Who Benefits From Cognitive Skills Programs?

Abstract

Cognitive skills programs, which teach problem-solving skills and perspective taking, have a strong evidence base for their ability to reduce recidivism with convicted populations. This study explored whether the Enhanced Thinking Skills program, delivered over several years to 21,000 male prisoners in England and Wales, reduced reoffending for some categories of offenders more than others by comparing predicted with actual reconviction rates. Consistent with earlier research, attending the program was associated with significantly reduced reoffending for sexual offenders (a 13 percentage point reduction), violent offenders (17 point reduction), and other non-acquisitive offenders (10-12 points), but not for offenders convicted of burglary or robbery. After controlling for risk, age, previous offenses committed, sentence length, and program completion, current offense type persisted as an independent and significant predictor of reoffending. Implications for the targeting of cognitive skills programs from this evaluation of a real world, large-scale implementation, and directions for future research, are discussed.

Keywords

offender offending behavior program rehabilitation cognitive-behavioral intervention reducing reoffending

Cognitive skills programs are cognitive-behavioral programs designed to help offenders solve problems and make personal decisions more effectively by helping offenders learn how, rather than what, to think. Originating in Canada, this approach to offender rehabilitation is based on research demonstrating that offenders have distinctive thinking styles (Ross & Fabiano, 1985). These programs are now widespread (Hollin & Palmer, 2009). Two recent U.K. studies on the effectiveness of cognitive skills programs in reducing the reoffending of prisoners (Sadlier, 2010; Travers, Wakeling, Mann, & Hollin, 2013) confirmed earlier findings, in both the U.K. and other jurisdictions, of a significant treatment effect (Friendship, Blud, Erikson, & Travers, 2002; Robinson, 1995; Tong & Farrington, 2006, 2008). In England and Wales, using different methodologies and different tranches of data, the Sadlier (2010) and Travers et al. (2013) studies concluded that the Enhanced Thinking Skills (ETS) program (Clark, 2000) reduced reconviction rates by between 6 and 8 percentage points. Although some earlier studies did not identify such a treatment effect (Cann, Falshaw, Nugent, & Friendship, 2003; Falshaw, Friendship, Travers, & Nugent, 2003) and others were challenged by high attrition rates in differentiating the positive treatment and completion effects in the community setting (Hollin et al., 2008; McGuire et al., 2008; Palmer et al., 2008), the weight of evidence seems now to support a positive treatment effect for this type of intervention.

Cognitive Skills Programs and the Risk Principle

Cognitive skills programs are commonly associated with the popular Risk Needs Responsivity (RNR) model of offender rehabilitation as proposed by Andrews and Bonta (2010). The RNR model predicts that reductions in reoffending will occur from participation in programs that (a) are proportionate in dose to the participants’ risk of reoffending, (b) target known criminogenic needs, and (c) are structured and skills-based, target multiple needs, and follow cognitive-behavioral or social learning principles (see, for example, MacKenzie, 2000).

The targets of cognitive skills programs are generally impulse control, emotional regulation, consequential thinking, and problem-solving skills; which Andrews and Bonta (2010) identified as part of the antisocial personality pattern, one of the Big Four criminogenic factors. Cognitive skills programs tend to be short in dose, providing about 50 to 100 hours of intervention. This level of dose has been found to be sufficient to reduce recidivism for moderate risk offenders (Bourgon & Armstrong, 2005). However, Bourgon and Armstrong concluded that for high-risk or high-need offenders, a 200-hour program was sufficient, and for high-risk and high-need offenders, a 300-hour program was required. In their study, “risk/need” was determined by LSI-OR score (Level of Service Inventory–Ontario Revision; Andrews, Bonta, & Wormith, 1995), but the report did not state what the expected reoffending rates of each risk group were. Therefore, we must estimate these from the reoffending rates of the untreated control group, which were 28% for the low-risk group, 43.8% for the moderate risk group, and 59.1% for the high-risk group. Sperber, Latessa, and Makarios (2013) also concluded that 200 hours of programming were required to reduce reoffending in high-risk offenders. Again in this study, risk was classified using the LSI-R (The Level of Service Inventory–Revised; Andrews & Bonta, 1995), and no estimates of reoffending were given for the different risk groups, thus limiting the applicability of the findings to jurisdictions that do not use the LSI-R.

The RNR model also predicts that programming will not reduce reoffending in low-risk offenders; indeed, several studies have found a detrimental effect of providing interventions to this group (Latessa, Brusman-Lovins, & Smith, 2010; Lovins, Lowenkamp, & Latessa, 2009; Lowenkamp & Latessa, 2005). In contrast, in their first meta-analysis of the Reasoning and Rehabilitation (R&R) cognitive skills program, Tong and Farrington (2006) reported that the program had a similar positive impact with both low and high-risk offenders. In the 2008 update of this analysis, however, they described a significant impact for high-risk prisoners only (Tong & Farrington, 2008). There was little discussion, unfortunately, of what these risk labels mean in terms of the actual likelihood of reoffending encompassed by each risk level. This omission makes interpretation of these findings problematic particularly if we want to translate this evidence into practical targeting guidance. Landenberger and Lipsey (2005), in their review of cognitive-behavioral interventions with offenders, also reported that larger effect sizes were associated with higher risk participants but provided little detail on the definitions and boundaries of “higher risk” applied either in the individual studies or in the synthesis of results.

For consistency with the RNR model, then, the prediction for cognitive skills programs such as ETS would be that they would reduce reoffending in moderate risk offenders, would be insufficient to change reoffending in higher risk individuals, and would be ineffective with lower risk individuals. However, several evaluations of cognitive skills programs have identified some departures from the predicted effect of these programs. On one hand, cognitive skills programs may have an impact on both lower and higher risk offenders than the RNR model would predict. On the other hand, they may not reduce reoffending for all types of offenders, even those who have the deficits that the programs tackle.

Differential Responding by Risk of Reoffending

Several evaluations of cognitive skills programs have attempted to investigate differential effects across offenders’ risk levels on reoffending. Robinson (1995) investigated the impact of cognitive skills training along several dimensions, including risk and offense type, in his random allocation study of 4,072 Canadian offenders referred for the R&R course while in custody. To assess the interaction of risk and program impact, Robinson split his treatment and control samples into two equal sized groups according to recidivism risk predictor scores derived from static criminal history variables, and labeled these high and low risk. The average reconviction rate observed for the “low-risk” controls in the Robinson study was 20%, while for the “high-risk” controls, it was 30%. Robinson reported a significant impact of R&R completion, equivalent to a 7 percentage point reduction in the recidivism rate for the low-risk group compared with controls. Tong and Farrington (2006) have subsequently presented this effect size as an odds ratio of 1.53, equivalent to a 35% reduction in recidivism. There was no significant treatment effect for the higher risk group.

More in line with the RNR model, Friendship et al. (2002) observed better outcomes for the low-medium- and high-medium-risk groups of ETS participants compared with the lowest or highest risk bands. In this study, the treatment group as a whole had initially a much higher mean 2-year risk of reconviction of 60% and a significant treatment effect was reported only for “medium-risk” offenders (i.e., those with a 2-year recidivism prediction score of between 25% and 75%). Similarly, Sadlier (2010) reported greater impact of the same program for those who fully met the program’s selection criteria (one of which pertained to predicted risk), although the actual risk scores of the participants and control group are not cited in the study.

However, Travers et al. (2013) observed an apparent treatment effect of the ETS program on all but the very highest risk offenders (those whose predicted reoffending rate exceeded 90%). The observed impact on high-risk offenders, whose predicted reoffending rate was between 75% and 90%, would not have been predicted by the RNR model.

Differential Responding According to Offense Type

Only one study, to our knowledge, has examined differential responding to cognitive skills programs by offense type (i.e., the nature of the current conviction). While it could be argued that categorizing individuals by their current offense masks a history of diverse forms of offending in most cases, there is reasonable support in the literature for the notion that most offenders specialize to a certain extent in particular forms of criminal behavior (Howard, Barnett, & Mann, 2014; Soothill, Fitzpatrick, & Francis, 2009). Hence, the issue of differential needs and responding according to conviction seems at least reasonable to explore. Examining this question, Robinson (1995) found that offenders with current convictions for sexual, violent (excluding robbery), and drugs offenses all showed a positive response to cognitive skills training. For instance, sexual offenders who completed the program showed a 57.8% drop in recidivism compared with the control group, and violent and drugs offenders who completed the program showed reductions of 35.3% and 36.3%, respectively. However, robbery offenders and non-violent property offenders appeared not to benefit from R&R, with near identical recidivism rates observed in treatment and control groups at the end of the follow-up period. The sample size in Robinson’s study precluded any further differentiation by offense type within these broad categories. Robinson suggested several possible explanations for this lack of treatment effect with robbery and property offenders. He noted, for instance, that these groups consisted of higher risk individuals than did the other offense type groups. He hypothesized that offenders in these groups may hold stronger, more entrenched pro-criminal attitudes, or have more serious substance abuse problems, or feel less driven to change their offending lifestyle.

While Robinson was not able to test any of these hypotheses, evidence has accrued from the wider literature that can shed some additional light on these potential explanations for failure to respond to treatment. Wilson, Attrill, and Nugent (2003) concluded from their examination of psychometric change over the course of cognitive skills training that acquisitive offenders (robbery was included in this class of offending) were generally the most needy offender type, as demonstrated by higher scores on measures of criminogenic beliefs and attitudes, in comparison with other offender types at pre-intervention testing. Acquisitive offenders also demonstrated significant positive shifts in the desired direction on most measures over the course of the program to the same or greater degree than other offender types. In addition, Debidin (2009) reported data from the national Offender Assessment System (OASys) database in England and Wales (on which assessments of offenders’ risk of reconviction, risk of harm and criminogenic needs are routinely recorded), which showed that around two thirds of burglary and robbery offenders were assessed as having thinking skills deficits—further evidence of the presence of this criminogenic need in acquisitive offenders. The Wilson et al. study was taken at the time to indicate that acquisitive offenders would benefit from cognitive skills training, but this assumption has not yet been tested by examining reconviction outcomes further to the psychometric impact observed post course.

Zamble and Quinsey (1997) reported considerable differences in motivation for offending between non-violent property offenders and violent offenders. Where violent offending was associated with a range of emotional and cognitive triggers, the only emotional trigger significantly associated with property offending was “frustration.” Property offenders saw their greatest problems as being a combination of deprived economic circumstances and substance misuse. Similarly, Willott and Griffin (1999) discussed the beliefs of working-class economic criminals who perceived that the state had reneged on them and had failed properly to provide for them, which left them no option but to commit crime to provide for their families. Nee and Meenaghan (2006) have described burglars as “Experts” in decision making, who process information rapidly and effectively and do not report impulsive or opportunistic choices as part of their criminal behavior. A similar theme was identified by Brezina and Topalli (2012) who described a strong sense of criminal self-efficacy among their sample that was particularly associated with crimes committed for monetary gain and which reduced the intention to desist from crime.

If these insights are correct, they imply that cognitive skills training might not address the major risk factors associated with acquisitive offending, even though, as indicated by Wilson et al.’s (2003) study, such offenders appear to present with impulsivity and problem-solving deficits. If property offenders are seen as making what are essentially rational choices to commit crime, they may well be less motivated to apply learning from a cognitive skills program for pro-social purposes, because desistance from offending is not their goal. In other words, property offenders may be less motivated to change following the intervention than, for example, individuals convicted of violence, for whom the costs of their offending might be more obvious than the benefits.

Of further relevance here is the finding that attending the ETS program has no significant impact on the self-reported impulsivity scores of prisoner participants convicted of acquisitive offenses of burglary, theft, and fraud (McDougall, Perry, Clarbour, Bowles, & Worthy, 2009). In that study, robbery was classed as a violent, not an acquisitive, offense. Participants with current convictions for non-acquisitive crimes were reported to have significantly lower (better) impulsivity scores after ETS compared with the acquisitive group. The acquisitive offenders in this study were demonstrating a similar or more marked level of need on a key program target compared with offenders with other types of current conviction, but did not appear to respond to the program in the same way demonstrating an absence of treatment effect on relevant short-term outcomes. Wilson et al. (2003) also reported that while acquisitive offender participants in ETS scored at least as highly at pre-test on self-report psychometrics of impulsivity, criminal thinking styles, and problem-solving skills, and post-course assessments indicated that some significant positive changes were observed, there was relatively little impact on impulsivity among acquisitive offenders, particularly those with a more established history of acquisitive offending. Whether changes on short-term outcomes such as impulsivity scales are relevant to longer term reoffending outcomes following cognitive skills interventions like ETS remains to be demonstrated, but these studies indicate some differential short-term responsivity to ETS by current offense type.

There are therefore some clues in the existing literature that some offenders may be less likely to reoffend following a cognitive skills program than others, but there has been insufficient systematic investigation of that differentiation. In difficult economic climates, there are only limited resources to direct to reduce reoffending among the offender population. Thus, the key aim of this study is to explore whether there is an empirical basis for better targeting of limited resource to best effect.

Aims of the Current Study

The current study aimed to identify whether differential patterns exist in the impact of cognitive skills training. The study examined the ETS program, a cognitive skills intervention delivered to prisoners and probationers across England and Wales between 1996 and 2010. ETS was designed and overseen by centrally located staff in the National Offender Management Service. The program has a range of treatment targets including impulse control, flexible thinking, values and moral reasoning, social perspective taking, critical reasoning, and interpersonal problem-solving. In this respect, ETS targeted mainly what Andrews and Bonta (2010) described as antisocial personality, and had less focus on antisocial attitudes, such as beliefs that crime is worthwhile. That is, the focus of ETS was on changing how people reason, not the content of their thoughts or attitudes. The course lasts for 20 sessions of about 2 hr each, typically delivered 2 or 3 times a week. Sessions are interactive and involve course members in discussion, role-playing, exercises, and assignments. ETS is manualized and program implementation is monitored via a comprehensive annual audit of each treatment site. A treatment manager is situated in each site, responsible for treatment integrity, staff management, and local adherence to RNR principles. There is a high degree of confidence that treatment integrity and treatment quality are consistently acceptable. ETS is targeted at medium- and high-risk offenders assessed as having the thinking styles targeted by the program. Toward the end of the period under study here, the risk of recidivism band targeted by the program was defined as those with a 2-year predicted likelihood of reoffending of at least 40%. Prior to this there appears to have been rather less precise instructions on the targeting of the program by risk level. ETS has been extensively evaluated, with the most recent evaluations being reported by Sadlier (2010) and Travers et al. (2013).

The current study used essentially the same participant sample as the Travers et al. (2013) ETS reconviction study, and compared actual offending with predicted offending using the Offender Group Reconviction Scale (OGRS; Copas & Marshall, 1998)—a high-quality predictor tool widely used in the criminal justice system in England and Wales. Predicted versus actual designs have varying levels of support as to their robustness. Sherman et al. (1998) devised the Maryland Scale of Scientific Methods to use in appraising the internal validity of outcome studies across five levels, where Level 5 represents the most robust. Harper and Chitty (2005) described the predicted versus actual design as equivalent to Level 2 on this Maryland Scientific Methods Scale, whereas others have argued that when the predictor is well validated, such a design can be considered closer to Level 4 (Sherman & Strang, 2007). As there was no systematic capture of risk and need data in this period, it was not possible to conduct a retrospective matched comparison study. Instead, a within-group design was used where the predicted reconviction rate, as assessed by OGRS, provided the counterfactual for the differential impact of ETS on key offender characteristics.

Based on the existing evaluation studies of cognitive skills programs, the following were hypothesized:

Hypothesis 1: Offenders with a main current conviction for a sexual or violent offense will show greater reductions in reoffending after completing a cognitive skills program than those with an acquisitive offense.

Hypothesis 2: A differential treatment effect for acquisitive offenders can be explained by higher predicted risk in this group.

Hypothesis 3: Program completion rates will be lowest among acquisitive offenders indicating issues with readiness change for this group in particular.

Hypothesis 4: The predictive value of current offense type for program impact will persist even when controlling for age, sentence length and heterogeneity of offending history.

Method

Participants

This study explored the reconviction rates of 21,373 male offenders aged 18 and over who had attended the ETS program (Clark, 2000) in prison, were released from custody between 1997 and 2005, and had been followed-up for at least 2 years following their release. This study includes all participants, including those who started but failed to complete the program. The small minority of offenders who had also attended the Controlling Anger and Learning to Manage It (CALM) accredited anger management program or one of the accredited sex offender treatment programs (SOTPs) were excluded from this sample. It is worth noting that those prisoners who had attended more than one program, and who were therefore excluded from this study, were only slightly higher risk than those attending just one program; the mean 2-year predicted recidivism risk score for the ETS-only sample was 56% (using the OGRS2 score), for ETS + CALM it was 58%, and for ETS + SOTP it was 26%. The mean predicted 2-year recidivism rates for those attending CALM-only or SOTP-only were 56% and 24%, respectively. Thus, while this study focuses on the ETS-only group, it is clear that their risk of general reoffending was broadly similar to those who participated in further offending behavior programs in the same prison sentence. There was little provision specifically for violent prisoners at that time with only a few hundred prisoners a year attending the CALM anger management program, compared with several thousand completing ETS. For many prisoners, ETS was the only program available to them.

A profile of the participants’ characteristics is presented in Table 1. This sample includes those prisoners who constituted the sample in the Travers et al. (2013) study of reconviction rates following ETS. To maximize sample size and generalizability, the current study also included further ETS participants released between 1997 and 2000 and those across the whole period sentenced to fewer than 12 months in prison, who had been excluded from the previous study. To have a sentence length value for every participant in the logistic regression on reconviction outcomes, offenders on indeterminate life sentences were imputed a sentence length of 20 years. The average time in custody served by male offenders on mandatory life sentences was between 13 and 14 years over the period of this study (Ministry of Justice, 2010b). This can be approximated by imputing a fixed sentence length of 20 years under which offenders become eligible for release on parole conditions between the 10- and 15-year points of their sentence.

Table 1:

Male ETS Participants in Custody: 2000 to 2005 (N = 21,373)

	M	SD
Age at program start—Years	29.06	9.69
Age at release—Years	30.03	9.90
Count of previous convictions	9.98	7.96
Previous custodial sentences	3.61	3.58
Previous custodial sentences <18 years	0.49	1.05
OGRS2 risk of reconviction—2 years	55.61	21.30
	N	%
Ethnicity
White	17,183	80.40
Black	2,695	12.61
Asian	982	4.59
Other/not recorded	513	2.40
Current offense
Sex	1,824	8.53
Violence	6,358	29.75
Robbery	3,007	14.07
Acquisitive	4,969	23.25
Drugs	4,072	19.05
Other	1,143	5.35
Length of current sentence
Less than 1 year	2,687	12.57
1-2 years	3,362	15.73
2-4 years	7,212	33.74
4 years or more	7,184	33.61
Indeterminate	297	1.39
Not known	631	2.95

Note. ETS = Enhanced Thinking Skills; OGRS = Offender Group Reconviction Scale.

Offense Categorization

Participants were assigned to an offense type category according to their main current offense. Where participants had more than one current conviction, they were categorized according to the offense that had been awarded the most severe penalty at court. Different studies and different risk assessment tools have applied a variety of offense typologies. Sadlier (2010) used the same five categories as Robinson (1995) but included an additional “other” category. A six-category typology similar to Sadlier was applied in this study (Table 2). The definition of violence included public order and criminal damage offenses.

Table 2:

Offense Type Categorization

Offense Type Category	Offenses Included
Sexual offenses	All sexual offenses against adult and child victims
Violence	Murder, attempted murder, all assaults, firearms and offensive weapon offenses, arson, public order, criminal and malicious damage, aggravated burglary
Robbery	Robbery
Acquisitive (non-violent)	Burglary both domestic and non-domestic, theft, handling, fraud and forgery, theft of and from motor vehicles
Drugs	Import/export, production, supply, and possession
Other	All other offenses, including soliciting and prostitution, motoring, and drink driving

Measures

Reconviction Data

Reconviction data were provided by the U.K. Home Office Police National Computer (HOPNC). The actual yes/no reconviction rate measures the proportion of the sample convicted or cautioned within a specified time period with reconviction defined, following Lloyd, Mair, and Hough (1994), as “an appearance in court where there has been at least one finding of guilt, irrespective of how many offenses were dealt with on a single appearance” (p. 5). There was a fixed 2-year follow-up for all participants starting at the date of release from custody. Pseudo-reconvictions were taken into account in that any reconvictions which related to an offense which occurred before the index offense were not counted as reconvictions. Some participants will have returned to custody following their first release while still in the at-risk period either for a breach of the conditions under which they were released or on remand or on new custodial sentences but none of these is discernible from the data available to this study. A study by the Ministry of Justice for England and Wales (Ministry of Justice, 2010a) aimed to quantify the amount of at-risk time lost to periods back in custody during a 12-month follow-up. The analysis found that 20% of offenders spent an average of 81 days in custody in the 12-month follow-up, but the bulk of this group were those who were also reconvicted in that 12 month period. Only 4.5% of those who were not reconvicted spent any time back in custody in the first 12 months. If the outcome in this study were a frequency of reoffending measure, at-risk time lost to returns to custody during follow-up would have greater impact than appears to be the case when using a simpler binary reconviction measure as applied here.

The OGRS risk of recidivism assessment

The OGRS (Copas & Marshall, 1998) is a risk-prediction tool based on the “static” variables of an individual’s history of offending combined with demographic variables such as age and gender. The OGRS produces a statistical risk score, based on a weighted sum of certain covariates, which undergo a logistic transformation. The scale is periodically refreshed and recalibrated and new calculators and guidance are issued to operational staff. During the period of this study, the second version, OGRS2, was in operation (Taylor, 1999). OGRS3 (Howard, Francis, Soothill, & Humphreys, 2009) is currently operational across the prison and probation services in England and Wales and OGRS4 is in preparation. The variables that inform OGRS2 include age at the time of sentence, gender, number of youth custodial sentences, current offense, age at current conviction, age at first conviction, and the Copas rate variable (the rate at which offenders are convicted, such that an offender with five convictions within 5 years from first to current conviction will have a higher rate of conviction than an offender with five convictions within 10 years). The variables contributing to the OGRS2 score are weighted to reflect their predictive power in relation to reconviction. The scale estimates the probability (presented as a percentage score) that an offender will be reconvicted of any offense within 2 years of release from custody or from the start of a community sentence.

OGRS2, the version used in this study, has an Area Under the Curve estimate (AUC) of 0.77 for general reoffending (Howard et al., 2009), which Kraemer et al. (2003) would describe as a larger than typical association in the behavioral sciences. Coid et al. (2007) administered a range of actuarial risk-prediction tools to a sample of prisoners convicted for sexual and violence offenses and demonstrated that OGRS2 yielded the highest AUC scores for all the reconviction outcomes surveyed. Maden et al. (2006) similarly demonstrated the utility of the OGRS2 scale in accurately predicting the recidivism of clients in a forensic mental health setting.

Travers et al. (2013) described how the OGRS2 predicted rate of reconviction was within 1.6% of the actual 2-year reconviction rate for a large national cohort of adult male prisoners over the same time period as this study. This demonstrates the potential utility of the predicted rate as the counterfactual for a national intervention delivered at the same time. There is also some evidence that the accuracy of the OGRS2 prediction is consistent across different offense type sub-groups of the offender population with an AUC of 0.78 for general reoffending and AUCs ranging from 0.66 to 0.78 (with the exception of an AUC of 0.52 for threat/harassment) for different types of violent reoffending (Debidin, 2009). Therefore, there is some assurance that any within-group variations observed will represent real differences in the responsivity of different types of offender to the ETS program.

The Program

The ETS program is a cognitive-behavioral intervention designed to provide offenders with new skills to interrupt their impulsive, short-term thinking with more successful social problem-solving leading to positive interpersonal interactions. The program was accredited for delivery in Her Majesty’s Prisons in 2000 and was delivered widely in both prison and probation settings until 2010, when it was replaced with an updated cognitive skills intervention, the Thinking Skills Program (Harris & Riddy, 2010). The ETS theory manual (Clark, 2000) describes the skills the program was designed to boost as problem-solving, perspective taking, empathy, impulse control, and critical reasoning. ETS consists of 20 two-hour sessions delivered to groups of participants by two trained facilitators. Sessions begin with an ice-breaker exercise and a description from the facilitator of the aims of the session. A variety of cognitive-behavioral techniques are used including practical tasks, discussions, role-play, and games. Sessions end with a plenary where the facilitator aims to emphasize the main learning objectives of the day and participants are encouraged to complete homework tasks in their own time before the next session. Facilitators are trained to make the training materials relevant to the everyday lives of the participants and to make the sessions as interactive and as little like school as possible. The course was designed so that more complex skills are introduced only after the basic constituent skills have been introduced and a degree of over-learning and repetition is used to allow for the assimilation of these new skills.

ETS is targeted at medium- and high-risk offenders who must also be assessed as having the thinking deficits targeted by the program. From 2004, program staff were able to use the OASys risk and need assessment (Howard, Clark, & Garnham, 2006) to ascertain an offender’s risk of reconviction using either the OASys likelihood of reconviction score (combines both static and dynamic items) or the OGRS2 score (uses only static items; Copas & Marshall, 1998). The OGRS risk assessment can be calculated even when a full OASys assessment is not available, as it is derived from static details concerning current offense, age, gender, and criminal history. Before these risk tools were routinely available, program staff were directed to prioritize medium-risk prisoners through an assessment of their current offense type, age, previous criminal history, and evidence of more than one dynamic risk factor. With the introduction of OASys toward the end of this period of study, the specification of risk thresholds for ETS became more precise and program staff were directed to select prisoners who scored 56 or above on the total OASys score, or 40 and above on OGRS2. In practice, this means that the target group for ETS came to be those whose likelihood of reoffending is estimated at 40% or higher with exceptions made for prisoners convicted of sex offenses or those serving indeterminate sentences, many of whom would have relatively low OGRS scores. Once an offender had passed the risk threshold, their need for ETS was formally assessed via a semi-structured interview to elicit their thinking habits or cognitive styles. Since 2004, this interview has been gradually superseded by the introduction of the OASys assessment in which the Thinking and Behavior section includes seven items relevant to suitability for ETS.

Analysis Plan

This study aimed to observe the actual reconviction outcomes for a cohort of ETS participants and compare these with the expected rates of reconviction for the group derived from their scores on the OGRS2 risk tool. No experimental manipulations were made nor control groups generated as participants’ own predicted risk of reconviction was used as the counterfactual for observed rates in this within-group design. During this period, there was no consistent capture of offenders’ risk levels or criminogenic needs; therefore, we were unable to create an appropriate control condition for ETS participation retrospectively. We identified individuals’ predicted reconviction rates as an adequate, if less than optimal, counterfactual which allowed us to explore responsivity to the program among this large group of program participants which would otherwise remain untested.

A series of chi-squares were planned to test for an association between the binary reconviction outcomes observed for all ETS participants over the period of study and various typologies of offense type (with a correction applied for multiple comparisons) and also between reconviction and levels of risk (deciles of OGRS2 score). A logistic regression would test the relative predictive influence of current and previous offense types along with age, OGRS2 score, sentence length, program completion, ethnicity, and year of release on the binary 2-year reconviction rate. These further variables were intended to capture that variance in the relationship between predicted and actual reconviction that will relate to differences in age, sentence length, ethnicity, previous offending, and changes in the criminal justice system over time. For the logistic regression analysis, the reference category for offense type was taken to be violence as that represented the modal offense type for the whole group and would allow for a direct test of the hypothesis that sex and violence offenders would be more responsive to the program than would acquisitive offenders. White offenders were used as the reference category for ethnicity as again they represented the modal ethnic group. For year of release, 2005 was selected as the reference category to make that test the most relevant to the current picture. A staged entry regression would test the added value of offense type over risk in predicting reconviction rates.

Results

Table 3 shows actual and predicted 2-year reconviction rates for the sample by offense category, for all levels of risk combined. A chi-square test was applied to investigate whether the differences from predicted to actual reconviction rates presented in Table 3 were associated with current offense type. This analysis found the association between offense type and reductions in reconviction rate to be statistically significant, χ²(5) = 559.98, p < .001. The absolute percentage point drop in reconviction rates (column C in Table 3) ranged between 10 and 17 points across all offense types with the exception of acquisitive and robbery offenders where the reduction was close to zero.

Table 3:

Actual Versus Predicted 2-Year Reconviction Rates Following ETS by Offense Type

Offense Type	N	Mean Age at Start of ETS (Years)	Mean Sentence Length (Months)	ETS Attrition Rate%	APredicted Reconviction Rate (OGRS2)%	BActual Reconviction Rate%	CAbsolute Difference(A − B)%	DRelative Difference(C / A) × 100%
Sexual	1,824	42.13	51.91	4.7	26.30	13.60	12.71	48.31
Violence	6,358	28.31	47.96	5.5	57.86	40.58	17.28	29.87
Robbery	3,007	25.25	48.61	6.8	49.78	52.84	−3.06	−6.15
Acquisitive	4,969	26.91	22.84	10.1	71.80	71.48	0.32	0.45
Drugs	4,072	30.33	49.19	4.5	46.07	35.95	10.11	21.95
Other	1,143	27.31	19.24	7.9	68.83	56.87	11.96	17.38
Total	21,373	29.06	41.46	6.6	55.61	47.18	8.44	15.17

Note. ETS = Enhanced Thinking Skills; OGRS = Offender Group Reconviction Scale.

The observed relative reduction from predicted to actual reconviction after ETS diminished as risk increased, but the magnitude of the absolute reduction appeared fairly consistent from the second up to the eighth risk deciles (Table 4). The average reduction from predicted reconviction rates remained fairly constant across all levels of risk except those at the extremes of the risk scale (i.e., where predicted reconviction rates were less than 10% or greater than 70% or, more clearly, above 80%). Having observed some variation in the pattern of change from predicted to actual reconviction at different risk levels, the offense type analysis was then further broken down by risk category within each offense type.

Table 4:

Actual Versus Predicted 2-Year Reconviction Rates Following ETS by Risk Band—All Offense Types

Risk Band (OGRS2)	N	PredictedReconviction Rate (OGRS2)%	ActualReconviction Rate%	Absolute Difference%	Relative Difference%
0 ≤ 10	583	9.11	3.26	5.85	64.23
11-20	1,217	17.08	8.22	8.86	51.89
21-30	1,433	26.18	14.45	11.74	44.83
31-40	2,387	36.10	22.83	13.26	36.75
41-50	2,715	46.32	35.10	11.22	24.23
51-60	3,125	56.16	46.59	9.57	17.04
61-70	3,772	66.06	57.56	8.50	12.87
71-80	3,804	75.82	70.45	5.37	7.08
81-90	2,152	84.96	82.90	2.06	2.43
91+	185	92.49	90.81	1.68	1.81
Total	21,373	55.61	47.18	8.44	15.17

Note. ETS = Enhanced Thinking Skills; OGRS = Offender Group Reconviction Scale.

Reconviction Rates by Offense Category

Table 5 presents the predicted and actual reconviction rates for the six major offense type categories by risk level, and indicates those observed rates of reconviction which were significantly different to those predicted by average OGRS2 scores. This goodness of fit was tested with a series of chi-square tests applying an adjusted critical value of chi-square to correct for multiple comparisons. To correct for running multiple chi-square tests and to minimize spurious findings, an adjusted critical chi-square value of 11.24 was applied here which has an associated p value of .0008 (i.e., a p threshold of .05 divided by the 60 tests conducted). The reduction from predicted rates of reconviction was seen to vary across both offense type and risk level. The significant change for sex offenders was seen with those whose OGRS2 scores fell between 0 and 60. For violence and drugs offenders, the impact on predicted rates was statistically significant with those whose OGRS2 scores fell between 11 and 80. In the smaller “other” category, the apparent effect of ETS was seen with those scoring 31 to 90. In the robbery and acquisitive categories, there were no significant differences between predicted and actual reconviction rates at any level of risk but for one small group of lower risk acquisitive offenders.

Table 5:

Actual Versus Predicted 2-Year Reconviction Rates by Risk Band and Current Offense Type Following ETS

OGRS2 Risk Band	Sex			Violence			Robbery			Acquisitive			Drugs			Other
OGRS2 Risk Band	N	Predicted Rate	Actual Rate	N	Predicted Rate	Actual Rate	N	Predicted Rate	Actual Rate	N	Predicted Rate	Actual Rate	N	Predicted Rate	Actual Rate	N	Predicted Rate	Actual Rate
≤10	470	9.23	2.55*	14	9.1	0	6^a	—	—	1^a	—	—	91	8.51	6.59	1^a	—	—
11-20	519	16.93	6.74*	155	17.8	3.23*	75	17.64	16	12	15.9	8.33	442	16.98	10.41*	14	15.79	7.14
21-30	264	25.32	9.09*	304	26.55	7.89*	277	26.06	28.52	38	26.27	0.00*	508	26.34	15.35*	42	27.72	4.76
31-40	212	35.64	21.23*	845	36.07	13.37*	517	35.88	37.72	95	37.05	26.32	656	36.31	24.54*	62	36.11	9.68*
41-50	136	45.93	17.65*	894	46.42	24.05*	648	46.29	50.62	240	46.49	40.83	710	46.29	38.03*	87	46.08	20.69*
51-60	89	55.59	32.58*	1,064	56.19	35.71*	718	55.88	57.52	527	56.6	56.55	626	56.1	47.92*	101	56.39	35.64*
61-70	74	65.1	48.65	1,358	66.2	47.79*	539	65.43	69.94	1,084	66.42	66.61	541	65.74	53.97*	176	66.11	53.98*
71-80	49	74.98	75.51	1,205	75.72	64.40*	203	74.72	80.3	1,640	76.04	76.89	384	75.48	59.11*	323	76.3	66.87*
81-90	11	84.42	54.55	500	84.57	80.2	24	82.7	87.5	1,213	85.22	85.49	113	83.81	74.34	291	85.23	80.76
91+	0^a	—	—	19	91.95	89.47	0^a	—	—	119	92.58	92.44	1^a	—	—	46	92.5	89.13
Total	1,824	26.3	13.6	6,358	57.86	40.58	3,007	49.78	52.84	4,969	71.8	71.48	4,072	46.07	35.95	1,143	68.83	56.87

Note. Shaded cells indicate significantly lower actual reconviction rates than would be predicted from the average OGRS2 scores for offenders in that cell. ETS = Enhanced Thinking Skills; OGRS = Offender Group Reconviction Scale.

Reconviction rates are not reported for cells where N < 10.

Significant goodness of fit chi-square after Bonferroni correction, χ²(1) ≥ 11.24, p < .0008.

Versatility or Specialization of Offending

There could be a challenge to a categorization of offenders by their main index offense alone. There is debate about the extent to which offenders tend to be heterogeneous in their offending (Soothill et al., 2009) and specialization is certainly not so marked that a current offense of an acquisitive nature could not be associated with a history of violent offending or vice versa. Moreover, Table 3 indicates that both the age and sentence length of the current offense types varied considerably, with sex offenders being the oldest and longest serving prisoners in the sample, robbers the youngest, and acquisitive and other offenders serving the shortest sentences.

The heterogeneity, or versatility, of this sample’s offending is shown in Table 6, where the average number of previous offenses falling in each of the offense type groups is plotted against each current offense type. Although versatility of offending was apparent in the range of previous convictions in every current offense category, it was also clear that the most prevalent previous offending type was the same as the current offense type in every category. Thus, those with the highest number of sexual previous convictions were current sexual offenders, those with the highest number of violent previous convictions were current violence offenders, and so on. Although current offense appeared to serve quite well as a proxy for an offender’s predominant offending history, there is also evidence that previous acquisitive convictions were common across all offenders. If those types of offenses are associated with lower responsivity to cognitive skills programs, then this would need to be controlled for in a study of differential impact. The numbers of different types of previous convictions were therefore included in the final regression analyses.

Table 6:

Mean Previous Convictions by Current Offense Type

Current Offense Type	PreviousSexual Convictions	PreviousViolent Convictions	Previous Robbery Convictions	Previous Acquisitive Convictions	PreviousDrugs Convictions	Previous Convictions—Other
Sex	1.32	0.82	0.07	1.81	0.10	0.58
Violence	0.04	3.51	0.19	3.41	0.44	0.69
Robbery	0.03	1.83	1.36	4.21	0.55	0.66
Acquisitive	0.04	1.99	0.31	9.07	0.64	1.09
Drugs	0.03	1.53	0.15	3.64	2.09	0.83
Other	0.09	2.19	0.28	5.43	0.60	2.42
Total	0.15	2.24	0.37	4.85	0.80	0.89

Impact of ETS—More Detailed Offense Types

Tables 7 and 8 present predicted and actual reconviction rates for different offense categories within each offense type domain. These tables identify more precisely the nature of the different convictions that were included in the overarching offense type categories (sex, violence, robbery, acquisitive, drugs, and other) and are useful in prompting hypotheses around the relevant offender characteristics in these sub-groups that appear to be associated with varying levels of responsivity to cognitive skills programs.

Table 7:

Actual Versus Predicted 2-Year Reconviction Rates by Sex Offense Type and Risk

Risk Band(OGRS2)	N	PredictedReconviction Rate (OGRS2) %	ActualReconviction Rate%	AbsoluteDifference%	RelativeDifference%
Sex offenders—Adult victim
≤10	10	8.15	10.00	−1.85	−22.67
11-20	81	18.06	3.70	14.36	79.49
21-30	79	26.96	6.33	20.63	76.53
31-40	138	35.89	15.94	19.95	55.58
41-50	94	46.21	12.77	33.44	72.37
51-60	60	55.89	30.00	25.89	46.32
61-70	69	65.22	47.83	17.39	26.67
71-80	47	74.98	78.72	−3.74	−4.99
81-90	11	84.42	54.55	29.88	35.39
91+	0	—	—	—	—
Total	589	42.92	23.26	19.66	45.80
Sex offenders—Child victim
≤10	460	9.25	2.39	6.86	74.15
11-20	438	16.72	7.31	9.42	56.31
21-30	185	24.62	10.27	14.35	58.28
31-40	74	35.17	31.08	4.09	11.63
41-50	42	45.31	28.57	16.73	36.94
51-60	29	54.99	37.93	17.06	31.02
61-70	5^a	—	—	—	—
71-80	2^a	—	—	—	—
81-90	0	—	—	—	—
91+	0	—	—	—	—
Total	1,235	18.38	8.99	9.39	51.11

Note. OGRS = Offender Group Reconviction Scale.

Reconviction rates are not reported for cells where N < 10.

Table 8:

Actual Versus Predicted 2-Year Reconviction Rates by Category of Non-Sexual Current Offense

Offense Type	N	PredictedReconviction Rate (OGRS2)%	ActualReconviction Rate%	Absolute Difference%	Relative Difference%
Violence
Murder and manslaughter	689	44.42	17.13	27.30	61.45
Wounding, GBH, and firearms	2,201	55.99	38.26	17.73	31.67
Malicious wounding and offensive weapon	1,558	60.24	45.12	15.12	25.10
Arson	302	58.92	34.44	24.49	41.56
Public order	544	68.40	54.78	13.62	19.92
Other violence	1,064	61.25	48.40	12.85	20.98
Robbery
Robbery	3,007	49.78	52.84	−3.06	−6.15
Acquisitive
Domestic burglary	2,753	68.11	73.19	−5.09	−7.47
Other burglary	649	77.19	75.19	1.99	2.58
Theft	624	80.49	69.87	10.62	13.19
Handling	311	75.35	69.77	5.58	7.40
Fraud/forgery	257	60.17	44.36	15.81	26.27
Taking and driving away	270	80.29	77.04	3.26	4.05
Theft of vehicles	105	80.09	70.48	9.61	12.00
Drugs
Drug import/export/production	2,006	35.22	34.00	1.22	3.46
Drug possession/ supply	2,066	56.60	37.85	18.75	33.12
Other
Soliciting or prostitution	10	62.83	10.00	52.83	84.08
Absconding bail offense	150	83.26	70.00	13.26	15.93
Motoring offense	618	73.35	63.75	9.60	13.09
Drink driving	189	55.71	46.03	9.68	17.37
Other	176	55.08	35.80	19.29	35.02

Note. OGRS = Offender Group Reconviction Scale; GBH = grievous bodily harm.

Sex Offenders

Whether sex offenders’ victims had been adults or children, attending ETS was associated with a virtual halving of the expected general reconviction rate (Table 7). However, sex offenders with adult victims were initially of considerably higher risk of general offending than those with child victims (43% vs. 18%). Most sex offenders with child victims had general offending risk scores below 20 in contrast to a wider spread of risk among those who offended against adults. The smaller numbers of sex offenders in the higher risk bands should lead to caution in asserting patterns of change following ETS with these offenders. Moreover, these sex offenders had not attended any further offense-specific interventions while in custody and may not be typical of the risk and need profile of the custodial sex offender population as a whole.

Violent Offenders

A breakdown of the violence category into smaller sub-groups, presented alongside robbery for comparison, revealed some clear variation in patterns of change (Table 8). Those convicted of robbery (i.e., offenses of theft using force or threats of force) were separated in this study from other violent offenders to test for the variable pattern of change first reported in Robinson’s (1995) study. As found in that earlier research, there was no positive difference seen here between actual and predicted reoffending rates following ETS for those with a main current offense of robbery. For all other categories of violent offending, the positive difference between predicted and actual reconviction rates was at least 12 percentage points. Moreover, when robbers were included in the violence category, the overall reduction from predicted rates fell from 17 percentage points (as per Table 3) to just under 11.

Acquisitive Offenders

Table 8 indicates that there were some sub-groups of non-violent acquisitive offender with whom ETS may have had some impact. Offenders convicted of theft, fraud, or theft of vehicles showed greater drops in comparison with predicted reconviction rates than other acquisitive offenders. The magnitude of this fall indicates that cognitive skills training could be cost-effective as an intervention to reduce reoffending with these offense types. Nonetheless, for the most prevalent acquisitive offense (domestic burglary), there appeared to be no benefit and possibly even an iatrogenic effect on reconviction rates after attending ETS.

Other and Drugs Offenders

Table 8 also describes the reconviction outcomes for those offense types categorized as committing drugs-related or “other” offenses, the most frequent of which were motoring offenses. Those convicted of offenses around more organized, larger scale drug dealing (with notably lower levels of predicted reoffending than others in this category at 35%) did not show reductions in reconviction following ETS. In contrast, large reductions from the predicted rate were seen for those convicted of smaller scale supply or possession, and the range of other offense types included in this category.

Program Completion by Offense Type

Table 3 describes the ETS attrition rates for each offense type. A chi-square test of these completion rates indicated a significant association between offense type and program completion, with the worst attrition rates found for acquisitive and “other” offenders, χ²(5) = 151.82, p < .001. Nonetheless, a completion rate of 90% is not an indication of an attrition problem and compares very well with the completion rates observed for offender programs in general (McMurran & Theodosi, 2007). While there seemed to be some indication that engagement may be more of an issue with acquisitive offenders than others (assuming completion is an indicator of engagement), the high completion rate overall suggested this was not a sufficient explanation of the differential impact of ETS on reconviction rates for that group of offenders.

Relative Influence of Offense Type on Reconviction

A logistic regression was conducted to ascertain the relative influence of current offense type, risk, number of previous convictions for different offense types, ethnicity, age, sentence length, and program completion status on the binary 2-year reconviction outcome of ETS participants (Table 9). To control for falling rates of reconviction in England and Wales over this period (Ministry of Justice, 2008), year of release was also added to the model with 2005 as the reference category and, to reflect the findings in Table 8, the drugs category was further divided into import/export and possession/supply.

Table 9:

Within-Group Logistic Regression on 2-Year Post-Custody Reconviction Rate Among Male ETS Participants

	B	SE	Wald	df	Significance	Exp(B)	95% CI Exp(B)
	B	SE	Wald	df	Significance	Exp(B)	LL	UL
ETS non-completion	.309	.067	21.040	1	.000	1.362	1.194	1.554
OGRS2 score	.037	.006	41.880	1	.000	1.038	1.026	1.049
OGRS2 squared	.000	.000	0.002	1	.961	1.000	1.000	1.000
Offense—(Ref: Violence)			249.762	6	.000
Offense—Sex	−.058	.113	0.268	1	.605	0.943	0.756	1.177
Offense—Robbery	.692	.068	103.163	1	.000	1.997	1.747	2.282
Offense—Acquisitive	.528	.053	98.740	1	.000	1.695	1.527	1.881
Offense—Drugs import/export	.738	.080	85.062	1	.000	2.092	1.788	2.447
Offense—Drugs possession	−.029	.067	0.190	1	.663	0.971	0.852	1.107
Offense—Other	.007	.085	0.007	1	.935	1.007	0.853	1.189
Ethnicity (Ref: White)			15.184	4	.004
Ethnicity—Black	.103	.051	4.000	1	.046	1.108	1.002	1.225
Ethnicity—Asian	−.015	.077	0.039	1	.843	0.985	0.847	1.145
Ethnicity—Other	−.545	.303	3.226	1	.072	0.580	0.320	1.051
Ethnicity—Not known	−.402	.151	7.070	1	.008	0.669	0.498	0.900
Sentence length (months)	−.007	.001	113.803	1	.000	0.993	0.992	0.994
Age at release	−.035	.004	79.493	1	.000	0.966	0.959	0.973
Previous sex	.141	.051	7.662	1	.006	1.151	1.042	1.271
Previous violent	.052	.009	34.973	1	.000	1.053	1.035	1.071
Previous robbery	.183	.030	37.232	1	.000	1.200	1.132	1.273
Previous acquisitive	.055	.005	104.199	1	.000	1.056	1.045	1.067
Previous drugs	.033	.016	4.094	1	.043	1.034	1.001	1.067
Previous other	.074	.013	32.658	1	.000	1.076	1.049	1.104
Released in (Ref: 2005)			140.517	8	.000
Released in 1997	.403	.193	4.346	1	.037	1.496	1.024	2.185
Released in 1998	.558	.092	37.039	1	.000	1.747	1.460	2.091
Released in 1999	.353	.078	20.571	1	.000	1.423	1.222	1.657
Released in 2000	.432	.067	42.100	1	.000	1.540	1.351	1.754
Released in 2001	.337	.063	29.009	1	.000	1.401	1.239	1.584
Released in 2002	.436	.060	53.317	1	.000	1.546	1.375	1.738
Released in 2003	.181	.059	9.261	1	.002	1.198	1.066	1.346
Released in 2004	−.041	.058	0.494	1	.482	0.960	0.857	1.076
Constant	−2.039	.238	73.163	1	.000	0.130

Note. Model statistics: −2 log likelihood = 22,157.729; Nagelkerke R² = .360; Model χ²(29) = 6,518.489, p < .001. ETS = Enhanced Thinking Skills program; CI = confidence interval; LL = lower limit; UL = upper limit; OGRS = Offender Group Reconviction Scale.

Risk (as captured by OGRS2 score), failing to complete the whole ETS program, and the number of each type of previous conviction were each significantly and independently associated with higher reconviction rates among ETS participants. Being Black, or where ethnicity was not known, was associated with significantly higher reconviction rates than seen in the reference category of White offenders. Reconviction rates also varied significantly between years of release from custody, with releases in every year but 2004 being predictive of higher reconviction rates than those in the reference year of 2005. This reflected what we know about national reoffending rates falling over this same period. Being older at release and serving a longer sentence were significantly and independently predictive of lower reconviction rates. Even when controlling for these other influences on recidivism, the nature of the current offense still had a significant influence on reconviction. Those with robbery, acquisitive, and drugs import/export offenses had significantly higher reoffending rates than those in the violent offense reference category. Those in the sexual, possession/supply of drugs, and other offense categories were seen to follow a pattern similar to violent offenders.

A further regression was conducted to ascertain the singular influence of offense type on reconviction outcomes following ETS. Using a series of three forced entry steps, a regression was conducted introducing first risk alone (OGRS2 score), then current offense type, and finally all the remaining variables as in our original regression. This analysis allowed for an assessment of how each additional step improved the model. The −2 log likelihood (−2LL) for the risk only model was 23,541.85, Block χ²(1) = 5,134.37, p < .001; Nagelkerke R² = .293. With the addition of offense type, the −2LL fell to 22,711.294, Block χ²(6) = 830.55, p < .001; Nagelkerke R² = .334, and with all variables added to the model, the −2LL was 22,157.73, Block χ²(22) = 553.57, p < .001; Nagelkerke R² = .360. While the 2LL and Nagelkerke R² statistics indicate that there is much variance still unaccounted for in this model (not unexpected, as we know reoffending to be multiply determined by more variables than those captured in this study), it seemed that offense type brought significant added value to a prediction of reoffending based on risk alone, and that current offense type continued to predict differential reoffending among program participants when other influences were accounted for.

Discussion

In summary, the analyses in this study replicate Robinson’s (1995) findings that responsiveness to a cognitive skills program appears accounted for in part by the participant’s current offense. The actual reconviction rates for all offense types in the ETS participant group were significantly lower than the expected rates, except for those with current convictions for robbery, non-violent acquisitive offenses and larger scale drug dealing crimes. Responsivity across risk level was fairly consistent but for those in the lowest and highest risk bands, although this pattern varied by offense type. A regression to control for other influences on reconviction rates such as previous offending, age, and sentence length demonstrated that outcomes still varied significantly by current offense type. As there is no matched, untreated control, we cannot attribute definitively any observed reductions from predicted reoffending rates to attendance on ETS. Nonetheless, recent evaluations of the same program also in the custodial setting can reassure that this reduction is at least in part a response to the intervention.

Treatment Impact Across Offense Types

The first hypothesis tested by the study was that cognitive skills participants with a main current conviction for a sexual or violent offense will show greater reductions in reoffending than participants with an acquisitive offense. The findings support this hypothesis: Sex offenders, both those with adult and child victims, were reconvicted at lower rates than expected in nearly all but the highest risk bands and violent offenders were reconvicted at lower rates across all types of violence except robbery. When robbers are observed separately from other violent offenders, the difference between predicted and actual reconviction for violent offenders increases from 10 to 17 percentage points where the latter represents a relative reduction in reconviction rates of 29% (Table 3). However, acquisitive offenders who attended cognitive skills training were not reconvicted at lower rates than expected, providing further support to the earlier findings of Robinson (1995) in Canada. While there were some apparent benefits of ETS on reconviction for theft and fraud offenders, these together represent a relatively small proportion of the acquisitive group of whom over one-half are convicted of burglary.

The impact of cognitive skills training on sex offenders’ reconviction rates is worthy of additional comment. Traditionally, sexual offenders are thought to require considerable offense-specific programming, involving a greater dose of treatment than provided by ETS (McGrath, Cumming, Burchard, Zeoli, & Ellerby, 2010). However, the data here suggest that a much shorter and non-offense-specific program may see significantly lower general reconviction rates than predicted. There are two caveats to introduce. First, this study reports reoffending for any offense and information was not available on sexual crimes specifically. Second, it is unclear why the sexual offender sample in this study had not completed the longer, offense-specific SOTP. Policy during the period of this study dictated that sex offenders in prison should complete both ETS (when a thinking skills deficit was assessed) and SOTP. Thus, the sexual offenders in the current sample have followed an atypical treatment route, or at least one that appears contrary to the guidance of the time, and that should prompt some caution in generalizing from these findings to all sex offenders in custody. Inspection of the sample’s treatment location indicates that some of the sexual offenders studied were located in prisons where the SOTP was not available and hence possibly completed ETS as an alternative to SOTP. However, the majority of the sexual offenders in this study were located in an SOTP prison and a plausible hypothesis is that they completed ETS rather than SOTP because they denied their offense. SOTP excludes those who deny their sexual offending but ETS is able to accept such offenders because the program does not require any discussion of personal offending. Unfortunately, there are no records of which offenders were offered but refused a place on SOTP; therefore, this hypothesis is untestable on this data set.

The impact of denial on reconviction risk is uncertain, with studies showing mixed findings (Mann, Hanson, & Thornton, 2010) but some researchers (e.g., Nunes et al., 2007) have suggested that denial may act as a protective factor, reducing risk of reconviction rather than raising it as may be assumed. Another complication in trying to understand the impact of ETS on sexual offenders is that the risk predictor used in this study predicts general reoffending, and therefore reflects general criminality rather than specifically sexual deviance. It is possible those more sexually deviant offenders are not well represented in this sample or that they are scattered across the risk bands, preventing detailed conclusions about differential effects. Given these two limitations, it may be yet unwise to draw definitive conclusions about the viability of cognitive skills training as an alternative treatment strategy to more intensive SOTPs.

We do not know the risk of sexual recidivism for the sex offenders in this sample and that puts a real constraint on how far we can compare these findings with those from studies on the impact of other rehabilitative interventions with sex offenders. The observed OGRS2 scores suggest these offenders had perhaps lower predicted rates of general recidivism than is typical for this offender type (Barnett, Wakeling, & Howard, 2010). However, the sample for this study was extracted from a larger sample consisting of all those participating in offending behavior programs in custody over the years 2000 and 2005. In that group, participants on SOTP had an average OGRS2 score of 24% and those doing both ETS and SOTP had an average score of 26%. In that respect, then, the sex offenders in this study who undertook only ETS were comparable with those who participated in SOTP or in both programs in terms of their risk of general recidivism.

It is known that a number of sex offenders’ dynamic risk factors are common to other types of offender (Hanson & Morton-Bourgon, 2005), and there is evidence that more intense programs are only appropriate for higher risk sex offenders (Hanson, Bourgon, Helmus, & Hodgson, 2009; Harkins & Beech, 2007). Therefore, it seems that these findings, replicating Robinson (1995) as they do, should prompt further exploration of the efficacy of a general cognitive skills program to reduce reoffending with some sexual offenders, not least because cognitive skills training is a much shorter, and therefore financially cheaper, intervention.

Treatment Impact Across Risk Levels

The second hypothesis tested by the study was that higher predicted risk might account for a reduced treatment effect with acquisitive offenders. The Risk Principle (Andrews & Bonta, 2010) states that offending behavior programs work best when targeted at higher risk offenders, and that program dose should be proportionate to risk. As ETS is a moderate dose program, it was expected to be most effective with medium-risk offenders. However, the patterns of change observed across risk groups within offense types appear to be quite distinct (Table 5). For some groups, such as sex or violent offenders or those convicted for drugs or other offenses, the patterns of change are consistent with Andrews and Bonta’s risk principle; that the program appears to reap least benefit with those at the very lowest and highest risk of recidivism. In contrast, the benefit to acquisitive offenders of attending ETS appeared to fall away once offenders reached OGRS2 risk scores of 40% or so (note: the selection criteria for ETS cite an OGRS2 score of 40% as the minimum risk threshold for the program), although the majority of acquisitive offenders in this sample had OGRS2 scores of 60% or greater. Robbers were reconvicted at a rate 3 percentage points above the expected rate and this apparent detrimental effect was evident in all but one of the risk bands; hence, robbers appeared unresponsive to ETS regardless of their risk level. There may be some further challenge to the usual interpretation of the risk principle here in that a moderate intensity program such as ETS appears associated with substantial reductions in predicted reoffending for relatively high-risk offenders. There is also the consideration that relying on an actuarial risk tool as the counterfactual in this way assumes that the tool is equally reliable across offender characteristics such as offense type or sentence length. We were not able to test that assumption here and would encourage further work on this issue.

Acquisitive offenders had the highest average predicted risk scores of all the offense type groups at 71.80, but higher risk offenders in all the other categories (barring robbery) appeared to respond to ETS to a degree that acquisitive offenders did not. This pattern of findings suggests that risk level alone may not be a sufficient explanation for the absence of a treatment effect in the acquisitive group, and the clear interaction between offense type and risk level signals the influence of moderator variables that remain to be identified.

We know that higher risk levels are associated with a greater spread of criminogenic need (Andrews & Bonta, 2010), but this study has not been able to explore the nature or degree of the needs presented in this participant group. It will be important for future research to understand how the risk of participants is associated with specific criminogenic needs and explore whether those needs are successfully addressed in programs such as ETS to bring about reduced reoffending. Serin, Lloyd, Helmus, Derkzen, and Luong (2012) have reviewed the literature on the relationship between changes observed on dynamic risk factors across the course of rehabilitative programs and reoffending outcomes. They point to the many methodological shortcomings in this area of work but conclude that there are signs that measured changes on key constructs such as pro-criminal attitudes and antisocial personality can be signals of eventual desistance from crime. Serin et al. (2012) called for better measurement strategies of more sophisticated constructs on a timeline that additionally accesses the circumstances a prisoner experiences post-custody. Such an approach would allow for a better understanding of the mechanisms underlying positive change over the course of programs such as ETS with offenders presenting with different levels of risk and different constellations of criminogenic need.

Treatment Compliance

The third hypothesis tested by the study was that readiness to change, as evidenced through program completion rates, will be lowest among acquisitive offenders. This hypothesis was supported by the finding that acquisitive offenders had the highest attrition rate from ETS of all the offense types. However, as attrition was generally low across all groups, it would be wrong to conclude that acquisitive offenders were poorly motivated to address their offending. It may also be that attrition among acquisitive offenders was not caused by poor motivation to cease offending or by otherwise being unprepared to change, but by other factors such as a realization that the program was not meeting their most pressing needs. Attrition rates are furthermore a blunt measure of whether participants are fully engaging with an intervention and there is doubtless more to learn on how well different offenders respond to and engage with a program such as ETS. It might be expected that age and sentence length are also associated with a readiness to change; we know that acquisitive offenders in this study tended to be younger and serving shorter sentence lengths. However, when these additional variables were included in the regression, the effect of current offense on reconviction rates was seen to persist.

Interpretation

These results suggest that while the implementation of ETS as a whole is associated with a marked and significant decrease in reoffending compared with expected reconviction rates, there are some groups of offenders who appear not to have benefited. Replicating findings from Robinson’s (1995) evaluation of a cognitive skills intervention in Canada, ETS participants with current convictions for robbery or non-violent acquisitive offenses appear least responsive to the program. This disparity of impact for different offense types remained apparent even after the effects of age, sentence length, ethnicity, risk, year of release, and previous offending were accounted for in a logistic regression on reconviction. This in turn lends support to our fourth hypothesis that offense type appears to be an independent influence on program impact.

Nor can the relative weakness of a methodology that relies on predicted rates rather than a matched control sufficiently explain the differential impact by offense type. OGRS2 was validated on data from 1995, but national reoffending rates dropped consistently in the years 2000 to 2005 (Ministry of Justice, 2008) and therefore some drop from predicted to actual rates can be expected independent of any specific intervention as was seen in an earlier evaluation of the ETS program (Travers et al., 2013). Yet there is no obvious nor plausible reason why this general downward trend in reoffending would explain the differential outcomes of ETS by offense type. Nonetheless, a more robust test of this question would apply the predicted/actual methodology alongside a control group design.

It therefore seems most likely that acquisitive offenders (specifically those with current convictions for burglary and robbery), despite apparently having the thinking deficits that cognitive skills programs target (Debidin, 2009; Wilson et al., 2003), have other criminogenic needs that are stronger or more urgent than their thinking skills deficits. For example, it may be that the established link between problematic drug use and acquisitive offending (Bennett, Holloway, & Farrington, 2008) means that drug dependency is a greater driver of acquisitive offenders than poor self-management or problem-solving skills. In further support of this argument, the Drug Treatment Outcomes Research Study (Jones et al., 2009) described how all self-reported acquisitive offending dropped from 40% at baseline to 16% at 3-month follow-up after treatment for substance misuse. Self-reported acquisitive offending that was motivated specifically to fund substance misuse dropped from 22% to 7%. These observations signal an explicit link between levels of acquisitive offending and substance misuse, supporting Robinson’s hypothesis that substance misuse may be a more important driver of acquisitive offending than cognitive deficits.

Additionally or alternatively, it might be that of all types of offending, acquisitive offending is the most likely to be a group behavior. This might imply that antisocial peers are a more influential risk factor than individual level factors such as poor problem-solving. Another possibility, as noted in the beginning of this article, is that acquisitive offending is a rational choice rather than an impulsive behavior; hence reducing impulsivity, even when it is present, does not lower reconviction rates. If the majority of these acquisitive offenders have identified themselves as career criminals, making a rational choice about how to source the lifestyle they aspire to, then that might account for the apparent lack of impact for an approach that seeks to address impulsive behavior and poor problem-solving. Acquisitive offenders may experience these problems but it seems they are not the main drivers of their offending behavior. Crank and Brezina (2013) described a sub-group of offenders who take pride in their criminality and hold a distaste for a conventional lifestyle. Furthermore, these offenders do not find prison hard, seeing it instead as a badge of honor that enhances their image. If these attitudes are more prevalent among the burglars, robbers, and drug traffickers who participated in ETS, that might explain their apparent lack of responsivity to the cognitive skills approach.

It may be that a lack of responsiveness to programs such as ETS may follow from a longer history of contact with the criminal justice system with a consequent loss of hope that change is possible, or that effort will make a difference, and this would be a useful dimension to capture in future work on this topic. We know that the acquisitive offenders in this sample had the highest average risk scores (see Table 3), implying the longest and densest criminal histories of all the offense type groups. Previous experience of failure may interfere with the readiness to change necessary for an intervention such as ETS to have a positive impact on the offender’s choices and behavior. Age and sentence length also have an independent impact on reoffending outcomes among program participants, and how these interact with risk and offense type need to be better understood.

Further research on this topic needs to better profile these offense types who are unresponsive to cognitive skills programs for us to get a fuller understanding of their criminal histories, the possible motivations for their offending, and their risk and need characteristics. This further information would allow both for a better targeting of the cognitive skills approach and for the development of more responsive programming for those who currently appear not to benefit. If current offense continues to stand as a useful typology to apply in understanding responsivity, then this study indicates some refinements over the appropriate categories to apply. The term acquisitive is perhaps too broad when the outcomes for those convicted of theft, fraud, and handling are apparently so different to those for domestic burglars. Similarly, there appears to be a useful distinction to make between those convicted of larger scale drug supply crimes and those convicted of possession or supply. This study was constrained by having no data available to it on offenders’ dynamic risk factors. Further research on this topic will need to include that dimension if we are to develop our understanding of differential impact and specifically, whether current offense type is a useful dimension to consider in our targeting of interventions. A final caveat is that we need more research to explore possible differential performance of static risk tools for individuals with different offending types.

The congruence between this study and Robinson’s (1995) findings provides grounds for policy makers to exercise caution about providing cognitive skills programs to acquisitive offenders, including those convicted of robbery and drug import/export. When resources are finite, it seems defensible to shift the targeting of cognitive skills programs from acquisitive to more violent offenders. If ETS had not been delivered to those convicted of robbery or acquisitive offenses, the overall change from predicted reconviction rates would have risen from 8 to 14 percentage points. Figure 1 demonstrates how the impact on reconviction rates at each risk level would have been greater had the program not included individuals with convictions for robbery or other acquisitive crimes. As ETS showed an impact at all risk levels below OGRS2 80, even lower risk violent offenders could potentially benefit from cognitive skills programs. This should be borne in mind when identifying services for those who are low risk of reoffending but present with a raised risk of committing serious harm. For violent offenders with a very high likelihood of reoffending (equivalent to those with an OGRS2 score of 80 or above), cognitive skills training is not an effective nor a sufficient intervention.

Figure 1:

Reduced Reconviction Rates After ETS by Risk Level: With and Without Acquisitive Offenders

Further research into the impact of cognitive skills training with sexual offenders is urgently needed. In particular, research should examine differential responding to cognitive skills training for sex offenders who admit and deny their offending, and should segment sex offenders into risk bands using a sex offending predictor rather than a general offending predictor. Research that compares the effectiveness of offense-specific treatment for sexual offenders against cognitive skills training alone should be considered. This would best be explored through a randomized controlled trial comparing the two approaches.

Finally, it should be noted that this study focused on the response of adult male offenders to a cognitive skills intervention. Further work is needed to see whether these findings are replicated with women and young offenders who receive cognitive skills training. This study has focused on the custodial setting, while Robinson’s (1995) research included both community and custody settings. It would be useful to see a replication of the differential response to ETS observed here with a sample of offenders participating in current cognitive skills interventions in the community.

In conclusion, cognitive skills training is one of the best evidenced approaches worldwide in terms of its ability to reduce reoffending. However, the Responsivity Principle of offender rehabilitation (Andrews & Bonta, 2010) reminds us that not all offenders respond equally well to the same interventions. This study has provided further evidence that cognitive skills training seems to be of particular benefit with violent offenders but does not appear to be such a successful approach with serious acquisitive offenders. While the reasons for this differential impact have yet to be established, the fact that the current study with a very large sample size so precisely replicated an analysis conducted in a different country, with a different cognitive skills program, almost 20 years earlier, indicates that it may be time to revisit the targeting of cognitive skills programs. It seems likely that current offense type is a proxy for offender characteristics relevant to responsivity; further work is required to refine our understanding of those characteristics.

Footnotes

Acknowledgements

We would like to thank Ralph Serin, James McGuire, Philip Howard, and Robin Moore for helpful comments on earlier versions of this article.

Rosie Travers is a chartered and registered forensic psychologist working in the Commissioning Strategies Group evidence team for the National Offender Management Service. She is interested in the evaluation of rehabilitative interventions with offenders and the application of the risk, need, and responsivity principles to a national strategy for correctional rehabilitation.

Ruth E. Mann is a chartered and registered forensic psychologist. She leads the Commissioning Strategies Group evidence team for the National Offender Management Service which promotes evidence-informed strategies for offender rehabilitation and correctional practice.

Clive R. Hollin is professor of criminological psychology at the University of Leicester. After a career spanning both practice and academic inquiry concerned with offender rehabilitation, he is very much looking forward to his imminent retirement. He has contributed several key works to the What Works canon and has enjoyed collaborations with some of the preeminent scientist practitioners in the field. He has presented widely at conferences and has advised government departments in several countries on correctional theory and practice.

References

Andrews

D. A.

Bonta

(1995). The Level of Service Inventory–Revised. Toronto, Ontario, Canada: Multi-Health Systems.

Andrews

D. A.

Bonta

(2010). The psychology of criminal conduct (5th ed.). Newark, NJ: LexisNexis/Matthew Bender.

Andrews

D. A.

Bonta

Wormith

(1995). The Level of Service Inventory-Ontario Revision (LSI-OR): Interview and Scoring Guide. Toronto, Canada: Ontario Ministry of the Solicitor General and Correctional Services.

Barnett

G. D.

Wakeling

H. C.

Howard

P. D.

(2010). An examination of the predictive validity of the Risk Matrix 2000 in England and Wales. Sexual Abuse: A Journal of Research and Treatment, 22, 443-470.

Bennett

Holloway

Farrington

(2008). The statistical association between drug misuse and crime: A meta-analysis. Aggression and Violent Behavior, 13, 107-118.

Bourgon

Armstrong

(2005). Transferring the principles of effective treatment into a “real world” prison setting. Criminal Justice and Behavior, 32, 3-25.

Brezina

Topalli

(2012). Criminal self-efficacy: Exploring the correlates and consequences of a “successful criminal” identity. Criminal Justice and Behavior, 39, 1042-1062.

Cann

Falshaw

Nugent

Friendship

(2003). Understanding what works: Accredited cognitive skills programmes for adult men and young offenders (Research Findings No. 226). London, England: Home Office.

Clark

D. A.

(2000). Theory manual for Enhanced Thinking Skills (Prepared for the Joint Prison Probation Accreditation Panel). London, England: Home Office.

10.

Coid

Yang

Ullrich

Zhang

Roberts

. . . Farrington

(2007). Predicting and understanding risk of re-offending: The Prisoner Cohort Study (Research Summary 6). London, England: Ministry of Justice.

11.

Copas

Marshall

(1998). The Offender Group Reconviction Scale: The statistical reconviction score for use by probation officers. Journal of the Royal Statistical Society: Series C (Applied Statistics), 47, 159-171.

12.

Crank

B. R.

Brezina

(2013). “Prison will either make ya or break ya”: Punishment, deterrence and the criminal lifestyle. Deviant Behavior, 34, 10, 782-802.

13.

Debidin

(Ed.). (2009). A compendium of research and analysis on the Offender Assessment System (OASys) 2006-2009 (Ministry of Justice Research Series 16/09). London, England: Ministry of Justice.

14.

Falshaw

Friendship

Travers

Nugent

(2003). Searching for what works: An evaluation of cognitive skills programmes (Home Office Research Findings No. 206). London, England: Home Office.

15.

Friendship

Blud

Erikson

Travers

(2002). An evaluation of cognitive behavioural treatment for prisoners (Home Office Research Findings No. 161). London, England: Home Office.

16.

Hanson

R. K.

Bourgon

Helmus

Hodgson

(2009). The principles of effective correctional treatment also apply to sexual offenders: A meta-analysis. Criminal Justice and Behavior, 36, 865-892.

17.

Hanson

R. K.

Morton-Bourgon

K. E.

(2005). The characteristics of persistent sexual offenders: A meta-analysis of recidivism studies. Journal of Consulting and Clinical Psychology, 73, 1154-1163.

18.

Harkins

Beech

A. R.

(2007). A review of the factors that can influence the effectiveness of sexual offender treatment: Risk, need, responsivity, and process issues. Aggression and Violent Behavior, 12, 615-627.

19.

Harper

Chitty

(2005). The impact of corrections on re-offending: A review of “what works” (Home Office Research Study 291). London, England: Home Office.

20.

Harris

Riddy

(2010). Theory manual for the thinking skills programme (Prepared for the Correctional Services Accreditation Panel). London, England: Ministry of Justice.

21.

Hollin

C. R.

McGuire

Hounsome

J. C.

Hatcher

R. M.

Bilby

C. A. L.

Palmer

E. J.

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

22.

Hollin

C. R.

Palmer

E. J.

(2009). Cognitive skills programmes for offenders. Psychology, Crime & Law, 15, 147-164.

23.

Howard

P. D.

Barnett

G. D.

Mann

R. E.

(2014). Specialisation in and within sexual offending in England and Wales. Sexual Abuse: A Journal of Research and Treatment, 26, 225-251.

24.

Howard

P. D.

Clark

Garnham

(2006). An evaluation of the Offender Assessment System (OASys). London, England: Home Office.

25.

Howard

P. D.

Francis

Soothill

Humphreys

(2009). OGRS 3: The revised Offender Group Reconviction Scale (Research Summary 7/09). London, England: Ministry of Justice.

26.

Jones

Donmall

Millar

Moody

Weston

Anderson

. . . D’Souza

(2009). The Drug Treatment Outcomes Research Study (DTORS): Final outcomes report (Research Report 24). London, England: Home Office.

27.

Kraemer

H. C.

Morgan

G. A.

Leech

N. L.

Gliner

J. A.

Vaske

J. J.

Harmon

R. J.

(2003). Measures of clinical significance. Journal of the American Academy of Child & Adolescent Psychiatry, 42, 1524-1529.

28.

Landenberger

N. A.

Lipsey

M. W.

(2005). The positive effects of cognitive-behavioral programs for offenders: A meta-analysis of factors associated with effective treatment. Journal of Experimental Criminology, 1, 451-476.

29.

Latessa

Brusman-Lovins

Smith

(2010). Follow-up evaluation of Ohio’s community based correctional facility and halfway house programs—Outcome study. Cincinnati, OH: University of Cincinnati.

30.

Lloyd

Mair

Hough

(1994). Explaining Reconviction Rates: A critical analysis (Home Office Research Study 136). London, England: Home Office.

31.

Lovins

Lowenkamp

C. T.

Latessa

E. J.

(2009). Applying the risk principle to sex offenders: Can treatment make some sex offenders worse? The Prison Journal, 89, 344-357.

32.

Lowenkamp

C. T.

Latessa

E. J.

(2005). Increasing the effectiveness of correctional programming through the risk principle: Identifying offenders for residential placement. Criminology & Public Policy, 4, 262-290.

33.

MacKenzie

D. L.

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

34.

Maden

Rogers

Watt

Lewis

Amos

Gournay

Skapinakis

(2006). Assessing the utility of the Offenders Group Reconviction Scale-2 in predicting the risk of reconviction within 2 and 4 years of discharge from English and Welsh medium secure units. Retrieved from http://www.liv.ac.uk/fmhweb/Maden%20Lay%20sum-mary.doc

35.

Mann

R. E.

Hanson

R. K.

Thornton

(2010). Assessing risk for sexual recidivism: Some proposals on the nature of psychologically meaningful risk factors. Sexual Abuse: A Journal of Research and Treatment, 22, 191-217.

36.

McDougall

Perry

A. E.

Clarbour

Bowles

Worthy

(2009). Evaluation of HM Prison Service Enhanced Thinking Skills Programme: Report on the outcomes from a randomised controlled trial (Ministry of Justice Research Series 3/09). London, England: Ministry of Justice.

37.

McGrath

R. J.

Cumming

G. F.

Burchard

B. L.

Zeoli

Ellerby

(2010). Current practices and trends in sexual abuser management: The Safer Society 2009 North American Survey. Brandon, VT: Safer Society Press.

38.

McGuire

Bilby

C. A. L.

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.

39.

McMurran

Theodosi

(2007). Is treatment non-completion associated with increased reconviction over no treatment? Psychology, Crime & Law, 13, 333-343.

40.

Ministry of Justice. (2008). Re-offending of adults: New measures of re-offending 2000-2005 England and Wales. Retrieved from http://webarchive.nationalarchives.gov.uk/20100505212400/http://www.justice.gov.uk/publications/reoffendingofadults.htm.

41.

Ministry of Justice. (2010a). Compendium of reoffending statistics and analysis. Retrieved from https://www.gov.uk/government/publications/compendium-of-reoffending-statistics-and-analysis-2010.

42.

Ministry of Justice. (2010b). Offender management caseload statistics 2009. Retrieved from http://www.justice.gov.uk/downloads/publications/statistics-and-data/mojstats/omcs-2009-complete-210710a.pdf

43.

Nee

Meenaghan

(2006). Expert decision-making in burglars. British Journal of Criminology, 46, 935-949.

44.

Nunes

K. L.

Hanson

R. K.

Firestone

Moulden

H. M.

Greenberg

D. M.

Bradford

J. M.

(2007). Denial predicts recidivism for some sexual offenders. Sexual Abuse: A Journal of Research and Treatment, 19, 91-105.

45.

Palmer

E. J.

McGuire

Hatcher

R. M.

Hounsome

J. C.

Bilby

C. A. L.

Hollin

C. R.

(2008). The importance of appropriate allocation to offending behavior programs. International Journal of Offender Therapy and Comparative Criminology, 52, 206-221.

46.

Robinson

(1995). The impact of cognitive skills training on post-release recidivism among Canadian federal offenders (Report No. R-41). Ottawa, Ontario, Canada: Public Safety Canada.

47.

Ross

R. R.

Fabiano

E. A.

(1985). Time to think: A cognitive model of delinquency prevention and offender rehabilitation. Johnson City, TN: Institute of Social Sciences and Arts.

48.

Sadlier

(2010). Evaluation of the impact of the HM Prison Service Enhanced Thinking Skills programme on reoffending: Outcomes of the Surveying Prisoner Crime Reduction (SPCR) sample (Ministry of Justice Research Series 19/10). London, England: Ministry of Justice.

49.

Serin

R. C.

Lloyd

C. D.

Helmus

Derkzen

D. M.

Luong

(2012). Does intra-individual change predict offender recidivism? Searching for the Holy Grail in assessing offender change. Aggression and Violent Behavior, 18, 32-53.

50.

Sherman

L. W.

Gottfredson

D. C.

MacKenzie

D. L.

Eck

J. E.

Reuter

Bushway

S. D.

(1998). Preventing crime: What works, what doesn’t, what’s promising. Washington, DC: Department of Justice, National Institute of Justice.

51.

Sherman

L. W.

Strang

(2007). Restorative Justice: The evidence. London, England: The Smith Institute.

52.

Soothill

Fitzpatrick

Francis

(2009). Understanding criminal careers. Cullompton, UK: Willan.

53.

Sperber

K. G.

Latessa

E. J.

Makarios

M. D.

(2013). Examining the interaction between level of risk and dosage of treatment. Criminal Justice and Behavior, 40, 338-348.

54.

Taylor

(1999). Predicting reconvictions for sexual and violent offences using the revised Offender Group Reconviction Scale (Home Office Research Findings No. 104). London, England: Home Office.

55.

Tong

L. S. J.

Farrington

D. P.

(2006). How effective is the “reasoning and rehabilitation” programme in reducing re-offending? A meta-analysis of evaluations in four countries. Psychology, Crime & Law, 12, 3-24.

56.

Tong

L. S. J.

Farrington

D. P.

(2008). Effectiveness of “reasoning and rehabilitation” in reducing reoffending. Psicothema, 20, 20-28.

57.

Travers

Wakeling

H. C.

Mann

R. E.

Hollin

C. R.

(2013). Reconviction following a cognitive skills intervention: An alternative quasi-experimental methodology. Legal and Criminological Psychology, 18, 48-65.

58.

Willott

Griffin

(1999). Building your own lifeboat: Working-class male offenders talk about economic crime. British Journal of Social Psychology, 38, 445-460.

59.

Wilson

Attrill

Nugent

(2003). Effective interventions for acquisitive offenders: An investigation of cognitive skills programmes. Legal and Criminological Psychology, 8, 83-101.

60.

Zamble

Quinsey

V. L.

(1997). The criminal recidivism process. New York, NY: Cambridge University Press.