The Treatment Group and Recidivism: A Multilevel Analysis of Prison-Based Substance Abuse Treatment

Abstract

This study contributes to the substance abuse treatment literature by examining recidivism across treatment groups, the level of aggregation between the individual and the program. The sample consisted of 618 drug-involved offenders who participated in 12 different treatment groups within a single prison. Reincarceration significantly varied across groups, controlling for time at risk and treatment modality. This between-unit variation was explained by individual level treatment response measures and control variables. Results demonstrated that treatment effectiveness is impacted by the group in which the individual participates. These findings have substantial implications for correctional treatment theory, research, and policy.

Keywords

correctional treatment recidivism substance abuse treatment group

Introduction

The research on correctional treatment, including prison-based substance abuse treatment, has matured over the past several decades. Scholars have been dedicated to discovering the relative effectiveness of treatment interventions, for which individuals or groups they are effective, and under what conditions. The breadth of factors, for instance, in the Texas Christian University treatment process model (Simpson, 2004) and the research supporting it demonstrates how far the field has come in understanding what works in substance abuse treatment. In this study, we sought to contribute to the correctional intervention literature by examining how the social context of prison-based treatment may influence postrelease recidivism among a sample of inmates who participated in a 12-month treatment program.

Research has generally demonstrated the effectiveness of prison-based therapeutic community (TC) drug treatment followed by community aftercare in reducing recidivism (e.g., Inciardi, Martin, & Butzin, 2004; Knight, Simpson, & Hiller, 1999; Prendergast, Hall, Wexler, Melnick, & Cao, 2004; Welsh, 2007, 2011). To better assess client-level factors on treatment progress, some studies have examined self-reports of psychosocial variables during treatment (Welsh, 2010), while others have examined treatment process variables such as duration (Joe, Knezek, Watson, & Simpson, 1991), therapeutic engagement (Hiller, Knight, Leukefeld, & Simpson, 2002), relationships with others (Broome, Knight, Knight, Hiller, & Simpson, 1997; Welsh, 2010; Welsh & McGrain, 2008), and counselor-reported measures (Stanley, Kelly, & Welsh, 2017). More recently, a randomized field experiment examined interactions between treatment modality and individual characteristics to account for variations in treatment outcomes, although this study found that prison-based TC participants had higher rates of reincarceration compared with their group counseling counterparts (Welsh, Zajac, & Bucklen, 2014).

Several studies have examined the group context of the treatment experience using multilevel statistical methods to simultaneously model and control for individual and contextual factors (e.g., Broome, Flynn, Knight, & Simpson, 2007), including the Dimensions of Change Instrument (DCI) within TCs (Mandell, Edelen, Wenzel, Dahl, & Ebener, 2008), client perspectives on treatment motivation and readiness across two treatment modalities (Melnick, Hawke, & De Leon, 2014), treatment and program level factors on posttreatment drug use (Ghose, 2008), and psychosocial and behavioral changes post treatment (Joe et al., 2012; Rowan-Szal, Joe, Simpson, Greener, & Vance, 2009). Others have assessed client-level asociality (e.g., lack of motivation) as a predictor of treatment engagement by using nested analysis of variance (ANOVA) models (i.e., clients nested within prison programs) that found highly asocial clients reported significantly lower engagement levels when compared with low and medium asocial clients (Pankow & Knight, 2012). Two of the studies controlling for organizational characteristics as group context (Ghose, 2008; Hser, Joshi, Anglin, & Fletcher, 1999) examined abstinence from drug use as the dependent variable, and two examined posttreatment recidivism (Kissin, Tang, Campbell, Claus, & Orwin, 2014; Rempel, Green, & Kralstein, 2012).

The recent attention to group dynamics and functioning within drug treatment is related to the central processes of De Leon’s (2000) TC model: “The quintessential element of the therapeutic community approach is community . . . [The] TC uses community as a method to help individuals change themselves” (emphases original; 2000, p. 85). Elsewhere, De Leon (1995) stated that “all activities are designed to produce therapeutic and educational change in the individual participants, and all participants are mediators of these therapeutic and educational changes” (p. 1611). In theory, some groups may function at high levels, whereas others may fall far short of it. Presumably, then, the functioning of the group will have a significant effect—positive or negative—on individual outcomes. Although the model recognizes the importance of the group, the contextual aspects of group-based treatment such as TC have received comparatively little attention, as existing research recognizes that all groups are not created equally.

Focusing on De Leon’s (1995, 2000) “community-as-method” approach, some researchers have begun to examine the internal processes of TCs, sometimes referred to as the “social climate” of the TC unit (Day, Casey, Vess, & Huisy, 2012; Schalast & Laan, 2017; Warren et al., 2013), “social network” (Doogan & Warren, 2017a, 2017b), or as “atmosphere” (Carr & Ball, 2014). Research focusing on the treatment climate has found that TCs are more conducive to rehabilitation than other settings, particularly those modified to a correctional setting. For example, heightened perceptions of the orderliness of the therapeutic environment have been found to increase treatment completion (Carr & Ball, 2014). There is also evidence that prisoners who are more satisfied with the social climate of their units are more likely to engage with treatment (Day et al., 2012).

Thus, an important aspect of climate within substance abuse programs is the treatment group (Kelly & Welsh, 2016). The treatment group is a collection of individuals in a shared space and time (e.g., a distinct housing unit within a prison) who have a common treatment experience. Kelly and Welsh examined group level differences on several measures of treatment climate between 12 distinct treatment groups participating in either a TC or group counseling program within a specialized drug treatment prison. Differences in climate were analyzed after 1 and 6 months of treatment. Results indicated that no differences in climate were detectable within the first month of treatment. At 6 months, however, treatment climate began to diverge across treatment groups. In fact, three treatment climate measures—program structure, counselor rapport, and counselor competence—revealed that some groups were functioning at significantly higher levels than others.

In this study, we adopt the concept of treatment group as operationalized by Kelly and Welsh (2016). As differences were found across groups on treatment climate variables, we explore the relationship between the treatment group and postrelease recidivism. Using nested data from a randomized sample of prison-based substance abuse treatment participants, we examined two sequential research questions. First, are there differences in reincarceration rates between distinct treatment groups, even when controlling for the type and intensity of treatment received? Second, can these differences be explained in statistical models that include individual level predictor and control variables?

Method

Setting and Participants

The State Correctional Institution (SCI) at Chester is a 1,200-bed medium security facility for men. It is a highly specialized treatment prison that houses only inmates who have been identified as having a serious substance abuse problem. Although the Pennsylvania Department of Corrections (PADOC) provides all the usual staffing for the prison (administration, security, etc.), the treatment services are contracted to a single well-established and accredited substance abuse treatment provider. SCI-Chester provided a unique setting for research, as it allowed for random assignment to one of two treatment modalities (a 12-month TC or a 12-month group counseling program) and the ability to control for institutional variables that have been shown to influence outcomes in between-prison studies of correctional treatment (Pelissier, Camp, & Motivans, 2003).

TC participants received roughly 30 hr per week of treatment, ultimately receiving over 1,500 hr of drug-treatment programming over the year. Treatment included one or more therapy groups led by professional addictions counselors as well as self-help (e.g., Alcoholics Anonymous/Narcotics Anonymous [AA/NA]) meetings run by the members of the unit themselves, and one-on-one counseling sessions with their unit counselor. Individual TC members are expected to take charge of their own recovery as well that of their peers (i.e., mutual self-help) through both formal and informal mechanisms (De Leon, 2000) such as inmate committees, morning meetings, and daily routine interactions. The other treatment modality within SCI-Chester was an outpatient group counseling program. Group counseling consisted of approximately 6 to 8 hr of treatment per week throughout the 12-month program. As opposed to the total immersion of a TC, group counseling affords the flexibility for the inmate to participate in other prison programming, such as educational and vocational classes and prison industries, and participants have more opportunities for downtime and recreation.

Over a 15-month period, all inmates admitted to SCI-Chester were asked to participate in a study with the cooperation of PADOC and the third-party treatment provider. A randomized experiment utilizing a response-dosage design was possible because no prior studies had directly compared or demonstrated the superiority of one approach over the other. TC participants (µ = 1,233 hr) received roughly 8 times the treatment dosage of group counseling participants (µ = 151 hr). TC was the experimental group and group counseling was the control group, with random assignment based on the first digit of the offender’s inmate number (even = TC, odd = group counseling). Following random assignment to treatment modality, the assignment to prison unit can be described as quasi-random. There was no systematic process associated with placing offenders in any of the treatment units. When a bed opened on a unit, it was filled by the next person awaiting a housing assignment within either treatment modality. All inmates were advised of the study procedures and asked to participate. Upon agreement, they gave their informed consent; 95% of the inmates approached agreed to participate in the study (Welsh, 2006a). The total sample size of the original study was 731 inmates.

Three additional criteria were used to select subjects for this study. First, only offenders whose treatment group assignments could be confirmed were included in the current analyses, resulting in a loss of 72 cases (9.8% of total). Second, to control for potential differences in postrelease supervision, the sample included only those inmates who were released to parole supervision. Those who maxed out on their sentences (n = 32, 4.4%) were removed from the sample. Finally, several inmates (n = 9, 1.2%) were transferred to another facility prior to completion of their treatment for reasons unrelated to their performance or behavior (e.g., to be in a facility closer to their homes and families). The final sample size for this study was thus 618.¹

The individuals in the sample represent the level-1 grouping in this multilevel study. The participants were nested within 12 distinct treatment groups (five TC and seven group counseling) at level-2 which was also the housing unit to which an inmate was assigned. This was the physical setting where they lived and participated together in their 12 months of treatment. In other words, each unit was a self-contained treatment program in that the inmates participated in the various treatment activities with the same group of inmates, similar to students within classrooms. As shown previously (Kelly & Welsh, 2016), treatment groups varied with respect to several treatment climate variables, justifying the need to control for differences in functioning across groups.

Measures and Analysis Plan

The outcome examined in this study was reincarceration (0 = no, 1 = yes) over a 3-year follow-up period, and the data were furnished by PADOC. Although other measures of recidivism such as rearrests and reconvictions may also be useful, reincarceration is generally viewed as the most consistent and reliable indicator of recidivism by researchers (De Leon, Melnick, Cao, & Wexler, 2006; Knight et al., 1999; Prendergast et al., 2004; Welsh, 2007).

To assess the participants’ responses to treatment, the Resident Evaluation of Self and Treatment (REST; Knight, Simpson, Chatham, & Camacho, 1997), was administered to all offenders in the final month of treatment. This survey contained 18 subscales measuring psychological and social functioning, motivation, and perceptions of the treatment process and program measured on a 1 (strongly disagree) to 7 (strongly agree) Likert-type scale. Simpson and Knight (1998) have demonstrated the instrument’s reliability and validity with numerous prison treatment samples (see also Garner, Knight, Flynn, Morey, & Simpson, 2007; Joe, Broome, Rowan-Szal, & Simpson, 2002).

A second-order principal components factor analysis with varimax rotation was performed using 17 of the 18 REST subscales (external pressures were excluded due to its low coefficient alpha). Reducing the REST subscales results in more parsimonious predictor variables and minimizes any multicollinearity among the 17 subscales. The first of the three factors that emerged from the REST scales was labeled treatment satisfaction (eigenvalue = 7.21; α = .92). As an indicator of the inmates’ overall satisfaction with the treatment unit, this factor included subscales such as program staff and structure, peer support, and counselor rapport. Next, the positive attitude factor (eigenvalue = 3.05; α =.85) measured the optimism the offender felt about his recovery and included subscales such as social conformity, personal progress, and treatment engagement. Finally, negative affect (eigenvalue = 1.17; α = .80) included the subscales anxiety, hostility, depression, and risk-taking. Only one subscale, therapeutic engagement from the positive attitude factor, loaded highly on more than one factor (.567 on treatment satisfaction as opposed to .601 on positive attitude), otherwise, the final solution was clean and easily interpretable. These three factors were the primary predictors of recidivism at the individual level in this study.

Control variables

Level 1

Treatment completion was used to control for those who did not receive the full 12 months of treatment (0 = failure, 1 = complete). Pretreatment risk was used to control for any pretreatment differences in client risk and motivation, measured by Offense Gravity Scores (OGS), Current and Prior (offense severity and criminal history, respectively) to control for the offender’s pretreatment criminal history (ranging from 0 to 10). Higher OGS current scores indicate a more severe controlling offense, while higher OGS prior scores indicate a lengthier and/or more severe criminal history. To control for the offender’s need for treatment at the time of admission to the program, scores on the Texas Christian University Drug Screen–II (TCUDS-II) were included. Time at risk was created to control for the amount of time (in months) that the offender was at risk for recidivating following their release from prison to the community. Demographic variables included age at admission and race.

Level 2

Treatment modality (TC vs. group counseling), on the contrary, as a property of the group, was entered as the sole variable at level-2 to control for the type of treatment received (0 = group counseling, 1 = TC). Although modality is but one of many possible group level properties that may affect an inmate’s probability of reincarceration, due to the relatively small number of level-2 units and the resultant low statistical power, the model could only allow one variable at this level.

The appropriate analytic method to explore the influence of the treatment group with nested data and a dichotomous outcome was a two-level binary hierarchical generalized linear modeling (HGLM) analysis, and Hierarchical Linear Modeling software (HLM 6.0.6) was used. The binary outcome is most useful to advance our understanding of group influences, as log odds ratios are more easily interpretable and useful for making policy recommendations about recidivism (for a survival analysis, see Welsh et al., 2014). First step in multilevel analysis was to examine variation in reincarceration between treatment groups at level-2. Ordinarily the unconditional ANOVA models would include no predictor variables, but because of the established positive relationship between time at risk and the probability of recidivating, the exposure variable was included in these models control for the variation in time at risk between treatment groups at level-2. In addition, because the primary focus of this study was the concept of the treatment group, the treatment modality variable (TC vs. group counseling) was entered at level-2. Should significant variation be detected between the units, controlling for both the time at risk and the type of treatment received, this initial analysis would support for the argument that the treatment group influences reincarceration.

A fixed effects analysis including all predictor and control variables was performed to establish which individual level variables were associated with reincarceration and to determine if any group level differences could be explained by individual level factors. In this model, the three factor scores, age of the offender, current and prior offense severity scores, the TCUDS-II, and time at risk variables were grand mean centered. Because race of the offender, treatment completion, and treatment modality (at level-2) were dichotomous variables, they were uncentered.

Results

Descriptive Statistics and HGLM ANOVA

In addition to the level-1 descriptive statistics of the outcome and independent variables presented in Table 1, it should be noted that all participants in this sample were male. Furthermore, approximately half of the sample (n = 301, 48.7%) was sentenced to their current term of incarceration for a drug offense, and 30.6% (n = 189) of the sample’s controlling offense was robbery, burglary, or theft. The rate of reincarceration for the overall sample was 37.1%. Disaggregated by treatment modality, 40.4% of TC participants were reincarcerated and 34.3% of group counseling participants were reincarcerated. In a purely descriptive sense, the TC inmates had somewhat higher rates of recidivism (see Welsh et al., 2014).

Table 1.

Level-1 Descriptive Statistics.

Outcome variables	N	Minimum	Maximum	M	SD
Reincarceration (0 = no, 1 = yes)	618	0	1	0.37	0.48
Demographic and control variables
Treatment modality (0 = GC, 1 = TC)	618	0	1	0.45	0.50
Age (years)	618	18	73	32.31	8.84
Race (0 = all others, 1 = Black)	618	0	1	0.62	0.49
OGS-current	618	2	10	6.75	1.27
OGS-prior	551	0	10	5.95	1.92
TCUDS-II	600	0	9	4.17	2.84
Time at risk reincarceration (month)	618	0.93	55.30	30.79	10.66
Treatment completion (0 = no, 1 = yes)	618	0	1	0.88	0.33
REST factors
Treatment satisfaction	618	−3.01	1.96	0	0.92
Positive attitude	618	−3.47	2.29	0	0.92
Negative affect	618	−2.63	3.42	0	0.92

Note. GC = group counseling; TC = therapeutic community; OGS = Offense Gravity Scores; TCUDS-II = Texas Christian University Drug Screen–II; REST = Resident Evaluation of Self and Treatment.

About 45% of the sample (n = 280) was distributed across the five TC units, and the remainder (n = 338) participated in one of the seven group counseling units. The mean number of study participants in each unit was approximately 52 (SD = 18.4), ranging from a low of 31 to a high of 101. As stated previously, assignment to the treatment unit was quasi-random in that inmates were placed onto their unit as beds opened up. Indeed, one-way ANOVA and chi-square tests supported this assertion, as there were no significant differences between the units on the pretreatment measures of age, offense severity, criminal history, TCUDS-II, and race of the offender. Thus, the results related to the treatment group presented below cannot be attributed to any pretreatment differences between the groups.

Figure 1 shows the rates of reincarceration (group means) across the 12 level-2 treatment groups. The rates of reincarceration varied across the groups, as the means ranged from a low of 20% (GC5) to a high of 48% (TC5) for reincarceration, and the results of the ANOVA model controlling for the time at risk and treatment modality are presented in Table 2. The level-1 control variable, time at risk, was significant and positively related to reincarceration (p < .001), controlling for treatment modality. Treatment modality at level-2 was a nonsignificant predictor of reincarceration, meaning that there was no difference in reincarceration between TC and group counseling participants. Although the implications of this nonsignificant finding will be discussed in the next section, the focus of this study was the influence of the treatment group on reincarceration and its moderating impact on individual level predictors.

Figure 1.

Reincarceration rates by treatment group.

Table 2.

Unconditional ANOVA Hierarchical Linear Model: Reincarceration.

Fixed effects	b	SE	t ratio	OR	p
Level-1
Intercept	−.760	.180	−4.297	0.468	<.01
Time at risk	.048	.009	5.487	1.05	<.001
Level-2
Treatment modality	.310	.268	1.157	1.36	ns
Random effects	Variance	SD	χ²	df	p
Treatment group	.1176	.343	23.54	10	<.01

Note. ANOVA = analysis of variance; OR = odds ratio.

The random effects portion of Table 2 addresses differences across treatment groups. Reincarceration significantly varied across the treatment groups at level-2 (p < .01), controlling for the time at risk in the community and the type of treatment delivered on the unit. The rates of reincarceration in Figure 1 that appeared to vary across the 12 treatment units are indeed statistically different, supporting the hypothesis that a participant’s odds of being reincarcerated depended on which unit he was assigned to.

Three-Factor Fixed Effects Model

The results of the full fixed effects model using the three REST factors (treatment satisfaction, positive attitude, and negative affect) are presented in Table 3. Controlling for treatment received at level-2 and all the other controls at level-1, negative affect significantly predicted reincarceration (p < .01) and was in the expected direction—the higher the individual scored on negative affect, the more likely he was to be reincarcerated. The control variables age, criminal history, time at risk, and treatment completion were also significant predictors of reincarceration and in the expected directions.

Table 3.

REST Three-Factor Fixed Effects Model Predicting Reincarceration.

Fixed effects	b	SE	t ratio	OR	p
Level-1
Intercept	−.331	.319	−1.036	0.718	ns
Treatment satisfaction	.137	.109	1.262	1.147	ns
Positive attitude	−.064	.104	−0.615	0.938	ns
Negative affect	.331	.111	2.974	1.392	<.01
Time at risk	.052	.010	5.294	1.053	<.001
Race	.146	.218	0.667	1.157	ns
Age	−.054	.013	−4.338	0.947	<.001
OGS-current	.059	.078	0.756	1.061	ns
OGS-prior	.162	.055	2.925	1.176	<.01
TCUDS-II	.023	.037	0.607	1.023	ns
Treatment completion	−.491	.277	−1.774	0.612	<.10
Level-2
Treatment modality	.208	.232	0.897	1.231	ns
Random effects	Variance	SD	χ²	df	p
Treatment group	.0418	.2044	13.089	10	ns

Note. REST = Resident Evaluation of Self and Treatment; OR = odds ratio; OGS = Offense Gravity Scores; TCUDS-II = Texas Christian University Drug Screen–II.

Of particular note in this analysis is the random effects portion of Table 3 that indicates that after entering all level-1 variables, reincarceration no longer significantly varied across treatment groups at level-2. This answers our second research question regarding compositional differences across treatment groups on the predictor and control variables that account for the observed variation in the initial ANOVA model. This, too, is support for the argument that the treatment group is an important construct to be considered in the empirical literature, as failing to control for it may produce misleading results.

Discussion

The central theme in this study was that the group context of substance abuse treatment, also known as the treatment group (Kelly & Welsh, 2016) that has been relatively unexplored in the treatment literature, matters when examining treatment outcomes. This study was not an evaluation of the treatment programs at SCI-Chester per se; instead, several related analyses were conducted to examine the influence of the treatment group within the prison. The first analysis explored differences in reincarceration between treatment groups and controlling for both time at risk in the community and treatment modality, reincarceration varied significantly across treatment groups. No previous research has demonstrated that otherwise equivalent treatment groups produce significantly different reincarceration rates. By controlling for the type of treatment received in the ANOVA model and with the successful random assignment to treatment modality and quasi-random assignment to the unit, this analysis showed that merely participating in one treatment unit or another, independent of treatment modality (TC vs. group counseling), affected the individual’s odds of reincarceration.

Next, in the fixed effects models including the full complement of predictor and control variables, the between-unit variation on reincarceration was no longer significant. This is explained by differences across the treatment groups on these within-group constituent variables. These two findings justify the rationale for this study and make a strong argument for future research examining the influence of group level dynamics on substance abuse treatment outcomes. Again, the small number of level-2 units in this study precluded entering further predictors at level-2, and that is a challenge that future multilevel studies of prison treatment will need to address (Welsh, 2006b).

These findings are underscored by the standardization of many contextual factors across the 12 treatment groups. Because the treatment was delivered by the same organization in the same prison setting, many of the factors that previous multilevel research used to explain variation in treatment outcomes were naturally controlled for in this study. For instance, factors such as different sources of accreditation, staffing and management structures, funding sources, and qualifications for employment (see, for example, Lehman, Greener, & Simpson, 2002) that have been examined as predictors of participant success were the same across all the treatment groups in this study. This is not to say that organizational variations are unimportant, but rather, at the organizational level, these factors were held constant in this study.

Likewise, previous research on prison-based treatment has controlled for variations across correctional institutions such as security level, geographic region, and institutional adherence to disciplinary standards (Pelissier et al., 2003, 2001). The present sample, therefore, had the added benefit of being drawn from a single prison under a single administration and security regime, overseen by a single state level agency with uniform policies regarding correctional practices, which also controlled for other institutional factors that have been shown to influence outcomes in between-prison studies.

Implications for Theory, Research, and Practice

In evaluating the long-term recidivism for a sample of offenders who participated in prison-based substance abuse treatment, this study supports De Leon’s (1995, 2000) TC theory as well. The findings that reincarceration varied across the 12 treatment groups support De Leon’s (2000) arguments that the TC, or treatment group as it has been referred to here, affects individual behavior during treatment and afterward, and that the community’s influence over individual behavior varies from group to group. Although previous research has compared program level differences, this is one of the first studies of its kind to operationalize and evaluate the treatment context at the treatment group level. Theoretical models should further investigate the potential effects of the treatment group on long-term outcomes such as relapse and recidivism. To date, research has not adequately fleshed out what the social environment of treatment constitutes or how it should be measured (Broome et al., 2007; Broome et al., 1999). This study attempts a step forward in the measurement and understanding of the treatment group, though much more detailed work is needed to describe and measure group processes (see the Therapeutic Community Prison Inmate Networks Study [Justice Center for Research, n.d.] for recent advances).

The next logical research questions in exploring group processes of substance abuse treatment, then, are as follows: What are the most relevant group level dynamics or dimensions in substance abuse treatment, and what effect do they have on later outcomes? To understand group processes within the treatment experience, researchers might aggregate individual level measures of treatment responsiveness and psychosocial functioning and enter them into full multilevel models where sufficient statistical power exists with larger numbers of level-2 groups.

Measures that could be aggregated to the group level include those that directly measure the participant’s responses to treatment. Whether it is the REST, as used in this study and others (Hiller, Belenko, Welsh, Zajac, & Peters, 2011; Welsh, 2010), or other instruments that measure the individual’s experiences in treatment, the critical factor required to enable the study of group dynamics is that the compositional measures come from the treatment participants themselves. These are the measures best suited to quantitatively describe treatment group functioning and to use as level-2 predictors of individual behavior.

To perform this type of research, where aggregated measures can be useful indicators of group functioning and processes, a large number of level-2 units are likely to be needed to achieve adequate statistical power. This is unlikely to be accomplished within a single prison, because most institutions simply do not have a sufficient number of treatment units to support analyses with adequate statistical power at the group level (Welsh, 2006b). Thus, future research will need to sample from multiple prisons and/or community-based correctional facilities. In turn, prisons, community treatment providers, and perhaps states can offer additional levels of aggregation that can be accounted for and examined in multilevel studies. The very substantial challenge in such studies, however, is to be able to measure and control for the many organizational and jurisdictional differences that inevitably emerge when one travels outside the walls of a single prison.

In terms of service delivery, perhaps the most important recommendation that follows from these results is that counselors, supervisors, and administrators need to be mindful of how the treatment group is functioning. As Simpson (2004) argued, understanding the functioning of the program is essential to improving it. The findings of this study support the importance of the treatment group in addition to the individual and organizational factors that have previously been demonstrated in the literature to individual outcomes.

Hostility, part of the negative affect factor, within a treatment group can manifest itself in various ways. Elevated animosity in interactions between clients may lead to verbal or even physical altercations, or at a minimum, impaired communication with one another and with the group’s counselors. Passive-aggressive actions may include negative body language and avoidance. In all group activities, there will undoubtedly be isolated incidents of this nature, but what must be encouraged is an awareness of when these isolated incidents begin to occur with regularity or ultimately come to define the group as a whole.

Conversely, whereas hostility and negative affect ought to be minimized, trust among the group (part of the treatment satisfaction factor) needs to be assessed and maximized. Possible indicators of trust could include levels of personal disclosure of sensitive information that are received without judgment, a large number of supportive friendships that develop between the clients, or clients being receptive to the leadership roles that other members of the group take on. Again, these and other indicators of trust likely happen every day in every treatment group, but ongoing, proactive strategies need to be developed to assess and promote these dynamics at the group level.

Limitations and Conclusion

Due to the small number of level-2 units (n = 12), there was not adequate statistical power to enter level-2 predictors other than the treatment modality variable. This is often the case with multilevel studies, as sampling a sufficient number of level-2 units, especially when researching correctional populations and programming, is difficult and costly (Welsh, 2006b). The present data, however, allowed for statements about how measures varied across the level-2 units and how the group context of the treatment unit moderated individual level factors and their influence on recidivism.

Another limitation is the lack of data on aftercare completion. In previous studies, in-prison TC combined with community aftercare produced the greatest reduction in recidivism (Hiller, Knight, & Simpson, 1999; Inciardi et al., 2004; Prendergast et al., 2004). Although it was not feasible to track aftercare completion for this sample (DOC did not collect such data), it is important to emphasize that 6 months of outpatient-level aftercare was a mandatory (not optional) component of all participants’ treatment and a condition of their parole and release to the community. Furthermore, general research on community aftercare programs for released offenders is limited, and there is no standardized conceptualization of what constitutes aftercare (McKay, 2001). Overall, neither the core intervention components nor the core implementation components associated with the aftercare phase of the continuum of care are well understood (Pelissier, Jones, & Cadigan, 2007).

Although organizational factors are standardized in this study and the assignment to treatment unit within the institution was random, observed differences across treatment units could still be the result of other, unmeasured factors. These could include compositional staff or other unmeasured dynamics that might vary across units. For instance, Welsh (2006a) and McGrain (2006) reported that a number of inmates received a new counselor in the middle of their treatment episode who was perceived as less effective than the previous counselor. Within a TC, counselors may be secondary to the peer group; however, experienced, well-trained professionals are still important to the functioning of the unit (De Leon, 2000) and could have significant influences on observed differences between units.

The foregoing study was an exploration of the importance of the treatment group in correctional theory, research, and practice. An empirical test of this idea used multilevel data from an experimental study of prison-based substance abuse treatment. The results from 12 treatment units within a single prison supported the conclusion that, in addition to individual factors, the treatment group can influence future outcomes such as reincarceration. Researchers can and should pay greater attention to the treatment group as a unit of analysis, not just individuals or programs, as treatment group effects may be important moderators of individual outcomes.

Footnotes

Acknowledgements

The authors wish to thank Ralph Miller, Matthew Hiller, and Jamie Fader for their invaluable assistance on other drafts.

Declaration of Conflicting Interests

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

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The research reported here was supported by Grant 2002-RTBX-1002 from the U.S. Department of Justice, National Institute of Justice (NIJ) and the Graduate School at Temple University. Opinions expressed here are those of the authors and not necessarily of the U.S. Department of Justice. Any errors or omissions are the responsibility of the authors alone.

Notes

Author Biographies

Christopher E. Kelly, PhD, is an associate professor in the Department of Sociology and Criminal Justice, Saint Joseph’s University, where he studies interrogation, jury decision-making, and correctional treatment. His work has been published in Law and Human Behavior, Justice Quarterly, and Annual Review of Law and Social Science.

Wayne N. Welsh, PhD, is professor of criminal justice at Temple University. His research focuses on violence, organizational change, and substance abuse treatment in criminal justice settings. Recent articles have appeared in the Journal of Experimental Criminology, Criminal Justice and Behavior, and Drug and Alcohol Dependence.

Jennifer N. Stanley, MS, is an instructor for the Department of Law and Justice Studies, Rowan University. Her current research focuses on correctional substance abuse treatment. In addition, she has also contributed to studies exploring formal and informal processing of juveniles in the justice system and a field experiment regarding interview rooms with the Philadelphia Police Department.

References

Broome

K. M.

Flynn

P. M.

Knight

D. K.

Simpson

D. D.

(2007). Program structure, staff perceptions, and client engagement in treatment. Journal of Substance Abuse Treatment, 33, 149-158.

Broome

K. M.

Knight

D. K.

Knight

Hiller

M. L.

Simpson

D. D.

(1997). Peer, family, and motivational influences on drug treatment process and recidivism for probationers. Journal of Clinical Psychology, 53, 387-397.

Broome

K. M.

Simpson

D. D.

Joe

G. W.

(1999). Patient and program attributes related to treatment process indicators in DATOS. Drug and Alcohol Dependence, 57, 127-135.

Carr

W. A.

Ball

S. A.

(2014). Predictors and treatment outcomes of perceived ward atmosphere among therapeutic community residents. Journal of Substance Abuse Treatment, 46, 567-573.

Day

Casey

Vess

Huisy

(2012). Assessing the therapeutic climate of prisons. Criminal Justice and Behavior, 39, 156-168.

De Leon

. (1995). Therapeutic communities for addictions: A theoretical framework. International Journal of the Addictions, 30, 1603-1645.

De Leon

. (2000). The therapeutic community: Theory, model, and method. New York, NY: Springer.

De Leon

Melnick

Cao

Wexler

H. K

. (2006). Recovery-oriented perceptions as predictors of reincarceration. Journal of Substance Abuse Treatment, 31, 87-94.

Doogan

N. J.

Warren

(2017a). A network of helping: Generalized reciprocity and cooperative behavior in response to peer and staff affirmations and corrections among therapeutic community residents. Addiction Research and Theory, 25, 243-250.

10.

Doogan

N. J.

Warren

(2017b). Dynamics of peer and staff corrections among therapeutic community residents. Substance Use and Misuse, 52, 1429-1438.

11.

Garner

B. R.

Knight

Flynn

P. M.

Morey

J. T.

Simpson

D. D.

(2007). Measuring offender attributes and engagement in treatment using the client evaluation of self and treatment. Criminal Justice and Behavior, 34, 1113-1130.

12.

Ghose

(2008). Organizational- and individual-level correlates of posttreatment substance use: A multilevel analysis. Journal of Substance Abuse Treatment, 34, 249-262.

13.

Hiller

M. L.

Belenko

Welsh

W. N.

Zajac

Peters

R. H.

(2011). Screening and assessment: An evidence-based process for the management and care of adult drug-involved offenders. In Leukefeld

C. G.

Gullotta

T. P.

Gregrich

(Eds.), Handbook of evidence-based substance abuse treatment in criminal justice settings (pp. 45-62). New York, NY: Springer.

14.

Hiller

M. L.

Knight

Leukefeld

Simpson

D. D.

(2002). Motivation as predictor of therapeutic engagement in mandated residential substance abuse treatment. Criminal Justice & Behavior, 29, 56-75.

15.

Hiller

M. L.

Knight

Simpson

D. D.

(1999). Risk factors that predict dropout from corrections-based treatment for drug abuse. The Prison Journal, 79, 411-430.

16.

Hser

Y. I.

Joshi

Anglin

M. D.

Fletcher

(1999). Predicting posttreatment cocaine abstinence for first-time admissions and treatment repeaters. American Journal of Public Health, 89, 666-671.

17.

Inciardi

J. A.

Martin

S. S.

Butzin

C. A.

(2004). Five-year outcomes of therapeutic community treatment of drug-involved offenders after release from prison. Crime and Delinquency, 50, 88-107.

18.

Joe

G. W.

Broome

K. M.

Rowan-Szal

G. A.

Simpson

D. D.

(2002). Measuring patient attributes and engagement in treatment. Journal of Substance Abuse Treatment, 22, 183-196.

19.

Joe

G. W.

Knezek

L. D.

Watson

D. D.

Simpson

D. D.

(1991). Depression and decision-making among intravenous drug users. Psychological Reports, 68, 339-347.

20.

Joe

G. W.

Knight

Simpson

D. D.

Flynn

P. M.

Morey

J. T.

Bartholomew

N. G.

O’Connell

D. J.

(2012). An evaluation of six brief interventions that target drug-related problems in correctional populations. Journal of Offender Rehabilitation, 51(1-2), 9-33.

21.

Justice Center for Research. (n.d.). Therapeutic community prison inmate networks study (TC-PINS). Retrieved from http://justicecenter.psu.edu/research/projects/therapeutic-community-prison-inmate-networks-study-tc-pins

22.

Kelly

C. E.

Welsh

W. N.

(2016). Examining treatment climate across prison-based substance abuse treatment groups. Substance Use and Misuse, 51, 902-911.

23.

Kissin

W. B.

Tang

Campbell

K. M.

Claus

R. E.

Orwin

R. G.

(2014). Gender-sensitive substance abuse treatment and arrest outcomes for women. Journal of Substance Abuse Treatment, 46, 332-339.

24.

Knight

Simpson

D. D.

Chatham

Camacho

(1997). An assessment of prison-based drug treatment: Texas’ in-prison therapeutic community program. Journal of Offender Rehabilitation, 24, 75-100.

25.

Knight

Simpson

D. D.

Hiller

M. L.

(1999). Three-year reincarceration outcomes for in-prison therapeutic community treatment in Texas. The Prison Journal, 79, 337-351.

26.

Lehman

W. E. K.

Greener

J. M.

Simpson

D. D.

(2002). Assessing organizational readiness for change. Journal of Substance Abuse Treatment, 22, 197-209.

27.

Mandell

Edelen

M. O.

Wenzel

S. L.

Dahl

Ebener

(2008). Do dimensions of therapeutic community treatment predict retention and outcomes? Journal of Substance Abuse Treatment, 35, 223-231.

28.

McGrain

P. N.

(2006). An examination of therapeutic engagement in a prison-based drug treatment therapeutic community (Unpublished doctoral dissertation). Temple University, Philadelphia, PA.

29.

McKay

J. R.

(2001). Effectiveness of continuing care interventions for substance abusers: Implications for the study of long-term treatment effects. Evaluation Review, 25, 211-232.

30.

Melnick

Hawke

De Leon

(2014). Motivation and readiness for drug treatment: Differences by modality and special populations. Journal of Addictive Diseases, 33, 134-147.

31.

Pankow

Knight

(2012). Asociality and engagement in adult offenders in substance abuse treatment. Behavioral Sciences and the Law, 30, 371-383.

32.

Pelissier

Camp

Motivans

(2003). Staying in treatment: How much difference is there from prison to prison? Psychology of Addictive Behaviors, 17, 134-141.

33.

Pelissier

Jones

Cadigan

(2007). Drug treatment aftercare in the criminal justice system: A systematic review. Journal of Substance Abuse Treatment, 32, 311-320.

34.

Pelissier

Wallace

O’Neil

J. A.

Gaes

G. G.

Camp

Rhodes

Saylor

(2001). Federal prison residential drug treatment reduces substance use and arrests after release. The American Journal of Drug and Alcohol Abuse, 21, 315-337.

35.

Prendergast

M. L.

Hall

E. A.

Wexler

H. K.

Melnick

Cao

(2004). Amity prison-based therapeutic community: 5-year outcomes. The Prison Journal, 84, 36-60.

36.

Rempel

Green

Kralstein

(2012). The impact of adult drug courts on crime and incarceration: Findings from a multi-site quasi-experimental design. Journal of Experimental Criminology, 8, 165-192.

37.

Rowan-Szal

G. A.

Joe

G. W.

Simpson

D. D.

Greener

J. M.

Vance

(2009). During-treatment outcomes among female methamphetamine-using offenders in prison-based treatments. Journal of Offender Rehabilitation, 48, 388-401.

38.

Schalast

Laan

J. M.

(2017). Measuring social climate in German prisons using the Essen Climate Evaluation Schema. The Prison Journal, 97, 166-180.

39.

Simpson

D. D.

(2004). A conceptual framework for drug treatment process and outcomes. Journal of Substance Abuse Treatment, 27, 99-121.

40.

Simpson

D. D.

Knight

(1998). TCU data collection forms for correctional residential treatment. Fort Worth: Institute of Behavioral Research, Texas Christian University.

41.

Stanley

J. N.

Kelly

C. E.

Welsh

W. N.

(2017). Counselor knows best? Clinician assessments of in-prison substance abuse treatment participants. Criminal Justice and Behavior, 44, 1462-1476.

42.

Warren

K. L.

Doogan

De Leon

Phillips

G. S.

Moody

Hodge

(2013). Short-run prosocial behavior in response to receiving corrections and affirmations in three therapeutic communities. Journal of Offender Rehabilitation, 52, 270-286.

43.

Welsh

W. N.

(2006a). Evaluation of drug treatment programs at the State Correctional Institution (SCI) at Chester. Washington, DC: National Institute of Justice.

44.

Welsh

W. N.

(2006b). The need for a comprehensive approach to program planning, development, and evaluation. Criminology and Public Policy, 5, 603-614.

45.

Welsh

W. N.

(2007). A multisite evaluation of prison-based therapeutic community drug treatment. Criminal Justice & Behavior, 34, 1481-1498.

46.

Welsh

W. N.

(2010). Inmate responses to prison-based drug treatment: A repeated measures analysis. Drug and Alcohol Dependence, 109, 37-44.

47.

Welsh

W. N.

(2011). Prison-based substance-abuse programs. In Gideon

Sung

(Eds.), Rethinking corrections: Rehabilitation, reintegration and reentry (pp. 157-192). Thousand Oaks, CA: Sage.

48.

Welsh

W. N.

McGrain

P. N.

(2008). Predictors of therapeutic engagement in prison-based drug treatment. Drug and Alcohol Dependence, 96, 271-280.

49.

Welsh

W. N.

Zajac

Bucklen

K. B.

(2014). For whom does prison-based drug treatment work? Results from a randomized experiment. Journal of Experimental Criminology, 10, 151-177.