Article Evaluation Review 2013, Vol. 37(3-4) 274-313 ª The Author(s) 2014 Reprints and permission: sagepub.com/journalsPermissions.nav DOI: 10.1177/0193841X13519105 erx.sagepub.com

The Impact of Prison Reentry Services on Short-Term Outcomes: Evidence From a Multisite Evaluation

Pamela K. Lattimore1 and Christy A. Visher2

Abstract Background: Renewed interest in prisoner rehabilitation to improve postrelease outcomes occurred in the 1990s, as policy makers reacted to burgeoning prison populations with calls to facilitate community reintegration and reduce recidivism. In 2003, the Federal government funded grants to implement locally designed reentry programs. Adult programs in 12 states were studied to determine the effects of the reentry programs on multiple outcomes. Research design: A two-stage matching procedure was used to examine the effectiveness of 12 reentry programs for adult males. In the first stage, ‘‘intact group matching’’ was used to identify comparison populations that were similar to program participants. In the second stage, propensity score matching was used to adjust for remaining differences between groups. Propensity score weighted logistic regression was

1 2

RTI International, Research Triangle Park, NC, USA University of Delaware, Newark, DE, USA

Corresponding Author: Pamela K. Lattimore, RTI International, 3040 Cornwallis Rd, Research Triangle Park, NC 27709, USA. Email: [email protected]

Lattimore and Visher

275

used to examine the impact of reentry program participation on multiple outcomes measured 3 months after release. Subjects and data: The study population was 1,697 adult males released from prisons in 2004–2005. Data consisted of interview data gathered 30 days prior to release and approximately 3 months following release, supplemented by administrative data from state departments of correction and the National Crime Information Center. Results and conclusions: Results suggest programs increased in-prison service receipt and produced modest positive outcomes across multiple domains (employment, housing, and substance use) 3 months after release. Although program participants reported fewer crimes, differences in postrelease arrest and reincarceration were not statistically significant. Incomplete implementation and service receipt by comparison group members may have resulted in insufficient statistical power to identify stronger treatment effects. Keywords reentry, recidivism, SVORI, drug use, service receipt, two-stage matching Nationwide, more than half of the individuals who are released from prison are reincarcerated within 3 years (Langan and Levin 2002). Programs and services for men and women leaving prison are designed to stop this revolving door and encourage individuals to desist from offending. Imprisonment without such preparation for community reintegration may reduce human capital and impede the acquisition of prosocial skills and behaviors, thus lessening the probability of a successful transition from prison to the community (Visher and Travis 2003; Western 2007). However, in comparison to 20 years ago, men and women leaving prison are less prepared for reintegration, less connected to community-based social structures, and more likely to have health or substance abuse problems than prior cohorts (Lynch and Sabol 2001; Petersilia 2005). Until recently, the majority of rehabilitation and reentry strategies were dominated by service providers who represented a single domain from among the possible correlates of desistance. However, the needs of individuals returning to the community usually span multiple problem domains suggesting that individual offenders often require more than a single program or intervention. To address this dilemma, many reentry specialists are encouraging a broader focus on comprehensive reentry strategies, not specific programs (Lattimore 2007; National Research Council 2007; Re-entry Policy Council 2005; Visher 2007). Such strategies involve multiple levels of government, coordination of efforts across agencies, and involvement of

276

Evaluation Review 37(3-4)

organizations that are traditionally not part of the reentry discussion (e.g., public health, local businesses, and community colleges). Moreover, these coordinated efforts may improve reintegration across a broader range of outcomes (e.g., employment, substance use, health) than simply reductions in recidivism. In an effort to improve the transition experiences of men and women exiting prison, the Federal government funded the Serious and Violent Offender Reentry Initiative (SVORI). The Initiative was broadly conceived to help states better use their correctional resources to address a variety of outcomes for former prisoners, including employment, education, health, housing, and recidivism. SVORI grantees were required to establish and support a partnership between institutional and community agencies and to create a three-phase continuum of services that began in prison, moved to a structured reentry phase before and during the early months of release, and was to continue for 1 or more years as released prisoners became reestablished in the community. Although conceptually straightforward, this model was far from ‘‘business as usual’’ as it required state and local agencies to collaborate in ways that had been rare in the past. In this article, we address the issue of whether a broadly conceived reentry initiative such as SVORI, which aimed to provide a wide range of services to address prisoners’ needs, was effective in delivering enhanced services in multiple domains and identify the effects of program participation on outcomes during the critical, initial postrelease period. We next briefly review the research that provided the context within which SVORI was implemented. Second, we describe SVORI and our evaluation methodology. Third, we describe our subjects and present our findings concerning the delivery of services to adult male SVORI participants and a comparison sample. We then examine the impact of SVORI on a broad array of reentry outcomes. We conclude with a discussion about the implications of our findings for the development and implementation of reentry programs and strategies.

Reentry Research Context and the SVORI Over many decades, the overwhelming majority of evaluations of rehabilitative programs for offenders focused on programs designed to address specific individual needs, such as reducing drug and alcohol use, addressing mental health issues, or finding a job. But, more recently, reentry services and programs have focused on easing the transition of individuals exiting prison, addressing multiple needs with individualized approaches. Many,

Lattimore and Visher

277

if not most reentry programs for individuals exiting prison are relatively new, having begun as a result of federal funding in the early years of the 21st century. As a result, there are few impact evaluations of programs focused specifically on reentry (Petersilia 2004). Numerous challenges characterize the extant research assessing the effectiveness of programs for formerly incarcerated individuals, whether focused on reentry or general rehabilitation. Foremost among the challenges is the lack of theoretical models that articulate behavior change among former prisoners. Within any particular substantive area, there are also problems of fidelity in that a particular service approach may manifest itself in different ways under different programs and circumstances. As a result, it is often difficult to generalize research findings from one program to others, and substantial variability exists among the outcome variables examined (e.g., employment, homelessness, substance use). The numerous combinations of program types unique to each study also render comparisons difficult. Finally, there are problems related to the research itself, as rigorous experimental designs—including the use of comparison groups (randomly assigned or otherwise)—are rare in this research literature (National Research Council 2007). Several recent reviews of reentry program evaluations have examined the available research on what works with regard to reentry and/or rehabilitative programming (Aos, Miller, and Drake 2006; Gaes et al. 1999; Lipsey and Cullen 2007; MacKenzie 2006; Petersilia 2004; Seiter and Kadela 2003). The evidence has been consistent in establishing that contact-driven supervision, surveillance, and enforcement of supervision conditions have a limited ability to change offender behavior or to reduce the likelihood of recidivism (MacKenzie 2006; White 2006). However, intensive supervision programs with a clear treatment component show a sizable impact on recidivism (Aos, Miller, and Drake 2006; Gaes et al. 1999; Petersilia 2004). In community corrections, there is a solid research base suggesting what works at the individual level. MacKenzie (2006) recently summarized the ‘‘what works’’ literature in corrections, with specific chapters on various community corrections programs (e.g., life skills, cognitive behavioral therapy, education, drug treatment, and intensive supervision). She concluded that human service-oriented programs were much more effective than those based on a control or deterrent philosophy. In particular, there is growing consensus that practices focusing on individual-level change, including cognitive change, education, and drug treatment, are likely to be more effective than other strategies, such as programs that increase opportunities for work, reunite families, and provide housing (see also Andrews and

278

Evaluation Review 37(3-4)

Bonta 2006). All of the strategies MacKenzie (2006) identified as effective focus on dynamic criminogenic factors, are skill oriented, are based on cognitive/behavioral models, and treat multiple offender deficits simultaneously. These conclusions are consistent with several large meta-analyses of the corrections evaluation literature (Aos, Miller, and Drake 2006; Andrews et al. 1990; Lipsey and Cullen 2007). Nonetheless, selection of program type may be less important than proper implementation of the program. Delivering a program in the wrong context (e.g., intensive substance abuse treatment to casual drug users) or poor implementation is common and may explain most of the weak or null findings in the research studies. Despite advances in knowledge and best practices, studies of programs for offenders have documented persistent problems in implementation and adherence to the fidelity of evidencebased practice models (Farabee 2005; Lowenkamp, Latessa, and Smith 2006; Petersilia 2004). One recent reentry program, Project Greenlight, was developed from research and best practice models to create an evidencebased reentry initiative which was evaluated with a random assignment research design (Wilson and Davis 2006). However, the program participants performed significantly worse than a comparison group on multiple measures of recidivism after 1 year and the evaluators concluded that the New York program did not replicate past best practice. Instead, Project Greenlight modified past practice to fit institutional requirements, was delivered ineffectively, did not match individual needs to services, and failed to implement any postrelease continuation of services and support (Wilson and Davis 2006; see also Rhine, Mawhorr, and Parks 2006; Visher 2006; Marlowe 2006). The evaluators attributed the findings to a combination of implementation difficulties, program design, and a mismatch between participant needs and program content. Another line of research has focused on identifying the principles of effective treatment (as opposed to the substantive content of the program) in assessing evidence-based practices (e.g., Andrews and Bonta 2006; Cullen and Gendreau 2000; Gendreau et al. 1996). MacKenzie (2006) summarizes this work into five principles of effective rehabilitation strategies: having strong program integrity, identifying criminogenic factors, employing a multimodal treatment approach, using an actuarial risk classification, and ensuring responsivity between an offender’s learning style and mode of program delivery. One of the failings of Project Greenlight was poor management of the program according to these principles that help guide or maximize program effectiveness (see Andrews 2006). In her review of what works in reentry programming, Petersilia (2004) discusses the striking

Lattimore and Visher

279

disconnect between the published ‘‘what works’’ literature and the efforts of governmental reentry task forces to develop programs that are thought to improve offender transitions from prison to the community. Finally, the goal of most reentry programs is to develop a seamless transition from prison to the community. However, the challenges in this regard are enormous. Corrections departments and community supervision agencies often have conflicting incentives, and community–justice partnerships linking these organizations with community groups face even larger hurdles. An important barrier to effective reentry strategies in many communities is the lack of information sharing between the criminal justice system and the community because of institutional barriers and privacy rules. Effective service delivery after release requires coordinated actions by government agencies, nongovernment service providers, and the community to ensure that returning prisoners do not fall through service gaps between agencies. Yet, knowledge about how to develop and manage these partnerships is lacking (Rossman 2003). Thus, it was in this context of relatively little concrete scientific evidence about the components and implementation of effective, comprehensive reentry programs that states and communities were asked to develop local programs with funding from the Federal SVORI at the beginning of the 21st century.

The SVORI and the SVORI Evaluation The SVORI sought to help states better use their correctional resources to address a variety of outcomes, including housing, health, education, employment, substance use, and criminal behavior. In fiscal year 2003, SVORI grants averaging between US$500,000 and US$2,000,000 were awarded to 69 grantees in the 50 states, the District of Columbia, and the U.S. Virgin Islands. These grantees conceived a total of 89 individual SVORI programs including 52 adult programs and 37 juvenile programs (see Lattimore et al. 2004). In this article, we address only the adult programs and more specifically the programs for adult male offenders. The SVORI grantees were expected to partner with other government agencies and community and faith-based organizations to provide programs and services prior to and following release from incarceration or juvenile detention. While the SVORI programs shared the common goals of improving service coordination, systems collaboration, and individual outcomes across numerous dimensions, the Initiative focused on the outcomes to be achieved rather than prescribing specific services or programs. Consistent

280

Evaluation Review 37(3-4)

with the emerging research on principles of effective treatment, the services and programming received by SVORI program participants were to be based on risk and needs assessments administered first in the institution and then updated after release. In-prison treatment and postrelease reentry plans were to be tailored to the risk level and needs of each offender and managed by a case manager or transition team. Depending on risk and needs, treatment and reentry plans could contain multiple services including substance abuse treatment, mental health treatment, housing, educational services, batterer intervention programs, training to improve job skills, a restitution mechanism and restorative justice plan, aftercare programs (including peer support), counseling on avoidance of criminal behavior and behavior triggers, consequences for noncompliance with the plan, and involvement of family and other natural support systems. The focus of the evaluation was the enhanced reentry programs that emerged with SVORI funding. The federal guidance accompanying SVORI funding placed few restrictions on the state agencies with respect to the design of the individual SVORI programs. Because there was not a specified SVORI program model, each program was locally designed, and the programs varied in approach, services provided, and target populations. These programs were a combination of extant programs and services and new services. Although the specific content could and did vary from site to site, each locally designed program was supposed to improve housing, health, education, employment, substance use, and criminal behavior. In an effort to successfully describe and compare programs and to determine what worked in the event of positive treatment effects, the evaluation team collected extensive data from program participants and comparison subjects who participated in interviews on the types of services they received. These data and the variability in service receipt across sites are described after an overview of the evaluation design and data collection procedures.1

Evaluation Design and Data A multimethod evaluation protocol was implemented in 2003 (Lattimore and Steffey 2009; Lattimore et al. 2005; Winterfield and Lindquist 2006) that included a two-stage matching quasi-experimental design. Of critical importance for the evaluation team was developing a research design that would assure strong internal and external validity of the evaluation results. External validity was maximized by including multiple sites in the evaluation. Internal validity is often maximized by the use of an experimental design with random allocation of subjects into different conditions.

Lattimore and Visher

281

However, some criminal justice programs cannot be evaluated with random assignment designs (Lum and Yang 2005). In many sites, the SVORI programs were already underway by the time the evaluation was funded and approved to begin. Of the 12 programs chosen for the evaluation, 2 sites randomly assigned individuals to the SVORI program. In the remaining sites, a rigorous quasi-experimental design was developed which created treatment and comparison groups with a high degree of preintervention comparability. The two-step procedure entailed selecting matched comparison groups for each site based on site-specific criteria that essentially replicated the selection process for SVORI participants in each site (see Cook, Shadish, and Wong 2008; Cook and Steiner 2010). In a second step, propensity score models were used to further remove any differences between SVORI and comparison groups. The resultant groups showed a high degree of balance prior to propensity score weighting, which eliminated remaining differences between the groups on observed variables. This combination of methods to reduce selection bias in a quasi-experimental design is a unique feature of the evaluation of SVORI programs and may be useful to other evaluation researchers when random assignment is not feasible. Six criteria guided the selection of sites for the evaluation: (1) program had clearly defined elements and goals, (2) program was implemented (or was likely to be implemented), (3) program target population was accessible and of sufficient size, (4) an appropriate comparison population was available and accessible for inclusion in the study, (5) administrative data were of good quality and available for the evaluation, and (6) the program was amenable to and able to participate in the evaluation. These criteria were addressed following (1) review of SVORI grantee proposals and work plans and follow-up telephone interviews with program directors to obtain information not gleaned from the review and clarification and updates on the programs’ status; (2) visits to 39 programs; and (3) review and synthesis of all information to develop a list of recommended programs for inclusion in the impact evaluation that was submitted to National Institute of Justice for approval. At the conclusion of this process, 13 adult programs and 5 juvenile programs were tentatively selected for the evaluation; 1 adult program was subsequently dropped because of logistical issues with respect to interviewing the comparison group and 1 juvenile program was dropped because of extremely low case flow (see Lattimore and Steffey 2009 for additional information). Here we focus on the 12 final adult impact evaluation programs. There were two pathways to inclusion in SVORI: (1) random assignment to SVORI programming or standard programming after a decision to

282

Evaluation Review 37(3-4)

participate in SVORI was made by the offender and (2) program and offender determination of SVORI program eligibility and participation. For two sites, those randomly assigned to standard programming constituted the pool of potential comparison subjects. For the other 10 sites, a quasiexperimental design was developed in close collaboration with site personnel to identify a matched pool of individuals who met the individual-level program eligibility criteria (e.g., offense type, expected release date, age), but who were not included in SVORI. In many cases, these individuals met all local SVORI program eligibility criteria except for those related to where they were housed (i.e., in a facility offering SVORI) or where they were returning at release (i.e., to a community with a postrelease SVORI program). Potential comparison subjects were identified by management information system or research office staff based on algorithms developed to simulate SVORI program selection criteria (for additional details on the site-by-site creation of comparison groups, see Lattimore and Steffey 2009). As will be shown later, the rigorous identification of non-SVORI participants resulted in a high degree of preintervention comparability between the SVORI and non-SVORI groups. The evaluation took an ‘‘intent-to-treat’’ approach. Practically, this meant that an individual was classified as SVORI or non-SVORI depending upon whether he or she was enrolled in a SVORI program at any time during the period between when the site first provided the case information to the evaluation team and when the case was fielded. Every 1–2 weeks, contact individuals at the selected departments of corrections provided the evaluation team the names of potential respondents who were SVORI program participants or comparison subjects with anticipated release dates within the next 30–90 days. This information was used to field the cases. Data collection consisted of four waves of in-person interviews, oral swab drug testing in the community, and administrative arrest and incarceration data. Wave 1 interviews were conducted between July 2004 and November 2005 with 1,697 adult male prisoners. A total of 2,564 potential interviews with eligible adult male prisoners were fielded. Of these, interviews were completed with 1,697 or 66.2% of the pool of eligible respondents. The primary reason for no interview was that the individual was released before an interview could be scheduled and completed, which occurred for 538 (21.0%) of the cases. This was particularly true during the initial enrollment period; adjustments to the identification process in which we extended the expected time to release reduced this problem. There was no evidence that this ‘‘early release,’’ which amounted usually to just a week or two, was related to SVORI program participation. Refusals among

Lattimore and Visher

283

those approached for interviews comprised 11.5% (N ¼ 295). Interviews were not conducted for the remaining 34 eligible respondents for a variety of reasons including denied access by prison (N ¼ 14), the potential respondent had absconded (N ¼ 5), and language barrier (N ¼ 7). One potential respondent died before the Wave 1 interview could be conducted. The Wave 1 interviews were conducted about 30 days prior to the respondents’ expected release dates to collect information on preincarceration experiences and history, as well as information on in-prison service receipt. Follow-up interviews were conducted at 3 time points following (the initial) release from prison—Wave 2 interviews were conducted 3 months postrelease; Wave 3, 9 months postrelease; and Wave 4, 15 months postrelease. All interviews were conducted by trained interviewers using computer-assisted interviewing technology. Prerelease interviews were conducted in prison facilities in areas where responses could not be seen or heard by anyone other than the interviewer. The postrelease interviews were conducted in the community or in jails or prisons in cases where subjects were reincarcerated. Subjects were not compensated for the Wave 1 interviews; nonincarcerated subjects were compensated for Wave 2, 3, and 4 interviews.2 State agencies provided data on return to prison after being released. The National Crime Information Center (NCIC) provided data on arrests, as well as information on convictions and reincarcerations for some states.3 The analyses in this article focus only on Wave 1 (in-prison) service receipt and outcomes measured shortly after release from prison, that is, Wave 2 findings. The impact of services received after release from prison will be examined in future analyses. Wave 2 interviews were conducted with 984 of the 1,697 adult males (58%) who completed a Wave 1 interview (6.2% of the SVORI program participants and 8.4% of the non-SVORI respondents were interviewed in prison or jail).4 Arrest data were obtained for 1,581 of the 1,697 adult males (93%) and reincarceration data were available for all 1,697 men. We discuss issues related to potential selection and attrition bias following a description of the study subjects.

Subjects Table 1 compares the characteristics of the adult male subjects on a variety of preincarceration and current measures. As can be seen, the first-stage matching resulted in two groups that are similar on most but not all measures.

284

Evaluation Review 37(3-4)

Table 1. Characteristics of Evaluation Sample at Wave 1 Interviews, Means (Standard Deviations) and t-Statistics. Variable Demographic characteristics Age at interview White* Black* Other race Homeless/shelter/no set place to live before incarceration Completed 12th grade or GED/other HS equivalent Married Are currently in steady relationship Have living children Employment Ever held a job* Employed during 6 months before incarceration* Hours per week last job Hourly salary Last job permanent Received formal pay last job Health insurance last job Expect to return to a previous job Family and peers Anyone in family ever convicted of a crime Anyone in family ever in a correctional facility Anyone in family problems w/drugs/alcohol Friends before incarceration, any convicted Friends before incarceration, any correctional facility Friends before incarceration, any drug/ alcohol problems Alcohol and other drug use Any victimization (6 months before incarceration) Age first time you drank alcohol (years) Used alcohol 30 days before incarceration Ever used drugs during lifetime Used drugs 30 days before incarceration

N

SVORI

1,697 28.89 (7.14) 1,697 0.31 (0.46) 1,697 0.56 (0.50) 1,697 0.08 (0.27) 1,695 0.12 (0.33)

Non-SVORI 29.30 (7.48) 0.37 (0.48) 0.50 (0.50) 0.09 (0.29) 0.12 (0.33)

1,695

0.61 (0.49)

0.58 (0.49)

1,696 1,677 1,684

0.09 (0.28) 0.38 (0.49) 0.62 (0.49)

0.10 (0.30) 0.39 (0.49) 0.64 (0.48)

1,696 1,696

0.89 (0.31) 0.64 (0.48)

0.92 (0.27) 0.68 (0.47)

1,107 41.72 (13.86) 41.76 (14.07) 1,083 10.91 (8.51) 10.13 (6.87) 1,117 0.75 (0.43) 0.73 (0.44) 1,120 0.74 (0.44) 0.72 (0.45) 1,094 0.37 (0.48) 0.34 (0.47) 1,483 0.57 (0.50) 0.57 (0.50) 1,574 1,602

0.75 (0.43) 0.75 (0.44)

0.76 (0.43) 0.74 (0.44)

1,591 1,540 1,556

0.72 (0.45) 0.83 (0.37) 0.81 (0.39)

0.74 (0.44) 0.83 (0.37) 0.81 (0.39)

1,572

0.82 (0.39)

0.83 (0.38)

1,696

0.59 (0.49)

0.58 (0.49)

1,616 13.71 (3.85) 1,697 0.68 (0.47) 1,697 0.94 (0.24) 1,697 0.66 (0.48)

13.64 (3.76) 0.67 (0.47) 0.96 (0.21) 0.69 (0.46) (continued)

Lattimore and Visher

285

Table 1. (continued) Variable Ever used marijuana Ever used hallucinogens Ever used cocaine* Ever used heroin* Ever used amphetamines Before incarceration, ever received treatment for AOD Criminal history Duration of incarceration at baseline (years)* Currently serving time for parole violation?* Ever jail/prison more than 24 hr at one time* Conviction offense: person/violent crime Conviction offense: property crime Conviction offense: drug crime* Conviction offense: public order crime* Conviction offense: other crime How old the first time arrested (years) Times in your life arrested Times in your life convicted of a crime Ever juvenile detention facility for committing a crime Times in life been sent to prison*

N

SVORI

Non-SVORI

1,695 1,695 1,694 1,695 1,692 1,696

0.92 0.43 0.53 0.18 0.26 0.42

(0.27) (0.50) (0.50) (0.38) (0.44) (0.49)

0.94 (0.24) 0.49 (0.50) 0.58 (0.49) 0.23 (0.42) 0.30 (0.46) 0.41 (0.49)

1,697

2.76 (2.46)

2.26 (2.63)

1,694

0.23 (0.42)

0.31 (0.46)

1,694

0.83 (0.38)

0.87 (0.33)

1,687 0.41 (0.49) 0.38 (0.49) 1,687 0.23 (0.42) 0.26 (0.44) 1,687 0.36 (0.48) 0.31 (0.46) 1,687 0.17 (0.37) 0.24 (0.43) 1,687 0.02 (0.14) 0.01 (0.12) 1,679 15.95 (4.75) 16.07 (5.07) 1,584 12.43 (11.46) 13.14 (11.39) 1,658 5.48 (6.05) 5.70 (6.26) 1,696 0.51 (0.50) 0.49 (0.50) 1,434

1.45 (1.82)

1.69 (2.05)

Note. SVORI ¼ Serious and Violent Offender Reentry Initiative; GED ¼ general educational development; HS ¼ high school; AOD ¼ alcohol and other drug. *p < .05, two-tailed test.

Demographic Characteristics The subjects in both groups were about 29 years of age. SVORI program participants were less likely than non-SVORI to be White (31% vs. 37%, respectively) and more likely to be African American (56% vs. 50%, respectively). Members of the two groups were equally likely to have been homeless prior to their current incarceration and about 60% reported that they had completed the 12th grade or obtained a general education diploma (GED). On average, the two groups were similar in terms of reporting currently being married (about 10%) or in a steady relationship (38%) and slightly more than 60% reported that they had children. Most reported

286

Evaluation Review 37(3-4)

having worked at some point, with 92% of the non-SVORI and 89% of the SVORI subjects reporting having held a job during their lifetimes (t ¼ 2.2).

Family and Peers Most members of both groups reported having family members and peers who had been convicted of crimes and incarcerated and who had problems with alcohol and drugs. More than 75% reported that someone else in their families had been convicted of a crime or incarcerated in a correctional facility; more than 80% reported that prior to this incarceration they had friends who had been convicted and incarcerated.

Alcohol and Other Drug Use Nearly all of the respondents reported having drunk alcoholic beverages (96% of the SVORI participants and 97% of the non-SVORI subjects) and to have had their first drink at about 13.7 years of age. About two thirds reported that they had a drink during the 30 days prior to their current incarceration. Self-reported drug use was also very high among these two groups, with about 95% of the respondents reporting that they had used illegal drugs in their lifetimes and about two thirds reporting that they had used drugs in the 30 days prior to their current incarceration. More than 40% of each group reported that they had received alcohol or other drug treatment prior to the current incarceration.

Criminal History Members of the two groups differed on several measures of criminal history. Members of the non-SVORI group had served, on average, about 6 months less at the time the Wave 1 interview was conducted (2.26 years vs. 2.76 years). The non-SVORI subjects were also much more likely to report that they were serving time for a parole violation (31% vs. 23%) and that their conviction offense/offenses included a public order crime (which includes probation and parole violations). The non-SVORI subjects also were slightly more likely than SVORI participants (87% vs. 83%) to report having served a prison/jail sentence other than the current incarceration. The two groups were similar on a variety of other measures of criminal history.

Lattimore and Visher

287

Method Following the preintervention first-stage group matching, a propensity score approach was used to address potential selection bias due to the quasi-experimental design (see Rubin 2006, for a collection of seminal articles in propensity score modeling; see D’Agostino, 1998, for an accessible tutorial). Propensity scoring methods are not without limitations. For example, use of propensity scores can only adjust for included covariates (Glynn, Schneeweiss, and Sturmer 2006; Rosenbaum and Rubin 1983). However, if many of the covariates believed to be related to treatment assignment are measured, propensity score approaches should yield consistent and approximately unbiased estimates of treatment effects (D’Agostino 1998; Rosenbaum and Rubin 1983). Although it is impossible to determine whether members of the SVORI and non-SVORI groups differed on unobserved variables, we were able to draw upon an extensive set of data from the Wave 1 interview to estimate the propensity scores. These data included variables related to eligibility for program selection (e.g., proximity to release and measures of risk), as well as other measures of the backgrounds of evaluation participants, including (1) immutable characteristics (e.g., race), (2) preprison characteristics and behaviors (e.g., drug and alcohol use), and (3) lifetime experiences (e.g., ever treated for substance abuse or mental health problems; currently have a GED or high school diploma). In developing our propensity model, we excluded measures that could reflect differences that could have been due to program assignment (e.g., have GED or high school diploma as a result of programs completed in prison following SVORI program assignment). Table 2 shows the t-statistics and standardized differences between the unweighted SVORI and non-SVORI group means (or proportions) for the variables included in the propensity models. As can be seen, there are few statistically significant differences (p < .05, two-tailed test). Rosenbaum and Rubin (1985, 34) suggest examining standardized differences as an alternative to t-statistics when assessing balance between study groups. The standardized difference is defined as ð x1  x2 Þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ðs21 þ s22 Þ=2 Although we observe statistically significant differences for 8 of these 24 variables, none of the standardized differences is larger than 0.20, suggesting good balance between the two groups prior to adjustment.

288

Evaluation Review 37(3-4)

Table 2. Initial Balance Check. Variable Age at incarceration Race_White Race_Black (reference category) Race_other Homeless prior to prison Employed 6 months prior to prison Steady relationship 6 months prior to prison AOD treatment prior to incarceration Mental health treatment prior to prison Victimized 6 months prior to prison Violence perpetration 6 months prior to prison Alcohol use 30 days prior to prison Marijuana use 30 days prior to prison Other drug use 30 days prior to prison Person offense Property offense Drug offense Public order or other offense Parole violator Age first arrest Arrest rate prior to prison Conviction rate prior to prison Number of juvenile detentions Incarceration rate prior to prison

t-Statistic

Standardized Difference

Number of Missing Values

2.57* 2.30* 2.74* 0.83 0.12 2.04* 0.28

0.12 0.11 0.13 0.04 0.01 0.10 0.01

0 3 3 3 2 1 4

0.42

0.02

1

0.52

0.02

4

0.61 0.80

0.03 0.04

1 0

0.43 0.76

0.02 0.04

4 3

1.92

0.09

1

0.92 1.35 2.36* 2.58* 3.18* 0.47 0.50 0.25 0.86 3.28*

0.04 0.07 0.11 0.13 0.16 0.02 0.02 0.01 0.04 0.16

9 9 9 9 2 12 113 39 17 9

Note. AOD ¼ alcohol and other drug. t-Statistics and standardized differences from the comparison of unweighted mean values for Serious and Violent Offender Reentry Initiative (SVORI) and non-SVORI groups for selected variables from the Wave 1 interview. *p < .05, two-tailed test.

Propensity Score Estimation and Missing Data Although item missingness was relatively rare in the data, imputation procedures were employed so that no observations had to be dropped from the

Lattimore and Visher

289

outcome analyses because of missing propensity scores. The adult male sample included 1,697 observations, 1,500 (88.4%) of which had no missing values on any of the variables. Table 2 lists the variables that were included in the propensity score models. Also shown are the numbers of missing values for each of the included variables. Logit models to generate the probability of assignment to SVORI were estimated within the framework of SAS 9.1.3 PROC MI and PROC MIANALYZE (see Lattimore and Steffey 2009, for details). These SAS procedures accommodated item missingness by imputing values for missing data (Allison 2001; SAS Institute 2004). The multiple imputations (MI) and MI ANALYZE procedures generated five data sets—each data set included different estimates for the missing values—then used logistic regression to generate parameter estimates for each data set. In traditional applications of imputation with logistic regression, the five sets of parameter estimates are used to produce a single set of parameters that are then reported to reflect the impact of the independent variables on the dependent, outcome variable. For the propensity score model that is our objective, however, the parameter values themselves are not of interest—what is of interest are the propensity scores (or p-hats or p^) that are generated by applying the parameters to the data. Each set of parameter estimates was applied to the data to generate five p-hat values for each individual. A review of the literature identified no applications in which missing value imputation was used in the production of propensity scores. After a thorough discussion among the analysis team and our expert panel, it was determined that the average of the five p^ estimates was an appropriate value to use. Therefore, we averaged the five values to generate the final p^ that were used in the outcome analyses.

Selection Bias The propensity scores were used to develop weights that were used to examine the population average treatment effect (PATE) for the outcome models. The PATE is the average treatment effect one would expect if the entire population were treated.5 The PATE weights were calculated as follows: If subject i was a SVORI participant, wi ¼ or else

1 ; p^i

290

Evaluation Review 37(3-4)

wi ¼

1 : 1  p^i

These weights were applied to the data to assess whether the propensity score resulted in improved balance on the variables included in the model. To assess balance in the weighted data, PROC Survey Logistic in SAS1 9.1 was used to regress the SVORI indicator on each of the variables included in the model. Results of these procedures showed that in all cases the Wald’s chi-square test statistics were effectively zero, ranging between 0.0000 and 0.0080, for all of the significance tests, suggesting that the propensity score weights generated good balance for the Wave 1 data. Balance results for the Waves 2, 3, and 4 data sets were similar—none of the coefficients differed significantly from zero, suggesting balance between the SVORI and non-SVORI groups at each wave of data collection.

Attrition Bias The primary validity threat of attrition is that, if the people who complete follow-up interviews differ from the people who do not complete followups, the population to whom findings can be generalized is no longer known. This problem is particularly serious if there is evidence that there is differential attrition from the study groups that could be correlated with outcomes. Throughout the evaluation, various approaches were used to investigate attrition. Informally, differences between groups at successive follow-up waves were similar to those observed at Wave 1, which suggested that attrition was either random or similarly affecting the SVORI and nonSVORI groups. Additionally, as noted previously, balance between the SVORI and the non-SVORI subjects after propensity score adjustment was observed at each wave. The issue of attrition bias then was addressed from the perspective of whether nonresponse was an issue after controlling for selection into SVORI (see Imai, King, and Stuart 2008; also Fitzmaurice and Laird 2000). The following equation was used to examine the pattern of nonresponse for the adult male subjects. Y ¼ b0 þ b1  S þ b1  pðRÞ þ b1  S  pðRÞ; where S equals 1 if the individual is a SVORI participant, 0 otherwise; p(R) is the estimated likelihood of having a follow-up for each observation (based on a logistic regression propensity score model); and S*p(R) is an

Lattimore and Visher

291

interaction term. The dependent variable was coded to indicate whether a subject was a completer (completed all follow-up interviews), an attriter (did not complete successive interviews), or missing (had prior and after responses).6 If the main effect for SVORI and the interaction term are not significant, the findings suggest that there is no differential attrition conditional on SVORI. The model was estimated as a nonordered multinomial logistic regression using SAS1 9.1.3. Once the main effect of response was controlled, SVORI program participation was not related to whether a response was obtained (i.e., none of the parameter estimates for SVORI or the interaction term was statistically significant at any usually accepted level). Given that the propensity score weights generated good balance between the SVORI and non-SVORI groups on data at each wave for all three demographic groups, the determination was made that it was not necessary to control for nonresponse in addition to SVORI program participation because no differential attrition was identified between the two groups. Of course, as is always the case, this determination is related only to observables.

Testing for Program Effects Tests for program effects were conducted incorporating the PATE weights generated from the propensity scores as described previously. Mean values were generated using Proc Survey Means. Tests of significance of differences between SVORI participants and non-SVORI comparison subjects were conducted using Proc Survey Logistic to regress the SVORI indicator on each of the outcomes, as a two-sample test of difference between means is not available in SAS Proc Survey Means. The test statistic on the resulting parameter estimate was used to determine whether SVORI differed significantly from non-SVORI participants. We look first at the effect of SVORI program participation on service receipt and subsequently on outcomes.

Findings The logic model underlying the SVORI is that increases in service receipt will lead to improved outcomes, including improvements in recidivism. SVORI programs were intended to provide services that would address multiple outcome domains, including employment, education, health, housing, and recidivism. We look first at the effect of SVORI program participation on service receipt, measured at the Wave 1 interview, and subsequently on outcomes measured at Wave 2.

292

Evaluation Review 37(3-4)

Service Receipt SVORI program participants were to receive needs and risk assessments which would be used to develop a reentry plan that included services to be provided prior to and after release from prison. Thus, an important component of the evaluation was the careful measurement of reentry services received by participants and any ‘‘treatment as usual’’ received by the comparison respondents. In each interview, respondents were asked whether they had received a variety of services. Table 3 presents the results for respondents’ reports of service receipt during the current incarceration at the interview conducted about 30 days before release. As can be seen, with only a few exceptions, SVORI program participants were significantly more likely to report receiving each of these services.7 The findings also show, however, that in most cases fewer than half of the SVORI program participants reported having received a particular service during their current incarceration. Indeed, only about two thirds of SVORI program participants reported having received a needs assessment (63%), met with a case manager (66%), worked with anyone to prepare for release (66%) or taken a class specifically for release (65%), and only 57% reported having developed a reentry plan. SVORI participants were more likely to report having participated in programs to prepare for release (75% compared with 51%). Relatively few reported receiving specific transition services. Although 41% reported receiving help obtaining documents, only 22% reported receiving assistance getting a driver’s license. Similarly, only 28% reported receiving assistance finding a place to live and 19% finding transportation. There was no difference in receipt of medical or dental services (services required by law); non-SVORI respondents were more likely to report receiving mental health treatment, which was consistent with these respondents being more likely than SVORI program participants to report having received mental health treatment prior to incarceration (data not shown). SVORI program participants were more likely to report receiving substance use treatment (48%) and anger management programming (34%). Most of the SVORI program directors reported that employment was a primary focus of their reentry program (see Lattimore et al. 2004). However, only 37% of SVORI program participants reported receiving any of eight specific employment-related services.8 A somewhat larger percentage (53%) reported that they had received educational services such as GED classes and more than half (52%) reported receiving training to change their criminal behavior and attitudes (e.g., cognitive behavioral therapy). About 42% reported having received life skills training with fewer reporting

Lattimore and Visher

293

Table 3. Service Receipt at Prerelease Interview, Weighted Means (Standard Deviations) and t-Statistics.a Variable Coordination services Received needs assessment Received release-specific needs assessment Met with case manager Developed reentry plan Worked with anyone to plan for release Transition services Participated in programs to prepare for release Took class specifically for release Received legal assistance Received assistance accessing financial assistance Received assistance accessing public financial assistance Received assistance accessing public health care assistance Received mentoring services Received assistance obtaining documents Received assistance finding transportation Received assistance finding place to live Received assistance getting driver’s license Received assistance accessing clothing/food banks Health services Received any medical treatment Received dental services Received any mental health treatment Received any substance use treatment Participated in groups for victims of abuse Participated in anger management program

N

SVORI

NonSVORI

tStatistica

1,690 0.63 (0.48) 0.45 (0.50) 1,678 0.49 (0.50) 0.23 (0.42)

7.43* 11.61*

1,694 0.66 (0.47) 0.40 (0.49) 1,663 0.57 (0.50) 0.24 (0.43) 1,695 0.66 (0.48) 0.31 (0.46)

11.05* 14.69* 15.22*

1,696 0.75 (0.43) 0.51 (0.50)

10.64*

1,695 0.65 (0.48) 0.37 (0.48) 1,697 0.12 (0.32) 0.08 (0.27) 1,697 0.13 (0.34) 0.04 (0.19)

11.89* 2.38* 7.11*

1,696 0.14 (0.35) 0.11 (0.31)

1.81

1,695 0.13 (0.34) 0.09 (0.29)

2.46*

1,697 0.20 (0.40) 0.08 (0.27) 1,693 0.41 (0.49) 0.26 (0.44)

6.92* 6.66*

1,696 0.19 (0.39) 0.12 (0.32)

4.30*

1,697 0.28 (0.45) 0.13 (0.33)

7.82*

1,696 0.22 (0.41) 0.08 (0.27)

8.46*

1,696 0.21 (0.41) 0.11 (0.32)

5.54*

1,691 1,696 1,675 1,696 1,696

0.58 (0.49) 0.50 (0.50) 0.16 (0.36) 0.48 (0.50) 0.07 (0.25)

0.55 (0.50) 1.55 0.47 (0.50) 1.38 0.20 (0.40) 2.17* 0.38 (0.48) 4.44* 0.03 (0.16) 4.02*

1,696 0.34 (0.48) 0.26 (0.44)

3.88*

(continued)

294

Evaluation Review 37(3-4)

Table 3. (continued) Variable Employment/education/skills services Received any employment services (of list below) Participated in employment readiness program Participated in job training program Talked to potential employer Given advice about job interviewing Given advice about answering questions about criminal history Given advice about how to behave on the job Given names of people to contact in community to find job Put together a resume Received any educational services Received money management services Received other life skills training Received assistance with personal relationships Received training to change criminal behavior attitudes

N

SVORI

NonSVORI

tStatistica

1,696 0.37 (0.48) 0.19 (0.39)

8.71*

1,693 0.23 (0.42) 0.09 (0.28)

8.06*

1,696 1,696 1,696 1,695

0.04 (0.20) 0.06 (0.23) 0.14 (0.35) 0.13 (0.34)

9.16* 6.37* 9.01* 8.53*

1,696 0.31 (0.46) 0.13 (0.34)

9.12*

1,695 0.27 (0.44) 0.13 (0.33)

7.37*

1,696 1,697 1,696 1,693 1,697

0.10 (0.30) 0.43 (0.50) 0.08 (0.27) 0.21 (0.41) 0.17 (0.37)

8.01* 4.06* 9.28* 9.84* 4.32*

1,697 0.52 (0.50) 0.36 (0.48)

6.76*

0.17 (0.38) 0.15 (0.35) 0.32 (0.47) 0.30 (0.46)

0.24 (0.43) 0.53 (0.50) 0.24 (0.43) 0.42 (0.49) 0.25 (0.43)

Note. SVORI ¼ Serious and Violent Offender Reentry Initiative. a t-Statistics are from weighted logistic regression models and test whether the parameter estimate on the SVORI indicator is significantly different from zero. *p < .05, two-tailed test.

receiving money management classes (24%) and assistance with personal relationships (25%). To examine variability in SVORI implementation across sites, individual service receipt was aggregated within sites and by SVORI and non-SVORI status. (For details on the methodology of developing measures of overall service receipt, see Lattimore, Visher, and Steffey 2008; Winterfield et al. 2006). We expected to observe considerable variability in service receipt across sites because sites designed SVORI programs to best fit their needs. Moreover, because the types and amounts of services provided on a routine basis to prisoners vary considerably across correctional systems, we also expected to observe considerable variation in the services delivered to our non-SVORI respondents who were receiving ‘‘treatment as usual’’

Lattimore and Visher

100 90 80 70 60 50 40 30 20 10 0

295

SVORI Non-SVORI 56 41

39 24

23 13

IA*

IN*

15 18

KS*

22

ME

16 14

MD

40 17

31 15

MO* NV*

22 25 20

OH*

OK

31 33 19

PA

27

20

10

SC*

WA*

Figure 1. Average in-prison service receipt bundle scores for all services, by site and group. *p < .05 for test of significant difference between SVORI and non-SVORI within site. SVORI ¼ Serious and Violent Offender Reentry Initiative.

while in prison. Figure 1 shows the service receipt scores across all prerelease services by site and group. The service receipt scores for SVORI respondents in the 12 sites ranged from 13 to 56, whereas the scores for non-SVORI respondents ranged from 10 to 33. Only respondents from the Iowa SVORI program reported receiving more than 50% of the services (average score of 56), more than double the score of the non-SVORI respondents in that site (24). SVORI respondents in 7 of the 12 sites reported receiving significantly more services than their non-SVORI counterparts. However, in four sites, there was not a significant difference in reported service receipt, and in one site, SVORI respondents reported receiving significantly fewer services, on average, than the non-SVORI respondents. Figure 1 clearly illustrates the variability of SVORI implementation across the 12 adult male evaluation sites. This article focuses on the overall impact of SVORI participation on reentry outcomes. Future analyses will focus on the impact of specific types of services received on outcomes.9 Despite the much-less-than universal reports of service receipt, in most cases SVORI program participation substantially increased the likelihood of receiving specific services—often 2- to 3-fold. At minimum, however, we should conclude that the SVORI programs were only partially implemented.10

Outcomes The SVORI programs were intended to improve outcomes as individuals reintegrated into the community following release from prison, including

296

Evaluation Review 37(3-4)

employment, education, health (particularly substance use), housing, and recidivism. Evaluation participants were asked 3 months following release about their circumstances with respect to each of these outcomes. Table 4 provides results, showing the PATE-weighted means for the SVORI and non-SVORI groups, the percentage difference between the SVORI and non-SVORI group means,11 and results from the logistic regressions of outcomes on the SVORI indicator (sample size, test statistic, and p value). All outcomes are self-report except for arrest and reincarceration, which came from administrative data, and some of the substance use measures that incorporate self-report and oral swab drug test results (refusal to consent to the drug test was counted as a positive test).

Housing Differences in housing outcomes between the SVORI and non-SVORI groups were positive, albeit not always statistically significant. SVORI program participants were about 35% more likely than non-SVORI respondents to report that they had their names on leases and 22% more likely to report that they were living in their own house or apartment. Although there were small positive differences between SVORI and non-SVORI respondents with respect to reporting ‘‘no housing challenges’’ and that they were contributing to housing costs, these differences were not statistically significant at normally considered p values.

Employment SVORI program participants were more likely than non-SVORI respondents to report that their current or most recent job offered formal pay (84% compared with 74%) and that their current or most recent job provided benefits (health insurance and/or paid leave; 47% compared with 39%). Nearly two thirds of SVORI respondents (64%) compared with 59% of non-SVORI respondents reported that they were currently supporting themselves with a job (p ¼ .09). Other positive differences, although not statistically significant, between the SVORI and non-SVORI groups included not having problems finding a job (30% vs. 25%), being currently employed (77% vs. 73%), and current job being a permanent job (68% vs. 67%).

Drug Use In addition to being asked about drug use since release and in the past 30 days, respondents were asked to consent to an oral swab drug test;

297

Housing Name is on lease Lives in own house or apartment Contributes to housing cost No housing challenges Employment Current/most recent job provides benefits No problems finding job Current/most recent job has formal pay Currently supports self with job Currently employed Current/most recent job permanent Worked each month since release Drug use No drug use since releaseb No drug use past 30 daysb No marijuana use No marijuana use past 30 days Drug test negative for all drugs Criminal behavior Committed no crimes since release

Outcome

Table 4. 3-Month Outcomes by Group.

0.1643 0.2054 0.6687 0.8149 0.3897 0.2536 0.7440 0.5880 0.7349 0.6686 0.3896 0.4877 0.5200 0.7583 0.8150 0.6407 0.7296

0.4721 0.3009 0.8417 0.6423 0.7692 0.6793 0.3763 0.5142 0.5432 0.7965 0.8609 0.6582 0.7891

Non-SVORI Mean

0.2209 0.2507 0.6973 0.8371

SVORI Mean

Weighted Means

8.14

5.43 4.46 5.03 5.62 2.74

21.14 18.65 13.13 9.24 4.67 1.60 3.42

34.51 22.05 4.29 2.72

% Difference

984

984 984 984 983 854

727 981 733 983 984 729 733

891 984 979 984

N

4.4975

0.6644 0.5113 1.9679 3.5945 0.2766

4.8166 2.6379 10.1580 2.9527 1.4930 0.0910 0.1327

4.3572 2.7767 0.9004 0.8204

.0339

.4150 .4746 .1607 .0580 .5989

.0282 .1043 .0014 .0857 .2217 .7629 .7157

.0369 .0956 .3427 .3651

p Value

(continued)

Wald’s Chi-Square Test Statistic

Logistic Regression Resultsa

298 0.8207 0.9877 0.9904 0.9065

0.8228 0.5864

0.8164 0.9643

0.9897 0.9871 0.8999

0.8257 0.6138

0.8396 0.9695

Non-SVORI Mean

0.8644

SVORI Mean

Weighted Means

2.85 0.54

1581 1697

983 808

982

0.73

0.35 4.68

979 980

981

N

0.21 0.33

5.33

% Difference

1.4331 0.3393

0.0132 0.6112

0.1175

0.0886 0.2533

3.3527

Wald’s Chi-Square Test Statistic

Logistic Regression Resultsa

.2313 .5602

.9084 .4343

.7317

.766 .615

.0671

p Value

a Logistic regression of outcome as a function of Serious and Violent Offender Reentry Initiative (SVORI) indicator variable; sample size, Wald’s chisquare test statistic testing that the coefficient for the SVORI indicator variable differs from zero; p value for the chi-square statistic. b No self-reported drug use, no positive drug test, consented to drug test.

Committed no drug possession crimes since release Committed no lesser crimes since release Committed no property crimes since release Committed no violent crime, didn’t carry weapon since release Not booked into jail since release Complied with all conditions of supervision since release No arrest since release No reincarceration since release

Outcome

Table 4. (continued)

Lattimore and Visher

299

incarcerated respondents were asked about use prior to their current incarceration and were not drug tested. Lab test results were available for 854 of the 984 subjects interviewed at 3 months following release. The level of drug use following release among these groups was high, although again, SVORI respondents were somewhat more likely to report no use. Eighty percent of SVORI and 76% of non-SVORI reported that they had not used marijuana since their release from prison and slightly more reported no marijuana use in the past 30 days (86% of SVORI, 82% of non-SVORI, p ¼ .06). Drug test results were negative for all substances for about two thirds of the sample (66% of SVORI, 64% non-SVORI). On the measure that combined self-reports and drug test results (with refusal to consent considered a positive test result), about half had no indication of use since release (51% of SVORI, 49% of non-SVORI) or of use in the past 30 days (54% of SVORI, 52% of non-SVORI).12

Criminal Behavior Measures of criminal behavior included self-reported criminality and other indicators of criminal conduct and arrest and reincarceration obtained from administrative sources. SVORI program participants were more likely to report having been crime free since release than non-SVORI respondents (79% compared with 73% reported having committed no crimes since release), a difference largely driven by differences in SVORI participants reporting no drug possession crimes (86% vs. 82%); self-reports of other crime types were similar. SVORI program participants were somewhat more likely to report complying with conditions of supervision since release (61% vs. 59%), but this difference is not statistically significant. (82% of SVORI program participants and 85% of non-SVORI respondents reported that they were on probation or parole at the time of the first postrelease interview.) Self-reported booking into jail was similar for the two groups—83% of SVORI and 82% of non-SVORI said that they had not been booked into jail since release. Arrest data from NCIC were available for 1,581 subjects and reincarceration data were available for everyone. There was a small but insignificant difference in remaining arrest free (84% of SVORI compared with 82% of non-SVORI); the percentages with no arrests were similar to the percentages who reported not having been booked in jail. The reincarceration data showed that few had been returned to jail so soon after release, with 97% of SVORI and 96% of non-SVORI remaining free in the community.

300

Evaluation Review 37(3-4)

2.0

Odds Rao

1.5 1.0 0.5 0.0

Figure 2. Odds ratios from weighted logistic regressions, showing the effect of SVORI program participation on outcomes. Note. Black bars indicate significance at the .05 level; dark gray bars indicate significance at the .10 level.

Overall, the evaluation revealed modest positive effects in most postrelease outcomes measures, although only a few effects were statistically significant at usual significance levels. Figure 2 shows the odds ratios from the logistic regression models estimated for the outcomes shown in Table 4. As can be seen, 22 of the 25 are greater than 1 (although 3 are effectively 1); 4 of the 22 are significantly different from 1 at the .05 level and another 4 are significantly different from 1 at the .10 level. Clearly, under the null hypotheses of no SVORI program effect, we would expect to see a more symmetrical distribution of the odds ratios around 1 and few, if any, significant effects. The analyses presented here did not control for site. The goal of the evaluation was to assess the effect of the SVORI on prisoner reentry outcomes. The target population for the Initiative was ‘‘serious and violent’’ state prisoners. The programs, although locally designed, were intended to address the panoply of needs that are ubiquitous among a prison population—meager education, few job skills, problems with substance use, and, in many cases, mental health symptoms or diagnoses. Thus, the programs were uniform in intent. We did, however, conduct analyses not reported here that controlled for site in two ways—in the first instance, we included site along with other control variables in the outcome regression models. In the second instance, we controlled for site in a hierarchical modeling framework. In both cases, although site indicators were sometimes significant, the effects of SVORI program participation on the outcomes were unchanged from those reported here from the simplest bivariate model.

Lattimore and Visher

301

Discussion and Conclusions The results suggest that these SVORI programs were able to increase the number and types of services provided to male prisoners before release, but that overall level of service provision was well below 100%. Of the array of in-prison services and programs shown in Table 3, SVORI program participants were significantly more likely to receive all of these except medical and dental treatment, which prisons are required to provide, and help accessing public financial assistance. Although some services would not be appropriate for all SVORI program participants (e.g., mental health treatment), other services should have been provided to all. For example, needs assessment and reentry planning should have been provided to everyone. Although these services were among the most likely to be provided, less than two thirds of SVORI program participants (63%) reported having received a needs assessment (49% reported having a release-specific needs assessment) and only 57% reported having a reentry plan 30 days prior to release from prison. Between 65% and 75% of SVORI respondents reported working with anyone to prepare for release or participating in programs or classes to prepare for release. The only other services reported received by about half of the SVORI program participants were services or programs focused on individual change. These are the types of programs that MacKenzie (2006), among others, has identified as more effective than more pragmatic strategies that, for example, increase opportunities for work. In particular, 52% of SVORI program participants reported receiving training to change criminal behavior attitudes (e.g., cognitive behavior therapy), 48% reported receiving substance use treatment, and 53% reported receiving educational services. Very few received employment-related services, although most of the adult SVORI programs had reported that employment was a primary focus of their programs (Lattimore et al. 2005). Only 37% reported that they had received any employment services, most commonly reporting receiving advice about job interviewing and responding to questions about their criminal history. These measures of service receipt are, as noted earlier, based on respondent self-reports of receiving services during their current incarceration. There are two potential limitations to these measures. First, our measures of receipt depend upon subject recall. Although some services may be more salient than others, we have no reason to expect that receiving a service would be more salient to someone who was in a SVORI program than

302

Evaluation Review 37(3-4)

someone who received the service as part of the standard treatment received by inmates. Thus, we have no reason to expect differences due to under- or overreporting errors between the two groups. Second, we have only a dichotomous indicator of whether the service or program was received. Because of the number of different types of services participants could potentially receive and the number of other types of information that needed to be collected during the interviews, as well as the cognitive imposition of attempting to obtain dosage information on each of the programs (e.g., how many group drug treatment sessions did you attend?), the decision was made during the instrument development stage to ask only whether someone received a service (e.g., drug treatment). Thus, we have no measures of dosage—one treatment session is treated the same as 2 years of treatment. Limited implementation of the SVORI programs, as indicated by selfreported service receipt, coupled with receipt of some services by those in the non-SVORI comparison group—albeit at levels lower than those in programs—reduces the power of the evaluation to identify treatment effects if they exist. In the case of the current evaluation, we find that service provision prior to release was not only incomplete, but other analyses reported elsewhere show that service provision was below expressed needs for service and declined substantially following release from prison (see Lattimore, Steffey, and Visher 2009; Lattimore, Visher, and Steffey 2010). Even so, we identified modest effects during the initial, some would call ‘‘critical,’’ early period following release. For example, the SVORI program participants were more likely to report living independently and having a job with formal pay and benefits shortly after release. The SVORI program participants were also more likely to report that they had not used drugs and to test negative on the oral drug screen, although few of these differences were statistically significant. Table 4 includes two self-reported measures of marijuana use (any since release, any in the past 30 days). About 80% of the SVORI group compared with 76% of the non-SVORI group reported no marijuana use since release—a 5% difference (p value ¼ .16) similar to the 6% difference (p value ¼ .06) for the past 30 days measure. SVORI program participants also were somewhat more likely to have indications of no use on any drug on the three measures shown in Table 4, although overall use as indicated by the combined measure (no self-report, negative drug test, consented to test) suggests that about half of all subjects were again using drugs within 3 months of release from prison. There was about a 5% difference between SVORI (51% no use) and non-SVORI (49% no use) reports for no use since release, but this difference was not

Lattimore and Visher

303

statistically significant. There was a 3% difference on the drug test results alone (66% of SVORI and 64% of non-SVORI were negative for all substances). SVORI program participants were less likely to report having committed crimes since release, driven primarily by a lower reporting of drug possession crimes. On the other hand, official measures of criminal behavior indicated little difference between SVORI and non-SVORI participants. Official measures of arrest showed that 16% of SVORI and 18% of non-SVORI evaluation participants (N ¼ 1,581) had been arrested at least once during the 3 months following release, comparable to the 17% of SVORI and 18% of non-SVORI 3-month interview respondents who reported having been jailed since release (N ¼ 983). Very few had been reincarcerated within 3 months of release—only about 3% of both groups (N ¼ 1,697), suggesting few revocations for either group in the early months following release. To see the potential erosion of power, we can perform a ‘‘thought experiment’’ using our data and results from the literature with respect to treatment effects. We will consider the impacts of four treatments, listed here with the treatment effect estimates (based on meta-analyses) from Drake, Aos, and Miller (2009): cognitive behavior therapy (6.9%), substance abuse treatment in prison (6.4%), vocational training (9.8%), and general education (8.3%). From our data, we use the ‘‘received training to change criminal behavior’’ for the CBT measure, ‘‘participated in job training program’’ as a proxy for vocational training, and ‘‘received any educational services’’ for the general education measure. It can be shown that for any group for which some receive treatment and others do not, that the observed recidivism rate for the group overall is R ¼ r  ð1  T  pÞ; where R ¼ observed recidivism rate for the group, r ¼ recidivism rate in the absence of treatment, T ¼ percentage of group that is treated, and p ¼ the percentage reduction in recidivism due to treatment (the treatment effect). Table 5 includes the treatment effects from Drake, Aos, and Miller (2009), the percentages of each of our groups who reported receiving each treatment while in prison, and the observed recidivism rate for each group under the assumption that the recidivism rate in the absence of treatment is 20%, slightly higher than what we observed in the initial 3 months following release for our subjects.13 The dilution in the treatment effect due to the two factors of incomplete treatment of the treated and partial treatment of the comparison group is stark. The modest treatment effects of 6.4–9.8%

304

Evaluation Review 37(3-4)

Table 5. Hypothetical Treatment Effects With Incomplete Treatment of the Treatment Group and Partial Treatment of the Comparison Group, Assuming Untreated Recidivism Rate is 20%. Treated (%) Treatment Cognitive behavior therapy Substance abuse treatment Vocational education General education

Observed Recidivism Rate (Untreated Rate ¼ 20%)

SVORI Non-SVORI Difference Treatment (%) (%) Effecta (%) SVORI Non-SVORI (%) 6.90

52

36

19.28

19.50

1.13

6.40

48

38

19.34

19.48

0.71

9.80

17

4

19.77

19.94

0.90

8.30

53

43

19.27

19.41

0.71

Note. SVORI ¼ Serious and Violent Offender Reentry Initiative. a Estimates from Drake, Aos, and Miller (2009).

diminish to an expected percentage difference between the two groups in the 1% or so range. We assumed in the calculations for Table 5 that the treatment effect was the average treatment effect on the treated. Given that both groups include ‘‘treated’’ subjects, we would still need to make an adjustment. In this case, we assume that the treatment effects from the literature apply to the SVORI group and that the treatment effect for the non-SVORI group is determined by multiplying the treatment effect from the literature by the ratio of non-SVORI to SVORI participation. Under these assumptions, cognitive behavior therapy is expected to reduce recidivism by 6.9% for the SVORI program participants and by 4.78% (or 69% of 6.9%) for the non-SVORI respondents. Again, in Table 6, we see that the observed differences between SVORI and non-SVORI recidivism is 2% or less except for the difference for vocational training where we would expect to see nearly an 8% difference. However, only 17% of the SVORI group and 4% of the nonSVORI group reported any job training so it seems unreasonable to assume that you could get an almost 10% reduction in recidivism by treating only 17% of your program participants so we should take this particular estimate with a grain of salt. Not surprisingly, even with reasonable sample sizes, the

305

52 48 17 53

6.90

6.40

9.80 8.30 4 43

38

36

Note. SVORI ¼ Serious and Violent Offender Reentry Initiative. a Estimates from Drake, Aos, and Miller (2009).

Cognitive behavior therapy Substance abuse treatment Vocational education General education

Treatment

Treatment Effecta SVORI Non-SVORI (%)

Treated (%)

23.53 81.13

79.17

69.23

NS/S (%)

2.31 6.73

5.07

4.78

NS Weighted Treatment Effect (%)

18.04 18.34

18.72

18.62

SVORI (%)

19.54 18.65

18.99

19.04

Non-SVORI (%)

7.67 1.68

1.40

2.23

Difference (%)

Observed Recidivism Rate (Untreated Rate ¼ 20%)

Table 6. Hypothetical Treatment Effects With Incomplete Treatment of the Treatment Group and Partial Treatment of the Comparison Group Based on Ratio of Treatment in Each Group, Assuming Untreated Recidivism Rate Is 20%.

306

Evaluation Review 37(3-4)

statistical power of any hypothesis testing is going to be quite low with differences such as those shown in Tables 5 and 6. For example, with the observed recidivism shown for cognitive behavior therapy in Table 6 (SVORI ¼ 18.62% and non-SVORI ¼ 19.04%) and assuming 800 subjects in each group and a ¼ .05, one-tailed test, we have statistical power of 5.2%. In this case, we have looked at the direct effect of treatment on recidivism. As Lattimore, Visher, and Steffey (2010) point out, some treatments and services are focused on other outcomes and reduced recidivism is supposed to follow improvements in those intermediate outcomes—for example, drug treatment reduces drug use which reduces recidivism. As they demonstrate, under this two-stage process, we once again confront substantial challenges to generate observable, testable reductions in recidivism. If participation in job training boosts postrelease employment by 20% from 50% to 60% and being employed reduces recidivism by 20% from 50% to 40%, 45 members of a 100-person control group and 44 members of a 100-person treatment group will fail—a reduction of 2.2%.14 The findings reported here suggest that fielding a sufficiently powerful test of reentry programs that focuses on multiple domains when the control condition also includes some services may be more formidable a challenge than originally anticipated. The findings here point to the need for more research into what works in reentry programming and for whom. Since the implementation of the SVORI programs in the early 2000s there has been additional work, including the work of MacKenzie (2006), to suggest that ‘‘practical services’’ may be less successful than ‘‘individual change’’ services in reducing the recidivism of releasing prisoners. This conclusion is supported by an emerging focus on desistance (Laub and Sampson 2001; LeBel et al. 2008; Paternoster and Bushway 2009), including an anticipation that cognitive transformation (or ‘‘individual change’’) may be a necessary condition for effecting positive change in the most serious criminal offenders. This emerging research also may call into question the underlying logic model to programs like the SVORI reentry programs (see Lattimore, Visher, and Steffey 2010)—specifically, the need to develop a better understanding of the relationship of services to intermediate outcomes and recidivism. In future work, we will explore the impact of individual service receipt on the SVORI program participants. Given the ongoing focus on needs-based programming in criminal justice settings, we are likely to continue to face challenges similar to those posed by the SVORI evaluation. Indeed, the Second Chance Act (SCA) funding through the Bureau of Justice Assistance that is a sequel to SVORI funding for prisoner reentry programs continues to place emphasis on needs

Lattimore and Visher

307

assessment and programs and services responsive to the assessed needs. Multisite evaluations of SCA-funded programs are underway and the lessons with respect to program/service delivery definition and measurement and statistical power provided by the SVORI evaluation should be taken into account as these new evaluations go forward. Acknowledgment The authors wish to thank the SVORI program directors, staff at the facilities where interviews were conducted, and the respondents who participated in the longitudinal study.

Authors’ Note Points of view are those of the authors and do not necessarily represent those of the U.S. Department of Justice.

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 Multisite Evaluation of the Serious and Violent Offender Reentry Initiative (SVORI) was supported by grant numbers 2003-RE-CX-K101 and 2004-RE-CX-0002 from the National Institute of Justice (U.S. Department of Justice).

Notes 1. Data were also collected from program directors; these data are not used in the analyses reported here. 2. Nonincarcerated subjects who completed Wave 2 and Wave 3 interviews received US$50 for each interview plus an additional US$5 if they called in to schedule the interview. Subjects who completed a Wave 4 interview received US$50 plus US$5 if they called into schedule the interview plus an additional US$50 if they had completed all four interviews. In addition, nonincarcerated subjects were asked to provide an oral swab sample for drug testing at the conclusion of the Wave 2 and Wave 4 interviews for which they received an additional US$15. Although the original protocol prescribed no compensation for incarcerated individuals, we did work with the states to provide compensation for Wave 4 in order to be able to provide the extra compensation for individuals who had completed all interviews.

308

Evaluation Review 37(3-4)

3. One reviewer questioned the timeliness of submission of arrest data to National Crime Information Center (NCIC), suggesting this could ‘‘jeopardize the integrity of the findings.’’ Our subjects were released between late 2004 and early 2006; NCIC data for these analyses were obtained in the summer of 2008, with a supplemental request in March 2009. We had a postrelease arrest follow-up period of at least 21 months for all subjects, suggesting that reporting timeliness should not be an issue for analyses focused on the initial 3 months following release (see Lattimore and Steffey 2009 for additional details). 4. Response rates improved across the remaining waves during which interviews were attempted with all eligible Wave 1 respondents. The response rate was 61% at 9 months (Wave 3) and 66% at 15 months (Wave 4). Overall, we were able to complete at least one postrelease interview with 79% of the subjects and all three postrelease interviews with 42%. 5. In contrast, the average treatment effect on the treated provides an estimate of the effect of treatment on the treated. 6. This approach comes from the clinical trials literature, where the regimen of treatment is set and a subject can miss the last treatment. Because the three data collection points in this study are somewhat arbitrary compared with a clinical trial, alternative coding schemes were applied for the dependent variable in this analysis; results were similar. 7. Due to the breadth of the interview, length of treatment or dosage was not gathered and was not available consistently in administrative records. 8. The respondents were asked whether they had received ‘‘any employment services or assistance with finding a job for when you are released.’’ If they responded ‘‘yes,’’ they were then asked specifically about the following services: employment readiness training, trade or job training, talked to potential employer, advice about interviewing, advice about questions about criminal history, advice on how to behave on the job, given names of people to contact in the community about a job, and putting together a resume. 9. Across the Serious and Violent Offender Reentry Initiative (SVORI) sites, service receipt after release dropped substantially, with participants reporting receiving about half of the services they received while in prison (see Lattimore, Steffey, and Visher 2009). Thus, this article focuses only on receipt of in-prison services. 10. Of course, service provision was to be needs driven and not all subjects would need all of the possible services. We asked respondents about their need for individual services and the proportion reporting needing a service exceeded the proportion reporting receiving a service in virtually all cases—often by substantial levels. For example, nearly everyone reported needing educational programs. Interested readers are referred to Lattimore, Steffey, and Visher (2009).

Lattimore and Visher

309

11. Percentage difference is calculated as ½ðSVORI  non  SVORIÞ=non SVORI  100 to show the percentage improvement of the SVORI group over the non-SVORI group. 12. The difference between the ‘‘no use since release’’ and ‘‘no use in the past 30 days measure’’ is the self-reports of no use, as the drug test results reflect recent use—the past few days for most drugs (Center for Substance Abuse Treatment 2006; Retrieved from http://www.ncbi.nlm.nih.gov/books/NBK14448/). 13. Note that the percentage differences are constant with respect to the untreated recidivism rate. 14. Fifty percent of the unemployed will fail or 25 controls and 20 treatment; 40% of the employed will fail or 20 controls and 24 treatment.

References Allison, P. D. 2001. Missing Data. Sage university papers series on quantitative applications in the social sciences, No. 07-136. Thousand Oaks, CA: Sage. Andrews, D. A. 2006. ‘‘Enhancing Adherence to Risk-need-responsivity: Making Quality a Matter of Policy.’’ Criminology and Public Policy 5:595–602. Andrews, D., and J. Bonta. 2006. The Psychology of Criminal Conduct. 4th ed. Newark, NJ: Lexis/Nexis/Matthew Bender. Andrews, D. A., I. Ziner, R. D. Hoge, J. Bonta, P. Gendreau, and F. T. Cullen. 1990. ‘‘Does Correctional Treatment Work: A Clinically-relevant and Psychologicallyinformed Meta-analysis.’’ Criminology 28:369–404. Aos, S., M. Miller, and E. Drake. 2006. Evidence-based Adult Corrections Programs: What Works and What Does not. Olympia: Washington State Institute for Public Policy. Center for Substance Abuse Treatment. 2006. Substance Abuse: Clinical Issues in Intensive Outpatient Treatment. Treatment Improvement Protocol (TIP) Series 47. DHHS Publication No. (SMA) 06-4182. Rockville, MD: Substance Abuse and Mental Health Services Administration. Cook, T. D., W. R. Shadish, and V. C. Wong. 2008. ‘‘Three Conditions under Which Experiments and Observational Studies Produce Comparable Causal Estimates: New Findings from Within-study Comparisons.’’ Journal of Policy Analysis and Management 27:724–50. Cook, T. D., and P. M. Steiner. 2010. ‘‘Case Matching and the Reduction of Selection Bias in Quasi-experiments: The Relative Importance of Pretest Measures of Outcome, of Unreliable Measurement, and of Mode of Data Analysis.’’ Psychological Methods 1:56–68. Cullen, F., and P. Gendreau. 2000. ‘‘Assessing Correctional Rehabilitation: Policy, Practice, and Prospects.’’ In Criminal Justice 2000, Vol. 3, edited by J. Horney, 109–75. Washington, DC: National Institute of Justice.

310

Evaluation Review 37(3-4)

D’Agostino, R. B., Jr. 1998. ‘‘Tutorial in Biostatistics: Propensity Score Methods for Bias Reduction in the Comparison of a Treatment to a Non-randomized Control Group.’’ Statistics in Medicine 17:2265–81. Drake, E. K., S. Aos, and M. G. Miller. 2009. ‘‘Evidence-based Public Policy Options to Reduce Crime and Criminal Justice Costs: Implications in Washington State.’’ Victims and Offenders 4:170–96. Farabee, D. J. 2005. Rethinking Rehabilitation: Why Can’t We Reform Our Criminals? Washington, DC: AEI Press. Fitzmaurice, G. M., and N. M. Laird. 2000. ‘‘Generalized Linear Mixture Models for Handling Nonignorable Dropouts in Longitudinal Studies.’’ Biostatistics 1:141–56. Gaes, G., T. Flanagan, L. Motiuk, and L. Stewart. 1999. ‘‘Adult Correctional Treatment.’’ In Prisons, edited by M. Tonry and J. Petersilia, 361–426. Chicago: University of Chicago Press. Gendreau, P., T. Little, and C. Goggin. 1996. ‘‘A Meta-analysis of the Predictors of Adult Offender Recidivism.’’ Criminology 34:575–607. Glynn, R. J., S. Schneeweiss, and T. Sturmer. 2006. ‘‘Indications for Propensity Scores and Review of Their Use in Pharmacoepidemiology.’’ Basic Clinical Pharmacology and Toxicology 98:253–59. Imai, K., G. King, and E. A. Stuart. 2008. ‘‘Misunderstandings between Experimentalists and Observationalists about Causal Inference.’’ Journal of Royal Statistical Society 171:481–502. Langan, P., and D. Levin. 2002. Recidivism of Prisoners Released in 1994. NCJ 193427. Washington, DC: U.S. Department of Justice, Bureau of Justice Statistics. Lattimore, P. K. 2007. ‘‘The Challenges of Reentry.’’ Corrections Today 69:88–91. Lattimore, P. K., and D. M. Steffey. 2009. The Multi-site Evaluation of SVORI: Methodology and Analytic Approach. Research Triangle Park, NC: RTI International. Lattimore, P. K., D. M. Steffey, and C. Visher. 2009. Prison Reentry Experiences of Adult Males: Characteristics, Service Receipt, and Outcomes of Participants in the SVORI Multi-site Evaluation. Research Triangle Park, NC: RTI International. Lattimore, P. K., C. A. Visher, S. Brumbaugh, C. Lindquist, and L. Winterfield. 2005. ‘‘Implementation of Prisoner Reentry Programs: Findings from the Serious and Violent Offender Reentry Initiative Multi-site Evaluation.’’ Justice Research and Policy 7:87–109. Lattimore, P. K., C. Visher, and D. M. Steffey. 2008. Pre-release Characteristics and Service Receipt among Adult Male Participants in the SVORI Multi-site Evaluation. Research Triangle Park, NC: RTI International. Lattimore, P. K., C. Visher, and D. M. Steffey. 2010. ‘‘Prisoner Reentry in the First Decade of the 21st Century.’’ Victims and Offenders 5:253–67.

Lattimore and Visher

311

Lattimore, P. K., S. Brumbaugh, C. Visher, C. H. Lindquist, L. Winterfield, M. Salas, and J. Zweig. 2004. National Portrait of the Serious and Violent Offender Reentry Initiative. Research Triangle Park, NC: RTI International. Laub, J., and R. Sampson. 2001. ‘‘Understanding Desistance from Crime.’’ In Crime and Justice: A Review of Research, Vol. 28, edited by M. Tonry, 1–69. Chicago: University of Chicago Press. LeBel, T. P., R. Burnett, S. Maruna, and S. Bushway. 2008. ‘‘The ‘Chicken and Egg’ of Subjective and Social Factors in Desistance from Crime.’’ European Journal of Criminology 5:131–59. Lipsey, M., and F. Cullen. 2007. ‘‘The Effectiveness of Correctional Rehabilitation: A Review of Systematic Reviews.’’ Annual Review of Law and Social Science 3: 297–320. Lowenkamp, C. T., E. J. Latessa, and P. Smith. 2006. ‘‘Does Correctional Program Quality Really Matter? The Impact of Adhering to the Principles of Effective Intervention.’’ Criminology & Public Policy 5:575–94. Lum, C., and S. Yang. 2005. ‘‘Why Do Evaluation Researchers in Crime and Justice Choose Non-experimental Methods?’’ Journal of Experimental Criminology 1: 191–213. Lynch, J. P., and W. J. Sabol. 2001. Prisoner Reentry in Perspective. Crime Policy Report 3. Justice Policy Center. Washington, DC: The Urban Institute. Mackenzie, D. L. 2006. What Works in Corrections: Reducing the Criminal Activities of Offenders and Delinquents. New York: Cambridge University Press. Marlowe, D. B. 2006. ‘‘When ‘What Works’ Never Did: Dodging the ‘Scarlet m’ in Correctional Rehabilitation.’’ Criminology and Public Policy 5:339–46. National Research Council. 2007. Parole, Desistance from Crime, and Community Integration. Committee on Community Supervision and Desistance from Crime, Committee on Law and Justice, National Research Council of the National Academies. Washington, DC: The National Academy Press. Paternoster, R., and S. Bushway. 2009. ‘‘Desistance and the ‘Feared Self’: Toward an Identity Theory of Criminal Desistance.’’ Journal of Criminal Law & Criminology 99:1103–56. Petersilia, J. 2004. ‘‘What Works in Prisoner Reentry? Reviewing and Questioning the Evidence.’’ Federal Probation 6:4–8. Petersilia, J. 2005. ‘‘From Cell to Society: Who Is Returning Home?’’ In Prisoner Reentry and Crime in America, edited by J. Travis and C. A. Visher, 15–49. Cambridge, England: Cambridge University Press. Re-Entry Policy Council. 2005. Report of the Re-entry Policy Council: Charting the Safe and Successful Return of Prisoners to the Community. New York: Council of State Governments.

312

Evaluation Review 37(3-4)

Rhine, E. E., T. L. Mawhorr, and E. C. Parks. 2006. ‘‘Implementation: The Bane of Effective Correctional Programs.’’ Criminology & Public Policy 5:347–58. Rosenbaum, P., and D. B. Rubin. 1983. ‘‘The Central Role of the Propensity Score in Observational Studies for Causal Effects.’’ Biometrika 70:41–55. Rosenbaum, P., and D. B. Rubin. 1985. ‘‘The Bias Due to Incomplete Matching.’’ Biometrics 41:103–16. Rossman, S. B. 2003. ‘‘Building Partnerships to Strengthen Offenders, Families, and Communities.’’ In Prisoners Once Removed: The Impact of Incarceration and Reentry on Children, Families, and Communities, edited by J. Travis and M. Waul, 343–80. Washington, DC: The Urban Institute Press. Rubin, D. B. 2006. Matched Sampling for Causal Effects. New York: Cambridge University Press. SAS Institute. 2004. SAS/STAT 9.1 User’s Guide. Cary, NC: SAS Institute. Seiter, R. P., and K. R. Kadela. 2003. ‘‘Prisoner Reentry: What Works, What Doesn’t, and What’s Promising.’’ Crime and Delinquency 49:360–88. Visher, C. A. 2006. ‘‘Effective Reentry Programs. Editorial Introduction.’’ Criminology and Public Policy 5:299–302. Visher, C. A. 2007. ‘‘Returning Home: Emerging Findings and Policy Lessons about Prisoner Reentry.’’ Federal Sentencing Reporter 20:93–102. Visher, C. A., and J. Travis. 2003. ‘‘Transitions from Prison to Community: Understanding Individual Pathways.’’ Annual Review of Sociology 29:89–113. Western, B. 2007. ‘‘The Penal System and the Labor Market.’’ In Barriers to Reentry? The Labor Market for Released Prisoners in Postindustrial America, edited by S. Bushway, M. A. Stoll, and D. F. Weiman, 335–59. New York: Russell Sage Foundation. White, T. F. 2006. ‘‘Evidence-based Practice in Probation and Parole: The Implementation Challenge.’’ APPA 30:22–27. Wilson, J. A., and R. C. Davis. 2006. ‘‘Good Intentions Meet Hard Realities: An Evaluation of the Project Greenlight Reentry Program.’’ Criminology and Public Policy 5:303–38. Winterfield, L., P. K. Lattimore, D. Steffey, C. Lindquist, and S. Brumbaugh. 2006. ‘‘The Serious and Violent Offender Reentry Initiative: Measuring the Effects on Service Delivery.’’ Western Criminology Review 7:3–19. Winterfield, L., and C. Lindquist. 2005. Reentry Research in Action: Characteristics of Prisoner Reentry Programs. Research Triangle Park, NC: RTI International.

Author Biographies Pamela K. Lattimore is Principal Scientist in RTI’s Crime, Violence, and Justice Research Program, which she directed from 2000 until 2003. She has 30 years of

Lattimore and Visher

313

experience conducting research focused on the evaluation of interventions; investigation into the causes and correlates of criminal behavior, including substance use and mental health; and development of approaches to improve criminal justice operations. Dr. Lattimore is a nationally recognized expert on prisoner reentry and was Co -Principal Investigator of the Multisite Evaluation of the Serious and Violent Offender Reentry Initiative. Christy A. Visher is Professor at the University of Delaware and Director of the Center for Drug and Alcohol Studies, a research center within the Department of Sociology and Criminal Justice. She has 30 years of experience in policy research and evaluation on crime and justice issues. While at the Urban Institute, she was CoPrincipal Investigator of the Multisite Evaluation of the Serious and Violent Offender Reentry Initiative. Her research interests focus on criminal careers, substance abuse, communities and crime, violence, and the evaluation of strategies for crime control and prevention.

The impact of prison reentry services on short-term outcomes: evidence from a multisite evaluation.

Renewed interest in prisoner rehabilitation to improve postrelease outcomes occurred in the 1990s, as policy makers reacted to burgeoning prison popul...
341KB Sizes 0 Downloads 0 Views