Men's impulsivity underpins gender differences in aggressive behaviour

Abstract

The relationship between gender and aggression is multi-faceted, with survey research showing that the aggression varies with the gender of both aggressor and target. Here, using a lab-based paradigm that controls the capacity to cause harm and the risk of retaliation, we reveal pronounced gender-specific differences in aggression.Reactive aggression was assessed using a competitive reaction-time task that allowed dyadic pairs to interact face-to-face during conflict. Across two studies (n = 162/trials = 5099) same- and mixed-gender familiar pairs blasted each other repeatedly with noxious sounds, producing an oscillatory pattern of escalating and de-escalating aggression. Analysis revealed lower aggression in dyads made up of two women than mixed dyads or those made up of two men. Crucially, men were an order of magnitude more likely than women to initiate increases in aggression, and brief experimentally imposed delays designed to block impulsive behaviour significantly reduced male aggression. Put simply, when higher levels of aggression were seen, men (usually) started it.

Fig. 1

Gender differences in aggression. (A) Behaviour, averaged across all rounds and gender pair combinations, reveals that overall mean aggression levels were higher for men than women. (B) Overall differences in aggression between men and women reflect an interaction between aggressor and target gender. Specifically, all pairs containing at least one man behaved similarly, but women-only pairs were less aggressive (and these pairings pull down the overall average for women). (C) Aggression over time, averaged by round, shows that women only pairs (white) consistently exhibit less aggression than pairs containing at least one man (dark grey). (D) The probability of men and women initiating an escalation in aggression, averaged across mixed gender pairs, reveals a large gender difference in who is most likely to initiate an increase in aggression. Specifically, men are an order of magnitude more likely than women to raise aggression above the pair’s overall mean.

Fig. 2

Metrics for modelling conflict behaviour. When accounting for an aggressor’s next blast level, we are interested in their previous behaviour, as well as the target’s previous behaviour. We introduce three variables to model previous behaviour in each member of a pair: (A) Level refers to aggression (i.e., blast selection) in the most recent round; (B) Velocity is the change in aggression (i.e., level) across the two most recent rounds; and (C) Acceleration is the change in velocity across the three most recent rounds. We examined each pairs behaviour, over all successive rounds, using a Cross-Lagged Coupled Linear Oscillator Model to produce a stationary account that shows the relationship between current behaviour and each individual’s previous aggression (i.e., level), trajectory (i.e., velocity), and volitivity (i.e., acceleration).

Fig. 3

General cross-lagged, coupled oscillatory model of aggression. A general model representing overall aggression trends based on data from all pairs. The model attempts to account for the aggressor’s next blast level selection (A_R3) based on their preceding level, velocity and acceleration, and their target’s preceding level, velocity, and acceleration. The number to the right of the aggressor’s next blast level is the intercept of the model (2.00), indicating the predicted blast level if all input variables are zero. The number along the paths from each predictor to the aggressor’s next blast level represent the increase in predicted blast level for every one unit increase in the associated predictor. For example, if the aggressor’s velocity increases one unit, we expect a 1.03 unit increase in their subsequent blast. This means that if the aggressor’s velocity is 1, and all else in the model is zero, the predicted blast is 3.03 (i.e., 2.00 + 1.03). All predictors are strongly significant, and the general model explains ~ 74% of variance in the aggressor’s next blast level selection.

Fig. 4

Oscillatory pattern of escalation and de-escalation. A prototypic pattern of aggressive behaviour in one couple, illustrating the contributions that prior level, velocity and acceleration make to the ongoing dynamics of conflict. (A) Each individual tends to match their opponent, but one person may initiate an increase in aggression; (B) Increases in aggression tend to lead to further increases, but the opponent may initially attempt to de-escalate by not retaliating; (C) If attempted de-escalation fails and aggression continues the opponent will then likely retaliate, matching the aggressor; (D) Increases in aggression from the opponent tend to lead to a reduction in aggression from the original aggressor, but the opponent may continue even after the original aggressor has de-escalated; (E) After the opponent has retaliated they may then de-escalate, resulting in a return to matching, and allowing the cycle to continue; (F) The cyclic nature of the conflict is consistent enough across dyads to be visible in a grand-mean plot representing the average blast level at each round.

Fig. 5

cross-lagged, coupled oscillatory model of women’s aggression. A model of aggression with only trials where the aggressor is a woman, allowing us to examine how women behave depending on whether the target is a man or another woman. The model can be read and interpreted as described in Fig. a3. In addition, all predictor paths (and the intercept) include two estimates. Values shown in grey represent women-women pairs, acting as the reference pair absorbed into the intercept. Values in black represent women-man pairs, representing the difference between woman-woman and woman-man pairs (i.e., how the predictors change, rather than an absolute value of the estimate for woman-man pairs). For example, when all predictors are zero and we have a woman-woman pair, the predicted blast level is 1.28. The predicted blast for a woman-man pair (all else being zero) is the base estimate (1.28) plus the difference for woman-man pairs (1.55), resulting in a blast of 2.83. The same addition rules hold for estimates of the influence of predictor variables. For example, a one unit increase in aggressor velocity predicts a 1.03 unit increase in blast level for woman-woman pairs (bringing the estimated blast to 2.31). For woman-man pairs, there is an additional predicted increase of 0.03, such that the estimated blast is 3.89 (i.e., 1.28 + 1.55 + 1.03 + 0.03). Significant predictors are indicated using ** for p < 0.01; *** for p < 0.001 and the model explains ~ 75% of variance in the aggressor’s next blast level selection.

Fig. 6

Men’s impulsivity or women’s non-physical aggression?. We predicted that men would be more impacted by our Forced Break condition due to their greater propensity to engage in impulsive behaviour. (A) As predicted, aggression in pairs containing at least one man was significantly decreased by forced breaks, and aggression in woman-woman pairs was not affected. Note that since our data was between participants, the subtraction visually represented in Panel A is a simulated subtraction built from our distributions (i.e., using bootstrapping), but all reported stats are from our raw data, not the simulated subtraction. (B) Given women’s reported aversion to specifically physical forms of aggression, we analysed verbal aggression from audio recordings of our task. We found that the patterns observed for physical aggression by aggressor-target pairing is identical for verbal aggression, where woman-woman pairs used significantly less aggression than pairs containing at least one man. Data represents counts of verbalizations.

Fig. 7

Breaking the mould. While we present evidence of both gender differences in aggression, and interactions between the gender of the aggressor and target, it is important to recognise that these patterns are averages across a large sample of data. In recognition of the variability within the data, and to avoid misleading claims about the average patterns being a complete description of behaviour, here we highlight patterns of behaviour that break the mould. Illustrative examples of real conflicts between pairs taking part in our studies show that: (A) Although aggression in mixed-gender pairs was predominantly initiated by men (black lines), not all aggression was initiated by men – sometimes women (grey lines) started it. (B) While most woman-woman pairs did not escalate above minimal aggression levels, some women used sustained, high levels of aggression. (C) While aggression was bi-directional and maintained by both individuals in most dyads, sometimes an individual sustained their own aggression in the absence of provocation (and in this case, it was the women in the pair, not the man, maintaining her own aggression).