A Dark Figure of Corrections

Abstract

Given recent fiscal issues and the continual struggle to reduce the nation’s overuse of incarceration, a renewed focus has been placed on the efforts of community corrections and alternative sanctions. Halfway houses represent a common and, until recently, infrequently evaluated intervention for inmates returning to the community. Although the model has advanced over the years, often providing an array of treatments and services, scant research has examined the impact such programs have on participants’ success in the community. These and other interventions like them, although providing a needed service, create additional avenues for failure and recidivism. However, failures that result in a return to prison are rarely disentangled, representing a “dark figure” of corrections. The current study explores failure types, prevalence, and competing risk predictors for a sample (N = 580) of halfway house participants. Findings both explore and describe the added and varying risks associated with participation community corrections interventions.

Keywords

reentry recidivism rehabilitation competing risks halfway house

For most states, the process of offender reentry has become a primary area of concern. Overall program effectiveness, efficiency, and reducing costs are among the chief concerns facing state departments. Justice administrators are perpetually reminded of the importance of successful reentry, especially in light of recent fiscal problems plaguing governments and correctional systems nationwide. Generally, reentry programming has been developed to meet the needs of correctional participants in an attempt to prevent future reintegration failures. Although types and styles of intervention vary from state to state, the practices of monitoring and evaluating interventions follow much of the same methodology.

From a state or Department of Corrections (DOC) perspective, the assessment of failure in reentry is understood in dichotomous terms: fail by way of returning to prison no matter the reason, or succeed by not returning and staying out (Ostermann, 2011). Thus, effectiveness is evaluated by whether the offender returns to prison. However, the inherent problem is the various types of failure or ways in which an individual can return to prison. Return types can range from violating a condition of supervision such as breaking community corrections system or intervention rules to committing a new crime while on parole. These specific distinctions among return types have been widely ignored, representing a “dark figure” in corrections research. Such practice is not surprising from a DOC perspective, as the distinction does not require a critical view; that is to say, a return is a return no matter how one codes it. This perspective tends to overlook, however, the potential for each return type to have a different antecedent cause and therefore different preventive needs depending on when and where they occur during the reentry process.

The type of potential returns, or failures, may take many forms, such as not reporting, breaking curfew, absconding from treatment, fighting, dirty urine analysis, insubordination, or failing to adhere to the conditions or rules of the program itself. Furthermore, failures that parolees can face often depend on their supervision/reentry plan. Community corrections conditions are often tailored to the risks and needs of the participants. As is the case of offenders with substance abuse issues, the provision of treatment in postrelease supervision adds another opportunity for failure by requiring the individual to abide by the rules of the intervention and eventually complete the programming. Thus, by attending to the substance abuse treatment need of the individual, one increases the amount of observation and adds special supervision conditions for a participant to violate. This creates complications when attempting to capture an accurate rate of failure and even more so when trying to account for an individual’s risk of recidivating. There are essentially two additional ways people can fail simply by their placement in a residential intervention: program rule violations and escapes or “walk-aways” from treatment (Culp, 2005; Wojtowicz & Liu, 2006). These two failure types are often understood by officials to be the same as technical violations of parolees. Correctional logic views these failures separately from returns for new crimes (or new commitments) as they are believed to be predictors of future recidivism and not, in themselves, criminal events. In actuality, they are measures of noncompliance with conditions of supervision or status offenses (behaviors prohibited for adult parolees), not predictors of recidivism.

In the current study, we examine whether the various types of returns are categorically different? More specifically we investigate differences between types of violations, the predictive nature of the violations, and prevalence of returns to prison. The study confines its examination to those individuals transitioning from prison to the community via a correctional halfway house. Viewed as a common transition from prison, halfway houses provide unique opportunities for a prison return. In the sections that follow, we will first explore prior research featuring the general need to examine specific failure types as well as the major discrepancies between new crimes and technical violations. Next we examine varying timelines and predictors of reentry failures and how failure-predicting risks “compete” with one another as the individual continues through the residential intervention.

Considering different failure types

State correctional systems all measure outcomes of offender performance with regard to rule or condition adherence. It is assumed that if offenders progress through their prescribed stages of supervision without violating their conditions of release, they will be far less likely to recidivate upon termination of supervision (Latessa & Lovins, 2010). A major problem, however, is a general lack of understanding on the part of the state with regard to what failures predict. It is believed that an offender’s ability to refrain from violating conditions of release will predict later recidivism and that supervision provides a deterrent effect for recidivism (MacKenzie, Browning, Skroban, & Smith, 1999). Such an assumption relies on broadly defined outcomes (i.e., technical violations, rearrests, or reconvictions) (Maltz, 2001; Spivak & Sharp, 2008), which limits a general recognition that offender performance varies across supervision settings and strategies (Olson & Lurigio, 2000). This limitation can lead to two problematic situations: (a) inaccurate predictions of parolee behavior and (b) a neglected need for tailored parolee supervision. That is to say, an individual may be returned to prison simply because he or she refuses to complete a residential treatment program or due to continued association with other released felons. Both acts represent qualitatively different behaviors but are filed under the same failure type—technical violation. Such reasoning ignores the idea that the various predictors of the technical violations are still unaccounted (Gray, Fields, & Maxwell, 2001). Furthermore, there is scant evidence indicating if technical violations (of any sort) predict future criminal acts (Petersilia & Turner, 1993).

If we are to believe that community supervision and observation of conditional violations are a proxy for future criminal acts, some would argue that revoking participants for technical violations is a sound crime prevention strategy. The empirical support for this assumption, however, is sparse at best. Gray et al. (2001) found that specific characteristics predict technical violations, many of which are unrelated to new criminal violations. Additionally, Olson and Lurigio (2000) found that predictive factors associated with probation revocations and technical violations often vary in strength, depending on the type of outcome measure used (revocation, new arrest while on supervision, and technical violation). Without empirical justification, one is left to believe that revocations are an indication of a failing community corrections approach and thereby threaten the legitimacy of their use (Wodahl, Ogle, & Heck, 2011).

In addition, the rationale of using release conditions as indicators of supervision do not reflect the scope of standardized risk instruments currently implemented. The few tests of this assumption were not launched until the 1990s. One such examination conducted by Hartmann, Friday, and Minor (1994) focused on predictors of successful discharge and recidivism of 156 participants in a probation halfway house. Findings revealed less than one-third of the participants received a successful discharge. The remainder were “unsuccessfully” (59.6%) or “administratively” (8.3%) discharged. Ultimately, Hartmann et al. (1994) recognized that predictors of “in-program performance do not necessarily relate to long-term post-program performance” (p. 512). Similarly, Latessa and Travis (1991) compared the outcomes of halfway house participants and probationers under general supervision. Results showed that, despite halfway house residents being more likely to need and receive treatment, due to returns via technical violations, participants were significantly less likely than traditional probationers to successfully complete their probation. Therefore, it follows that to ensure the proper application and efficiency of supervision and/or treatment, one needs to recognize the added risk of failure associated with community corrections participation and its associated conditions of release.

Disentangling Reentry Failure

In order to adequately understand and adhere to the diversity of needs among parolees, the issue of operationalizing reentry failures must be addressed. Across most state DOCs and the Bureau of Prisons, assessments of failure often begin when there is some type of starting event (or release) followed by a return event (Sabol, Adams, Parthasarathy, & Yuan, 2000). Releases may include those granted community corrections or reentry by way of “maxing out” or being released without supervision (Donnelly & Forschner, 1984; Ostermann, 2011). When measuring failures, it is common practice to monitor participants during their time-at-risk in the community, including both time on supervision and the period following successful termination; although the duration of the follow-up period can vary, typical outcome evaluations range from 6 months to 3 years (Bureau of Justice Statistics, 2011).

However, as it is difficult to obtain lengthy follow-up periods of parolees, when citing the prevalence of recidivism most researchers rely on a select few, large, and well-recognized studies. One such piece is Langan and Levin’s (2002) Bureau of Justice Statistics report “Recidivism of Prisoners Released in 1994.” As a frequently cited piece in recidivism research, this report followed a large sample of released inmates from 15 states. Their main (and often cited) finding stated that two-thirds of released persons fail within 3 years (Langan & Levin, 2002). However, this “two-thirds” figure is in reference to the 68% of released prisoners who are simply rearrested, whereas only half return to prison. Furthermore, the piece fails to account for the rationale behind the returns or failure types, as only a fraction were returned for new crimes committed.

Though their methodology was relatively strong and their findings well received, there are two essential problems with the frequent citation of this report: the sample used 1994 releasees, and the data do not take into consideration specific return types. Considering these issues, there are a number of pitfalls for current reliance on its findings. First, being based on releases from the 1990s, the study can be viewed as a “snapshot” of reentry policies and practices of that era. For instance, the effects of new policies substantially affected the use of discretionary parole boards, leaving more inmates to max out their sentence through determinate sentencing and relatively less on parole supervision (Travis & Petersilia, 2001). Similarly, it does not account for the increase of max-outs that occurred due to the “truth in sentencing” clause of the Violent Crime Control and Law Enforcement Act of 1994. Second, policies, interventions, and other rehabilitative services have been developed and implemented since the release of their report (i.e., the continuum of care) documenting increased success with reentry populations. Third, the size of the community corrections population nationally has continued to grow as more and more offenders are released from incarceration. Fourth, studies have yet to make an attempt to replicate their findings. Lastly and of great concern for the current study is the inclusion of all technical violation and program failure types in their measure of recidivism. However, in spite of such shortcomings, most researchers needing to cite a credible recidivism study are likely to utilize this report without a detailed understanding of how failure is defined. This has led many policy makers to infer (likely inadvertently) that most released prisoners commit new crimes, fail, and return to prison within 3 years, thus suggesting that reentry populations are far riskier than the evidence indicates.

Because of this lack of understanding of failure specificity, vague definitions and measures persist in restricting corrections officials’ ability to tailor strategies to the risks of each participant. By their nature, residential halfway houses require more intense and stringent supervision than general community corrections supervision. Subsequently, as the released individual progresses through the intervention, there is an increased likelihood of failure due to violations specific to halfway house participation (English & Mande, 1991; Olson & Lurigio, 2000). Prior research has identified that the high-intensity supervision provided in residential treatment (e.g., halfway houses) can negatively affect low-risk persons whom the intervention was not designed to serve (Lowenkamp & Letessa, 2005; Solomon, Kachnowski, & Bhati, 2005). Recent findings from California have revealed that predictors and prevalence of parolees returned to prison for noncriminal events (i.e. technical violations) largely include those with substance abuse issues (Grattet, Petersilia, & Lin, 2008). Findings from specialty courts and intensive supervision interventions have also described the potential for iatrogenic effects of intensive correctional programming through what has been termed supervision effects (Hamilton, 2011a; Petersilia, 1999; Solomon et al., 2005). Although well intended (i.e., to ameliorate addiction), these special conditions place additional barriers to successful termination of supervision. Unfortunately, current community corrections practices often fail to recognize the unique risks of those released to such interventions, thus increasing the potential for iatrogenic effects resulting from increased supervision (Hamilton, 2011a; MacKenzie, 2000; Petersilia & Turner, 1993).

The Impact of Halfway Houses on Reentry Timelines

When the decision is made to release an inmate to supervision, there is typically a decision as to how, who, and where the person will be monitored. Whether decided by a discretionary parole board or given mandatory release, many states use halfway houses, where a facility provides housing and sometimes programmatic needs to recently released offenders. Halfway houses play a critical role in the correctional continuum of care, bridging the gap between prison and community life. Originally, halfway houses served as boarding houses or shelters; as they have grown and become more specialized, this venue can often be viewed as a staging ground for many of the goals achieved during in-prison treatment (Moos, Pettit, & Gruber, 1995). Here, goals and lessons learned while incarcerated are then applied in conjunction with supervision in community settings. Treatment at this stage of the continuum is often the most critical as the reintroduction in an autonomous environment may trigger relapse or cause other temptations to reappear (Mears, Winterfield, Hunsaker, Moore, & White, 2002). For many offenders, special requirements based on assessed risks are applied to their release conditions, including obtaining employment, paying arrearages and fines, returning to school, and remaining substance free (Hser, Evans, Huang, & Anglin, 2004).

While incorporating the resources of halfway houses, there are a few ways of structuring the methods of supervision. Perhaps one of the best (though dated and not necessarily exhaustive) depictions of halfway house models was that presented by Seiter and colleagues (Allen & Seiter, 1976; Seiter, Carlson, Bowman, Grandfield, & Beran, 1977). One common model presented is where upon being granted early release from prison, the individual spends the initial period of his or her supervision in the halfway house. In this model, individuals are supervised by a community corrections officer but reside in the halfway house rather than a personal residence. Here, failure timelines begin when a participant is released from prison.

A second model releases individuals through the halfway house, whereby participants are to spend a portion of the prescribed supervision time in the halfway house as a type of “test” regarding their readiness for the community (Latessa & Allen, 1982). Following this readiness test, the individual may then be granted parole/community corrections release. Failure timelines associated with this model often do not begin until the inmate is released from the halfway house. A final model follows the same recidivism tracking process as the first. However, instead of residing in the halfway house for the initial part of release as a way of aiding reintegration, the halfway house is used as a “safeguard,” or alternative sanction, and is provided only if releasees regress while under general community corrections supervision.

Competing Risks of Reentry Failure

The concept of the competing risk was best described by Allison (1984) in which two or more failure events compete with one another. In the current model, we focused on comparing noncriminal failure with criminal failures. As discussed previously, the extended duration of treatment and increased supervision creates a series of additional failure risks that compete and may result in a return to prison that is unrelated to the commission of new offenses. Among the possible failure types, the current study is concerned with understanding how noncriminal failures compete with criminal failures. Noncriminal failures are due to community corrections condition violations, which may occur either in a halfway house or under traditional supervision. With the increased observational element of residential treatment, there is a greater risk for specified early failure types, such as violation of programming rules and walk-aways or “escapes” (see Culp, 2005; Wojtowicz & Liu, 2006). For those persons under traditional supervision, noncriminal failures are classified among the various types of technical violations, which result from violating the conditions of traditional community corrections (parole) supervision. Criminal failures, on the other hand, involve re-incarceration resulting from new crimes (most often felonies) occurring during and following supervision. Although other studies may conceptualize and disentangle violations differently, considering them in their totality, the described failures can be generalized into four distinct return types: escapes (walk-aways), halfway house (program) violations, (parole) technical violations, and new commitments (resulting from new crimes).

Gaps in the Literature

The current research is bifurcated, dated, and incomplete on many fronts when addressing reentry failure, understanding technical violations, operationalizing time-at-risk, and recognizing the impact of competing risks. Research that examines reentry often lacks specificity and does not attempt to disentangle failure beyond the simple dichotomy of returning or staying out of prison. In the limited number of studies that parsed out technical violations from new criminal commitments, the type of violation is rarely considered and often relies on vague operationalizations. Similarly, most supervision and risk levels are accepted as assigned, with little regard to the potential interaction of offender needs and provided interventions (Solomon et al., 2005). Moreover, the broad category of “technical violations” is assumed to predict future recidivism despite the fact that few have actually challenged this notion. In mainstream application, researchers and practitioners rely on such assumptions that may lead to inaccurate predictions of participant behavior and ignore the need for tailored supervision.

Furthermore, research tends to overlook the true effectiveness of intense postrelease supervision, particularly among substance abusers. The majority of research that focuses on risks associated with supervision applies the analysis toward probationers rather than those released from long-term incarceration. The few studies that have emphasized released inmates do so in broad terms and do not address the impact of specific competing failures, let alone for those mandated to participate in residential treatment. Lastly, studies that highlight the role of halfway houses are sparse at best. Those studies that do exist do not consider the timing of failure patterns and how they differ across failure types.

The Current Study

The current research connects these related themes by examining multiple types of failure events among community corrections participants that are both criminal and noncriminal. Our intent is to explore and describe both the timing and predictors of each event type within the halfway house context described by the first model presented by Seiter et al. (1977). Thus, our primary research questions investigated are as follows: What is the prevalence and timing of each failure type for halfway house participants? When attempting to predict prison returns, do predictors differ based on type of failure? Related to the primary study questions, we also discuss three rarely contested correctional ideals: (1) Is there a specified distribution of reentry failure? (2) Are the various types of returns categorically different? (3) Are noncriminal failures predictive of new criminal events or nothing more than an unrelated umbrella of behaviors representing noncompliance?

Method

Study Setting and Data

In 2000, the New Jersey DOC, Department of Drug Programs instituted a policy to provide a “continuum of care” for all incarcerated substance-abusing offenders. The continuum is provided through several stages of treatment interventions. First, offenders were screened for substance abuse need upon prison entry. Need for treatment was determined by a score of 5 or greater on the Addiction Severity Index (ASI). If need was identified, within the last 9 to 24 months, inmates were placed in any 1 of the 10 possible prison-based therapeutic community (TC) treatment programs.¹ Following completion, participants applied for community release and were sent to a community assessment center, where a battery of instruments and measures were used to collect participant information within several behavioral domains. Participants were then placed in 1 of 16 halfway houses in the state. All halfway houses employed were accredited by the state to provide substance abuse treatment services. Although each program may differ in terms of the programmatic philosophy and ancillary services, all programs provided an intensive treatment regimen requiring residency. The duration of halfway house participation ranged from 3 to 18 months. If compliant with program conditions, participants were then placed on parole supervision. There is no pre-established range of parole duration as the amount of time an offender participates is based on sentence length. Following the completion of a parole term, correctional supervision is terminated.

There were four eligibility criteria for study inclusion. All participants (1) participated in the New Jersey DOC continuum of care (attended prison-based TC treatment, received an assessment, followed by halfway house treatment, and were eligible for parole), (2) did not possess a diagnosed Axis I mental illness, (3) were male (omitting females as they did not all possess assessment needed for study analyses), and (4) were released from a New Jersey State prison facility between 2001 and 2003. The study used a retrospective purposive sample of participants based on the described eligibility criteria. New Jersey DOC offender record files provided participant movement and failure data, and community-based assessments provided pre-intervention risk factors.

Predictors of Failure

Participants’ assessments provided a detailed description of pre-intervention risk and protective factors. Prior findings of predictors indicated by Andrews, Bonta, and Wormith’s (2006) Central Four and Big Eight, in addition to proper model building procedures, were used to select the assessment items used in the multivariate analysis; operational definitions are described in Table 1.²

Table 1:

Predictors of Failure

Item – Description	Measurement Type – Values/Range
Age – Age at time of release measured in years	Continuous – Range 19 to 60
Children – Does the participant have children	Dichotomous − 0 = no, 1 = yes
HS/GED – Possesses a high school diploma or GED at time of release	Dichotomous − 0 = no, 1 = yes
History of education problem – History of grade repetition, learning disorder, or expulsion	Dichotomous − 0 = no, 1 = yes
Total number of arrests – Total number of prior arrests	Continuous – Range 1 to 22
Family deviance history – Immediate family member has prior conviction and/or substance abuse history	Dichotomous − 0 = no, 1 = yes
Prior treatment – Participated in substance abuse treatment prior to current incarceration	Dichotomous − 0 = no, 1 = yes
Prior incarceration – Two or more prior incarcerations	Dichotomous − 0 = no, 1 = yes
Drug age – Participants’ age of first drug use	Continuous – Range 7 to 22
Daily use – Prior to incarceration participant was using substances daily	Dichotomous − 0 = no, 1 = yes
LSI-R – General recidivism risk score computed with the Level of Service Inventory–Revised	Continuous – Range 0 to 54
ASI – Score on the Addition Severity Index	Continuous – Range 1 to 10
Prior felony – Convicted of a felony prior to the current incarceration	Dichotomous − 0 = no, 1 = yes
Prior juvenile – Convicted as a juvenile prior to the current incarceration	Dichotomous − 0 = no, 1 = yes
Companion risk – Rated as high risk on the LSI-R subscale	Dichotomous − 0 = no, 1 = yes
Instant offense – Conviction type for current incarceration	Nominal − 0 = drug, 1 = violent, 2 = weapon, 3 = property, or 4 = parole violation
Primary substance – Preferred substance of choice reported	Nominal − 0 = heroin, 1 = cocaine/crack, 2 = marijuana, 3 = alcohol, or 4 = other drug
Sentence length – Duration of incarceration in years	Continuous – Range 1 to 7
Co-occur – History of co-occurring mental health issue	Dichotomous − 0 = no, 1 = yes
Treatment benefit – Participant reports he or she would benefit from further treatment	Dichotomous − 0 = no, 1 = yes
WPT – Wonderlic Personnel Test rating of vocational aptitude	Continuous − 1 to 50

Failure Events

The outcomes collected during the halfway house intervention, parole, and postparole, represent the study’s dependent variables. Consistent with the definition presented by Langan and Levin (2002), we operationalize a failure to be any return to incarceration following the initial study prison release date, either due to a new crime or a violation of community corrections conditions (i.e., noncriminal failure). Failure outcomes are further broken down by type, including halfway house violations, escapes/walk-aways, technical violations, and new commitments (felony conviction). Study definitions for each failure type are presented in Table 2. The absence of failure, during or following the intervention, was operationalized as participant success. Although ideally one should utilize the date of the criminal or noncompliance event as the date of failure, this type of data is not routinely collected and rarely available to a DOC; hence, the use of recapture date was used as the date of failure and is a typical measure for many analyses of reentry populations.³

Table 2:

Prison Return Outcomes

Item – Description	Measurement Type – Values/Range
New commitment – New criminal conviction incurred following release, which results in a return to prison	Dichotomous − 0 = no, 1 = yes
Technical violations – Violations of parole conditions resulting in a return to prison, which may include failure to report, substance use, insubordination, association with failure to obtain employment or stable housing, and failure to participate in treatment/services	Dichotomous − 0 = no, 1 = yes
Halfway house violation – Violation of halfway house facility rules incurred while residing in the halfway house resulting in a return to prison, which may include substance use, sexual misconduct, insubordination, and failure to participate in treatment/services	Dichotomous − 0 = no, 1 = yes
Escape – An unauthorized exit from the halfway house (i.e., walk-away), where upon recapture, results in a return to prison	Dichotomous − 0 = no, 1 = yes
Time to failure – Days from release to recapture	Continuous – Range 0 to 1,825

Analytic Plan

As the timing of failure types and their predictors are hypothesized to be important, event history models were selected because of their precision in estimation of predictor variables, the determination of predictors’ temporal order, and their ability to provide an appropriate model for nonreturning participants (censored cases) (Kruttschnitt, Uggen, & Shelton, 2000). To address the first research question, we examine the prevalence and timing of each event through an examination of the cumulative hazard plots of each event type. It was anticipated that the timing and shape of the cumulative hazards will differ. We therefore examine the potential failure distributions, identifying the underlying rate of each failure. To investigate the second study question, we fit multivariate survival models. First we fit a single model with the general failure event: Any Return. This initial model represented an examination of risk without the consideration of failure type. We then fit competing risks survival models comparing each noncriminal prison return failure event to new commitments. In this article, the model created could be conceived as a multinomial version of a survival model, where model predictors indicate a greater hazard for failing of one type of event versus another. We model nonrepeatable competing events, where, in a given model, if one failure event occurs for a particular participant, the other failure events are unobservable for that participants and the case is right-censored in those models. Competing risks models were computed utilizing the mstate package in R.⁴

Time to Failure

The current study operationalizes the start time for the evaluation of failure as the day of release from prison. “Time to Failure” is measured in days, and the failure event is measured as the day of recapture (i.e., the arrest or capture from an escape that resulted in a return to prison). Although some have operationalized “time at risk” differently, claiming that the time spent in a halfway house should not count as community corrections failure (Lowenkamp & Letessa, 2005), the intent of this analysis was to examine failure during and following halfway house participation.

Missing Data

Likely due to minor errors in data entry, predictor variables collected from community assessments were not entirely complete. Nearly 33% of the sample possessed at least one missing value; however, the vast majority (66%) of participants with missing values was missing two or fewer predictor values. All participants possessed complete data with regard to release and recapture dates as well as failure type. Overall, the data set was found to be nearly 88% complete, which is well within the acceptable range for imputation identified by Rubin (1987). Missing patterns were analyzed, and based on theoretical and empirical findings (i.e., Little’s test for Missing Completely At Random [MCAR]), it was determined that missingness was of the “missing at random” type. Consequently, to maintain sample size and statistical power, multiple imputation procedures were conducted using the MICE package in R. Values that were missing on continuous measures were imputed with linear regression, dichotomous measures were imputed with binary logistic regression, and nominal measures were imputed with multinomial logistic regression. Based on Graham, Olchowski, and Gilreath’s (2007) recommendations, the imputation process was repeated 20 times, with each repetition producing a separate data set. The analyses presented represent combined results of the collective imputed data sets using pooling procedures.

Results

Bivariate Patterns

To answer the first study question, we began by examining the prevalence of each failure event and conducted bivariate comparisons of predictor characteristics among the five possible outcomes. Several significant findings were revealed and are presented in Table 3. Over one-third (37%) of participants were “successes” and did not return within the 5-year follow-up, whereas 50% were returned for noncriminal revocations and just over 13% returned due to a new commitment based on a new felony conviction (i.e., new crime). Age was found to vary significantly among outcome events, with younger participants being more likely to walk away and older participants less likely to be returned to prison (p < .001). When comparing the instant offense, or the offense that resulted in the initial incarceration, participants who committed drug offenses were more likely to commit new crimes and less likely to have parole revoked for technical violations (p < .001). On average, participants who were not returned were also older when they began their drug use (p < .001). Those participants who fail due to the commission of a new crime were also more likely to have a prior escape documented (p = .039). Participants revoked on technical violations were more likely to have experienced education problems during their youth by comparison to other event types (p = .004). Those who walk away from halfway house interventions were more likely to have been arrested as a juvenile (p < .001). Parolees who were not returned were less likely to possess a felony conviction prior to the initial incarceration (p = .001). Those who violated while participating in the halfway house were more likely to have experienced a family history of abuse, and those who did not return were less likely to have such a history (p = .002). Finally, the median days-to-event varied significantly, with failure resulting from halfway house violations occurring first, followed by escapes/walk-aways, revocations for technical violations, and prison commitments for new crimes (p < .001).

Table 3:

Bivariate Relationships Among Five Outcome Types (N = 580)

	Halfway House Violation (n = 49)			Escape/ Walk-Away (n = 89)			Technical Violation (n = 152)			New Commitment (n = 77)			Not Returned (n = 213)
Domain / Measure	%	M	SE	%	M	SE	%	M	SE	%	M	SE	%	M	SE	χ² / F Test	P
Demographics
Age		31.5	0.9		29.5	0.7		31.9	0.6		31.0	0.8		33.7	0.6		<.001
Race/ethnicity
Black	76.9			78.2			71.1			75.0			71.6			.908
Hispanic	12.8			8.0			11.2			11.9			11.5
White	10.3			13.8			17.8			13.1			16.8
Family/background
Children	71.4			67.9			67.5			75.4			75.3			.515
Family physical and/or sexual abuse	27.3			29.4			12.8			9.4			17.4			.002
Parental drug abuse and/or conviction	72.7			85.9			78.7			70.6			72.8			.093
Criminality history
Total number of arrests		11.6	1.1		11.0	0.6		11.6	0.5		11.6	0.6		10.7	0.4	.915
Arrested as a juvenile	48.5			74.1			58.5			60.0			45.1				<.001
Prior adult felony	54.5			50.6			55.5			47.1			35.2			.001
Sentence length (in years)		5.6	0.2		5.6	0.1		5.6	0.1		5.5	0.1		5.4	0.1	.268
Instant offense Drug	60.6			54.1			46.3			75.3			65.3				<.001
Violent	15.2			16.5			17.1			10.6			11.7			.480
Property	12.1			8.2			14.6			4.7			7.5			.069
Weapon	3.0			4.7			3.0			1.2			2.8			.758
Violation	9.1			15.3			16.5			7.1			11.7			.232
Prior incarceration	61.4			64.7			61.8			58.6			56.1			.703
Prior community corrections attempt	45.5			42.1			40.4			42.9			43.7			.945
Corrections escape history	12.1			9.4			12.8			20.0			7.5			.039
Substance use
Age first drug use		15.0	0.5		15.4	0.3		15.5	0.2		15.0	0.3		16.5	0.2		<.001
Daily drug use	75.8			58.8			65.2			69.4			62.4			.361
ASI score		5.9	0.2		6.1	0.2		5.9	0.1		6.0	0.2		6.0	0.1	.667
Primary drug
Heroin	27.3			16.5			31.1			18.8			24.4			.077
Cocaine	12.1			11.8			11.0			10.6			12.2			.994
Marijuana	39.4			50.6			35.4			48.2			40.8			.131
Alcohol	15.2			9.4			16.5			12.9			13.1			.477
Other	0.0			8.2			4.9			4.7			6.1			.636
Prior treatment episode		0.6	0.1		0.3	0.1		0.5	0.0		0.5	0.1		0.4	0.0	.060
Treatment motivated	51.5			61.2			59.1			49.4			46.5			.064
Co-occurring mental health issue	12.1			10.6			17.7			7.1			12.2			.166
Education/vocation
High school diploma/GED	63.6			48.2			62.2			57.6			63.4			.157
History of education problems	39.4			41.2			56.1			32.9			40.4			.004
Employed 12 months prior	69.7			63.5			70.7			61.2			74.2			.163
WPT vocational aptitude		15.8	0.8		14.3	0.7		13.4	0.5		13.8	0.6		14.0	0.4	.276
Peers
LSI-R Companions subscale	33.3			32.1			40.0			32.3			38.0			.694
Risk assessment
LSI-R Overall score		31.5	0.9		30.9	0.6		30.4	0.4		30.3	0.2		29.9	0.4	.410
Outcome
Days until prison return (median)		170.0	33.9		182.0	22.0		476.0	38.6		1,187.0	129		1,825.0	11.4		<.001

Note: WPT = Wonderlic Personnel Test; LSI-R = Level of Service Inventory–Revised.

Comparison of Event Hazards

Next, we examined the timing of each failure event. Given the significant trend in the median time-to-event, we then investigated the underlying hazard ratios of each failure type without consideration of covariates. This allowed us to examine the timing of each event. Figure 1 presents a simple hazard plot describing the fluctuations of risk over time. The major finding to take away from this plot relates to the primary study questions. Specifically, during the 5-year follow-up, noncriminal failure events gradually decreased over time, whereas the hazards for new crime gradually increased. Given that halfway house participation does not typically last beyond 24 months, it is expected that hazards relating to failure in the halfway house will become constant following the 2-year mark. One unanticipated finding was the spike in hazards for revocations by way of technical violations at Year 1 and a staggered decline thereafter. A similar but inverted hazard trend for new crimes was also observed, with a shallow trend observed within the first 2 years and a sharp incline in the hazard at Year 3 and again at Year 5.

Figure 1.

Hazard Plot of Prison Return Type

As simple hazard plots are often too erratic to decipher, Figure 2 presents cumulative hazard plots for easier pattern recognition. Again, flat plot lines for failures can be seen within the halfway house cohort at around 18 months. For technical violations, the cumulative hazard continues to increase for the entire study period. Although this increase is sharper within the first 24 months, it then slows through Year 5. The reverse is shown for new commitments, where there is a slow increase observed for the first 30 months and a sharper increase for the remaining study period.

Figure 2.

Cumulative Hazard Plot of Prison Return Types

Assessing Survival Distributions

An often uncontested aspect of survival modeling in criminal justice research is the appropriate use of Cox regression models. When an underlying distribution of the dependent variable is not known, analysts will typically use a Cox regression model, which makes no assumption as to the baseline hazard. The variations displayed in Figures 1 and 2 indicate that distribution assumptions among failure types likely differ, and thus, a survival model distribution type (other than the Cox model) must be identified. To identify the appropriate distribution model, Akaike Information Criteria estimates were compared for each failure type and are presented in Table 4. Estimates indicated similar findings for selecting an underlying distribution for failure types. The general (Any Return) model as well as the three noncriminal failure types were identified to possess a log-normal distribution. This finding is consistent with the cumulative hazard plot findings displayed previously, as log-normal distributions are said to more adequately accommodate baseline hazards that initially increase and then decrease over time (Hosmer, Lameshow, & May, 2008; Mills, 2011). A Gompertz distribution was found to be the best fit for new commitment failures. The Gompertz distribution was originally proposed to model human mortality and, like the better known Weibull distribution, assumes a monotonic hazard shape, where the hazard may only increase or decrease over time. Consistent with this assumption, and again supported by the findings of the cumulative hazard plot, new commitments possess an accelerated failure time, where events accumulate slowly and progressively increase in frequency as follow-up time extends. However, the study follow-up time ends at Year 5 and hence does not capture the anticipated decline in hazard for new crimes over time. Although statistically appropriate, we feel the use of this hazard distribution, which indicates that risk of any failure may only increase overtime, is theoretically inappropriate. Findings of life course theory have repeatedly indicated that as an individual continues to age, the likelihood of any criminal event decreases (Farrington, 2008). That is, one would not expect the hazard to continue to rise if the follow-up was extended another 5 or 10 years and that our follow-up period, although sufficiently long by most standards, does not extend far enough to capture the eventual decline in hazard for this failure type. Therefore, to aid in interpretation between models and to provide theoretical consistency, a log-normal distribution was utilized for all regression models.

Table 4:

Comparison of Akaike Information Criteria Model Fit Estimates

Distribution	Any Return	Halfway House Violation	Escape/Walk-Away	Technical Violation	New Commitment
Exponential	6,002.7	1,050.2	1,710.2	2,977.0	1,673.9
Weibull	5,993.3	1,024.7	1,675.5	2,978.9	1,627.8
Log Normal	5,954.7	1,019.2	1,661.0	2,955.9	1,636.3
Log Logistic	5,964.3	1,023.2	1,670.1	2,962.2	1,630.0
Gaussian	6,394.1	1,144.1	1,882.0	3,155.6	1,634.8
Gompertz	6,004.7	1,052.2	1,712.2	2,979.0	1,621.9
Gen. Gamma	6,018.5	1,026.0	1,679.0	2,992.9	1,624.8

Note: Bolded figures identify the best model fit for a given outcome.

Competing Risks Models

To examine the second study question (Do predictors differ based on type of failure?), four multivariate survival regression models were fit. A global model for Any Return was fit first, which analyzes event times by collapsing all four event types into a single dichotomous measure of failure. This model served as a comparison for the more specified event failure models and represents a typical formulation of recidivism (or community corrections failure) when event type is ignored. The remaining models were utilized to examine if predictors of prison return differed by specified failure type. That is to say, do participants who commit noncriminal failures possess qualitatively different risk predictors than those who return on a new crime (i.e., new commitment)? If it is appropriate to group all failures singularly, then one would expect the significant predictors of each noncriminal failure event type to be similar for both Any Return and New Commitments.

The first two columns of Table 5 present results of the Any Return model. Overall, the model was found to be significant (LL = -3003.6, p < .001), with 14% of the variance in prison return explained by the model predictors. A total of 367 of the 580 study participants (63%) returned to prison within 5 years of release. However, of the 23 included predictors, only 4 (static) measures were found to significantly predict prison return where younger participants (p < .001), those with a prior adult felony conviction (p < .05), a prior incarceration (p < .01), and prior treatment episode (p < .01) possessed a greater propensity for return.

Table 5:

Competing Risk Analysis of Failure Types (N = 580)

	Any Return		Halfway House Violation		Escape / Walk-Away		Technical Violation		New Commitment
Parameters	Logit	SE	Logit	SE	Logit	SE	Logit	SE	Logit	SE
Age	−.05	.01***	−.07	.06	−.12	.04**	−.03	.02⁺	−.01	.01
Children	−.22	.17	−.37	.70	−.16	.44	−.28	.21	.18	.18
Family physical and/or sexual abuse	−.09	.24	.57	.83	.25	.63	−.45	.33	−.16	.18
Parent drug abuse and/or conviction	.27	.20	.49	.73	1.12	.60⁺	.14	.28
Total number of arrests	.03	.02⁺	.03	.06	.09	.04*	.01	.02*	.01	.02
Arrested as a juvenile	.12	.16	−.75	.66	.85	.48⁺	−.01	.21	.20	.16
Prior adult felony	.39	.17*	.78	.65	.38	.43	.43	.20*	.02	.15
Sentence length (in years)	.04	.09	.36	.36	.23	.23	−.02	.11	−.08	.07
Instant offense (reference = drug)	—		—		—		—		—
Violent	.09	.24	−1.35	1.09	.74	.63	.23	.31	.00	.22
Property	.49	.26⁺	.45	1.01	−.11	.76	.60	.49**	−.31	.32
Weapon	.40	.40	−1.19	−1.87	1.58	.96⁺	.85	.31	−.56	.52
Violation	.14	.22	−.96	1.06	.44	.57	.41	.28	.04	.21
Prior incarceration	.49	.18**	.13	.78	1.44	.50**	.27	.25**	.15	.18
Prior community corrections attempt	−.14	.16	.11	.70	−.43	.45	−.08	.21	−.15	.15
Corrections escape history	.33	.26	.73	1.26	−.47	.73	.47	.33	.42	.23⁺
Age first drug use	−.02	.03	−.11	.12	−.01	.08	.01	.04	−.05	.03*
Daily drug use	−.09	.16	.15	.75	.15	.75	−.21	.21	.15	.15
ASI score	−.04	.06	−.17	.24	.26	.15⁺	−.14	.07*	−.02	.05
Primary drug (reference = heroin)	—		—		—		—*		—
Cocaine	−.21	.25	−.84	1.13	.19	.70	−.53	.32⁺	.25	.24
Marijuana	−.36	.22	−.84	.98	.34	.62	−.83	.29**	.18	.22
Alcohol	.07	.24	−.37	.97	.22	.68	.00	.29	.20	.22
Other	−.26	.35	−2.83	1.86	1.44	.89	−.77	.46⁺	.21	.34
Prior treatment episode	−.40	.15**	−.60	.61	−1.12	.42**	−.07	.19	.02	.13
Treatment motivated	.09	.16	−.96	.65	.42	.44	.18	.19	.08	.13
Co-occurring mental health issue	−.12	.21	−.36	.92	−.35	.59	.18	.25	−.38	.24
High school diploma/GED	−.31	.16⁺	.27	.67	−.92	.44*	−.12	.21	−.20	.15
History of education problems	.05	.16	.03	.66	−.45	.45	.54	.20**	−.33	.24*
L12 employment	−.04	.16	.46	.71	−.14	.42	−.02	.21	−.24	.15
WPT vocational aptitude	.01	.01	.11	.06⁺	.03	.04	−.01	.02	.01	.01
LSI-R Overall score	−.01	.02	.03	.66	−.02	.06	−.03	.03	.02	.02
Companions Subscale	−.02	.16	−.17	.66	−.36	.44	.29	.20	−.13	.15
Number of events	367		49		89		152		77
Model fit
Log likelihood^a	–3,003.60***		–474.28		–799.46***		–1,354.69**		–711.71*
Akaike Information Criteria	6,070.47		1,012.33		1,665.15		2,771.19		1,487.91
R ²	.14		.05		.13		.13		.09

Note: LSI-R = Level of Service Inventory–Revised; WPT = Wonderlic Personnel Test; ASI = Addiction Severity Index.

The likelihood ratio test of model significance is a pooled computation of the 20 data sets and uses the Li, Raghunathan, and Rubin (1991) formula for multiple imputed data and approximates and F distribution.

p < .1. *p < .05. **p < .01. ***p < .001.

When examining the competing risks models, one finds significant predictors vary between return types, providing evidence that the failure events are categorically different. The Halfway House Violation model was not found to be significant (LL = -474.28, p = .69), with only 5% of the variance in prison returns explained by the model predictors. The Escape/Walk-Away model was found to be significant (LL = -799.46, p < .001), with 14% of the variance in escapes/walk-aways from treatment explained via the model predictors. Specifically, participants who are younger (p < .01), have more prior arrests (p < .05), were previously incarcerated (p < .01), and previously attended treatment (p < .01) were predicted to escape/walk-away. Also, those lacking a high school diploma or GED (p < .05) are more likely to return to prison based on a recapture for this failure type.

The most frequent return type, technical violations (26%), revealed a significant model with several significant predictors (LL = -1,354.69, p < .01), and 13% of the variance is explained by the model predictors. Specifically, participants returning for technical violations were more likely to possess a greater number of prior arrests (p < .05), were more often convicted of a property offense (in comparison to drug offenders) (p < .01), had a lower ASI score (p < .05), were less likely to report marijuana as their drug of choice (compared to heroin) (p < .01), and possessed a history of educational problems.

Finally, only 77 participants (13%) were returned for a new commitment, and the model was also found to be significant (LL = -711.71, p < .05), with 9% of the variance in failures for new crimes explained by the model predictors. Only two of the predictors were identified as significant: Participants with a younger age of first drug use (p < .05) and those without a history of education problems being more likely to commit a new offense resulting in a return to prison (p < .05). The lack of significant predictors and a lack of overlap among predictors between this failure type and the others measured suggest that noncriminal returns are not strong predictors of returns for criminal events.

Discussion

The results provided substantial evidence to answer the two primary study questions: What is the prevalence and timing of each failure type for halfway house participants? When attempting to predict recidivism, do predictors differ based on type of failure? The analyses revealed four key findings associated with (1) recognizing the existence of a “dark figure” in community corrections data, (2) realizing the problems with generalizing technical violations and time-at-risk, (3) distinguishing and measuring the different distributions of reentry failures, and (4) properly sorting and predicting failure types.

A Dark Figure in Community Corrections

As mentioned previously, a well-cited study in reentry research is that of Langan and Levin (2002), which identified an often-repeated statistic that two-thirds of reentering inmates fail within 3 years of release. As mentioned previously, the lack of attention given to disentangling failure types prevents a true understanding of the effectiveness of common community supervision practices. Overclassification of offenders and extended observation can lead to deleterious effects. Although supervision effects have been discussed elsewhere (Hamilton, 2011a; MacKenzie, 2000; Olson & Lurigio, 2000; Petersilia & Turner, 1993), it is a concept that is understood but rarely investigated. Only recently has research attempted to examine the impact of technical violations on the community correction system (Grattet et al., 2008; Petersilia, 2005; Piehl & LoBuglio, 2005), suggesting that, if the objective of community corrections is to keep a person from committing new crimes, are we sure the system is truly achieving that objective by revoking community supervision for noncriminal events? Similar to the “dark figure of crime” that refers to the way methods used to collect criminological data leave us with a void of unknown criminal activity (Biderman & Reiss, 1967), we seek to shed some light on a similar dark figure of community corrections. In other words, this study describes how the general lack of understanding regarding community corrections failure has become an unobserved and thus an unknown quantity of reentry research. Like the dark figure of crime, which has less to do with crime than it does with implementation and empirical operationalization, the dark figure of community corrections has little to do with recidivism and a lot to do with additional constraints placed on offenders and how we measure them. Although restricted to halfway house participants, our study suggests (and likely others will confirm) that less than one-sixth (13%) of inmates return to prison for a new criminal event, and roughly half (50%) of all reentering inmates are returned because of noncriminal revocations. This finding of noncriminal return prevalence is comparable but somewhat higher than those previously reported for general parolee populations (Grattet, Petersilia, Lin, & Beckman, 2009; Herrschaft & Hamilton, 2011), where differences in findings are likely due to the specialized treatment conditions of current study participants. We hope that these prevalence figures provide the basis for future citations and less or at least a better informed presentation of Langan and Levin’s (2002) “two-thirds” failure figure.

Specificity and Timing

Expounding on the prevalence and reporting of noncriminal failure, we identify the trend that studies typically focus (obtusely in our opinion) on technical violations generally, without providing an understanding of why and when an offender is returned. As compliance with treatment and other rehabilitative objectives are added to a participant’s conditions of supervision, practitioners should be concerned with how these additions will affect the likelihood of revocation and return to prison. Within a continuum of care system, added special conditions (such as treatment) often provide greater intensity of observation and thus the potential for supervision effects (Hamilton, 2011a). As one would expect, these noncriminal returns have heightened risk of early occurrence. What is unexpected is the relative lack of risk for new crimes early in the release process, where all other noncriminal returns typically occur between 6 months and just over 1 year. This finding is confirmed by Grattet et al. (2009), who indicate that the riskiest time for all technical violations occurs within the first 180 days of release.

With the absence of a control group, one might claim that the crime control aspect of technical and other noncriminal revocations cannot be evaluated, as participants at risk of committing a crime were revoked before having the chance to do so. Although this is a valid argument, it is nonetheless quite remarkable that the median time for new crimes is a full 2 to 2.5 years beyond that of the noncriminal failure types. Prior findings suggest that the riskiest time for reentry failure is the first 6 to 12 months following release (Grattet et al., 2008; Langan & Levin, 2002; Solomon et al., 2005). If a noncriminal event revocation was to predict the occurrence of a future criminal conviction, then the likelihood of return should be similar for all types of failure events during this early “riskiest” time period. But this is not the case for recommitments resulting from new crimes. These findings are confirmed by Ostermann (2011), who found that “max-outs” spend significantly greater time in the community before reconviction by comparison to community supervision participants. However, like Ostermann, absent a randomized control group, it is difficult to do more than describe the pattern in which new crimes are typically infrequent and late-occurring by comparison to revocations. Understandably, the ethical issues of assigning a comparison group of participants to “no supervision” for the purposes of analysis make such a study design infeasible. Despite this limitation, what we can say with some certainty is that the “riskiest” time period suggested by prior research (Petersilia, 2005; Travis, 2005; Wright & Rosky, 2011) likely only reflects a risk of failure for noncriminal revocations and not for new criminal events (which, if they do happen, often occur much later).

Event Distributions

Related to the previous point, we also discovered a variation in the distribution of events. We find that failure times take a log-normal distribution shape as they accommodate the process where the risk of each event is allowed to both increase and decrease over time. Particularly in the case of technical violations, the log-normal shape allows for an initial increase (in Year 1) and then decrease (for Years 2 to 5) in the modeling of hazards. Although some have attempted to use alternate distributions to model the timing of recidivistic events (see Maltz, 1984, 2001), the vast majority of attempts to fit multivariate survival models use Cox Proportional Hazards models (e.g., Escarela, Francis, & Soothill, 2000; Ostermann, 2011; Wright & Rosky, 2011). A central reason for the popularity of the Cox model is that it does not require an analyst to specify a distribution. These models are less appropriate when one is concerned with testing how the event hazards change over time (Mills, 2011).⁵ Hence, if one is uncertain about the shape of the probability distribution, fitting a Cox model is a good start (Hosmer et al., 2008). However, if prior findings indicate that the timing of events takes on a particular distribution, using that specified distribution will allow for more accurate estimates and provide greater insight into the interpretation of events and coefficient estimates. Our hope is that the current findings will provide evidence as to how recidivism failure distribution times should be modeled.

Prediction and Failure Type

The second study inquiry examined if predictors differed by failure type. General and competing risks analyses were conducted, and findings revealed that predictors do in fact differ by return type. That is, if all failure types were simply collapsed into a dependent measure of “Any Return,” only four (static) items were found to be significant predictors. Some overlap of static items was identified where age, prior incarceration, and prior treatment predicted both Any Return and Escape/Walk-Away. Similarly, the item “prior adult felony” was a predictor of Any Return and Technical Violation. Though, in spite of the overlap identified, there were several items that were either only predictors of a specific failure event or were not identified as a significant predictor using the Any Return conceptualization of failure, including total number of prior arrests, instant offense, age of first drug use, ASI score, primary drug used, possessing a high school diploma/GED, and history of education problems. The main finding to take away from these analyses is that predicting a community corrections participant’s return to prison appears to be more complicated than a simple dichotomization of failure.

Practitioners must be made aware of these distinctions prior to placement in treatment, residential corrections facilities, and while monitoring individuals on community corrections supervision. Furthermore, although our findings may be less generalizable to nonsubstance-abusing participants, of all 23 items selected, only 2 (age of first drug use and history of education problems) were found to significantly predict new commitments. As mentioned, all predictors included in the model were carefully selected. Although most risk-prediction instruments weigh prediction parameters so that unique item significance with regard to outcome prediction is not a requirement for inclusion, it is surprising that standard predictors, such as the Level of Service Inventory–Revised (LSI-R) score, did not predict re-incarcerations for new crimes. This supports the notion presented by Gottfredson and Moriarty (2006) that using risk assessment tools to predict treatment outcomes is an improper application of such tools. Instead, separate needs assessments should be constructed to address the prediction of treatment success or failure.

Limitations

There were a few notable study limitations. The first and most obvious was a lack of a control group. Although unlikely, it may be that those who are not placed in halfway houses fail at the same rate and frequency as those who reenter directly to supervision. The current study provided substantial efforts to describe the patterns of failure, however without a control group, it is difficult to disentangle if the substantial portion of the noncriminal failures observed can be attributed to supervision effects of treatment, parole, and extended portions of intensive observation generally. However, given the ethical complications and the retrospective nature of the data collection, creating such a control group was not feasible.

Second, failure event times are not ideal. When one uses data available to a DOC, it is often subject to certain limitations—the primary one being the accuracy of the measure “date of recapture.” For escapes, technical violations, and new commitments, the lag time for recapture is variant and based on the speed of justice system processing for each participant. Although our operationalization of failure event time is consistent with the majority of prior reentry research, it still contains a certain amount of unaccounted variability. It is very unlikely that the significant differences between median survival times and general timing patterns found are due entirely to lag time of justice processing. Finally, a known (but only anecdotally discussed) issue is that of discretionary revocation resulting from new crimes. That is, community corrections officers have the ability to revoke an individual based on conditions violations but may choose not to, allowing participants additional chances to complete supervision and remain in the community. When evidence arises of a new crime, an officer may allow the state an end-around the prosecutorial procedures by instead revoking parole on previously identified technical violations. Although this practice is known, it is difficult to determine how often the procedure is actually used. This concept is worthy of further exploration.

Conclusion

Despite the listed limitations of our study, the development and recognition of differences across reentry failure types for reentering populations provides distinct policy implications. Our findings yield practical application and impact in the sheer number of people who could be successfully reentered rather than returned to incarceration. If our results are found to hold high external validity, it can be assumed that if halfway house placement, treatment, and supervision providers were to integrate our findings into the fulfillment of identified intervention principles (risk, need, and responsivity), then the general effectiveness of halfway house use would increase (Listwan, Cullen, & Latessa, 2006). From this it can be estimated that the average cost of imprisonment and the number of parolees under state supervision would both decrease while the number of available halfway house beds would increase, as more people successfully reentered upon receiving the proper risk assessment, supervision, and treatment (Hamilton, 2011b).

The second policy implication involves a call for administrators and policy makers to rethink the purpose and implementation of postrelease supervision. Considering the notion that certain parolees may be in need of specific treatment and at risk for different technical and halfway house violations, then the major hindrance of successful reentry is actually the manner in which the state assigns postrelease control by way of halfway houses, supplies certain treatment and available resources properly, and structures discretion in parole revocation (see Burke, 1997). Moreover, we believe that the overarching process of evaluating community corrections failure must be overhauled to encompass the actual failure taking place and provide subsequently necessary reforms to remedy the overuse of noncriminal revocation.

Finally, a renewed call for a guideline system of graduated sanctions must take place, with an emphasis on keeping nonviolent offenders in the community at all costs (Travis & Petersilia, 2001; Wright & Rosky, 2011; see also Makkai & Braithwaite, 1994). Washington State recently implemented a policy opting for short-term jail confinement sanctioning (30 to 120 days) for most violations, restricting revocations to sex offenders as well as serious and repeated violation of release conditions (4% of sanctioned confinement) (State of Washington DOC, 2012), and California has gone so far as to create formal sanctioning guidelines (Parole Violation Decision Making Instrument) for parolees to prevent revocations (Grattet et al., 2009). Either of these policies (or a combination of both) should go a long way toward decreasing back-door prison admissions and reduce the need and costs associated with incarceration.

Future research should explore the interactions of supervision intensity and reentry intervention participation. Although the methods of offender classification have continued to advance (Andrews et al., 2006), practitioners are provided little guidance in how to use this information for the provision and tailoring of supervision plans. As described by Thanner and Taxman (2003), little research has investigated this issue, known within corrections as the Principle of Responsivity. This will ultimately require studies to create and define offender types and link such findings to the effectiveness of intervention provision. Although some recent research has advanced the creation of typologies (Brennan & Brietenbach, 2009; Dembo et al., 2011), there is an absence of efforts attempting to disentangle how offender risk and needs interact with supervision and the provision of interventions and services. Research of this nature will go a long way to guide practitioners to the most efficient means of treatment matching and supervision planning.

Footnotes

Authors’ Note:

We would like to thank our colleagues, the reviewers, and the editor for their helpful and insightful input in the development of this article. Additional thanks go to James Wojtowicz, Ruth Steinruck, Renee Willitts, and Ralph Fretz for data access and assistance. Data for this article were gathered with the cooperation of New Jersey Department of Corrections–Office of Drug Programs. Points of view and conclusions expressed herein are those of the author and do not necessarily reflect the positions or policies of the New Jersey Department of Corrections. The research is in compliance with federal and university human subjects protections (Rutgers University, Office of Research and Sponsored Programs, Approval No. 07-010Mp).

Notes

Zachary K. Hamilton, PhD, is a professor of criminal justice and criminology at Washington State University. He completed his PhD from Rutgers University in criminal justice and is the author of Treatment Matching for Substance-Abusing Offenders. His research interests include offender quantitative methods, risk assessment, reentry, specialty courts, and substance abuse programming.

Christopher M. Campbell, MA, is a PhD candidate at Washington State University in the Department of Criminal justice & Criminology. He completed his Master of Arts degree from Ohio University in sociology and has worked as a criminal justice consultant for the Ohio Justice Policy Center and the Vera Institute of Justice. His research interests include offender risk assessments, the impact of technical violations, and juvenile corrections.

References

Allen

H. E.

Seiter

R. P.

(1976). The effectiveness of halfway houses: A reappraisal of a reappraisal. Chitty’s Law Journal, 24, 196-200.

Allison

(1984). Event history analysis: Regression for longitudinal event data. Thousand Oaks, CA: SAGE.

Andrews

D. A.

Bonta

Wormith

J. S.

(2006). Recent past and near future of risk and/or needs assessment. Crime & Delinquency, 49, 7-27.

Biderman

A. D.

Reiss

A. J.

(1967). On exploring the “dark figure” of crime. Annals of the American Academy, 374, 1-15.

Brennan

Breitenbach

(2009). The taxonomic challenge to general theories of delinquency: Linking taxonomic development to delinquency theory. In Sahin

Maier

(Eds.), Delinquency: Causes, reduction and prevention (pp. 1-38). Hauppauge, NY: Nova Science Publishers, Inc.

Bureau of Justice Statistics. (2011). Prisoner recidivism. Retrieved from http://bjs.ojp.usdoj.gov/index.cfm?ty=datool&surl=/recidivism/index.cfm

Burke

(1997). Policy-driven responses to probation and parole violations. Silver Spring, MD: National Institute of Justice, Center for Effective Public Policy.

Culp

R. F.

(2005). Frequency and characteristics of prison escapes in the United States: An analysis of national data. The Prison Journal, 85, 270-291.

Dembo

Briones-Robinson

Ungaro

Karas

Gulledge

Greenbaum

Belenko

(2011). Problem profiles of at-risk youth in two service programs: A multigroup exploratory latent class analysis. Criminal Justice and Behavior, 38, 988-1008.

10.

Donnelly

P. G.

Forschner

(1984). Client success or failure in a halfway house. Federal Probation, 38, 38-44.

11.

English

K. K.

Mande

M. J.

(1991). Empirical support for intervention strategies in community corrections. Journal of Contemporary Criminal Justice, 7, 95-106.

12.

Escarela

Francis

Soothill

(2000). Competing risks, persistence, and desistance in analyzing recidivism. Journal of Quantitative Criminology, 16, 385-414.

13.

Farrington

D. P.

(2008). Building developmental and life-course theories of offending. In Cullen

F. T.

Wright

J. P.

Blevins

K. R.

(Eds.), Taking stock: The status of criminological theory (Vol. 15, pp. 335-364). New Brunswick, NJ: Transaction Publishers.

14.

Gottfredson

S. D.

Moriarty

(2006). Statistical risk assessment: Old problems and new applications. Crime & Delinquency, 52, 178-200.

15.

Graham

Olchowski

Gilreath

(2007). How many imputations are really needed? Some practical clarifications of multiple imputation theory. Prevention Science, 8, 206-213.

16.

Grattet

Petersilia

Lin

(2008). Parole violators and revocations in California. Washington, DC: Bureau of Justice Statistics.

17.

Grattet

Petersilia

Lin

Beckman

(2009). Parole violators and revocations in California: Analysis and suggestions for action. Federal Probation, 73, 2-11.

18.

Gray

M. K.

Fields

Maxwell

S. R.

(2001). Examining probation violations: Who, what, and when. Crime & Delinquency, 47, 537-557.

19.

Hamilton

(2011a). Adapting to bad news: Findings from the Harlem parole reentry. Journal of Offender Rehabilitation, 50, 385-410.

20.

Hamilton

(2011b). Treatment matching for substance abusing offenders: An assessment of responsivity. El Paso, TX: LFB Publishing.

21.

Hartmann

D. J.

Friday

P. C.

Minor

K. J.

(1994). Residential probation: A seven-year follow-up study of halfway house discharges. Journal of Criminal Justice, 22, 503-515.

22.

Herrschaft

Hamilton

(2011). Recidivism among parolees in New York City, 2001-2008. New York: Center for Court Innovation.

23.

Hosmer

Lameshow

May

(2008). Applied survival analysis: Regression modeling of time-to-event data. Hoboken, NJ: John Wiley & Sons, Inc.

24.

Hser

Evans

Huang

Anglin

(2004). Relationship between drug and treatment services, retention, and outcomes. Psychic Services, 55, 767-774.

25.

Kruttschnitt

Uggen

Shelton

(2000). Predictors of desistance among sex offenders: The interaction of formal and informal social controls. Justice Quarterly, 17, 61-88.

26.

Langan

P. A.

Levin

D. J.

(2002). Recidivism of prisoners released in 1994. Washington, DC: U.S. Department of Justice.

27.

Latessa

E. J.

Allen

H. E.

(1982). Halfway houses and parole: A national assessment. Journal of Criminal Justice, 10, 153-163.

28.

Latessa

E. J.

Lovins

(2010). The role of offender risk assessment: A policy maker guide. Victims and Offenders, 5, 203-219.

29.

Latessa

E. J.

Travis

(1991). Halfway house or probation: A comparison of alternative dispositions. Journal of Crime & Justice, 14, 53-75.

30.

Listwan

S. J.

Cullen

F. T.

Latessa

E. J.

(2006). How to prevent prisoner re-entry programs from failing: Insights from evidence-based corrections. Federal Probation, 70, 19-25.

31.

Raghunathan

Rubin

(1991). Large sample significance levels from multiply imputed data using moment-based statistics and an F reference distribution. Journal of the American Statistical Association, 80, 1065-1073.

32.

Lowenkamp

C. T.

Latessa

E. J.

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

33.

MacKenzie

D. L.

(2000). Evidence based corrections: Identifying what works. Crime & Delinquency, 46, 457-471.

34.

MacKenzie

D. L.

Browning

Skroban

S. B.

Smith

D. A.

(1999). The impact of probation on the criminal activities of offenders. Journal of Research in Crime and Delinquency, 36, 423-453.

35.

Makkai

Braithwaite

(1994). The dialectics of corporate deterrence. Journal of Research in Crime & Delinquency, 31, 347-373.

36.

Maltz

M. D.

(2001). Recidivism (2nd ed.). Retrieved from http://www.uic.edu/depts/lib/forr/pdf/crimjust/recidivism.pdf

37.

Mears

Winterfield

Hunsaker

Moore

White

(2002). Drug treatment in the criminal justice system: The current state of knowledge. Washington, DC: Urban Institute Press.

38.

Mills

(2011). Introducing survival and event history analysis. Thousand Oaks, CA: SAGE.

39.

Moos

Pettit

Gruber

(1995). Characteristics and outcomes of three models of community residential care for substance abuse patients. Journal of Substance Abuse, 7, 99-116.

40.

Olson

D. E.

Lurigio

A. J.

(2000). Predicting probation outcomes: Factors associated with probation rearrest, revocations, and technical violations during supervision. Justice Research and Policy, 2, 73-86.

41.

Ostermann

(2011). How do former inmates perform in the community? A survival analysis of rearrests, reconvictions, and technical parole violations. Crime & Delinquency. Retrieved from http://cad.sagepub.com/content/early/2011/02/17/0011128710396425.abstract

42.

Petersilia

(1999). A decade of experimenting with intermediate sanctions: What have we learned? Washington, DC: National Institute of Justice.

43.

Petersilia

(2005). Hard time: Ex-offenders returning home after prison. Corrections Today, 67, 66-71.

44.

Petersilia

Turner

(1993). Intensive probation and parole. Crime and Justice, 17, 281-335.

45.

Piehl

A. M.

LoBuglio

S. F.

(2005). Does supervision matter? In Travis

Visher

C. A.

(Eds.), Prisoner reentry and crime in America (pp. 105-138). New York: Cambridge University Press.

46.

Rubin

(1987). Multiple imputation for nonresponse in surveys, New York: John Wiley.

47.

Sabol

W. J.

Adams

W. P.

Parthasarathy

Yuan

(2000). Offenders returning to federal prison, 1986-97 (No. NCJ 182991). Washington, DC: Bureau of Justice Statistics.

48.

Seiter

R. P.

Carlson

E. W.

Bowman

H. H.

Grandfield

J. J.

Beran

N. J.

(1977). Halfway houses. Washington, DC: U.S. Department of Justice.

49.

Solomon

A. L.

Kachnowski

Bhati

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

50.

Spivak

A. L.

Sharp

S. F.

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

51.

State of Washington Department of Corrections. (2012). Public safety: Community supervision (DOC 308.200). Olympia: Washington State Governmental Management Accountability & Performance and the Department of Corrections.

52.

Thanner

Taxman

(2003). Responsivity: The value of providing intensive services to high-risk. Journal of Substance Abuse Treatment, 24, 137-147.

53.

Travis

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

54.

Travis

Petersilia

(2001). Reentry reconsidered: A new look at an old question. Crime & Delinquency, 47, 291-313.

55.

Wodahl

E. J.

Ogle

Heck

(2011). Revocation trends: A threat to the legitimacy of community-based corrections. The Prison Journal, 91, 207-226.

56.

Wojtowicz

J. P.

Liu

(2006). A seven-year analysis of walk-away rates of the New Jersey halfway house program. Corrections Compendium, 31, 6-23.

57.

Wright

K. A.

Rosky

J. W.

(2011). Too early is too soon: Lessons from the Montana department of corrections early release program. Criminology & Public Policy, 10, 881-908.