Even a Little Bit Helps

Abstract

Cognitive-behavioral therapy (CBT) is one of the most promising and widely used therapeutic approaches to reducing recidivism among criminal populations. Although many studies have evaluated CBT for this express purpose, few have done so in a community correctional environment. This article reports findings from a randomized field trial evaluating, “Choosing to Think, Thinking to Choose,” a CBT program designed specifically for a community correctional setting, and its impact on the recidivism of high-risk offenders. High-risk probationers were assigned to either standard, intensive probation (n = 447) or to the treatment condition (n = 457), where they received the same supervision intensity while also being directed to a classroom-based, 14-week CBT program. Twelve months after random assignment, intention-to-treat (ITT) analyses indicate that the overall CBT group was significantly less likely to reoffend, although this effect is concentrated in measures of nonviolent offending.

Keywords

probation parole community corrections CBT cognitive-behavioral therapy high-risk offenders

Introduction

Community supervision is one of the most frequently relied-upon sanctions in the criminal justice system (Glaze & Kaeble, 2014). This dependence places enormous pressure on probation and parole agencies, especially as staff and budgets in many jurisdictions have not kept pace with changes in the population (Gifford, 2002). Community correctional agencies have responded to this challenge by adjusting the nature and intensity of supervision to address the specific characteristics and needs of the individual being supervised.

Reflecting a desire to use evidence-based programs (EBPs), these agencies have also sought out interventions with a demonstrated history of effectiveness (Skeem & Manchak, 2008; Taxman, 2002). Many follow the principles of risk-needs-responsivity, intensifying supervision for certain groups of high-risk offenders while decreasing it for individuals who do not pose as much danger (Hyatt & Barnes, 2014; MacKenzie, 2000; Petersilia, 1998; Petersilia & Turner, 1990, 1993). Agencies have also responded with therapeutic programming that specifically targets the unique criminogenic needs of different offender populations. Although there are no definitive lists of EBPs, almost every review of evidence-based practices (e.g., Aos, Miller, & Drake, 2006; Guevara & Solomon, 2009) recommends the expanded application of cognitive-behavioral therapy (CBT).

While the delivery of CBT is time-intensive, it offers agencies the opportunity to apply a noninvasive form of treatment in a manner that can provide both cost-effectiveness and reduced offending. Evaluations of various CBT methods have been largely positive, but these studies have been primarily limited to fully incarcerated populations and to a narrow range of commonly applied programs. True randomized experiments are also rare in community supervision environments, and few studies have focused attention on the subset of probationers who present the highest risk of serious violent offending. This study directly addresses these deficits within the literature.

CBT

CBT is one of the dominant, and perhaps most widely studied, psychological approaches to behavior modification (Baker, McFall, & Shoham, 2008; J. Beck & Beck, 2011; Lipsey, Chapman, & Landenberger, 2001; Milkman & Wanberg, 2007). This approach has been successfully used to manage depression, anxiety, and a variety of other psychological disorders (A. T. Beck, 1964; Butler, Chapman, Forman, & Beck, 2006; DeRubeis & Beck, 1988; Dobson & Khatri, 2000). More recently, the same techniques have been applied to the treatment of anger, violence, and criminal activity (A. T. Beck, 1999; R. Beck & Fernandez, 1998). CBT uniquely addresses undesirable behaviors—including criminal activities—by focusing exclusively on reforming maladapted and antisocial thought patterns.

CBT brings about change by modifying the way that participants respond to the automatic thoughts and emotional reactions encouraged by stressful, external stimuli. Participants are taught to observe, manage, and reform these cognitive relationships (Matthews, 1997; Milkman & Wanberg, 2007). CBT can also assist individuals in learning new coping skills and cognitive models to supplant the distortions in their thinking. In turn, these new approaches to emotional and cognitive management may change subsequent behaviors, as well as improve social skill and community integration (Kazemian, 2007).

Criminogenic thought patterns and conduct, from a cognitive-behavioral perspective, are not unique from those that often underlie depression and other psychological disorders (A. T. Beck, 1999). Distortions common among an offender population include self-justification, perceptions of dominance and victimization, misinterpretation of social cues, and failures in moral reasoning; all of these may cause offenders to respond inappropriately to stimuli (A. T. Beck, 1999; Clark, Beck, & Alford, 1999; Wanberg & Milkman, 2006). The clinical response remains largely unchanged since CBT was first introduced, with activities focused on identifying problematic schemata, automatic emotional responses, and cognitive reactions while providing prosocial coping and management skills (A. T. Beck, 1999; R. Beck & Fernandez, 1998).

CBT programs that target criminal activity, however, are also not simple analogs to traditional CBT programming, and may have several unique features. When dealing with an offender population, CBT treatment can integrate training focused on both interpersonal and social skills, two distinct skill-sets thought to influence the propensity to commit crime. This approach includes reinforcing the attitudes necessary to encourage responsible conduct, develop empathy, and gauge consequences (Little, 2000). The environmental and emotional stimuli known to trigger pro-delinquent schemata can be addressed with an offender population, while also targeting issues of anger management not commonly found in depressive cases. The ability to combine these (often court-mandated) elements into a program with a history of prompting behavioral change is one of the primary reasons that CBT remains at the vanguard of EBP in corrections.

Applied CBT in Corrections

CBT is not a standardized intervention, but rather a diverse family of treatment programs united by a common hypothesis about the relationship between cognition and action. As applied to recidivism reduction, the principles of CBT have been operationalized through a wide variety of environments (Lösel, 1995), and have coalesced into a number of different “brand name” treatment programs. Examples of such packages include Reasoning and Rehabilitation (Ross, Fabiano, & Ewles, 1988), Moral Reconation Therapy (Little & Robinson, 1993), Aggression Replacement Training (Goldstein & Glick, 1994), and Thinking for a Change (Bush, Glick, Taymans, & Guevara, 2011).

Given CBT’s broad appeal and popularity, it is unsurprising that the literature is rife with evaluations.¹ Although variation between programs makes it difficult to draw conclusions about the overall effectiveness of CBT, a meta-analytic approach can help synthesize these findings, given their strong theoretical and practical commonalities (Lipsey & Wilson, 1993). An early meta-analysis included 14 studies, including eight randomized experiments, which specifically evaluated the impact of cognitive-behavioral programs designed to reduce recidivism. Overall, the analysis reported a weighted mean odds ratio of .66 (α = .05; Lipsey et al., 2001) favoring the use of CBT to lower the prevalence of offending. An update to this study, published several years later, identified a larger sample of 14 randomized trials (Lipsey & Landenberger, 2005; Lipsey, Landenberger, & Wilson, 2007), finding a mean reduction in recidivism of approximately 27%. These findings were not anomalous, as other meta-analytic studies identified similar positive effects for programming that followed the CBT model (Pearson, Lipton, Cleland, & Yee, 2002; Redondo, Sanchez-Meca, & Garrido, 1999; Wilson, Bouffard, & Mackenzie, 2005).

Although a review of these meta-evaluations hints at the efficaciousness of CBT, the impact of individual programs varies considerably, depending on characteristics such as program size and external funding. These divergent effects have often been attributed to the difficulties in program attrition, instructor training, and challenges in treating larger groups of offenders over extended periods of time (Spruance, Van Voorhis, Listwan, Pealer, & Seabrook, n.d.). In addition to these program elements, the composition and structure of the evaluations themselves was strongly predictive of effect size. For example, demonstration projects (e.g., those led by researchers) returned a 49% reduction in recidivism, while practitioner-led programs produced a lower mean reduction of approximately 11% (Lipsey & Landenberger, 2005).

This pattern suggests that some mechanism common to varying CBT programs has the potential to meaningfully reduce recidivism rates, but that the need for targeted interventions, more rigorous evaluations, and replication is clear. Many varying programs incorporate ideas from CBT into treatment protocols designed to prevent offending. There have been few evaluations, however, that use a population, context, and experimental design similar to the research we present here. Notably, the present study provides a (a) large-scale, (b) randomized experimental design to evaluate the impact of a (c) practitioner-led, (d) CBT-based intervention that is designed to be (e) delivered in a community correctional setting for (f) high-risk offenders (g) identified with a powerful, actuarial forecasting tool. We now turn to how each of these elements was addressed within the research design.

Method

Setting

This study was conducted in Philadelphia, Pennsylvania. In this jurisdiction, the city operates its own local system of probation and parole that is separate from the state system. In fiscal year (FY) 2013, the city’s Adult Probation and Parole Department (APPD; 2014) supervised a total of 55,946 unique offenders who were serving 80,314 different terms of supervision, with an average daily caseload of 39,485 offenders under the supervision of 276 case-carrying officers.

Since 2009, Philadelphia’s APPD has used a series of random forest forecasting models to stratify its offenders into three different risk groups. Although the full details of these models are discussed elsewhere (Barnes & Hyatt, 2012; Berk, Sherman, Barnes, Kurtz, & Ahlman, 2009), they use machine learning techniques (Berk, 2012) to predict both the prevalence and form of criminal behavior during the first 2 years after a new term of supervision begins. Offenders designated as high risk are those forecasted to commit a new serious offense—defined as murder, attempted murder, aggravated assault, robbery, or a sexual crime—during this period of time. Forecasts are performed whenever a new probation² case is initiated, regardless of whether the offender is already under supervision by APPD.

The introduction of random forest forecasting in Philadelphia coincided with a series of randomized controlled trials (RCTs) to test APPD’s stratified response to the different levels of predicted risk. These experiments began by focusing on offenders with a lower risk³ of offending, finding that increased caseloads and reduced supervision intensity had no effect on offending (Barnes et al., 2010; Barnes, Hyatt, Ahlman, & Kent, 2012). The agency then turned to the question of what to do with high-risk offenders.

APPD created distinct high-risk units to provide strongly increased supervision intensity for these offenders, largely mirroring the Intensive Supervision Probation (ISP) programs tested by other researchers (Gill, 2014; Petersilia & Turner, 1993). In FY 2013, this high-risk division supervised just 10.3% of the agency’s daily caseload while employing 26.6% of its officers. High-risk units carried an aggregate caseload of 45.9 offenders per officer, compared to 122.5 in moderate and 206.0 in the lowest risk⁴ division. Replicating the relatively consistent findings in the literature (Petersilia & Turner, 1990, 1993; but see Jalbert, Rhodes, Flygare, & Kane, 2010; Paparozzi & Gendreau, 2005), an experimental evaluation of ISP for APPD’s high-risk offenders showed little or no impact on offending. However, rates of absconding, returns to custody and technical violations all increased when offenders were simply supervised more intensively (Hyatt & Barnes, 2014).

The present research stems from a further randomized trial with APPD’s high-risk population. This experiment examines two randomly assigned groups of offenders, both of whom were supervised under this same ISP protocol. One of these groups, however, was additionally targeted for a classroom-based program of CBT.

Participants

The experiment enrolled offenders over a single full-year period, with random assignment taking place when the offenders were beginning a new instance of supervision with APPD. All offenders enrolled in the trial, regardless of their engagement with the treatment program, are included in the current intention-to-treat (ITT) analyses. It is important to note that the goal of this experiment was limited to evaluating the policy of offering CBT to this population of high-risk probationers, as opposed to a direct test of CBT’s effects.

The same computer application that APPD used to produce the random forest risk forecasts was also used to assess the offenders’ eligibility for the study. During the year that new cases were brought into the RCT, this system produced predictions for a total of 27,196 new case starts, spread across 19,998 unique offenders. Each of these case starts was screened using at least 10 eligibility tests, the results of which are shown in Table 1.

Table 1:

Tests of Eligibility for Random Assignment

			Percent of cases eligible
Eligibility test	Number eligible	Number ineligible	Of all forecasts	Of high-risk forecasts
Forecasted as high risk	5,492	21,704	20.2%
Male	21,903	5,293	80.5%	90.2%
City resident	5,941	21,255	21.8%	98.1%
At least 9 months of supervision remaining	20,970	6,226	77.1%	83.2%
Not already randomly assigned into RCT	26,732	464	98.3%	91.6%
Not previously graduated from CBT^a	10,436	9	99.9%	99.6%
Not currently on high-risk supervision	25,858	1,338	95.1%	76.7%
No previous high-risk forecasts in last year	25,490	1,706	93.7%	69.1%
Not currently on specialized supervision	23,589	3,607	86.7%	88.1%
No judicial order for specialized supervision	20,705	6,491	76.1%	78.1%
Not eligible for YVRP	25,997	1,199	95.6%	78.2%
All tests combined	1,289	25,907	4.7%	23.5%

Note. RCT = randomized controlled trial; CBT = cognitive-behavioral therapy; YVRP = Youth Violence Reduction Partnership.

This eligibility test was introduced 7 months after the start of random assignment.

The most basic eligibility criterion for this experiment required that the offenders receive a high-risk result from the agency’s forecasting system. As shown in Table 1, just 20% of forecasted case starts successfully met this requirement. The eligibility rules also excluded offenders who were already assigned to one of the high-risk units, or who had been scored as high-risk within the previous year.

Offenders placed into the agency’s specialized supervision units, such as those devoted to drug treatment, mental health, sex offenders, and domestic violence, were also barred from enrollment. This restriction applied to those who were already assigned to specialized supervision, as well as those who were judicially ordered into it at as a result of their presenting case. The agency also requested that the participants in the city’s Youth Violence Reduction Partnership (YVRP)—a multiagency, grant-funded program that included intensive supervision and mentoring for young offenders living in specific neighborhoods—be kept out of the research sample (Jucovy & McClanahan, 2008).

Most of these individual eligibility requirements were easily met by those in APPD’s high-risk population. The primary barrier for eligibility was therefore not any individual tests that were applied to incoming case starts, but was instead the need to meet all of these criteria simultaneously. Of the 5,492 case starts which were forecasted as high risk, only 1,289 (23.5%) screened as fully eligible for the RCT. Of these, 457 were randomly assigned to the experimental CBT group, while 447 were placed into the control group.⁵ This sample size was sufficient to provide the desired degree of statistical power, even under the assumption that the effect of the randomly assigned treatment might be comparatively weak.⁶ With a small effect size of d = 0.20, the experiment’s statistical power was calculated at 85.3% (Cohen, 1988).

One advantage of integrating the random assignment process into the agency’s risk forecasting procedure was the wealth of data collected to generate these predictions. The forecasting model used 48 different predictor variables (Barnes & Hyatt, 2012), each of which can be used to compare the nature of the two treatment groups. A selection of the more informative values from these forecasts is presented in Table 2. The study’s enrolled participants averaged less than 30 years of age and were overwhelmingly African American. Based on their self-reported residential zip codes at the time of random assignment, they also tended to come from neighborhoods of relative poverty, with a mean household income at the 29th percentile of all Philadelphia zip codes.

Table 2:

Treatment Group Demographics and Equivalence Prior to Random Assignment

Variable	CBT	Control	p ^a
Age at random assignment	30.27	29.38	.167
Offender race
African American	0.72	0.73	.718
Census household income	26,919.83	26,733.99	.709
Census housing value	50,090.31	49,787.25	.837
Presenting charges
Violence	1.68	1.45	.146
Firearm	0.38	0.22	.228
Prior adult charges
Any crime	60.84	58.04	.380
Violence	21.18	19.23	.124
Firearm	4.95	4.26	.180
Drugs	6.17	5.81	.428
Onset age
Juvenile any crime^b	14.99	14.83	.394
Juvenile violence^b	14.96	14.98	.886
Prior terms of APPD supervision	2.36	2.45	.628
Prior county prison
Prevalence	0.98	0.94	.008*
Admissions	4.82	4.80	.911
Days	539.61	518.13	.563
Prior sentences to incarceration	3.72	3.39	.322

Note. CBT = cognitive-behavioral therapy; APPD = Adult Probation and Parole Department.

Two-tailed independent sample t tests, assuming unequal variances. ^bIncludes only those offenders with prior history of these juvenile offenses.

p < .05.

As high-risk probationers, the participants tended to have substantial prior histories of violent offending. Just under half (45.3%) of them were beginning their new term of APPD supervision due to a violent offense, and nearly all (92.7%) had at least one prior violent crime in their adult criminal history. Most of the offenders (65.4%) had been under the agency’s supervision before, and prior stays in the county’s prison⁷ system were practically universal (96.1%). The typical participant had previously spent 17.3 months incarcerated in the county prison system alone.

Table 2 also demonstrates that random assignment succeeded in producing two treatment groups that were statistically equivalent at the moment of random assignment. Out of 79 different comparisons on measures collected at the moment of random assignment, only one—the prevalence of one or more prior stays in the Philadelphia prison system—produced a statistically significant difference. This single significant difference is well below the number of such findings that would be expected due to chance alone, and a divergence of such a small degree (0.978 vs. 0.944) likely has little practical importance. It is therefore not controlled for in any of the analyses that follow.

Experimental Treatment

The Philadelphia CBT curriculum was titled, “Choosing to Think, Thinking to Choose,” and consisted of 14 sequential classroom sessions. Classes were 2 hours in length and were held on a weekly basis. The curriculum was designed by experienced psychologists who were well versed in CBT, but the classes themselves were delivered by probation officers.⁸ Although these probation officers had no prior experience with this form of therapy, they were provided with approximately 100 hours of training (Hyatt, 2013) from a psychologist who helped design the treatment. The classroom setting and the use of probation officer instructors ensured that the agency could continue to deliver this program once the researcher-led experiment had ended.

Although numerous CBT programs exist for criminal offenders, the Philadelphia curriculum was uniquely customized for APPD’s high-risk population. The 14 weekly sessions covered a variety of topics, including anger management, dealing with stressful situations, successful management of criminal justice and community correctional interactions, and management of interpersonal and professional relationships. These themes were illustrated through topically relevant scenarios and examples, and were presented in a manner designed to be accessible to the young, urban, and male offenders who make up most of the agency’s high-risk caseload.

The classes operated continuously, with a new block of instruction beginning every 7 or 8 weeks. Each of the instruction blocks was divided into three sections, each containing a maximum of 15 enrolled students (i.e., 45 students per block). Although students had some flexibility concerning the day of the week, the time of day was rigidly fixed, with classes available only during the late morning or early afternoon. The coursework was also cumulative in nature, meaning that each week of instruction assumed that the students had attended all of the previous classes. Offenders were permitted to miss a single class in the 14-week sequence, but a second absence almost invariably led to their removal from that block of instruction.

The experimental offenders were told that their participation in this curriculum—referred to within the department as the “Life Skills Class”—was a mandatory part of their supervision. Like all Philadelphia probationers (including the control group), they signed a document which acknowledged this requirement, which was further reinforced by a letter from the judge who presided over the vast majority of probation violation cases in Philadelphia.

There were certain conditions, however, that could excuse these offenders from attending individual blocks of instruction. Given the demonstrated relationship between employment and recidivism (Bushway, 2011; Lageson & Uggen, 2013; Sampson & Laub, 2003), offenders were not required to enroll in the class if attending it would put them at risk of losing their job or dropping out of school. Other reasons for excused absences included offenders who were deemed unsuitable for the CBT classroom (usually due to language or mental health issues), offenders whose in-office reporting requirements were less frequent than the weekly class schedule, and those who were verifiably scheduled to begin some other form of treatment in the near future. Whenever these conditions existed, the offenders remained eligible for placement into future classes, and their suitability was reevaluated at the start of each subsequent block of instruction.

The mechanics of placing CBT-assigned offenders into the treatment classroom were complex. Three or 4 weeks prior to the start of each instruction block, a combination of database searches and personal meetings with probationers were used to determine their availability and eligibility for the forthcoming class. Those who were successfully placed into the next CBT class then had to wait a week or more until the block began. Effectively, this process meant that graduates of the CBT class were required to remain on active supervision for 16 to 18 consecutive weeks, with no periods of incarceration longer than a few days, and without absconding from supervision. As will be shown below, this requirement often proved challenging for this particular population of probationers.

To make the administration of the CBT treatment more manageable, a separate high-risk supervision unit was created for these offenders. Only probationers who were assigned to this unit were eligible to attend CBT. The control group, meanwhile, was supervised by three other high-risk units. All four participating units also supervised a large number of offenders who were not enrolled in the research. One result of this design was that those supervising the participants had no means to distinguish which members of their caseloads were in the experiment and which ones were not.

Although some variations between officers and units certainly existed, the same supervision protocol applied equally to all officers and probationers in these units, regardless of their participation in the randomized trial. The only salient difference in the treatment of the two randomly assigned groups should therefore be the application of the classroom-based CBT program.

Assignment Integrity

To receive their randomly assigned treatment, the participating offenders in both groups first needed to be assigned for supervision in the appropriate high-risk units. The seemingly simple task of keeping the offenders under active high-risk supervision in the correct units proved to be somewhat challenging. Figure 1 presents a day-by-day record of the supervision placement histories of all 904 randomly assigned offenders during their first year after random assignment. The combined proportion of these offenders who were being supervised in their randomly assigned units peaked at just 72.5%, almost exactly 1 month after random assignment. During this first follow-up year, the combined sample experienced 330,864 offender-days of potential supervision, with just 194,763 (58.9%) of them spent assigned to an experimentally appropriate supervision unit.

Figure 1:

Daily Unit Supervision Assignment of Experimental and Control Cases for First Year After Random Assignment

Although these values may seem somewhat low, they also mask a large amount of movement over time. In an aggregate sense, the agency was exceptionally compliant with random assignment, in that more than 95% of the participating offenders spent at least some portion of this year under the supervision of a unit which matched their randomly assigned treatment. The real challenge—which was often outside of APPD’s control—was to keep them there. Numerous forces combined to pull offenders out of their experimentally appropriate placements, including judicial orders into specialized supervision units (5.4% of the offender-days), absconding (7.0%) and especially local incarceration. On any given day, 23.3% of the sample was in the local prison system. This incarcerated portion of the sample, however, was also constantly in flux, and a large majority (66.8%) of the participants found themselves in custody at some stage during this 1-year period. This pattern suggests that even those offenders who largely stayed in their “as assigned” units often experienced short breaks in the continuity of their supervision.

One important point in these supervision patterns is that they appear very similar in both the experimental and control groups. Table 3 compares the two treatment groups on the mean number of days spent in different supervision categories, and finds no significant differences between them. Offenders in the CBT group spent, on average, 220 days under supervision by the unit that managed enrollment in the experimental therapy classes. Although this figure suggests that most offenders had more than enough time to complete the CBT treatment program (which required just 112-126 days), it does not account for the frequent interruptions which many participants experienced in supervision. A substantial majority (69.9%) experienced at least one break in the continuity of supervision before the year was out.

Table 3:

Days of Supervision Provided During the First Year After Random Assignment

Variable	CBT	Control	p
Days supervised as assigned	220.31	210.47	.252
Days of treatment crossover	0.30	0.44	.552
Days of specialized supervision	21.50	18.12	.437
Days of supervision in other units	1.85	3.04	.172
Days absconded	23.67	27.73	.337
Days in local prison system	84.03	86.78	.690
Days no longer under supervision	14.34	19.42	.200

Note. CBT = cognitive-behavioral therapy.

p < .05.

These interruptions are important, because the CBT treatment required that an offender remain on continuous supervision in the appropriate unit for four sequential months. Less than two thirds (63.7%) of the CBT treatment group experienced a consecutive period of availability that would have allowed them to complete the full 14-week program. This proportion was not significantly different in the control group (61.3%), suggesting that the same mechanisms were systemic across the agency’s high-risk offenders. The frequent incarcerations, increasing use of specialized supervision, and growing prevalence of absconding combined to make the experimental CBT program—along with any other form of long-term treatment—very difficult to deliver, and these difficulties grew more acute as time went on.

Delivery of CBT

Given the difficulty of maintaining high-risk offenders in continuous and consistent supervision conditions, it is not surprising that successfully delivering a multimonth sequence of CBT classes proved troublesome. Of the 457 offenders who were randomly assigned into the CBT treatment group, a small majority of the experimental offenders (54.7%) participated in at least some component of the classroom program within this 1-year period, with 35% completing the entire sequence. Table 4 shows the percentage of the CBT-assigned offenders who successfully reached each phase of this treatment, and reveals some reasons to hope that the Philadelphia program was delivered as well as possible with this population of high-risk probationers.

Table 4:

Delivery of CBT Treatment to Randomly Assigned Offenders During the First Year After Random Assignment, Across Different Stages of the Treatment Protocol

Amount of CBT treatment completed	Count	Percentage of all assigned	Percentage of attendees
Randomly assigned	457	100.0
Assessed for eligibility	328	71.8
Scheduled to attend	291	63.7
Attended Week 1	250	54.7	100.0
Attended Week 2	241	52.7	96.4
Attended Week 3	233	51.0	93.2
Attended Week 4	222	48.6	88.8
Attended Week 5	214	46.8	85.6
Attended Week 6	196	42.9	78.4
Attended Week 7	190	41.6	76.0
Attended Week 8	185	40.5	74.0
Attended Week 9	180	39.4	72.0
Attended Week 10	175	38.3	70.0
Attended Week 11	173	37.9	69.2
Attended Week 12	169	37.0	67.6
Attended Week 13	161	35.2	64.4
Graduated	160	35.0	64.0

Note. CBT = cognitive-behavioral therapy.

As described above, each new block of instruction required the prescreening of all available offenders to determine their eligibility for the forthcoming class. Over their first year after random assignment, each of the CBT-assigned offenders experienced the start of six or seven instruction blocks, and each of these blocks presented up to three different opportunities for an offender to be assessed in this manner. As Table 4 shows, however, only 72% of these offenders completed even this first assessment stage of the CBT treatment process. The most common reason for these absences, accounting for 45% of these missed opportunities, was incarceration in the local prison system. Other reasons included the transfer to a different (often specialized) supervision unit (19% of assessment opportunities), the natural expiration of all active supervision cases (13%), and absconding (9%). By a very large margin, these unreachable offenders accounted for the largest single source of attrition in the CBT treatment program.

Within the subset of CBT-assigned offenders who were successfully assessed, a relatively small number (8.1% of the full treatment group) were repeatedly deemed ineligible for the class. The most common reasons for ineligibility were verified scheduling conflicts with either work or education. By the end of their first year, 63.7% of the CBT treatment group was successfully scheduled to attend at least one block of classroom instruction. Merely being scheduled for a class session, however, was no guarantee of attendance. As noted earlier, there was often a several-week gap between the eligibility assessment and the first week of CBT instruction, which gave the offenders numerous opportunities to become unavailable for therapy. Another 9.0% of the CBT-assigned group experienced attrition at this stage, and never arrived for their first class.

Once offenders successfully made it to this first class, attrition began to moderate somewhat. Of the 250 offenders who participated in at least one session of CBT training, 160 (64%) managed to complete all 14 weeks and graduate from the program within 1 year of their random assignment. Across all offenders assigned to the experimental treatment group, however, these figures amount to a graduation rate of 35%. This level of attrition during treatment is roughly in line with those seen in similar evaluations in the literature (e.g., McGuire et al., 2008; Robinson, 1995; Van Voorhis et al., 2004). For example, in one evaluation of 735 offenders court-ordered to behavioral treatment, 29.2% completed the full program, 24.6% participated in some treatment, and 46.1% of the sample received no treatment at all (McGuire et al., 2008).

Moreover, there is every reason to believe, based on the analyses presented above, that the two groups contained very similar sorts of offenders, and that the control offenders would have experienced the same attrition profile if they had been targeted with a similarly burdensome treatment regime. The intention to treat analyses presented here, which include every randomly assigned offender regardless of their engagement with the CBT treatment, should therefore provide valid estimates despite these levels of treatment delivery. In a way, our attrition rate also presents something of an opportunity. If this form of treatment—which was so imperfectly delivered—can be shown to have any positive effects on the experimental offenders, then perhaps even greater impact can be produced by a CBT treatment program that is better designed for delivery to such a troublesome population.

Results

Violations and Sanctions

Both the CBT and control offenders were supervised in fairly rigorous manner, with an average of 32.4 officer contacts during the first year of follow-up. In some ways, the two groups appear to have exhibited very similar reactions to this intensive supervision. Table 5 shows that offenders in both groups were equally likely to return a positive⁹ drug screening result, had the same prevalence and frequency of incarceration, and presented the same probability of absconding. Moreover, as shown earlier in Table 3, they also spent equivalent numbers of days both in local prisons and in an absconded status.

Table 5:

Violations and Sanctions During the First Year After Random Assignment

Variable	CBT	Control	p
Proportion positive on any drug tests	0.60	0.55	.153
Number of positive drug tests	2.80	1.91	.000*
Proportion with local incarceration	0.68	0.69	.785
Number of stays in incarceration	0.98	0.97	.845
Proportion absconded	0.26	0.26	.903
Proportion with violation hearings	0.52	0.43	.004*
Number of violation hearings	1.14	0.82	.000*

Note. CBT = cognitive-behavioral therapy.

p < .05.

One notable difference between the two groups, however, is in the use of probation and parole violations. Those assigned to the CBT group were significantly more likely to experience a violation hearing during their first year after random assignment than the offenders in the control group. CBT-assigned offenders also faced a significantly higher number of violation hearings during this time. In theory, these additional violations could have been directly related to the CBT treatment, because a refusal to attend class was, in itself, considered a violation of probation. Every offender who was violated for this reason, however, also had additional violations at the same time, suggesting that these hearings would have been pursued even if the offenders had complied with their CBT treatment.

Reoffending

Recidivism data were gathered by counting all new criminal charges lodged against the offenders in the Philadelphia court system for offenses that were committed during their first year after random assignment. These charges do not include any probation or parole violations, and therefore cover only new criminal conduct which took place after the offenders were randomly assigned.

The proportions of offenders who were charged with one or more new offenses during this time frame (i.e., the prevalence of reoffending) are presented in Table 6. Assignment to the CBT group significantly reduced the percentage of offenders who reoffended during the 1-year follow-up period. Across all offense types, 33.7% of the CBT-assigned offenders were charged with a new offense, compared with 40.5% of the control group (p = 0.035). When the recidivism data are split into more specific categories of offending, however, the significance of these differences largely disappears, most likely due to much smaller base rates of offending within these categories. Nevertheless, the between-group difference in the prevalence of nonviolent offending comes tantalizingly close to the traditional boundaries of statistical significance (p = 0.065).

Table 6:

Prevalence of Reoffending During the First Year After Random Assignment

Variable	CBT	Control	p
Any kind of offenses	0.337	0.405	.035*
Serious offenses	0.118	0.103	.465
Violent offenses	0.144	0.141	.881
Firearm offenses	0.055	0.045	.491
Weapons offenses	0.061	0.054	.625
Nonviolent offenses	0.326	0.385	.065
Property offenses	0.155	0.174	.439
Drug offenses	0.136	0.170	.152

Note. CBT = cognitive-behavioral therapy.

p < .05.

Although a smaller proportion of the CBT offenders engaged in new criminal activities, the overall count of new offenses (i.e., the frequency of reoffending) was unaffected by the randomly assigned treatment. These comparisons can be seen in Table 7, and reveal no significant differences between the two groups. This absence of an effect held true for both general offending and for all of the more specific offense categories considered in the analysis. Nevertheless, the direction of these differences generally suggests a lower amount of reoffending in the CBT group, although this was not true for either firearm or weapon offenses.

Table 7:

Frequency of Reoffending During the First Year After Random Assignment

Variable	CBT	Control	p
Any kind of offenses	3.72	4.18	.535
Serious offenses	0.64	0.73	.670
Violent offenses	1.43	1.51	.842
Firearm offenses	0.46	0.37	.585
Weapons offenses	0.48	0.41	.653
Nonviolent offenses	2.29	2.67	.348
Property offenses	0.79	1.00	.304
Drug offenses	0.47	0.64	.138

Note. CBT = cognitive-behavioral therapy.

p < .05.

Although there was no reduction in the number of new offenses, the timing of the observed improvements in the prevalence of general and nonviolent recidivism provides some intriguing reasons for optimism. Figure 2 displays the results of a survival analysis and plots the speed at which the offenders committed their first new offense (of any kind) after random assignment. For the first 2 months of the follow-up period, the CBT and control groups initiated reoffending at almost exactly the same rate. Beginning roughly 2½ months after random assignment, however, the prevalence of reoffending began to diverge between the two randomly assigned groups, with the CBT group suddenly becoming much less likely to commit new offenses. This gap then widened throughout the rest of the follow-up year, until finally reaching the proportions shown in Table 5.

Figure 2:

Prevalence of Reoffending Based on the Amount of Time Since Random Assignment

Across the entire period, this survival analysis reveals a significant difference between the two groups. The mean survival estimates (CBT = 294.9 days, control = 279.0 days) are significantly different (Kaplan–Meier log rank, χ² = 4.59, df = 1, p = 0.032) from one another, and we can be confident that the two treatment groups followed different survival curves. Similar effects, which approach statistical significance, are found when the analysis is limited to nonviolent offenses (χ² = 3.58, df = 1, p = 0.058). Given the prevalence differences presented earlier, this result is far from surprising.

What is more intriguing, however, is that the two curves did not begin to diverge until more than 2 months had passed after random assignment. As described above, the classroom-based CBT treatment used in Philadelphia operated on a fixed schedule, with new blocks of instructions beginning only every 7 to 8 weeks. For those offenders who successfully made it into treatment, the median amount of time between random assignment and the first available CBT class was 69 days. This same time lag is also reflected in the survival analysis. Not only did the two groups exhibit significantly different likelihoods of reoffending over the entirety of the 1-year follow-up period, but this difference did not begin to manifest itself until the moment when CBT treatment finally became available for the majority of the experimental group. This timing lends support to the idea that this form of CBT, however imperfectly delivered, is largely responsible for the observed difference in reoffending.

Discussion

The results of this experiment indicate that a policy of enrolling high-risk probationers in the, “Choosing to Think, Thinking to Choose,” CBT program significantly reduces the prevalence of overall and (with near-significance) nonviolent offending, and lengthens the times to failure for those types of offenses. Although higher levels of treatment delivery would have been ideal, the trial was able to clear many of the hurdles that have limited other community corrections studies to nonexperimental designs (Hollin et al., 2008; McGuire et al., 2008). It is important to remember that our findings portray the combined impact of the full package of supervision policies deployed in Philadelphia. The impact of these treatment delivery rates cannot be separated from that of the CBT program itself, particularly in an ITT analysis that is inherently conservative in estimating treatment effects.

The evaluation presented here therefore tests the overall policy of providing these offenders with the opportunity to receive this form of CBT and should not be considered a direct test of the treatment regimen itself. This is not without value, as it presents a faithful, real-world picture of what practitioners could expect to observe upon replication of this particular curriculum at their agencies (Hollis & Campbell, 1999). Moreover, the results generated during this trial also are divorced from any reliance on the external funding, academic researchers, and trained psychologists that have, in earlier reviews, obscured the magnitude and nature of the effects in other interventions (Lipsey et al., 2001; Welsh & Farrington, 2000).

A comparison of the limited effects observed here with the synthesized results of the meta-analytic literature—which has primarily examined the use of CBT during incarceration—also has bearing on these findings (e.g., Lipsey et al., 2007). Community corrections and incarceration are two very different environs in which to deliver treatment. In prison, it is generally easy to locate your participants and engineer their attendance in treatment; they are (quite literally) a captive audience. The results reported here are largely in line with previous studies of CBT for high-risk offenders under community supervision (e.g., Robinson, 1995; Van Voorhis et al., 2004), and the experimental methodology used contributes to the robustness of these findings.

The random forest methodology used to identify the targeted high-risk population may also inform these results. This approach to actuarial forecasting was strictly limited to predicting the future dangerousness of the offenders (Barnes & Hyatt, 2012; Berk, 2008; Berk et al., 2009), and their criminogenic needs were neither assessed, nor considered. Assessing risk alone, however, while leaving needs unexamined, is a potential conflict with the evidence-based criteria for correctional programming (Thanner & Taxman, 2003). These requirements specify that offenders receive the particular services, in an appropriate format, that meet their bespoke psychological and criminogenic needs (Andrews, Bonta, & Wormith, 2011; Taxman, Thanner, & Weisburd, 2006). The Philadelphia framework also only partially addressed the important dimension of risk-needs-responsivity, which suggests that programming will be the most effective when access is limited to those who are best suited to it (Taxman & Marlowe, 2006).

The maturity of the intervention itself should also be considered. The CBT protocol being evaluated here was relatively new and was developed specifically for use by APPD during this study. With the exception of a nonexperimental pilot study, these are the first reported outcomes of this curriculum. As might be expected, these findings warrant further refinement of this approach. Program length, frequency of instruction, the number of classroom hours, instructor training and background, and the quality of implementation have all been shown to relate to the magnitude of effects in CBT treatment (Lipsey et al., 2007). The treatment used for this study may not have been sufficient to affect the most severe forms of offending and overcome the long-standing cognitive distortions that influence violent recidivism among high-risk offenders.

The base rate of offending for violent crimes, even in such a high-risk population of offenders, may also be too low to detect a difference over a relatively brief 1-year follow-up period. If so, then stronger results may be revealed once the offenders have had a longer time in which to reoffend. Although a 1-year follow-up is common in program evaluation, it remains to be seen if additional analyses, with longer follow-up and different analytical structures, will return more nuanced results; the survival analyses presented here seem to suggest that this may be the case.

The key question is what lessons Philadelphia’s CBT program offers in making this kind of treatment more accessible to the offenders who could benefit from it the most. An ideal form of treatment would, for example, be immediately available for delivery at the moment when offenders first begin a new term of supervision. A revised treatment regime could also deliver CBT in an incremental but nonrigid manner, making treatment available for delivery whenever an offender is available to receive it. The central idea, in an ideal program, would be to seize every moment that presents itself for delivery of CBT and to develop an intervention that could be implemented during these brief windows of opportunity.

Conclusion

CBT offers a significant promise for community corrections, but it is not without challenges or limitations. As contemporary trends in sentencing begin to move away from incarceration, it seems likely that the use of community supervision will continue to expand over the next decade. Moreover, risk forecasting will become more widespread and more accurate. Diversion from incarceration will inescapably result in high-risk offenders being placed into community correction caseloads, and forecasting techniques will provide agencies with advanced warning of the risk these individuals pose to public safety.

With these predictions in hand, agencies will need to respond. It is therefore crucial that we develop methods of treatment that are tested and validated in this challenging environment, and that are available for use in conjunction with these forecasts. Intensive supervision alone will not be enough to reduce the offending of high-risk offenders. Instead, supervision will need to be combined with proven evidence-based practices that work with this specific population.

Our results suggest that CBT can be an effective option under these conditions but that additional work needs to be done to marry the requirements of evidence-based supervision with the pragmatic challenges of delivering treatment to a high-risk population. CBT will not be a panacea for the reduction of offending, but these results—considered in light of the broader evidence-based literature—offer additional empirical support and highlight a path forward for community correctional research.

Footnotes

Acknowledgements

The authors express their deepest thanks to Richard Berk, Ellen Kurtz, Robert Malvestuto, Charles Hoyt, and Laurie Robinson for their support of this research. The authors would also like to acknowledge the many other members of the Philadelphia Adult Probation and Parole Department, including the cognitive-behavioral therapy officers and leaders, Jillian Eidson, Carrinnia Woodson, Chiquita Brice, Keith Nigro, Steve Austin, Vince Fiorentino, and Kathres Davis, without whom this work could never have taken place.

This research was funded by Grant 2008-IJ-CX-0024 from the National Institute of Justice and by a grant from the Smith Richardson Foundation. The authors also thank Jerry Lee and his support for the Jerry Lee Center of Criminology at the University of Pennsylvania from 2001 to 2015, and the Jerry Lee Centre of Experimental Criminology at the University of Cambridge.

Notes

Geoffrey C. Barnes, PhD, is a Lecturer in Evidence-Based Policing and Fellow of the Jerry Lee Centre of Experimental Criminology at the University of Cambridge. Prior to this, he served as a Research Assistant Professor in the Department of Criminology at the University of Pennsylvania. His work specializes in the conduct of randomized controlled trials within the criminal justice system, and on the forecasting of criminal behavior through machine learning techniques. His research has included collaboration with numerous criminal justice agencies in Pennsylvania, Maryland, the United Kingdom, and Australia.

Jordan M. Hyatt, JD, PhD, is an Assistant Professor in the Department of Criminology and Justice Studies at Drexel University in Philadelphia, Pennsylvania. His research focuses on the evaluation of correctional programs, with an emphasis on randomized designs, and the development of evidence-based policies in policing, corrections, reentry, and community supervision.

Lawrence W. Sherman, PhD, KNO, FRSA, is Director of the Institute of Criminology and Wolfson Professor of Criminology at the University of Cambridge, and Distinguished University Professor at the University of Maryland. His research interests are in the fields of crime prevention, evidence-based policy, restorative justice, police practices, and experimental criminology. He has conducted multiple field experiments in the areas of homicide, gun violence, domestic violence, robbery, burglary, and other crime problems, in collaboration with such agencies as the Metropolitan, Northumbria and Thames Valley Police, London’s Crown Courts, HM Prisons, the Crown Prosecution Service, the Youth Justice Board of England and Wales, and the National Probation Service.

References

Adult Probation and Parole Department. (2014). Population report. Philadelphia: First Judicial District of Pennsylvania.

Andrews

D. A.

Bonta

Wormith

J. S.

(2011). The Risk-Need-Responsivity (RNR) Model: Does adding the Good Lives Model contribute to effective crime prevention? Criminal Justice and Behavior, 38(7), 735-755.

Aos

Miller

Drake

(2006). Evidence-based public policy options to reduce future prison construction, criminal justice costs, and crime rates. Federal Sentencing Review, 19, 275-290.

Baker

T. B.

McFall

R. M.

Shoham

(2008). Current status and future prospects of clinical psychology: Toward a scientifically principled approach to mental and behavioral health care. Psychological Science in the Public Interest, 9, 67-103.

Barnes

G. C.

Ahlman

Gill

Sherman

L. W.

Kurtz

Malvestuto

(2010). Low-intensity community supervision for low-risk offenders: A randomized, controlled trial. Journal of Experimental Criminology, 6, 159-189.

Barnes

G. C.

Hyatt

J. M.

(2012). Classifying adult probationers by forecasting future offending. Washington, DC: National Institute of Justice.

Barnes

G. C.

Hyatt

J. M.

Ahlman

L. C.

Kent

D. T.

(2012). The long term effects of low intensity supervision for lower risk probationers: Updated results from a randomized controlled trial. Journal of Criminal Justice, 35, 200-220.

Beck

A. T.

(1964). Thinking and depression II: Theory and therapy. Archives of General Psychiatry, 10, 561-571.

Beck

A. T.

(1999). Prisoners of hate: The cognitive basis of anger, hostility and violence. New York, NY: HarperCollins.

10.

Beck

(2011). Cognitive behavior therapy: Basics and beyond. New York, NY: Guilford Press.

11.

Beck

Fernandez

(1998). Cognitive-behavioral therapy in the treatment of anger: A meta-analysis. Cognitive Therapy and Research, 22, 63-74.

12.

Berk

R. A.

(2008). Forecasting methods in crime and justice. Annual Review of Law and Social Science, 4, 219-238.

13.

Berk

R. A.

(2012). Criminal justice forecasts of risk: A machine learning approach. New York, NY: Springer.

14.

Berk

R. A.

Sherman

L. W.

Barnes

G. C.

Kurtz

Ahlman

(2009). Forecasting murder within a population of probationers and parolees: A high stakes application of statistical learning. Journal of the Royal Statistical Society, Series A: Statistics in Society, 172, 191-211.

15.

Bush

Glick

Taymans

Guevara

(2011). Thinking for a change: Integrated cognitive behavior change program, Version 3.1. Washington, DC: National Institute of Corrections.

16.

Bushway

S. D.

(2011). Labor markets and crime. In Wilson

J. Q.

Petersilia

(Eds.), Crime and public policy (pp. 183-209). New York, NY: Oxford University Press.

17.

Butler

A. C.

Chapman

J. E.

Forman

E. M.

Beck

A. T.

(2006). The empirical status of cognitive-behavioral therapy: A review of meta-analyses. Clinical Psychology Review, 26, 17-31.

18.

Clark

D. A.

Beck

Alford

B. A.

(1999). Scientific foundations of cognitive theory and therapy of depression. New York, NY: John Wiley.

19.

Cohen

(1988). Statistical power analysis for the behavioral sciences. Hillsdale, NJ: Lawrence Erlbaum.

20.

DeRubeis

R. J.

Beck

A. T.

(1988). Cognitive therapy. In Dobson

K. S.

(Ed.), Handbook of cognitive-behavioral therapies (pp. 273-306). New York, NY: Guilford Press.

21.

Dobson

K. S.

Khatri

(2000). Cognitive therapy: Looking backward, looking forward. Journal of Clinical Psychology, 56, 907-923.

22.

Gifford

S. L.

(2002). Justice expenditures and employment in the United States, 1999. Washington, DC: Bureau of Justice Statistics.

23.

Gill

C. E.

(2014). Intensive probation and parole. In Bruinsma

Weisburd

(Eds.), Encyclopedia of criminology and criminal justice (pp. 2581-2592). New York, NY: Springer.

24.

Glaze

L. E.

Kaeble

(2014). Correctional populations in the United States, 2013. Washington, DC: Bureau of Justice Statistics. Retrieved from http://www.bjs.gov/index.cfm?ty=pbdetail&iid=5177

25.

Goldstein

Glick

(1994). Aggression replacement training: Curriculum and evaluation. Simulation & Gaming, 25, 9-26.

26.

Guevara

Solomon

(2009). Implementing evidence-based policy and practice in community corrections. Washington, DC: Crime and Justice Institute and The National Institute of Corrections.

27.

Hollin

C. R.

McGuire

Hounsome

J. C.

Hatcher

R. M.

Bilby

C. A.

Palmer

E. J.

(2008). Cognitive skills behavior programs for offenders in the community: A reconviction analysis. Criminal Justice and Behavior, 35, 269-283.

28.

Hollis

Campbell

(1999). What is meant by intention to treat analysis? Survey of published randomised controlled trials. British Medical Journal, 319, 670-674.

29.

Hyatt

J. M.

(2013). The impact of cognitive-behavioral therapy on the recidivism of high risk probationers: Results from a randomized trial (Doctoral dissertation) (Publicly Accessible Penn Dissertations, Paper 644). Retrieved from http://repository.upenn.edu

30.

Hyatt

J. M.

Barnes

G. C.

(2014). An experimental evaluation of the impact of intensive supervision on the recidivism of high-risk probationers. Crime & Delinquency. Advance online publication. doi:10.1177/0011128714555757

31.

Jalbert

S. K.

Rhodes

Flygare

Kane

(2010). Testing probation outcomes in an evidence-based practice setting: Reduced caseload size and intensive supervision effectiveness. Journal of Offender Rehabilitation, 49, 233-253.

32.

Jucovy

McClanahan

W. S.

(2008). Reaching through the cracks: A guide to implementing the youth violence reduction partnership. Philadelphia, PA: Public/Private Ventures.

33.

Kazemian

(2007). Desistance from crime: Theoretical, empirical, methodological, and policy considerations. Journal of Contemporary Criminal Justice, 23(5), 5-27.

34.

Lageson

Uggen

(2013). How work affects crime—and crime affects work—over the life course. In Gibson

Krohn

M. D.

(Eds.), Handbook of life-course criminology (pp. 201-212). New York, NY: Springer.

35.

Lipsey

M. W.

Chapman

G. L.

Landenberger

N. A.

(2001). Cognitive-behavioral programs for offenders. Annals of the American Academy of Political and Social Science, 578, 144-157.

36.

Lipsey

M. W.

Landenberger

N. A.

(2005). Cognitive-behavioral interventions: A meta-analysis of randomized controlled studies. In Walsh

B. C.

Farrington

D. P.

(Eds.), Preventing crime: What works for children, offenders, victims, and places (pp. 57-71). Berlin, Germany: Springer.

37.

Lipsey

M. W.

Landenberger

N. A.

Wilson

S. J.

(2007). Effects of cognitive-behavioral programs for criminal offenders. Oslo, Norway: Campbell Systematic Reviews.

38.

Lipsey

M. W.

Wilson

D. B.

(1993). The efficacy of psychological, educational, and behavioral treatment: Confirmation from meta-analysis. American Psychologist, 48, 1181-1209.

39.

Little

(2000). Cognitive-behavioral treatment of offenders: A comprehensive review of MRT outcome research. Addictive Behaviors Treatment Review, 2(1), 12-21.

40.

Little

Robinson

(1993). Cognitive behavioral treatment of felony drug offenders: A five-year recidivism report. Psychological Reports, 73, 1089-1090.

41.

Lösel

(1995). The efficacy of correctional treatment: A review and synthesis of meta-evaluations. In McGuire

(Ed.), What works: Reducing reoffending (pp. 79-103). Chichester, UK: John Wiley.

42.

MacKenzie

D. L.

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

43.

Matthews

(1997). Information processing biases in emotional disorders. In Clark

D. M.

Fairburn

C. G.

(Eds.), Science and practice of cognitive behaviour therapy (pp. 47-67). New York, NY: Oxford University Press.

44.

McGuire

Bilby

C. A.

Hatcher

R. M.

Hollin

C. R.

Hounsome

Palmer

E. J.

(2008). Evaluation of structured cognitive-behavioural treatment programmes in reducing criminal recidivism. Journal of Experimental Criminology, 4, 21-40.

45.

Milkman

H. B.

Wanberg

K. W.

(2007). Cognitive-behavioral treatment: A review and discussion for corrections professionals. Washington, DC: National Institute of Corrections, U.S. Department of Justice.

46.

Paparozzi

M. A.

Gendreau

(2005). An intensive supervision program that worked: Service delivery, professional orientation, and organizational supportiveness. The Prison Journal, 85, 445-466.

47.

Pearson

F. S.

Lipton

D. S.

Cleland

C. M.

Yee

D. S.

(2002). The effects of behavioral/cognitive-behavioral programs on recidivism. Crime & Delinquency, 48, 476-496.

48.

Petersilia

(1998). Community corrections: Probation, parole, and intermediate sanctions. New York, NY: Oxford University Press.

49.

Petersilia

Turner

(1990). Intensive supervision for high-risk probationers. Santa Monica, CA: RAND.

50.

Petersilia

Turner

(1993). Intensive probation and parole. In Tonry

(Ed.), Crime and justice (pp. 281-335). Chicago, IL: University of Chicago Press.

51.

Redondo

Sanchez-Meca

Garrido

(1999). The influence of treatment programmes on the recidivism of juvenile and adult offenders: A European meta-analytic review. Psychology, Crime & Law, 5, 251-278.

52.

Robinson

(1995). The impact of cognitive skills training on post-release recidivism among Canadian federal offenders. Ottawa, Ontario: Correctional Research and Development, Correctional Service Canada.

53.

Ross

R. R.

Fabiano

E. A.

Ewles

C. D.

(1988). Reasoning and rehabilitation. International Journal of Offender Therapy and Comparative Criminology, 32, 29-36.

54.

Sampson

R. J.

Laub

J. H.

(2003). Life-course desisters? Trajectories of crime among delinquent boys followed to age 70. Criminology, 41, 319-339.

55.

Skeem

J. L.

Manchak

(2008). Back to the future: From Klockars’ model of effective supervision to evidence-based practice in probation. Journal of Offender Rehabilitation, 47, 220-247.

56.

Spruance

L. M.

Van Voorhis

Listwan

S. J.

Pealer

Seabrook

(n.d.). The Georgia cognitive skills experiment: Phase II process evaluation. Cincinnati, OH: University of Cincinnati. Retrieved from http://www.uc.edu/content/dam/uc/ccjr/docs/reports/project_reports/Georgia_Phase_2_process_eval.pdf

57.

Taxman

F. S.

(2002). Supervision-exploring the dimensions of effectiveness. Federal Probation, 66(2), 14-27.

58.

Taxman

F. S.

Marlowe

(2006). Risk, needs, responsivity: In action or inaction? Crime & Delinquency, 52, 3-6. doi:10.1177/0011128705281757

59.

Taxman

F. S.

Thanner

Weisburd

(2006). Risk, need, and responsivity (RNR): It all depends. Crime & Delinquency, 52, 28-51.

60.

Thanner

Taxman

F. S.

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

61.

Van Voorhis

Spruance

L. M.

Ritchie

P. N.

Johnson-Listwan

Seabrook

Pealer

. (2004). The Georgia cognitive skills experiment: Outcome evaluation, Phase II. Retrieved from https://www.uc.edu/content/dam/uc/ccjr/docs/reports/project_reports/Georgia_Phase_II_final_report.pdf

62.

Wanberg

K. W.

Milkman

H. B.

(2006). Criminal conduct and substance abuse treatment: Strategies for self-improvement and change—The participant’s workbook. Thousand Oaks, CA: SAGE.

63.

Welsh

B. C.

Farrington

D. P.

(2000). Monetary costs and benefits of crime prevention programs. Crime and Justice, 27, 305-361.

64.

Wilson

Bouffard

Mackenzie

(2005). Quantitative review of cognitive behavioral programs. Criminal Justice and Behavior, 32, 172-204.