The Mediating Role of Hostile Attribution Bias in the Relationship between Cluster B Personality Traits and Reactive Aggression: An Event-Related Potentials Study

Abstract

The aim of this study was to better understand the role of hostile attribution bias (HAB) in the relationship between cluster B personality traits and reactive aggression. Sixty-three French-speaking adults were asked to complete online questionnaires assessing their personality traits, hostile attribution bias, and aggressive behaviors. While brain activity was recorded, they were asked to read scenarios involving daily life interactions and to imagine why the characters (whose intentions were ambiguous) behaved in a provocative manner toward them. Following each scenario, we analyzed the N400 component of the event-related brain potential associated with the presentation of unexpected hostile or nonhostile intentions after each scenario. Results showed a stronger N400 amplitude during the presentation of unexpected nonhostile intentions (hostile expectancy violations) in the centro-parietal regions. There was no mediating effect of hostile or nonhostile expectancy violation in the relationship between cluster B personality characteristics and reactive aggression. Further studies are needed to better understand the psychological processes underlying aggressive behaviors in cluster B personality disorders.

1. Introduction

Personality disorders are conditions that can have a destructive impact on an individual’s quality of life and social interactions. Indeed, a person with a personality disorder will experience serious difficulties managing emotions, behaving according to culturally acceptable cognitions, and interacting normally in everyday life [1]. In the United States, 9–15% of people develop a personality disorder and in most cases, this disorder is accompanied by numerous comorbid conditions [2]. The situation is even more worrisome considering that personality disorders are commonly associated with aggression, violence, criminal behavior, and violent recidivism worldwide [3]. According to a systematic review, 65% of incarcerated men and 42% of incarcerated women have a personality disorder [4]. These epidemiological data underline the importance of better understanding and treating personality disorders, particularly those in cluster B. According to the DSM-5, cluster B personality disorders are characterized by relational disorders and impulsive, emotional, and/or unstable behavioral manifestations [1]. They include disorders such as antisocial personality disorder (ASPD) and borderline personality disorder (BPD) and tend to be strongly associated with a variety of maladaptive behaviors, including addictive, suicidal, or aggressive behavior.

BPD is characterized by pervasive instability of affects, self-representations, and interpersonal relationships [1]. It also includes the presence of impulsiveness, paranoia, feelings of emptiness, and/or suicidal gestures. BPD has a lifetime prevalence of approximately 6% among both sexes [5]. In a population of adolescents, BPD traits are associated with high levels of delinquency, antisocial behavior, and all forms of aggression (e.g., sexual harassment, overt aggression, and violence). As such, the diagnosis of borderline personality is, according to some authors, a good predictor of violence and aggression [6]. Moreover, in hospital settings, 65% of patients with BPD report having used physical, verbal, or relational gestures that were aggressive [7]. According to several authors, aggressive behaviors among borderline patients are guided by emotions [8]. In fact, BPD patients are prone to overreact, which leads to irritability, outbursts of anger, and subsequent physical aggression.

The DSM-5 describes ASPD as a pattern of violation of, and disregard for, the rights and interests of others [1]. It is expressed through a lack of social conformity, use of deception for personal gain, lack of remorse, and irresponsible, irritable, or impulsive behavior. In the United States, the prevalence of ASPD is 3.63% in the general population and the prison population, as high as 21–47% [4]. Moreover, in young adults, self-reports of two antisocial characteristics (i.e., sensation-seeking and egocentricity) have been associated with relational aggression [9]. More generally, ASPD diagnosed in clinical populations has been shown to be a strong predictor of violence and aggression [6]. Further, high levels of aggression have been associated with ASPD regardless of gender [3]. Authors suggest that violent behavior by antisocial patients can be explained as being part of an instrumental goal, such as for the purpose of obtaining gratification [8].

Conceptually, aggression refers to intentional and observable action directed toward someone with the goal of physically or mentally harming them [10]. Aggression is said to be reactive when it occurs under provocation, threat, or frustration. It is expressed through outbursts of uncontrolled anger and cognitive scripts involving distinct expectations and hostile perceptions. Aggressive behaviors have disastrous economic, legal, and social consequences [11]. Since the impacts are observable at the individual, family, community, and national levels, many programs have been developed to prevent and reduce aggression. One potential area of intervention could consist of decreasing aggressive cognitions that cause the individual to perceive the world as a dangerous environment and to reconsider the use of aggression when a conflict occurs.

Relatedly, a meta-analysis of studies conducted with people without BPD or ASPD showed a strong relationship between aggression and the hostile attribution bias (HAB) [12]. According to Crick & Dodge’s [13] theory of social information processing, HAB refers to a tendency to attribute hostile intentions to others despite the intention behind their behavior being ambiguous [14]. In more than 100 studies, the positive relationship between reactive aggression and HAB has been demonstrated in clinical and normal samples of individuals of different ages and ethnicities [14, 15, 16].

In an ambiguous and provocative situation, people with BPD tend to interpret events (such as abuse or rejection) as threatening [17]. This leads them to be overly sensitive to rejection, behave impulsively, and feel negative emotions. According to several authors, dysregulation of affect and behavior, which is characteristic of BPD, is associated with various cognitive biases, such as the HAB [18]. In fact, a study by Arntz et al. [19] found that people with BPD showed a readiness to perceive a person as negative, aggressive, malicious, abusive, and rejecting. Thus, it is quite possible that the HAB can explain why people with BPD act aggressively toward others. Smeijers et al. [20] have shown that patients with BPD often produce a lot of hostile interpretation biases.

With regard to ASPD, few studies have tested the HAB as an explanatory variable for reactive aggression [20, 21]. According to Lobbestael et al. [21], ASPD traits and HAB (measured using thumbnails and images) were good predictors of reactive aggression. Further Smeijers et al. [20] found that people with ASPD performed many HABs when looking at facial expressions.

The HAB can be measured using self-reports [22], written vignettes [21], video vignettes [23], and computer tasks [20]. For example, in the study by Lobbestael et al. [21], HAB was measured using eight images from the thematic apperception test and eight text vignettes describing ambiguous and provocative scenes from daily life. Participants were asked to describe the scenes and rate the hostile, positive, negative, and neutral character of each scene on a 4-point scale, ranging from most plausible to least plausible. While all of the previously mentioned HAB measures provide interesting results, they are not without flaws. Indeed, these methods do not allow for the measurement of spontaneous inferences and rapid intention attribution processes that are characteristic of the HAB. The latter occurs in the early stages of social information processing [13]. Before providing their responses, participants have time to consider other, more socially acceptable interpretations.

To capture the first cognitive processes of real-time intention attribution, Gagnon et al. [24] developed an innovative measurement method based on the recording of brain signals. The aim was to present different scenarios on a screen that, in written form, describe a character performing ambiguous behavior toward the reader in a context-specific manner (see Table 1). The context was either hostile or nonhostile and the reader was asked to read the scenarios while imagining the intention of the character. Subsequently, the character’s actual intention was revealed through a final target word and event-related potentials (ERPs) were recorded. The intention could be either hostile or nonhostile. In principle, when the hostile or nonhostile nature of the intention was at odds with the hostile or nonhostile nature of the context, expectations about the intention of the character being portrayed were violated. According to Gagnon et al. [24], the ERP component N400 was observable when hostile expectations were violated. In the literature, N400 is described as a negative deflection occurring around 200–500 ms poststimulus presentation [24, 25]. Its amplitude is maximal in the centro-parietal regions of the brain and is triggered when the word presented is unexpected or inconsistent with the context in the scenario [25]. In a study by Gagnon et al. [26], the N400 directly measured expectation violation, and its amplitude was stronger among aggressive individuals compared to nonaggressive individuals during the hostile expectations violation than during the nonhostile expectations violation.

List	First sentence—context	Second sentence—behavior	Third sentence—intention	Condition
1	You’re playing soccer against a team that has an aggressive style	On a breakaway, the defender trips you up	The defender wants to hurt you^a	Hma
2	You have soccer practice with your team			Hmi
1	You’re having dinner with friends and Sylvie, who’s obnoxious	She does not mention that your shirt is stained	Sylvie does not want to embarrass you^b	NHmi
2	You’re having dinner with friends and Sylvie who’s nice			NHma

Table 1.

Examples of scenarios under the four conditions of the hostile expectancy violation paradigm.

NHma = nonhostile match; NHmi = nonhostile mismatch; Hma = hostile match; Hmi = hostile mismatch. Here, the target word is in bold. Translation in English of ^a “Le défenseur veut vous blesser”, ^b “Sylvie ne veut pas vous embarrasser”.

The main goal of this study is to examine the mediating role of the HAB (measured by EEG and self-report) in the relationship between cluster B personality traits and reactive aggression. To achieve this, we present several objectives and hypotheses. (1) First, we want to replicate and validate the HAB measurement method developed by Gagnon et al. [24]. Our first hypothesis is that N400 will be more pronounced in the right posterior brain regions during the hostile expectations violation. (2) Secondly, we aim to evaluate the predictive role of ASPD traits, BPD traits, and HAB (as measured by EEG and self-report) on self-reported reactive aggressive behaviors. (a) We hypothesize that ASPD and BPD traits will positively and significantly predict self-reported reactive aggression. (b) We hypothesize that ASPD and BPD traits will significantly predict self-reported HAB and hostile expectations violation. (c) We expect a neurophysiological and self-reported measure of HAB to significantly predict reactive aggression. (d) Finally, we expect that self-reported and neurophysiological HAB will mediate the relationship between cluster B personality traits and reactive aggression.

2. Methods

2.1 Participants

Seventy-two French-speaking adults were recruited from university classes in two metropolitan universities, a list of former patients who consulted in a personality disorders clinic, and the general population through posters and announcements on Facebook and Kijiji. Interested individuals were then contacted by email to receive information about the study and to make an appointment for a laboratory visit. All participants were between 18 and 65 years of age, had normal vision with or without correction and had no history of psychosis, neurological disorder, or severe brain damage. Seventeen of them had been taking a central nervous system medication (e.g., anxiolytic, stimulant, SSRI, SNRI, and antidepressant) for at least 2 weeks prior to the day of the experiment. Before the visit, participants were asked not to use other drugs or alcohol for at least 1 week and 24 h prior to the experiment, respectively. Failure to comply with any of these instructions resulted in the postponement of the appointment. All participants received financial compensation of $25 at the end of the appointment. Nine participants were excluded due to attrition, a significant amount of missing data, a mother tongue other than French, or excessive artifacts on the EEG signals caused by eye movements. The final sample consisted of 63 participants (46 females and 17 males) with an average age of 29 (SD = 1.44) and 15 years of education (SD = .40).

2.2 Measure

Personality Assessment Questionnaire [27, 28]. Only scales of the French adaptation assessing borderline and antisocial personality traits were included. Each subscale consisted of 24 items. The Antisocial Characteristics Scale (ASPD features) consisted of eight items measuring antisocial behavior, eight items measuring egocentricity, and eight items measuring stimulus seeking. The Borderline Characteristics Scale (BPD features) consisted of six items assessing affective instability, six items assessing identity problems, six items assessing negative relationships, and six items assessing self-harm. Each item was scored on 4 Likert-type points, ranging from 0 = False, not at all true to 3 = Very true. For each subscale, the scores on the 24 items were added together to form a total score for antisocial traits and a total score for borderline traits. Higher scores reflected the greater degree of personality traits. According to Morey [28], several studies have demonstrated the reliability and validity of the PAI subscales in normal, clinical, and student populations [29]. In our study, the internal consistency was excellent for the ASPD features scale (α = .91) and the BPD features scale (α = .90).

Brief Symptom Inventory (BSI; [30]). Two scales from the French version of the BSI [31] were used to measure the level of depression and paranoia and used as control variables as both traits are associated with HAB [32, 33]. The depression scale consisted of six items, while the paranoid ideation scale consisted of five items. Each item was answered using a 5-point Likert-type scale, ranging from 0 = not at all to 4 = extremely. For each scale, a total score was calculated by adding up all item scores. The higher the score, the greater the level of traits. The reliability and validity of the BSI dimensions have been demonstrated in normal and clinical population samples [34, 35]. In our study, Cronbach’s alpha was good for the depression scale (α = .84) and paranoid ideation (α = .81).

Reactive-Proactive Aggression Questionnaire (RPQ; [36]). The French version of the RPQ [16] was used to assess aggression behaviors. The questionnaire included an 11-item scale measuring reactive aggression (e.g., getting angry at the provocation of others) on a 3-point Likert-type scale, ranging from 0 = never to 2 = often. The reactive aggression scores were calculated by adding the item scores. Higher scores indicated greater aggressive behaviors. Reliability and validity were tested in multiple samples of incarcerated and nonclinical individuals ages 6–64 years old [36, 37]. In our sample, the internal consistency was good for the reactive aggression scale (α = .82).

Social Information Processing – Attribution and Emotional Response Questionnaire (SIP-AEQ; [22]). The French version of the SIP-AEQ [38] was administered to measure the HAB. The SIP-AEQ included eight vignettes depicting scenes of everyday life where a character acts provocatively and has ambiguous intentions. For each vignette, participants were asked to rate the likelihood that the character’s intention was directly hostile, indirectly hostile, neutral, or instrumental (four items per vignette). Each item was rated on a 4-point Likert-type scale, ranging from 0 = not at all likely to 3 = very likely. Hostile attribution biases were calculated by averaging the responses to the eight vignettes for each hostile intention type (HAB-direct; HAB-indirect). The HAB score was determined by adding HAB-direct and HAB-indirect. The higher the score, the higher the HAB. According to Coccaro et al. [23], the reliability and validity of the SIP-AEQ have been demonstrated in multiple samples. In our sample, Cronbach’s alpha was excellent for HAB (α = .93).

The last questionnaire administered assessed age, gender, mother tongue, and education status.

2.3 Stimuli

The stimuli constituted 320 scenarios depicting social interactions encountered in everyday life and was developed by Gagnon et al. [24] to test hostile and nonhostile expectancy violations. Each scenario consisted of three sentences (see Table 1). The first sentence described a typically hostile or nonhostile context. The second sentence depicted a character whose intention was ambiguous, thus committing potentially provocative behavior to the reader. The last sentence included a final target word that resolved the ambiguity by clarifying the intention behind the behavior. The scenarios were created under four conditions—hostile match (Hma), hostile mismatch (Hmi), nonhostile match (NHma), and nonhostile mismatch (NHmi). When the conditions were hostile, the target word indicated hostile intent on the part of the character’s behavior. Conversely, when the conditions were nonhostile, the intention was depicted as nonhostile. Conditions were said to be a match when the hostile or nonhostile nature of the intention was consistent with the hostile or nonhostile nature of the context. Similarly, conditions were said to be mismatched when the hostile or nonhostile nature of the intention differed from the hostile and nonhostile nature of the context. Two lists of 160 scenarios (i.e., 2 × 40 scenarios for each of the four conditions) were used to balance the match and the mismatch conditions with the hostile and the nonhostile conditions across participants. For a given scenario, the match and mismatch versions shared the same behaviors and intentions but differed in the hostile or nonhostile nature of the context. The first two sentences were composed of a maximum of 25 words and the last sentence a maximum of eight words. The third sentence was phrased negatively in almost 50% of the scenarios for each condition. The two lists were administered alternately and equally across participants. A list of 20 additional scenarios (i.e., 5 × 4 scenarios for each of the four conditions) was developed for the purpose of practice and comprehension trials.

2.4 Procedure

After completing the online questionnaire and giving their written consent, participants were invited to the laboratory to perform the experimental task. While their brain activity was recorded, they were asked to read the daily life interaction scenarios and visualize them as though they were actually experiencing them. As they read the first two sentences, the reader had to imagine why the characters were behaving in such a way toward them (intention attribution process). Once ready, they could initiate the presentation of the third sentence. For each scenario, a trial consisted of presenting the first two sentences for at least 1500 ms. After pressing the space bar on the keyboard, a delay of 500 ms without stimuli was followed by a fixation cross appearing in the center of the screen for 1000 ms. A third sentence was then displayed, word by word, in the center of the screen and ended with the target word. Each word was presented for 300 ms, with a delay of 200 ms between words. Finally, a fixation cross was displayed in the center of the screen for 2000 ms. The participant had to keep his eyes focused on the center of the screen and refrain from blinking from the appearance of the first cross until the disappearance of the second cross. In total, there were four practice trials followed by 10 blocks of 17 trials (170 trials). Each block consisted of 16 experimental trials (four scenarios for each of the four conditions: Hma, Hmi, NHma, and NHmi) and one trial used as a comprehension test. The comprehension trial was followed by a true or false question. The purpose of this question was to ensure that the participant was reading and understanding the scenarios. The participant could answer by pressing the letter N (true) or M (false). A correct/incorrect answer was followed by feedback (green or red cross, respectively). For our sample, the average rate of correct answers was 91.1%, indicating a high rate of comprehension. The experimental trials were presented in random order and without repetition. The blocks were separated by a break, the duration of which was determined by the participant. The words and fixation crosses were written in white, Helvetica font, size 14, bold, on a 17-inch (43.18 cm) black screen. The distance between the screen and the participant’s eye was 70 cm. Three characters corresponded to approximately 1° of visual angle. The experimental task was created using E-Prime 2.0 software (E-Prime, Psychology Software Tools, Pittsburgh, PA) [39].

2.5 Electrophysiological methods

The electroencephalography took place in a Faraday cage and under medium brightness. The brain activity of the participants was captured using 64 Ag/AgCI electrodes in an elastic cap. The position of the electrodes was done according to the International 10–10 System [40]. The right and left mastoids were used as references. One electrode was placed below the left eye to capture blinking and vertical eye movements. Two other electrodes were placed at the outer canthi of the eyes to capture horizontal eye movements. The signals were processed and recorded via a Biosemi ActiveTwo amplifier system (Amsterdam, Netherlands) at a sampling frequency of 512 Hz. Online, a 0.16 Hz high-pass filter and a 100 Hz low-pass filter were applied to the EEG signals. On Matlab, a 0.1 Hz high-pass filter and a 30 Hz low-pass filter were applied during offline analyses. The resulting signals were segmented in trials according to a time window of from 200 ms before, to 800 ms after the target word onset. The baseline time window ranged from −200 ms to 0 ms. Trials containing too many artifacts (i.e., eye or muscle movements) were rejected using an independent component analysis [41]. Rejection thresholds were applied for blink (i.e., > 80 mV within a time window of 150 ms) and for eye movement (i.e., > 35 mV within a time window of 300 ms). Electrodes with a noisy EEG signal (i.e., exceeding +/− 100 mV voltage) were interpolated by spherical spline. When more than seven electrodes were noisy in a trial, the trial was rejected. When the number of rejected trials was greater than 20 per condition, the participant was excluded from the sample. In our final sample, the percentage of rejected trials was less than 17.5% in the four conditions (i.e., 0–17.5% for Hma, 0–15% for Hmi, 0–12.5% for NHma, and 0–12.5% for NHmi). The trials were then averaged by condition (Hma, Hmi, NHma, and NHmi) and for each participant. On average, there were 39 trials per condition. The ERP amplitudes captured by the electrodes were averaged over six lateral regions and three midline regions on the scalp. The lateral electrodes were separated as follows: anterior left (AF3, AF7, F1, F3, F5, F7, FT7, FC1, FC3, FC5), central left (TP7, T7, C1, C3, C5, CP1, CP3, CP5), posterior left (P1, P3, P5, P7, PO3, PO7, O1), anterior right (AF4, AF8, F2, F4, F6, F8, FT8, FC2, FC4, FC6), central right (TP8, T8, C2, C4, C6, CP2, CP4, CP6), and posterior right (P2, P4, P6, P8, PO4, PO8, O2). The midline electrodes were analyzed as follows—anterior median (AFZ, FZ, FCZ), central median (CZ, CPZ), and posterior median (PZ, POZ, OZ).

2.6 Statistical analyses

Statistical analyses were performed to evaluate the voltage of the ERP amplitudes (dependent variable) according to the conditions (Hma, Hmi, NHma, and NHmi) and location of sensors on the scalp. Each subject being its own control, two repeated measures ANOVAs with Huynh-Feldt corrections were performed. The first ANOVA was for the lateral electrodes. The independent variables were intention (hostile, nonhostile), Consistency (match, mismatch), Hemisphere (left, right), and Location (anterior, central, posterior). Mean ERP amplitudes observed at midline regions were analyzed in a second ANOVA. The independent variables were Intention, Consistency, and Location. Given that our first objective was to demonstrate the presence of an N400 during expectancy violations (mismatch-match conditions), interaction effects involving the Consistency factor were looked at in the ANOVAs. To assess the role of the N400 in our mediation models, we selected regions showing greater negative amplitude (as shown in [24, 25]). Pearson’s correlations were performed between all variables. Therefore, several multiple linear regressions were conducted to assess whether antisocial characteristics, borderline characteristics, and the HAB (as measured by self-report or EEG) predicted scores on reactive and proactive aggression. A product-of-coefficient test for mediation analyses was performed by using bootstrapping procedures, a nonparametric resampling technique to test for indirect effects [42]. This method has been recommended as bootstrapping was found not to inflate Type I and Type II error rates and to have higher power [43]. In addition, bootstrapping does not assume multivariate normality. The significance of the mediation effect is determined when the 95% bias-corrected confidence intervals (CIs) do not contain zero. In the current study, estimates are based on 5000 bias-corrected bootstrap samples. All analyses were two-tailed, with an α level set at .05.

3. Results

3.1 N400

Figure 1 shows differences in mean ERP amplitudes between mismatch and match conditions for the nine scalp regions. Mean amplitude differences indicate a negative deflection (N400) at around 350–650 ms during hostile expectancy violations (NHmi-NHma conditions). Based on visual inspection, the greatest deflections were at midline and right sites in the central and posterior regions. When nonhostile expectancies were violated (Hmi-Hma conditions), mean perceived amplitudes neared zero in the central and posterior regions. Figure 2 shows the topography of mean amplitude differences observed on the scalp from 350 to 650 ms (post target onset) during hostile and nonhostile expectancy violations. During the hostile expectancy violations, the N400 seems to appear in the central and posterior regions of the right hemisphere and the midline sites.

Figure 1.

Difference between the mismatch and match conditions of the grand ERP averages obtained after presentation of the hostile or nonhostile target word for 9 brain regions. LA = anterior left; LC = central left; LP = posterior left; MA = anterior median; MC = central median; MP = posterior median; RA = anterior right; RC = central right; RP = posterior right.

Figure 2.

The topographic map of ERP mean differences between mismatch and match conditions from 350 to 650 ms after presentation of hostile or nonhostile target words. On the left, nonhostile expectancy violation (hostile mismatch-match). On the right, hostile expectancy violation (nonhostile mismatch-match).

For the lateral electrodes ANOVA, there was an interaction effect between Intention, Consistency, and Location (F (2.124) = 5.90; p = .01), and between Intention, Consistency, and Hemisphere (F (1,62) = 5.21; p = .03). For these interactions, effect sizes were moderate (partial R² = .08; partial R² = .09, respectively). Simple effects for these last two interactions were assessed for Consistency factor by paired comparisons with post hoc Bonferroni adjustment. The levels of the Consistency factor (match and mismatch) differed significantly for the nonhostile intention in the central and posterior regions, with an adjusted alpha of .004. There was no difference between hostile mismatch and hostile match at anterior, central, and posterior sites.

For the midline regions ANOVA, there was an interaction effect between factors of Intention and Consistency (F (1,62) = 16.16; p = .00), between factors of Consistency and Location (F (2,124) = 10.59; p = .00), and between factors of Intention, Consistency, and Location (F (2,124) = 5.60; p = .01). Effect sizes for these interactions were high to moderate (partial R² = .21; partial R² = .15; partial R² = .08). Simple effects for the last interaction were evaluated for the Consistency factor by paired comparisons with post hoc Bonferroni adjustment. The level of Consistency factor differed significantly for nonhostile intention at central and posterior regions on the scalp with an adjusted alpha of .004. There was no difference between mismatch and match for hostile Intention in anterior, central, and posterior regions.

These results confirm the presence of the N400 in central and posterior regions in the nonhostile intention condition (i.e., when hostile expectations were violated). In the hostile Intention condition (i.e., during nonhostile expectancy violations), the N400 was not significantly visible. Since the ERP waveform differences and the topographic map indicated a stronger N400 effect in the central and posterior regions of the right and midline sites, we selected MC, RC, MP, and RP regions for further analysis.

3.2 Prediction of reactive aggression

The scores of all self-report scores were normally distributed, except for antisocial behavior that had positive skewed distributions as observed in the general population.

Correlation coefficients of the variables of interest are presented in Table 2. Reactive aggression was significantly correlated with age, depression, antisocial traits, borderline traits, and nonhostile expectancy violations (hostile Intention condition) in the MC (r = −.29, p £ .05 two-tailed), MP (r = −.25, p £ .05 two-tailed), RC (r = −.35, p £ .01 two-tailed), and RP (r = −.32, p £ .01 two-tailed) region. In addition, antisocial traits were significantly correlated with gender, paranoid ideation and borderline traits, and indirect hostile attribution bias (r = .30, p £ .05 two-tailed). Borderline traits were significantly correlated with paranoid ideation, depression, and self-reported HAB. Reactive aggression was not significantly correlated with hostile expectancy violations (nonhostile intention condition). Because the correlation between hostile or nonhostile expectancy violations and aggression scores was more strongly consistent in the RC region than in the MC, MP, and RP regions, regression analyses were performed in the RC region.

Variables	1	2	3	4	5	6	7	8	9	10	11	12
1. Age	—
2. Gender	.27*	—
3. Education	−.11	−.12	—
4. Paranoid idea	.13	.01	.09	—
5. Depression	.07	.06	.08	.49**	—
6. ASPD features	.08	.39**	.04	.26*	.19	—
7. BPD features	.05	.03	.21	.58**	.68**	.48**	—
8. REAG	.26*	.23	.06	.24	.41**	.47**	.52**	—
9. HAB	.13	.11	−.11	.58**	.37**	.25	.34**	.14	.16	—
10. HN400RC	−.01	−.08	.03	.00	.10	−.10	−.11	−.35**	−.27*	.18	—
11. NHN400RC	.14	.26*	−.09	−.13	−.01	−.00	−.21	.19	.01	−.08	−.03	—

Table 2.

Correlation matrix.

^*

p ≤ .05.

^**

p ≤ .01.

REAG: reactive aggression; HAB: hostile attribution bias – SIP-AEQ; HN400RC: N400 effect in RC region for nonhostile expectancy violations; NHN400RC: N400 effect in RC region for hostile expectancy violations.

A first regression was conducted with ASPD features as the independent variable (Figure 3). Hostile expectancy violation (nonhostile intention condition) and self-reported HAB were the mediator variables and age, sex, education, BPD features, paranoid ideation, and depression traits served as covariates. Results showed a nonsignificant indirect effect for the hostile expectancy violation (indirect = −.00, SE = .01, 95% CI [−.02; .02]) and self-reported HAB (indirect = −.00, SE = .01, 95% CI [−.02; .01]). The model explained 44% of the variance of reactive aggression, F(9, 53) = 4.56, p < .001 with antisocial characteristics and hostile expectancy violation as significant predictors. The same regression was assessed with nonhostile expectancy violation (hostile Intention condition) and self-reported HAB as mediator variables (Figure 3). Results showed a nonsignificant indirect effect for nonhostile expectancy violation (indirect = −.01, SE = .02, 95% CI [−.05; .04]) and self-reported HAB (indirect = .00, SE = .01, 95% CI [−.02; .02]). The model explained 50% of the variance of reactive aggression, F(9, 53) = 5.73, p < .0001 with antisocial characteristics and nonhostile expectancy violation as significant predictors. These results indicated that neither electrophysiological (hostile and nonhostile expectancy violations) nor self-report measures of HAB mediated the relationship between ASPD traits and reactive aggression.

Figure 3.

Mediation of antisocial characteristics—reactive aggression relationship by the hostile attribution bias and the N400 in hostile and the nonhostile conditions.

A third regression was conducted with BPD features as the independent variable, hostile expectancy violation (nonhostile intention condition), and self-reported HAB as mediator variables and age, sex, education, ASPD features, paranoid ideation and depression traits served as covariates (Figure 4). Results showed a nonsignificant indirect effect for the hostile expectancy violation (indirect = −.02, SE = .02, 95% CI [−.06; .01]) and self-reported HAB (indirect = .00, SE = .01, 95% CI [−.01; .03]). The model explained 44% of the variance of reactive aggression, F(9, 53) = 4.56, p < .0001 with borderline characteristics and hostile expectancy violation as significant predictors. Also, borderline characteristics predicted hostile expectancy violation. A final regression was assessed with nonhostile expectancy violation (hostile Intention condition) and self-reported HAB as mediator variables and the same covariates (Figure 4). Results showed a nonsignificant indirect effect for the nonhostile expectancy violation (indirect = .04, SE = .03, 95% CI [−.00; .10]) self-reported HAB (indirect = −.00, SE = .01, 95% CI [−.02; .02]). The model explained 50% of the variance of reactive aggression, F(9, 53) = 5.73, p < .0001 with only nonhostile expectancy violation as a significant predictor. Therefore, neither hostile and nonhostile expectancy violations nor self-report measures of HAB had a mediating effect on the relationship between borderline characteristics and reactive aggression. There was no other significant relationship.

Figure 4.

Mediation of borderline characteristics—reactive aggression relationship by the hostile attribution bias and the N400 in hostile and nonhostile condition.

4. Discussion

The first objective of this study was to replicate the measurement method of Gagnon et al. [24] and to validate their results. The aim was to present scenarios describing social interactions while measuring brain activity. In each scenario, characters acted in a provocative and ambiguous manner in both hostile and nonhostile contexts. Participants were asked to read the scenarios on a screen and imagine the intentions behind the behaviors presented. Subsequently, the characters’ hostile or nonhostile intentions were revealed through a final target word. As in the study by Gagnon et al. [24], we were able to observe the N400 ERP component in a time window ranging from 350 to 650 ms post-stimulus onset. Moreover, the amplitude of this deflection was more pronounced during the hostile expectancy violation in the central and posterior cerebral regions at the medial and right electrodes. This implied that participants attributed hostile intent to the characters when the context was hostile. This result has been corroborated by several other studies reporting a maximum amplitude N400 in the centro-parietal regions when expectations are violated [24, 25].

During the nonhostile expectancy violation, ERP amplitudes neared zero at approximately 350–650 ms. Therefore, when a nonhostile context was followed by ambiguous and provocative behavior, participants did not attribute a nonhostile intent to the behavior. Although consistent with findings reported in Gagnon et al. [24], this result appears inconsistent with the established assumption that the N400 would reflect an expectancy violation. Gagnon et al. [24] suggest this phenomenon possibly reflects a cautious interpretation, based on perceived cues, on the part of nonaggressive students. It is indeed possible that, in our study, nonhostile contextual cues conflicted with the ambiguous and provocative nature of the behavior. Therefore, the type of intent attribution depended on the weight the participant gave nonhostile cues versus provocative cues. In the end, in scenarios designed to violate nonhostile expectations, the participant may have had mixed views and not been systematically surprised to see hostile intent appear after a nonhostile context.

The second objective of this study was to demonstrate the predictive role of ASPD traits, BPD traits on self-reported aggressive behaviors. As expected, ASPD traits positively predicted reactive aggression in both models, which is consistent with the scientific literature [9, 21]. BPD traits were highly correlated with reactive aggression. However, when controlling for age, gender, education, depression, paranoid ideation, and ASPD traits, they did significantly predict reactive aggression in one model only. This result was surprising given that several studies have shown BPD to be a good predictor of reactive aggression [6, 7, 9]. In a recent longitudinal study, however, Penson et al. [29] showed that BPD characteristics were not sufficient in significantly predicting aggressive behaviors and rather, that ASPD characteristics were better predictors. Thus, it is likely that, in our regressions, ASPD traits were more effective predictors of reactive aggression than BPD traits. In addition, BPD traits and ASPD traits shared a high percentage of common variance (r² = .24), possibly explaining the nonsignificant coefficient for BPD traits in the regression.

ASPD traits failed to predict both HAB, as measured by self-report, and hostile or nonhostile expectancy violations. These findings are not consistent with the few studies evaluating HAB in ASPD [20, 21]. However, it is important to mention that the methodology used to measure the HAB could explain the conflicting data. We used the SIP-AEQ questionnaire to measure self-reported HAB and an electrophysiology method developed by Gagnon et al. [24] to measure hostile and nonhostile expectancies violations. As such, it is possible to expect different results across studies. Given that ASPD is characterized by a lack of conformity to societal norms [1], it is also possible that the individuals with ASPD in our study did not relate to the characters or that they experienced difficulty imagining the situations described in our task.

BPD traits also did not predict self-reported HAB and nonhostile expectancy violation. In contrast, they positively predicted the hostile expectancy violation (i.e., BPD traits negatively predicted N400). Thus, the higher the BPD traits, the stronger the hostile expectancy violation. In other words, when the context was hostile, people with high BPD traits made more hostile intent attributions than people with lower BPD traits. This result partially confirmed our expectations and was consistent with findings in Smeijers et al. [18]. In addition, several researchers have provided arguments regarding the meaning of such a prediction [17, 18, 19]. For example, Lobbestael and McNally [17] demonstrated that people with BPD were subject to interpretive biases related to rejection and anger. According to Baer et al. [18], people with BPD have negative beliefs about themselves and their environment. They also interpret and evaluate neutral and ambiguous stimuli negatively. Finally, according to Arntz et al. [19], people with BPD judge other people as negative, aggressive, and malicious.

Self-reported HAB did not predict reactive aggression in all models, which is in contrast with the numerous studies showing that self-reported HAB is positively related to reactive aggression [15, 16, 44]. It is possible that variability in HAB scores was too small in our study to observe correlations. Regarding the hostile and nonhostile expectancies violations as measured by the N400 effect, results showed an opposite relationship with reactive aggression than expected. First, given that HAB and reactive aggression are positively associated as reported in the literature [15, 16, 44], we assumed that N400 effect in the nonhostile intention condition (hostile expectancies violation) would negatively predict reactive aggression (more negative amplitude associated with higher aggression score). However, because of the chronic accessibility to hostile patterns, it is possible that an aggressive person would see aggression in all their social interactions [45]. Since mismatch and match conditions would have a similar effect in this case, their subtraction should have the effect of reducing the N400 (more positive amplitude going up) as aggressive traits increase. Second, we assumed that nonhostile expectancy violation would negatively predict reactive aggression as nonaggressive individuals would be more surprised to see a hostile intention appear after a hostile context. However, given that the N400 effect was nonsignificant when the intention words were hostile, it is difficult to infer the nature of the cognitive processes underlying this relationship. Also, it appears that two other studies have found a negative relationship between HAB and reactive aggression [46, 47] suggesting that the relationship between HAB and reactive aggression could be more complex than we may think and difficult to predict.

Finally, our final hypothesis that self-reported HAB, the hostile expectancy violation, and the nonhostile expectancy violation were mediators of the relationship between cluster B personality traits and reactive aggression, could not be confirmed. These findings were inconsistent with the few studies that have evaluated the mediating role of HAB in the relationship between these personality features and reactive aggression [20, 21]. When it comes to cluster B personality, it is possible that the N400 effect (expectancy violation) may be influenced by other mediators, like sensitivity to rejection, impulsivity, and dysfunctional beliefs [17, 18], which were not included in our study. Further studies are therefore needed to better understand the cognitive and affective processes underlying aggressive behavior in antisocial and borderline personality disorders.

This research project had several methodological limitations, such as sample size and heterogeneity. Future analyses using a larger sample would be warranted to better understand the nature of the observed relationships. In addition, our sample potentially over-represented students in the general population. Out of 63 participants, 49 were from an academic background. It would be interesting and beneficial to evaluate our measures on samples more representative of the clinical population.

5. Conclusion

In conclusion, our study replicated the measurement of expectation violations by electrophysiology and validated the presence of a strong negative deflection of ERP amplitudes at the time of hostile expectation violations, as demonstrated in the study by Gagnon et al. [24]. Additionally, our results show that antisocial traits and borderline traits were positively associated with self-reported reactive aggressive behaviors. Our mediation models involving intention attribution processes as mediators could not be confirmed and the unexpected results suggested that HAB and reactive aggression sustain a complex relationship. To better understand the meaning of the relationship between hostile and nonhostile expectancy violation and reactive aggression, more studies are in need to verify how N400 effect among aggressive and nonaggressive participants varies according to various parameters of the ERP task. Nonetheless, this study indicates that electrophysiological measurements can be more sensitive than self-report questionnaires when investigating the nature of cognitive processes associated with reactive aggression. Considering the contribution of socio-cognitive treatments that are offered to aggressive individuals (e.g., [48]), we believe that this study can help to open the way to other empirical studies using ERP tasks to understand the cognitions associated with reactive aggression among cluster B personality disorders.

Acknowledgments

This study was supervised by Dr. Jean Gagnon, PhD, from the Department of Psychology at the University of Montreal. The study was funded by a portion of the supervisor’s research funds. The project was supported by Dr. Monique Bessette, PhD, psychologist and director of the Victoria Institute for participant recruitment. The author would like to thank Dr. Pierre Jolicoeur, PhD, of the Department of Psychology for his support in the analysis of EEG signals. This study was supported by a research grant to JG and PJ from the Social Sciences and Humanities Research Council (SSHRC) (no 435-2018-0963).

Conflict of interest

There is no conflict of interest. This project was approved by the Research Ethics Board of Education and Psychology of the University of Montreal. All procedures were consistent with the Énoncé de politique des trois conseils (EPTC-2, 2018).