Both partners' negative emotion drives aggression during couples' conflict

Abstract

Researchers examining conflict between intimate partners believe that the experience and expression of emotion drives aggressive behaviour. Intra-personally, increases in negative affect make aggression more likely. Inter-personally, theoretical models suggest that each individuals' perception of their partners' emotion also influences aggression, potentially creating a Violence Escalation Cycle. Here, using a lab-based aggression task across a primary study (n = 104, number of trials = 3095) and a replication (n = 58, number of trials = 3167), we show that both intra- and inter-personal experiences of negative emotion predict reactive aggression within couples, revealing retaliation but not escalation. Critically, analyses of facial affect reveal that prototypic displays of negative emotions have a compounding effect, leading to dramatic changes in aggression depending on whether one, both, or neither partner expressed negative emotion. We propose a mechanism by which temporal delays (i.e., experimentally imposed forced breaks) reduce aggression by decreasing negative emotional arousal and limiting impulsive action. Our results show that both forced breaks and elective breaks (i.e., extra participant-initiated extensions of the forced break time) reduce aggression, providing exciting evidence that interventions focused on preventing impulsive action when people are in a provoked state can reduce aggression within couples.

Fig. 1. Experimental Set Up and Affect Coding.

a Volunteers (romantic couples) participated together, face-to-face. Each participant had a monitor, a big red button, a keyboard, and a pair of headphones. To record facial expressions an Insta360 camera was positioned centrally, below the eyeline, allowing participants to see each other unobstructed. b Each round began with the prompt “Ready” (in white text on a black screen), followed by “Set” exactly one second later, and finally “GO!!” after 0 to 8 s (varied randomly to increase task attention). Screens then revealed who had pushed their button first and prompted the winner to select a blast level (either immediately or delayed by a 5, 10, or 15 s forced break, depending on the experimental condition). Once a blast level was selected by the winner, the noise was delivered at the chosen volume to the loser’s headphones for 2 s. Once the blast was over, the next round began immediately. Participants played 30 rounds. c Video recordings were analyzed frame-by-frame using a machine learning program designed to automate annotation with the Facial Action Coding System, allowing prototypic displays of emotion to be identified. d Composite facial images illustrate differences in expression for prototypic displays of positive and negative emotion. Images were generated by averaging together stills exhibiting the characteristic Action Units (AUs) associated with each effect. AUs illustrated are 6 (cheek raiser) and 12 (lip corner puller) for positive effect; 4 (brow lowerer [sic]), 9 (nose wrinkle) and 23 (lip tightener) for negative effect.

Fig. 2. Illustration of the aggression eliciting effects expected from our face-to-face CRTT.

According to the I³ Model and Perfect Storm theory. In the last round (i-1), a player loses, is faced with their partner’s negative emotions, and is blasted with a loud, unpleasant sound. In the next round they win (i): having been instigated by their last loss, they are impelled toward aggression by their own negative emotions, their partner’s negative emotions, and the fact they won the round. Finally, the task demands of the paradigm, along with a sense of justified retaliation for previous blasts and latent trait factors, act to disinhibit participants. Theoretical models predict that this environment should produce aggression, resulting in high negative emotional expression and high blast level selections.

Fig. 3. Trait aggression and emotional expression.

a Behavioural performance data revealed that our self-report measure of trait aggression (the BPAQ-SF) did not correlate with our in-task measure of aggression (mean blast level), consistent with previous research showing mixed results regarding the relationship between the BPAQ and the Competitive Reaction Time Task (CRTT). b Our assessment of emotional expressions during the blast initiation period (blast initiation ±1 s) explains the lack of relationship between BPAQ and aggression measures seen here and in previous studies—participants displayed significantly more positive affect (green) than negative affect (red) during task performance. Defining Plot Elements. Grey-shaded regions in regression plots represent standard errors. Boxplot features are as follows: centre line, median; upper and lower box boundaries, Inter Quartile Range (IQR); upper and lower whiskers, smallest or largest value within 1.5x the corresponding IQR; dots, outliers.

Fig. 4. Identifying aggressive behaviour.

a Using facial affect data we isolated responses with high positive or negative emotional expressive power. We then examined aggressive behaviour (the proportion of responses at each blast level) as a function of emotion when participants were able to act immediately, illustrated as density plots. The distribution of responses reveals significantly greater use of high blast levels when high negative affect was present (compared to high positive affect without high negative affect). b Further examination of behaviour associated with high negative affect revealed significantly lower use of high blast levels when responses were made after a forced break (compared to immediate responses). c By contrast, examination of behaviour associated with high positive affect revealed no effect of a forced break (compared to immediate responses) and no skew towards high blasts. d Similarly, examination of behaviour associated with low negative affect also revealed no effect of a forced break (compared to immediate responses), reflecting greater use of low blasts. e Given that we observed a reduction in aggressive responding in the forced break condition when participants expressed negative affect, we reasoned that negative expression should have been greater when the winner was announced (Time 1), compared to after the break, when the winner selected a blast level (Time 2). As predicted, box plots show a significant reduction in mean negative emotional expressive power over time in the forced break – the reduction in highly aggressive behaviour seen following forced breaks was associated with a reduction in negative affect. f Further, the difference in negativity change between the immediate response and the forced break condition is also significant, strengthening the claim that forced breaks produced the observed reduction in aggression. g When players elected to wait even longer than their forced break time, the extra time they waited negatively correlated with negative expressive intensity. h Extra time waited after the forced break period is over also negatively correlated with aggression. Together, these data suggest that taking a break can be effective whether it is forced or voluntary. Defining Plot Elements. Grey shaded regions in regression plots represent standard errors. Boxplot features are as follows: centre line, median; upper and lower box boundaries, Inter Quartile Range (IQR); upper and lower whiskers, smallest or largest value within 1.5x the corresponding IQR; dots, outliers.

Fig. 5. Negative emotion compounds within couples.

To assess the interaction between partners we examined aggressive behaviour (the proportion of responses at each level of blast) as a function of whether neither, one, or both members of a couple expressed high negative affect. Density plots from the immediate response condition revealed that when neither player expressed high negative affect (white: m = 3.4; sd = 2.0) behaviour was dominated by low blast level responses. A significant increase in higher blast level responses occurred when either the loser (grey: m = 4.3; sd = 2.5) or winner (red: m = 5.1; sd = 2.6) displayed high negative affect. Furthermore, blast level selections were significantly higher again when both the winner and loser displayed high negative affect (burgundy: m = 6.2; sd = 2.3). Aggressive behaviour was significantly higher when both members of a couple expressed high levels of negative emotion.

Fig. 6. Dyadic conflict behaviour.

a Counter to expectation, results do not reveal a tendency for couples in conflict to escalate over time. b When compared directly against each other, however, partners tended to closely match each other’s grand mean blast levels. c Couples also tended to roughly match each other’s blast selections at a trial level, meaning that partners choose blast levels (in R(ound)_i) similar to their partner’s previous blast level selection (in R_i-1). d, e, f The pattern of blast levels within three separate couples illustrates the matching of blast levels over time. g Trial-to-trial changes in aggression were used to calculate the proportion of trials associated with matching, escalation and de-escalation, in each participant, shown as density plots illustrating the prevalence of each behaviour. h An example couple’s behaviour, showing the dynamics of aggression over time. For each behaviour (matching/escalation/de-escalation) an arrow points to the second of two responses (one from each player) that exemplify that behaviour. Player 1 (red) consistently makes higher blast level responses than Player 2 (black). Player 1 exhibits both matching and escalation over time, whereas Player 2 only exhibits de-escalation. i Comparison of the escalation tendencies within couples reveals a significant asymmetry between escalation and de-escalation when partners are compared against each other. Couples shown in (d, e, f and h) are identified, illustrating the different patterns of behaviour observed as a function of the escalation/de-escalation asymmetry. Defining Plot Elements. Grey shaded regions in regression plots represent standard errors. Boxplot features are as follows: centre line, median; upper and lower box boundaries, Inter Quartile Range (IQR); upper and lower whiskers, smallest or largest value within 1.5× the corresponding IQR; dots, outliers.

Fig. 7. Actor-partner interdependency.

Actor-Partner Interdependence Models allow us to see the impact of each player’s emotion on their own, and their partner’s, future emotion. These models are longitudinal (illustrated with a fading continuation) and can be read as follows: the black numbers represent the predicted negativity level at time Round i (R_i) given that R_i-1 negativity is zero; each horizontal line represents the “actor effect” (the impact of a given player’s negativity on themselves at a later time), and each diagonal line is the “partner effect” (the impact of one’s negativity on their partner at a later time); for actor and partner effects, relationships can be read as “for every one unit increase in negativity from [player] at R_i-1 we expect a unit increase of [number] in [player] negativity at R_i” (e.g., in (a): for every unit increase in negativity from P1 at R_i-1, we expect a unit increase of 0.29 in P1 negativity at R_i). a This model of our immediate response condition shows that the winner of each round (P1 is the winner at R_i in the present model) is influenced by the emotions of both themselves and their partner from the previous round (R_i-1). b Indeed, negativity from both the winner and loser of a given round predict higher blast level selections (grey regression lines show the pair-level of our multi-level analysis). c This model of our forced break condition shows that the winner of each round (P1 is the winner at R_i in the present model) is influenced by their emotions from the previous round (R_i-1). d In contrast to the Immediate Response model, the Forced Break model only displays a significant impact of actor effects on future negativity. d Along these same lines, regression shows no significant impact of players’ emotions on blast level selection (see Table 3 for Bayes factors).

Fig. 8. Mediation analysis of the impact of forced breaks on blast selections.

Mediation Analyses allow us to assess the impact of an independent variable on a dependent variable in isolation and when mediated through another independent variable. In the present model, we are testing to see if the forced break condition influences blast level selections directly (bottom arrow) or indirectly (through negativity). Analysis reveals that, in isolation, the forced break condition predicts lower blast levels (by ~0.36 units on average) than the immediate response condition. However, forced breaks also predict lower negativity than the control group (by ~0.11 units on average). When we assess the impact of forced breaks on blast level mediated through negativity, we see that a one unit increase in negativity predicts an increase in blast of about 1.26 units, while the influence of forced breaks directly is not significant (−0.23) in the mediation model. This suggests that the impact of forced breaks on blast is mediated by its impact on negativity (i.e., the mechanism behind the forced break effect likely lies in negativity).

Fig. 9. Replication of the primary study.

A follow-up to the original study replicated key findings. Specifically, a In the forced break condition, we observe a significant decrease in negative emotion from the time the winner is announced to the time the winner selects a blast level (10 s later). b Under conditions of high negative emotion, blast selections are lower in the forced break condition than the immediate response condition. c Negative expression from both players predicts higher blast level selections from the winner. d Players closely matched each other’s overall mean blast levels, suggesting retaliation as shown in the primary study. Defining Plot Elements. Grey shaded regions in regression plots represent standard errors. Boxplot features are as follows: centre line, median; upper and lower box boundaries, Inter Quartile Range (IQR); upper and lower whiskers, the smallest or largest value within 1.5x the corresponding IQR; dots, outliers.