trouble until they are discharged. In short, if the expectation of parole boards is to ensure that parolees do not recidivate for the relatively short period of time that they are under active supervision, and nothing further, an agency’s pursuit of this sort of mission may undermine larger rehabilitative goals that have demonstrated to have, at times, dramatic impacts upon criminal recidivism (Dowden & Andrews, 1999b).

Pursuing this short-term goal would likely encourage a reactive case management style, whereas pursuing long-term rehabilitative goals would likely encourage a proactive case management style. Within the first of these managerial styles, officers would likely be encouraged to place parolees who are starting to demonstrate problem behaviors into programs so that they can be taken out of the community and, as a result, out of risk of potentially failing. This would likely occur irrespective of the parolee’s assessed risks and needs. For example, under this model, if an inpatient program offered 90 days of pseudoincapacitation in the community, it could guarantee 90 days of a parolee not being at risk of being involved in new criminal activity. This could encourage sending parolees to programs unnecessarily. This strategy would not only squander resources, it may also result in iatrogenic effects if low-risk parolees receive programs unnecessarily (Lowenkamp & Latessa, 2005). If a parole board were held responsible for the long-term impacts of their efforts upon criminal recidivism, officers would likely be encouraged to identify risks and needs that their caseloads demonstrate, and attach appropriate individuals to appropriate interventions proactively. These efforts would likely take place shortly after release.

It is imperative that parole boards explicitly state how they define success for those they supervise, and these definitions must be realistic representations of the public’s expectations about how these agencies should perform. Although parole boards do not typically concern themselves with the performance of parolees after they are discharged, it is likely that the public expects supervision agencies to work toward effectuating long-lasting positive changes in the lives of those they supervise. The public likely does not expect that parolees should only abstain from criminal behavior during the course of their supervision. Rather, they expect parole boards to keep them safe and to work toward successfully rehabilitating individuals. Based on the underlying theoretical tenets on which parole was originally implemented, these rehabilitative efforts are expected to be transformative for former offenders and should encourage them to not engage in criminal activities after their term of supervision is completed.

These expectations are not unreasonable, but allowing parole boards to only concern themselves with recidivism that occurs solely during the course of supervision can likely have deleterious impacts on realizing these types of long-lasting results. While previous studies have explored a parole construct that is essentially an ad infinitum status, parole boards likewise do not have the capacity to supervise parolees ad infinitum. Parole boards must strive to have impacts upon parolees beyond the term of supervision to ensure public safety over the long term. To that end, further investigation into the mechanisms that are used to produce positive outcomes for supervised parolees is needed. These types of research endeavors will shed additional light onto whether parole agencies primarily focus their resources toward realizing long-term rehabilitative goals by programming parolees according to established evidence-based principles that can have long-lasting impacts or toward realizing short-term diversions geared at successfully meeting dates of discharge.

Footnotes

Declaration of Conflicting Interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

Author Biography

Michael Ostermann is a research professor at the Rutgers University School of Criminal Justice where he serves as the director of the Evidence Based Institute. The institute’s function is to address the research and policy analysis needs of the New Jersey criminal justice system and to assess and improve the services that are directed toward the state’s criminal justice-involved populations. His research interests primarily lie within the fields of corrections and reentry and how they intersect with public policy.

References

Andrews

(1982). The Level of Supervision Inventory (LSI): The first follow-up. Ottowa, Canada: Carleton University.

Andrews

Bonta

Wormith

J. S.

(2006). The recent past and near future of risk and/or need assessment. Crime & Delinquency, 52, 7-27.

Bales

W. D.

Bedard

L. E.

Quinn

S. T.

Ensley

D. T.

Holley

G. P.

(2005). Recidivism of public and private prison inmates in Florida. Criminology & Public Policy, 4, 57-82.

Beck

Shipley

(1989). Recidivism of prisoners released in 1983 (Special Report No. NCJ 116261). Washington, DC: Bureau of Justice Statistics.

Dowden

Andrews

D. A.

(1999a). What works in young offender treatment: A meta-analysis. Forum on Corrections Research, 11, 21-24.

Dowden

Andrews

D. A.

(1999b). What works for female offenders: A meta-analytic review. Crime & Delinquency, 45, 438-452.

Fox

(2002). Cox proportional-hazards regression for survival data: Appendix to an R and S-PLUS companion to applied regression. Retrieved from http://cran.r-project.org/doc/contrib/Fox-Companion/appendix-cox-regression.pdf

Gottfredson

M. S.

Mitchell-Herzfeld

Flanagan

(1982). Another look at the effectiveness of parole supervision. Journal of Research in Crime & Delinquency, 19, 277-298.

Hughes

T. A.

Wilson

D. J.

Beck

A. J.

(2001). Trends in state parole, 1990-2000 (Special Report No. NCJ 184735). Washington, DC: Bureau of Justice Statistics.

10.

Jackson

P. C.

(1983). Some effects of parole supervision on recidivism. British Journal of Criminology, 23, 17-34.

11.

Langan

P. A.

Levin

D. J.

(2002). Recidivism of prisoners released in 1994 (Special Report No. NCJ 193427). Washington, DC: Bureau of Justice Statistics.

12.

Lowenkamp

C. T.

Latessa

E. J.

(2005a). Increasing the effectiveness of correctional programming through the risk principle: Identifying offenders for residential placement. Criminology & Public Policy, 4, 263-290.

13.

Lowenkamp

C. T.

Latessa

E. J.

(2005b, April). Developing successful reentry programs: Lessons learned from the “What Works” research. Corrections Today, 72-77.

14.

Lowenkamp

C. T.

Latessa

E. J.

Holsinger

A. M.

(2006). The risk principle in action: What have we learned from 13,676 offenders and 97 correctional programs? Crime & Delinquency, 52, 1-17.

15.

Lowenkamp

C. T.

Lovins

Latessa

E. J.

(2009). Validating the Level of Service Inventory–Revised and the Level of Service Inventory: Screening version with a sample of probationers. The Prison Journal, 89, 192-204. doi:10.1177/ 0032885509334755

16.

Lowenkamp

C. T.

Pealer

Smith

Latessa

E. J.

(2006). Adhering to the risk and need principles: Does it matter for supervision-based programs? Federal Probation, 70(3), 3-8.

17.

Marciniak

L. M.

(1999). The use of day reporting as an intermediate sanction: A study of offender targeting and program termination. The Prison Journal, 79, 205-225.

18.

Morgan

K. D.

Smith

(2005). Parole release decisions revisited: An analysis of parole release decisions for violent inmates in a southeastern state. Journal of Criminal Justice, 33, 277-287.

19.

Ostermann

(2009). An analysis of New Jersey’s day reporting center and halfway back programs: Embracing the rehabilitative ideal through evidence based practices. Journal of Offender Rehabilitation, 48, 139-153.

20.

Ostermann

(2011a). How do former inmates perform in the community? A survival analysis of rearrests, reconvictions, and technical parole violations. Crime & Delinquency. Advance online publication. doi:10.1177/0011128710396425

21.

Ostermann

(2011b). Recidivism and the propensity to forgo parole release. Justice Quarterly. Advance online publication. doi:10.1080/07418825.2011.570362

22.

Ostermann

(2011c). Parole? nope, not for me: Voluntarily maxing out of prison. Crime & Delinquency, 57, 686-708.

23.

Petersilia

(2003). When prisoners come home: Parole and prisoner reentry. New York, NY: Oxford University Press.

24.

Sabol

W. J.

West

H. C.

Cooper

(2009). Prisoners in 2008 (Special Report No. NCJ 228417). Washington, DC: Bureau of Justice Statistics.

25.

Sacks

H. R.

Logan

C. H.

(1979). Does parole make a difference? Hartford, CT: University of Connecticut School of Law Press.

26.

Sacks

H. R.

Logan

C. H.

(1980). Parole: Crime prevention or crime postponement? Hartford, CT: University of Connecticut School of Law Press.

27.

Schlager

Robbins

(2008). Does parole work?—Revisited: Reframing the discussion of the impact of postprison supervision on offender outcome. The Prison Journal, 88, 234-251.

28.

Shadish

W. R.

Cook

T. D.

Campbell

D. T.

(2002). Experimental and quasi-experimental designs for generalized causal inference. Boston, MA: Houghton Mifflin Company.

29.

Solomon

A. L.

Kachnowski

Bhati

(2005). Does parole work? Analyzing the impact of post-prison supervision on rearrest outcomes. Washington, DC: Urban Institute Justice Policy Center.

30.

Spivak

A. L.

Damphousse

K. R.

(2006). Who returns to prison?: A survival analysis of recidivism among adult offenders released in Oklahoma, 1985-2004. Justice Research and Policy, 8(2), 57-88.

31.

Spivak

A. L.

Sharp

S. F.

(2008). Inmate recidivism as a measure of private prison performance. Crime & Delinquency, 54, 482-508.

32.

Taxman

F. S.

Soule

Gelb

(1999). Graduated sanctions: Stepping into accountable systems and offenders. The Prison Journal, 79, 182-204.

33.

Tonry

(1997). Intermediate sanctions in sentencing guidelines. Washington, DC: U.S. Department of Justice.

34.

Travis

(2005). But they all come back: Facing the challenges of prisoner reentry. Washington, DC: Urban Institute Press.

35.

West

H. C.

Sabol

W. J.

Greenman

S. J.

(2010). Prisoners in 2009 (Bureau of Justice Statistics Bulletin, NCJ 231675). Washington, DC: Bureau of Justice Statistics.