UvA Clinical Master Psychology
___________________________________________________________________________
Reducing Aggression in Incarcerated
Violent Offenders by Increasing the
Recognition of Happy Emotions: Testing an
Intervention
Lotte Habets MS Clinical Psychology
Supervisor: Henk Cremers
Internal supervisor: Niki Kuin
ABSTRACT
Background In existing literature it is found that aggressiveness is associated with a tendency to interpret others’ behaviors as having hostile intent, even when the behavior is ambiguous, a bias called the hostile attributional bias (HAB). This study investigated the hostile attributional bias (HAB) among violent offenders, using a facial-affect recognition task. In this experiment, HAB was modified to encourage the perception of happiness over anger in ambiguous facial expressions. It was tested if this modification resulted in reduced self-perceived and staff-observed aggression. Methods 95 male adult incarcerated violent offenders in the Penitentiary Institution (PI) Vught in the Netherlands were randomly assigned to either the experimental or control condition. HAB was assessed before and after an intervention program. In the experimental condition HAB was altered via a feedback-based training. Self-reported and staff observed aggression were measured before and after the intervention program and with six weeks follow-up. Results All participants reported less aggression directly after the intervention program and six weeks follow-up, regardless of the condition. The results suggest that training facial recognition in the happy-angry continuum positively effects self-perceived cognitive, emotional and behavioral aggression. Future studies should focus on the effects of emotion perception interventions on both self-perceived as observed aggression, in order to reduce violence in incarcerated populations.
TABLE OF CONTENTS
ABSTRACT ... 2
TABLE OF CONTENTS ... 3
1. INTRODUCTION ... 4
Violence Prevention ... 4
Neuropsychology and Behavioral change ... 4
Defining Aggression ... 5
Etiology and Hostile Attributional Bias ... 5
Neuropsychological Assessment and Emotion Perception ... 6
Modifying Emotion Perception ... 7
The present study ... 8
2. METHODS ... 9
2.1 Participants ... 9
2.2 Materials ... 11
Screening Checklist Intelligence (SCIL): ... 11
Novaco Anger Scale-Provocation Inventory -Dutch (NAS-PI): ... 11
Observation Scale for Aggressive Behavior (OSAB):... 12
Semi Structured Interview (SSI): ... 13
Criminal Records: ... 14
2.3 Procedure ... 14
3 RESULTS ... 16
3.1 Preliminary Analysis ... 16
Descriptive Characteristics ... 16
Assessment of Statistical Assumptions ... 17
3.2 The Association between HAB and Aggression... 18
3.3. Manipulation Check ... 19
3.4a Effectiveness Intervention - Self-reported Aggression ... 20
3.4b Effectiveness Intervention - Staff-rated Aggression ... 22
3.5 Exploratory analysis ... 23
4 DISCUSSION ... 24
Summary of Results ... 24
Interpretation of Results ... 25
Limitations and Future Recommendations ... 26
5 REFERENCES ... 28
1. INTRODUCTION Violence Prevention
Human violence is an ever-present problem with substantial consequences for society. The prevention of crime has been on the political agenda for years. Interventions based on primary and secondary prevention, are seen as fundamental in crime prevention. According to Marcus (2007) primary prevention focusses on preventing the onset of violence within the general population, whereas secondary prevention can be seen as interventions that are implemented selectively with individuals being high risk violent offenders. Unfortunately, there is also a great deal of offenders that reoffend. Among violent offenders, the percentage of recidivists with severe violent offences (legal requirements of >8 years of custody) increases in the course of time: three years after release it is 9, 5% and 18 years after release it is 20.5% (Schönberg & de Kogel, 2012). Therefore, the establishment of tertiary prevention is highly demanded. This can be defined as strategies that reduce rather than reverse violence, in individuals that have already engaged in violent behavior, e.g. people with criminal records (Marcus, 2007). In order to generate evidence based violence prevention strategies, a clear understanding of violence and its nature is needed. Therefore, multiple disciplines are involved; criminology, forensic studies, psychology, neuroscience etc. As an example, the Justice Department in court makes its judgments on both juridical and psychological information. Moreover, a great deal of the people that come into contact with the Justice Department suffer from psychiatric or behavioral disorders (De Kogel, 2008).
Neuropsychology and Behavioral change
Over the years, there have been relevant developments surrounding behavioral- and neuroscientific influences in the clinical forensic psychological area of science. Cornet et al. (2015) propose a summary of research that reveal evidence that specific neurobiological measures, such as brain activity, hormones and HRV, change in response to anger interventions, with some of the studies linking neurobiological change to behavioral improvement. However, there were only few studies that investigated adults with anti-social/aggressive behavior. According to Parrot and Giacola (2007) the lack of evidence based anger interventions is due to limitations in measurements of aggression and violence, referred to as a criterion problem. This problem is twofold; first, definitions of aggression and its manifestations, such as violence, are frequently misinterpreted. Researchers inconsistently differentiate between aggressive behaviors (violence) and related emotional/cognitive constructs of aggression (anger, hostility). The various routes by which aggression may be expressed are often inappropriately discerned in research (direct vs. indirect). Second, assessment tools designed to measure aggressive acts do not adequately correspond to these existing definitions. For example, a self-report questionnaire that measures “aggression”
might actually assess the cognitive experience of the individual, whilst observation rating scales might actually assess direct forms of aggression (verbal, physical). Using only one of these instruments to assess “aggression” therefore, would give an incomplete image. The lack of precision in measurement created challenges for investigators attempting to assess risk factors (underlying factors) that predict aggressive behavior and moreover, to develop risk prevention strategies.
Defining Aggression
Taking this in consideration, the first attempt for the current study was to clearly define aggression and violence. Violence and aggression are often used in the same context, although there is a difference between them. Baron and Richardson (1994, ref. Krahé, p. 9) proposed one of the first concise definitions of the term “aggression” that is still widely used today (Parrott & Giancola, 2007). They suggested that aggression should be defined as “any form of behavior directed toward the goal of harming or injuring another living being who is motivated to avoid such treatment”. Rippon (2000) criticizes this definition and argues that there is a myriad of definitions, inclusions and exclusions of the term aggression in scientific research. According to Rippon (2000) there are some factors in the aggression literature that seem to be constant. He adds two parts to the definition of Baron and Richardson (1994); “aggression can be physical or verbal, active or passive, and can be focused on the victim(s) directly or indirectly. Aggression can be the manifestation of anger and can be directed either toward oneself or other persons” (Rippon, 2000, p.5). Moreover, the term “violence” refers to a form of aggressive behavior that is active, direct and physical, either to oneself or others. Violence is less difficult to detect, since it is observable by others. Other more passive subtypes of aggression, for example ‘irritation’, might only be assed via self-report questionnaires (Parrot & Giancola, 2007). Another widely used classification as observed in current scientific research, is the division into proactive and reactive aggression.
Proactive aggression is characterized as instrumental, planned behavior, not accompanied by autonomic arousal, whereas reactive aggression, in contrast, is accompanied with high levels of autonomic arousal and emotions such as fear and anger (Card & Little, 2007; Dodge, 1991, ref. Fite et al., 2009). Proactive aggression is found to be less common and unlike reactive aggression it is closely tied to psychopathic personality traits (Raine et al., 2006; Fite et al., 2009). Reactive aggression is associated with more general psychopathology (Walsh, Swogger & Kosson, 2009). It is also referred to as hostile, impulsive aggression. The current study had taken the above definitions into account while trying to further capture the etiology of aggression. A model that contributes to this understanding is the Social Information Processing Model of Dodge and Schwartz’s (1997 ref. Miller & Lynam, 2006).
According to the Social Information Processing model behavior in a particular social situation will occur as a direct reflection of a person’s mental processing of that situation. These processes start in early childhood. Dodge and colleagues state that aggressive children are more likely to engage in aggressive behavior than nonaggressive children because they process information differently. According to these researchers aggressive children make several mistakes in the various steps of the social information processing model. First, they tend to encode environmental cues in a selective an inaccurate manner, attending selectively to hostile or threatening cues (Dodge & Frame, 1982, ref. Miller & Lynam, 2006). Second, they show misinterpretations in other’s behavior, especially when this behavior is ambiguous. Hence, aggressive individuals quickly make the assumption that ambiguous human actions are hostile in nature, even before specific hostile cues could be encoded in the particular situation (Wilkowski et al., 2007). These biases together fall under the term hostile attributional bias (HAB; Dodge, 2006). This phenomenon was found in replication studies with both children and adult populations (De Castro et al., 2002).
In earlier scientific studies on HAB, researchers discriminated between different subtypes of aggression and their relation to HAB. Some of them found that HAB was only displayed by reactive aggressive individuals and not by proactive aggressive individuals (Dodge and Coie, 1987, Schwartz et al. 1998, ref. Helfritz-Sinville & Stanford, 2014). However, more recent studies about HAB are divergent and therefore it is unclear whether HAB is related to either or both reactive and proactive aggression (Sinville & Stanford, 2014; Bailey and Ostrov, 2008). For example, Helfritz-Sinville and Stanford (2014) found that signs of HAB were evident in the behavior of individuals that show both types of aggressors, as compared to a control group without these behavioral tendencies. More interestingly, Helfriz-Sinville and Stanford (2014) found that individuals with signs of HAB reported to act out in aggressive behavior in a threatening situation, whilst individuals without HAB would not. To continue, some researchers argue that studies on HAB should primarily be approached from a behavioral point of view. In fact, HAB has been studied extensively in the physical aggression literature. These studies have posited that HAB leads to the development behavior problems, leading to physical violence or predict aggressive behavior (Godleski and Ostrov, 2010; Dodge et al., 1990). Therefore, HAB has been established to be essential in the development and maintenance of physical aggression (De Castrio et al. 2002), which is commonly found in behavioral disorders such as conduct disorder and antisocial behavior in both children and adults (Best, Williams & Coccaro, 2002) and in psychopathy and is constituent dimensions (Vitale et al., 2005). These studies suggest that HAB is a complex construct that can be approached to from multiple perspectives. Therefore, the current study will be focusing partly on these different classifications of aggression in relationship to HAB.
A variety of neuropsychological paradigms have been used to investigate cognitive attention biases in various types of psychopathology characterized by emotional dysfunction. Very popular ones are the dot-probe task, emotional strooptasks and facial expression tasks. For example, the dot-probe task has been investigated extensively in disorders such as anxiety, depression, post-traumatic stress disorder and chronic pain. It is found that anxious individuals show a quicker reaction time to threatening stimuli in the task than non-anxious individuals. These results are replicated many times in other studies (Salemink et al., 2007). Until now, a number of studies have found that anger-prone and aggressive clinical populations exhibit a bias in attention (HAB) towards violence-related words (Domes et al., 2013; Smith & Waterman, 2003). However, only few studies have studied HAB in relationship to facial expression tasks. Facial expressions carry a broad range of socially relevant information reflecting the internal state of the sender. Therefore, processing facial affect is crucial for socialization and normal social interaction (Corden et al., 2006; Fridlund, 1991). It is plausible that the disability to correctly decode subtle facial cues during social interaction leads to misinterpretations of other’s emotions and intentions. In fact, compelling evidence of replicated studies shows that ambiguous facial expressions are more negatively interpreted (as ‘angry’ or ‘hostile’) by individuals with aggressive tendencies than by individuals without those tendencies (Marsh & Blair, 2008; Schönenberg, 2014). Given that aggressive individuals experience these deficits in emotion recognition, they are more likely to experience ambiguity in social interactions and to fall into their maladaptive characteristic schema of expecting the worst from other people (Social Information Processing Model, Dodge, 2006). Moreover, researchers found that these types of deficits are commonly associated with socially deviant behavior seen in behavioral disorders, such as conduct and anti-social disorders (Best, Williams, & Coccaro, 2002; Bowen & Dixon, 2010; Fairchild, et al.,2009; Sato et al., 2009). Modifying Emotion Perception
Some researchers have tried to repair deficits in emotion processing, particularly in young people with behavioral problems. Dadds et al. (2006) found that children with psychopathic traits had an impairment in fear recognition, however, this deficit could be temporarily reversed by directing attentional focus to the eye region of the facial stimuli depicting emotion. In response to this, Penton- Voak et al. (2013) proposed the idea of modifying biases in emotional processing through neuropsychological techniques, to influence mood in aggressive individuals. Their theory was partly based on an experiment of Harmer et al. (2009) in which cognitive neuropsychological models of antidepressant drug’s action indicated that antidepressants have their effect on people’s mood via changes in emotional processing (Harmer et al., 2009). Penton-Voak et al. (2013) suggested that experimentally manipulating biases in facial emotion recognition in happy and angry faces might correct deficits in emotional processing and this could influence mood and reduce aggression. In their experiment, biases in emotion processing and recognition regarding ambiguous facial cues on a happy-anger continuum were
modified, using a feedback training. A group of British adolescents with high risk of criminal offending received feedback on how well they could rate the facial cues as ‘happy’ or ‘angry’. The direction of the feedback they received, determined how their emotion perception would change over time. The experiment group received feedback directed towards more ‘happy’ judgements of the stimuli, expecting the participants to respond differently than before they received feedback. The control group received feedback directed towards their original responses, therefore, no change in their responses was expected. The results showed a decrease in self-reported aggression and observed aggressive behavior in the experiment group, compared to no change in the control group (Penton-Voak et al., 2013). The strength of this study design is that the training establishes changes in HAB on an implicit level. Therefore, the feedback training used in this experiment has showed to be effective in a way that it has made beneficial changes in emotion perception and reduced anger and aggression in young high risk criminal offenders. However, no clear understanding of which neuropsychological mechanisms underlie these effects exists yet in the current field of research.
There is some evidence that certain neural pathways are activated when processing angry facial expressions. Tsang et al. (2015) investigated neural correlates of face emotion processing using functional magnetic resonance imaging (fMRI). Young patients with severe mood dysregulation (SMD) exhibited increased activation in the parahippocampal gyrus (PHG) and superior temporal gyrus relative to healthy volunteers when processing angry faces. SMD patients showed decreased activation in the insula, PHG and thalamus compared to healthy volunteers when processing happy faces. This suggests that perturbed activation in emotion processing areas might be important in the maintenance of irritability and aggression symptoms in SMD patients. Moreover, Stoddard et al. (2016) replicated the study of Penton-Voak et al. (2013) and found the same results with disruptive mood dysregulation disorder (DMDD) in youth; a shift in classifying ambiguous faces more often as happy rather than angry, using the feedback training. More importantly, their results suggested that the training was associated with decreased irritability and changes in activation in the lateral orbitofrontal cortex.
The present study
These studies support the idea that emotion recognition modification tasks stimulate and influence the underlying mechanisms that are active when processing angry faces. There is some evidence that these modifications are generalizable to changes in daily mood and behavior (Penton-Voak et al., 2013; Stoddard et al., 2016). However, there is insufficient evidence that changing emotion perception in aggressive individuals results in behavioral and cognitive changes in relation to aggression. Therefore, the present study will be focusing on the possible effects of the facial recognition task of Penton-Voak et al. (2013) on aggressive behavior among a population of adult incarcerated male violent offenders. Since most studies have studied HAB and aggression interventions in youth populations, it seems relevant to investigate adult
populations as well. Moreover, this study has not been replicated in a prison environment yet. The results could have great implications for the rehabilitation of violent offenders and secondary prevention of criminal acts.
Another strength of this study design is that it uses implicit measures of HAB. In previous research, HAB has mostly been operationalized and measured via self-report questionnaires or Conflict Situation Vignettes (Tremblay & Belchevski, 2004, ref. Helfritz-Sinville & Stanford, 2014) So far, only few studies have utilized implicit techniques. According to Bluemke et al. (2010) implicit measures do not rely on conscious self-report (explicit) but on hard-to-control spontaneous associations and therefore are considered automatic responses. Moreover, implicit measures are less susceptible to biasing factors such as social desirability than explicit measures (Bluemke et al., 2010).
In order to answer the research questions: ‘Is a computer based facial-affect recognition task able to make changes in HAB and moreover, make changes in cognitive and behavioral aggression?’ this study will investigate four hypotheses. First, this study investigates whether HAB is present among Dutch violent offenders using a facial-affect recognition task (Penton-Voak et al., 2013). Then, a feedback intervention is used, attempting to influence HAB, encouraging more happy perceptions of ambiguous faces on the task. After two weeks and after six weeks follow-up, the study will determine the long-term effect of changes in HAB on self-rated and staff-observed aggression. The first hypothesis is that HAB is related to self-reported and staff-rated aggression in incarcerated violent offenders. A second hypothesis states that the intervention training will modify a shift in facial emotion recognition towards the happy- dimension over time in the experimental condition. The third and fourth hypothesis states that participants in the experimental condition improve significantly more during training than participants in the control condition on resp. self-perceived and staff-rated aggression. This will be measured two and six weeks after the intervention. Exploratory analysis will investigate whether the changes in HAB are related to possible changes in aggression.
2. METHODS
2.1 Participants
All participants were adult males (18 years and older), who were incarcerated in the Penitentiary Institution (PI) Vught in the Netherlands. Penton-Voak et al. (2013) found significant effects for the experimental (n=23) condition on staff-observed and self-rated aggression after the intervention compared to the control condition (n =23) with effect sizes of Cohen’s d = resp. 1.26 and 1.11. The current study used a sample size of 95 (experimental=46; control=49) with an effect size of Cohen’s d =1.25, as estimated with G*Power to generate enough statistical power (0.8) for the intervention study (Faul, et al., 2009). Participants received
no incentive for participation in the study. However, soda and cookies were provided during the intervention week to make participating more attractive.
Participants had a history of or present conviction for a violent/ aggressive crime. These participants were recruited from two departments: the Institution of Systematic Offenders (ISD) and regular prison department. Inclusion took place when participants were stable enough to participate (e.g. not suffering from a psychotic, manic or major depressive episode during at least six months prior to testing, according to their self-report and confirmation from their psychologists). Illiteracy was an exclusion criteria, because of the use of questionnaires. Additionally, participants were excluded if staff members expressed their concerns for the wellbeing of the participant or for the safety of the administrator/trainer. Moreover, participants needed to have a remaining detention durance of at least eleven weeks to be able to participate.
Another exclusion criteria was if participants had been diagnosed with DSM-IV autism spectrum disorders (ASD). According to a meta-analysis of Harms, Martin and Wallace (2010) several studies provided evidence that individuals with ASD decode facial expressions differently than individuals without ASD. They experience difficulty labeling and matching emotions (Macdonald et al., 1989; Tantam et al., 1989; Ozonoff et al., 1990, Ashwin et al., 2007; Golan et al., 2010; ref. Harms et al, 2010). It is unclear whether the computer task in the current study is applicable to ASD individuals, therefore these individuals were excluded from the study population.
It is expected that personality disorders (especially cluster B personality disorders; Brazão et al., 2015) and psychopathic personality traits (Kuin & Mastohoff, 2016) are very common among the current population. The association between personality disorders and violent offenses is widely known; a meta-analysis of Brazão et al. (2015) shows that this relationship is reported in several studies (Duggan & Howard, 2009; Gilbert & Daffern, 2011; Roberts & Coid, 2010; Short, Lennox, Stevenson, Senior, & Shaw, 2012; Warren & South, 2009; Yu, Geddes, & Fazel, 2012, ref. Brazão et al., 2015). Moreover, it is suggested that proactive aggression is prevalent in antisocial personality disorder and developmental psychopathy (Blaire, 2001). It is expected that aggression rates are high for participants with cluster B personality disorders. However, diagnostic information on personality disorders from existing psych-medical files were often unavailable so it was not possible to evaluate the association between personality disorders and HAB and the effectiveness of the intervention.
Mild intellectual disability (ID) does not form a contra-indication for participating in the study. However, mild ID will be monitored to investigate if participants with mild ID respond differently on the task than participants without mild ID. A study of Hayes et al., (2007) with 140 adult prisoners found that ID was very common: 21.7% had a score of lower than 79 IQ points on the WAIS-III and VABS, indicating ID. Mild ID was found in 24.5 % of all prisoners, with IQ scores between 80-89 on the WAIS-III and VABS (Hayes et al., 2007). Therefore, it is expected
that ID is common among the current study population. According to Rose, West and Clifford (2000) there has been some consideration as to whether people with intellectual disabilities can report their emotional states accurately, however, evidence for this is growing. The current study will investigate whether the training is applicable to populations wild mild ID.
2.2 Materials
Screening Checklist Intelligence (SCIL): The Screening Checklist Intelligence (SCIL; Kaal, Nijman & Moonen, 2013) was used to detect mild ID, differentiating between below or above 85 IQ-points. The SCIL includes fourteen subtests (math, telling time, writing and reading). A score of 19 or below indicates a mild intellectual disability, meaning an IQ score below 85 IQ-points. The SCIL is often used in a justice, rehabilitation and psychiatric context to make quick decisions about the intellectual abilities of a patient. Kaal, Nijmand and Moonen (2015) found good internal consistency measures with a Cronbach’s alpha of .80. Test-retest reliability over six weeks was stable with a significant correlation, Pearson’s r = .92. Validity measures showed good specificity (83%) and sensitivity (82%) and an AUC-value of .93, supporting the predictive value of the test scores.
The Reactive Proactive Aggression Questionnaire – Dutch (RPQ): The RPQ is a self-report questionnaire, consisting of 23 items with a 3-point Likert scale developed by Raine (2006) that assesses reactive and proactive aggression. It takes 5-10 minutes to complete. Scores are comprised to two factor scales: proactive (12 items; e.g. ‘How often did you use physical violence to get people to do what you want?’) and reactive aggression (11 items; e.g. ‘How often did you respond aggressively when someone nagged you?’). The total score for each scale was calculated by summing the items per scale, with higher scores reflecting higher self-reported reactive and proactive aggression. The RPQ has demonstrated excellent internal consistency in adult offender populations, with Cronbach’s alphas of .83 and .87 for the proactive and reactive items and Cronbach’s alpha of .91 for the total RPQ. Test-retest reliability over three year follow-up showed acceptable stability (rho= .78) (Cima et al., 2013). Multiple studies with juvenile and adult populations showed that both reactive and proactive aggression subscale scores of the RPQ have adequate convergent and discriminant validity, supporting the construct validity of the RPQ (Cima & Raine, 2009; Fung et al., 2009; Raine et al., 2006; ref. Cima et al., 2013).
Novaco Anger Provocation Inventory -Dutch (NAS-PI): The Novaco Anger Scale-Provocation Inventory (NAS-PI, Novaco, 1994, ref. Hornsveld et al., 2011) is composed of two parts and contains 71 items in total. The NAS contains 48 items, divided over three interrelated subscales that represent anger dispositions; Cognitive (16 items, e.g., “I feel like I am getting a raw deal out of life”), Arousal (16 items, e.g., “I feel agitated and unable to relax”) and Overt behavior (16 items, e.g., “When someone yells at me, I yell back at them”). Items are rated on a 3-point Likert scale (1 = never true, 2 = sometimes true and 3 = always true). The PI exists of 25
items that focus on situations that lead to anger in five content areas – disrespectful treatment, unfairness, frustration, annoying traits of others, and irritations – using a 4-point Likert scale (1 = not at all angry, 2 = a little angry, 3 = fairly angry, 4 = very angry). Hornsveld et al. (2011) found good internal consistency and test-retest reliability in a population with forensic clinical and policlinic psychiatric patients. The internal consistency (Cronbach’s α) in clinical patients for the NAS and the three subscales was respectively .93, .77, .82, .86 and in policlinic patients respectively .95, .81, .88, .90. The internal consistency of the PI in clinical patient group was .90 and in policlinic patient group was .94. The test-retest reliability of the NAS in a subgroup of 90 policlinic patients was respectively .80, .71, .78, and .79.
Observation Scale for Aggressive Behavior (OSAB): The Observation Scale for Aggressive Behavior (OSAB; Hornsveld, Nijman, Hollin, & Kraaimaat, 2007) was used to assess aggressive mood and behavior as weekly observed by staff members. The tests includes 40 items divided over 6 subscales, however, the current study used the subscales Irritation/ Anger (nine items, e.g.: ‘The person was tensed’) and Aggressive behavior (ten items, e.g.: ‘The person scolded the staff’). Items are scored by the staff, each score indicates the frequency of the observed behavior during the past week, using a five-point Likert scale (1 = never, 5 = often). In a study of Hornsveld et al. (2007) with a group of 220 adult delinquents, reliability measures of the three subscales were respectively .82, .79 and .93 (Cronbach’s α) for internal consistency, .79, .81 and .70 for interrater reliability and .59, .57 and .76 for test retest reliability. Convergent validity was supported by significant correlations between analogous subscales of the Forensic Inpatient Obersvation Scale (FIOS; Timmerman, Vastenburg, & Emmelkamp, 2001, ref. Hornsveld, 2007) and with the self-report questionnaires measuring hostility, anger and aggression (Hornsveld, Nijman, Hollin, & Kraaimaat, 2007).
Agressie Vragenlijst – Aangepast versie (AVL-AV): A shorter version of the Dutch Aggression Questionnaire (AVL; Meesters, Muris, Bosma, Schouten, & Beuving, 1996), the customized Aggression Questionnaire (AVL-AV; Hornsveld, Muris, Kraaimaat, & Meesters, 2009) is a twelve item self-report questionnaire that measures aggression, divided over four subscales; Physical aggression (four items, e.g.: ‘Sometimes I feel the need to hit someone’), Verbal aggression (four items, e.g.: ‘I have threatened someone I know in the past’), Anger (four items, e.g.: ‘I get angry easily, but I cool down afterwards’ and Hostility (four items, e.g.: ‘I sometimes feel life has not been fare to me’). Items were answered on a 5-point Likert scale, varying from 1 (= strongly disagree) to 5 (= strongly agree). Hornsveld, Muris, Kraaimaat, en Meesters (2009) found an acceptable fit of the four-factor model of the AVL-AV in their study with 138 clinical patients, 206 policlinic patients and a control group of 160 high school students. The internal consistency coefficients (Cronbach’s α) of the 12 items varied between .72 and .88, the average inter-item correlations between .19 and .27 and the average item-total correlations between .35 and .50. Cronbach’s α for the four subscales varied between .38 and .74, the average
inter-item correlations between .18 and .49 and the average of the item-total correlations between .23 and .57. Test retest correlations (four weeks interval) were significant for the AVL-AV total and for the subscale scores, except for the subscale Physical aggression. Validity of the AVL-AV was supported by the significant correlations with levels of observed aggressive behavior (OSAB), anger (NAS) and agreeableness (NEO-FFI; Hornsveld, Muris, Kraaimaat, & Meesters, 2009). These empirical results correlate with those of the original AVL questionnaire (Diamond et al., 2005; Diamond & Magaletta, 2006).
Semi Structured Interview (SSI): The SSI is administered by the researcher and takes about 20 minutes to complete. The participant is asked to answer questions regarding drug abuse (defined as weekly use for at least 3 months), ethnicity and highest completed education. This interview is not tested on its validity. It is comprised of frequently asked questions in a number of studies. Parameters are all widely accepted and used in other studies (Penton-Voak et al., 2013).
The Facial Recognition Task (Penton-Voak et al., 2013): Penton-Voak et al. (2013) used a computer-based task that was designed to modify the perception of (ambiguous) facial expressions of emotion. The task consisted of pictures of happy and angry facial expressions that were presented on the center of the computer screen. The original images came from the Karolinska Directed Emotional Faces study (Lundqvist, Flykt & Öhman, 1998). These were images of 20 individual male faces showing a happy facial expression and the same 20 individuals showing an angry expression. With established techniques for prototyping and transforming facial textures (Tiddeman, Burt & Perrett, 2001), one male prototypical face was generated from these images. A linear sequence was generated that consisted of a morphed continuum of 15 images of the prototypical face that changed incrementally from unambiguously happy to unambiguously angry, with emotionally ambiguous in the middle (see Figure 1). Faces were presented in random order, for 150 ms, followed by a fixation cross (1,500–2,500 ms, randomly jittered). Participants rated these images as happy or angry, in a two-alternative forced-choice procedure administered by a computerized test in E-prime 2.0 (pressing ‘C’ means ‘happy’, pressing ‘M’ means ‘angry’.).
Figure 1. Illustration of the design of the computer task, based on the study of Penton-Voak et al. (2013). The top row shows an example of the balance point of a participant at baseline phase. The bottom row shows an example of the balance point of a participant in the Experimental condition, at posttest phase. At posttest, a shift in balance point is
expected directed towards the “angry”- end of the continuum, meaning that the threshold to respond “angry” was higher and the tendency for the participant to respond “happy” increased.
For participants in the control condition, no shift in balance points was expected. Adapted from ‘Increasing Recognition of Happiness in Ambiguous Facial Expressions Reduces Anger and Aggressive Behavior’ by I.S. Penton-Voak, J. Thomas, S.H. Gage, M. McMurran, S. McDonald and M.R. Munafò, 2013. Psychological science, 24(5), p. 690.
Criminal Records: These records are used to obtain information about the crime history (based on prior convictions) and the current crime, current regime within the prison setting and age. Although there are number of reasons to assume that the number of convictions is not a good reflection of total committed crimes, it is the most factual information about criminal behavior. Self-report on this topic is often difficult for participants, especially when asked for within a juridical context and therefore leads to unreliable results.
Psychomedical records: To make sure if participants don’t meet the exclusion criteria for severe mental illness, we can check existing psychomedical files with psychiatric diagnoses, if these are available. This way, the burden for participants, is the lowest.
2.3 Procedure
After inclusion of the participants, they were randomly allocated to either the intervention or the control condition. The trainer and participants were not able to differentiate which condition was activated participant (double blind). Before participating a trainer has a personal conversation and informs the participant about the procedure. After 24 hours of deliberation an informed consent form is signed. Intakes are done to query autobiographical data (age, nationality, work experience, educational level) using a standardized interview. Moreover, possible neurocognitive damage was determined by asking participant’s history of drug use and by asking if they ever had head injuries throughout their lives. Finally, a SCIL test (see materials) was taken to control for mild intellectual disabilities.
During eleven weeks, staff members wrote observational registrations (OSAB, see materials) every week, once a week. Before the intervention week started, self-report questionnaires (NAS-PI, AVL-AV, RPQ) were scored. This way a baseline score for staff-observed and self-perceived aggression was estimated. In week five, the training started. (See Appendix A for an overview of the tests in a total time frame). During five days (Monday – Friday) training took place on a daily basis. Each day the emotion perception task, developed by Penton-Voak et al. (2013), was applied (see materials).
Each training consisted of three phases: baseline, intervention and pretest. The baseline phase was one session that consisted of 45 trials, in which each face of the continuum was presented three times in randomized order. Each participant’s baseline balance point was estimated, by calculating the number of ‘happy’ responses as a proportion of the total number of responses. Statistical analysis were conducted to determine differences in baseline balance points between the experiment group and control group.
In the intervention phase, there were six sessions, consisting of 30 instead of 45 trials. Each face from the 15-face continuum was presented twice in randomized order within each session. Participants received feedback following their responses. This was a written message on their screen saying “Correct/Incorrect! That face was happy/angry.” The control condition received feedback that was formed on the participant’s baseline balance point, meaning that responses were “correct” only when the participant identified faces below the original balance-point as ‘happy’, and faces above the balance point as ‘angry’. Otherwise, responses were “incorrect.” The experimental condition received feedback that was also formed on the participant’s baseline balance point, but the “correct” classification was shifted two faces toward the ‘angry’- end of the continuum. This way, the two faces nearest the balance point in direction of the ‘angry’- end of the continuum had to be classified as ‘happy’ in order to be “correct”. Therefore, participants learned that these faces (previously responded to with ‘angry’) were considered to be ‘happy’ as a result of the feedback (Penton-Voak et al., 2013). At posttest phase, the session was identical to the baseline phase, but it was employed to assess the magnitude of the shift and to confirm whether the modification training was successful in shifting the balance point in the experimental condition.
The training week took place in groups with a maximum of eight participants, each participant worked individually on his training session. Each training took about 20 minutes to complete. The training week was administered by two trainers, who were master students in (neuro-, forensic or clinical) psychology. In week eleven, another posttest phase computer task was done to assess the balance point (HAB) at follow-up and participants filled out a final self-report aggression questionnaire (NAS-PI). Participants got a debriefing of the study.
3 RESULTS 3.1 Preliminary Analysis Descriptive Characteristics
From the total sample of 116 participants 19 participants dropped out before the intervention week started. A total of 95 participants completed the first day of the intervention, in which baseline HAB was measured (hypothesis 1). 82 participants completed the total intervention week with N=41 in the experimental condition and N=41 in the control condition. This population was used for hypothesis 2, 3 and 4. Of these participants, demographics and descriptive characteristics are presented in Table 1.
Table 1
Descriptive characteristics for participants that finished the intervention week (N=82)
Characteristics Country of birth Netherlands 77.8% Morocco 4.6 % Suriname 3.7 % Curacao 4.6 % Belgium 3.7 % Other 5.6 % Average Age 38.8 Minimum age 20 Maximum age 75 No education 12 % Education 88 % Primary education 23 % Vocational 19.4 Community college 34.3 Pre-university/university 10.2 SCIL score ≤19 38.9 % Medication 56.4 % Antipsychotics 20.2 % Anti-depressives 11.7 % Benzodiazepines 21.3 % Somatic medication 41.5 %
As seen in Table 1, the current population consists mostly out of adult man (age, M=38.8) with Dutch nationalities (77.8%). Interestingly, the percentage of offenders that have a score of 19 or below on the SCIL test, indicating high change of mild ID, is 38.9%. This is notably higher than the percentage of 24,5 with mild ID (tested with WAIS-III and VABS tests) that was found in the study of Hayes et al. (2007). Another interesting finding is that a lot of participants use mood stabilizers and antipsychotics. This data will not be included in the analysis. It is unclear how these medications could impact the emotion recognition task, further research is needed. During the eleven weeks of the study, more participants dropped out (N=18) and a total of 77 participants finished the follow-up of six weeks. Drop-out was mostly due to practical implications (leave, displacement) or personal reasons (motivation). The participant’s test scores on the three self-report aggression questionnaires (RPQ, NAS-PI and AVL-AV physical aggression), the staff’s scores on the OSAB and the HAB balance point scores at baseline day one are presented in Table 2.
Table 2
Means and Standard Deviations of Pre-test Scores at on OSAB, RPQ, NAS-PI, AVL-AV and Balance Point on Day 1 of Intervention Week (N=95)
Test scores Mean SD
OSAB 10.26 2.68 RPQ Total 13.98 10.17 RPQ Reactive 8.67 5.60 RPQ Proactive 5.31 5.18 NAS-PI Total 128.91 29.32 NAS 82.16 17.20 PI 48.10 13.12 AVL-AV physical 7.20 3.23
HAB Balance Point1 7.80 1.75
Assessment of Statistical Assumptions
For the first hypothesis a Shapiro-Wilk test was performed to check the assumption of normality for the distribution of scores on the aggression questionnaires at pretest (OSAB, NAS-PI, RPQ, AVL-AV physical) and scores on balance point on day 1 (HAB) of the intervention week. The scores at pre-test on OSAB, W(95) = .817; RPQ, W(95) = .939, p<.05 and balance point at day 1, W(95)= .960, p< .05 were non-normally distributed. Scores at pre-test on NAS-PI, W(95) = .975, p > .05 and AVL-AV physical, W(95) = .991, p>.05 were normally distributed. Since the variable ‘balance point at baseline’ had violated the parametric assumption of normally
distributed data, Spearman's rank-order correlations were run to assess the relationship between scores on all the aggression questionnaires and balance point on day 1.
For hypothesis two, there was homogeneity of variances (p > .05) as assessed by Levene's test of homogeneity of variances. Mauchly’s test indicated that the assumption of sphericity had been violated for the main effect of Time, χ2(9) = 35.07, p < .0005. Therefore, degrees of freedom were corrected using Greenhouse-Geisser estimates of sphericity (ε = .83).
When testing the third and fourth hypothesis, one outlier in the data was found. This outlier had a studentized residual value of 3.83, which is greater than the maximum of 3.29. The outliers represented a participant who scored almost the maximum amount of points on each NAS-PI questionnaire. This participant’s score might not be a valid representation of reality. Therefore, this outlier was removed from the analysis. This resulted in a total of 77 participants for hypothesis three and four, with N=38 in the experimental condition and N=39 in the control condition. There was homogeneity of variances (p > .05) and covariances (p > .05) as assessed by Levene's test of homogeneity of variances and Box's M test, respectively. Scores on the NAS-PI were normally distributed for the conditions on pre-, post- intervention and follow-up, as assessed by Shapiro-Wilk's test (p= .172; p= .071, p=.442 resp.). Moreover, Normal Q-Q assessed normally distributed data. Scores on OSAB were non-normally distributed for both conditions on each of the three time levels, as assessed by Shapiro-Wilk's test (p<.001). Mauchly's test of sphericity indicated that the assumption of sphericity was met for the main effect of Time, χ2(2) = 1.56, p = .46; χ2(2) = 5.821, p = .054
The assumption for homogeneity of regression slopes had been violated for the covariates ‘baseline aggression’ and ‘baseline reactive aggression’ (RPQ), as their interaction term with ‘condition’ was statistically significant, F(2,66)= 8.12, p<.005, partial η2 = .20 and F(2,66)= 12.04, p<.001, partial η2 = .27 resp. For the other covariates, all the other assumptions were met (homogeneity of regression slopes, homoscedasticity, homogeneity of variances and normality). 3.2 The Association between HAB and Aggression
The first hypothesis states that aggression is related to HAB. To test this, a correlation analysis was done. The Spearman’s rho analysis revealed a negative correlation between balance point at baseline and scores on NAS-PI, rs= -.29, p < .01, RPQ scores, rs= -.25, p < .05 and
AVL-AV physical, rs= -.39, p < .005. No significant correlation was found between balance point at
baseline and scores at baseline on OSAB, rs= .06, p=.59. These results imply that participants
with a higher HAB balance point (stronger tendency to recognize faces as ‘happy’ rather than ‘angry’) score lower on self-reported aggression questionnaires than participants with a lower HAB balance point (stronger tendency to recognize faces as ‘angry’). Since both the RPQ and NAS-PI exist out of two factors, another Spearman's rank-order correlation was run with these factors. There was a negative correlation between HAB balance point at baseline and scores on
Reactive Aggression items on the RPQ, rs= -.25, p < .05. There was no relationship between
HAB balance point at baseline and Proactive Aggression items on the RPQ, rs= -.18, p=.08.
These results suggest that HAB was related to more reactive rather than proactive forms of aggression. For the NAS-PI, both scores on NAS, rs = -.23, p < .05 and scores on PI, rs= -.30, p
< .01 were negatively correlated with balance point at baseline. These results suggest that participants with high HAB tendencies report more anger evoking situations and cognitive experiences of anger than participants with low HAB tendencies.
In conclusion, there was a statistically significant relationship between HAB balance point at baseline and scores on reactive aggression items of RPQ and scores on NAS-PI. There was no relationship between HAB balance point and staff-observed aggression.
3.3. Manipulation Check
The second hypothesis states that the intervention training will modify a shift in facial emotion recognition towards the happy- dimension over time in the Experimental condition. A Repeated Measure ANOVA was performed to compare the effect of Training Condition (independent variable) on shift in balance point (dependent variable) at five different time levels (day 1 – day 5) of the intervention week.
The Repeated Measures ANOVA revealed that there was a significant interaction effect between Time and Training condition, F(3.31, 261.61)=9.44, p<.001, partial η2=.12. Looking at the interaction graph (Figure 3), these effects reflect that shifts in balance point at time level one were significantly greater in the Experimental condition than in the Control condition. There was a significant main effect of Time on shift in balance point, F(3.31, 261.62) = 4.38, p<.005, partial η2 = .05. There was also a significant main effect of Training condition on shift in balance point, F(1, 79)=6.66, p<.05, partial η2= .08. This indicates that time had different effects on the shifts in balance point, depending on which condition a participant was assigned to.
Figure 2 and 3. Mean shifts in balance points on day one
to five of the intervention week and HAB balance points
on day one to five for the two conditions (experimental
Table 4
Means and Standard Deviations (in brackets) of HAB balance point on Day One (HAB1) to Five (HAB5) for the Experimental Condition (N=40) and Control Condition (N=41)
Training Condition HAB1 HAB2 HAB3 HAB4 HAB5
Experimental 7.89 (1.65) 9.77(1.97) 11.33 (1.73) 11.81(1.64) 12.18(1.51) Control 7.78 (1.74) 8.16 (1.92) 8.04 (2.38) 8.00 (2.82) 8.44 (3.12)
In the Experimental condition mean threshold on day five was 4.05, 95% CI [3.05, 5.05] higher than the mean threshold on day one, a statistical difference, p<.05. This means that there was a morph shift of 4.05 (+/- four faces) in the Experimental Condition over 5 days of training (see Figure 3 and Table 4). As seen in the graph of Figure 3, the shift in the Experimental Condition gets smaller each day. This can be explained by a ceiling effect of the training. If a participant would have rated all ambiguous faces as ‘happy’ instead of ‘angry’, the next step would be to rate angry faces as ‘happy’ as well, considering he receives feedback that is directed 2 morph shifts towards the happy continuum away from each day’s balance point. Therefore, there is a limit to which participants can shift in their balance point. It is concluded that the manipulation was successful; there was a shift in HAB balance point in the Experimental condition was due to the effect of the training.
3.4a Effectiveness Intervention - Self-reported Aggression
In the third analysis, the effect of condition in the intervention week on self-reported aggression on the NAS-PI was tested in participants over time. A two-way mixed ANOVA was performed to calculate the effect of "training condition" and "time" and their interaction. Means and standard deviations of scores on NAS-PI are shown in table 5.
Table 5
Mean scores on NAS-PI on and Standard Deviations (in brackets) for Experimental Condition (N=38) and Control Condition (N=39) at Pre-test (NAS-PI 1), Post-test (NAS-PI 2) and Follow-up of 6 Weeks (NAS-PI 3)
Training Condition NAS-PI-1 NAS-PI 2 NAS-PI 3
Experimental 130.21 (23.23) 124.26 (20.98) 123.72 (18.90) Control 122.58 (31.81) 121.55 (27.71) 115.58 (22.57)
In contrast with the expectations, the results of the two-way mixed ANOVA revealed that there was no statistically interaction between training condition and time on self-reported aggression (NAS-PI), F(2, 150) = 1.221, p = .30, partial η2 = .016. There was a significant main effect of Time, F(2, 150)= 6.167, p < .005, partial η2 =.076, there was no main effect of Training Condition, F(1, 75)= 1.44, p= .234, partial η2 =.02. This means that the effect of Time on self-reported aggression, did not depend on the effect of Training Condition, e.g. participants reported less aggression on the NAS-PI on time point 3 than at time point 1, no matter which condition they were in. Mean self-reported aggression score (NAS-PI) at time level one (pre-test) was 6.76, 95% CI [2.13, 11.4] higher than at time level three (after 6 weeks follow-up), a statistically significant difference, p < .005. As expected, in the Experimental condition, there was a significant difference in self-reported aggression directly post-training, compared to pre-training (p<.05). Mean self-reported aggression in the Experimental condition at pre-training was 5.95, 95% CI [.50, 11.40] higher than at post-training. There was no significant difference between the other time points. This effect is seen in Figure 4, where the line of the Experimental condition first drops and then stabilizes. Interestingly, scores on self-reported aggression in the Control condition changed significantly (p<.05) at 6 weeks follow-up. Mean self-reported aggression in the Control Condition at pre-and post-training was resp. 7.00, 95% CI [1.27, 12.73] and 5.98, 95% [.91, 11.03] higher than at 6 weeks follow-up. This corresponds with the graph as seen in Figure 4, where the line of the Control Condition is descending towards the end. This could be explained by a possible placebo-effect of the training.
It can be concluded that a significant decrease in self-reported aggression on the NAS-PI was found after six weeks follow-up for all participants, indicating a main effect of time on self-reported aggression (see Figure 4). However, effect sizes were small (η2 =.076 for time, η2 =.016 for interaction time*condition).
Figure 4. Mean scores on NAS-PI for the Experimental condition and Control condition on the three time level; pre-test(1), post-test(2) and at 6 weeks of follow-up(3)
Furthermore, analysis of covariance had been performed. The covariates included in the two-way mixed ANOVA were age and mild intellectual disability (ID). After adjustment for ‘age’ and ‘mild ID’ the two-way mixed ANCOVA revealed a statistically significant main effect of time,
F(2, 148) = 4.80 p< .05, partial η2 = .06. No interaction was found between the covariate age * training condition *time and mildID*condition*time, p>.08 and p> .10. So far, it can be concluded that the covariates did not influence the effect of Time on self-reported aggression on NAS-PI for both conditions.
3.4b Effectiveness Intervention - Staff-rated Aggression
For the last hypothesis a two-way mixed ANOVA was performed to assess the effect of training condition on staff-rated aggression in participants over time. Means and standard deviations are shown in table 6.
Table 6
Mean scores on staff-rated aggression scale (OSAB) and Standard Deviations (in brackets) for Experimental Condition (N=41) and Control Condition (N=43) at Pre-test (OSAB 1), Post-test (OSAB 2) ad Follow-up OSAB 3)
Training Condition OSAB 1 OASB 2 OSAB 3
Experimental condition 10.57 (2.59) 9.83 (2.74) 10.13 (2.44)
Control condition 10.20 (2.75) 10.10 (2.91) 10.51 (3.89)
Despite this violation of normality, it was decided to run the two-way mixed ANOVA regardless. In contrast with the expectations, the results of the two-way mixed ANOVA revealed that there was no statistically interaction between training condition and time on staff-rated aggression (OSAB), F(2, 160) = 1.336, p = .266, partial η2 = .016, and that there was no significant main effect of time, F(2, 160)= 2.638, p = .075, partial η2 .032. The mean scores on staff-rated aggression between groups were not statistically different, F(1, 80)= 9159.8, p< .890. However, when looking at contrasts for the main effect of Time, there is a statistically significant difference between OSAB scores at pre-test and at post-test for all participants. The mean of OSAB at pre-test was .42, 95% CI [.04, .80] higher than at post-test, p< .05. This corresponds with the lines as shown in Figure 5. The lines of both conditions drop down from time level on to two. This could mean that there is a direct result of the training on staff-rated aggression for both groups. However, the effect sizes were small (Cohen’s d=.32).
Figure 5. Mean scores on OSAB on the three time levels pre-test (1), post-test (2) and 6 weeks follow-up (3) for the experimental and control condition
After adjustment for age and mild ID, there was still no statistically significant main effect of time nor condition, p >.05. There was no interaction between the covariates *condition*time, p> .05.
In summary, there were no main effects of time nor condition on staff-rated aggression on the three time levels, pre-test, post-test and six weeks follow-up (p>.05) and there was no interaction between time and training condition (p>0.05). However, there was a significant difference in staff-rated aggression between pre- and post-test, indicating less staff-reported aggression for all participants. These results remained after controlling for age, ID and (mild) brain injury. Therefore, the alternative hypothesis that the experimental condition improved significantly more/ scored lower than the control condition after the intervention week on staff-rated aggression cannot be accepted.
3.5 Exploratory analysis
An exploratory analysis was done to investigate the association between changes in self-reported aggression (NAS-PI) and changes in HAB. A Spearman’s rho correlation analysis revealed that there was no significant correlation between changes in HAB and changes in scores on NAS-PI between pre-test and post-test of six weeks for the experimental condition, rs= .113,
p= .34 and control condition, rs= .227, p= .18. Contrary to our predictions, the finding of
decreased self-reported aggression (hypothesis 3) for all participants was not associated with changes in HAB balance point.
4 DISCUSSION Summary of Results
This study analyzed the effectiveness of a new aggression intervention program (Penton-Voak et al., 2013). The intervention program’s goal was to reduce aggressiveness in violent offenders, by making changes in emotion processing via a facial affect recognition task (Penton-Voak et al., 2013; Lundqvist, Flykt & Öhman, 1998; Tiddeman, Burt & Perrett, 2001). To accomplish this, a study population of 95 Dutch incarcerated violent offenders was used. This is the first longitudinal study that has tested the intervention on a prison population.The research questions (‘Is a computer based facial-affect recognition task able to make changes in HAB and moreover, make changes in cognitive and behavioral aggression?) was divided over four hypotheses that together cover a valid answer. These results are discussed below.
One of the main findings was that HAB was related to different types of aggression. Both physical and reactive aggression were related to HAB, as established through the correlations between HAB balance point and scores on the AVL-AV physical aggression and RPQ-Reactive aggression. However, it must be noted that these correlations were weak. Nevertheless, these correlations correspond with the results of previous research in which participants with as stronger tendency to interpret other person’s actions as hostile showed significant more expressions of physical aggression (Helfritz-Sinville & Stanford, 2014, Godleski and Ostrov, 2010) in comparison to a control group without these tendencies. As for HAB and the proactive/reactive debate, the findings of the current study support the early work of (Dodge and Coie, 1987, Schwartz et al. 1998, ref. Helfriz-Sinville and Stanford, 2014), as they indicate no association between proactive aggression and HAB. However, it must be noted that proactive forms of aggression were infrequently reported in this study population. Moreover, scores on both the NAS and PI were correlated to HAB. The NAS-PI gives a meaningful understanding of the cognitive mechanisms that underlie aggression, e.g. the cognitive experience of aggression and anger provoking factors. This correlation says something about the cognitive elements that are related to HAB, therefore, it helps creating a deeper understanding of HAB.
The manipulation check demonstrated a significant shift in HAB balance point for the experimental condition, with a total of 4 morph faces. These results corresponds with the studies of Penton-Voak et al. (2013). As mentioned above, one important difference between these studies and the current one is the study population, as Penton-Voak (2013) used a general population of adults and adolescents being at high risk of committing a crime. The results of the current study support the idea that the training is applicable to a broader population, supporting the generalizability of the training. It lays the groundwork for examining this training in forensic populations. The manipulation check demonstrated a significant shift in HAB balance point in the experimental condition, with a total of 4 morph faces. These results corresponds with the studies of Penton-Voak et al. (2013), where an effect size of .36 was found. As mentioned above,
one important difference between these studies and the current one is the study population, as Penton-Voak (2013) used a general population of adults and adolescents being at high risk of committing a crime. The results of the current study support the idea that the training is applicable to a broader population, e.g. adult man (age 20 to 75) possible mild ID, supporting the general utility of the training. Moreover, it lays the groundwork for examining this training in forensic populations.
An overall effect of time on self-reported aggression on the NAS-PI for all participants was found, this remained after controlling for age and mild ID. This effect was evident at the eleven weeks of follow-up, indicating long-term change in self-reported aggression. No main effect of training condition and no interaction with time was found and therefore, it can be concluded that the decrease in self-reported aggression was not due to changes in HAB. This is supported by the exploratory analysis, in which no correlation appeared between changes in HAB and changes in self-reported aggression. Lastly, staff-observed aggression (OSAB) did not significantly decrease over the eleven weeks of study, however, there was a decrease for all participants two weeks after the training. These results remained when controlling for age and mild ID.
Interpretation of Results
Thus, all participants reported less aggression during the eleven weeks of the study and the staff observed less aggressive behavior in the first week after the study, for all participants, but this effect disappeared after six weeks follow-up. How should these results be interpreted? Since both groups benefited from the training and the changes in self-reported aggression were not related to changes in HAB, another factor should account for this effect. One possible explanation is that all participants believed they would improve due to the experiment and therefore reported less anger and aggression, comparable to a placebo-effect. Thinking you will improve, will actually cause improvement. It was noticed that mainly all men were happy to participate, at a post-test evaluation interview they reported to believe the training was designed to help them rehabilitate. However, they weren’t aware of the operation of the intervention itself. Evidently, the participants were motivated and hopeful for the training to establish beneficial changes in their behavior and current emotional state (the study was presented as an ‘emotion training’). This is plausible, considering the sometimes hopeless situation these prisoners are in. Another possible explanation is that their answers were biased by a tendency to respond socially desirable. This is highly probable in a criminal justice setting, in which offenders are constantly observed by juridical authority. In this case, participants would give inaccurate answers on anger-sensitive topics in order to present themselves in the best possible light. This might also explain the overall low scores on the questionnaires (floor-effect), e.g. low scores on the proactive aggression scales. Admitting to have proactive aggressive tendencies is not socially desirable. Lastly, there might be a lack of self-insight for participants in their own (violent) behavior. As stated
earlier, individuals with pathological aggression may experience or rationalize their violence or aggression as being within the boundaries of normal protective or defensive aggression (Siever, 2008). This could also explain the incongruence between self-reported and staff-observed aggression.
As for the scores on the staff reports, it was concluded that these scores were non-normally distributed and a great deal of the scores were close to the zero point of the scale, implying a floor-effect. These scores imply a floor-effect and this has negative consequences for the test’s validity. Several biases in staff-observations might lead to this effect. Besides the risk of an observer-expectancy bias to occur, which is almost inevitable in observational research methodology, there were two main causes to the problem. First, there were practical issues. Due to understaffing and displacement, it was problematic for the staff to make representative observations of the general behavior of a participant during a particular week. Moreover, understaffing resulted in miscommunication with staff and therefore it was hard to include them in the study and motivate them that their observations were of great value. There was a lot of of missing data due to these practical complications, which has a negative impact on the reliability. Second, there are certain standards of conduct in prison and if these are violated, there are (negative) consequences. Therefore, instinct behavior is often suppressed by prisoners and aggressive acts might therefore not be observable. This could explain the low scores and floor-effect of the OSAB.
The results of the current study are inconsistent with those found in the study of Penton-Voak et al. (2013), in which young high risk offenders showed significant changes in aggression as measured through self-report questionnaires and staff-observation scales within two weeks after the training. A possible explanation is the difference in aggression assessment; staff-observed and self-reported aggression was measured using diaries in the study of Penton-Voak et al. (2013), this is an unstructured way of reporting and might therefore be more subjective than a structured questionnaire used in the current study. Moreover, these differences in assessment instruments make it difficult to compare results between the studies.
Limitations and Future Recommendations
There were some limitations to the current study design. First, as stated before, the self-measurement instruments were highly prone to all kinds of biases (social desirability, lack of insight, floor-effects etc.). These biases might have an impact on the validity and reliability of the test scores. A possible solution for this is to measure aggression on a more implicit level, for example, aggression could be measured through social role play or social skill evaluation tasks. Aggressive behavior could be evaluated by experienced behavioral experts.
There are some limitations to the computer task used in the current study to measure HAB. First, only one type of male face was used in the task, this was a computer generated face. Therefore, HAB was tested and modified solely on this type of face. It is unclear
whether responses of participants would differ between types of faces, as seen in everyday life, such as female, young old, dark, white faces etc. Moreover, responses might be culture dependent. Therefore, it is important to include all kinds of faces, in order to make the task more generalizable to the real world. Second, the computer task in the current study was tested on an offender population only. Therefore, it is unclear how a control group of non-violent offenders would score on the task and on measures of HAB. If ambiguous faces were to be valued as ‘angry’ by a control group as well, the computer task might not be able to detect HAB and therefore, modifying HAB via this task would not succeed in actual modifications of HAB. That is, modifying HAB with this computer task, might not alter actual hostile biases towards human faces. Finally, it was noticed that the computer task had a very long duration for all participants. At the end of the training, participants lost their focus and therefore participants might not perform as well as they could have. In further research, the task might be more effective if the duration was shorter or more short pauses were included. Further, the participants did receive feedback from the computer, but they could learn to perform the way the computer wants them to perform. Therefore, the task might not cause an actual learning effect. Future studies should focus on other feedback methods, such as psycho-education from a psychologist or other professionals. In conclusion, the duration and effectiveness of the computer task training is questionable and researchers should look at solutions for problems such as generalizability,
culture influence, duration and the learn effect. When
looking at the population and the descriptive characteristics as seen in Table 1, it is concluded that a lot of participants used medication. Anti-psychotics and mood-stabilizers were used in 20 % of all participants, indicating psychotic and mood disorders were noticeably common. Considering these medication influence emotions and affect, their influence on the recognition of and response to facial emotion expressions might be affected by this. Further, emotion regulation problems are very common in personality disorders (especially cluster B), however, diagnostic information was not included in the current study. Future research should include diagnostic information on scale one and two of the DSM-V and medical information in order to investigate their influences on computer task responses.
This study was not able to establish long-term behavioral changes in an offender population. However, the training probably did cause cognitive changes. Moreover, practical implications should be taken in consideration. First, future studies could focus on intervention strategies that combine this training in combination with other intervention methods. This training could be used to create awareness and insight in cognitions and emotions of anger. This awareness is the groundwork of any treatment that focusses on the prevention of aggression and anti-social behavior in high risk offender populations. Second, the training turned out to be user friendly in this population, in which almost all offenders were motivated. This leads to hopeful
