How do Former Inmates Perform in the Community? A Survival Analysis of Rearrests,Reconvictions,and Technical Parole Violations

Prior Recidivism Studies of Release Cohorts

Criminological literature that explores the postprison performance of release cohorts is extremely limited. The two releases by the BJS (Beck & Shipley, 1989; Langan & Levin, 2002) and the Urban Institute’s (UI) reanalysis of the latter of these two reports are the only national-level studies of recidivism. The first of these BJS studies analyzed the behaviors of 108,580 individuals released across 11 different states in 1983 (Beck & Shipley, 1989); these data represented over half of all released state prisoners in that year. The second analyzed 272,111 individuals released across 15 different states in 1994 (Langan & Levin, 2002), these data represented two thirds of all state prisoners released in that year (Petersilia, 2003). Both used state and federal criminal record checks to track felony and/or serious misdemeanor arrests and convictions and analyzed data for a 3-year postrelease period. The data used for the 1994 cohort provided inquiries into returns to custody for either a new sentence or a parole violation.

Both BJS studies did not evidence promising results. The study of the 1983 release cohort found that 62.5% of those released from state prisons were rearrested within 3 years, and the study of the 1994 cohort found that 67.5% were rearrested within a similar follow-up period. Among those released in 1983, 46.8% were convicted of a new crime within the follow-up period, whereas 46.9% of those within the 1994 cohort were convicted. Reconviction rates did not significantly differ between the two studies. The 1994 study found that within 3 years of release, 51.8% of prisoners were returned to state custody either for new crimes or because of technical parole violations.

In 2005, the UI reanalyzed the data from the recidivism study of the 1994 cohort (Solomon et al., 2005) and disaggregated the total group of prisoners released across those 15 states according to the release type that they experienced. Release type groups included those who were released to discretionary parole (i.e., selected for release because of assessed “readiness” by a parole board), those released to mandatory parole (i.e., required supervision as a part of a determinant sentence), and those who were released unconditionally (i.e., served either all or the last part of a sentence in a prison and released without supervision). The UI’s (2005) research focused on rearrest outcomes over a 2-year follow-up period. Findings suggested, at a macro level, that after 2 years, a lower proportion of those released to parole via a discretionary function were rearrested when compared with unconditional releases and mandatory parolees.

Discretionary parolees were rearrested at a rate of 54%, whereas those released to parole via statutory mandate were arrested at a rate of 61%. Unconditional releases fared slightly worse with 62% of this group experiencing a new arrest within 2 years of release from prison. In an attempt to isolate the impact of postrelease supervision, the researchers controlled for predictors of rearrest, including demographic, criminal history, and instant offense information, through multivariate regression modeling. After statistical controls were in place, differences in postrelease performance were much less striking: Both unconditional releases and mandatory parolees had a predicted probability for rearrest at 61%, whereas discretionary parolees exhibited a predicted probability of only slightly lower at 57% (Solomon et al., 2005). These studies are useful because they highlight the postprison performance of many individuals across many different states and allow for a rough macro-oriented perspective into the ability of prisons to deter future criminal transgressions and how parole, in its various iterations, affects future criminality. However, these studies are also limited in their utility to public policy makers because of the enormous amount of state-to-state variation that exists in parole release and supervision policy.

As highlighted in the UI (2005) report, the types of prisoners that are released to parole as well as the types of interventions that are available to parole officers during supervision and overall supervision guidelines vary considerably from state to state. Additional follow-up reports using these data have suggested that significant variation occurs across states regarding how much ex-prisoners contribute to overall crime rates (Rosenfeld, Wallman, & Fornango, 2005), and some have argued that multistate aggregate level reports do not account for disparate state-level policy and practices from outcomes (Schlager & Robbins, 2008; Travis, 2005; Travis & Visher, 2005).

Furthermore, in these types of multistate analyses, jurisdictions with large corrections populations have greater impact on overall recidivism rates when compared with those with smaller populations. This impact can skew recidivism outcomes. For example, after California was eliminated from the BJS’s (1994) study, the national return to prison rate fell from 51.8% to 40.1%. Similarly, when California was excluded from the UI’s (2005) reanalysis of these data, rearrest rates for unconditional releases rose to 63% whereas rates for mandatory and discretionary parolees fell to 60% and 56%, respectively. Finally, the data that were utilized in both BJS reports are relatively outdated by today’s standards. Because of these shortcomings, there is a need for research into the postrelease behaviors of former prisoners according to the release type they experience that use large samples of current context-specific data.

Study Context

All discretionary release decisions and the supervision of all conditional releases are the responsibility of the state’s lead reentry agency: the New Jersey SPB. For the first of these functions, the SPB employs a mixed approach between discretionary and mandatory release mechanisms. Although most of the prison population of New Jersey are eligible to serve a portion of their sentence in the community if they are granted early release through parole, many individuals leave prisons unconditionally or “max out.” Members of this group serve time in prison until their sentence ends and are not supervised by parole after their release. Approximately 37% of New Jersey prison releases in a given year are max outs. About two fifths of this unconditionally released population voluntarily opts out of the parole process (Ostermann, 2010).

Offenders who are paroled in New Jersey can experience either a discretionary or mandatory release process. Discretionary releases are interviewed, typically by a two member Board Panel, and are either granted parole on their eligibility date or are denied parole and are issued a future eligibility term. Eligibility dates are calculated through an analysis of in prison time earned and/or given through work, commutation, jail, and good-time credits versus the terms of the individual’s sentence (i.e., mandatory minimum requirements). Inmates are typically eligible for parole release after serving approximately one third of their sentence after any mandatory minimums have been served and after accounting for any credits that have been earned and/or given to the prisoner. Future eligibility terms serve as parole “hits” and, when issued, are added to the eligibility date for which the prisoner is being actively reviewed for release (parole hearings occur approximately 3-to-6 months prior to the eligibility date). Hits typically range from 17 to 27 months (plus or minus 9 months if deemed appropriate by the Board Panel).

Some New Jersey inmates must serve a mandatory term of community supervision through parole as a part of their sentence. These inmates’ offenses include certain violent and/or sex offenses that require them to be supervised under either the No Early Release Act (for particular classes of violent crimes) or Parole Supervision for Life (for particular classes of sex crimes). The former is aimed at offenders convicted of first- or second-degree violent crimes in which the actor caused death, serious bodily injury, or threatened the immediate use of a deadly weapon (New Jersey Statutes Annotated 2C:43-7.2(d), 2009). Those convicted under the No Early Release Act are required to serve 85% of their sentence in custody before being eligible for parole. After parole release, the individual must serve either a 5 year (if convicted of a first-degree offense) or a 3 year (if convicted of a second-degree offense) term of mandatory parole supervision.

The latter form of mandatory supervision, aimed at those convicted of certain sex offenses (e.g., aggravated criminal sexual assault/contact, endangering, kidnapping, luring/enticing, etc.), requires parole supervision after release for a period of at least 15 years. After the individual has served their 15 year parole term, they are required to petition a court to be released from supervision. If the court finds that the individual does not pose a threat to the safety of others, they will release the individual from supervision (New Jersey Statutes Annotated 2C:46-6.4(c), 2009).

Data and Method

Archival information from multiple state-level criminal justice agencies were used to construct the data sets for these analyses. Release-event data were obtained from the New Jersey SPB and the New Jersey DOC. Information pertaining to rearrests and reconvictions were obtained through a database provided by the New Jersey Department of Criminal Justice (DCJ) within the Office of the Attorney General. This database highlighted formal criminal justice involvement through a query of Criminal Case History (CCH) reports.

The CCH reports are maintained by the New Jersey State Police and provide information on arrests, convictions, and adjudications that occur within the state. As a result, any criminal history/recidivism that either went unreported or occurred across state lines would consequently remain undetected throughout the analyses. Complications with using FBI Interstate Investigative Index (III) reports that highlight criminal involvement in other states resulted in the exclusive use of CCH reports within this research. While the use of these reports would have resulted in a potentially more complete picture of the criminal involvement of cases, data would not be as reliable as exclusively relying on CCH reports. Prior research that utilized FBI reports both within this context and in others found that many cases contained within this information system were missing data on adjudications (Langan & Levin, 2002) and that the proportion of persons released from one state who were subsequently rearrested in another within a 3-year period was very small (Spivak & Damphousse, 2006).

Technical parole violation information was obtained from the SPB because they are not considered rearrests and as a result are not highlighted on CCH reports. All cases were considered to be at risk of experiencing a postrelease arrest or conviction; however, technical parole violations were only considered for individuals who were supervised by the SPB. The SPB’s information system was queried to highlight all release events that occurred between January 1, 2006, and December 31, 2006. Each event was linked to an individual offender. Offenders were identified by their State Bureau Identification (SBI) number. The SBI number is attached to an individual at the time of arrest through fingerprinting. The link between an inmate’s SBI number and his or her fingerprint makes this identification unique to the individual.

According to SPB data, 12,555 individuals were released from the custody of the DOC in 2006. Criminal history and recidivism information were attached to 12,187 of the 12,555 unique individuals released in this year through a match on the SBI number. SPB information systems provided both the SBI number and release type of all of these reintegrating former inmates. Release types included unconditional (n = 4,438) and parole releases (n = 7,749). The latter category was further disaggregated to highlight the specific type of parole release event the individual experienced. This resulted in two broad categories: discretionary (n = 6,339) or mandatory release (n = 1,410) as well as two specific types of mandatory release: mandatory sex offender cases (n = 511) and mandatory violent offender cases (n = 899). These latter categories were used for the multivariate analyses that explored the likelihood of experiencing a postrelease technical parole violation. Recidivism information provided by the DCJ highlighted all arrest charges and offenses for which the individual was convicted throughout their criminal career up to October 31, 2009. Arrest charges and conviction offenses were collapsed by date to highlight unique criminal events. Total events were contrasted against the individual’s 2006 release date to highlight activities that occurred prior to and postrelease.

Cases were considered to be at risk if they were in the community between their 2006 release date and the date that data were gathered. This research exclusively utilizes Cox proportional hazards survival regression for its multivariate analyses because this method allows for the actual time an individual is at risk of experiencing a dependent variable to be accurately explicated. These considerations are especially pertinent when researching the experience of an event such as criminal recidivism because data sources utilized to capture information pertaining to traditional criminal justice involvement such as rearrests and reconvictions do not typically highlight the experience of technical parole violations (e.g., CCH reports).

If it precedes the commission of a crime, experiencing a technical parole violation makes an individual not at risk for being involved in new criminal activity. However, technical violations of parole are not considered criminal acts. This study accounts for the time an individual spent not at risk for recidivism in the community by calculating the number of days the former inmate spent in prison due to a technical violation of parole. This time is subtracted from the overall follow-up time if the person did not criminally recidivate. In addition, this time was subtracted from the time to criminal failure (rearrest and/or reconviction) if a technical parole violation preceded the failure event.

Bivariate contrasts of groups were accomplished through the use of t tests and cross-tabulations (chi-square). Throughout the analyses, several major contrasts are presented: (a) How those who are released to any form of parole supervision compare to those who are unconditionally released, (b) how those released via parole’s discretion compare to unconditional releases, (c) how mandatory parole releases compare to unconditional releases, and (d) how discretionary releases compare to mandatory supervision cases. These comparisons mirror those found in Solomon et al. (2005). Multivariate analyses also explore these comparisons and were constructed to predict the likelihood of experiencing recidivism while considering time at risk and statistical controls. These analyses, unlike the bivariate tests, also provide contrasts between discretionary parole releases and each of the two disaggregated mandatory parole release types (i.e., sex offenders and violent offenders) regarding the likelihood of experiencing a postrelease technical parole violation. Controls include (a) age, (b) minority status, (c) gender, (d) marital status, (e) county of conviction, (f) the number of instant offenses, (g) crime type of the instant offense, (h) the individual’s actuarial risk score, and (i) the number of prior arrests. Independent variables were tested for colinearity through the use of a correlation matrix prior to their inclusion in multivariate models.

Categorical variables were dummy coded within multivariate models. Minorities served as the variable of interest and Whites served as the reference category throughout the models. Females were the reference category for the gender variable. Single was the category of interest for the marital status variable with all other relationship statuses serving as the reference category. The county of conviction served as a proxy measure for the county to which the individual was released. This variable was collapsed so that individuals who were convicted in the counties that contain New Jersey’s largest and most depressed urban centers (i.e., the city of Newark in Essex County, the city of Camden in Camden County, and the city of Trenton in Mercer county) served as the category of interest and all other counties of conviction served as reference categories. A similar coding scheme was utilized for the number of instant offenses (which, because of the nature of the data, required categorical rather than continuous coding), where those with one instant offense conviction served as the category of reference for the categories of interest. Crime type communicates the most serious instant offense conviction with administrative offenses serving as the reference category. This variable was omitted from the final two multivariate models due to high correlations between the most serious instant offense and each of the disaggregated mandatory parole supervision types. Colinearity was not an issue when these supervision types were combined to form the mandatory parole supervision category.

Results

Results from bivariate analyses can be viewed in Table 1. Most of the people reintegrating back into the community were single (82.3%) Black (61.3%) males (92.5%). The average age was approximately 35. Almost a third of the reentering population was convicted of their instant offense in one of three of New Jersey’s counties that contain major urban centers (31.8%) and were classified as either low/moderate (42.8%) or moderate (47.1%) risk on a prerelease Level of Supervision Inventory–Revised (LSI-R) assessment. Average risk score across all offenders was about 23, which is on the threshold between low/moderate and moderate categories. Risk category bands were ascertained according to original Canadian data (Andrews, 1982).

OPEN IN VIEWER

Table 1.

Differences in Means and Proportions by Release Type Using t Test or Chi-Square.

	Significant Differences	All Released Prisoners (n = 12,187)	Max Outs(n = 4,438)	All Parolees (n = 7,749)	Discretionary Parolees(n = 6,339)	Mandatory Parolees(n = 1,410)
Demographics
% Male		92.5	92.5	92.4	92.2	93.5
% Ethnicity	a, c, d
Black		61.3	63.7	60.0	62.6	47.8
White		20.4	19.6	20.8	18.6	30.6
Hispanic		18.1	16.5	18.9	18.5	20.8
Asian		0.2	0.2	0.4	0.2	0.9
Age	a, c, d	34.98 (9.51)	35.81 (9.30)	34.51 (9.60)	34.34 (9.29)	35.23 (10.85)
% Age group	a, b, c, d
Below 25		13.4	10.7	14.9	14.9	14.5
25 to 30		25.7	24.0	26.7	26.3	28.5
31 to 40		33.4	34.3	32.4	33.2	28.7
Over 40		27.8	31.0	26.0	25.5	28.4
% Marital status	a, b, c, d
Single		82.3	83.8	81.5	82.7	75.8
Widowed		0.7	0.8	0.6	0.6	0.8
Married		8.7	7.0	9.6	9.0	12.0
Divorced		4.8	5.2	4.6	4.3	6.4
Separated		3.5	3.2	3.7	3.4	4.9
% County	a, b, c	31.8	37.9	32.4	32.8	30.1
Risk
LSI-R score	a, d	23.49 (6.08)	24.71 (5.93)	22.85 (6.06)	22.92 (5.87)	22.01 (8.03)
% Risk group	a, b, c, d
Low		5.7	3.0	7.2	6.5	16.3
Low/moderate		42.8	38.7	44.9	45.4	38.1
moderate		47.1	51.6	44.7	45.3	37.0
Moderate/high		4.2	6.3	3.1	2.7	8.0
High		0.2	0.4	0.1	0.1	0.7
Instant offense
% Number of instant  offense convictions	a, b, c, d
One		39.8	38.1	40.7	39.0	49.1
Two		27.5	26.5	28.0	28.3	26.7
Three		15.0	15.8	14.6	15.3	11.2
Four or more		17.7	19.6	16.6	17.4	13.0
% Instant offense includes
Administrative   offense	c, d	32.3	35.2	34.8	33.3	41.9
Property offense	a, b, c, d	27.9	34.9	27.4	30.1	14.4
Drug offense	b, c, d	51.3	55.8	55.5	64.2	13.8
Sex offense	a, b, c, d	3.1	1.6	4.4	0.8	21.8
Violent offense	a, b, c, d	19.3	18.2	22.4	14.8	58.6
% Most serious instant offense
Administrative offense	b, c, d	8.3	8.9	8.0	6.9	13.3
Property offense	a, b, c, d	17.8	21.3	15.7	18.1	4.6
Drug offense	b, c, d	50.2	50.4	50.0	59.7	3.7
Sex offense	a, b, c, d	2.8	1.1	3.8	0.5	19.8
Violent offense	a, b, c, d	20.9	18.2	22.4	14.8	58.5
Time served	a, b, c, d	1,076.01 (1,203.42)	1,256.38 (1,244.57)	970.95 (1,166.09)	883.69 (1,211.24)	1,389.71 (795.79)
Criminal history and recidivism
Total arrests	a, b, c, d	9.87 (7.54)	11.60 (8.23)	8.87 (6.92)	9.55 (7.08)	5.81 (5.17)
Total arrest charges	a, b, c, d	20.68 (14.56)	23.84 (15.29)	18.87 (13.80)	20.43 (13.89)	11.83 (10.89)
Prerelease arrests	a, b, c, d	8.37 (6.78)	9.73 (7.41)	7.60 (6.26)	8.16 (6.42)	5.06 (4.70)
Postrelease arrests	a, b, c, d	1.49 (1.94)	1.87 (2.21)	1.27 (7.73)	1.39 (1.80)	0.76 (1.19)
% Rearrested	a, b, c, d	60.3	67.8	56.0	59.2	41.6
% Time to rearrest
1 month	a, b, c	3.5	5.1	2.6	2.6	2.5
2 months	a, b, c	6.9	10.5	4.7	4.8	4.5
3 months	a, b, c	10.1	15.8	6.9	6.9	6.7
6 months	a, b, c	20.2	30.2	14.5	14.7	13.1
1 year	a, b, c, d	33.3	45.5	26.3	27.3	21.4
2 years	a, b, c, d	49.3	59.5	43.4	45.9	31.9
3 years	a, b, c, d	58.2	66.2	53.7	56.9	38.7
Total convictions	a, b, c, d	8.58 (7.02)	10.07 (7.74)	7.72 (6.42)	8.24 (6.55)	5.36 (5.23)
Prerelease convictions	a, b, c, d	7.46 (6.43)	8.58 (7.06)	6.82 (5.94)	7.27 (6.06)	4.78 (4.86)
Postrelease convictions	a, b, c, d	1.08 (1.73)	1.45 (2.11)	0.87 (1.43)	0.94 (1.48)	0.56 (1.12)
% Reconvicted	a, b, c, d	49.1	59.1	43.3	46.1	30.6
% Time to reconviction
1 month	a, b, c	2.3	3.3	1.7	1.7	1.4
2 months	a, b, c	3.6	5.2	2.8	2.8	2.6
3 months	a, b, c	4.7	7.1	3.3	3.4	3.0
6 months	a, b, c	8.9	13.9	6.0	6.0	5.4
1 year	a, b, c, d	19.0	28.1	13.8	14.1	12.1
2 years	a, b, c, d	35.8	47.8	29.0	30.5	21.7
3 years	a, b, c, d	46.7	57.1	40.8	43.4	28.7
Postrelease technical  parole violations	d	—	—	1.12 (0.37)	1.11 (0.34)	1.27 (0.53)
% Technical violation	d	—	—	19.4	21.0	12.3
% Time to tech PV
1 month	d	—	—	1.2	1.3	0.5
2 months	d	—	—	2.8	3.2	1.1
3 months	d	—	—	4.5	5.1	2.1
6 months	d	—	—	9.7	11.0	3.8
1 year	d	—	—	16.0	17.8	8.2
2 years	d	—	—	19.3	20.9	11.9
3 years	d	—	—	19.4	21.0	12.3

Note: LSI–R = Level of Supervision Inventory–Revised; PV = parole violation. Time is measured in days. Standard deviations of means are presented in parentheses. Significant differences are as follows:

a = Significant difference between max outs and all parolees.

b = Significant difference between max outs and discretionary parolees.

c = Significant difference between max outs and mandatory parolees.

d = Significant difference between discretionary parolees and mandatory parolees.

The majority of prior inmates returning to New Jersey communities in 2006 were serving sentences for one (39.8%) or two (27.5%) instant offenses, the most serious of these instant offenses was usually a drug-related crime (50.2% of the population’s most serious instant offense was for a drug-related crime, and 51.3% of the population had an instant offense that included a drug-related criminal conviction). On average, offenders served almost 3 years (1,076.01 days) for their instant offense conviction(s) prior to being released in 2006. Most offenders were released to some sort of supervision through parole (63.6%) with 52% being released per the discretion of the Board.

Offenders who were released in 2006 had an average of 8.37 arrests prior to their release date. About 60% experienced an arrest and about 50% experienced a conviction for a new crime after their 2006 release date. About 20% of all prior inmates were rearrested within 6 months of release, with a third of the entire population experiencing a rearrest within a year. About 20% of those who were released to parole supervision did not successfully complete and were revoked for a technical parole violation during the follow-up time. These violations typically occurred early in the supervision process with 9.7% of parolees experiencing a technical violation within 6 months of release and 16% experiencing this event within a year of release.

Max outs significantly differed from parolees on several demographic variables, including ethnicity, average age, and marital status. Max outs and parolees also significantly differed on their LSI-R score as well as risk-band classification. This group evidenced approximately two points higher (connoting greater measured risk) on the prerelease LSI-R when compared with parolees. Risk category classification also evidenced significant differences with 51.6% of max outs being classified as moderate risk and 44.9% of parolees being classified as low/moderate risk. Instant offense makeup significantly differed between max outs and parolees when considering the number of instant offenses these individuals were serving prison terms for, what the most serious offense was classified as, and what sorts of crimes these instant offenses included.

Parolees, on average, served significantly less time for their instant offenses when compared with max outs and experienced fewer total arrests and convictions throughout their criminal careers. Parolees had, on average, two fewer arrests and two fewer convictions prior to their 2006 release when compared with those who were released unconditionally. A significantly smaller proportion of parolees were either rearrested or reconvicted for new crimes between their release from prison in 2006 and the date of data gathering.

Many of the differences between unconditional and parole releases remained significant when the latter group was disaggregated according to whether the individual experienced discretionary or mandatory parole release. However, while unconditional releases differed from mandatory parole releases regarding ethnicity and actuarial risk, these differences did not reach a level of statistical significance when compared with discretionary parolees. Discretionary and mandatory parole groups significantly differed according to each of the three recidivism criteria, but mandatory releases evidenced significantly lower rates of rearrest, reconviction, and postrelease technical parole violation. Time to event experience differed between the two groups but only after 1 year of follow-up time.

Multivariate analyses consist of 11 models that provide for contrasts between the release groups according to the predicted likelihood of experiencing the three forms of recidivism after controlling for established predictors. All 11 models provided statistically significant good fits for predicting their respective dependent variables. The variable for group membership was a significant predictor for the dependent variables in Models 1, 2, 5, 6, 9, and 10. As highlighted by the odds ratios (ORs) in Table 2, parolees were about 25% less likely to experience a postrelease arrest (Model 1) and 32% less likely to experience a postrelease conviction (Model 5) when compared with those who were released unconditionally (OR = 0.751, p = .000 and OR = 0.681, p = .000, respectively) after controlling for the predictor variables. However, in an attempt to highlight absolute rate differences between groups, when the models are expressed as conditional probabilities parolees were predicted to be rearrested at a rate of approximately 57% and were predicted to be reconvicted at a rate of 51% versus 61% and 58% (respectively) for those released unconditionally in the presence of the control variables. Those who were released to parole supervision via the discretion of the SPB were about 26% less likely than max outs to experience a postrelease arrest (Model 2; OR = 0.745, p = .000) and about 33% less likely to experience a postrelease conviction (Model 6; OR = 0.670, p = .000) according to OR analyses. These models translate to a predicted probability that 55% of discretionary parolees would be rearrested and 52% would be reconvicted and a predicted probability that 59% of max outs would be rearrested and 59% would be reconvicted while controlling for all other variables in the models.

OPEN IN VIEWER

Table 2.

Hazard Ratios of Cox Survival Regressions.

	1	2	3	4	5	6	7	8	9	10	11
Comparisons
1. AP vs. MAX  (rearrest)	0.751***	—	—	—	—	—	—	—	—
2. DP vs. MAX  (rearrest)	—	0.745***	—	—	—	—	—	—	—	—	—
3. MP vs. MAX  (rearrest)	—	—	0.855	—	—	—	—	—	—	—	—
4. DP vs. MP  (rearrest)	—	—	—	0.930	—	—	—	—	—	—	—
5. AP vs. MAX  (reconvict)	—	—	—	—	0.681***	—	—	—	—	—	—
6. DP vs. MAX  (reconvict)	—	—	—	—	—	0.670***	—	—	—	—	—
7. MP vs. MAX  (Reconvict)	—	—	—	—	—	—	0.858	—	—	—	—
8. DP vs. MP  (Reconvict)	—	—	—	—	—	—	—	0.841	—	—	—
9. DP vs. MP (PV)	—	—	—	—	—	—	—	—	1.577***	—	—
10. DP vs. MP-  SO (PV)	—	—	—	—	—	—	—	—	—	4.001***	—
11. DP vs. MP-  VO (PV)	—	—	—	—	—	—	—	—	—	—	0.881
Controls
Age	0.965***	0.965***	0.971***	0.961***	0.964***	0.964***	0.966***	0.964***	0.996	0.997	0.998
Minority	1.146***	1.155***	1.225***	1.079	1.093*	1.107*	1.165*	1.024	0.777***	0.740***	0.754***
Gender	1.433***	1.410***	1.468***	1.428***	1.370***	1.342***	1.438***	1.341***	1.100	1.097	1.108
Marital status	1.083	1.070	1.049	1.118*	1.052	1.055	0.984	1.099	1.111	1.100	1.108
County	0.981	0.979	0.967	0.994	1.019	1.027	1.003	1.016	1.188**	1.168*	1.180**
Instant offenses
Two	1.100**	1.102**	1.104	1.087	1.045	1.037	1.032	1.066	0.968	0.976	0.961
Three	1.095*	1.106*	1.125	1.054	1.054	1.059	1.056	1.046	0.916	0.918	0.895
Four or more	1.134**	1.131**	1.125	1.149**	1.004	0.996	1.017	1.004	0.830*	0.811*	0.823*
Crime type
Property	1.129*	1.159*	1.105	1.135	1.286***	1.376***	1.263*	1.233*	1.388*	—	—
Drug	1.127*	1.157*	1.171*	1.086	1.219***	1.300***	1.304**	1.107	1.130	—	—
Sex	0.692**	0.674	0.580***	0.702*	0.850	0.839	0.639*	0.793	1.004	—	—
Violent	1.028	1.056	1.004	1.005	1.085	1.167*	1.028	0.987	1.091	—	—
LSI-R score	1.025***	1.024***	1.015***	1.033***	1.025***	1.023***	1.010**	1.039***	1.057***	1.058***	1.058***
Prior arrests	1.048***	1.048***	1.041***	1.053***	1.049***	1.049***	1.043***	1.053***	1.018***	1.019***	1.019***
Model statistics
-2 Log likelihood	93,818	89,636	34,983	55,309	77,409	73,947	30,796	43,654	19,521	18,789	19,202
Chi-square	1,325	1,237	419	893	1,170	1,112	388	696	216	207	184
Significance	***	***	***	***	***	***	***	***	***	***	***

Note: AP = all parolees; DP = discretionary parolees; MP = mandatory parolees; MAX = max outs; MP-SO = mandatory parolee-sex offenders;MP-VO = mandatory parolee-violent offenders; LSI-R = Level of Supervision Inventory–Revised.

*

p ≤ .05. **p ≤ .01. ***p ≤ .001.

The comparisons between unconditional releases and mandatory parole supervision cases did not reach statistical significance for either the rearrest or reconviction models. Mandatory and discretionary parole releases also did not differ in the predicted likelihood of experiencing either of these forms of recidivism. When technical parole violations were considered, ORs indicate that discretionary parolees evidenced about a 57% increased likelihood of being revoked on parole when compared with mandatory supervision cases (Model 9; OR = 1.577, p = .000). This difference is likely attributed to the postrelease behaviors of mandatory supervision sex offender cases. When compared with this group, discretionary parolees were predicted to be four times as likely to experience a technical parole violation (Model 10; OR = 4.001, p = .000).

Control variables that obtained statistical significance in their ability to predict the various dependent variables throughout the iterations of the different regression models remained relatively consistent in terms of direction and, to a lesser extent, effect size. The variables of age, minority status, gender, the number of instant offenses, crime type, LSI-R score, and prior arrests obtained significance in most of the models. The age variable was a significant predictor for all models that were used to predict involvement in new criminal activities. Results generally indicated that for every additional year of age at the time of release, the predicted likelihood of experiencing a new arrest or conviction significantly decreased. Offenders classified as minorities were at an increased risk of experiencing a postrelease arrest or conviction when compared with nonminorities; however, minorities were at less risk of experiencing technical parole violations when compared with nonminorities. Minority status was not significant in models that compared unconditional and mandatory parole supervision releases. Males were more likely than females to experience postrelease recidivism for criminal transgressions but not technical parole violations. The county variable was not a significant predictor of experiencing new criminal involvement; however, it did significantly predict technical parole violations. Additional points on a pre-release LSI-R assessment and additional arrests on an individual’s criminal record prior to their release put a former inmate at an increased hazard of experiencing all three types of recidivism events.

Discussion

Results generally indicated that those who were released unconditionally were predicted to be more likely to experience new criminal involvement when compared with those who were released to parole supervision. Max outs significantly differed from discretionary parolees, but these differences were not evident when compared with mandatory supervision cases within multivariate models. Those released under discretionary parole were predicted to be at an increased hazard of experiencing a technical parole violation when compared with the mandatory parolees in general and mandatory sex offender cases in particular. Mandatory and discretionary parole releases did not differ in the predicted likelihood of experiencing a rearrest or a reconviction after controlling for the predictor variables.

Findings from the present research are thematically similar to results found in other studies within this context that investigated the postrelease performance of parolees and max outs in general (Schlager & Robbins, 2008); parolees who receive postrelease community program interventions (Ostermann, 2009); the performance of parolees, max outs, and specialized parole caseloads (Veysey & Lanternman, 2008); and the performance of parolees and subpopulations of the max out population (Ostermann, 2010). These previous studies generally found that those released unconditionally were not as successful at reintegrating as those released to parole supervision. Results were also similar to the recidivism outcomes of the BJS (2002) study of a 1994 release cohort in that findings suggested that the first 6 months to 1 year after release are the most crucial to an individual’s success in the community.

When conditional probabilities were calculated, findings were similar to the UI’s (2005) reanalysis of the BJS’s (2002) data in that the predicted likelihood of a former inmate experiencing new criminal involvement did not substantially differ according to release type after controlling for pertinent predictors of recidivism. Discretionary releases indicated a slightly higher likelihood of succeeding in the community when compared with max out cases; however, these differences did not reach a level of statistical significance when mandatory releases were considered. This is likely due to the mandatory supervision policies that exist in this context, a nuance that was largely lost due to the multijurisdictional nature of the UI’s data. Because of the policies that are currently in place in this jurisdiction, the resources that the SPB allocates to mandatory supervision cases both prior to and after release are likely affected.

Within the present study context, only individuals who are convicted of certain sexual or violent offenses are required to serve a time of mandatory parole supervision, and, as evidenced by the bivariate results, these offenders are supervised after serving relatively long prison terms. It would appear that these individuals are mandated supervision because they are perceived as “harder” criminals when compared with those who are released via parole’s discretion. However, the results of this study do not empirically support this consensus. Mandatory supervision cases were on average lower risk according to the LSI-R, were involved in less crime prior to their release, and ultimately had lower levels of recidivism.

An effect of the mandatory supervision policies is that these inmates have no parole decision rendered by the releasing authority. Because these cases technically max out from prison and subsequently serve long periods of supervision, it is likely that the Board does not dedicate the level of effort toward case preparation and rehabilitative planning for mandatory cases when compared with cases for which a discretionary release decision must be reached. If a Board Panel does not have the authority to affirm or deny parole for these cases, it is unlikely that resources are steered toward analyzing prerelease indicators of recidivism that would be used to inform a release decision. This in turn would likely negatively influence the amount of information at an officer’s disposal about individual-level risks and needs as they attempt to supervise these mandatory parolees in the community.

Many of the potential fruits of these prerelease analyses are likely lost for these cases. This in turn could adversely affect rehabilitative plans, effectuating these plans and ultimately managing and mitigating risks during the supervision phase of parole. Consequently, it is unlikely that rehabilitative resources are steered toward these cases at the time of their release. Officers must spend time developing intelligence about these individuals that would have otherwise been available if a discretionary decision would have been rendered. This is especially salient given the results of this study and the findings from similar research efforts: The most challenging time for a reintegrating individual is the first 6 months after release.

Results from this study ultimately beg the question of whether parolees experience greater success because those who are targeted for release actually benefit from community supervision and an increased access to community programs or because those who are selected by the Parole Board are de facto more likely to succeed. While this research did not address the “black box” of parole, namely, what sorts of community interventions were offered to and/or experienced by parolees after release, results from both sets of analyses are indicative that parole seems to “work,” at least when considering its release function.

On average, those who were extended discretionary parole release had less risky prerelease indicators when compared with those who maxed out. Discretionary releases were generally incarcerated for less egregious crimes, had less previous involvement in criminal activities, and had lower scores on an actuarial assessment. Multivariate models were constructed to control for these differences in an attempt to isolate the effects of parole on performance in the community. Although these models are inherently limited by data availability in their abilities to control for all confounding explanations for postrelease criminal involvement, between-group differences could have been mitigated through an experimental design. However, studies that have used this approach have similarly found few differences in the postrelease performance of those who are supervised versus those who are not (Jackson, 1983).

Results indicate that, when afforded the opportunity to make decisions, this release authority is skilled at triaging offenders into groups but that the management of these cases after release needs substantial improvement. The majority of inmates who were released to parole were rearrested within 3 years and many were also convicted of new crimes. Furthermore, the absolute predicted differences in rearrests and/or reconvictions between parolees and unconditional releases were negligible in the presence of statistical controls. These findings paired with the facts that discretionary based parole systems, in most cases, have the ability to choose which individuals receive their interventions, have various rehabilitatively oriented evidence based interventions at their disposal in the community, and that once release is granted have the ability to send a problem case back to prison without the individual committing a new crime does not reflect desirable outcomes about the abilities of parole boards to ensure that those they supervise successfully desist from crime.

These findings, however, should be cautiously consumed. These data are limited in their ability to tell a complete story about the successes and failures of parole because several key variables were not available for analysis. Variables that connote criminal risk such as involvement in in-prison disciplinary infractions and measures of antisocial/criminogenic personality, exposure to criminal peers after release, social support systems, family rearing practices, substance abuse issues, and so on (Gendreau, Little, & Goggin, 1996) were unavailable at the time of data gathering, and were thus impossible to control for in multivariate models.

Future research should both investigate those who are released to parole and subsequently fail for their prerelease indicators as well as investigate postrelease supervision interventions that are offered to those who do not fail (holding prerelease failure indicators constant), akin to the research currently being conducted by the UI for release cohorts in the state of Massachusetts (Kohl, Hoover, McDonald, & Solomon, 2008). The effectiveness of parole programs should be further studied. Programs found to be effective should be targeted at those who pose the greatest risk and should be consistent with demonstrated need. These types of services should be frontloaded (Austin & Hardyman, 2004; Binswanger et al., 2007; Lowenkamp, Pealer, Smith, & Latessa, 2006). Although it is not expected that releasing authorities making discretionary parole decisions will ever be able to make perfect predictions about who will succeed and parole officers charged with supervising those who are released will never prevent all individuals from failing in the community, having the knowledge of how those who succeed and those who fail in the community differ according to risk, need, and appropriate service reception will allow the former group to target decision-making practices more effectively and will allow for the latter group to target postrelease interventions with greater impact.