Aggression in the Face of Trauma and the Role of Hostile Attention Bias in a Sample of Psychiatric Patients ()
1. Introduction
Violence is a major problem in all socioeconomic and demographic groups with widespread societal costs with regard to economics, health and social functioning (World Health Organization, 2002). Globally, it is estimated that one out of two children aged 2 - 17 years, experience some form of violence each year (Hillis et al., 2016; World Health Organization, 2020). Thirty-five percent of women worldwide have experienced intimate partner violence in their lifetime (World Health Organization, 2019).
Aggressive behaviors are highly prevalent in long-term psychiatric inpatient care. De Bles et al. (2020) reported five incidents per day in an average psychiatric ward with 20 inpatients, whereby the annual direct cost for the setting would amount to €140,000 annually. Physical, mental and behavioral health consequences can persist long after the violence has stopped (World Health Organization, 2012). Reduction of behaviors on the violence-aggression continuum is highly relevant, given the detrimental societal and health consequences (Allen & Anderson, 2017).
The causes and mechanisms of aggression are complex and manifold and there remains no clear consensus among therapists and researchers on the best way to treat angry patients (Glancy & Saini, 2005). Research to further elucidate mechanisms underlying aggression is therefore crucial, because it may lead to the designment of more adequate, tailored and effective prevention and treatment programs for aggressive patients (Klein Tuente et al., 2019).
Violence and aggression often occur in patients with symptoms of posttraumatic stress disorder (PTSD) (LaMotte & Taft, 2017; Taft et al., 2012). This association is found in various populations and established for several components of aggression among trauma exposed adults (Luijkx et al., 2024; Orth & Wieland, 2006; Taft et al., 2011). PTSD symptoms predicted subsequent levels of anger (the emotional component of aggression), but anger did not predict subsequent PTSD symptoms in a sample of crime victims (Orth et al., 2008; Semiatin et al., 2017; Shorey et al., 2018). Traumatic childhood experiences may lead to a non-psychopathic aggressive personality, that is marked by impulsivity and reactive aggressive behavior (Cima et al., 2008). Therapeutic interventions like trauma-focused treatment and reducing hostile attribution bias are considered as promising avenues for improving clinical treatment of aggressive patients (LaMotte & Taft, 2017; Lobbestael et al., 2013). However, more rigorous testing of hypotheses that aim to explain the relation between PTSD and anger is needed (Orth & Wieland, 2006; Semiatin et al., 2017).
For more than forty years, a prominent theoretical model explaining the relation between anger and PTSD is the survival mode theory (Chemtob et al., 1997; Chemtob et al., 1988). This theory states that the perception of threat activates a biologically predisposed survival mode, including both fear and flight reactions and aggressive responses, such as anger and fight reactions (Orth & Wieland, 2006). Aggressive responses are considered to be the result of several sequential steps in processing social information. Within these steps, inferring hostile intentions to the behavior of other persons is seen as a key component in increasing the likelihood to engage in aggressive behaviors (Calvete & Orue, 2012; Klein Tuente et al., 2019). Hostile attribution of intent, also called hostile attribution bias, is the phenomenon in which individuals tend to see hostile intent in others, despite the fact that environmental cues do not clearly support such an intent (Dodge, 1980; Dodge & Pettit, 2003).
Angry facial expressions are primarily perceived as threatening (Marsh et al., 2005) and thereby one of the most well-established nonverbal signals of hostile intent (Wilkowski & Robinson, 2012). Research suggests a robust relation between hostile attribution of intent and aggressive behavior in child studies (Orobio de Castro et al., 2002) and small to medium associations in a study among adults (Klein Tuente et al., 2019).
The hostile attribution bias was reinterpreted by Wilkowski and Robinson (2012) introducing the concept of perceptual sensitivity based on observations in students. This leads to the hypothesis that physically aggressive individuals are more sensitive to subtle differences in facial anger than non-aggressive individuals (Qiu et al., 2016; Wilkowski & Robinson, 2012). In other words, aggressive individuals may not be more biased but are more perceptually sensitive to hostile cues in ambiguous situations. Thus, what may appear to be a bias is actually a finely tuned skill. These results however have been observed in relatively healthy populations. It is not sure whether these processes also occur in psychiatric populations. Schönenberg and Jusyte (2014) for example, found that aggressive antisocial violent male prisoners (mis)interpreted ambiguous facial cues as hostile and showed a strong tendency to systematically overrate the perceived intensity of anger. This raises the question to what extent the findings of perceptual sensitivity to hostile cues in a student population can be generalized to patient populations, especially those with more extreme levels of aggression than students.
Moreover, Wilkowski and Robinson (2012) suggested an association of perceptual sensitivity to facial cues of anger and aggression with the existence of trauma. This is in line with literature indicating that the existence of trauma in childhood seems to influence the ability of adults to recognize facial expressions (Catalana et al., 2020). However, trauma symptoms were not measured in the study of Wilkowski and Robinson (2012). Williams et al. (2018) noticed that surprisingly little research has examined the relationship between current trauma symptoms and the perceptual sensitivity to subtle facial cues of anger in adult samples, and that the results of the few studies examining the relation between PTSD and facial affect recognition have been inconsistent, with a tendency to find PTSD being associated with poorer performance. For a better understanding of the association between trauma symptoms and the perceptual sensitivity to facial cues of anger, more research is needed, especially in patient populations.
For the first time, we therefore explored the relationship between trauma symptoms, the perceptual sensitivity to subtle facial cues of anger, and aggression in a population of psychiatric patients. We hypothesize that trauma symptoms affect aggression through its effect on perceptual sensitivity to subtle facial cues of anger. More specifically, the relation between trauma symptoms and the levels of aggression may be mediated by perceptual sensitivity to angry faces.
2. Methods
2.1. Participants
Eighty-three patients from two mental health institutions in Netherlands (25 female, 58 male) participated on a voluntary base. The mean age was 40 years (SD = 10.79, range = 21 - 77). Sample demographics are presented in Table 1. Fifty-five percent of the participants were patients with a forensic legal status, forty-five percent of the participants were receiving treatment in a clinic for patients with addiction, trauma and personality disorders. To classify the level of education, the Dutch Verhage scale was used (Verhage, 1964). Its seven categories were merged to three ordinal categories: low (Verhage 1 - 3), middle (Verhage 4 - 5), and high educational level (Verhage 6 - 7). The level of education of participants varied from low (19%) to middle (63%) and high (18%). The vast majority of participants had the Dutch nationality (93%). Most of the participants were unmarried, widowed or divorced (93%), and not employed (86%).
Table 1. Sample demographics.
Variable |
|
Total (N = 83) |
Gender |
Male |
58 (70%) |
Female |
25 (30%) |
Setting |
Forensic mental health care |
46 (55%) |
Mental health care |
37 (45%) |
Education |
Low |
16 (19%) |
Medium |
52 (63%) |
High |
15 (18%) |
Marital status |
Married/common law |
6 (7%) |
Divorced/widowed |
15 (18%) |
Unmarried |
62 (75%) |
Nationality |
Dutch |
77 (93%) |
Other |
6 (7%) |
Work status |
Employed |
12 (14%) |
Unemployed |
71 (86%) |
2.2. Stimuli
For the measurement of perceptual sensitivity to subtle facial cues of anger we used stimuli created by FaceGen Modeller, a software program that provides images of people in which facial expressions can be adjusted in emotional intensity. Both stimuli with facial cues of anger and stimuli with facial cues of happiness were included, as this study is part of a larger project. The faces displayed 40%, 50%, or 60% of either anger or happiness, with the remaining percentage left emotionally neutral. Stimuli created by this program have shown to be both valid and to provide adequate experimental control for research purposes. Wilkowski and Robinson (2012) used the stimuli in their study and characterized the faces as non-Hispanic Caucasian and gender neutral in appearance. Participants completed two blocks of trials, one involving 60 mixtures of anger and neutral expressions and the other involving 60 mixtures of happy and neutral expressions. Each block consisted of twenty exemplars of each blend (i.e., 40% emotional, 50% emotional, 60% emotional) (See Figure 1).
![]()
Top row faces display (from left to right) 40% anger, 50% anger, and 60% anger. Bottom row faces display (from left to right), 40% happiness, 50% happiness, and 60% happiness.
Figure 1. Example stimuli (Wilkowski & Robinson, 2012).
For each block, one of four versions was given to the participant. The version of the block, the order in which the participants completed the blocks, and the order of the response columns (emotional-neutral, or neutral-emotional), were randomly assigned to the participants to prevent biases. Participants were asked to indicate which expression each face displays in a two-alternative forced-choice task (i.e., anger vs. neutral for the anger/neutral blends; happy vs. neutral for the happy/neutral blends). Participants clicked at their own pace through the stimuli presented on a laptop, and crisscrossed their perceptions with a pen on a paper.
2.3. Measures
For the measurement of symptoms of trauma and aggression we used two self-report questionnaires.
The Trauma Screening Questionnaire (TSQ) (Brewin et al., 2002) is a 10-item instrument consisting of five re-experiencing and five arousal items from the DSM IV (American Psychiatric Association, 2000) PTSD criteria. The TSQ achieved very good performance in a systematic review of screening instruments for adults at risk of PTSD (Brewin, 2005). The Dutch version of the Trauma Screening Questionnaire is an effective screening instrument to distinguish between subjects with PTSD, without PTSD and with subthreshold PTSD (Dekkers et al., 2010). According to Dekkers et al. (2010) sensitivity and specificity of the Dutch version of the TSQ are balanced best at a cut-off score of 7, although a cut-off score of 6 is also suggested (Brewin et al., 2002). Participants were asked whether they had experienced each symptom at least twice in the past week.
Furthermore, participants completed the Dutch version of the Buss-Perry Aggression Questionnaire-Short Form (AQ-SF; Bryant & Smith (2001)/AVL-AV; Hornsveld et al. (2009)). Participants indicated how accurately each of the 12 statements described them using a 1 (totally disagree) to 5 (totally agree) response scale. Scores on the scale may range from 12 to 75. The four subscales measure different components of aggression: Physical Aggression, Verbal Aggression, Anger, and Hostility. The AQ-SF is a reliable and valid instrument to measure components of aggressive behavior in university students of both genders (Meesters et al., 1996) and has adequate reliabilities in a sample of male and female federal offenders (Diamond & Magaletta, 2006). In a student sample, Cronbach’s alpha for the total score was .84, indicating good internal consistency (Meesters et al., 1996). Construct validity of the Dutch version of the AQ was demonstrated in adolescent male offenders (Morren & Meesters, 2002). The AQ-SF is a reliable and valid instrument with even better psychometric properties than the original full-length version, which is found in a sample of Dutch violent forensic psychiatric male patients (Hornsveld et al., 2009). Internal consistency coefficients (Cronbach’s alpha) for the total score of the AQ-SF were between .72 and .88.
2.4. Procedures
Participants were recruited among patients of two mental health institutions in Netherlands. Patients were verbally informed about the study during a visit of the researcher in their treatment program, and received detailed information about the study on paper. Participants were ensured that all their data would only be stored anonymously. Volunteers willing to participate provided their contact details and were invited by the researcher afterwards for an individual appointment to participate. Participants signed the informed consent. Subsequently, they completed the facial expression perception task. Once the participants completed the facial expression perception task, demographic information was collected by a short interview, and afterwards participants completed the TSQ and the AQ-SF. Tasks were always completed in this order. The procedure was followed in an individual face-to-face contact with the researcher. Two participants were seen by a trainee. Due to COVID-19 provisions, two participants completed the questionnaires by video calling.
2.5. Ethics
Ethics approval was provided by the Commission Scientific Research of Foundation Mental Health Care Western North-Brabant, Netherlands. This study is part of a larger study design for which ethics approval was granted by the Medical Ethics Committee of Maastricht University (protocol no. NL60187.068.17). Given COVID-19 safety measures, we were limited in our abilities to recruit participants. Preliminary analysis indicated that our hypotheses would not be confirmed. Out of ethical considerations we therefore decided to stop the data collecting at 83 participants.
3. Results
We analysed the data with SPSS, version 26, IBM Corp. (2019). We reported all manipulations, measures, and exclusions in the study. Prior to analyses, the accuracy of data entry was examined by comparing double entered data. FaceGen scores were computed by dividing the sum of item scores by the number of completed items, in order to correct for four missing items on the FaceGen Angry-Neutral and three missing items on the FaceGen Happy-Neutral. Two forensic male participants had not completed the TSQ. For one of them, the TSQ-items were scored as zero because he reported that he has never experienced any traumatic event. Testing for outliers in FaceGen scores yielded no significant outliers defined by z-scores more than 3.29 standard deviations beyond the mean, following Tabachnick and Fidell (2013). One case was identified with a multivariate outlier (FaceGen Angry-Neutral score 5%, FaceGen Happy-Neutral score 85%, p ≤ .001) and was excluded from analysis for that reason. A double scored answer on the AQ-SF (answer 2 and 4) was replaced by answer 3 (“don’t know”). Table 2 presents mean scores and standard deviations of the FaceGen stimuli and the questionnaires.
Table 2. Means and Standard Deviations of FaceGen stimuli and measures.
Stimuli/Measures |
Subschale |
Total (N = 82) |
M |
SD |
FaceGen A-N |
Total |
37.24 |
12.26 |
60% angry |
49.23 |
15.87 |
50% angry |
35.87 |
15.07 |
40% angry |
26.63 |
16.34 |
FaceGen H-N |
Total |
48.16 |
12.59 |
60% angry |
74.70 |
16.17 |
50% angry |
48.54 |
17.12 |
40% angry |
21.28 |
13.38 |
AQ-SF |
Total |
28.48 |
8.12 |
Physical |
7.34 |
3.37 |
Verbal |
6.05 |
1.98 |
Anger |
7.07 |
3.03 |
Hostility |
8.01 |
3.42 |
TSQ* |
Total |
4.37 |
3.28 |
Re-experiencing |
2.14 |
1.86 |
Hyperarousal |
2.23 |
1.70 |
Note: AQ-SF = Aggression Questionnaire-Short Form; FaceGen A-N = FaceGen Angry-Neutral (% angry faces); FaceGen H-N = FaceGen Happy Neutral (% happy faces); TSQ = Trauma Screening Questionnaire. *TSQ: N = 81.
Preliminary analyses were performed to check for violation of the assumptions of linearity, homoscedasticity, and normality of estimation error. The FacGen A-N score, D (82) = .093, p = .077, the FaceGen H-N score, D (82) = .082, p = .200, and the AQ total score, D (82) = .068, p = .200, did not deviate significantly from normal; however, TSQ total, D (81) = .125, p ≤ .01 was significantly non-normal. Also the subscales of the TSQ; TSQ arousal, D (81) = .141, p ≤ .001, TSQ re-experiencing, D (81) = .161, p ≤ .001, and the subscales of the AQ; AQ physical, D (82) = .120, p ≤ .01, AQ verbal, D (82) = .193, p ≤ .001, AQ anger, D (82) = .156, p ≤ .001, AQ hostility, D (82) = .116, p ≤ .01, deviated significantly from a normal distribution. For TSQ scores and AQ-SF subscales, non-parametric measures were used. Coefficients between the stimuli, measures and their subscales were computed and are presented in Table 3. The relation between trauma symptoms (as measured by the total score of the TSQ) and aggression (as measured by the total score of the AQ-SF) was investigated using Spearman’s rank-order correlation. There was a small, positive correlation between the two variables r (82) = .22, p < .05.
Table 3. Correlations table.
|
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
1. FaceGen A-N |
|
|
|
|
|
|
|
|
|
2. FaceGen H-N |
.269* |
|
|
|
|
|
|
|
|
3. AQ-SF (total) |
.104 |
.015 |
|
|
|
|
|
|
|
4. AQ-SF Physicala |
.022 |
.026 |
.640** |
|
|
|
|
|
|
5. AQ-SF Verbala |
−.044 |
−.024 |
.652** |
.238* |
|
|
|
|
|
6. AQ-SF Angera |
.171 |
−.028 |
.712** |
.273* |
.365** |
|
|
|
|
7. AQ-SF Hostilitya |
.120 |
−.011 |
.730** |
.182 |
.449** |
.339** |
|
|
|
8. TSQ (total)a |
−.014 |
.137 |
.223* |
.016 |
.187 |
.098 |
.289** |
|
|
9. TSQ Re-experiencinga |
−.054 |
.120 |
.142 |
.004 |
.079 |
.045 |
.209 |
.926** |
|
10. TSQ Hyperarousala |
.037 |
.134 |
.275* |
.018 |
.262* |
.135 |
.342** |
.917** |
.704** |
aSpearman correlation instead of Pearson’s correlation. *Correlation is significant at the .05 level (2-tailed). **Correlation is significant at the .01 level (2-tailed).
To test for mediation, a series of regression models was estimated, following Baron and Kenny (1986). Since TSQ scores violated the assumption of normality of estimation error, the TSQ was entered as a dichotomous variable in the analyses (trauma: TSQ ≥ 7, no trauma: TSQ ≤ 6).
First, a standard multiple regression was performed between FaceGen Angry-Neutral as dependent variable and the presence of trauma symptoms as dichotomous independent variable. The variance in FaceGen Angry-Neutral scores explained by trauma was .0%, F (1, 79) = .022, p = .88. This already illuminates that the FaceGen is not the hypothesized mediator between TSQ-scores and AQ-scores. To complete the conditions for mediation, second, a standard multiple regression was performed to assess the ability of the presence of trauma symptoms to predict levels of aggression (AQ-SF). The variance in aggression explained by the presence of trauma symptoms was 5.8%, F (1, 79) = 4.836, p < .05. Third, a standard multiple regression analysis was performed between aggression (AQ-SF) as dependent variable and FaceGen Angry-Neutral scores as independent variable. The variance in aggression explained by FaceGen scores was 1.1%, F (1, 80) = .872, p = .35.
As the analyses did not confirm our hypothesis on mediation, we performed post-hoc analyses to check whether the results could be explained by heterogeneity within the population. Independent-sample t-tests were conducted to compare the aggression scores for males and females, as it is known that the relationship between gender and aggression is a complex one (Padgett & Tremblay, 2020). No significant difference between males and females in the total aggression score of the AQ-SF was found. There was a significant difference in hostility scores for males (M = 7.47, SD = 3.17) and females (M = 9.24, SD = 3.72; t (80) = −2.20, p = .03, two-tailed). The magnitude of the differences in the means (mean difference = 1.77, 95% CI = −3.36 to − .17) = was moderate (eta squared .057). No significant gender differences on the other subscales of the AQ-SF were found. Previous studies on PTSD have found a significant gender effect with an approximately twofold higher rate for women than for men (De Vries & Olff, 2009). A Mann-Whitney U Test revealed a significant difference in reported trauma symptoms levels of females (Mdn = 8, n = 25) and males (Mdn = 3, n = 56), U = 1085, z = 3.96, p ≤ .0001, r = .43. No significant differences in FaceGen scores for males and females were found. Because of the broad inclusion of patients in our study, we explored the data by comparing some potential subgroups within the population. A Mann-Whitney U Test also revealed a significant difference in reported trauma symptoms levels of forensic patients (Mdn = 3, n = 45) and regular patients (Mdn = 5, n = 36), U = 595, z = −2.052, p = .040, r = .23. Regarding the scores on stimuli and measures, no other significant differences between forensic and regular mental health patients were found. The post-hoc analyses indicate that heterogeneity within the population could be relevant for trauma symptoms, but not for the hypothesized mediator.
4. Discussion
In this study, we explored the relationship between trauma symptoms, the perceptual sensitivity to subtle facial cues of anger, and aggression in a broad population of psychiatric patients. Our results demonstrate a cross-sectional association between trauma symptoms and aggression. However, perceptual sensitivity to subtle cues of facial anger measured by FaceGen stimuli did not mediate the association between trauma symptoms and aggression in our population. We discuss the four main outcomes of our study to contribute to a better understanding of the mechanisms underlying aggression.
First, the findings of our study indicate that the clear link between (physical) aggression and perceptual sensitivity to facial anger reported in students (Wilkowski & Robinson, 2012), is not necessarily generalizable to a patient population. This finding is reminiscent of the increasing interest in issues of replicability and generalizability of clinical science in recent years (Tackett et al., 2019). Braslow et al. (2005) found strong empirical support for the belief that mental health literature has paid scant attention to external validity. Therefore, they recommended researchers to study patients and context to which they wish to extend their findings, as we attempted partially by including a broad sample of psychiatric patients. Differences between students and patients may provide an explanation for the absence of a relation between self-reported aggression and perceptual sensitivity to angry faces in our population of psychiatric patients, whereas Wilkowski and Robinson (2012) had found a clear link in students. For example, it is known that traumatization often leads to schemas marked by Mistrust/Abuse (Rafaeli et al., 2011). Evidence has been found that, among others, schemas of Mistrust/Abuse uniquely differentiate (Borderline Personality Disorder) patients from healthy adults (Bach & Farrell, 2018). We recommend researchers and clinicians to take into account concerns about the generalizability of findings from one population to another.
Second, we found no relation between trauma symptoms and perceptual sensitivity to facial cues of anger. This result is in contrast with the findings of three recent studies. Passardi et al. (2018) found that (childhood) traumatization can be linked to enhanced emotion recognition abilities among adults. Other authors found that in addition to traumatic events that might culminate in PTSD symptomatology, all trauma occurring during childhood, including trauma types that would not qualify as a traumatic event based on criterion A of DSM-5 (PTSD) diagnosis, may have far-reaching consequences and entail greater distrust and threat perception later in life (Luijkx et al., 2024; Hepp et al., 2021). In a relatively healthy student population, Williams et al. (2018) found a clue that might help to explain the inconsistency between the results of those two recent studies among adults, and the results of our study. They suggested that the enhanced attention of people with higher levels of PTSD to affective information can be either beneficial or detrimental, depending on the precise circumstances, either various and diverse sources of information in their environment (Williams et al., 2018). Future research should include such other sources of information that may influence the performance of patients with trauma symptoms.
Given the abovementioned results, we also considered the concern of measurement errors in our study due to the use of FaceGen stimuli to measure perceptual sensitivity. For three reasons we do not see a clear indication that the unexpected lack of relation between scores on the FaceGen, trauma and aggression questionnaire should be explained by measurement errors due to the use of FaceGen stimuli. First, angry facial expressions are useful in research for a deeper understanding of the (perceptual) processes underlying aggressive behavior (Teige-Mocigemba et al., 2016). Second, in accordance with Wilkowski and Robinson (2012), faces in our study were endorsed as displaying an emotion (i.e., anger or happiness) on a higher proportion of trials as the intensity of the displayed emotion increased. Third, in line with Wilkowski and Robinson (2012) and Becker et al. (2011), emotion perceptions were more frequent in the happy/neutral block than the angry/neutral block. To explain that compared to angry faces, happy facial expressions are detected better when all potential confounds have been removed or controlled for Becker et al. (2011) speculated that given most people are happy most of the time (Diener & Diener, 2016), the detection of the happy face is more practiced.
Third, we noticed that the effect size between trauma and aggression we found was considerably smaller compared to the results of the meta-analysis of Orth and Wieland (2006). Literature suggests that the strength of the relation between trauma symptoms and aggression depends on the type of traumatic event. The relation appeared to be stronger in samples with military war experience than in samples with other types of traumatic events, such as criminal victimization (Orth & Wieland, 2006). We assume that the traumatic experiences participants of our study reported were relatively mild compared to those the participants in the meta-analysis of Orth and Wieland (2006) had experienced, or that the traumatic experiences concerned a different type of trauma. In our study, most participants self-categorized the types of events they had experienced as physical abuse or mental abuse, but they were not specified nor defined by Diagnostic Criteria A1 of the DSM-IV (American Psychiatric Association, 2000) as they were in the meta-analysis of Orth and Wieland (2006). Another explanation for the relatively small effect size we found is that the level of measurement of aggression in our study differs from the meta-analysis of Orth and Wieland. Buss and Perry (1992) stressed the need to assess overall aggression and its individual components; physical aggression, anger, hostility, and verbal aggression. We measured the relation between trauma symptoms and overall aggression, whereas the meta-analysis of Orth and Wieland (2006) concerned anger and hostility. Although trauma and aggression obviously are related, effect sizes seem to differ among samples and variables.
Fourth, the mean scores on aggression were not high in our study, despite our sample including criminal offenders. According to Holtzworth-Munroe et al. (2000) court-referred and clinical samples contain more generally violent man than a general community sample (Serie et al., 2017; Thijssen & De Ruiter, 2011). However, participants in our study did not perform differently from students in the study of Wilkowski and Robinson (2012) when comparing the mean scores of the subscale of physical aggression. This corresponds with the study of Hornsveld et al. (2009) in which inpatients were not found to display higher scores on the AQ-SF than students. They suggest that the restricted living environment of inpatients, which gives them lesser opportunities to exhibit aggressive or violent behavior, might explain this. It could also be explained by socially desirability since participants may be less open about their aggression within a treatment context (Taft et al., 2011). Moreover, Hornsveld et al. (2009) questioned the validity of comparing adult inpatients with adolescent students due to differences in age and a number of other variables such as marital status.
Noteworthy, the measure of trauma and aggression used in our study was based on self-report. Self-reporting of aggressive behavior relies heavily on the honesty of respondents about their tendency to become angry and behave aggressively, and on their recall of past events (Nijman et al., 2006). Patients with (severe) psychiatric disorders might lack insight into their own symptoms and behaviours. Despite this general belief that self-reports are questionable, it remains interesting that empirical literature suggests that self-report does not undermine the relationship between self-report and salient outcomes (Diamond & Magaletta, 2006; Mills & Kroner, 2005). Moreover, a recent Swedish study among incarcerated violent offenders found that self‐ratings and clinician‐ratings of aggression were highly convergent and concordant, especially regarding physical aggression (Berlin et al., 2021). Their results indicated that either self-reports and clinician-ratings yield such similar information that either alone would be sensitive enough. Furthermore, all our participants participated on a voluntary base without consequences for their treatment or forensic status.
All things considered, we assume that it is most likely that the unexplained direct relation between trauma symptoms and aggression indicates an omitted mediator (Zhao et al., 2010). Hypotheses and theories regarding the neurobiological and environmental determinants and their interaction of aggressive behavior, such as the influences of stress hormones, could be interesting to consider (Cima et al., 2008; De Wit-de Visser-et al., 2023; Tonnaer et al., 2016). A variety of mental disorders are characterized by deficits in facial emotional recognition, including schizophrenia (Gao et al., 2021) and dysthymia/depression (Krause et al., 2021). Literature demonstrates that high anxiety students can identify threat stimuli from faces more accurately and faster than low anxiety students (Mustapha et al., 2019). Taft et al. (2012) mentioned that information processing models for aggression highlight the role of additional factors such as alcohol use problems that may affect the traumatized individual cognitively and further increase risk for aggression and impaired information processing. Briefly, further research is needed to investigate whether comorbid symptoms such as mood or anxiety disorders or substance abuse mediate the relation between trauma symptoms and aggression.
Some limitations of the present study should be noted. First, the current study focused on PTSD symptoms rather than diagnoses. Thus, associations may have been inflated because PTSD symptom measures not keyed to specific trauma may be capturing other nonspecific distress in addition to true PTSD symptoms (Taft et al., 2011). Second, the use of only self-report data for the assessment of aggression and trauma may have led to uncontrolled bias and inflation or deflation of associations among study variables. Third, due to the cross-sectional nature of the data for this study, directionality between aggression and trauma symptoms cannot be assumed. Fourth, the ratio of male to female participants was heavily in favor of males, and there was a gender effect on trauma symptoms. We had not enough data to run models for males and females separately. Fifth, although our sample was highly diverse with respect to setting, we were underpowered to examine whether findings varied across certain settings. Sixth, since we did not classify patients according to the Diagnostic Criteria of the DSM-5 (American Psychiatric Association, 2013), use of medication or participation in psychotherapy, we were not able to explore differences among the different types of patients. Seventh, the emotional intensity levels (40%, 50%, 60%) in the FaceGen stimuli were not validated for this particular psychiatric population, therefore, we do not know whether perceptual thresholds might differ from the general population.
The present study leaves some potentially important avenues for future research. Future research should utilize more comprehensive measures to examine associations between different symptom clusters of PTSD (e.g., hypervigilance, avoidance), several components of aggression (general aggression, anger, hostility, verbal and physical aggression), and the perceptual sensitivity to emotional facial cues. For PTSD symptoms, a structured diagnostic interview is recommended. Besides that, it would be fruitful to include a more representative sample of men and women in future research. The issue that self-report questionnaires of aggression do not seem to discriminate between forensic psychiatric inpatient or inmate populations and control samples requires further study (Hornsveld et al., 2009). Future studies should consider designing the materials to be more ecological, such as photographed facial expressions of real people and small videos with contextual backgrounds to create stimuli as close to life as possible for patients, so that they can be more realistic and accurate for basic research or clinically targeted treatments (Gao et al., 2021).
Despite its limitations, we believe that this investigation adds to research to the mechanisms underlying aggression. We advise clinicians designing therapeutic interventions for aggressive patients to be careful in using the results of studies performed in student populations, because these may not fit into patient populations. More research to the complex and manifold mechanisms underlying aggression is needed, especially in populations of psychiatric patients.