CBT and Informed Anger Management

A systematic review and meta-analysis on the effectiveness of CBT informed
anger management
Kevin sammut henwood, shihning Chou, Kevin D. Browne
PII: S1359-1789(15)00126-3
DOI: doi: 10.1016/j.avb.2015.09.011
Reference: AVB 960
To appear in: Aggression and Violent Behavior
Received date: 20 April 2015

Revised date: 10 August 2015
Accepted date: 23 September 2015
Please cite this article as: henwood, K., Chou & Browne, K.D., A systematic review and
meta-analysis on the effectiveness of CBT informed anger management, Aggression and
Violent Behavior (2015), doi: 10.1016/j.avb.2015.09.011
This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
ACCEPTED MANUSCRIPT
Running head: A SYSTEMATIC REVIEW AND META-ANALYSIS ON THE EFFECTIVENESS OF CBT
INFORMED ANGER MANAGEMENT
T
A systematic review and meta-analysis on the effectiveness of CBT informed anger
IP
management
R
SC
Authors:
Kevin Sammut Henwood, MSc, Trainee Forensic Psychologist
Shihning Chou, PhD, Assistant Professor of Forensic Psychology
NU
Kevin D. Browne, Professor of Forensic Psychology & Child Health
Affiliation:
Centre for Forensic and Family Psychology, Division of Psychiatry and Applied Psychology,
MA
School of Medicine, University of Nottingham
Address:
Floor B, Yang Fujia Building, University of Nottingham, Wollaton Road, Nottingham, NG8
D
1BB
TE
Corresponding author:
Shihning Chou
P
CE
Centre for Forensic and Family Psychology, Division of Psychiatry and Applied Psychology,
School of Medicine, University of Nottingham
Floor B, Yang Fujia Building, Wollaton Road, Nottingham, NG8 1BB
AC
Tel: +44 115 8466623; Email: [email protected]
ACKNOWLEDGEMENTS
We thank Kathleen Green, Assistant Professor of Forensic Psychology, for her advice on the
meta-analysis.
ACCEPTED MANUSCRIPT
2
A SYSTEMATIC REVIEW AND META-ANALYSIS ON THE EFFECTIVENESS OF CBT INFORMED ANGER
MANAGEMENT
Highlights
T
CBT based anger management achieved a risk reduction of 23% for general recidivism.
R IP
CBT based anger management achieved a risk reduction of 28% for violent recidivism.
SC
Treatment completion may result in a 42% risk reduction in general recidivism.
NU
Treatment completion may result in a 56% risk reduction in violent recidivism.
MA
Anger management may be effective in reducing risk, especially violent recidivism.
D
P TE
CE
AC
ACCEPTED MANUSCRIPT
3
MANAGEMENT
Abstract
T
This meta-analysis sought to investigate the effectiveness of CBT based anger management
IP
interventions on reducing recidivism among adult male offenders. Studies were selected after
R
a bibliographic database search, a hand-search of references from similar studies and an
SC
electronic search on apposite Correctional websites. The outcome measures of interest were
NU
general and violent recidivism rates. These were considered to be evidence of long term
behavioral change. Studies that included appropriate data were analysed using risk ratio
MA
analysis. The analysis of the effect of exposure to CBT based treatment on general recidivism
showed an overall effect of 0.77, indicating a risk reduction of 23%, whereas the overall
D
effect on violent recidivism was 0.72, indicating a risk reduction of 28%. The meta-analysis
TE
also explored the effects of treatment completion in comparison to attrition groups. The
effects of treatment completion on general recidivism through risk ratios was 0.58, indicating
P
CE
a 42% risk reduction. For violent recidivism, the risk ratio was 0.44, indicating a 56% risk
reduction. Subgroup analysis based on the treatment modality and the analysis of the risk of
AC
bias carried out on the selected studies was conducted to explore the significant heterogeniety
noted in the results. Overall, anger management appeared to be effective in reducing the risk
of recidivism, especially violent recidivism. Moderate-intensity anger management were
associated with larger effect than the high-intensity correctional programmes for violence
reduction.
Keywords: Anger management, Cognitive Behavioral Therapy, systematic review, meta-
analysis, violence, violent recidivism

ACCEPTED MANUSCRIPT
4
MANAGEMENT
Background
T
The aim of this study was to explore whether CBT based anger management reduces
IP
recidivism amongst offender populations. The link between anger and offending behavior is
R
not clearly understood. Anger does not always manifest itself in aggression, as it could help
SC
invigorate the person to take action against the object, person, or event causing the
NU
frustration. Also, not all aggressive acts necessitate anger. This instrumental form of
aggression might be present in a number of violent offenders, explaining why some violent
MA
offenders were found to have no pathological or problematic levels of anger when assessed
(Howells, 2004). Dysfunctional anger typically is more frequent, more intense, of longer
D
duration and comprised of more adverse action schemas (Novaco, 2011). A recent study in a
TE
forensic setting (Gilbert, Daffern, Talevski, & Ogloff, 2013) noted that individuals with high
trait anger were prone to activate aggressive behavioral scripts. Furthermore, they suggest
P
CE
that intensity, frequency and the duration of the anger problem were more salient than
normative beliefs about violence and aggressive script rehearsal in determining future
AC
violence. Thus, anger and associated feelings like rage can be considered as precipitators to
violent offending (Novaco, 2011) activating aggression related knowledge structures which
justify aggression, reduce inhibitions to violence, and disabling cognitive reappraisal while
also activating aggressive scripts. This might imply that anger dysfunction can be involved
in recidivism.
Group-based cognitive behavioral programs seem to be the most widely used intervention for
dysfunctional anger. Typically, such programs are brief and aim to increase the client’s
ability to control anger and limit arousal (Gilbert & Daffern, 2010). This is achieved by
replacing the dysfunctional cognitions, inferences, and evaluations with anger inhibiting ones
ACCEPTED MANUSCRIPT
5
MANAGEMENT
such as seeking alternative reality-based explanations for the antecedent events (Howells,
1998; Trower, Casey, & Dryden, 2008); addressing aggression related knowledge structures
T
IP
such as schemas and behavioral scripts (Gilbert & Daffern, 2010); imparting arousal
reduction techniques are aimed at reducing the client’s physiological state of readiness such
R
SC
as breathing and visualization (Novaco, 2011); and teaching behaviors that are functionally
equivalent to their dysfunctional behavior (Deffenbacher, 2011).
Previous systematic reviews NU

MA
A scoping search was carried out on five databases (CENTRAL, Campbell, Medline,
D
PsychInfo and SCOPUS) and the Ministry of Justice website in February 2014. Six
TE
systematic reviews were published between 1995 and 2009, seeking to determine the
effectiveness of psychological therapies and cognitive behavioral interventions specifically

P
on problematic anger; Tafrate (1995); Edmonson and Conger (1996); Beck and Fernandez,
CE
(1998); Di Guiseppe and Tafrate (2003); Del Vecchio and O’Leary (2004) and Saini (2009).
AC
They all reported mean treatment effects of between .70 and .76 on measures of anger.
However, most of these reviews did not include offender populations, as some focused
exclusively on college students or clinical samples.
A further four systematic reviews focused specifically on offenders. Dowden and Andrews
(2000) explored anger management and offending behavior and found significant positive
treatment effect for programs focusing on anger management and relapse prevention had
sizes in the offenders. Other programs such as those dealing only with antisocial attitudes
also showed positive effect but the results were not significant. This systematic review did
not examine risk of bias of the 35 included studies.

ACCEPTED MANUSCRIPT
6
MANAGEMENT
Lipsey et al. (2007) systematic review analyzed the effects of CBT on offenders with anger
management being one type of the interventions explored and found positive treatment
T
IP
effects. Moderator variables were also explored. For example, higher risk offenders fared
R
significantly better and good quality program implementation was associated with greater
SC
treatment effect. This study, however, provided little information on the risk of bias of the
included studies except to state that only 19 out of 58 studies included were randomized
NU
control trials (RCTs).
MA
A review from the UK Ministry of Justice Research Series focused on the effectiveness of
interventions in general for violent offenders and included studies that did not administer
D
psychological interventions such as electronic monitoring. Jolliffe and Farrington (2007)

TE
concluded that despite violent offenders being more difficult to engage in therapy and having
extensive offending histories, the overall results showed an 8-11% post-treatment reduction
P
CE
of general re-offending and a 7-8% reduction of violent re-offending. Jolliffe and
Farrington’s (2007) review also found evidence that the length of treatment was negatively
AC
correlated with re-offending rates. It should be noted that the effects were significantly
smaller in good quality studies. This finding could not be explored further due to
heterogeneity of their included studies. In fact, only one out of the 11 included studies was a
random control trial.
Ross, Quayle, Newman, and Tansey’s (2013) narrative review aimed at determining the
effectiveness of psychological therapies on violent behavior. The participants included for
analysis consisted of offenders but also included offenders with mental health issues. This
narrative review included 10 studies, consisting of randomized controlled studies, controlled
before and after studies and case series studies. Ross et al. (2013) concluded that most of the
ACCEPTED MANUSCRIPT
7
MANAGEMENT
studies showed a reduction of aggressive behavior amongst those who had received
psychological therapies. However, high levels of heterogeneity between their included studies
T
IP
may have confounded the overall conclusions of this narrative review. Jolliffe and Farrington
R
(2007) and Ross et al. (2013) were the only 2 systematic reviews analyzed that used a quality
SC
analysis of the included studies but neither focused exclusively on anger.
NU
Objectives
MA
The aim of this systematic review was to assess the effectiveness of CBT informed
interventions or anger management interventions on adult offenders sentenced to probation

D
orders or incarceration. Unlike previous reviews, it quantitatively explored the treatment

TE
effects on recidivism by conducting a meta-analysis based on a stringent review process,
including a systematic literature search and an assessment of risk of bias that explored the
P
common methodological flaws to help explain the findings.

CE
AC
Methods
Criteria for study inclusion
An Inclusion Criteria Checklist (available upon request) was used to assess the studies for
eligibility.
Type of Studies: RCT or NRCT with a matched control or a waitlist control, case control and
cohort studies. Single case studies and qualitative designs were excluded.
ACCEPTED MANUSCRIPT
8
MANAGEMENT
Participants: At least 50% of sample had to be comprised of adult male offenders (>18 years)
with a history of violence or screened for dysfunctional anger. Studies that focused
T
IP
exclusively on domestic violence abusers or offenders with mental health diagnosis were
R
excluded to limit heterogeneity.
SC
Interventions: All modalities of CBT based treatment for anger or violence receiving
treatment in prison or the community.
NU
Comparators: Randomized or non-randomized control group receiving alternate treatment to
MA
anger management, waitlist control, intent to treat, or comparison to attrition group.
Outcomes: Reconviction as a measure of general and/or violent re-offending.

D
TE
Search methods for identification of studies

P
CE
The following electronic databases were searched in June 2014; the Cochrane Library; the
Campbell Collaboration; Medline; PsychInfo; ASSIA; SCOPUS; and Web of Science. The
AC
same concatenation of key words was used in all searches, although minor modifications to
the search syntax were made for each particular database.
(anger OR aggression OR violen*) AND (offend* OR crimin* OR perpetrat*) AND (CBT
OR cognitive OR behavio* OR treatment OR “anger management”)
In addition, the reference lists of the previous 10 systematic reviews were hand searched.
Relevant Governmental portals were also searched, such as the Canadian Correctional
Service portal, the Australian Institute of Criminology portal, and the UK Ministry of Justice.
ACCEPTED MANUSCRIPT
9
MANAGEMENT
Data collection and analysis
T
IP
The references yielded in the database search were exported onto Endnote Online (N=3362).
After excluding duplicates (n=273), the titles of the remaining references were scanned to
R
SC
remove book chapters, reviews or meta-analyses, opinion papers, and clearly irrelevant
papers (n=2985). The abstracts of the 104 remaining references were compared to the
NU
inclusion/exclusion criteria and 43 did not meet the inclusion criteria. The full-text of the
remaining 61 articles was assessed using the Inclusion Criteria Checklist, resulting in the
MA
exclusion of 47 references, with 13 matching all the inclusion criteria. One article by
Marquis, Bourgon, Armstrong, and Pfaff (1996) contained a report on two studies using
D
different samples, baseline measures, and results. Thus, this report was considered to be two
TE
studies. The reasons for study exclusion can be made available from the authors. The sorting
P
is in Figure 1.
CE
AC
ACCEPTED MANUSCRIPT
10
MANAGEMENT
T
R IP
SC
NU
MA
D
P TE
CE
AC
Figure 1. The Selection Process.
Data extraction
A pre-defined Data Extraction Sheet was used to extract the information relevant to the
review question (available upon request). This was adapted from Cochrane's EPOC as a
guideline (Cochrane Public Health Group, 2011).

ACCEPTED MANUSCRIPT
11
MANAGEMENT
Assessment of risk of bias in included studies
T
IP
The included studies were either case control studies (n= 7) or controlled before and after
studies (n = 7).
R
SC
The checklist was adapted from the Risk of Bias Tool in the Cochrane Handbook for
Systematic Reviews of Interventions (Higgins, Atman, & Sterne, 2011). The checklist
NU
covered selection bias, performance bias, attrition bias, detection bias, and reporting bias.
MA
Where possible, an appraisal of the overestimation or underestimation of the reported
treatment effects of the study was attempted. A number of criteria were identified in each
D
category of bias whose presence would increase or decrease the risk of bias. The assessor
TE
could also rate an item unclear or inconclusive from the information provided in the article.
The included studies were also assessed using the Maryland Scientific Methods Scale
P
(MSMS). This MSMS has been widely used in criminological studies to assess the
CE
methodological quality of studies (Sherman, Gottfredson, Mackenzie, Reuter, & Bushway,

AC
1997).
Statistical Analysis
A meta-analysis was carried out to explore the reduction of risk of reconviction after
treatment exposure or completion. Not all the included studies contained the data needed for
meta-analyses. Although these studies were not included in the meta-analyses, their findings
were used to compliment the meta-analytic synthesis. Subgroup analysis based on treatment
modality was used to assess the effects associated with anger management, correctional
programs for violent reduction, and other CBT-based treatments. This would compare the
effect of anger management to more intensive correctional programs. Subgroup allocation

ACCEPTED MANUSCRIPT
12
MANAGEMENT
was based on the treatment components reported in the study and the intensity or dosage of
the respective programs.
T
IP
Risk Ratio Analysis (RR) was performed on the recidivism rates of treatment groups and
R
their respective controls to estimate the overall risk of general or violent recidivism. This was
SC
calculated through Review Manager 5.3 (RevMan) computer software (Review Manager,
2014). Analyses were carried out to determine:
NU
1. The overall effect of treatment on general recidivism and violent recidivism. Subgroup
MA
analyses compared the magnitude of effect and treatment modalities.
2. The overall effect of treatment completion on recidivism and violent recidivism in

D
comparison to attrition groups with the same subgroups of treatment modality.

TE
Any relative risk reductions of the analyses being conducted were calculated by computing
P
100% x (1-RR).
CE
The above analyses were performed using the inverse variance random effects model
AC
(DerSimonian & Laird, 1986). This would take into account the amount of heterogeneity
across the studies. Such a method would generate more conservative estimations of statistical
significance producing an average treatment effect (Deeks et al., 2008). A random effects
model was used on the assumption that any treatment effects in the studies would take the
form of a distribution of effect sizes with the mean effect size being the center point of this
distribution. Heterogeneity would then be the spread of the distribution (Deeks et al., 2008).
The Q statistic and the I² estimates were used to estimate the probability of heterogeneity in
the included studies. The I² indicates the amount of variance in a pooled effect size that could
ACCEPTED MANUSCRIPT
13
MANAGEMENT
be explained through heterogeneity (Higgins, Thompson, Deeks, & Altman, 2003). An I²
value between 0 and 40% indicates low variance, scores in the region of 30%-60% indicate
T
IP
moderate variance and those score from 50% to 90% indicate substantial variance (Deeks,
R
Higgins, & Altman, 2008). The magnitude, direction, and possible reasons for heterogeneity
SC
were explored in the results section.
NU
Results MA
Characteristics of included studies
The characteristics and results of the 14 reviewed studies are in Table 1. Each study was
D
assigned a study number and will be referred to by their study number in text or in superscript
TE
when mentioned in the results, discussion, and conclusion. An extended narrative synthesis
P
exploring the settings, samples, and interventions used can be made available upon request.
CE
In brief, this narrative synthesis of the included studies confirmed the main findings of the
meta-analysis.
AC
ACCEPTED MANUSCRIPT
14
A SYSTEMATIC REVIEW AND META-ANALYSIS ON THE EFFECTIVENESS OF CBT INFORMED ANGER MANAGEMENT
Table1 Relevant characteristics and findings of included studies.
T
IP
CR
Study No. Sample composition Intervention Type and brief Setting and Percentage of offenders involved in Percentage of offenders involved in Follow up
and Mean Ages (M) description duration of Recidivism Violent Recidivism period
Author, treatment
US
Country of
Publication and
N
Design
MA
1. Henning & Freuh N = 196 Cognitive Self Change Program – Prison Treatment – 50% Control – 70.8% Treatment – 17% Control – 24% 2 yrs
(1996) USA Adult male offenders components similar to anger
Before and After 60% violent offenders management – designed for Min. 6
D
Study months –
M age treatment = 32 yrs offenders with history of
3/5 times
TE
M age control = 29.6 yrs interpersonal violence
weekly
P
2. Hughes (1993) N= 79 Anger Management and Drama Prison Treatment – 56% Control – 68.8% Treatment – 40% Control – 65.8% 9mths – 4 yrs
Canada therapy – components of anger
Before and After
Study
Adult male offenders
screened for anger
CE
management combined with 24 hours
problems drama
AC
M age not reported
3. Motiuk, Smiley, & N = 120 Intensive program for violent Prison Treatment – 40% Control – 35% Treatment – 18% Control – 15% Average of 2
Blanchette (1996) Adult male violent offending – a cognitive yrs
Canada offenders – 95% of behavioral program with 8 months
Case Control Study 3 mths – 6
treatment group and 86% psychosocial components aimed
of control with violent to address criminogenic needs years
index offence
M age 35 yrs
4. Marquis, Bourgon, N1 = 216 Substance Abuse Treatment – Prison Relapse Matched control – N/A N/A Not Reported
Armstrong, & Pfaff Adult male offenders relapse prevention components Based prevention** - 59%
(1996) Violent and non-violent Treatment
Canada
ACCEPTED MANUSCRIPT
15
Case Control Study offenders included Anger Management Treatment – Facility at 51%
T
M age not reported programs containing anger least 20days
Matched control –
IP
management components but not 4/5 times Relapse prevention
enough details are provided. weekly anger management 59%
CR
– 34%
5. Marquis, Bourgon, N = 190 Substance Abuse Treatment – Prison Relapse Prevention Untreated Control N/A N/A Not reported
US
Armstrong, & Pfaff Adult male offenders relapse prevention components Based – 48% – 60%
(1996) screened for anger Treatment
Canada Anger Management Treatment – Anger
problems Facility
N
Before and After programs contained anger Management –
Study M age not reported min. of
management components but not 33%
MA
20days 4/5
enough details are provided. times
weekly Combined – 36%
D
6. Boe, Belcourt, N = 74 Violent Unit Program - intensive Community Treatment – 0%* Control – 15% N/A N/A 6 months – 2
Ishak, & Bsilis Adult male offenders – community supervision –
TE
Min. 6 years
(1997) Canada 96% had violent index components of the program could months with
Before and After
offence – high risk not be ascertained 2 sessions
P
Study
offenders (SIR-R1) weekly
M age not reported CE
7. Dowden, N = 220 Anger and Other Emotions Prison Treatment – 10% Control – 30% Treatment – 5% Control – 17% 3 years
AC
Blanchette, &Serin Adult male violent Management Program – anger 50 hours
(1999) Canada Case offenders management focused treatment
Control Study
M age 35.6 yrs.
8. Dowden & Serin N = 220 Anger and Other Emotions Prison Treatment -1 10% Attrition Group – Treatment – 5% Control – 17% 3 years
(2001) Canada Adult male violent Management Program – anger 50 hours 52%
Case Control Study offenders – management focused treatment Attrition Group –
M age 35.6 yrs. Control – 30% 40%
9. Cortoni, Nunes, & N = 966 Violence Prevention Program – Prison Treatment – 24.6% Attrition Group – Treatment – 8.5% Attrition Group – 5 years
Latendresse (2006) Adult male violent program contains components of 188 hours 37.7% 24.5% Average 1
Canada offenders anger management but also year
M age 30.3 yrs.
ACCEPTED MANUSCRIPT
16
Case Control Study All had min. of 2 violent focused on criminogenic needs. Control -41% Control – 21.8%
T
offences
IP
High Risk Offenders
(SIR-R1)
CR
10. Serin, Gobeil, & N = 256 Persistently Violent Offender Prison PVO Group -17% AEMP Group – PVO Group -8% AEMP Group – 5 yearsrs
Preston (2009) Adult male violent Program – cognitive behavioral 144 hours 21% 7%
US
Canada offenders program based on social AEMP PV Group AEMP PV Group Average of
Before and After M Age 31.75 yrs. information processing. – 11% Drop-out Group – – 8% Drop-out Group – 3.29 years
Study 38% 21%
N
Anger and Other Emotions
MA
Management Program - anger
management focused treatment
11.Hatcher,Palmer,M N = 197 Aggression Replacement Therapy Community Treatment – 39.2% Control – 50.9% N/A N/A 10 mths
D
acGuire, Hounsome, Adult male violent – program contained components Min. 10
TE
Belby, & Hollin offenders with established of anger management, behavioral week
(2008) UK pattern of aggression and and affective components and an
Before and After anger problems additional component on values.
P
Study M Age 27.42 yrs.
Medium – Medium/High
risk on OGRS-2
CE
AC
12. Berry (2003) N = 164 Montgomery House Violence Community Treatment – 40% Control – 62% Treatment – 26% Control – 44% 17 mths
New Zealand Adult violent offenders Prevention Program – cognitive Rehab
Case Control Study M Age 28yrs. behavioral program with violence 10 weeks
prevention components,
addictions and communication.
13. Polaschek, N = 104 Violence Prevention Unit Prison VPUP – 73% Control – 85% VPUP – 32% Control – 63% 3 years
Wilson, Townsend & Adults male violent Program– program contains 330 hours
Daly (2005) offenders components of anger management 28 weeks
New Zealand High risk offenders but also focused on criminogenic
Before and After Mean age treatment 23..5 needs e.g. victim empathy and
Study yrs offence supporting thinking.
ACCEPTED MANUSCRIPT
17
Mean age control 24.4 yrs
T
High risk control –
IP
14. Polaschek (2010) N = 224 Violence Prevention Unit Prison High Risk Treatment High High Risk Control 3.5 years
New Zealand Adult male violent Program– program contains 330 hours 95% Risk 62% - 72% Drop out
Treatment – 83%
CR
Case Control Study offenders components of anger management 28 weeks Control 75%
M age 28 yrs. but also focused on criminogenic Medium risk Drop-out 71%
Medium and High risk needs e.g. victim empathy and control – 67% Medium Risk
US
offenders offence supporting thinking. Treatment Control 48%
Medium Risk Drop-out control – Medium Risk –
Treatment - 76% 89% 33%
N
Drop-out 93%
MA
* Study examined failure rates with 10 out of 11 returns to custody were due to technical breaches of licence.
**Results presented for violent offenders only
D
P TE
CE
AC
ACCEPTED MANUSCRIPT
18
MANAGEMENT
General characteristics
T
6, 11 1, 4, 5, 12
Two studies focused on offenders in the community while four studies were
IP
conducted in secure rehabilitative centers. The remaining studies were carried out in prison
R
settings 2, 7-10, 13, 14. The number of participants varied from 74 to 892.
SC
The mean ages of all participants across studies ranged from 23.5 years to 35.6 years. There
NU
was considerable variation in participants’ offence history and other demographic variables.
Only three of the 14 studies 2, 4, 5 screened offenders for dysfunctional anger before treatment.
MA
Thirteen studies focused on violent offenders with the exception of Study 1 which included
40% of the participants with no violent history. In terms of risk, some diversity was noted.
D
Studies 9, 10, 12, 13 included high-risk offenders with limited capacity of making or
TE
maintaining progress. For example, offenders in Study 6 were responsible for over 1300
P
offences, with 222 being violent offences, 33 homicides, 103 robberies, and 81 assaults. This
CE
indicated that the seriousness of offending history amongst some offenders included in this
meta-analysis. Other studies 7, 8, 14 explored how risk levels might affect treatment outcomes,
AC
thus included offenders with different levels of risk.
The mean maximum follow-up period reported in the study was 3.21 years (SD 1.55).
However, the period of follow up for some offenders ended once they were returned to
custody or reconvicted and this varied considerably across studies.
Intervention types
The included studies adopted a mixture of intervention modalities that might be broadly
CBT-based. All the interventions in the studies were manualized and had provided training
for the program facilitators. However, the level of reported integrity and fidelity varied across
ACCEPTED MANUSCRIPT
19
MANAGEMENT
the studies. Study 14 distinguished between low intensity treatment, typically of a shorter
duration and frequency, such as traditional anger management and more intensive
T
IP
correctional programs such as the violence reduction programs typically run in correctional
R
institutions. Although the level of intensity varied in these violence reduction programs, the
SC
mean length of the program was of 190 hours.
Five studies had specifically focused on anger management. Studies 7, 8, 10 used the Anger
NU
and Other Emotions Program (AEMP). This program comprised of typical anger
MA
management components such as developing self-management, increasing problem solving,
improving communication, challenging dysfunctional thinking, and relapse prevention. It

D
consisted of approximately 50 hours of therapeutic engagement. Study 2 also explored anger

TE
management but was supplemented with psychodrama. This intervention was shorter, with 24
hours of therapy delivered to participants. Studies 4 and 5 were described in the same report.
P
Both studies compared recidivism rates of offenders who received a combined intervention of
CE
relapse prevention and anger management to those offenders receiving relapse prevention
AC
only or anger management only. Participants in Study 4 comprised of violent and non-violent
offenders, whereas Study 5 included exclusively offenders screened for anger dysregulation.
Study 11 focused on Aggression Replacement Therapy (ART). This program typically
imparts new skills to address the behavioral, affective, and moral components. It aims to
deliver arousal and anger control training and moral reasoning modules. Study 1 examined
the effects of a Cognitive Self Change Program which attempted to address interpersonal
aggression through challenging cognitive distortions and offending thought patterns.
Five studies 9-10, 12-14 explored the effectiveness of intensive violence reduction programs such
as the Violence Prevention Program 9; Persistently Violent Offender program 10, Montgomery
ACCEPTED MANUSCRIPT
20
MANAGEMENT
House Violence Prevention Program 12; and the Violence Prevention Unit Program 13-14
. The
content of the programs were typical of anger management interventions such as arousal
T
IP
reduction, communication skills, relationships, addressing cognitive distortions, and problem
R
solving. They would also include other elements such as victim empathy and risk
SC
management deemed essential in correctional programs. The length of treatment ranged from
144 hours to 330 hours, indicating high intensity.
Outcome Measures
NU
MA
The outcome measure of interest was general and violent recidivism. The percentage rates of
the reported outcomes in the individual studies are summarized in Table 1.

D
TE
Excluded studies
P
When full text studies were examined, 22 out of 61 studies were excluded as they focused on
CE
psychometric or observational measures only. Although these might yield valuable
information on the efficacy of treatment, the focus of the review was to explore the long-term
AC
effectiveness in terms of recidivism of anger management treatment. Another 15 studies were
excluded for not meeting the population criteria due to sampling offenders with no anger
problems or history of violence or offenders with acute personality disorders or adolescents.
Ten studies were excluded as they were multiple case studies, narrative reviews, or opinion
papers.
ACCEPTED MANUSCRIPT
21
MANAGEMENT
Risk of bias in included studies
T
IP
Previous research on offender populations highlight that it is inherently difficult to conduct
RCTs with offender populations and in settings such as prisons (Jollifee & Farrington, 2009;
R
SC
Ross et al., 2013). This limits the potential number of RCTs included in the review. Focusing
on recidivism analysis also makes random allocation of participants difficult, reducing the
NU
probability of finding RCTs that answered the review question. Table 2 summarizes the
levels of risk of bias.

MA
D
P TE
CE
AC
ACCEPTED MANUSCRIPT
22
MANAGEMENT
Table 2. Risk of Bias in Included Studies
T
Authors; Publication Maryland Selection Performance Attrition Detection Reporting
IP
Date and Study No Scientific Bias Bias Bias Bias Bias
Methods
R
Scale level
SC
Henning & Frueh Level 3 High Low High High Low
(1996)
Study 1
NU
Hughes (1993) Level 3 High High High Low High
Study 2
MA
Motiuk, et al. (1996) Level 4 High Unclear High Unclear High
Study 3
Marquis et al. (1996) Level 3 Unclear Unclear High Unclear High

D
Study 4 and 5
TE
Boe et al. (1997) Level 3 High Unclear Unclear Low High

Study 6
Dowden et al. (1999) Level 4 High Low High Low Low

P
Study 7
CE
Dowden & Serin Level 4 High Low Low Low Low

(2001) Study 8
AC
Cortoni et al. (2006) Level 4 Low Unclear Low Low Low

Study 9
Serin et al. (2009) Level 3 High Low Low Low Low

Study 10
Hatcher et al. (2008) Level 4 Low Unclear Low Low Low

Study 11
Berry (2003) Level 4 Low Low Low Low Low

Study 12
Polaschek et al. (2005) Level 3 High Low Low Low Low

Study 13
Polaschek (2010) Level 4 Low Unclear Low Low Low

Study 14
ACCEPTED MANUSCRIPT
23
MANAGEMENT
Selection bias
The analysis of selection bias focused on whether the study adequately controlled for baseline
T
IP
differences between the control and treatment groups. NRCTs are particularly prone to this
R
bias potentially introducing substantial amount of error in the review (Reeves et al., 2008).
SC
Eight (57.1%) of the included studies were deemed to be high risk, with a further study being
classed as unclear or undetermined. All the studies classed as Level 3 of the Maryland
NU
Scientific Methods Scale (MSMS) were deemed to have a high risk of selection bias. This
judgment was based mainly on the basis that the studies used an unmatched control group
MA
and in most instances did not control for known variables associated with recidivism. Two
studies 6, 13 in particular used comparison groups that were used as a matched control group in
D
another study. Although some of these reported that analysis revealed no pre-treatment
TE
significant differences on some important confounding variables, the analysis was not
P
exhaustive, and other differences between the treatment and control groups could have
CE
introduced bias, for example, differences in offence histories 1,10 or ethnic composition 13.
AC
3, 7, 8
Level 4 studies also displayed high risks of selection bias . This was primarily because
the matching process resulted in the exclusion of a large number of participants that had
received treatment. This could have potentially affected the treatment outcomes.
Furthermore, significant differences were observed between groups even after the matching
process in terms of offence type (e.g. Study 3) and ethnic composition (e.g. Studies 7, 8).
The studies deemed as low risk for selection bias were those that conducted case-by-case
matching on confounding variables such as risk of reconviction, previous offence history, and
salient demographic variables. Also, they reported no significant differences before

ACCEPTED MANUSCRIPT
24
MANAGEMENT
treatment. These studies included almost all of the participants that had received the treatment
intervention and had no candidates excluded due to the matching process 9, 11, 12, 14.
T
IP
Through matching participants on common confounding variables such as age, variables
R
associated with recidivism and level of risk or analyzing pre-treatment differences meant that
SC
some extraneous variables were controlled for. However, even when no significant
NU
differences exist at baseline, there is still a substantial amount of residual confounders such as
misclassification of offenders, which could have affected the matching process and
MA
consequently, the results (Reeves et al., 2008).
The lack of information in the report of Studies 4 and 5 rendered the assessment of these
D
4
studies difficult. One report contained two studies; a case control study and a before-and-
TE
after study 5. Selection bias could not be determined as detailed descriptions of participants
P
were omitted, resulting in a rating of unclear despite reporting that analysis revealed no
CE
significant pre-intervention differences on salient confounding variables.

AC
Direction of bias was difficult to ascertain adequately. For example, Study 2 used non-
completers as a control group and conducted no analysis before or after interventions to
assess possible differences that could have affected the outcome. In this case, the results
could have favored the treatment group since treatment non-completers could have failed to
complete treatment due to more complex needs and entrenched psychosocial difficulties.
However, in Studies 1 and 3, the potential bias might have favored the control group, as more
high risk and violent offenders were included in the treatment group than in the respective
controls.
ACCEPTED MANUSCRIPT
25
MANAGEMENT
Performance bias
T
IP
Performance bias tends to occur due to inconsistency in treatment delivery. Elements
assessed included whether interventions were delivered in different locations, times or by
R
SC
different facilitators. It also explored the reported integrity and fidelity of the program
2
through the use of supervision and training of facilitators. Only one study was deemed to
NU
have a high risk of performance bias. Although the study reported that the treatment was
delivered by professional and experienced staff, the treatment group was deemed to have
MA
completed treatment after 6 of the planned 12 sessions. This could have undermined the
3-6, 9, 11, 14
integrity of the treatment. Seven studies were rated as unclear, as they did not
D
provide enough information on the integrity and fidelity of the program.

TE
Those studies that were classed as low risk of performance bias had reported adequate
P
measures of controlling for potential biases. For example, Studies 1, 7, 8, 10, 12, 13 reported
CE
that the treatment was administered under supervision, by trained personnel and using a
AC
manual. Only one study8, however, claimed to have controlled for the effect of parallel
interventions on the subjects included in the study.
Measurement Bias
Since the focus was on failure or recidivism rates, it can be argued that the evaluators (the
criminal justice system) act blind to whether the offender received treatment or not. However,
other sources of bias could have affected the measurement of outcomes such as different
follow-up periods or different legal jurisdictions. The latter could have introduced bias due to
potentially different rates of prosecution and conviction.

ACCEPTED MANUSCRIPT
26
MANAGEMENT
Most of the studies (n= 10) included were rated as being of low risk for measurement bias.
These studies were rated as low risk as they used multiple information sources to arrive at the
T
IP
recidivism rate of participants in the study by exploring national correctional and police data.
One study 1 was rated as high risk of measurement bias. This study was reported to have used
R
SC
only one state database, which would not reveal re-offending in different states. The
3, 4, 5
remaining studies were rated as unclear, as they did not provide enough information
NU
pertaining to the length of follow-up period or the provenance of the information concerning
conviction rates.
MA
Attrition bias
D
TE
Considering that recidivism rates can only be observed after a period of time, attrition bias
could have also significantly affected the studies. The principle method by which attrition
P
bias was considered was assessing the type of analysis the researchers conducted on the non-
CE
11 12
completer group. Attrition levels in the selected studies varied from 22% to 71.7% .
AC
Interestingly, the study that reported the most attrition was the only study conducted on
13
probationers. One study reported no attrition as the participants were selected after the
completion of the program but this procedure could also have introduced selection bias.
High risk of attrition bias was noted in six included studies1-5, 7. In Studies 3, 4, 5 and 7, non-
completer groups were removed from the analysis and subjected to no further analysis. Study
2 reported that as much as 50% of the treatment group did not complete the program. The
attrition group in this study was then compared to the treatment completers. No analysis of
group differences was conducted before comparing their outcomes. Furthermore, setting
ACCEPTED MANUSCRIPT
27
MANAGEMENT
program completion at half of the planned interventions could have been indicative of greater
attrition figures which were not reported in the study.
T
IP
8-14
The other seven studies were deemed to be of low risk of attrition bias, as these studies
R
reported the figure of attrition and analyzed the reasons for non-completion. Studies 8-10, 12
SC
and 14 explored characteristics of the attrition groups and in some cases compared them to
NU
the treatment completer groups. High attrition rates are a common problem with community-
based studies (McMurran & Theodosi, 2007). Study 11 was in fact conducted in the
MA
community and sustained serious attrition rates of over 70%. Nevertheless, this study was
able to compare the attrition group to the treatment group on the outcome measures and other
D
salient confounding variables associated with recidivism. Study 6 could not be rated due to
TE
insufficient information in the study report.

P
Reporting bias
CE
The quality of reporting in NRCTs is often very poor and difficult to assess (Reeves et al.,
AC
2008). Poor quality of reporting could hinder the effective analysis of quality and inhibit the
accurate determination of biases.
Considering that NRCTs would in most instances not have an a priori protocol to analyze the
risk of reporting bias, the assessment focused on the overall quality and depth of reporting in
2-6
the study. Five studies were deemed to be of high risk. This bias affected the rating of
other quality criteria, as not enough information could be extracted to effectively assess the
risk of bias. The other studies, however, were all classed as having satisfactory levels of
reporting where methodological and other limitations of the studies were also reported. These
ACCEPTED MANUSCRIPT
28
MANAGEMENT
studies explored clinically relevant features of their study even when the results were not
significant.
T
IP
However, it must be noted that although Study 7 was deemed satisfactory, there was not
R
enough information reported on the attrition or non-completer group in the study. Also, Study
SC
14 contained what could be interpreted as contradictory information between the fidelity and
NU
integrity of the program reported in the method section compared to analysis provided in the
discussion.
MA
Overall, only one study12 was rated as low risk in all five areas of bias and was also classed as
level 4 on the MSMS. Six studies8-11, 13-14 had only one area rated as high risk or unclear risk
D
of bias. Only two of these studies10, 13 were classed as level 3 on the MSMS. Study 7 was
TE
classed as level 4 but had been rated as high risk for selection and attrition bias. Study 4
P
another level 4 study had three areas rated as high risk for selection, attrition and reporting
CE
bias. Other areas of bias in Study 4 were rated unclear. Most of the studies1, 2, 4, 5 classed as
level 3 on the MSMS had been rated as having 3 or more areas which were considered as
AC
high risk of bias.
Effects of interventions
Exposure to Treatment
The first set of RRs was aimed to analyze the effect of treatment on general (Fig. 1) and
violent recidivism (Fig. 2). Studies that included appropriate data, a treatment and a matched
2, 4-6, 10
or intent-to-treat control were used to run the RR. Some studies did not report
appropriate data for a meta-analysis. Studies 2 and 10 used attritions as controls. Studies 4
ACCEPTED MANUSCRIPT
29
MANAGEMENT
and 5 did not report the number of participants in the experimental and treatment groups.
Study 6 reported failure rates or returns to custody such as withdrawal of licenses which
T
IP
might not indicate actual re-offending. Thus, these studies were not entered into this analysis.
R
Experimental Control Risk Ratio Risk Ratio
Study or Subgroup Events Total Events Total Weight IV, Random, 95% CI IV, Random, 95% CI
SC
1.1.1 Anger Management
Dowden et al (1999) 11 110 33 110 7.9% 0.33 [0.18, 0.63]
Subtotal (95% CI) 110 110 7.9% 0.33 [0.18, 0.63]
Total events 11 33
Heterogeneity: Not applicable
Test for overall effect: Z = 3.42 (P = 0.0006)
NU
1.1.2 Correctional Programmes
Cortoni et al (2006) 89 446 109 446 17.7% 0.82 [0.64, 1.05]
Motiuk et al (1995) 24 60 21 60 11.3% 1.14 [0.72, 1.82]
Polaschek (2010) 92 110 94 110 21.5% 0.98 [0.87, 1.10]
MA
Polaschek et al (2005) 16 22 51 60 16.7% 0.86 [0.65, 1.13]
Subtotal (95% CI) 638 676 67.1% 0.94 [0.86, 1.04]
Total events 221 275
Heterogeneity: Tau² = 0.00; Chi² = 2.85, df = 3 (P = 0.42); I² = 0%
1.1.3 Other CBT based

D
Hatcher et al (2008) 20 53 27 53 12.0% 0.74 [0.48, 1.14]

Henning & Freuh (1996) 18 55 89 141 13.0% 0.52 [0.35, 0.77]
Subtotal (95% CI) 108 194 25.0% 0.61 [0.43, 0.87]
TE
Total events 38 116

Total (95% CI) 856 980 100.0% 0.77 [0.62, 0.96]

P

0.01 0.1 1 10 100
Test for overall effect: Z = 2.32 (P = 0.02) Favours [experimental] Favours [control]
CE
Test for subgroup differences: Chi² = 15.23, df = 2 (P = 0.0005), I² = 86.9%
Fig. 2 Exposure to treatment and general recidivism

AC
Figure 1 presents the results using random effect of exposure to treatment on general
recidivism. The treatment groups consisted of all cases exposed to treatment and thus
included the attrition groups. The overall risk ratio for general reconviction of offenders
exposed to treatment against their respective control groups was 0.77 (p= .02), indicating a
significant risk reduction of 23% for general recidivism among offenders after treatment.
However, significant heterogeneity was observed through the Q statistic (p= .002). The I²
value was 72%, also indicating significant heterogeneity across the study results.
Interestingly, significant differences were noted between the different subgroups included in
the RR. This may indicate significant cumulative differences in the results calculated with the
ACCEPTED MANUSCRIPT
30
MANAGEMENT
three subgroups. Although this could indicate a real difference between the levels of
treatment efficacy across studies, it could also be a result of the clinical and methodological
T
IP
diversity of the included studies.
R
SC
Dowden et al (1999) 6 110 19 110 7.1% 0.32 [0.13, 0.76]
Subtotal (95% CI) 110 110 7.1% 0.32 [0.13, 0.76]
Total events 6 19
NU

Berry (2003) 24 79 33 79 18.5% 0.73 [0.48, 1.11]
MA
Cortoni et al (2006) 43 446 58 446 20.8% 0.74 [0.51, 1.08]
Motiuk et al (1995) 11 60 9 60 8.2% 1.22 [0.55, 2.73]
Polaschek (2010) 66 110 72 110 29.5% 0.92 [0.75, 1.12]
Polaschek et al (2005) 7 22 38 60 11.4% 0.50 [0.26, 0.95]
Subtotal (95% CI) 717 755 88.5% 0.81 [0.67, 0.99]
D

TE

Henning & Freuh (1996) 3 55 21 141 4.4% 0.37 [0.11, 1.18]
Subtotal (95% CI) 55 141 4.4% 0.37 [0.11, 1.18]
P
Total events 3 21
CE
Total (95% CI) 882 1006 100.0% 0.72 [0.55, 0.93]

0.01 0.1 1 10 100
AC
Test for subgroup differences: Chi² = 5.81, df = 2 (P = 0.05), I² = 65.6%
Fig. 3 Exposure to treatment and violent recidivism
Figure 2 presents the RR exploring the effects of exposure to treatment and the risk for
violent recidivism only. Six studies divided into subgroups based on the treatment modality
contain the appropriate data for this analysis. The results indicate a 28% significant risk
reduction of reconviction for violence (p=.01) if the offender receives treatment of any
modality. The Q statistic was not significant (p = .08). The I² value of 46% on the other hand
indicated moderate variance amongst the study results. Due to the limited number of studies
included in the meta-analysis, the Q statistic could be deemed to be have low power in
ACCEPTED MANUSCRIPT
31
MANAGEMENT
detecting heterogeneity. However, since the I² value was not 0%, heterogeneity was assumed
to be present (Heudo-Medina, Sanchez-Meca, Marin, & Botella, 2006).
T
IP
Results in Figures 1 and 2 may give the impression of a qualitative interaction, with Study 3
R
seemingly showing results in the opposite direction of the other included studies. However,
SC
this difference in recidivism between treated and untreated offenders was not statistically
significant. Nevertheless, the treatment group in Study 3 included offenders considered of
NU
higher risk than their controls, which might skew the results. Furthermore, the high attrition
MA
rates in this study could have also affected the result. A quantitative interaction was also
found with the anger management subgroups seemingly associated with larger effect sizes.
D
This shall be explored in the subgroup analysis.

TE
In these RR analyses, only Study 7 focused on the effects of traditional anger management,
P
with the other studies focusing on other treatment programs which included only some
CE
components of anger management. The largest effect of risk reduction of treatment on
recidivism and violent offending amongst the subgroups was found in Study 7. Results from
AC
2, 4, 5, 10
other studies on traditional anger management in this systematic review but not
suitable to be entered into the RR analysis support the result of this analysis, with anger
management being associated with a significantly lower rate of re-offending. For example,
Studies 4 and 5 showed that anger management seemed a key element for the treatment of
violent offenders and those with anger dysregulation. Study 4 reported that violent offenders
receiving combined relapse prevention and anger management fared better than their waitlist
control and those receiving only relapse prevention. Study 5 concentrated on offenders with
anger difficulties only. It also found significantly lower rates of re-offending among
offenders receiving the anger management interventions when compared to their controls.
ACCEPTED MANUSCRIPT
32
MANAGEMENT
The subgroup analysis for general and violent reconviction also explored the reduction of risk
3, 9, 13, 14
associated with exposure to correctional violence reduction programs. Four studies
T
IP
were deemed to contain appropriate data for the analysis of general reconviction and five
3, 9, 12-14
R
studies for violent reconviction. The Correctional Program subgroup showed the
SC
smallest effect.
Both tests for heterogeneity between the studies in the Correctional Program subgroups of the
NU
two RR analyses were not significant for general reconviction (p=.42) and violent
MA
reconviction (p=.30). The I² values of 0% and 18% respectively. This may indicate the
studies in these subgroups were not as diverse in clinical terms.

D
The subgroups exploring other CBT-based treatments on reduction of general recidivism

TE
included two studies 1, 11 and one study 1 for violent recidivism. This subgroup also seemed to
P
produce larger effects than the correctional programs.

CE
The Q statistic was not significant for the general reconviction analysis. The I² value was
29%, indicating a moderate level of heterogeneity. This might be explained by the differences
AC
1, 11
inherent in the samples of the two studies . Study 1 only included approximately 60%
incarcerated violent offenders in the sample, whereas study 11 consisted of medium risk
offenders in the community who were court mandated to attend treatment. A salient
1
difference between the studies was the treatment modality, with one study using the
Cognitive Self Change program and the other 11 using Aggression Replacement Therapy.
ACCEPTED MANUSCRIPT
33
MANAGEMENT
Treatment completion
Experimental Control Risk Ratio
T
Risk Ratio
IP
Dowden & Serin (2001) 11 110 41 78 15.4% 0.19 [0.10, 0.35]
Hughes (1993) 24 42 13 19 18.9% 0.84 [0.56, 1.25]
Serin et al (2009) 23 112 10 33 14.8% 0.68 [0.36, 1.28]
R
Subtotal (95% CI) 264 130 49.2% 0.48 [0.20, 1.18]
Total events 58 64
SC

Cortoni et al (2006) 49 300 40 146 19.5% 0.60 [0.41, 0.86]
NU
Polaschek (2010) 62 77 30 33 22.5% 0.89 [0.76, 1.03]
Subtotal (95% CI) 377 179 42.0% 0.75 [0.51, 1.10]
Total events 111 70
MA
Hatcher et al (2008) 3 15 17 38 8.9% 0.45 [0.15, 1.31]
Subtotal (95% CI) 15 38 8.9% 0.45 [0.15, 1.31]
Total events 3 17
D
Total (95% CI) 656 347 100.0% 0.58 [0.39, 0.87]

TE

Heterogeneity: Tau² = 0.18; Chi² = 27.26, df = 5 (P < 0.0001); I² = 82%
0.01 0.1 1 10 100
Test for subgroup differences: Chi² = 1.43, df = 2 (P = 0.49), I² = 0%
P
Fig. 4 Treatment Completion and Recidivism

CE
Figure 3 presents an RR analysis using random effects on the risk reduction of recidivism of
AC
treatment completers and non-completers. Subgroup analysis was also calculated based on
treatment modalities, with the studies divided into three subgroups; three studies2, 8, 10 in the
anger management subgroup, two studies9, 14

in the correctional programs for violence
reduction and one11 in the other CBT programs subgroup.
The RR analysis on recidivism revealed a significant overall effect of 0.58 (p = .008). This
represents a 42% reduced risk of reconviction if treatment is completed. The results also
indicate a significant Q statistic (p=.001) and an I² value of 82%, showing considerable
variance in the study results.

ACCEPTED MANUSCRIPT
34
MANAGEMENT
Similarly, Figure 4 shows the RR analysis on violent reconviction for completers and non-
completers. However, only three studies2, 8, 10

in the anger management group and three
T
IP
studies9, 12, 14 in the correctional programs group were deemed to have the appropriate data for
R
this subgroup analysis.
SC
Dowden & Serin (2001) 6 110 31 78 13.9% 0.14 [0.06, 0.31]
NU
Hughes (1993) 17 42 13 19 18.7% 0.59 [0.37, 0.95]
Serin et al (2009) 9 112 7 33 12.8% 0.38 [0.15, 0.94]
Subtotal (95% CI) 264 130 45.4% 0.33 [0.13, 0.80]
Total events 32 51
MA

Berry (2003) 16 62 8 17 16.2% 0.55 [0.28, 1.06]
Cortoni et al (2006) 17 300 26 146 17.3% 0.32 [0.18, 0.57]
Polaschek (2010) 47 77 24 33 21.2% 0.84 [0.64, 1.10]
Subtotal (95% CI) 439 196 54.6% 0.55 [0.30, 1.01]
D
Total events 80 58
TE
Total (95% CI) 703 326 100.0% 0.44 [0.27, 0.71]

P
0.01 0.1 1 10 100

Test for subgroup differences: Chi² = 0.88, df = 1 (P = 0.35), I² = 0%
CE
Fig. 5 Treatment Completion and Violent Recidivism

AC
The RR analysis showed a significant overall effect of 0.44 (p=.001). This indicates a 56%
reduction in risk of reconviction for violence if offenders had completed treatment. This set
of analysis had a significant Q statistic (p = .0003). The I² value (78%) was also highly
indicative of substantial heterogeneity amongst the study results.
The largest effect was noted in the ART study, closely followed by the Anger Management
subgroup. The smallest treatment effect was found in the Correctional Program subgroup. For
violent reconviction, the intensive correctional programs show the smallest effect. Again, the
anger management subgroup shows the greatest effect. Thus, this subgroup analysis found a
ACCEPTED MANUSCRIPT
35
MANAGEMENT
similar quantitative interaction as reported in the analysis for exposure to treatment with
anger management subgroups having the largest effect.
T
IP
Discussion
R
SC
The RR analysis found an overall risk reduction in recidivism of 23% for general recidivism
and 28% for violent recidivism after treatment. The overall risk reduction for treatment
NU
completion as opposed to non-completion was of a 42% reduction in general recidivism and
56% in violent recidivism. Quantitative interactions were noted in all the subgroup analyses,
MA
with anger management subgroups showing the largest effect. This quantitative interaction
between subgroups appears to be clinically plausible.

D
TE
Difference between anger management and violence reduction program
Most studies exploring the effects of traditional anger management on offenders found that
P
CE
treatment had significant effects on reconviction (e.g. Studies 4, 5, 7 and 8). The subgroup
analysis found greater effects for the Anger Management subgroups, compared to the
AC
Correctional Programs subgroup. These results may be due to the differences in severity of
risk of offenders included in the studies 9, 12-14, as violence reduction programs typically target
offenders with an extensive history of violence and criminal convictions. Studies on anger
management groups either did not report the risk level of the offenders (e.g. Studies 4 and 5)
or a mix low risk or high risk offenders (e.g. Studies 7 and 8). Study 10 controlled for the
difference in risk level when comparing violence reduction programs and anger management.
This study concluded that the PVO program was not superior in reducing re-offending to
anger management. However, it must be noted that most of the anger management studies
included were related to the AEMP which at 50 hours of treatment might be considered to be
ACCEPTED MANUSCRIPT
36
MANAGEMENT
more intense than traditional anger management and of moderate intensity in comparison to
violence reduction programs.
T
IP
Effect of treatment completion
R
All the studies that included the non-completer groups found that reconviction was higher for
SC
these groups than for the control groups. This applied for all types of programs reviewed,
NU
indicating that overall treatment completion plays a significant role in changing the
problematic behavior. It might be useful to explore in further research variables that increase
MA
the likelihood of treatment completion. Some notable differences at baseline were identified
from the studies that explored the differences between completer and non-completer groups.
D
Study 9 found that non-completers were significantly younger and had shorter sentences than
TE
the completer group. Marital status also approached significance with more single men in the
attrition group. This difference in marital status was statistically significant in Study 12.
P
CE
Study 8 found that ethnic minorities and offenders classed as medium risk were over-
represented in the attrition group, with these differences also being statistically significant.
AC
Study 11 had controlled for significant differences on salient criminogenic variables between
the completers and non-completers to ensure that any treatment effect can be attributed to
treatment completion. They found that completing treatment decreases odds for reconviction
by 42%. The RR analysis exploring the effects of treatment completion supports these
findings.
Clinical heterogeneity
ACCEPTED MANUSCRIPT
37
MANAGEMENT
The marked heterogeneity of the data indicated a need to explore clinical and methodological
variables that could moderate the overall results in the meta-analyses. Clinical heterogeneity
T
IP
was observed in the offenders included and the intervention modalities.
R
Studies that included offenders of different risk levels and criminal histories could have
SC
affected the outcome measure of recidivism. This could have introduced an artificial effect in
the meta-analysis and increased the amount of error in the overall results. This systematic
NU
review excluded those studies focusing only on domestic violent abusers or those with a
MA
mental health diagnosis, in order to limit the potential confounding effects of extraneous
variables other than anger. However, this could affect the generalizability of the findings, as
D
problematic anger may be comorbid with other mental health diagnoses. Another source of
TE
error could stem from the lack of pre-treatment screening and assessment for anger
dysfunction to ensure the allocated treatment actually meets this need. Indeed, Watt and
P
Howells (1999) and Howells (2004) recommend that offenders be assessed at intake and
CE
screened for anger problems prior to assigning them to a correctional program. Unfortunately,
AC
most studies included offenders on the basis of offending history and may have diluted the
treatment effect as a result.
Another considerable source of error that could have affected the results is the lack of
standardization of follow-up periods between the included studies. Periods of follow-up
varied considerably with a range between 6 months to 5 years.
Methodological differences
Various types of bias were present in the included studies. All the studies included were
NRCTs which are considered to be more prone to bias than RCTs. Selection bias was
ACCEPTED MANUSCRIPT
38
MANAGEMENT
particularly prominent. The direction of bias could not be ascertained, as some of the studies
might be deemed to underestimate or over-inflate the treatment effects at the same time.
T
IP
Therefore, the results must be interpreted with caution.
R
Some studies did not report the appropriate data for meta-analysis. Attempts at contacting
SC
authors also bore little results since some of the studies were at least 20 years old and most of
NU
the data had since been discarded.
Another potential moderating factor was the lack of integrity and treatment fidelity reported
MA
in some studies. This could have potentially moderated the overall effect. Such bias could
have been introduced through different staff competencies and different sites in which
D
interventions where delivered. For instance, Study 14 stated that psychologists were
TE
frequently replaced by students and rehabilitation staff who often had extensive experience in
P
social work but little experience in working with offenders. However, treatment integrity is in
CE
fact considered essential for an effective correctional system (Howells, Watt, Hall, &
Baldwin, 1997; Howells et al., 2002). These issues might have introduced considerable
AC
methodological diversity which could have affected the accuracy of the reported results.
A small sample size in some of the included studies may have introduced a within-study bias,
since the random effects model might give them more weight (Deeks et al., 2008; Kjaergaard,
Villumson, & Gluud, 2001). Furthermore, the lack of appropriate data in some included
studies limited the number of studies included in the meta-analysis and consequently the
possibility of further analysis such as meta-regression. This may have also introduced
artificial bias due to the exclusion of studies.
Conclusions
ACCEPTED MANUSCRIPT
39
MANAGEMENT
Most of the included studies indicate a decrease in re-offending especially when the focus is
on violent reconviction, albeit not always statistically significant (e.g. Study 2). This supports
T
IP
the results of the RR analyses carried out. In fact, Studies 1, 4, 5, 7, 8, 14 reported significant
R
differences in general re-offending for high-risk groups. Studies 2, 4, 5, 7, 8, 12 and 13 all
SC
reported significant differences on violent re-offending between treatment and control
9, 11
groups. Other studies reported that these differences became significant only when the
NU
attrition or non-completers were removed from the analysis.
MA
The analyses also indicate that the less intensive anger management seemed to be the most
effective treatment modality in reducing offending behavior especially violent offending.

D
This seems to contradict Joliffe and Farrington (2007) who had found that greater dosage of
TE
interventions was related with a greater response. Their review had a wider scope than our
study, as they included different types of interventions such as electronic tagging and
P
CE
comparatively brief interventions (e.g. approximately 15-20 minutes) when compared to the
studies included in our review. It is worth noting that our finding is based almost exclusively
AC
on one particular anger management program, the AEMP. The length and intensity (50 hours)
of this program may be classed as moderate in our study but considered more intense in the
Joliffe and Farrington (2007) review. It is possible that this program struck an optimum dose-
response relationship, as intensity beyond a certain threshold would no longer be beneficial.
Therefore, this finding may have significant funding implications for correctional programs,
as less intensive might be more cost-effective than the intensive violence reduction programs.
Further empirical research in this area could explore other moderators, for example, readiness
to change and pre-intervention screening for anger dysfunction. This would aid the
ACCEPTED MANUSCRIPT
40
MANAGEMENT
development of more consistent treatment protocols. This is especially important, considering
the amount of heterogeneity in the included studies in this systematic review.
T
IP
Anger control dysregulation often plays a significant role in violent offending. The findings
R
from this review seem to support this claim that when anger control is addressed, violent
SC
recidivism in particular would be reduced. This may apply for both intensive violence
NU
reduction programs whose program components place heavy emphasis on anger management
modules and the less intensive anger management.

MA
D
P TE
CE
AC
ACCEPTED MANUSCRIPT
41
MANAGEMENT
References
References marked with an asterisk indicate studies included in the meta-analysis.
T
IP
Anderson, C & Carnagey, N. (2004) Violent Evil and the general aggression model. In A G Miller
(Ed.). The Social Psychology of Good and Evil. New York: Guildford Publications.
R
Anderson, C. A. & Bushman, B. J. (2002). Human Aggression. Annual Review of Psychology, 53:
SC
27-51.
Beck, R. & Fernandez, E. (1998). Cognitive-behavioral therapy in the treatment of anger: A meta-
NU
analysis. Cognitive Therapy and Research, 22(1), 63-74.
*Berry, S. (2003). Stopping violent offending in New Zealand: Is treatment an option? New Zealand
Journal of Psychology, 32 (2), 92-100.
MA
*Boe, R., Belcourt, R., Ishak, K. & Bsilis, S. (1997). Follow-up of offenders from the Vancouver
district violent offenders program. Forum on Corrections Research, 9, 3 – 10.
Bowman Edmondson, C. & Cohen Conger, J. (1996). A review of treatment efficacy for individuals
D
with anger problems: Conceptual, assessment and methodological issues. Clinical

Psychology Review, 16(3):251-275.
TE
Cochrane Public Health Group, (2011). Guide for Developing a Cochrane Protocol. 24 November
2011 [cited in March 2014]. Retrieved from:
P
http://ph.cochrane.org/...cochrane.../Guide%20for%20PH%20protocol_Nov%202011_final
%20for%20website.pdf
CE
*Cortoni, F., Nunes, K. & Latendresse, M. (2006). An examination of the effectiveness of the
violence prevention program. Research Report R-178. Ottawa: Correctional Service of
AC
Canada.
Deeks, J.J., Dinnes, J., D’Amico, R., Sowden, A.J., Sakarovitch, C., Song, F., Petticrew, M. &
Altman, D.G. (2003). Evaluating non-randomised intervention studies, Health Technology
Assessment, 7, 27.
Deeks, J.T., Higgins, J.P.T. & Altman, D.G. (2008). Chapter 9: Analysing data and undertaking
meta-analyses. In J.P.T. Higgins & S. Green (Eds.) Cochrane Handbook for Systematic
Reviews of Interventions. Version 5.0.1 [Updated September 2008] The Cochrane
Collaboration, 2008. Retrieved from www.cochrane-handbook.org
Del Vecchio, T & O'Leary, K. D. (2004). Effectiveness of anger treatments for specific anger
problems: A meta-analytical review. Clinical Psychology Review, 24:15-34.
DerSimonian, R. & Laird, N. (1986). Meta-analysis in clinical trials. Controlled Clinical Trials, 7,
177-188.
De Wall, C. N., Anderson, C. A. & Bushman, B. J. (2011). The General Aggression Model:
Theoretical extensions to violence. Psychology of Violence. 1 (3), 245-258.
ACCEPTED MANUSCRIPT
42
MANAGEMENT
Di Guiseppe, R. & Tafrate, C. R. (2003). Anger treatment for adults: A meta-analytic review.
T
Clinical Psychology: Science and Practice, 10(1):70-84.
IP
Dowden, C. & Andrews, D.A. (2000). Effective correctional treatment and violent offending: A
meta-analysis, Canadian Journal of Criminology, 42, 449-467.
R
*Dowden, C., Blanchette, K. & Serin, R. (1999). Anger Management Programming for Federal
SC
Male Offenders: An Effective Intervention. Research Report R- 82, Ottowa, ON: Correctional
Services of Canada.
*Dowden, C. & Serin, R. (2001). Anger Management Programming for Offenders: The Impact of
NU
Program Performance Measures. Research Report R-106, Ottowa ON: Correctional Services
of Canada.
MA
Gilbert, F., Daffern, M., Talevski, D., & Ogloff, J.R.P. (2013). The role of aggression related
cognition in aggressive behaviour of offenders: A General Aggression Model perspective,
Criminal Justice and Behaviour, 40 (2), 119-138.
Gilbert,F. & Daffern, M. (2010). Integrating contemporary aggression theory with violent offender
D
treatment: How throughly do interventions target violent behaviour? Aggression and Violent
TE
Behavior , 15;167-180.
*Hatcher, R. M., Palmer, E. J., McGuire, J., Hounsome, J. C., Bilby, C. A. L., & Hollin, C. R.
(2008). Aggression replacement training with adult male offenders within community
P
settings: a reconviction analysis. The Journal of Forensic Psychiatry & Psychology, 19(4),
CE
517-532.
*Henning, K. R. & Frueh, B. C. (1996). Cognitive-behavioral treatment of incarcerated offenders:

An evaluation of the Vermont Department of Corrections’ cognitive self-change program.
AC
Criminal Justice and Behavior, 22, 523 – 541.

Heudo-Medina, T., Sanchez-Meca, J., Marin-Matinez, F. & Botella, J. (2006). Assessing
heterogeneity in meta-analysis: Q-statistic or I² index?. CHIP Documents Paper 19. Retrieved
from http://digitalcommons.uconn.edu/chip_docs/19
Higgins J.P.T., Altman, D.G., & Sterne, J.A.C. (2011). Chapter 8: Assessing risk of bias in included
studies. In: Higgins JPT, Green S (eds). Cochrane Handbook for Systematic Reviews of
Interventions Version 5.1.0 (updated March 2011). The Cochrane Collaboration, 2011.
Retrieved from www.cochrane-handbook.org.
Higgins, J.P.T., Thompson, S.G., Deeks, J.J., & Altman, D.G. (2003). Measuring inconsistency in
meta-analyses, British Medical Journal, 327, 557-560.
Howells, K. (2004). Anger and its links to violent offending. Psychiatry, Psychology and Law. 11 (2),
189-196.
ACCEPTED MANUSCRIPT
43
MANAGEMENT
Howells, K., Day, A., Bubner, S., Jauncey, S., Williamson, P., Parker, A. & Heseltine, K. (2002).
Anger management and violence prevention: Improving effectiveness, Australian Institute of
Criminology: Trends and Issues in Crime and Criminal Justice, No. 227.
T
IP
Howells, K., Day, A., Williamson, P., Bubner, S., Jauncey, S., Parker, A., & Heseltine, K. (2005).
Brief anger management programs with offenders: Outcomes and predictors of change.
Journal of Forensic Psychiatry & Psychology, 16(2), 296-311.
R
SC
Howells, K., Watt, B., Hall, G. & Baldwin, S. (1997). Developing programs for violent offenders,
Legal and Criminological Psychology, 2, 117-128.
NU
*Hughes, G. V. (1993). Anger management program outcomes. Forum on Corrections Research, 5,
5 – 9.
Jolliffe, D & Farrington, D. P. (2007). A systematic review of the national and international evidence
MA
on the effectiveness on interventions with violent offenders. Research Development Statistics:
Ministry of Justice Research Series, 16/07.
Kjaergaard, L.L., Villumsen, J. & Gluud, C. (2001). Reported methodologic quality and
discrepancies between large and small randomised trials in meta-analyses. Annals of Internal
D
Medicine, 135, 982-989.

TE
Lipsey, M.W., Landenberger, N. & Wilson, S. J. (2007). Effects of cognitive behavioral programs
for criminal offenders. The Campbell Collaboration 2007:6, DOI: 10.4073/csr.2007.6.
P
*Marquis, H.A., Bourgon, G.A., Armstrong, B. & Pfaff, J. (1996). Reducing recidivism through
CE
institutional treatment programs, Forum on Correctional Research, 8, 3-5.

Marshall, W.L. (2009). Manualisation. A blessing or a curse? Journal of Sexual Aggression,15, 109-
120.3
AC
McMurran, M. & Theodosi, E. (2007). Is offender treatment non-completion associated with

increased reconviction over no treatment? Psychology, Crime and Law, 13, 333-343.
*Motiuk, L., Smiley, R. & Blanchette, K. (1996). Intensive programming for violent offenders: A
comparative investigation. Forum on Corrections Research, 8, 10 – 12.
Novaco, R. W. (1997). Remediating anger and aggression in violent offenders, Legal and
Criminological Psychology, 2, 77-88.
Novaco, R.W. (2011). Anger dysregulation: driver of violent offending, The Journal of Forensic
Psychiatry and Psychology, 22 (5), 650-668.
*Polaschek, D. L. L. (2010). High-Intensity rehabilitation for violent offenders in New Zealand:
Reconviction outcomes for high and medium-risk prisoners, Journal of Interpersonal
Violence, Retrieved from:
http://jiv.sagepub.com/content/early/2010/05/26/0886260510365854
ACCEPTED MANUSCRIPT
44
MANAGEMENT
*Polaschek, D. L., Wilson, N. J., Townsend, M. R., & Daly, L. R. (2005). Cognitive-behavioral
rehabilitation for high-risk violent offenders: an outcome evaluation of the violence
prevention unit. Journal of Interpersonal Violence, 20(12), 1611-1627.
T
IP
Reeves, B.C., Deeks, J.J., Higgins, J.P.T, Wells, G.A. Chapter 13: Including non-randomized studies.
In J.P.T. Higgins & S. Green (eds.), Cochrane Handbook for Systematic Reviews of
R
Interventions Version 5.1.0 (updated March 2011). The Cochrane Collaboration, 2011.
Retrieved from www.cochrane-handbook.org.
SC
Review Manager (RevMan) [Computer program]. Version 5.3. Copenhagen: The Nordic Cochrane
Centre, The Cochrane Collaboration, 2014
NU
Ross, J., Quayle, E., Newman, E. & Tansey, L. (2013). The impact of psychological therapies on
violent behavior in clinical and forensic settings: A systematic review. Aggression and
Violent Behavior, 18, 761-773.
MA
Saini, M. (2009). A meta-analysis of the psychological treatment of anger: Developing guidelines for
evidence-based practice. The Journal of the American Academy of Psychiatry and Law, 37,
473-488.
D
Sherman, L. W., Gottfredson, D., Mackenzie, D., Eck, J., Reuter, P., & Bushway, S. (1997).
Preventing crime: What works, what doesn’t, what’s promising. A report to the United States
TE
Congress. College Park: University of Maryland, Department of Criminology and Criminal

Justice.
P
*Serin, R. C., Gobeil, R., & Preston, D. L. (2009). Evaluation of the persistently violent offender
treatment program. International Journal of Offender Therapy & Comparative Criminology,
CE
53(1), 57-73.
AC
Walker J. S & Bright, J. A. (2009). Cognitive therapy for violence: Reaching the parts that anger
management doesn’t reach. Journal of Forensic Psychiatry and Psychology. 20 (2), 174-201.
Watt, B. D., & Howells, K. (1999). Skills training for aggression control: Evaluation of an anger
management program for violent offenders. Legal and Criminological Psychology, 4(2), 285-
300.
ACCEPTED MANUSCRIPT
45
MANAGEMENT
Highlights
T
CBT based anger management achieved a risk reduction of 23% for general recidivism.
R IP
CBT based anger management achieved a risk reduction of 28% for violent recidivism.
SC
Treatment completion may result in a 42% risk reduction in general recidivism.
NU
Treatment completion may result in a 56% risk reduction in violent recidivism.
MA
Anger management may be effective in reducing risk, especially violent recidivism.
D
P TE
CE
AC

CBT and Informed Anger Management

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

CBT and Informed Anger Management

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

CBT and Informed Anger Management

Uploaded by

Copyright:

Available Formats

 

Kevin sammut henwood, shihning Chou, Kevin D. Browne

To appear in: Aggression and Violent Behavior

Received date: 20 April 2015

INFORMED ANGER MANAGEMENT

Tel: +44 115 8466623; Email: [email protected]

A SYSTEMATIC REVIEW AND META-ANALYSIS ON THE EFFECTIVENESS OF CBT INFORMED ANGER

A SYSTEMATIC REVIEW AND META-ANALYSIS ON THE EFFECTIVENESS OF CBT INFORMED ANGER

of recidivism, especially violent recidivism. Moderate-intensity anger management were

Keywords: Anger management, Cognitive Behavioral Therapy, systematic review, meta-

analysis, violence, violent recidivism

A SYSTEMATIC REVIEW AND META-ANALYSIS ON THE EFFECTIVENESS OF CBT INFORMED ANGER

A SYSTEMATIC REVIEW AND META-ANALYSIS ON THE EFFECTIVENESS OF CBT INFORMED ANGER

equivalent to their dysfunctional behavior (Deffenbacher, 2011).

Previous systematic reviews NU

effectiveness of psychological therapies and cognitive behavioral interventions specifically

exclusively on college students or clinical samples.

not examine risk of bias of the 35 included studies.

A SYSTEMATIC REVIEW AND META-ANALYSIS ON THE EFFECTIVENESS OF CBT INFORMED ANGER

psychological interventions such as electronic monitoring. Jolliffe and Farrington (2007)

of general re-offending and a 7-8% reduction of violent re-offending. Jolliffe and

random control trial.

effectiveness of psychological therapies on violent behavior. The participants included for

narrative review included 10 studies, consisting of randomized controlled studies, controlled

A SYSTEMATIC REVIEW AND META-ANALYSIS ON THE EFFECTIVENESS OF CBT INFORMED ANGER

interventions or anger management interventions on adult offenders sentenced to probation

orders or incarceration. Unlike previous reviews, it quantitatively explored the treatment

effects on recidivism by conducting a meta-analysis based on a stringent review process,

common methodological flaws to help explain the findings.

Criteria for study inclusion

A SYSTEMATIC REVIEW AND META-ANALYSIS ON THE EFFECTIVENESS OF CBT INFORMED ANGER

treatment in prison or the community.

Outcomes: Reconviction as a measure of general and/or violent re-offending.

Search methods for identification of studies

the search syntax were made for each particular database.

(anger OR aggression OR violen*) AND (offend* OR crimin* OR perpetrat*) AND (CBT

OR cognitive OR behavio* OR treatment OR “anger management”)

A SYSTEMATIC REVIEW AND META-ANALYSIS ON THE EFFECTIVENESS OF CBT INFORMED ANGER

Data collection and analysis

A SYSTEMATIC REVIEW AND META-ANALYSIS ON THE EFFECTIVENESS OF CBT INFORMED ANGER

Figure 1. The Selection Process.

guideline (Cochrane Public Health Group, 2011).

A SYSTEMATIC REVIEW AND META-ANALYSIS ON THE EFFECTIVENESS OF CBT INFORMED ANGER

Assessment of risk of bias in included studies

methodological quality of studies (Sherman, Gottfredson, Mackenzie, Reuter, & Bushway,

effect of anger management to more intensive correctional programs. Subgroup allocation

A SYSTEMATIC REVIEW AND META-ANALYSIS ON THE EFFECTIVENESS OF CBT INFORMED ANGER

the respective programs.

2014). Analyses were carried out to determine:

2. The overall effect of treatment completion on recidivism and violent recidivism in

comparison to attrition groups with the same subgroups of treatment modality.

A SYSTEMATIC REVIEW AND META-ANALYSIS ON THE EFFECTIVENESS OF CBT INFORMED ANGER

be explained through heterogeneity (Higgins, Thompson, Deeks, & Altman, 2003). An I²

Table1 Relevant characteristics and findings of included studies.

Mean age control 24.4 yrs

**Results presented for violent offenders only

A SYSTEMATIC REVIEW AND META-ANALYSIS ON THE EFFECTIVENESS OF CBT INFORMED ANGER

thus included offenders with different levels of risk.

(anger OR aggression OR violen) AND (offend OR crimin* OR perpetrat*) AND (CBT