Predictors of Incarceration of Veterans Participating in U.S. Veterans’ Courts

Veterans’ courts (VCs) were created to address the needs of veterans with mental illness who had been appearing in drug and mental health courts with increasing frequency. The VC concept was that for eligible offenses, a VC would permit the individual to be diverted from incarceration to a specified program of community-based treatment, regular court appearances, and veteran-specific interventions. Significant variability exists across jurisdictions regarding eligibility criteria, with some courts requiring a link between a veteran’s service, the illness, and the crime; some permitting violent felonies; and some accepting combat veterans from only certain eras. Some allow for pretrial diversion, whereas others are post plea (1). All require the veteran to have either a substance use disorder or other treatable psychiatric condition (2).

By definition, any participant in a VC program remains in the criminal justice system. Outcomes of successful participation are variable, with the most favorable allowing for dismissal of the relevant criminal charge. Unsuccessful participation might result in extension of program obligations, sanctions imposed by the court, or even removal from the program. Removal from the program would result in the veteran’s facing incarceration for the original criminal charge.

Since VCs began, growth has been exponential. For example, there were 24 VCs in January 2010 (1), 168 by December 2012 (3), and over 300 in more than 35 states as of January 1, 2014. To provide clinical and support services, the Veterans Justice Outreach (VJO) program links many U.S. Department of Veterans Affairs (VA) medical centers with the local VC. In addition, a key component of most VCs is peer mentorship provided by other veterans (4).

Only one previous study of national inventory data on U.S. VCs was located (5). It found that more than 7,700 veterans had been admitted to VCs as of mid-2012, and slightly over two-thirds had successfully completed both the court and the treatment regimen. The length of involvement in the VC averaged 15 to 18 months.

A records review from the original Buffalo, New York, VC program indicated a 0% rate of criminal recidivism at the one-year point (4). A subsequent recidivism analysis revealed a one-year criminal recidivism rate of less than 2% in 14 courts (6). A separate recidivism study of one VC found that the longer a veteran stayed in the program, irrespective of the number of sanctions issued by the judge, the lower the veteran’s likelihood of arrest following separation (7). Researchers studying the same court examined judicial sanctions and found an association between veterans with substance use relapse and subsequent discharge from the VC program (8). Furthermore, in the same study, the infractions of failure to complete a task, substance use relapse, unexcused absence, and a missed hearing were associated with court-sanctioned jail time. Notably, Marlowe and Kirby (9) have shown that judicial sanctions need not be humiliating or painful in order to be effective. However, they should be administered immediately and be of sufficient intensity.

Despite the aforementioned work, available data are sparse. VCs are left with limited guidance regarding which court model might result in the lowest cost to the taxpayer and the greatest efficacy. Fortunately, considerable outcomes research has been conducted in the years since drug courts were created. Although generalizing findings to VCs is uncertain, the drug court research serves as a guide. In that regard, Marlowe (10) has noted a pervasive perception among drug court staff that sanctions and rewards strongly motivate positive behavioral change. Two studies demonstrated that gradually escalating sanctions for infractions, including brief incarcerations, significantly improve outcomes among drug offenders (11,12). Individuals employed at the time of their enrollment in drug court were more likely to successfully complete the program (13).

On the basis of this prior research, we developed a phone survey based upon Russell’s (4) work on VC best practices and Festinger and colleagues’ (14) work regarding measurement methodology on sanctions and rewards. It was assembled with expert input from the VA’s VJO program, the Houston Veterans Treatment Court, and the National Association of Drug Court Professionals.

Our phone survey provided data on a range of VC program components, including sanctions and rewards. During the construction of the VC survey, informants knowledgeable about the courts reported that many VCs placed special emphasis on written and signed behavioral contracts that establish specific expectations of veteran participants and time frames for meeting them. Thus we hypothesized that VCs using behavioral contracts would have lower incarceration rates during veterans’ VC participation. Through this examination of program factors, it is our hope that this research will contribute toward increasing the understanding and efficacy of VC models.

Methods

Ethical Considerations

This study was approved by both the Institutional Review Board (IRB) of Baylor College of Medicine and the VA Research and Development Committee, and all processes used were approved IRB procedures.

Measures

HOMES database.

The VA’s Northeast Program Evaluation Center (NEPEC) manages the VA Homeless Operations Management and Evaluation System (HOMES) database, which is an integrated, primary data collection system for the VA’s specialized homeless programs. This database documents veteran participation in specialized homeless services, including baseline assessments, program entries, progress in services, and outcomes of program participation. Services to VC participants are recorded in HOMES as part of VA's homelessness prevention strategy. Any veteran who enters a VC program completes both a HOMES and VJO intake form; thus not all veterans included in the HOMES database receive housing assistance. During the two-year period from October 2011 to September 2013, a total of 4,495 veterans receiving services in VCs from VA's VJO specialists had exited their respective programs. This cohort represented 106 VA facilities affiliated with 286 VC programs from 53 states and territories. When veterans become justice involved, they complete standardized patient questionnaires administered by specialists throughout the process.

NEPEC provided access to deidentified HOMES data for the purpose of statistical analysis and research. Incarceration data were available only for veterans during the time of their enrollment in a VC program and prior to their exit. Therefore, our primary outcome variable was the likelihood of incarceration during VC program participation. Incarceration as an outcome variable reflects a veteran’s failure to complete the program because of incarceration for the original or new charge. This does not include incarceration as a sanction received in the context of ongoing program participation. Length of stay in a VC program included the number of days from the date of entry until the date of exit. Because our analyses blended information from the HOMES data and the VC phone survey (see below), only individuals who had exited the VC program, whether they completed the program successfully or not, and had court survey data available, were included in these analyses.

Phone survey.

A phone survey of the existing U.S. VCs was conducted over five months from 2013 to 2014 by the VJO National Program Office (phone survey). The survey involved a questionnaire administered by telephone by national VJO staff, who posed questions to each court’s VJO specialist. When that individual could not be reached, another suitable individual with access to court information was interviewed. Each interview lasted approximately 60 minutes, and the questionnaire consisted of 107 questions. Question categories included court structural classification; admission eligibility criteria; court sanctions and rewards; program admissions, terminations, and completions; standard procedures regarding treatment; peer mentor component; and perceived barriers to treatment.

Data Matching

The HOMES data consisted of deidentified individual-level data, whereas phone survey data consisted of data grouped by court. The two data sets were matched by using VA facility codes, which was the only information available in both HOMES and phone survey data. Because some VA facilities were linked to more than one VC, phone survey variables were calculated as the percentage of VCs per facility that endorsed each item. That percentage was then applied for all individual veterans at that facility. Of the 4,495 participants who exited the program, only 1,224 had exit forms with VA facility codes that also appeared in the court survey data set. Thus a total of 3,271 veterans could not be associated with a particular court and were excluded from the analysis at random. The difference in recidivism rates between the included and excluded samples was not significant. Therefore, the included sample is representative of the larger cohort regarding rates of recidivism.

Data Analysis

We first present a descriptive analysis, using HOMES data, of the sociodemographic and clinical features of the veterans in VCs. Continuous measures were evaluated by analysis of variance, and categorical measures were evaluated with the Kruskal-Wallis test. Next, with the goal of identifying which items were most predictive of a veteran’s incarceration during the VC program, we performed a series of four backward conditional logistic regressions by using the method described by Hosmer and Lemeshow (15).

The first regression focused on the initial HOMES assessment variables to identify which pre-VJO program factors were predictive of incarceration during the VC program. The second regression focused on the HOMES variables assessed during a veteran’s VC program involvement to identify which program factors were predictive of incarceration during the VC program. The third regression focused on the court characteristics obtained from the national VC survey to identify which court factors were predictive of incarceration during the VC program. Our fourth regression combined factors that met Bonferroni criteria identified in the sociodemographic characteristics and the first three regressions into a single model to identify which factors—sociodemographic, pre-VC, during VC, or court related—were predictive of incarceration during the VC program. To account for an increased risk of type 1 error in using four logistic regressions, we applied the Bonferroni correction with a new alpha of p<.013 to guide our interpretations.

Results

Demographic Characteristics

The final sample included 1,224 veterans from 73 VA facilities affiliated with 214 VC programs. Average length of stay in the VC program was 197.1±158.6 days. Of the nonincarcerated veterans, 60% were Caucasian, 28% African American, and 10% Hispanic (Table 1). The mean age of the nonincarcerated veterans was 44.1 years. Information regarding gender was not available. Only the number of prior arrests was significantly associated with subsequent incarceration during the VC program.

Associations With Incarceration During VC Participation

Prior to VC entry.

Evaluation of the regression diagnostics, including nonzero variance, homoscedasticity, independence of errors, independence of the outcome, and linearity, indicated that no corrections were needed and that the model had an acceptable goodness of fit (Hosmer and Lemeshow p=.559). The final model predicted an adjusted 15% of the total variance. The following variables were associated with higher rates of incarceration during VC program participation: homelessness lasting for six months to a year, compared with no homelessness; being a student, in the active military, or in VA vocational training, compared with working regular hours, either full-time or part-time; being moderately bothered by substance use cravings within the past month per subjective self-assessment, compared with not being bothered; having a history of substance use treatment, compared with having no such history; and the VJO specialist’s concerns about danger to the veteran from others, compared with no concern about such danger (Table 2).

After VC entry.

Of the 1,224 veterans for whom complete data were available, 908 (74%) had exited their VC program successfully by meeting all requirements and graduating from the program. These veterans no longer required VJO services, nor were they justice involved. Evaluation of the regression diagnostics indicated that no corrections were needed and the model had an acceptable goodness of fit (Hosmer and Lemeshow p=.180). The final model (Table 3) predicted an adjusted 47% of the total variance. The following variables were associated with higher rates of incarceration during VC program participation: charges of probation or parole violations, longer stays in the VC program; end of VC participation because of incarceration for a new arrest or case transfer by the legal system, and requiring mental health follow-up but not undergoing treatment. Lower rates of incarceration were found for veterans with stable housing.

Phone survey.

Evaluation of the regression diagnostics indicated that no corrections were needed and the model had an acceptable goodness of fit (Hosmer and Lemeshow p=.193). The final model (Table 4) predicted an adjusted 10% of the total variance. The following program variables were associated with higher rates of incarceration during VC program participation: use of expedited phase adjustments, phase progression based on measurable goals, and acceptance of cases from outside jurisdictions. Lower rates of incarceration were found for programs that regularly referred veterans to substance abuse treatment.

Combined Analysis of Associations With Incarceration

After evaluation of the regression diagnostics including nonzero variance, homoscedasticity, independence of errors, independence of the outcome, and linearity, the final stepwise model was examined (Table 5). Although number of prior arrests was associated with incarceration in our initial linear regression (Table 1), it was no longer statistically significant in the final analysis. The following variables were associated with higher rates of incarceration during VC program participation: charges of probation or parole violations, compared with no such charges; end of VC participation because of incarceration for a new arrest or case transfer by the legal system, compared with no subsequent legal involvement at the time of program termination; and requiring mental health follow-up but not undergoing treatment, compared with not needing any follow-up. Lower rates of incarceration were associated with having stable housing, compared with being homeless, and with program referrals to substance abuse treatment.

Discussion and Conclusions

The purpose of this study was to examine factors associated with incarceration among VC participants. Overall, three-quarters of veterans participating in these courts exited successfully. Our core findings were that stably housed veterans were at lower risk of incarceration, as were veterans referred for substance abuse treatment. Conversely, veterans who required mental health follow-up but did not undergo treatment had a higher risk of incarceration.

This study represented a thorough review of an extensive set of recidivism data. It provides expanded empirical information about VC programs and their structure that can serve as guidance for VCs, whereas little guidance is currently available. Results also indicate future areas for exploration in this burgeoning field of VC research.

Regarding our hypotheses, the use of behavioral contracts as a sanction was associated with lower rates of incarceration during VC participation. However, the level of association was no longer significant after the Bonferroni adjustment was applied. Prior research has shown that 55% of VC programs utilize behavioral contracts (16).

An unsurprising finding was that the stably housed were at lower risk of incarceration than homeless veterans during VC participation. This finding was demonstrated in both pre-VC entry and post-VC entry data, lending further support to national initiatives to provide stable housing for veterans.

In addition, the risk of incarceration was lower in VCs that referred veterans to substance abuse treatment. Prior research has shown that 99% of VC programs utilize this approach (16).

Of the variables associated with increased incarceration risk, one of the most impactful was requiring mental health follow-up but not undergoing treatment. This finding may suggest a need for increased efforts to coerce or entice veterans by more assertive means into needed mental health follow-up. However, information regarding treatment needed, available resources, and potential barriers to care remains unknown. Therefore, this finding should be interpreted with caution, because access to mental health treatment depends on a variety of factors.

The method of enforcing treatment participation and the specific treatments available in VCs were unknown. Future investigations should include a more detailed analysis of the individual court processes, treatment types and their availability to participants, and potential barriers and facilitators of treatment engagement. These court-related factors likely have implications for program outcomes.

In addition, the findings of lower incarceration rates for VCs that referred veterans to substance abuse treatment warrants further investigation. Specifically, 99% of the courts reported referring veterans to substance abuse treatment 100% of the time, and 1% reported referring to treatment 75% of the time. We view these high referral rates as a limitation, because it is difficult to accurately interpret the meaning of this finding. In the few courts that reported not referring veterans to substance abuse treatment, potential causes include a lack of education on the part of the court as to the value of a referral to substance abuse treatment, a lack of service availability, or documentation errors. In addition, the high level of referrals to substance abuse treatment suggests that further investigation is needed to determine how often a referral results in treatment and what those treatments are.

Limitations of our study also included the utilization of self-reports of representatives of VCs or VJO programs for the phone survey measure. VJO specialists are responsible for tracking admissions, exits, and terminations of veterans receiving VA services during VC participation, and because they are treatment team members, they generally have access to such information about a court’s defendants who are not receiving care within the VA system. Because they are not court employees, specialists may be more objective in their reports of court policy and procedure. However, the phone survey’s reliance on noncourt staff to describe court policy and procedures may also be seen as a limitation.

The lack of follow-up data post-VC involvement is another limitation and highlights the need for future investigation that would include databases of the criminal justice system. Criminal justice information would allow more specificity about reincarceration outcomes that could further inform the outcomes for these programs. Such information may offer further insight into veteran and court characteristics that affect the likelihood of recidivism after exit from a VC program.

This study used surveys completed by VC-affiliated VA staff regarding the participants with whom they worked. However, some VC participants may not be eligible for VA services or may opt to receive care outside of the VA, and the lack of data from these participants was a limitation of this study. Given the rapid proliferation of VCs, future studies will likely inventory larger numbers of courts and thereby benefit from enhanced statistical power.

Despite its limitations, this study offers an extensive review of national VC recidivism data. Furthermore, the data can be used in the development and improvement of this critical component of rehabilitation of veterans involved in the criminal justice system.