Evaluation of a Serious and Violent Offender Reentry Initiative (SVORI) Program in a Midwest State

Abstract

The Serious and Violent Offender Reentry Initiative (SVORI) paved the way for a new era of rehabilitation in corrections’ programming. However, published outcome evaluations of SVORI programs and their progeny are limited in number. The current article presents the multiyear outcome evaluation of one prisoner reentry initiative established in a Midwestern state, which was developed within the framework of the SVORI program model. A comparison group was identified using propensity score matching to evaluate program effectiveness on the recidivism outcomes of returns to prison and new convictions. Cox proportional hazards modeling found program participants to have significantly lower hazard to incur a new conviction than the comparison group but no difference in the hazard for reincarceration. The implications of these mixed findings in recidivism outcomes are discussed for the reentry program initiative.

Keywords

prison reentry program evaluation SVORI

In 2000, then U.S. Attorney General Janet Reno (2000) drew attention to the nation’s significant challenges in bolstering the successful reentry of offenders from prison back to the community. In the 15 years since Reno’s highlighting of the complexities of prisoner reintegration, the United States has seen a rapid build-up of programs intended to promote the successful reentry of incarcerated persons into communities across the country. Efforts led by the federal government to promote prisoner reentry programs through the Serious and Violent Offender Reentry Initiative (SVORI), implemented in 2002, and the Second Chance Act of 2007, paved the way for a new era of embracing the value and ideas of a rehabilitation agenda. The enactment of both of these initiatives greatly accelerated the development of programs for former prisoners by providing funding to state and local governments to either establish new or enhance existing prisoner reentry programming.

The multiyear SVORI granted approximately US$110 million to 69 correctional agencies located in all 50 states, the District of Columbia, and the U.S. Virgin Islands (Lattimore et al., 2004). An important legacy of the SVORI is its three-phase program model (Lattimore & Visher, 2009), which provided funded jurisdictions with a structure within which they could design and deliver reentry services. First, the SVORI programs enrolled participants prior to their release from prison and supported efforts to implement a system of structured inmate assessments to determine both the individual’s risks of recidivism and the services needed to address those risks. Second, the SVORI encouraged the intensification of program services to be delivered immediately before and on an inmate’s release from prison, as the individual transitioned from prison into the community. Third, reentry program efforts continued in the community for 6 months to 1 year after release, as participants established “productive and independent roles” (Lattimore & Visher, 2009, p. ES-1).

The freedom given to funded jurisdictions to create their own reentry programs within this three-phase SVORI structure proved to be both productive and perplexing. While certain commonly adopted program offerings were found to be of benefit to individuals in reentry programs, for example, housing and employment services (Lattimore, Steffey, & Visher, 2010; Lattimore & Visher, 2009), the broad array of offerings across the 69 agencies made wholesale rigorous evaluation impossible, primarily because the programs lacked uniformity across sites, particularly seen in variances in content, dosage, and duration. Consequently, evaluations of single-jurisdiction programs where information about specific program structure and content can be obtained are more useful for replication and implementation efforts.

This article presents the findings of an evaluation of one prisoner reentry initiative established in a Midwestern state, which was supported by SVORI funding and developed within the framework of the SVORI three-phase program model. Recidivism outcomes (e.g., returns to prison and new convictions) of the SVORI participants (hereinafter, Reentry, n = 467) are analyzed relative to a matched comparison group (n = 467) using Cox proportional hazards modeling.

SVORI Evaluation Literature

Despite the fact that the SVORI programs and their progeny are more than a decade old, published outcome evaluations of these programs are limited in number. Indeed, only five studies could be located that specifically reported on the outcomes of a SVORI program. The most comprehensive and detailed of the studies is that of Lattimore and Visher’s (2009) multisite evaluation of 12 adult program sites across the United States. This quasi-experimental investigation yielded mixed results when the SVORI participants were analyzed relative to a matched comparison group. The SVORI programs significantly increased participants’ access to services, but the recidivism outcomes varied by sex and by reason for reincarceration. When viewed with their matched comparisons, male SVORI participants had moderately lower rates of rearrest and no difference in reincarceration rates. Female SVORI participants were significantly less likely to be rearrested but significantly more likely to be reincarcerated as a result of violations of the terms of their post-release supervision.

Although Lattimore and Visher (2009) looked at multiple sites across a number of states, evaluation studies of single-state SVORI-guided reentry programs are more common. In one Midwestern state, participants in a SVORI program were matched to comparison individuals using propensity score matching. Results showed the SVORI participants to have a greater likelihood to return to prison but a significantly lower rate of new convictions (Severson, Bruns, Veeh, & Lee, 2011). Using an unmatched group of comparable individuals, a SVORI program in North Dakota found its participants to use community services at a greater rate and incur significantly fewer positive drug tests and rearrests post release (Bouffard & Bergeron, 2007). Similar to the findings in North Dakota, a SVORI program evaluation in New Jersey indicated participants were significantly less likely to be rearrested compared with two groups of randomly selected parolees and persons with unconditional releases who met the inclusion criteria for participation in the reentry program (Veysey, Ostermann, & Lanterman, 2014). Finally, in an unpublished evaluation from Nebraska, again using a comparison group of individuals who met reentry program eligibility criteria but who chose not to participate, the authors found the SVORI participants to have a moderately lower rearrest rate in the community (Sample & Spohn, 2008). However, with its small sample of 19 in SVORI and 53 in the comparison group, the results of this evaluation must be viewed with caution.

In sum, the findings of the evaluations of the SVORI programs are mixed at best. Although Lattimore and Visher (2009) provide the most rigorous investigation of SVORI programs to date, their findings also reveal the greatest ambiguity. Evaluations based in a single state provide a better picture of SVORI outcomes, but many of these studies have methodological flaws that raise questions about the validity of their results. Overall, to further understand its effectiveness in reducing criminal recidivism, there is a need for additional empirical investigation of programs that followed the SVORI model.

Reentry Program Description

The statewide prisoner reentry initiative examined here was located in the Midwestern region of the United States. Initial funding of the statewide initiative was provided by the seven federal partners who joined together to support the SVORI; thus, the program evaluation presented here largely followed the three-phase SVORI program model, with a particular emphasis on the in-prison and community phases of the program design (see Lattimore & Visher, 2009). Persons eligible for participation in the reentry program included males and females, 18 years of age or older, with at least 12 months remaining to serve on their current sentence, and who were scheduled to return to one of three jurisdictions in the state that followed a formal reentry program format. One final criterion, related to offenders’ composite Level of Service Inventory–Revised (LSI-R) scores, was used in an effort to maintain fidelity to the Andrews and Bonta’s (2010) Risk–Need–Responsivity (RNR) principles by targeting services to those at the highest risk, addressing dynamic risk factors, and emphasizing cognitive behavioral strategies throughout all interventions. The reentry program goals were created on the basis of matching service referrals to high LSI-R domain scores, and progress on these goals was tracked during the in-prison phase of the program. Individuals who met all other eligibility criteria and who had composite scores of 30 or higher were included in the sample. Certain individuals with LSI-R composite scores between 25 and 29 were also admitted on a case-by-case basis, if, for example, the participant was determined to have multiple incarceration episodes, heavy drug and alcohol use, or an inadequate education and employment history, making him or her an appropriate candidate for an enhanced level of services to assist in community reintegration.

The in-prison phase of the reentry program occurred in the 12 to 18 months prior to an individual’s scheduled release-from-prison date. The LSI-R assessment provided guidance for identifying an individual’s criminogenic needs during the in-prison phase of the program. Based on the LSI-R results, particularly related to those criminogenic domains wherein the individual was assessed as having high needs, referrals to prison-based programs were made. For example, a referral to the workforce development program for those high in LSI-R education/employment or a referral for a substance abuse assessment for those high in the LSI-R alcohol/drug domain could be made. Fidelity to the RNR principles within the day-to-day reentry program activities was not systematically examined.

In addition to the specific programs offered to the inmate during the in-prison phase, another component of this preparatory period included in-reach by the participant’s community-based case manager and parole officer. During in-reach, the case manager built the individual’s case plan based on the LSI-R results. The same case plan and the case manager followed the individual from the facility to the community with the intent of securing a seamless delivery of services during the reentry process. The LSI-R was again administered a few weeks after community reentry after which the case plan was revised.

Within the community phase of the program, needs for services and the corresponding referrals were again based on the individual’s criminogenic needs as identified by an updated, post-release LSI-R. A team of community providers including the case manager, parole officer, and an accountability panel composed of community stakeholders worked together to secure the long-term supports thought necessary to optimize the individual’s chance of success in the community. Participants identified as being successful in the community were recognized in an official graduation ceremony at approximately 6 months post release. Alumni services were also offered to any graduate following his or her disengagement from the program; however, data on alumni services were not captured for this evaluation.

Method

Sample

Program participants, hereinafter, the “Reentry” group, included those who were assessed to have the highest criminogenic risks and needs as measured by the LSI-R, who would voluntarily participate in an in-prison program for at least 12 months, and who would be released to one of three counties that hosted a designated reentry program.

To form the comparison group pool, the state’s department of correction provided the administrative data that included information relevant to the research on individuals released from the state prison system from 1967 through 2010. Individuals were selected for matching if they were released between 7/1/2006 and 6/30/2010, completed the LSI-R at least once, and were not transferred out of state during their incarceration. Next, individuals who died while incarcerated (n = 26) or who moved to an out-of-state location on release (n = 512) were eliminated from the pool. The final pool of comparison cases for matching was 12,704.

Propensity score matching

Propensity score matching (PSM) was used to obtain the study samples of the Reentry and comparison groups. The matching procedure followed Guo and Fraser’s (2010) guidelines. Propensity scores, which are the conditional probability of assignment to the intervention (Reentry) group versus a comparison group, were calculated for each individual from an estimated logistic regression model, which included the covariates of age, race, gender, months served on index incarceration, prior months served, community supervision, and the 10 individual LSI-R domain scores, including criminal history, education/employment, financial, family/marital, accommodation, leisure/recreation, companions, alcohol/drug, emotional/personal, and attitudes/orientation. Similar to Duwe and King (2012), it was viewed as an imperative to control for recidivism risks when matching cases for an evaluation where program selection was determined by an individual’s likelihood for criminal recidivism, and program effectiveness was based on differences in recidivism outcomes. However, instead of simply matching on the LSI-R total score, the current analysis aimed to achieve a better match by including the individual LSI-R domain scores that capture different attributes and situations relevant to the level of supervision provided and treatment decisions made. Overall, the PSM was limited by the administrative data available on both the Reentry and comparison groups. Informed by the work of Andrews and Bonta (2010) and Lipsey and Derzon (1998), the propensity score estimation included every theoretically relevant covariate (16) available in the administrative data that would affect an individual’s risk to recidivate and in turn be selected into the Reentry group. An important requirement of PSM is a large sample size; the current comparison pool of 12,704 resulted in a ratio (~27:1) above that recommended by Lunt (2013) of approximately 20 to 1. Furthermore, it was determined that propensity scores from both the Reentry and comparison groups must demonstrate significant overlap in values for the PSM to be effective (Shadish, Cook, & Campbell, 2002). Although propensity scores significantly differed between the two groups, t(476.57) = −14.01, p < .01, there was still substantial overlap. Both groups contained substantial proportions of cases with a propensity score less than 0.4 (Reentry = 94.7%, Comparison = 99.8%).

Moving forward with the analysis, a matched sample of comparison cases was drawn from the final pool described above, using the nearest neighbor matching algorithm with a caliper (see D’Agostino, 1998). A caliper value of 0.248 was determined by following the recommendation of Austin (2010)—that is, 0.20 × standard deviation of the logit of the propensity scores for the entire pre-match samples. Two matched samples of 467 cases in both the Reentry and comparison groups were obtained. The covariate balance was checked through three different methods: visual inspection of box plot and quantile–quantile (Q-Q) plot, comparison of standardized difference between before and after matching, and an omnibus multivariate test for standardized difference. As detailed in Table 1, the standardized difference (d) was computed between the two groups on each of the 16 covariates using Haviland, Nagin, and Rosenbaum’s (2007) formula for the absolute standardized difference in covariate means (see Guo & Fraser, 2010). After matching, the d value dropped below 0.1 for each of the covariates, indicating PSM improved covariate balance with less than 10% of differences in standard deviation. In addition, xBalance package in R was utilized to examine whether the overall standardized difference across all 16 covariates is significant (Bowers, Fredrickson, & Hansen, 2013). The results of this omnibus test revealed the matched samples to be similar across all the covariates, χ²(16) = 9.47, p = .893. Finally, a sensitivity analysis was conducted to assess whether the matching was robust to the possible presence of unobserved covariates (Rosenbaum, 2010). Because the outcome variables (return to prison, new conviction) were binary, this analysis was based on McNemar’s test that compares the proportions between the two matched samples. The lower and upper bounds of the p value were calculated over a range of sensitivity parameter Γ values using rbounds package in R. This parameter represents the magnitude of hidden bias due to the presence of unobserved covariates. The results showed that our inference on statistical significance would not change even if individuals who have the same values on the 16 observed covariates differed in their odds of receiving the intervention by as much as 30%, suggesting that the matching was fairly robust and unbiased.

Table 1.

Propensity Score Matching—Absolute Standardized Difference in Covariate Means.

Covariate	Standardized difference pre-match (n = 13,177)	Standardized difference post-match (n = 934)
Age	0.08	0.09
Gender	0.38	0.08
Race	0.33	0.04
Index incarceration months	0.62	0.02
Prior months in prison	0.64	0.02
Supervision	0.56	0.02
LSI-R criminal history	0.53	0.00
LSI-R education/employment	0.35	0.00
LSI-R financial	0.23	0.05
LSI-R family/marital	0.26	0.00
LSI-R accommodation	0.18	0.00
LSI-R leisure/recreation	0.12	0.02
LSI-R companions	0.23	0.01
LSI-R alcohol/drug	0.08	0.05
LSI-R emotional personal	0.32	0.01
LSI-R attitudes/orientation	0.19	0.06

Note. Pre-match χ² = 760.00, df = 16, p < .001. Post-match χ² = 9.47, df = 16, p = .893. LSI-R = Level of Service Inventory–Revised.

Six of the 473 Reentry cases that entered the PSM could not be matched with a case within the comparison pool. Bivariate comparisons (t test and chi-square test) were conducted to identify any significant differences between the matched and unmatched Reentry cases. The unmatched cases were significantly older (matched = 37.5 years, unmatched = 50.5 years) and spent considerably more months incarcerated both overall (matched = 71, unmatched = 256) and on the index incarceration (matched = 36, unmatched = 178).

Data Collection

Precautions were taken to ensure the accuracy and completeness of the data. All data provided by the state were repeatedly cross-checked with the program database at each reentry site, the public database of state inmates, and with reentry program staff. These data collection systems were brought together to provide information about relationships between a defined participant and program elements and recidivism outcomes, including the two measures of new convictions and returns to prison reported here. Comprehensive program data were collected on the Reentry participants who provided informed consent. Data collection for the comparison participants was limited to the data made available through the memorandum of understanding between the state and the study’s principal investigator.

Measures

Age, gender, and race

Age, gender, and race have robust associations with criminal recidivism; thus, this trio of variables was considered as covariates in the analyses. Race was coded dichotomously as either “White” or “non-White.”

Supervision

Supervision was coded “yes” if a participant was required to report to a state parole office on his or her return to the community.

Index incarceration

The index incarceration was defined as the number of months an individual served on the sentence from which he or she was initially released for this study.

Prior months in prison

The total number of months an individual served in a state prison between 1967 and the date of admission for his or her index incarceration. Only the time served in the state prison system where the evaluation was conducted counted toward the total.

Level of Service Inventory–Revised (LSI-R)

The LSI-R is an assessment tool designed to identify problem areas in an individual’s life and to predict risk of recidivism. The LSI-R comprises a total of 54 items across 10 domains. Those domains include criminal history, education/employment, financial, family/marital, accommodation, leisure/recreation, companions, alcohol/drug, emotional/personal, and attitudes/orientation (Andrews & Bonta, 2010). The LSI-R has extensive empirical support for both its reliability and predictive validity (see Flores, Lowenkamp, Holsinger, & Latessa, 2006).

Return to prison

The return to prison was coded “yes” if a participant returned to the state prison system in the state where the program was delivered and the evaluation conducted.

New conviction

The coding of “yes” indicated those participants who incurred new convictions in the same Midwestern state where the program and program evaluation were conducted.

Data Analysis

Statistical analyses include descriptive statistics, bivariate tests, survival analysis, and Cox proportional hazards modeling. All data analyses were performed using SPSS 22; statistical significance was determined at .05 alpha level.

Results

Descriptive Statistics

Demographics and recidivism outcomes are described in Table 2. Independent samples t test (for continuous or ordinal variables) and chi-square test (for categorical variables) were conducted to examine group differences.

Table 2.

Descriptive Statistics of Demographic Characteristics and Recidivism Outcomes.

	Article 1. Reentry		Comparison		t/χ²	p
	(N = 467)		(N = 467)
	M/N	SD/%	M/N	SD/%
Demographics
Age (in years)	37.46	9.48	38.34	9.80	1.41	.160
Gender					1.51	.219
Male	348	74.5%	364	77.9%
Female	119	25.5%	103	22.1%
Race					0.43	.512
White	247	52.9%	257	55.0%
Non-White	220	47.1%	210	45.0%
Supervision					0.05	.819
No	41	8.7%	43	9.2%
Yes	426	91.2%	424	90.8%
Index incarceration months	36.47	25.72	37.07	36.86	0.29	.773
Prior months in prison	71.04	59.58	69.59	57.29	−0.38	.705
LSI-R total score	31.48	5.78	31.60	7.57	0.29	.774
LSI-R criminal history	6.88	1.58	6.88	1.61	0.04	.967
LSI-R education/employment	7.30	1.70	7.31	1.90	0.04	.971
LSI-R financial	1.01	0.69	0.98	0.74	−0.69	.493
LSI-R family/marital	1.96	1.15	1.95	1.17	−0.06	.955
LSI-R accommodation	0.94	1.05	0.94	1.06	0.00	1.000
LSI-R leisure/recreation	1.61	0.63	1.60	0.64	−0.36	.719
LSI-R companions	3.14	1.02	3.13	1.08	−0.16	.877
LSI-R alcohol/drug	4.10	2.57	4.22	2.71	0.71	.480
LSI-R emotional/personal	1.73	1.54	1.72	1.56	−0.15	.882
LSI-R attitudes/orientation	2.79	1.16	2.87	1.14	0.97	.333
Recidivism
Any new conviction					21.27	.000
No	431	92.3%	384	82.2%
Yes	36	7.7%	83	17.8%
Any return to prison					0.30	.586
No	302	64.7%	294	63.0%
Yes	165	35.3%	173	37.0%

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

For the most part, there was no discernible difference in any of the demographic variables, which was expected given that most of the demographic variables had been used as covariates in the preceding sample matching. There were no missing data.

Although not statistically significant, the Reentry group had a lower rate of returns to prison than did comparison participants (35.3% vs. 37.0%). The rate of new convictions was significantly lower for Reentry participants over the follow-up period, χ²(1) = 21.27, p < .05, at less than half the rate of that incurred by comparison participants.

Survival Analysis

Survival analysis, as described below, was conducted separately for each of the recidivism outcomes of interest. First, cumulative probability of the event (not returning to prison, not receiving a new conviction) during days in the community (survival function) was estimated via the Kaplan–Meier product-limit method; the groups were then compared by a log-rank chi-square test. Next, Cox proportional hazards models were sequentially fitted to examine group difference in hazard function—that is, the risk (hazard) of returning to prison or incurring a new conviction at a given time period—adjusting for age, gender, race, supervision, and LSI-R total. A strength of survival analysis is the ability to analyze time-to-event data for individuals that released throughout the study time frame from 7/1/2006 to 6/30/2010. Reentry participants and their matched counterparts were included in the study if released into the community prior to the termination of the study. This study design resulted in a range of possible time in the community prior to study termination from 2 to 1,972 days among the matched sample of 934 participants. Despite the wide range in the possible time to experience a return to prison or a new conviction, 97.8% of the sample was released for at least 1 month prior to the termination of the study, and 76.6% was released for at least 1 year prior to study termination.

Figure 1 presents the estimated survival functions of not returning to prison for the Reentry (green line) and comparison (blue line) groups. There was no discernible difference in survival function between the two groups, log-rank χ²(1) = 0.008, p = .929.

Figure 1.

Survival function—prison return.

As shown in Figure 2, below, the groups significantly differed in the estimated survival function to incur a new conviction, log-rank χ²(1) = 18.82, p < .001. Reentry participants (green line) spent a mean of 1,797.42 days (SD = 28.68, 95% confidence interval [CI] = [1,741.21, 1,853.63]) in the community prior to incurring their first new conviction, while comparison participants (blue line) spent only 1,198.25 days in the community (SD = 24.64, 95% CI = [1,149.97, 1,246.54]). Thus, the Reentry group stayed in the community, on average, almost 600 days longer than the comparison group, before sustaining a new conviction.

Figure 2.

Survival function—new conviction.

Prior to Cox modeling, the proportionality hazards assumption—that is, shapes of survival functions are the same for all levels of covariates over time—was evaluated by inspecting the interaction between time and each of the covariates in the model (Tabachnick & Fidell, 2013). A Bonferroni-adjusted p value was used for statistical significance because of the number of interactions being evaluated (.05/8 = .008; Tabachnick & Fidell, 2013). Significant interaction was found between time to prison return and LSI-R total score; thus, the time by LSI-R total score interaction term was included in the final return to prison model. In contrast, none of the interactions between time to new conviction and the covariates were significant, and thus, none were included in the final new conviction model.

The estimates from the final return to prison model are detailed in Table 3. Because the outcome of return to prison was coded 0 for No and 1 for Yes, a positive value of the coefficient b indicates a greater hazard of returning to prison. After adjusting for the model covariates, no group difference was found in the hazard of returning to prison, which was similar to unadjusted results from the survival functions analysis (see Figure 1). Participants who were male, non-White, released to community supervision, and had a higher LSI-R total score were found to be at significantly greater hazard of returning to prison. Females had 36% lower hazard of returning to prison (hazard ratio [HR] = 0.64, p < .01, CI = [0.48, 0.84]). White participants (HR = 1.24, p < .05, CI = [1.004, 1.54]) and those released without community supervision (HR = 0.29, p < .01, CI = [0.17, 0.51]) had 19% (1/1.24 = 0.81) and 71% lower hazard of returning to prison, respectively. In terms of the LSI-R total score, for each 1-point increase, the hazard of returning to prison increased by 18% (HR = 1.18, p < .01, CI = [1.07, 1.29]).

Table 3.

Cox Proportional Hazards Model—Prison Return.

Parameter	b	SE	Wald χ²	p	HR	95% CI for HR
Parameter	b	SE	Wald χ²	p	HR	Lower	Upper
Age (in years)	−0.01	0.01	3.63	.057	0.99	0.98	1.00
Gender (Ref = Male)	−0.45	0.14	9.87	<.01	0.64	0.48	0.84
Race (Ref = White)	−0.21	0.11	3.98	<.05	1.24	1.004	1.54
Supervision (Ref = Yes)	−1.23	0.28	18.86	<.01	0.29	0.17	0.51
LSI-R total score	0.16	0.05	11.40	<.01	1.18	1.07	1.29
Group (Ref = Reentry)	−0.03	0.11	0.09	.769	0.97	0.78	1.20
Time × LSI-R total score	−0.02	0.01	7.12	<.01	0.98	0.96	0.99

Note. b = unstandardized regression coefficient; SE = standard error; HR = hazard ratio; CI = confidence interval; LSI-R = Level of Service Inventory–Revised; Ref= Reference.

The estimates from the final new conviction model are presented in Table 4. The outcome of new conviction was also coded 0 for No and 1 for Yes. After controlling for the covariates, the Reentry group significantly differed from the comparison group in the hazard to incur a new conviction following release. Reentry participants had 55% lower hazard of incurring a new conviction than the comparison group (HR = 2.22, p < .01, CI = [1.50, 3.29]; 1/2.22 = 0.45), which reflected the unadjusted results in Figure 2. The hazard for a new conviction was also significantly increased for male participants (HR = 0.44, p < .01, CI = [0.25, 0.77]; 1/0.44 = 2.27) and those assessed with a higher LSI-R total score (HR = 1.05, p < .01, CI = [1.02, 1.08]).

Table 4.

Cox Proportional Hazards Model—New Conviction.

Parameter	b	SE	Wald χ²	p	HR	95% CI for HR
Parameter	b	SE	Wald χ²	p	HR	Lower	Upper
Age (in years)	−0.02	0.01	3.24	.072	0.98	0.96	1.002
Female (Ref = Male)	−0.83	0.29	8.34	<.01	0.44	0.25	0.77
Non-White (Ref = White)	0.01	0.19	0.01)	.941	1.01	0.70	1.46
No Supervision (Ref = Yes)	−0.34	0.35	0.92	.337	0.72	0.36	1.42
LSI-R total score	0.05	0.01	11.38	<.01	1.05	1.02	1.08
Comparison Group (Ref = Reentry)	0.80	0.20	15.70	<.01	2.22	1.50	3.29

Note. b = unstandardized regression coefficient; SE = standard error; HR = hazard ratio; CI = confidence interval; LSI-R = Level of Service Inventory–Revised; ; Ref= Reference.

Discussion

This reentry program evaluation builds on the existing and limited literature available that reports SVORI program outcomes. The quasi-experimental design of this evaluation used propensity score matching to identify a comparison group similar both demographically and in the risk of recidivism. Only the multisite evaluation conducted by Lattimore and Visher (2009), which also used propensity score matching, conducted a similar procedure to systemically identify and reduce observed differences between program participants and the comparison group. Furthermore, like the Lattimore and Visher (2009) study, this current evaluation yields mixed findings in recidivism outcomes. However, before a discussion of the results ensues, certain limitations that may have influenced these findings are detailed.

Limitations

In this evaluation, recidivism is narrowly defined as returns to prison and new convictions in the state where the reentry program was active; new convictions in another state and admissions to local jails or other state or federal prisons are not accounted for in the outcome measures. It is quite possible, perhaps likely, that a certain small percentage of the samples incurred new charges or returns to prison. Yet this potential bias in the recidivism data is equally present for all study participants, and it is unlikely that the event of recidivism in another state or in a federal jurisdiction would accrue to only one group and not to the other groups; that is, this data bias would likely affect all groups equally.

Some participants may have been exposed to interventions not identified and not included and controlled for in this analysis. These unknown interventions might have had an effect, positive or negative, on the results of this evaluation. However, there is no reason to believe that one group would have had more opportunity for such extraneous exposure than any other group.

It is well documented that the failure to use random assignment increases the chance that a systematic difference between groups biases the results. Indeed, in most of the reentry research published to date, this limitation exists. As the pressure to use more rigorous sampling methods extends to the prison environment, the consequences of it cannot be ignored. Empty seats in the reentry program classroom and unfilled spaces in the community reentry program—the bane of prison and rehabilitation managers and politicians alike—mean that an aggressive and effective campaign must be waged to get the message across that an empty chair or empty space is not the same as a wasted effort. Although prisons across the country have been forced to make dramatic fiscal cuts to their programs in light of the Great Recession, it is clear that the use of more rigorous research methods ultimately leads to greater knowledge of what works and what does not. Until this patient approach to knowledge building is fully accepted, findings such as those presented here will be limited.

Finally, incarcerated persons interested in participating in reentry programming are conceivably both likely and motivated to volunteer to take part in programming, and that interest alone may skew the findings. Yet the findings of this research suggest that if personal motivation or some other force is at play in the group of reentry participants and not in the other sample groups, it is not evident in the mixed outcomes reported here.

Discussion of the Salient Findings

Returns to prison

As a whole, members of the comparison group were found to have a slightly higher rate of returning to prison (37.0%) than the Reentry group (35.3%). However, in Cox proportional hazard modeling, after adjusting for age, gender, race, and LSI-R total score, the hazard of returning to prison was statistically similar between the two groups. In fact, the demographic and risk factors of being female, White, not released to community supervision, and being assessed as low-risk on the LSI-R were the most significant predictors of reincarceration and beg additional investigation, particularly as the female incarceration rate continues its unrelenting rise.

Surprisingly, age was found to only marginally affect the hazard of returning to prison. This weak relationship might be explained by the advanced age of the average participant (Reentry = 37.4 years, comparison = 38.3 years) when seen in the context of the often discussed age–crime curve (see Hirschi & Gottfredson, 1983) that suggests that as chronological age increases, criminal activity decreases.

New convictions

In contrast to the findings on reincarceration, participants in the Reentry group performed significantly better on the new convictions outcome than did the comparison group. At the bivariate level, the difference was quite significant with the comparison group (17.8%) incurring a new conviction at more than twice the rate of the Reentry group (7.7%). This significant difference between groups still held even after adjusting for the variables in Cox proportional hazards modeling. Beyond the impact of the Reentry program, and again worthy of further investigation, only being female and having a low LSI-R score significantly reduced the hazard of a new conviction.

Why do reentry program participants have better outcomes on the new conviction measure? In truth, we do not really know. It may be that their returns to prison interrupt a cycle of criminal activity. It may also be that the reentry programming received in fact had some specific impact on criminal thinking and behaviors, but which part or parts of that programming may have had a beneficial effect remains unknown. It also may be that the criminal behavior that lands reentry program participants back in prison will never exceed the seriousness of the behavior that results in a violation of the terms of their post-release supervision, but this cannot be determined without more finely detailed information about the nature of the parole officer’s decisions and the decision-making process. These questions require further exploration, and the answers to them will guide the development and implementation of helpful interventions.

Conclusion

Looking at all the data and the outcomes together, one principle pathway back to prison is revealed. Reentry group participants were found to incur significantly fewer new convictions than participants in the comparison group but, at the same time, returned to prison at a statistically similar rate, most likely as a result of technical violations of community supervision. Thus, for example, a parolee may be returned to prison for not having a job. Although it is not a crime to be unemployed, it becomes a detainable offense if in the judgment of the parole officer, the parolee’s lack of work puts him or her and society at risk. However, the risk assessed is not necessarily based on the individual offender; rather, popular actuarial data such as that on which the LSI-R relies tell us there is an elevated risk of criminal offending for any and all offenders who are unemployed.

Juxtapose the return to prison data with the new convictions findings of this study and we note that Reentry participants return to prison for new convictions at a significantly lower rate than do members of the comparison group. This is important. Considered with the returns to prison findings, the following questions must be asked: Is the parole system the proper place to apply broad actuarial predictions? To what extent do the pre-emptive decisions of parole officers pre-empt something other than criminal behavior? Already underway in the country is a movement to reduce the number of parolees returned to prison for violations of the terms of their community supervision (Lawrence, 2008). Given this major pathway to reincarceration, more research is needed to help state correctional agencies understand when interdiction by incapacitation is necessary for public safety and when it is not, in the context of a prisoner reentry program.

Furthermore, and perhaps more important, the following questions must be asked: Why do certain persons never return to prison—on violations of post-release supervision or for new convictions? What strengths, resources, abilities, and supports might this group of offenders have that others do not? These important queries go far beyond the scope of a program evaluation and remind us of the important finding that most people—no matter what programming they received or did not receive—do not return to (one Midwestern state’s) prison as a result of new convictions. Furthermore, a certain percentage of people (60%-80%) never return to prison for technical violations of the terms of community supervision. Whether the outcomes of rehabilitation programming or the products of personal and social supports, these findings are important to more fully understand.

At the conclusion of this study, there is one major unanswered question: Could the SVORI participants in the current study have been kept out of prison while maintaining their low rate of new convictions? In other words, what are the alternatives to incarceration for technical violations of parole? The intersection between prisoner reentry programs with their rehabilitation focus and parole programs with their maintain-the-status quo supervision focus remains an underexplored domain in the criminal justice evaluation literature. These parallel systems may have conflicting goals that complicate the process of successful community reentry for individuals. Further inquiry is needed to determine how reentry programs and parole agents operate in support or hindrance of the long-term reintegration of ex-offenders.

Footnotes

Authors’ Note

Points of view in this document are those of the authors and do not necessarily represent the official position or policies of the state in which the research occurred. This manuscript has not been published, nor has it been submitted for publication elsewhere.

Declaration of Conflicting Interests

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

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The research referenced in this article was funded by state general funds with Principal Investigator Margaret Severson.

Author Biographies

Christopher A. Veeh is a doctoral candidate at the University of Denver Graduate School of Social Work. His research focuses on prisoner reentry programs and the positive development of young adults.

Margaret E. Severson is a professor at School of Social Welfare, The University of Kansas. In addition to evaluations of prisoner reentry and related rehabilitation programs, her research focuses on the interface of prisoners’ mental and physical health conditions and the social, civil, environmental, and justice responses to them.

Jaehoon Lee is a research associate professor of Educational Psychology and Leadership and Associate Director of the Institute for Measurement, Methodology, Analysis and Policy at Texas Tech University. His research focuses on latent variable modeling, multilevel modeling, item response theory, and Bayesian statistics.

References

Andrews

D. A.

Bonta

(2010). The psychology of criminal conduct (5th ed.). New Providence, NJ: Matthew Bender & Company/LexisNexis.

Austin

P. C.

(2010). Optimal caliper widths for propensity score matching when estimating differences in means and differences in proportions in observational studies. Pharmaceutical Statistics, 10, 150-161.

Bouffard

J. A.

Bergeron

L. E.

(2007). Reentry works: The implementation and effectiveness of a serious and violent offender reentry initiative. Journal of Offender Rehabilitation, 44(2/3), 1-29.

Bowers

Fredrickson

Hansen

(2013). Package “RItools.” Retrieved from http://cran.r-project.org/web/packages/RItools/RItools.pdf

D’Agostino

R. B.

Jr. (1998). Propensity score methods for bias reduction in the comparison of a treatment to a non-randomized control group. Statistics in Medicine, 17, 2265-2281.

Duwe

King

(2012). Can faith-based correctional programs work? An outcome evaluation of the innerchange freedom initiative in Minnesota. International Journal of Offender Therapy and Comparative Criminology, 57, 813-841.

Flores

A. W.

Lowenkamp

C. T.

Holsinger

A. M.

Latessa

E. J.

(2006). Predicting outcome with the Level of Service Inventory–Revised: The importance of implementation integrity. Journal of Criminal Justice, 34, 523-529.

Guo

Fraser

M. K.

(2010). Propensity score analysis: Statistical methods and applications. London, England: SAGE.

Haviland

Nagin

D. S.

Rosenbaum

P. R.

(2007). Combining propensity score matching and group-based trajectory analysis in an observational study. Psychological Methods, 12, 247-267.

10.

Hirschi

Gottfredson

(1983). Age and the explanation of crime. American Journal of Sociology, 89, 552-584.

11.

Lattimore

P. K.

Brumbaugh

Visher

Lindquist

Winterfield

Salas

Zweig

(2004). National portrait of SVORI: Serious and violent offender reentry initiative. Retrieved from http://www.rti.org/pubs/nationalportraitofsvori2004.pdf

12.

Lattimore

P. K.

Steffey

D. M.

Visher

C. A.

(2010). Prisoner reentry in the first decade of the twenty-first century. Victims and Offenders, 5, 253-267.

13.

Lattimore

P. K.

Visher

C. A.

(2009). The multi-site evaluation of SVORI: Summary and synthesis. Retrieved from http://www.urban.org/UploadedPDF/412075_evaluation_svori.pdf

14.

Lawrence

(2008). Probation and parole responses: State responses. Washington, DC: National Conference of State Legislatures.

15.

Lipsey

M. W.

Derzon

J. H.

(1998). Predictors of violent and serious delinquency in adolescence and early adulthood: A synthesis of longitudinal research. In Loeber

Farrington

D. P.

(Eds.), Serious and violent juvenile offenders: Risk factors and successful intervention (pp. 86-105). Thousand Oaks, CA: SAGE.

16.

Lunt

(2013). Selecting an appropriate caliper can be essential for achieving good balance with propensity score matching. American Journal of Epidemiology, 179, 226-235.

17.

Reno

(2000, February 10). Remarks of the Honorable Janet Reno Attorney General of the United States on reentry county initiative. Retrieved from http://www.justice.gov/archive/ag/speeches/2000/doc2.htm

18.

Rosenbaum

P. R.

(2010). Design sensitivity and efficiency in observational Studies. Journal of the American Statistical Association, 105, 692-702.

19.

Sample

L. L.

Spohn

(2008). Final report for the evaluation of Nebraska’s serious and violent offender reentry program. Omaha: University of Nebraska, Omaha.

20.

Second Chance Act of 2007: Community Safety through Recidivism Prevention, Pub. L. No. 110-199, 122 Stat. 657 (2008). Retrieved from http://www.gpo.gov/fdsys/pkg/PLAW-110publ199/html/PLAW-110publ199.htm

21.

Severson

M. E.

Bruns

Veeh

Lee

(2011). Prisoner reentry programming: Who recidivates and when? Journal of Offender Rehabilitation, 50(6), 327-348.

22.

Shadish

W. R.

Cook

T. D.

Campbell

D. T.

(2002). Experimental and quasi-experimental designs for generalized causal inference. Belmont, CA: Wadsworth.

23.

Tabachnick

B. G.

Fidell

L. S.

(2013). Using multivariate statistics (6th ed.). Needham Height, MA: Allyn & Bacon.

24.

Veysey

B. M.

Ostermann

Lanterman

J. L.

(2014). The effectiveness of enhanced parole supervision and community services: New Jersey’s Serious and Violent Offender Reentry Initiative. The Prison Journal, 94, 435-453.