Integrating the behavioral ecology, social psychology, and neurobiology of aggression

1 Introduction

Aggression is pervasive in animals, including humans. Its outcomes determine individual fitness and drive evolution because winners obtain more resources than losers (Elwood and Arnott, 2012). Tactics of threat, attack, defense, and quitting are controlled by central motivational states. Here, we integrate information from behavioral ecology (BE), behavioral neuroscience (BN), and human social psychology (HSP) to explore these states and their role in behavior.

BE uses game theory to model different tactics as evolutionarily stable strategies (ESS) (Maynard Smith and Price, 1973). An early example was the hawk–dove game, in which hawks threaten each other and then escalate to attack, resulting in costs to one of them. Doves threaten each other, but retreat without cost if attacked. Hawks win resources from doves at no cost and, thus, increase in number until fights between hawks become so common that hawks incur a greater net cost than doves. This settles into stable percentages of hawks and doves in the population that depend on the relative costs of fighting and the value of resources gained by winning. It is these percentages that comprise the ESS. More recent BE models address assessment of resource values, honest/dishonest displays of aggressive intent and resource holding potential (RHP; roughly, the fighting capacity), self and mutual assessment strategies, and escalation vs. quit decisions (Arnott and Elwood, 2008, 2009a). BE modeling has also begun to address offensive vs. defensive strategies (Palaoro and Briffa, 2017), which have very different costs and payoffs (De Dreu and Gross, 2019).

Neuroscience investigates the core subcortical structures and cortical networks that plan, execute, or inhibit aggression. Although this research is not directly grounded in evolutionary considerations, neurocircuits result from natural selection. However, the brain is not infinitely malleable, and these systems operate in “good enough” mode, with consequent constraints on behavior. Neuroscience may inform BE models of resource valuation, fighting ability assessment, fight tactics, energetic cost monitoring, and quit decisions, making them more realistic. However, there may not be a one-to-one correspondence between BE-inferred processes and the operation of specific circuits.

HSP has examined typologies of human aggression and their associated states, including observations in human ethology, research on the frustration–aggression hypothesis (Dollard et al., 1939; Potegal, 2023), and the general aggression model (GAM) (Allen et al., 2018; Kersten and Greitemeyer, 2024). Human aggressive states, including anger and dominance-related aggressive arousal, are instantiated in inherited neurobehavioral networks that are modified by early development and later experiences with peers. Although human aggression is made complex by language, cultural overlay, social organization, and socioeconomic status, it involves the same subcortical and cortical systems as in other animals.

It has long been argued that a range of phenomena, including the temporal persistence of aggressive behavior, aggression contagion, redirection, and motivational override, imply the existence of central states (Potegal and TenBrink, 1984; Potegal and Nordman, 2023). BN is making it clear that aggressive motivational states do exist and account for the escalation, persistence, and respective appetitive and aversive characteristics of offensive and defensive aggression. These circuits support dominance and change in defeat-induced subordination (winner/loser effects).

There have been recent attempts to merge these three areas of research. Mobbs et al. (2021) argue for linking behavioral ethograms to neural circuits and note the neurobehavioral similarities between humans and other animals. Battivelli et al. (2024) called for the integration of BE and BN in relation to fear, aggression, and dominance, and Crump et al. (2020) suggest that emotional/motivational states be incorporated into BE models. One problem for the merger is that the definitions of “aggression” may differ between research areas. A currently accepted definition in social psychology is that aggressive behaviors are those that “the perpetrator must believe … will harm the target and that the target is motivated to avoid …” (Anderson and Bushman, 2002, p. 28). This definition avoids the problem of earlier stipulations of aggression as the delivery of pain, which would have included various medical and dental procedures as “aggressive.” However, this definition is problematic for animals that cannot report their intent and for people who lie about it (“just kidding” is a favorite alibi of bullies). BN and BE study agonistic interactions involving species-typical behaviors of threat and attack in which “aggression” may be formally undefined but obvious to observers. BE in particular focuses on various strategies that include threats that influence the outcome of interactions and, thus, the gain or loss of resources. Our definition of aggression for this essay is: “interactions that may include threat displays and/or physical fights in which participants compete for a resource or status. These are normally intraspecific contests, but interspecific contests do occur.” Although predatory behavior has some features in common with aggression, this is not considered here; however, reference to defense against predators is made, where relevant.

Using this definition, we move to further integrate these three areas, as follows. Aggression typologies of humans and other animals are compared in Section 2. Section 3 presents the rationale for inferring the central motivational states from the respective behavioral phenomena. Section 4 examines the neurocircuitry of aggression and the instantiation of motivational states. Section 5 applies the foregoing information to BE models, showing how current developments in neuroscience illuminate and constrain the components of assessment strategies and subsequent decisions. The terms and concepts used in this essay are intended to serve as introductions to the respective research areas for interested readers who may be unfamiliar with them.

2 Aggressive behavior typologies of humans and other animals

2.1 Intraspecific offense and defense in other animals

Offense and defense, each with its own motivations, behavioral tactics, and neural substrates, are now recognized in BE and BN as two major typologies of aggression in animals. This recognition mostly grew out of observations that dominant male rats living in a colony would attempt an offensive bite-and-kick attack on an unfamiliar rat, which would flip the intruder onto its back (Blanchard et al., 1975, Wall et al., 2003). Intruders defensively reared up, which blocked the attack, and bit whatever part of the attacker came closest, usually its snout. Similar offensive and defensive tactics were subsequently compiled from reports on two dozen muroid rodent species inhabiting widely different ecological niches (Adams, 1980). The rat colony model has since morphed into the now widely used resident–intruder design for rodent aggression, in which a single resident animal lives for a while in an enclosure before a conspecific intruder is introduced.

During this formative period of research, it was rediscovered that two rats foot-shocked together in a small chamber would squeal, rear up, strike each other with their forelimbs, and bite each other’s snout (Ulrich and Azrin, 1962). Although shock-fighting was then proposed as a general model of aggression, these are clearly just the behaviors of defensive rats. Furthermore, these defensive reactions were limited to rats. Shock-triggered offensive fighting in pairs of hamsters, cats, and monkeys continued long past shock termination. Shock-fighting is no longer widely used. The same investigators noted that pigeons would attack conspecific targets when the food reward for key pecking was terminated (Azrin et al., 1966), an effect now labeled “frustrative non-reward” (FNR), which occurs in other species, including humans, and remains of great interest (Potegal, 2023).

Early experimental evidence for specific neurohormonal substrates for different types of aggression included observations that castration affected offense more than defense in rodents and that the opioid antagonist naloxone reduced offense, but increased defense (reviewed in Potegal and Nordman, 2023). Stimulation of the medial hypothalamus in opossums and cats elicited piloerection and a hissing and/or growling defensive attack directed to any nearby target (Wasman and Flynn, 1962; Roberts et al., 1967). Such defensive aggression was quite different from cats’ quiet, directed biting attack on small prey elicited from the lateral hypothalamus.

The offense–defense distinction extends to primates. Among chimpanzees, offense is signaled by piloerection, while defense is not. Defensive screams in vervet monkeys and chimpanzees are higher pitched than the respective offensive screams (Mercier et al., 2019; Slocombe and Zuberbuhler, 2005). The overall generality of these two typologies, at least across vertebrate taxa, is suggested by the conservation of threat–attack sequences among 18 species of Old World Cyprinodont fish, which accord with their phylogenetic relationships (Ewing, 1975). Taxonomically wider comparisons found similar agonistic sequences in rats and rhesus monkeys (Adams, 1981) and in African naked mole rats compared with cats (Nathaniel et al., 2006).

2.2 Offense establishes and maintains dominance

As suggested by the original observation that dominant rats mounted offensive attacks, offense is initiated in the context of obtaining/maintaining territory and/or social dominance. In contrast to asymmetrical offense/defense conflicts, the clash of antlered, horned, or tusked adult males is an example of mutual offense that determines which one will become dominant and have access to females. These offensive attacks are typically preceded by visual and/or auditory threat displays, especially so for animals with weapons (Palaoro and Peixoto, 2020). Direct observations of fighting and post-fight wound locations have shown that these attacks are directed at particular body parts, such as the flank in cervids and bovids, the dorsum for the bite-and-kick tactic in muroid rodents, the neck in cats and bears, and the head in pigs and birds (Brain and Hui, 2003). Analyses of fight recordings showed that species-typical offense postures and acts are tactics to gain access to the target areas (Pellis and Pellis, 2015). The head is the preferred target in human physical aggression, as observed in street fight videos (Potegal et al., 2025).

Offensive aggression maintains dominance. Alpha male rats begin each day with brief attacks on subordinates that are not contesting their dominance (Blanchard et al., 1988). Silk’s (2002) aptly titled review of baboon behavior, “Practice random acts of aggression and senseless acts of intimidation,” documents apparently arbitrary, unprovoked attacks by dominants on subordinates. It is argued that such intermittent offense communicates that the dominants are not bluffing and preemptively inhibits challenges from potential rivals (Table 1).

Table 1. Equivalence of aggression typologies.

In contrast to the relatively restricted repertoire of offense tactics, there are a multiplicity of fear-motivated defense maneuvers that depend on whether the threat is a conspecific rival or a predator (Table 1). The general defense against predators has several behavioral stages that are relevant to this essay because equivalent behavioral stages occur in people at risk of being assaulted, as noted below.

2.3 Typology of human aggression and equivalences to other animals

The development of human aggression typology lagged that of other animals. Dodge and Coie’s (1987) Reactive/Proactive Scale quantified previous informal distinctions between children’s anger-motivated affective reactions to frustration or perceived provocation, such as being teased, and their more deliberate, hostile but not angry acts, such as bullying and dominating others by force. These were denoted reactive and proactive aggression, respectively. Proactive aggression at any age now refers to deliberately harmful acts committed for some sort of gain and generally without emotion (at least, without anger). Importantly, the gain need not be material goods, but can be social status and a sense of control over others, i.e., dominance. Proactive and reactive aggression and aggressors are described in more detail below. Equivalences between human and animal typologies are shown in Table 1. In particular, we present the rationale for comparing offense in other animals to human proactive aggression, in contrast to the previously proposed analogies between offense and reactive aggression (Feltes and de Boer, 2020; Potegal and Nordman, 2023). Human defensive reactions against assault, which we describe below as a fear-motivated form of reactive aggression, were found to follow the sequence of defensive responses to predation in other animals considerably later (Blanchard, 2017).

2.3.1 Proactive aggression

Precursors to proactive aggression can be seen in infants as young as 6 months who hit and bite parents and sibs without being provoked and while showing no signs of distress (Dahl, 2016; Hay et al., 2018). The distinction between children’s provoked and unprovoked aggression can be reliably made by age 2 (Peets and Vecchio, 2023). Large size, stimulation-seeking, and fearlessness at age 3 predict aggressiveness at 11 (Raine et al., 1998). Some preschoolers threaten, push, hit, and kick other children and take their toys (Ostrov et al., 2023). Slightly more than a third of early adolescent boys around the world report having been in a physical fight in the last year (Smith et al., 2024). A mean 10% of male adolescents in 79 countries affirmed fighting frequently (Elgar et al., 2015). These fights can begin with unprovoked teasing and insults, disputes over rules and possessions, and dominance-related “honor” contests (Jackson-Jacobs, 2013; Malette, 2017; Runions et al., 2018).

Using verbal skills unique to humans, young men on the street, in bars, and elsewhere employ a repertoire of insults to challenge others and display dominance. Verbal challenge sometimes escalates to physical fighting. In other situations (Dane et al., 2022; Griffiths et al., 2011), men of roughly equal social status goad each other into fights over trivial matters, their real motivation being “impression management” (Felson, 1982), i.e., the assertion of dominance in these “character contests” (Toch, 2017).

2.3.2 Teasing and bullying as low-cost proactive aggression

Teasing is common across cultures and can be a mild (or not so mild) form of proactive aggression (Dawes and Andrews, 2025). Apes tease each other as well by offering and then withdrawing desired objects, by being provocatively non-compliant with routine expectations, and by disrupting the activities of others (Laumer et al., 2024). They sometimes engage in such potentially antagonistic teasing without assuming the usual placating playface.

The study of bullying was initiated by Olweus’ 1978 book “Bullies and whipping boys,” which describes the sociology and psychology of bullying in Swedish schools. Bullying is now defined as belittling, intimidating, and coercive or physically aggressive behavior repeatedly perpetrated by a more powerful person or group against a less powerful individual. Bullying is clearly an example of proactive aggression, but with little or no retaliation cost. It typically takes place in front of an audience and is a deliberate display of dominance for their viewing.

Animals bully as well. While reciprocal fighting in pigs can result in wounds on the anterior thirds of both animals, attacks by dominants on subordinates leave lesions on undefended posteriors (Turner et al., 2006). Indeed, dominants selectively harassing subordinates have been observed in nine mammalian and eight bird species, including common waxbills in which dominant birds attack the lowest-ranking members of the large flocks, especially when the audience includes relative strangers (Beltrão et al., 2023).

2.3.3 Valence of offense and proactive aggression

In early studies, fighting fish and cocks learned routes in mazes to threat-display to mirror images or conspecifics (Golden et al., 2019). Many subsequent studies have demonstrated that rodents would cross electrified grids, acquire “place preferences” in mazes that hold an attackable target, or perform operant responses of lever pressing or poking their noses into holes for the opportunity to attack a conspecific, if they could win (with sex differences in appetitive aggression in some species) (Aubry et al., 2022; Börchers et al., 2023). Offensive attack was found to be less reinforcing than food reward for the majority of animals, as indicated by the relative running (or swimming) speeds, choice frequencies, or operant schedule “breakpoints” (the schedule on which animals stop responding because the time or repeated effort required for reward has been made too great) (Bols and Hogan, 1979; May and Kennedy, 2009).

The positively reinforcing value of aggression is likely involved in the phenomena of “social instigation,” in which mere sensory exposure to a conspecific excites aggressive motivation, and that of “attack priming,” in which aggressive motivation is increased by the sensorimotor act of committing the first attack (Miczek et al., 2013; Potegal and Coombes, 1995). Motivational effects also appear in the converse phenomenon of aggression satiation after repeated ad lib attacks, as when initially aggressive female hamsters increasingly avoided successively introduced conspecific intruders and when initially muricidal male rats increasingly attempted to bury rather than bite successively introduced mice (Potegal, 2024).

Similar to offense in animals, proactive aggression is rewarding to some people who become aroused by and enjoy watching blood sports and playing violent video games. Engaging in actual aggression is only mildly rewarding for the majority of people, but more so for those with histories of antisocial behavior (Alvarado and Ramírez, 2014; Chester et al., 2021). Bullies do enjoy their acts of bullying and their victim’s fear and pain reactions (Boldizar et al., 1989; Connell et al., 2016; Fluck, 2017; Lee, 2010). They are “happy victimizers,” an attitude that may or may not be restricted to childhood (Gutzwiller-Helfenfinger and Latzko, 2020; Jambon et al., 2021). College campus “fight seekers” are a small, older group who frequently drink to inebriation, then instigate arguments that escalate into fights; they report feeling calmer and better after the fight. Similarly, a minority of mice choose to attack male conspecifics rather than copulate with a female; their motivation for aggression increases after they drink sweetened alcohol (Miczek et al., 2022). For other examples of the reward value of proactive aggression, see Quansah and Gagnon (2025).

Recreational fighting throughout history has been documented (Ingle and Cross, 2004). Similarly to a mouse nose-poking for the opportunity to aggress, contemporary bar fighters make deliberate choices such as going into a crowded bar where they know someone will bump into them sooner or later, giving them an excuse to start a fight. Such eagerness to fight likely involves some combination of increased self-esteem and/or sense of control, pride in fighting prowess, and showing off to peers. At a much greater extreme, Weierstall and Elbert’s (2011) Appetitive Aggression Scale quantifies the hedonic valence of severe aggression and outright cruelty for youth who were recruited or abducted into guerilla groups or warlord militias as children or adolescents, given drugs and brutally trained to maim and kill, and were traumatized by combat experience (cf. Bigirwa et al., 2025).

2.4 Human reactive aggression

As noted in Table 1, human reactive aggression can be elicited by discomfort and pain and/or motivated by anger, fear, or some combination. Stimuli that have been used to elicit aggression in the laboratory include foul odors, loud noise blasts, air blasts to the neck, finger pinch, or hand or foot immersion in ice water, which generate hostile feelings and thoughts and augment aggressive reactions (Groves and Anderson, 2016). Shock has also been used to elicit aggression in people, but often in competitive paradigms in which the participants’ attributions of their opponents’ hostile intent influence their level of aggression, sometimes more than the shock itself (Greenwell and Dengerink, 1973). Excessive heat is the most widely documented condition that increases real-world human aggression (Hsiang et al., 2013). In terms of emotions, human fear-motivated reactive aggression most closely resembles animal defense, but occurs relatively rarely. Anger is the most common motivator of reactive aggression.

2.4.1 Anger

Anger is provoked most frequently and severely by interpersonal negligence, disrespect or threat to self-esteem, and reputation or harm to the self, valued others, or possessions (Kashdan et al., 2016). In the classic discrete emotions theory, anger is one of six primary emotions, each of which is distinguished by co-occurring features including facial and body expression (e.g., lowered brows, compressed lips, clenched fists, rapid forward arm, and body movements for anger); autonomic reactions (increased heart rate, blood pressure, and facial flushing, sometimes associated with sensations of heat in the face, chest, and hands); action tendencies (approach or attack); and distinct subjective feelings (a “remarkable” energy, impulses to set limits on the offender and to “counteract or destroy something,” according to a team of psychologists trained in introspection) (Hackert et al., 2021). Objective behavioral studies confirm that impulsive risk-taking and optimism about the outcome of the risks taken are characteristic of anger in children, adolescents, and adults, particularly in men (Gambetti and Giusberti, 2016; Ferrer et al., 2017; Stephens and Groeger, 2011, but see Matsumoto and Wilson, 2023).

A tenet of the discrete emotions theory is that emotions evolved as biologically based aids to survival and are therefore pan-cultural. The majority of research supports this proposition in finding similar emotional facial and vocal expressions across cultures (Hwang and Matsumoto, 2014; Laukka et al., 2016). Other, more complex, views of emotion (Barrett, 2017) are beyond the scope of this narrative; however, the majority of neurophysiological studies of anger assume the simpler, classic version, either implicitly or explicitly.

2.4.2 Differences between reactive and proactive aggressors and aggression

Anger-prone, reactively aggressive youth are typically suspicious of others and exhibit “hostile attribution bias” (the tendency to interpret neutral or ambiguous behavior of others as being antagonistic). Easily redirecting their anger from one person to another, they often have poor social skills and meet with social rejection. When the reactive/proactive distinction was applied to Barbary macaques, reactive aggression was found to be characteristic of low-ranking animals (Paschek et al., 2019). Proactive aggressors have better social skills and affiliate with peers, e.g., as clique or gang members or leaders (Jung et al., 2017). Recent classification studies suggest that there are more or less purely reactive individuals, but that proactive aggressors also engage in reactive aggression (e.g., bullies also get angry). The latter combination yields more severe aggression overall and accounts for the relatively high correlation between the two types of aggression (Brugman et al., 2017; Euler et al., 2017; Evans et al., 2021; Smeets et al., 2017).

The genetic bases of reactive and proactive aggression overlap considerably, but with the latter showing greater age-related genetic influence (Paquin et al., 2017; Tuvblad et al., 2009). The two have rather different profiles of peripheral autonomic activity (Thomson et al., 2021). Centrally, reactive aggression is associated with hypo-serotoninergic function, while proactive aggression appears hyper-serotoninergic (Romero-Martinez et al., 2022). However, both are regulated by overlapping cortical/subcortical networks.

Laboratory paradigms that reliably provoke reactive aggression include the Taylor aggression paradigm (Taylor, 1967), which uses shock, and the newer point subtraction aggression paradigm (Cherek et al., 1997) and ultimatum game situations (White et al., 2014), which use point loss and unfair offers, respectively. Proactive aggression is largely internally driven; therefore, providing encouraging opportunities for those motivated to display dominance is a more difficult methodological challenge (Boccadoro et al., 2021; Lobbestael and Cima, 2021). It may be easier to distinguish the two types in real life, as indicated by the high consistency with which nurses and their aides categorized the aggressive acts of adolescents in a psychiatric hospital (Vitiello et al., 1990).

2.4.3 Do other animals experience anger? Phenomena of frustrative non-reward

The argument about emotions as evolved aids to reproduction and survival can be extended to animals that face similar challenges of predator avoidance and social competition (Crump et al., 2020; Zych and Gogolla, 2021). The facial action coding system used to assess human expressions has been applied to other primates. The action of the orbicularis oris pars medialis muscle in narrowing and tightening the red margins of the lips happens only in human anger (Kret et al., 2020). Bonobos show a tense mouth “lip press,” indicating retaliative aggressive intent that has been construed as equivalent to anger. However, anthropomorphizing the experiences of animals is risky. Thus, cats perceived to be “playing” with prey are actually switching between predatory and defensive actions (Pellis et al., 1988). Pigs winning or losing a fight did not differ in valence as judged by experienced pig farmers; counterintuitively, winning pigs were seen as more stressed (Camerlink et al., 2016). There is more support for the proposal that FNR can serve as an animal model of human anger (Potegal, 2023).

Frustration (“goal blockage”) is a special case of an aversive stimulus–response situation. The frustration–aggression hypothesis, originally proposed to explain all human aggression (Dollard et al., 1939), became tautological in reducing all aggressive motivation to frustration. Frustration is now studied in cases of unambiguous thwarting, e.g., “road rage” resulting from traffic delays. For other species, observers familiar with animal behavior report that hens give a long, harsh, species-typical “gakel” call when frustrated (Zimmerman et al., 2000), whereas horses make a particular rotational head toss and pawing the ground, kicking, and biting (Pannewitz and Loftus, 2023; Torcivia and McDonnell, 2021). Chimpanzees have temper tantrums (Proctor et al., 2013). However, the key cross-species finding is of FNR in equivalent situations, including responding on operant schedules that produce reactive aggression and other associated behaviors in humans, other mammals, birds, and fish (Potegal, 2023). FNR increases aggression in small rainbow trout facing a larger opponent (Vindas et al., 2014), while mice responding in an FNR situation attack more severely than they do in a standard resident–intruder encounter (Covington et al., 2019). FNR unequivocally makes many people angry, with effects so powerful that it has been adopted as a cross-species, transdiagnostic construct in the negative valence system domain within the National Institutes of Health (NIH) research domain criteria. Therefore, with caveats, the FNR of other species can be a stand-in for human anger-driven reactive aggression (Potegal, 2023).

2.4.4 Valence of FNR, anger, and reactive aggression

As is obvious, frustration is aversive, and animals avoid such situations when they can (Daly, 1969; Haskell et al., 2000). Anger is an aversive experience for the majority of people who tend to avoid anger-provoking situations. In this respect, it resembles defense. Even reactively aggressive people prefer monetary rewards over retaliating against someone who provoked them (Gan et al., 2016). However, anger can make people feel stronger, more powerful, and, for men, more in control (Litvak et al., 2010). They may experience relief and even satisfaction after angry retaliation against an offender (Chester and DeWall, 2016). Hence, anger valence is shown as −/+ in Table 1. Psychologists trained in introspection have reported that expressing their everyday feelings of anger produced a sense of satisfaction, followed by one of anergic fatigue (Hackert et al., 2021).

2.5 Defense against attack

To survive, animals and humans must evade, resist, or escape from attacks by predators and hostile conspecifics. Importantly, defensive reactions are generated in the exigent moment and therefore are unlike offense and proactive aggression, in which decisions to engage may be extended over time. In this section, we consider defensive reactions in animals and their human behavioral equivalents along with the associated human emotion states of anxiety, fear, and panic.

2.5.1 Attack imminence

The attack imminence sequence, also known as proximity-dependent defense, has replaced Cannon’s fight/flight dichotomy and is operative in many species (Fanselow and Hoffman, 2024). When first detecting a threat, animals respond with parasympathetically mediated freezing. They may hide if refugia are available. For example, rats exposed to a cat outside their burrow emerge after hours of hiding and explore warily in an anxious seeming “stretched attention” posture (Blanchard et al., 1991). After a fear-inducing stimulus is withdrawn from an experimental apparatus, mice crouch, pause intermittently while exploring, and prolong thigmotaxic contact with walls (Kennedy et al., 2020). If a potential threat approaches, animals flee. Both freezing and flight have been taken as signs of fear. Behaviors interpreted as akin to human panic include startling, then freezing in rats and abrupt, random-seeming zigzag flight and jumping in mice (Smith-Vickery et al., 2024; Hoffman et al., 2022). If the threat closes in, animals may turn and defensively counterattack, the classic example being the “cornered rat’s” leap into an attacker’s face. In the laboratory, the distance at which wild rats counterattack when no escape is possible is approximately 50 cm (Blanchard, 1997). These tactics work: 90% of cats’ withdrawals from prey were associated with the prey’s counterattack (Adamec et al., 1980).

2.5.2 Human defensive anxiety, fear, and panic

The sequence also applies to people who can report the associated emotional reactions (Blanchard, 2017). Surveys in Brazil, Hawaii, and the UK that presented people with threat scenarios found that, when a hypothetical threat was mild, distant, or ambiguous, such as hearing voices outside the window, people actually recalled or pictured themselves as being anxious and assessing the situation with tense alertness. If the real or imagined threat was unambiguous and intense, but escape was possible, such as someone at a distance coming toward them with a knife, people reported, or envisioned, running away in fear. People freeze in response to a sudden threat, which has been measured in the laboratory as a reduction in the subjects’ body sway when they were confronted by pictures of angry faces looking at them or guns pointed at them. In response to survey scenarios in which escape was not possible, such as being attacked in an elevator, people imagined screaming and fighting back by scratching and biting. In real life, human targets of violence are similar to defensive rats: they bite their attackers on the hand, arm, shoulder, or face (Potegal and Nordman, 2023). Under dire threat, however, some people experience maladaptive panic reactions, which can include frantic, ineffectual attempts at escape and various autonomic responses and subjective experiences.

2.5.3 Valence of defense and defensive aggression

Fear motivates all stages of the attack imminence response sequence. Defensive aggression motivated by fear and/or pain is avoided when possible. Rats in shock-fighting studies escaped the chamber rather than fight, at least when there was a way out (when “escape” required learned operants, defensive attack became the prepotent response) (Potegal, 1979).

2.6 Threat displays and retaliation costs

Offensive threat displays that cause opponents to quit (flee or submit) without fighting and defensive threats that forestall attack are extremely common. In fact, majority of contests consist of threats by one or both contestants and do not escalate to physical aggression. Some displays are “honest” advertisements of RHP. For example, the lateral displays of many fish show their size to potential challengers, as does deeper pitch in the calls of various frogs and mammals (Bee et al., 1999; Miller and Engstrom, 2012). Contests between red deer stags during the rut begin with roaring, which is energetically demanding and therefore an indicator of strength (Reby et al., 2005; cf. Raine et al., 2018). In contrast, threat behaviors including fish fin erection, lizard head bobs, and canid snarls convey aggressive intent. However, performable by all animals regardless of true RHP, such threats can be a “dishonest” bluff. Maynard Smith (1982) argued that, because bluffing is energetically “cheap” and may sometimes deter rivals, animals should evolve the strategy of threatening every challenger, whatever their own real intent. Threat would then contain no information about the animals’ true willingness to fight, prompting evolution of the counterstrategy of ignoring threat behaviors, after which these behaviors should disappear. Obviously, they have not. Investigations prompted by this apparent paradox showed that bluffing is kept in check by “retaliation cost.” That is, a threat may not only elicit a counter-threat but can also provoke an actual, pre-emptive counterattack (Maynard Smith, 1982). Consequently, threat is often an “honest” advertisement of intent that does predict subsequent attack, e.g., in the dramatic “zebra display” in cuttlefish (Adamo and Hanlon, 1996) and in the changes in body and/or eye color or gill erection in various fish (see Potegal, 1994, for citations). Eventual winners among green anoles have shorter latencies and higher frequencies, durations, and vigor of nearly all display components (Wilczynski et al., 2015). In several bird species, low-amplitude “soft songs” reliably predict escalation to attack (Ali and Anderson, 2018; Krieg and Burnett, 2017; Moran et al., 2018). The proposed general criteria for identifying aggressive intent in displays include: 1) displays that become more frequent in aggressive contexts; 2) they predict displayers’ subsequent aggressive escalation; and 3) opponents respond differently to them than they do to other signals (Searcy and Beecher, 2009). However, bluffing also occurs (Wilson and Angilletta, 2015). How escalation from threat to attack is mediated by central states and how BE models animals’ assessment of threats are addressed in Sections 3.4 and 5.5, respectively.

Retaliation cost also plays a key role in human aggression. For example, Sell and Sznycer (2021) conjectured that angry facial and vocal expressions force opponents to “re-calibrate” the cost of their encroaching or provocative behavior. Several lines of evidence suggest that we respond to perceived malice, unarmed attack, or potentially lethal armed assault with increasing counter-aggression when we feel adequately powerful or well-armed, but are deterred when the opponent is perceived as having a high formidability or when the cost of obtaining sufficient power to escalate increases. Retaliation cost estimations have been found in scenarios of college undergraduates imagining conflict with less or more formidable antagonists (Winstok, 2007) or playing hawk–dove games for monetary gain/loss against less or more challenging opponents (Benard et al., 2017) or supposedly shocking an insulting peer who might or might not be given the opportunity to retaliate (Diamond, 1977), as well as in actual bar fights between unarmed college students or between armed bikers (Graham et al., 2006). However, some aggressive boys are more likely to fight when retaliation is anticipated, which indicates the strength of their proactively aggressive motivational state (Peterson, 1971; Boldizar et al., 1989).

3 Arguments and evidence for the existence of aggressive motivational states

We propose that both defensive/reactive aggression and offensive/proactive aggression are instantiated in motivational states within the central nervous system (CNS). In humans at least, the main motivation for defensively reactive aggression is fear, but the motivation for most of reactive aggression is anger, which prompts engagement in risky conflicts (Fessler, 2010). However, offensive/proactive aggression is risky as well; therefore, motivation is required to overcome fear of fighting. Proximal neurobehavioral motivational states that overcome fear have evolved under selection for dominance (Potegal and Nordman, 2023). In humans, this state has been termed “non-angry aggressive arousal”; for other animals, the separate term “angriffsbereitschaft” has been used to acknowledge that the states in us and them might not be equivalent. However, for simplicity in this essay, we will use the term “aggressive arousal” to refer to this central state in other animals.

3.1 Central states: persistence, scalability, and intensity

Such aggression-related central motivational states are increasingly being invoked, both for animals and ourselves (Flavell et al., 2022; Hashikawa et al., 2017; Hoopfer et al., 2015; Kennedy et al., 2020; Kersten and Greitemeyer, 2024; Kunwar et al., 2015; Minakuchi et al., 2024; Nair et al., 2023). Central aggressive states operate at lower levels of physiological organization, e.g., adjusting respiration and preparing for appropriate muscular activity, reflexes, and modal action patterns (MAPs). At higher levels of behavioral decision-making, they: 1) direct attention to and heighten the sensitivity for aggression-relevant stimuli; 2) increase the probabilities of behaviors that anticipate and seek opportunities for offense and avoid defensive situations; and 3) reduce competing or conflicting behavioral tendencies for, e.g., exploration, feeding, or sexual behavior. Among their key properties, motivational states persist over time and are “scalable,” meaning that they can vary in intensity or magnitude. Higher-intensity states motivate more effortful, riskier, and/or more severe behaviors. The prepotency of these states is indicated by the fact that an intruding conspecific can elicit them in rodents whose heads are fixed in a stereotaxic apparatus (Ferris et al., 2008; Vinograd et al., 2024).

Accordingly, this section describes the ethological distinction between motivating and releasing stimuli, then briefly reviews five offensive/proactive aggression-related phenomena that are most parsimoniously explained by motivational state. Measures of motivational state are noted, as are conjectured increases in motivation that underlie threat–attack transitions.

3.2 Stimuli evoking appetitive aggressive motivation

Ethologists distinguished between motivating stimuli that somehow “activated” particular behavior sequences and “releasing” stimuli that triggered their execution. Thus, aggressive motivation was evoked in Siamese fighting fish by a fish-shaped model fitted with small lights located to emulate supernormal gill-flaring threat when lit (Romano et al., 2017). Similar motivation was elicited in pigeons by a model that emitted species-typical aggressive/defensive “deep-throated” coo vocalizations (Cross and Goodman, 1988), while the pheromones brevicomin and thiazoline secreted by the preputial glands on the tip of the penis have related motivational effects in mice (Trouillet et al., 2019). Movement of the model or real conspecific was the releasing stimulus for offensive attack in all three species [see “social instigation” effects (Miczek et al., 2013) and Section 2.3 in Potegal and Nordman (2023) for other examples].

We propose that such species-typical visual, auditory, and olfactory motivating stimuli interact with endogenous physiological factors, such as the hormone levels, to engender a central state of aggressive arousal. As ethologists also recognized, trade-offs can occur between the aggression-inspiring “adequacy” of stimuli and the intensity of the motivational state sufficient to trigger behavior. Thus, wild rats being foot-shocked attacked inanimate objects that they would otherwise not bite (Galef, 1970). Similarly, people who are already angry require only a little additional provocation to react aggressively, as in the phenomenon of triggered displaced aggression.

Such studies may create the impression that an intense central state motivation is irresistible, inevitably driving attack. Furthermore, in animals chosen for their propensity for aggression, confined in an experimental apparatus with no distractions or opportunities for alternative behaviors, and optimally challenging targets with minimal retaliation cost, offensive attack does indeed follow. However, in real-life situations with multiple opportunities and risks, animals with varied histories of victory and defeat have information about their own capabilities and some ability to estimate those of their antagonist. In these situations, motivational activation may increase the probability of threat and attack, but they remain contingent on multifactorial decision-making, which may take some time. BE has modeled such motivated decision-making, as described in Section 5.

3.3 Five phenomena associated with aggressive states

The following aggression-related phenomena are parsimoniously explained in terms of central states (Potegal and Nordman, 2023).

3.3.1 Warm-up, escalation, bouts, and persistence

The rates and the severity of the attack can increase in the presence of an unchanging provocation, and a heightened attack probability can outlast withdrawal of provocation. Classic demonstrations include increasing rates of attack directed by three-spined sticklebacks toward an unreactive conspecific in a small tube (Peeke and Dark, 1990). These short-term increases in aggression were originally conceptualized as “response sensitization,” but must reflect changes in aggressive arousal within subjects due to the provoking stimuli being non-reactive.

A variant of warm-up effects is an escalation of aggression when the same or similar provocation is presented in successive discrete trials. In rodents, a two-trial attack-priming paradigm showed that a single attack on a first “priming” intruder reduced the latency and increased the probability of attack on a second “probe” intruder (Potegal and Coombes, 1995). Fish also showed attack priming in discrete trial situations (Potegal and Nordman, 2023). In humans, demonstrations of two trial escalation included 7-month-olds who showed more intense anger the second time their mother took a biscuit from them and 15-month-olds whose protests became significantly more probable, rapid, and intense during a second 1-min trial of gentle arm restraint (Potegal et al., 2007).

In both resident–intruder situations and unconstrained fighting, attacks occurred in bouts separated by inter-bout intervals, during which animals engage in non-social behaviors (Potegal and Nordman, 2023). These pauses suggest that a need for repeated respites from peripheral and/or central fatigue causes temporal variations in aggressive arousal. Central state temporal characteristics in fish and rodents include aggressive arousal persisting past the removal of an opponent. After Matadi cichlids were primed by briefly viewing a conspecific intruder in Hieligenberg’s (1965) seminal study, their aggressive arousal persisted for approximately 10 min, as indicated by their increased attack rate on the small, blind fish swimming in the tank. Investigations of aggressive arousal persistence in rodents revealed two phases of shortened attack latencies: a first phase lasting up to several minutes and a second phase lasting up to a few hours (Potegal and Nordman, 2023).

Non-angry aggressive feelings, physiological reactions, and behavior evoked by violent movies or video games may last up to 5–10 min (Barlett et al., 2009; Sestire and Bartholow, 2010). Human non-angry aggressive arousal generated in more naturalistic situations might last longer, but has not yet been measured. More is known about the duration of anger.

3.3.2 Anger escalation and duration

Analogous to aggression warm-up with continuing provocation in animals, people whose efforts on a laboratory task were met by continuous non-cooperation by the experimenter’s confederate escalated from mild requests to impatient demands, then complaints, angry statements, threats, and harassment in a progressive fashion so systematic that it met the Guttman scalability criteria (Pruitt et al., 1997).

In discrete trial-like situations, a transient frustration or provocation generates anger that outlasts the event. Anger in these episodes rises quickly and falls slowly. The generation of anger involves assessment in a sequence of rapid, unconscious “appraisals” that determine the offender’s intention (e.g., joking or serious) and blameworthiness (e.g., sober or drunk). Once provoked, anger typically persists for a few minutes up to approximately 3 h (Verduyn, 2021). As might be expected, the duration increases with anger intensity (Verduyn and Lavrijsen, 2015). Abnormally persistent anger is a significant contributor to the antisocial problems of irritable children, adults with borderline personality disorder, and violent criminal recidivists (Acland and Cavanagh, 2023; Jacob et al., 2008; Waschbusch et al., 2023).

Anger drives almost all reactive aggression; nevertheless, importantly, many anger experiences are not outwardly expressed, and a few result in overt verbal or physical aggression (Averill, 1982). That is, the action tendencies of anger are often inhibited. Hence, anger is truly an internal state.

3.3.3 Redirection of aggression

The persistence of a motivational state is required to explain the redirection of aggression in which an animal in an aggressive interaction subsequently redirects an attack on a previously uninvolved conspecific. Experimentally, an intruder presented to one of a pair of peacefully neighboring fish or cohabiting rats elicited aggression redirected to the companion animal (Alberts and Galef, 1973; Peeke and Dark, 1990). When waxbills are attacked by unfamiliar birds, they redirect their aggression to known flockmates (Pacheco et al., 2025). Redirection occurs in primate troops when a dominant attacks a subordinate that then seeks an even lower-ranking animal to attack. Redirected attack protects reputation in these groups: animals that do not redirect aggression are attacked by others more often than those that do (Ito et al., 2018; Tokuyama and Furuichi, 2014). For redirection to occur, aggressive arousal must persist throughout the delay between the original provocation and locating an appropriate target. This delay can be at least 5 min in both hyenas and gorillas (Engh et al., 2005; Watts, 1995).

Redirection of reactive aggression has been observed in human children (Butovskaya and Kozintsev, 1999), adolescents (Kidman et al., 2025), and adults (Wang et al., 2020; Wiederhold and Bühren, 2025), including a variant “triggered displaced aggression” in which people respond with excessive anger to a minor provocation after previously being seriously insulted (Lagios et al., 2025). Bystanders to real fights, particularly those who intervene on behalf of losers, risk being targets of redirected aggression (Jacobs and Wright, 2010; Potegal et al., 2025). Police officers intervening in fights may be subject to redirected attack by one or both combatants (Toch, 2017). In animals and humans, too, aggression may be redirected to the original offender’s kin (Engh et al., 2005; Jacobs and Wright, 2010; Tiddi et al., 2017).

3.3.4 Contagion of aggression

Contagion is said to occur when the expressed emotions of one or more individuals in the group are emulated by other group members. Aggressive arousal in animals can be induced by seeing or hearing conspecific fights. Such contagion occurs in fish (Clotfelter and Paolino, 2003; Desjardins et al., 2015) and non-human primates (Schino and Sciarretta, 2015). Fighting in primates, which is triggered by auditory cues, is an especially clear case of aggressive arousal induction because such signals provide no visible behavior to be mimicked (Watson and Caldwell, 2010; Videan et al., 2005).

Aggression can be contagious in people, as observed in child daycare classrooms, in adult workplaces, and among sports fans and players, resulting in social disruption and injury (Block and Lesneskie, 2018; Goldstein et al., 2001; Petitta et al., 2019). Contagion effects may be why people in groups are often more violent than individuals acting alone (Meier et al., 2007; Nawata, 2024). Anger prompts some of this contagion. Studies have confirmed rapid, automatic mimicry of others’ angry facial expressions and the corresponding shifts in felt emotion (Kelly et al., 2016; Olszanowski et al., 2020).

3.3.5 Motivational override

“Spite” refers to the anger-motivated infliction of punishment that is costly to the punisher (Marcus and Norris, 2016). Anger-motivated spiteful behavior is reliably elicited from a subset of individuals in exercises such as the Ultimatum Game, in which a player who feels that his partner is offering too small a portion of the reward, such as a small amount of money, cancels reward for both parties, thereby depriving himself of any reward in order to deprive the partner of his unfairly large share (Martínez and Maner, 2024).

In animal fights, losers’ species-typical signals of submission are often effective in stopping winners’ attacks (Reddon et al., 2021). However, aggressive arousal may override these inhibitory effects, e.g., lobster and polecat winners continue their energetically costly attacks or chases despite losers’ submission or flight (Cushing and Reese, 1998; Poole, 1974). The strength of motivational override was demonstrated experimentally by pigeons and rats that were trained to attack a restrained target for access to food or water. When two such food- or water-deprived animals were placed together, they attacked each other, ignoring the needed resource (see Potegal, 1994, for more examples).

3.3.6 Reduction of anti-predator vigilance: narrowing of cognitive capacity

Another important form of motivational override is a maladaptive and potentially lethal reduction in anti-predator vigilance in animals engaged in intraspecific conflict. Such effects have been documented in fish, reptiles, birds, and mammals (Díaz-Uriarte, 1999; Dunn et al., 2004; Jakobsson et al., 1995; Ota, 2018). Reductions in vigilance are presumably due to heightened aggressive arousal that narrows the focus to the immediate antagonist and immediate situation and reduces the depth of cognitive processing (Alia-Klein et al., 2020; Gable et al., 2015; Gummerum et al., 2022).

3.4 Motivational states and threat-to-attack escalation

Some animals show “scaling” in display amplitude or intensity, e.g., the deepening pitch of frog calls (Reichert and Gerhardt, 2013). Other species employ a variety of threat signals (Hurd and Enquist, 2001). In the simplest linkage of motivational state intensity to overt behavior, these displays will be ordered according to increasing energetic demand, antagonist proximity, or other relevant dimension of effort and/or risk and be associated with a progressively higher probability of ensuing attack. Examples include early-stage, frontal display performed at a distance by cuttlefish followed by shovel and lateral displays at closer proximity (Schnell et al., 2016), “tret” calling and then high-syllable rate singing in common chiffchaffs (Sierro et al., 2023), and soft song followed by wing-quivering in male house wrens (Barnett et al., 2014).

In any case, the majority of threats are not followed by an attack. Animals may quit before the motivational state reaches the level to trigger attack, likely due to an assessment that the resources at stake are not worth the risk or that the opponent’s RHP is too great (Elwood and Arnott, 2012). Conversely, attack sometimes occurs without prior threat. This might happen when the resource is very valuable and/or if the animal has had multiple experiences of low-cost attacks or otherwise predicts a low-cost encounter (Cussen et al., 2025).

The majority of human conflicts, similar to those of many animals, do not involve physical contact (Averill, 1982; Felson, 1984). When escalation does happen, it is usually verbal threats leading to physical aggression. Such reports exist for older children (Geiger and Fischer, 2006), university students (Tuppen and Gaitan, 1989) and community members, ex-psychiatric patients, and ex-criminal offenders who described sequences of increasingly severe “reproaches” (e.g., complaints), “insults” (including yelling), and then verbal threats, which sometimes led to physical fighting (Felson, 1984). Retrospective surveys of aggressive incidents in healthcare workplaces revealed a hierarchy of escalation levels: Events at the lowest levels involved only verbal exchanges; however, increasingly more hostile, threatening, and physically damaging behaviors were added at successively higher levels of severity (Potegal et al., 2022). Because violent individuals in closed psychiatric wards are a serious threat to other patients and staff, considerable effort has been made to develop predictive instruments. One such effort found that patients who used abusive language, yelled, and made aggressive statements were, respectively, 6.0, 6.8, and 7.2 times more likely to assault someone subsequently than those who did not (Jackson et al., 2014). Just as in other animals, not all such threats were followed by attacks, but almost all attacks were preceded by threats (Whittington and Patterson, 1996). An interpretation of verbal-to-physical escalation as resulting from increasing anger is consistent with findings that more intense anger is associated with more severe aggression in a variety of situations (Neukel et al., 2022; Precht et al., 2017; Robinson et al., 2020).

3.5 Measures of aggressiveness in offense

3.5.1 Attack latency and probability and fight initiation

Some aggressive behavior characteristics are used to infer underlying motivational states. Thus, initial attack latency has been taken as a measure of aggressive intent (Camerlink et al., 2016). In the two-trial attack-priming paradigm, shortening of latency to attack the intruder on the second trial was used as the primary indicator of increased aggressive arousal (Potegal and Coombes, 1995). Attack latency is a consistent characteristic of individual animals and is under genetic control, as shown by the breeding of short-attack-latency strains of bugs (Royer et al., 2024), great tits (Carere et al., 2005), and mice. However, one concern about latency measures is that animals that quickly attack a standard, non-aggressive target may be slower when faced with a conspecific of equal size and aggressiveness (Miczek et al., 2001). Furthermore, shorter latency is not always associated with other indices of aggressive state, as shown by the counterexample of longer latency attacks in otherwise highly aggressive pigs (D’Eath and Pickup, 2002). Shorter latencies can also be associated with general arousal, such as the pathological hyper-reactivity induced by social isolation in normally colony-dwelling mice and rats (Potegal and Nordman, 2023).

Among the relevant observations of human behaviors, women who started a bar fight with another woman were also more aggressive during the fight (Collins et al., 2007). However, in competitive reaction time studies in the laboratory in which losers received actual shocks, women’s “flashpoint” latencies when they delivered the first retaliatory shocks were less correlated with several other measures of aggressive motivation (Zeichner et al., 2003). In an aggression paradigm in which subjects thought they were playing the role of “teacher,” their latency to first correctively shock a fictitious error-prone learner was uncorrelated with the shock intensities they chose (Geen and Stonner, 1973). Shock intensity was increased by prior exposure to a violent movie scene, which was the expected aggressive motivational effect, but latency was unaffected by the exposure. An additional complication of latency measures of human aggressive intent is that rapid, preemptive fight initiation may be based on fear, an “adversary effect” of facing a more dangerous opponent (Felson and Hullenaar, 2021).

In cases in which the attack latency does appear as an indicator of central state that is plausible enough for modeling, the transformation of latency data into probability measures may be more reliable and closer to the physiological processes involved. This approach can involve the analysis of hazard functions of standard survival analysis, as was done in a transform of attack latency to probability for hamster aggression (Potegal and Coombes, 1995) and for emotions and antisocial behavior in children (Stoolmiller and Snyder, 2014).

3.5.2 Fight vigor

The attack rate, bout duration, and strike or bite severity have been taken as measures of aggressive motivation in various animal species, as well as in referee’s judgments about who won a prize fight (Beccardi et al., 2023; Lane and Briffa, 2020; Szebik et al., 2025; Zubizarreta et al., 2015). However, fight vigor, similar to attack latency, may not be an independent trait, but may reflect some RHP-related variable. The axial muscles of winners of stickleback contests had increased glycolysis-regulating phosphofructokinase and reduced citrate synthase, which control muscle metabolism (Guderley and Couture, 2005). In fighting fish (Regan et al., 2015), mice (Costantini et al., 2008), and men, other behavioral measures of aggressiveness are also associated with RHP. For example, stronger men tend to be more aggressive in sports and politics (Hartin et al., 2020, Sell et al., 2017; Webster et al., 2021). A shorter attack latency, and/or greater fight vigor, might be a useful intrinsic indicator of high aggressive state; however, other extrinsic measures are needed.

3.5.3 Startle pause

An audible/visible startling stimulus will cause animals in an agonistic encounter to pause. The rapidity with which an animal then resumes fighting provides an inverse measure of motivation separate from contest actions, costs, or consequences. For example, the startle pause durations in hermit crabs attacking to evict opponents from their shells were shorter when the shell quality was high (Elwood et al., 1998). Hermit crabs that evicted their opponents had shorter pauses (Briffa and Elwood, 2001). The startle pause of fighting fish was shorter in larger fish and longer when their opponents were larger (Arnott and Elwood, 2009b, 2010). The pauses in cuttlefish also reflected own and opponent size (Schnell et al., 2015). However, some smaller fish facing a larger opponent have shorter pauses, indicating higher levels of aggression arousal, which may overcome their size disadvantage (Kareklas et al., 2019; Reddon et al., 2013).

3.5.4 Operant effort

As noted in Section 2, aggressive motivation has been measured with operants such as bar pressing by rodents, with higher schedule breakpoints indicating greater motivation and demonstrations that aggression is less motivating than food. Although cumbersome in requiring learned responses that are distinct from species-typical behaviors, the methodology is widely used and accessible to the majority of behavioral laboratories. Repeated nose poking by rodents and key pecking by pigeons to obtain an aggression reward are steps closer to a natural approach or attack-related response, respectively; nevertheless, they involve operant schedules (May and Kennedy, 2009). Troublingly, several different manipulations have dissociated aggression severity from operant response rate (Aubry et al., 2025; Golden et al., 2019).

In the widely used point subtraction paradigm (PSP) in human aggression research, subjects press the “A” button on an FR100 schedule to collect points for a monetary reward, while they press the “B” button on an FR10 to punish a (fictitious) opponent for provocatively stealing their points. The time lost on the “A” button is the response cost for the aggressive responding on the “B” button. Multiple PSP studies have shown reliable increases in aggressive motivation indicated by “B” button responding associated with ongoing provocation, acute alcohol intoxication, history of aggressive behavior, diagnoses of psychopathology, and/or male sex in different subject populations (Hyatt et al., 2019; Lee and Coccaro, 2019).

3.6 State aggressiveness in people: anger and non-angry aggressive arousal

Anger and non-angry, dominance-related aggressive arousal are the respective central states motivating reactive and proactive aggression, respectively, as proposed in Section 2 and as reified within neural circuitry in Section 4. Other conceptions of human aggressive motivation have also converged toward the idea of a central state, as in the I³ meta-theory (Finkel and Hall, 2018). However, the characteristics and the measurement of such aggressive arousal in I³ are generally undefined, although they have been estimated through self-report. Within the widely accepted GAM, central state has been identified as some combination of hostile cognition, felt affect, and physiological arousal (Becher et al., 2025). In one recent GAM-based study, the subjects’ momentary central states were self-reported on a scale whose items included feelings of annoyance, irritation and/or anger, and physiological arousal (Kersten and Greitemeyer, 2024). Their aggressive state was found to be associated with antagonistic personality traits and environmental frustrations and interpersonal provocations. Importantly, regression analyses showed that the influence of personality and provocation on the resulting aggressive behavior could be statistically subsumed by the central state variable. The items on the Kersten and Greitemeyer (2024) scale indicate that the GAM is biased toward the identification of anger as the central state. Unfortunately, there appears to be no unambiguous, purely behavioral “gold standard” measure of non-angry aggressive arousal at present.

4 Neurocircuitry of aggression types instantiate respective central aggressive states

4.1 Elements of neuroethology: gated reflexes, central pattern generators, modal action patterns, and command neurons

Lower levels of aggression-related function controlled by central motivational states include muscle activation, gating of reflexes, and activation of species-typical “modal action patterns.” The central nucleus of the amygdala (CeA), for example, is involved in higher-level switching from passive to active modes of defense, but also influences jaw muscle tension prior to biting (Han et al., 2017; Lubin et al., 2003; Zagrodzka et al., 1998). Threats posed to people by angry faces increase the motor neuron excitability in the motor cortex (Hortensius et al., 2016). Anger generally increases the force of voluntary movement, but may reduce its precision, creating problems for competitive athletes, among others (Davis et al., 2010; Rathschlag and Memmert, 2013; Zur et al., 2019). At higher organizational levels, aggression-related hypothalamic activity opens a neural gate for cutaneous touch on paw and mouth to trigger reflexes of striking and biting, respectively (see Kruk, 1991, for a review). The latter reflex makes defensive biting on oral contact with an antagonist automatic and rapid.

At the highest level of species-typical movements are ethologically identified “modal action patterns,” which include threat displays that are likely driven by genetically dedicated neurocircuitry, akin to the well-known “central pattern generators” that organize rhythmic activities such as chewing, scratching, and walking, as well as aperiodic behaviors like swallowing. A key feature of vertebrate central pattern generators is that the activity of particular sets of neurons, typically located in the spinal cord or brainstem, generates the entire movement sequence without the necessity of sensory feedback, as shown by deafferentation that eliminates input from the muscle and skin receptors. In actuality, sensory input orients and adjusts movement to the immediate environment. Temporal persistence is another feature of pattern generators; once triggered into action, circuit activity continues for seconds or minutes without further stimulation.

Unlike central pattern generators, MAPs are used in social signaling. The most complex MAP about which some physiology is known is birdsong, a stereotyped sequence of distinct syllables vocalized in a particular order (Berkowitz, 2019). An important component of the song pathway in zebra finches is a forebrain nucleus, the HVC. Indeed, the finch syllable sequence is embedded within the HVC as shown by the brief, high-frequency stimulation of in vitro slices of HVC, which triggers a seconds-long series of repetitive membrane potentials whose timing is that of normal song syllables.

Command neurons are individual or small clusters of nerve cells whose activity is necessary and sufficient to trigger complete MAPs. Majority of the known command neurons are in invertebrates. In fruit flies, for example, threat displays, but not aggression, are driven by several different small sets of neurons (Duistermars et al., 2018; Tao et al., 2024). There are 8–10 male-specific P1 interneurons in the lateral protocerebrum that regulate decisions to court or fight, dependent on the context (Hoopfer et al., 2015). Brief, low-level stimulation of the latter neurons produces a minute-long tendency toward aggression, thereby meeting the persistence criterion for a central state. Serotonergic A1 and T5 neurons in the lobster ventral nerve cord amplify the output of flexor or extensor command neurons, generating respective dominant/aggressive or submissive/defensive postures (Kravitz, 2000).

Classic examples of command neurons in vertebrates are the Mauthner cells in fish hindbrain that drive fast, directional tail-flip escapes (Hecker et al., 2020). A few locomotor command neurons have been identified in the mouse brainstem (Cregg et al., 2020; Bouvier et al., 2015). Conjecturally, command neurons coordinating jaw and leg movements for rodent bite-and-kick MAPs may be located in the brainstem pontine tegmentum (Potegal, 2023b). However, MAPs and command neurons are activated in appropriate motivational states.

4.2 Aggressive states instantiated as attractors in brain circuits

“Attractor” models are the latest addition to the array of neuroethological elements. As we have seen, persisting motivational states vary in intensity or magnitude, which, in the simplest cases, is reified in “population coding” and/or “frequency coding,” i.e., the number of neurons in the locus or circuit that are firing and/or the firing rate of these neurons. It is now quite clear that the brain function in a number of sensory, motor, and cognitive domains can be well modeled as an attractor, i.e., a series of discrete or continuous loci in a multidimensional space of relevant neural variables that can include the firing rates or patterns of all the neurons in the circuit (Khona and Fiete, 2022). Attractors are so named because the system characteristics tend to stay in or repeatedly return to these loci, which accounts for state persistence. Attractors can also account for central state scalability and behavior sequencing, as described below.

4.3 Subcortical structures instantiating aggressive states are controlled by cortical networks

The mammalian subcortical social behavior network has six interconnected nodes, the neurons of which contain receptors for testosterone and/or other steroid hormones (Newman, 1999). Together, these nodes control social behaviors including aggression. The six nodes are: 1) the medial amygdala (MeA), including its extension into the bed nucleus of the stria terminalis (BNST) and 2) the lateral septum (LS) in the telencephalon; the 3) preoptic area (POA); 4) the anterior hypothalamus (AH) and 5) the ventromedial hypothalamus (VMH) in the diencephalon; and 6) the periaqueductal gray (PAG) in the midbrain (see schematic Figure 41.1 in Fuxjager et al., 2017). In particular, the amygdala within the temporal lobe, the hypothalamus, and PAG are the core subcortical structures through which the signals shaping aggressive behavior may originate and descend. This subcortical chain is influenced or controlled by cortical areas that interact with each other and the subcortical structures in distributed brain networks. The next sections describe these structures and networks in more detail.

4.3.1 Amygdala

As the most rostral node of descending subcortical circuits for aggression, the amygdala detects threats from predators and challenges from conspecifics. The cortico-medial group of nuclei receives olfactory and vomeronasal inputs from the main and accessory olfactory bulbs, respectively, identifying prey and predator odors, as well as pheromones signaling the sex, dominance status, and alarm state of conspecifics (Raam and Hong, 2021). Projections of these nuclei to the medial pre-optic area (mPOA) prompt sexual behaviors, while axons reaching the VMH and ventral pre-mammillary nuclei provide input for offense (Zha et al., 2020). It is through these axons that the MeA contributes to states of offensive arousal.

The most consistent cross-taxa evidence for the localization of offense circuitry is the expression of the early/intermediate activity c-Fos gene in the MeA in nine different rodent species that have just attacked or fought with a conspecific (Reyes‐Arenas et al., 2025; Potegal and Nordman, 2023). Avian homologs of the extended MeA are activated during offense in some birds (Voigt et al., 2018). The amygdala contributes to the persistence of aggressive motivational states and the sequencing of associated behaviors. In Siamese fighting fish, the integrity of the caudal dorsomedial telencephalon, the fish homolog of the amygdala, is required for continuing participation in aggressive interactions (Everett et al., 2025). In mammals, lesions of the amygdala reduce the persistence of various types of aggression (Potegal and Nordman, 2023). Part of the amygdala contribution to persistence may be due to the short-term synaptic potentiation of MeA synapses on the VMH that underlies the escalation of resident rodents’ aggression towards intruders (Nordman et al., 2020). This potentiation develops over several minutes of interaction and lasts for tens of minutes.

The behavior sequences in resident male mice responding to intruders escalate from non-aggressive anogenital and facial sniffing to rough grooming and/or dominance mounting to tail rattling threat and then bite-and-kick attacks and chases (Leroy et al., 2018). Within the amygdala, neurons in the posterolateral cortical nuclei (CApl) track these transitions from the approach/investigate appetitive phase to consummatory attack (Aubry et al., 2025). Inhibition of CApl cells reduces aggression, but increases social contact, as if aggressive motivation is being specifically reduced. CApl neurons act through their projections to the CeA and VMH of the hypothalamus.

Projections from the MeA to the anterior hypothalamus and dorsal pre-mammillary nuclei transmit olfactory/vomeronasal inputs related to defense. This sensory channel is complemented by the processing in the basal and lateral amygdala nuclei of auditory and visual threat information, such as conspecific threat calls in bats and other animals and the acoustically “rough,” fear-inducing screams in people, as well as the visual looming of fear-inducing inanimate and animate stimuli in birds and rodents and threatening faces and people in monkeys and humans (Arnal et al., 2015; Khalil et al., 2023; Mennella and Grezes, 2023). The basolateral amygdala (BLA) mediates fear learning through reciprocal projections with the medial and ventral prefrontal cortex (PFC; the human lateral nucleus is 37% bigger than what would be allometrically predicted for an ape with a human-sized brain) (Barger et al., 2014).

Neurons in the lateral nucleus project to the CeA, from which there is direct, major output to brainstem structures including the PAG. Among the actions of the different CeA neuron subgroups are fear-related amplification of startle responses, adjustments of the jaw muscles for biting, and rapid transitions from passive freezing to active flight (Moscarello and Penzo, 2022). Connections of the CeA with the brainstem regions of motor control (e.g., the PAG) and autonomic regulation (e.g., the hypothalamus) show only minor differences between rats, monkeys, and humans (Mennella and Grezes, 2023).

Functional MRI (fMRI) studies have shown that the amygdala responses to threat or provocation are exacerbated in chronically angry, reactively aggressive people. In contrast, individuals prone to proactive aggression tend to have smaller and dysfunctional amygdala, which may be due to a specific reduction in the fear-mediating basolateral component (Potegal and Nordman, 2023). Although humans are microsmatic and our olfaction-processing medial nuclei have shrunk, neurosurgical destruction of the MeA reduces intractable aggression in patients whose medication-resistant behavior was secondary to moderate to severe cognitive impairments associated with a variety of etiologies (Mpakopoulou et al., 2008; Salazar et al., 2018).

4.3.2 Hypothalamus

Shaped by input from the amygdala, the activity in various nuclei distributed across the anterior–posterior extent of the medial hypothalamus motivates, organizes, and orients defensive and offensive behaviors (Gutierrez-Castellanos et al., 2024) (contrastingly, predatory motivation is instantiated in the lateral hypothalamus). As repeatedly confirmed by a number of independent laboratories, it is within these medial nuclei that neuronal activity instantiates aggressive states. In rodents, anterior medial hypothalamic activation is associated with active defense directed against aggressive conspecifics, including avoidance and jumping. Posterior hypothalamic sites, including the dorsal pre-mammillary nucleus (DPN), initiate more general defense, triggering passive freezing and/or active flight, dependent on the circumstances.

What was earlier identified as the “hypothalamic attack area” in the rat brain, based on electrical stimulation, has been parsed by more precise optogenetic methodology to the ventrolateral area of the mouse ventromedial hypothalamic nucleus (VMHvl), which lies between the anterior and posterior defensive areas. As replicated in multiple laboratories, activity in the 10,000 or so glutaminergic (i.e., excitatory) estrogen receptor-expressing cells of the mouse VMHvl instantiates a central aggressive state that motivates offense. That is, optogenetic stimulation of these cells in male mice initiates immediate, robust, pheromone- and hormone-independent offensive attacks on intact and castrated males, females, and even inanimate objects (Yang et al., 2017). Conversely, pharmacogenetic inhibition suppresses normal attacks, while mice with genetic modification of the estrogen receptors are non-aggressive. In short, activity in VMHvl neurons is both necessary and sufficient to initiate and terminate bouts of aggression.

Aggressive motivation instantiated in mouse VMHvl neuronal activity during natural fighting has been modeled as a recurrent switching dynamical system that approximates a line attractor in a two-dimensional plane with three or four distinct states located along the linear trajectory (Nair et al., 2023). One dimension of the plane is encoded by excitatory VMHvl neurons whose firing is time-locked to the entry of a conspecific intruder. A second “integration” dimension is instantiated by neurons whose slow increase in firing rate is the time-extended sum of their various exteroceptive and interoceptive inputs (Vinograd et al., 2024). Activity along the integration dimension scales the aggressive motivational intensity from one state to the next in partially separated subpopulations of VMHvl neurons (Nair et al., 2023). It is low during sniffing, ramps up at the onset of dominance mounting (a low-intensity aggressive behavior), and increases further to a stable plateau as the mouse attacks. However, this state outlasts any particular bout of attack and persists past the removal of the intruder, indicating that it is more about aggressive motivation than particular behaviors. Slow neurotransmission by oxytocin and vasopressin generates the integrative function of the attractor, which decays exponentially over time, terminating the aggressive state (Mountoufaris et al., 2024). In fact, the empirically determined decay time constant in each mouse predicts the behavioral persistence of aggression in that animal.

While attractor modeling of hypothalamic aggressive states has been restricted to mice, there are more general complementary observations across species. The homolog of VMHvl is activated during offense in fish (Jackson et al., 2025). Transitions from a non-agonistic behavior to threat display and then to overt attack in tree lizards involve increasing activation in VMHvl (Kabelik et al., 2008). Aggression-specific neurons in the anterior hypothalamus of finches are akin to those in mice (Goodson et al., 2012).

Accumulated clinical observations have shown that some hypothalamic lesions and tumors increase human aggression (Siegel and Victoroff, 2009). Experimental evidence that the hypothalamus is also part of our subcortical aggression core includes its activation by simulated aggression-related threats or actions, either explicit (viewer-directed) or implicit (other-directed) (Caria and Dall, 2022). Significant negative correlations between self-reported aggression and functional connections between lateral hypothalamic resting state activity and that of the thalamus in men and the middle temporal gyrus in both sexes suggest the importance of inhibitory control (Yao et al., 2025). At least 10 small studies have reported aggression reduction with a few side effects from deep brain stimulation of the posterior medial hypothalamic “triangle of Sano” in low-functioning, intractably aggressive patients (Cojazzi et al., 2024; Mofatteh et al., 2025). However, the most recent brain imaging and stimulation studies suggest a lateral hypothalamic locus for human aggression (Gouveia et al., 2023; Yao et al., 2025).

4.3.3 Periaqueductal gray

Both the POA and VMH project to the PAG, which is the most caudal locus from which organized, albeit stimulus-bound, attack may be elicited. Neuroanatomically, the PAG is a densely packed cylinder of neurons organized into columns surrounding the cerebral aqueduct along the axis of the brainstem. The neural circuits for attack and defense are discrete and highly localized within these columns. The circuitry for aggression lies within the rostral end of the columns, with affective defense located dorsally (Shaikh et al., 1997; Sukikara et al., 2006). The caudal ends of the PAG columns contain the circuitry for fear responses, with active flight elicitable from the dorsal loci (Bandler et al., 2000; Borelli et al., 2005). These highly active defensive behaviors are physiologically supported by the dorsal PAG column, which increases heart rate and blood pressure to provide greater blood flow to the limbs and viscera for both rostrally activated affective defense and caudally activated flight.

A ventrolateral column in caudal PAG mediates the opposite physiological passive behavior and effects of defeat. Mice and rats that have been severely and repeatedly attacked by a conspecific no longer try to flee. Instead, they assume an immobile, upright defense posture. They squeal when approached, but are less sensitive to pain due to defensive stress-induced analgesia. The immobility and analgesia, accompanied by reduced heart rate and blood pressure, are all mediated by the ventrolateral PAG column and associated structures. The respective dorsal vs. ventral PAG localization of active vs. passive coping may be true for people as well (Franklin, 2019).

The defense activated by the electrical or chemical stimulation of the PAG in rats and cats is highly reliable and very dramatic. Stimulation of the PAG in patients undergoing brain surgery elicited various autonomic responses including tachycardia, sweating, hyperventilation, chest pain, and urges to urinate (George et al., 2019). Some of these patients reported feelings of dread/terror and an urge to flee. While this system is predominantly directed caudally, a rostral flow of signals provides information about behaviors being executed back to the forebrain structures (Halász et al., 2002).

4.4 Neural bases of the reinforcing value of aggression and its satiation

It is well established that dopamine, which is released from the ventral tegmental area (VTA) of the midbrain into the nucleus accumbens (NAc) in the basal forebrain, is a primary mechanism of self-reward in offensive attack, as it is for other reward modalities (see Potegal and Nordmann, 2023, for a review and Dai et al., 2025, for a caveat). A comparable role in aggressive motivation has been reported for the left hemisphere accumbens in humans, the actions of which are communicated to the frontal cortex (Threadgill and Gable, 2020).

Dopamine release in the shell portion of the NAc also provides reward-initiated incentive motivation for the next attack (Ferrari et al., 2003; Saddoris et al., 2015). This mechanism may well contribute to the social instigation and attack-priming effects noted in Section 2.3. By extension, the aggression satiation effects, as well as the post-anger fatigue state (Hackert et al., 2021), might be explained by the inhibitory projections from the NAc to the anterior cingulate (ACC). That is, after sufficient accumbens activation, these projections suppress the cingulate’s energy-expending drive, as described below (Potegal, 2023). The reward value of aggression and the consequent intensity of aggressive motivation is modulated by input from the lateral habenula to the accumbens through multiple neural pathways in fish, mice, and humans (Gouveia and Ibrahim, 2022; Ogawa and Parhar, 2022; Takahashi et al., 2022). This circuit might also be involved in aggression satiation effects (Post et al., 2022).

4.5 Key cortical areas in aggression regulation and dysregulation

The descending chain of core subcortical structures is controlled by cortical areas and networks, including the prefrontal and temporal neocortex, the cingulate gyrus, and the insula. The dorsolateral prefrontal cortex (dlPFC) has a general, if neuroanatomically indirect, inhibitory effect on aggression as part of its executive function of restraining impulsivity and risk-taking. A swath of the ventral PFC, including the ventrolateral, orbitofrontal, and ventromedial tissue, is associated with the experience/expression of anger and also modulates various forms of aggression (Potegal and Nordman, 2023). The latter two areas have direct projections to the amygdala and hypothalamus, hence can affect aggressive motivational activity monosynaptically. Some regions within the rodent ventral and medial PFC exert inhibitory control over aggression (Takahashi et al., 2014), while other regions excite hypothalamic nuclei release or intensify aggression (Biro et al., 2018).

4.5.1 Ventral prefrontal cortex

The inferior frontal gyrus, ventrolateral on the surface of the right hemisphere (vlPFC), may be involved in the verbal labeling of various felt emotions, such as anger, and/or their coupling with impulses to action (Kim et al., 2021; Kirby and Robinson, 2017; Sorella et al., 2021; Puiu et al., 2020). On the under surface of the frontal lobe, just above the orbits, the orbitofrontal cortex (OFC) evaluates reward and punishment expectancies and outcomes, broadly speaking. In social context, it tracks the dominance status and determines the optimal affiliative or aggressive moves, registering social threat and mediating the aggression-inhibitory effects of empathy, guilt, and fear (e.g., Kätsyri et al., 2013; see Potegal and Nordman, 2023, for a review). Accordingly, the OFC output is mostly inhibitory, as shown by the typically negative correlation of OFC activity and aggression, e.g., the OFC is deactivated during FNR (Dugre and Potvin, 2023). Conversely, impulsivity, anger, and aggression increase following OFC damage. However, under some conditions of provocation, the OFC promotes retaliatory aggression, at least in men. Evidence for OFC function is somewhat mixed, depending as it does on the differences between the medial vs. lateral OFC and on sex, personality, and circumstance.

4.5.2 Cingulate gyrus

The elongated cingulate gyrus surrounds the corpus callosum on the medial surface of the hemisphere. The regions within the cingulate vary in their function. The perigenual ACC is the anterior extension of the gyrus, lying rostral to the “knee” of the corpus callosum. This area is activated when people make risky or emotionally laden “hot” decisions, with negative emotions localized to the more ventral regions and positive emotions localized more dorsally (Akiki et al., 2025; Pelliccia et al., 2023).

The dorsal anterior cingulate (dACC) is involved in estimating situational risks and payoffs and in amplifying response efforts according to both the motivational drive and demands of circumstance (Astori and Sandi, 2024). A few patients whose dACC was stimulated through electrodes implanted prior to epilepsy surgery reported experiencing a sense of “determination” (Parvizi et al., 2013) and the impulse to literally “get up and go” (Caruana et al., 2018). Several studies have associated dACC activation with an increased “commitment to aggression,” according to Bronstein (1981). However, determination is also required to resist prepotent impulses (White et al., 2024). ACC activity increased in men and women who were actively suppressing anger (Pelz, 2025) and most strongly with the inhibition of retaliatory responses in a social network aggression task by young adults who were desisting from previous antisocial behavior (van de Groep et al., 2022). In rodents, cingulate inhibition that keeps aggression from intensifying beyond the species-typical range is mediated by GABAergic interneurons and serotonergic terminals (Crew et al., 2021; Chaibi et al., 2023).

The mid-cingulate region becomes activated both during FNR imposed by an opponent and the retaliatory response (Dugre and Potvin, 2023). This region also contains a secondary motor area that controls non-volitional emotional vocalizations and facial expressions, independent of the more anterior primary motor area that controls voluntary movement. Proactive aggression has been associated with the posterior cingulate (Dugré and De Brito, 2024).

4.5.3 Insula

The insula in the human brain is hidden at the base of the Sylvian fissure separating the frontal and temporal lobes. Recording and stimulation in neurosurgical patients have shown that the insula receives basic visceral and autonomic inputs, including cardiac and gut sensations and warmth and pain signals. As sensory input is transmitted from the posterior to the anterior insula, it is transformed into negative and positive sensory-related emotional responses, including empathy. The anterior insula is also involved in motor impulsivity and reactive aggression, possibly related to a reduced consideration of harm avoidance in risky decision-making. fMRI studies have shown insula activation during anger processing and greater aggression associated with reduced insula gray matter (Tanzer et al., 2021; Seok and Cheong, 2020).

4.6 Brain networks

The structures noted above, along with a number of others, are organized into networks that control broad processes of perception, attention, motivation, and action. This section describes the brain networks that control aggression in rodents and people.

4.6.1 Aggression-activated brain networks in rodents

Patterns of c-Fos activation following an encounter with an intruder revealed two brain networks that were more tightly connected in aggressive than non-aggressive male mice (Aubry et al., 2025). The first network involved strengthened correlations among several cortical, basal ganglia, and thalamic areas, consistent with sensorimotor evaluation and feedback-guided decision-making associated with eventful social encounters. The second network included brainstem structures suggesting general arousal-related processes. These results are akin to the multiple activated areas seen in fMRI images of head-fixed male rats piloerecting in response to a male conspecific (Ferris et al., 2008). Both studies are a reminder that specifically aggression-motivated behaviors are accompanied by supportive sensory, motor, hormonal, and autonomic processes.

4.6.2 Three aggression-relevant canonical human brain networks and four others

Some of the most extensively characterized human brain networks are based on patterns of neuroanatomical connectivity, fMRI-derived brain activity reflecting cerebral blood flow, and EEG signals that identify multiple connected areas that are in synchrony at rest and become interactive during particular perceptual or cognitive tasks. Three of these canonical networks are the most relevant to aggression. The dlPFC is the rostral component of the frontoparietal executive function network that regulates impulsivity and inhibits reactive aggression, among its many actions. Transcranial direct current anodal stimulation of the dlPFC reduced the tendency of normal adults to commit aggression in hypothetical vignettes (Choy et al., 2018; but see Denson et al., 2025).

The salience network consists of the anterior insula and the ACC, both of which contain large von Economo neurons, suggesting some common function. Indeed, as might be expected from the insula’s detection of aversive bodily states and the cingulate’s facilitation of impulses to action, the salience network functions to identify and rectify urgent needs. Meta-analytic comparisons of non-aggressive and aggressive individuals showed that the salience network structures of the latter group exhibit abnormal gray matter volumes and also greater network activation during aggression-eliciting laboratory paradigms (Chen et al., 2025). The salience network is linked to frustration-induced aggression in particular (Dugre and Potvin, 2023; Yu, 2016; Morley et al., 2019).

The default/mentalizing network (D/MN) is active when people are attributing mental states, such as hostile intent, to others (and also when they are just daydreaming). The D/MN includes the frontal and parietal cortex on the medial surface of the hemisphere along with the lateral temporoparietal junction (TPJ). The D/MN has been identified as consistently abnormal in aggressive individuals (Chen et al., 2025). More specifically, some evidence suggests that the D/MN may suppress reactive aggression, perhaps associated with the calming parasympathetic input that it receives, but may also promote proactive aggression (Potegal and Nordman, 2023). Thus, the TPJ becomes activated when perceiving someone’s hostile intentionality, but also by thoughts of winning a contest. The whole D/MN may be partially activated when people contemplate proactive revenge; its activity declines when they decide to avoid provocation rather than fight.

In addition to the three canonical cortical networks, an emotion-processing subcortical network involving the amygdala and the lateral OFC has been associated with escalated aggression (Wang et al., 2024). Mapping of the brain regions that are associated with reactive aggression when damaged onto known connections among human brain structures identified two “lesion networks” (Miller et al., 2026). The first has a hub in the ventromedial PFC (vmPFC) with connections to the OFC and the frontal and temporal poles. Its associated aggression may result from generally disinhibited impulsivity. The second network’s hub includes the perigenual ACC and extends to the insula and the dlPFC. This network may foster aggression via emotional dysregulation. Both lesion networks may be contained in the structural/activational networks of aggression, including the perigenual ACC, the OFC, and the temporal poles, which were neurogenetically identified by their expression of 126 genes that have been associated with aggressive clinical phenotypes (e.g., irritability, conduct disorder in youth, and antisocial personality disorder in adults) (see Figure 3 in Dugré et al., 2025). The mapped genes, including those mediating sensory processing and pain, synaptic vesicle release, and axon guidance, also identified the salience network and the D/MN as part of the neurogenetically defined aggression network. Notably, this mapping identified the posterior hypothalamus as part of a subcortical human aggression network for the first time.

4.7 Neurocircuitry of defensive central states

4.7.1 Defensive behavior and central states in rodents

Defense-related neurocircuitry is distributed across the anterior–posterior extent of the rodent medial hypothalamus (Gutierrez-Castellanos et al., 2024). Attacker imminence is population coded in the posterior dorsal hypothalamic sites, including the DPN (Cheung et al., 2025). Progressively higher levels of defensive firing are associated with avoidance, freezing, and then flight (Kennedy et al., 2020). Similarly, in cats, increasing dorsal hypothalamic stimulation yielded a sequence of growling, hissing, then flight that mimicked natural defensive sequences (Brown et al., 1969). These responses are relevant to contest behavior in that freezing immobility can reduce aggressors’ attacks (Pellis and Pellis, 2015), while flight terminates agonistic encounters.

In contrast to the posterior dorsal hypothalamic generalized defensiveness, activity in the mouse anterior medial hypothalamus and/or ventral pre-mammillary nucleus is specifically associated with active self-protection against aggressive conspecifics, including upright boxing and jumping directed at them (Gutierrez-Castellanos et al., 2024; Melleu and Canteras, 2025). It is this central state–behavior relationship that is most relevant to BE modeling.

4.7.2 Neural bases of human anxiety, fear, and panic

A general distinction between the subcortical substrates that mediate human defensive states is that anxiety primarily involves the BNST, while the BLA is activated in fear (Sonkusare et al., 2025; van de Poll et al., 2023). In one of the paradigms used to elicit these states, large threatening virtual reality characters activated both the amygdala and the insula (de Borst and de Gelder, 2022). When such stimuli engage the PAG, subjects freeze, as shown by a reduced body sway (Melleu and Canteras, 2025). Notably, the right hemisphere structures are preferentially involved in defensive motivation and action (Patrick et al., 2019).

The last phase of the defensive imminence sequence is counterattack. For obvious reasons, real fear-induced counterattack has not been studied in people. Approximations to it are videogame shoot/no shoot paradigms in which participants receive an aversive noise blast or shock for not shooting the villain before he shoots them. The switch from a preparatory phase to active shooting in such games involves the perigenual ACC increasing its connectivity with the amygdala (Hashemi et al., 2019). Although the neural circuitry underlying the successive stages in human defensive behavior has been described in some detail, no particular locus of a central state that appears to scale to the gradations of anxiety, fear, and panic has yet been identified.

4.7.3 Stress analgesia

A common feature of defensive states is stress analgesia, in which the transmission of pain signals from the spinal cord to the brain is at least partially blocked. A short-lived, non-opioid, cutaneous analgesia occurring in anticipation of being attacked and through the early passive stages of defense is well replicated (Blanchard et al., 1998; De Souza et al., 1997; Pitcher et al., 2017). So is a longer-term, opioid-mediated visceral analgesia occurring during the later active stages of defense. In contrast, there is only limited evidence for an analgesic increase in the pain thresholds associated with rodent offense or human non-angry aggressive arousal (Ignatov et al., 1984; Siegfried and Frischknecht, 1988; Stephens and Allsop, 2012; Niedbala et al., 2018).

4.8 Aggression reward

The appetitive nature of the aggression motivated by central offense states is demonstrated in the increased activity of VMHvl neurons during operant performance to obtain an attackable target (Falkner et al., 2016). The same subpopulations in the VMHvl become active in the successive stages leading up to the attack in both the standard resident–intruder situation and in operant responding for an attack opportunity (Minakuchi et al., 2024). Accordingly, the optogenetic activation or inactivation of these neurons promoted or reduced, respectively, operant responding for attack opportunity.

4.9 Superordinate aggressive motivation and general arousal

A balance between excitation and inhibition that adjusts the motivational level of both defense and offense and also accounts for sex differences in aggression in mice reportedly involves the substantia innominata, which is neuroanatomically interpolated between the VMHvl and PAG (Zhu et al., 2024). Whether or not these findings can be replicated, specific central aggressive states are distinct from general arousal, which involves cortical activation from brainstem nuclei such as the nucleus gigantocellularis and the parabrachial nucleus (Kandel and Pfaff, 2019; Li et al., 2025).

4.10 Contagion of aggression

In accord with the contagion of aggression being a central state phenomenon, as noted in Section 3.3.3, the core amygdala and hypothalamic aggression neurocircuitry becomes activated in mice observing another mouse attacking an intruder. Interestingly, the observer mouse’s MeA becomes activated and the animal becomes aggressive only when the attacking mouse is familiar to the observing mouse (Adjei et al., 2025). Aggression “mirror” neurons that are activated by both attacking and seeing another mouse attack comprise a subpopulation of the VMHvl; these neurons are coupled to aggressive actions, as indicated by their optogenetic activation that triggers aggressive displays, even to the mouse’s mirror image (Yang et al., 2023). On a longer timescale, repeated observations of aggression produce quite similar behavioral changes and brain-wide alterations in activation and connectivity, as do repeated acts of aggression, with the exception of the effects on the NAc, suggesting that observation is not as rewarding as action (Iravedra-Garcia et al., 2024).

4.11 Inhibitory neuroanatomical nodes and neurochemical systems

Attacks are moderated or inhibited in a variety of conditions and circumstances in addition to the quit decisions or the fear that may constrain intention bites, bluff charges, and other attack feints. Pregnant/lactating female rodents fiercely attack and bite male and female intruders; however, during intrasexual fighting under other circumstances, they inhibit their bites, which may not even break the skin (Begg and Nelson, 1977; Blanchard et al., 1984). Play bites in all species are routinely inhibited.

4.11.1 Septal nuclei

Aggression-inhibitory neuroanatomical nodes include the lateral septum in birds and mammals. Lesions of the lateral septum in rats were infamous for producing “septal rage,” a transient but striking increase in defensive attack when handled or even approached. However, this lesion effect was muted or absent in other species. In mice and rats bred for low or high aggression, the low-aggression strains showed much more septal activity after agonistic encounters than did the high-aggression strains (Veenema and Neumann, 2007). More direct evidence is that lateral septal stimulation inhibited all types of aggression in the rodents and cats that were tested (Brutus et al., 1989). A major pathway for septal aggression inhibition is via its GABAergic projections to the VMHvl (Wong et al., 2016). Aggression-inhibitory signals from the lateral septum may help shape the social interactions of rodents more generally, including that of pups with their siblings and adults with unfamiliar vs. familiar conspecifics (Leroy, 2023; Rashid et al., 2024).

A particularly important form of inhibition is the suppression of aggressive motivation to allow sexual interaction and parental care of the young. Without such inhibition, pup care becomes pup killing, consistent with MAPs for infanticide of the progeny of others that support the aggressor’s own inclusive fitness (Chen et al., 2019). Input from the mPOA, which mediates reproductive behaviors, to the VMHvl imposes long-term inhibition of aggressive arousal that permits reproductive behaviors (Minakuchi et al., 2024).

The septal nuclei are vestigial in humans. However, neurological cases and experimental fMRI studies have shown that the oxytocin-rich region of our basal forebrain extending from the septal nuclei through the contiguous mPOA is activated in social scripts involving charitable donations, trust, and attachment and is involved in affiliative bonds with friends, social groups, objects, and cultural symbols (Bortolini et al., 2024; Caria, 2024).

4.11.2 Serotoninergic inhibition and excitation

Serotonin (5-hydroxytryptamine, 5HT) is an evolutionarily conserved monoamine neurotransmitter that regulates multiple functions. In invertebrates, serotonin increases aggressive behavior, arguably by altering aggressive motivation (Bergman and Moore, 2020; Liang et al., 2023; Hu et al., 2020). In vertebrates, the overall effect of serotonin is to reduce aggression. Serotonin in vertebrates is synthesized in the brainstem raphe nuclei, which lie in the midline of the midbrain and brainstem (except in birds where they are bilateral). The most anterior of these nuclei, the dorsal and median raphe, project rostrally to structures including the hypothalamus, thalamus, amygdala, and cortex. Among the inhibitory neurotransmitter systems, serotonergic projections from the dorsal raphe (DR) to various cortical and subcortical structures have the most suppressive effects on offense, lesser effects on defense, and opposite effects on maternal aggression in other animals (Blanchard et al., 1998; Carrillo et al., 2009; Vergnes et al., 1986; cf. Dong et al., 2025, for related 5HT aggression-type specificity in a spider). The suppression of offense may be mediated by an inhibition of aggression–reward in the NAc (Zhang et al., 2025).

More generally, 5HT projections to the cortex reduce impulsive actions. Accordingly, 5HT has inhibitory effects on human anger and reactive aggression, as shown by agents that increase its availability, e.g., by blocking its reuptake into presynaptic terminals, while dietary manipulations and drugs that reduce serotonin availability increase angry aggression (Fanning et al., 2020). In both rats and humans, a reduction in 5HT neurotransmission strengthens the connections between the amygdala and the PAG, thereby intensifying the bradycardia of fear (Schipper et al., 2019).

Beneath the two simple generalizations of the opposite 5HT effects in invertebrates and vertebrates and its inhibition of general impulsivity and some specific types of aggression in mammals lie layers of complexity. Dynamic regulatory function is suggested by the surge of 5HT release that occurs at the onset of aggressive episodes (Feltes and de Boer, 2020), while set point effects may be inferred from the 5HT receptor agonists that increase aggression in some low-aggressive invertebrates and vertebrates, but reduce it in high-aggressive individuals (Bubak et al., 2020; Malone et al., 2025). On a cellular level, the effects of 5HT, similarly to those of other neurotransmitters, depend on the particular receptor to which it binds. The seven families of 5HT receptors include approximately 15 different subtypes. Among these, pre-synaptic 5HT_1B autoreceptors on the 5HT terminals that limit 5HT neurotransmission through negative feedback are more numerous in highly aggressive rodents and humans, consistent with a serotonergic inhibition of aggression (Feltes and DeBoer, 2020). However, different receptors in different nervous system loci have opposite effects on one or another type of aggression (Bubak et al., 2020; Shaikh et al., 1997). The need for further research to parse this panoply of 5HT effects is clear because it modulates aggression in multiple important ways.

4.12 Central states of human frustration and anger

4.12.1 Frustrative non-reward

The few studies on anger-eliciting FNR agree that it affects multiple brain networks including the salience network (Dugré and Potvin, 2023). Curiously, the acute experience of frustration has been reported to partially deactivate the frontal-parietal executive function and D/M networks in healthy adolescents (Harlé et al., 2022). However, what may be rebound activation was observed in these networks following frustration in irritable adolescents (Linke et al., 2023). Post-frustration activation of the anterior D/M and the amygdala predicted self-reported irritability of these youth, and orbitofrontal, striatal, and somato-motor activation predicted parent-reported irritability. This difference between the self and parent irritability ratings suggests that the former network instantiates the felt subjective responses to FNR, while the latter structures generate overt observable behavioral responses.

4.12.2 Anger and reactive aggression

Similarly to frustration, anger activates multiple networks. Anger elicited by various provocations engages more neurocircuitry than the four other emotions with which it was meta-analytically compared (Wager et al., 2015). The appraisal stage of emotion processing, which includes consideration of the offender intent, activates the D/MN, as expected (Giardina et al., 2011; Riedel et al., 2018). In addition, five major meta-analyses combining published brain maps of emotion through 2017 all found the experience/expression of anger to be associated with ventrolateral PFC and/or orbitofrontal activation (Potegal and Nordman, 2023). The insula was also frequently activated during anger, as were different regions of the cingulate (Alia-Klein et al., 2020; Klimecki et al., 2018). The breadth of anger-related activation may be related to shifts in focus and depth of attention, cognitive processing, and judgment that typically accompany this emotion. Because expressing anger can be socially deleterious, e.g., when inappropriately redirected (see 3.3.2, cf., Wang et al., 2020), it is typically under some inhibitory control. The D/MN is involved in anger control and in early-stage appraisal (Sorella et al., 2022).

Although the right inferior frontal gyrus mediates both the perception of angry expressions and the subjective experience of anger (Sorella et al., 2021), anger largely activates the left hemisphere, particularly when there is implicit or explicit opportunity to act on that anger Harmon-Jones and Gable, 2018; Gallucci et al., 2020; Gramegna et al., 2025; Nikolic et al., 2022). Anger involves approach, and left hemisphere activation is generally associated with behavioral approach, even in other animals (Koboroff et al., 2008). Note the contrast with the right hemisphere fearful defense and withdrawal. However, when anger is likely to be suppressed for social or other reasons, cortical activation tends to shift toward the right hemisphere (Consolini et al., 2022).

It is noteworthy that anger is the least likely emotion to be elicited by direct human brain stimulation (Guillory and Bujarski, 2014). Perhaps any angry feelings are overridden by fear, the most typical response when vulnerable, anxious patients with electrodes in their heads are surrounded by operating room surgeons or lab-coated psychologists. Alternatively, anger is truly a network function that requires simultaneous or successive excitation of its various components to be consciously experienced and expressed.

The trait of reactive aggression has been associated with a reduced volume of the PFC, but an enlarged anterior and ventromedial temporal lobe within which the amygdala is located (Dugré et al., 2025; Bashford-Largo et al., 2025; Sorella et al., 2022). Reactive aggressive episodes, which were neuroimaged in adults, involve retaliation in response to provocation in competitive/cooperative laboratory situations. Activation of the dorsomedial PFC (dmPFC) and the adjacent ACC during both provocation and retaliation is the most consistent observation and has been interpreted as assessing the respective values of revenge vs. restraint (Fanning et al., 2017; Mathur et al., 2023). Equally consistent, aggressive individuals responding to threat and provocation show pronounced amygdala reactivity, which may not occur in non-aggressive comparison subjects (Bertsch et al., 2020). Similarly, in normally non-aggressive people, the amygdala activity during reactive episodes is often negatively correlated with the activity in its various neocortical inputs, consistent with inhibitory cortical control (Tonnaer et al., 2023). In aggressive individuals, the amygdala and cortical activity can become disconnected, indicating a loss of inhibition, or even positively correlated, suggesting aggression-promoting positive feedback (Dougherty et al., 2004). These differences between populations of interest are one of the problems in elucidating the neural bases of aggression.

4.13 Neurology of dominance-motivated proactive aggression

There are fewer studies of the neural bases of dominance-motivated proactive aggression than of anger and reactive aggression. To date, it appears that the same brain areas may be involved in both types of aggression (Romero-Martinez et al., 2022; Zhu et al., 2022). However, the neurogenetic analysis that defined the networks for reactive aggression failed to determine any for proactive aggression (Dugré et al., 2025). One feature that most consistently differentiates reactive from proactive aggressors is the latter’s association with a reduced amygdala and/or temporal pole volume. This reduction may be related to specific diminution in the fear-mediating BLA (Potegal and Nordman, 2023). In addition, the anterior and posterior cingulate may be more associated with reactive and proactive aggression, respectively. Among the issues complicating this analysis are that some samples of proactive aggressors include severely psychopathic and very violent individuals whose brains are clearly abnormal.

Activity in the dorsal and/or ventral striatum is sometimes reported in association with aggression. The dorsal striatum, i.e., the caudate nucleus, is on the input side of the basal ganglia, which generally functions to evaluate and transmit back to the cortex the outcomes of cortical decision-making, including, but certainly not limited to, decisions about aggression. The ventral striatum, i.e., the NAc, is activated in reward situations, including winning a fight and/or seeing an opponent lose a fight (Cikara et al., 2011; Skibsted et al., 2017). Widespread activity in the ancillary sensorimotor and other supportive structures also occurs during aggression provocation in rats (Ferrris et al., 2008).

Failure to find neurocircuitry for the dominance motivational state underlying proactive aggression may be related to the limitations of fMRI that is used for human investigations relative to the micro-level discoveries about aggressive states that are instantiated in the amygdala and the hypothalamus of other animals. Alternatively, in the larger, more diverse human brain with the multiple inputs and experiences that shape human aggression, motivation and behavior are controlled by complex networks rather than a localized “necessary and sufficient” node. Elements in such networks may be facultative in the sense that they are brought into play in some circumstances but not in others, making it that more difficult to determine the entire network and its contingencies. Research to clarify these issues is needed.

5 Behavioral ecology of contest factors and strategies and their neural bases

BE models predict the probability of fight initiation, the duration, and/or the outcomes of contests based on the information each opponent gathers (Arnott and Elwood, 2008, 2009a). Key factors include resource value, RHP, and potential or actual costs. The most common way to determine what information is used is by assessing if the quit time of the loser is influenced by the resource value, aspects of their own abilities, and the costs and aspects of their opponent’s abilities. In this section, we first describe the resource and RHP factors, then the three main BE models, and finally the processes involved in quit decisions about which some physiology is known and which could be used to advantage in future BE modeling. We then review how human aggression might be interpreted with concepts from BE.

5.1 Resource valuation

Although the hawk/dove model noted in Section 1 assumes no assessment of resource value, such assessments are central to the majority of other models and have been demonstrated in many studies (Arnott and Elwood, 2008). For example, when intruder male amphipods (Dick and Elwood, 1990) and newts (Verrell, 1986) compete for a guarded female, both competitors have contacted her and assessed her reproductive value. This is clear because both intruders and guarders that lost fought longer for higher-value females. Sometimes, only the original guarder or resident knows about the resource. Male orb web spiders compete for females, but only the resident fights for longer if the female is large and has more ova (Bridge et al., 2000). Female iguanas fight for burrows, but only residents fight longer for what they know is a deeper burrow (Rand and Rand, 1976). This asymmetry in information creates a resident advantage because only they can make decisions based on resource values.

These are examples of objective resource assessment because they relate to measurable values. However, the resource value can vary with the state of the animal. For example, food items are valued more by insects (Nosil, 2002) and birds (Hansen, 1986; Popp, 1987) that have not fed recently. Furthermore, territory residents fight harder and usually win against intruders (Sacchi et al., 2021). Here, state dependency refers to the time and the effort residents have invested in finding food sources, learning where danger lies and where refugia are located, and settling borders between neighbors (Brown, 2001; Stamps, 1983). Other examples include male crickets that have not recently mated being more likely than recently mated males to initiate aggression and win fights for females (Brown et al., 2006). Similarly, male rats that had just copulated (Flannelly et al., 1982) and men who viewed images of attractive women (Ainsworth and Maner, 2014; Mason et al., 2020) show heightened aggression. Furthermore, pregnant or lactating female rodents respond to intruders with a ferocious, maternal-aggressive mix of offensive and defensive attacks that protect offspring (Elwood et al., 1990; Rieger et al., 2022). In rats, cocaine- and amphetamine-regulated transcript (CART) neuropeptidergic neurons in the ventral pre-mammillary nucleus mediate maternal, but not inter-male, aggression (Basu et al., 2025). The CART neurons project to the VMHvl, as do the posterior amygdala cells expressing oxytocin receptors in mice (Yamaguchi et al., 2025). Oxytocin release during lactation activates the posterior amygdala and also strengthens the amygdala–VMHvl connection, thus increasing VMHvl excitability and facilitating attack. After pup separation, the oxytocin levels drop, reducing maternal aggression. These findings demonstrate motivational state effects on aggression.

Comparable parental behavior is seen in men and women who have children: visualize an angry male antagonist as larger and more muscular and imagine taking more aggressive actions if threatened than do non-parents (Fessler, 2010; Steiner, 2016). There is little direct evidence for people acting on these impulses, but nursing mothers react more punitively to a rude person than do formula-feeding mothers or women who have never been pregnant (Hahn-Holbrook et al., 2011).

5.2 Resource holding potential

The hawk–dove model does not consider RHP; hence, the chance of one hawk winning against another is 50%. However, in reality, the hawk with greater RHP, because it is larger, stronger, better armed, and/or fights more skillfully, will be more likely to win. For example, the claw size in crustaceans (Yasuda et al., 2014), the horn length in ibex, and the bite force in collared lizards (Lappin and Husak, 2005) contribute to RHP. Height and upper body strength are components of human RHP (Lukaszewski et al., 2016; Stulp and Barrett, 2016), with height being the key factor that determines the winner of street fights (Potegal et al., 2025). Many contests involve symmetrical offense. In these, the larger opponent usually wins. In contests involving offense and defense, different aspects of RHP are important. The main aspect for offense is the ability to inflict damage, while the RHP for defense may comprise both stamina and ability to endure damage. The importance of these components is likely to change during the contest, with stamina being important during the early display phases while the ability to endure damage becomes important during the later escalated fighting (Palaoro and Briffa, 2017). This asymmetry of fighting styles is seen in hermit crab fights for shells in which the attacker raps its shell against that of the defender, while the defender withdraws into its shell. Not only do their activities differ markedly, but so do the physiological constraints on them, e.g., attackers fare worse than defenders when the oxygen levels are reduced (Briffa and Elwood, 2000).

5.2.1 Self-assessment of RHP is influenced by previous wins or loses

Across species, animals that win a contest are more likely to win subsequent fights, while losers typically lose again (Yan et al., 2024). Thus, self-assessment is likely to be based on the outcomes of previous contests, at least in part (Hsu et al., 2006). In both human sports players and other animals, post-conflict testosterone release is maximal, and winner effects are the strongest following victory in the home territory, with the result that winning becomes associated with specific territorial sites. In vertebrates, the reinforcement of territorial winner effects depends on reward-related dopaminergic transmission from the VTA to the NAc.

5.3 Pure self-assessment

The energetic war of attrition (E-WoA) (Payne and Pagel, 1996) is an important self-assessment model wherein fighters give up when their cumulative energetic costs reach a threshold determined by resource value (Arnott and Elwood, 2009a). The model assumes that smaller contestants use a greater proportion of their energy stores per unit time. Hence, the smaller one will quit because it is the first to reach the threshold. The loser’s persistence is strongly correlated with its RHP (Prenter et al., 2008). However, winners do not influence losers’ quit decisions; therefore, there is no relationship between the winner RHP and the contest duration.

5.4 Cumulative assessment

Similar to self-assessment models, losers in the cumulative assessment model (CAM) quit when the accumulated costs exceed a resource-dependent threshold. In CAM, however, costs accrue from the opponents’ actions (Payne, 1998). Thus, the quit decisions in CAM are influenced by both the individual’s own RHP (poor-quality individuals tolerate fewer costs before quitting) and the opponent’s RHP (fitter individuals inflict costs more rapidly). Thus, the correlations between fight duration and both winner and loser size should be significant, but in opposite directions (Arnott and Elwood, 2009a).

5.5 Mutual assessments and the sequential assessment model

The central idea in the sequential assessment model (SAM) is mutual assessment, with each contestant gathering information about the other from observations or interactions and combining that with status updates about itself. This enables contestants to know whether they are likely to win or lose. Animals assessing that they are likely to lose do not need to continue until they reach a cost threshold, but quit immediately. The original SAM did not suggest how the pieces of information about the opponent and the self were integrated, but subsequent studies implied that they were compared and differences discovered. However, an internal RHP comparator is unnecessary. Information about both RHPs could be transmitted separately and directly to the central motivational neurocircuitry, with increasing self-assessed RHP heightening motivation and increasing opponent RHP inhibiting it (Elwood and Arnott, 2012). A pathway for the latter effect in mice is the input the VMHvl receives from the estrogen receptor alpha-expressing cells in the caudal aspect of the mPOA. This input inhibits aggressive motivation when an opponent’s RHP is superior, as the mice determined from prior fighting experiences or just by exposure to a larger mouse of a more aggressive strain (Wei et al., 2023). When this pathway was inactivated, male mice initiated more attacks toward superior opponents, which then defeated them.

The situation is different when animals contest high-value resources and require a higher certainty about relative RHPs. These contests are predicted to start with low-cost displays that provide limited information about the opponent’s RHP and progress to high-cost challenges in prolonged interactions that may escalate to fighting. The central state-mediated threat-to-attack transition is key in this BE model. It is therefore important to note that the threat–attack linkage can be modified, as shown in fighting fish by habituation of their display to a mirror that does not affect their subsequent intraspecific attack (Meliska et al., 1980) and in mice by the dependence of behaviors following tail-rattling on dominant/subordinate status (Lee et al., 2019). In flies, as noted, there are different command neurons for threat and for attack. Therefore, increasing threat, and the threat-to-attack transition, might involve linked and scaled central states that occur in only some species and may be dependent on circumstance. It remains an experimental and computational challenge to determine which species and what circumstances.

SAM requires an early prediction of outcome that is not found in pure self-assessment or CAM. The mPFC is proposedly involved in aggression outcome predictions in mice (Lischinsky and Lin, 2020). In humans, the activity in the OFC and the perigenual and posterior cingulate increased with heightened reward expectancy for proactive aggression (Wei et al., 2025).

5.6 Fight structure: matching and bouts

A characteristic of E-WoA and CAM is that combatants match the fight intensity up until one quits (Briffa and Elwood, 2009), as noted for zebrafish, Siamese fighting fish, and green anoles (Jenssen et al., 2005; Laan et al., 2017; Vu et al., 2020). One potential mechanism for matching is a tendency for attack to automatically trigger counterattack, as observed in pig and polecat fights (Rushen and Pajor, 1987; Poole, 1974). Similarly, attacks triggering counterattacks between human research subjects were described as “reflexive” (Vandermeer et al., 2019). Given that the many functions of the medial frontal cortex and the adjacent ACC cortex noted in Section 4.6 include the adjustment of effort to both motivational drive and situational demands, this structure could play a role in an E-WoA- or CAM-type strategy of matching the frequency and intensity of the opponent’s threats and attacks in mice and also in people, as suggested by studies involving provocation followed by immediate retaliation or delayed revenge (Crew et al., 2021; van de Groep et al., 2022; Fanning et al., 2017). The latter studies also reported frequent retaliation-related activation of the insula, with the combination of insula and ACC activation suggesting the operation of the salience network.

As noted in Section 3.3, many animal contests are segmented into bouts. Importantly, reductions in the duration of successive bouts in the fight are permissible within E-WoAs and CAM, but not SAM (Jennings et al., 2005). Some insight into the neural mechanisms that affect bout structure comes from the observation that the short-term potentiation that strengthens the MeA–VMHvl pathway increases the aggressive bout duration in mice (Nordman et al., 2020). While the bout is ongoing, it can be prolonged by stimulating the DR axons terminating in the medial aspect of the OFC (MeOFC) or shortened by stimulating the DR axons at the MeA (Nordman and Li, 2020). Notably, DR stimulation by itself does not induce aggression, strengthening the case that it is specifically involved in bout regulation. VMH stimulation in rhesus monkeys produced 10-s bouts of attack, after which the animal would move away despite continuing stimulation (Alexander and Perachio, 1973). Perhaps these were instances of DR bout-regulating inhibition overriding the VMH.

5.7 Discriminating BE models

Animals may display threats, but that does not mean that opponents pay any attention to them. To summarize the discrimination among models:

• If the duration is positively correlated with the loser size but not the winner size, the contest involved self-assessment (Taylor and Elwood, 2003; Arnott and Elwood 2009b).
• If loser size and fight duration are positively correlated, but winner size is negatively correlated, then it could be either CAM or SAM.
• These may be discriminated by examining the fight structure (see Section 5.6), vigor, and the startle responses (Arnott and Elwood, 2009a; Briffa and Elwood, 2009).

These methods have identified pure self-assessment in amphipods (Prenter et al., 2008), spiders (Bridge et al., 2000; Taylor et al., 2001), and fig wasps (Moore et al., 2008). Banded mongooses have been shown to use self-assessment of troop size when deciding to attack another troop (Rayner et al 2026). CAM was supported, at least in part, in studies of house crickets (Briffa, 2008), fiddler crabs (Morrell et al., 2005), and chameleons (Stuart-Fox, 2006). Mutual assessment was strongly supported in studies on squid (Schnell et al., 2015) and mantid shrimps (Green and Patek, 2018). A meta-analysis showed that no model clustered in a particular taxonomic group and that self-assessment was the most frequently identified model (Pinto et al., 2019).

5.8 Multiple strategies

It is becoming clear that assessment strategies can vary: 1) between environmental contexts; 2) among individuals within populations; 3) within individuals across development; and 4) within individuals during a contest (Mesterton-Gibbons and Heap, 2014; Stienecker et al., 2019). Animals that start with mutual assessment and switch to self-assessment include crickets (Lobregat et al., 2019), jumping spiders (Kasumovic et al., 2010), snapping shrimp (Dinh and Patek, 2023), killifish (Hsu et al., 2008), green anoles (Garcia et al., 2012), chameleons (Stuart-Fox, 2006), and fallow deer (Jennings et al., 2005). Identifying the moment and the nature of the change in the brain activity that underlies the shift in strategies would represent a significant melding of BE and BN approaches.

5.9 Fight costs and quit decisions: resisting or accepting defeat

Contest durations and outcomes are decided when losers quit and submit and/or flee. The factors that may go into quit decisions include relatively straightforward fear, pain, injury, and fatigue that might occur with all types of assessment models. These factors may act on various core subcortical structures that support aggression. In more cognitively endowed species, there may be some reckoning of increasing opportunity and/or physiological costs and diminishing chances of victory that suppress the motivation to fight (Reichert and Quinn, 2017).

5.9.1 Pain and injury

Cutaneous pain from being struck, clawed, or bitten can trigger defensive counterattack, as inferred from the observation that bites were rapidly followed by counter-bites in spontaneously fighting polecats and pigs (Poole, 1974; Rushen and Pajor, 1987) and as routinely occurred in rat shock-fighting studies. Bite force is associated with aggressiveness and/or contest advantage in some cricket and lizard species (Adeola and Lailvaux, 2023; Henningsen and Irschick, 2012) and can only be appreciated by being bitten. In mice, GABAergic neurons in the anterior hypothalamic nucleus encoded the intensity of the mechanical “bite” force on the mouse’s tail exerted by the alligator-clip mouth of a model snake (Xie et al., 2020). The pathway from these neurons to the PAG was necessary and sufficient to elicit a defensive counterattack.

However, cutaneous pain may be acutely blunted by non-opioid benzodiazepine and serotonin receptor-mediated analgesic systems that become activated as an animal is being defeated (Pitcher et al., 2017). Nonetheless, repeated cutaneous pain acts as a punishment that can be terminated by quitting. In mice, the pain of being bitten is processed by neurons in the anterior VMHvl, which suppress aggression and mediate the transition to defeat behavior (Osakada et al., 2022; Xie et al., 2022). Deep pain from injury to the tendons and the viscera may be blunted by rapid-onset opioid analgesia generated by neurons in the ventrolateral column of the posterior PAG that express μ-opioid receptors (Kimmey et al., 2025). A single nucleotide polymorphism in the human opioid receptor gene, OPRM1 A118G, which is expressed in the NAc and PAG, imparts an opioid-mediated reduction in pain and confers greater resistance to defeat (Briand et al., 2015). However, deep pain can trigger autonomic responses such as drops in the blood pressure and heart rate that would impair RHP and so, again, lead to defeat.

Aggressive arousal in people, induced by either swearing or playing a violent videogame, increases pain tolerance (Robertson et al., 2017; but see Wagener and Melzer, 2022), and anger can be accompanied by an opioid-mediated hypoalgesia (Adachi et al., 2022; Apkhazava et al., 2018; Bruehl et al., 2011; Yarns et al., 2022). Mixed martial arts (MMA) fighters are permitted the use of painful strikes and holds that are not allowed in Olympic wrestling. Whatever analgesia arousal and/or anger might provide to MMA fighters, the pain inflicted by a superior opponent forces them to “tap out” and accept defeat, just as pain inflicted by an ordinary bully leads vulnerable victims to more rapid capitulation.

5.9.2 Fear

Fear in the context of losing a fight has been extensively demonstrated in the form of conditioned defeat, in which previously defeated animals avoid, freeze, or adopt submissive postures in the presence of former or potential aggressors. It is more difficult to tease out the role of fear and its underlying neural structures, distinct from pain and fatigue, which occur in moments of acute defeat. However, there is a diencephalic circuit for processing conspecific fear in rodents that differs from the networks activated by pain and predator threat. This circuit includes the mPOA, a subpopulation of neurons within the VMHvl, the ventral pre-mammillary nucleus, and the dorsomedial aspect of the DPN (Canteras, 2018; Silva et al., 2016). The DPN is particularly important in that its activity is positively correlated with defensive orienting by the mouse under attack toward its attacker and negatively correlated with aggression-related VMHvl mobilization, but mice and rats differ in this regard (de Almeida et al., 2022). The DPN is critical for responses to threat and attack by a conspecific in terms of risk assessment, avoidance, freezing, and escape, but not defensive upright. This behavioral profile is identified with fear, as noted in Section 4.8, and suggests that the DPN is involved with the expression of this central state.

Relatedly, an fMRI comparison of the responses conditioned to audiovisual predator and conspecific threat, as well as electric shock, found differentially localized changes in blood flow for each type of threat within the human amygdala and hypothalamus (Bertram et al., 2023). Nonetheless, it is difficult to determine how much fear contributes to quitting.

5.9.3 Fatigue

Fighting requires energy and has metabolic costs. It raises the body temperature and creates costly changes in carbohydrate and lipid metabolism. In crabs, the increases in heart rate and oxygen consumption that occur during agonistic display and fighting can persist for 30–60 min (Rovero et al., 2000). Siamese fighting fish had compensatory higher oxygen consumption the night after the fight (Castro et al., 2006). In pigs, the blood lactate and glucose increased with contest duration and peaked during escalated fights (Camerlink et al., 2015). Large animals such as red deer and elephant seals can be so exhausted after a fight that rivals can freely “poach” the females they had fought to defend (Briffa and Sneddon, 2007). In wasps, winning a fight drains energy, downregulates the genes related to energy metabolism, and reduces the fighting ability for at least 3 h (Liu et al., 2020). That is, fatigue has long-lasting effects on the RHP.

Although the accumulation of lactic acid from anaerobic activity affects muscular function, muscles may not be the only, or even the primary, locus of fatigue effects (Potegal, 2023). For example, a small number of glucose-sensing neurons in the MeA may influence defeat transitions (Zhou et al., 2010). Other likely loci for such an effect are the dorsomedial and central parts of the VMH, in which neurons sense glucose and control the homeostatic responses to hypoglycemia. Evidence for such processes includes disinhibition of the dorsomedial hypothalamus by GABA antagonism that substantially increased rats’ time to fatigue on a treadmill, with an accompanying increase in body temperature (Zaretsky et al., 2018). Burst firing in these cells is associated with energy conservation mediated by a shift from glucose toward lipid metabolism (Shao et al., 2022).

Furthermore, the estrogen receptor-expressing neurons within the aggression circuitry of the VMHvl are also glucose-sensitive (He et al., 2020). Some are excited by them, while others are inhibited. A signal from the excited cells to the defeat-related neurons in the anterior VMHvl and/or an inhibitory signal to the aggression-promoting neurons of the posterior VMHvl would provide a route for direct metabolic influence on the switch from offense or defensive aggression to defeat.

The contribution of fatigue to quitting may be mediated at cortical levels by the salience network, in which the insula registers the unpleasant effects of increased body heat, fluid loss, and/or reduced CNS oxygenation and transmits this information to the ACC, thus reducing the determination to fight (McMorris et al., 2018; Potegal, 2023). This proposal is consistent with the changes in the salience network that have been observed in CNS conditions associated with fatigue (Almutairi et al., 2020). More direct evidence comes from observations of increased insula activity with unpleasant exercise and of the ACC activity just before fatigue-induced quitting of exercise (Noakes, 2012).

5.9.4 Resistance to defeat

Multiple behavioral and physiological factors contribute to resistance to defeat, including winner/loser effects. For example, animals that have achieved dominance are more willing to fight and are more resistant to defeat, as shown when dominant male hamsters that were used as intruders defended more actively against attacks by residents and had longer latencies to submission than did previously subordinate hamsters (Dulka et al., 2020). Social defeat engenders both short-term fear and longer-term social anxiety, both of which are mitigated by the dominance status. As noted in Section 4.6.1, the dominance status is tracked and maintained by neurons in the dmPFC. Fear mitigation in mice and male hamsters most likely involves inhibitory projections from the prelimbic and infralimbic dmPFC, respectively, to the BLA, as shown by the levels of c-Fos activation in the respective regions, as well as by GABA agonist micro-injection into this frontal cortical area (Dulka et al., 2020; Grossman et al., 2022). Equally, anxiety mitigation involves inhibitory projections from the androgen-sensitive cells in the posterior MeA to the BNST, as indicated by a combination of c-Fos and retrograde BNST labeling (Whitten et al., 2024).

Neurochemically, 5HT in invertebrates can contribute to defeat resistance, as shown by its infusions into the hemolymph of crayfish with smaller claws, which then fought longer against larger clawed conspecifics, but went on to lose anyway (Fuxjager et al., 2017). Alterations in the vertebrate dopaminergic VTA-to-NAc reward system are also involved in resistance to defeat. Activation of the lateral habenula in zebrafish increased the likelihood of winning a fight, while medial habenula activation reduced this (Chou et al., 2016). These manipulations likely affected the reward value of aggression and were interpreted as shifting the “threshold for surrender.”

5.9.5 Deciding to quit

At least three different inhibitory inputs to the mouse VMHvl have been identified to reduce aggressive motivation and/or behavior. Input from the mPOA reduces VMHvl activity and thus limits aggressive motivation, as noted in Section 4.11. Activity recorded in a peri-VMHvl zone, the VMHvl “shell,’ is associated with rapid termination of aggression: stimulation in the shell causes a prompt exit from an aggressive motivational state (e.g., the operant behavior to obtain opportunities to attack ceases) (Minakuchi et al., 2024). The conditions under which activity in the VMHvl shell occurs naturally remain undetermined. In contrast, a linkage between pain and aggression termination is provided by oxytocin receptor-expressing cells in the anterior VMH that control counterattack, but are also acutely activated in social defeat (Osakada et al., 2022). When activated, they suppress the firing of VMHvl aggression-promoting neurons and, thus, are sufficient for the transition from counter-aggression to defeat-related behavior. These anterior VMH neurons receive oxytocin input from a subpopulation of cells in and around the retrochiasmatic supraoptic nucleus, which respond to cutaneous pain, thus providing a neural pathway from injurious attack by an opponent to defeat.

In both crickets and humans, an aggressive interaction followed by an aversive stimulation that seemingly confirms defeat is sufficient to reduce aggressiveness and/or induce submission (Rillich and Stevenson, 2017; Wilkowski and Leki, 2019). In crickets, this effect is centrally mediated by nitric oxide acting as a gaseous, diffusible neurotransmitter. The neurophysiology of this effect in humans is unknown, but the phenomenon itself fits with an involuntary defeat strategy (IDS) that is conjectured to have evolved to automatically forestall defeated individuals from continuing potentially dangerous contests with superior rivals (Sloman et al., 2006). An IDS whole-body response involves reduced cardiac parasympathetic control, a downcast gaze and a slumped posture, and social withdrawal (Potegal, 2023), Subjectively, IDS is experienced as a “mental” defeat, with feelings of helplessness, hopelessness, social inferiority, physical inadequacy, and fatigue. Serious defeats also engender shame (Cabral et al., 2023). IDS is “involuntary” because it is triggered automatically and often against someone’s conscious wishes. It essentially functions to “de-motivate” competition. IDS presumably occurs following serious conflicts over important resources.

5.10 First strikes and sneak attacks

An alternative to the threat–attack escalation implied by SAMs is to attack first with maximum vigor, thereby winning quickly, as seen in Lake Victoria cichlids that were aggressively aroused by 15 min of mirror fighting (Dijkstra et al., 2012). The “strike first” strategy also worked for stomatopods (Green and Harrison, 2020) and for sticklebacks that had the physiological muscle reserve to keep fighting (Guderley and Couture, 2005). It worked for pigs when their opponent quit without fighting, but conveyed no advantage when the opponent fought back (Camerlink et al., 2016).

The first-strike strategy is documented in court cases of assault and homicide and is found in the fear-triggered preemptive attacks on more powerful or better-armed opponents (Felson and Hullenaar, 2021). These human “adversary effects” are reminiscent of the “Napoleon effect,” in which small males of several species initiate aggression against larger opponents (Jenssen et al., 2005). One hypothesis about the latter strategy is that the value of the resource at stake is assessed as higher than the costs of the contest (Kareklas et al., 2019; Svensson et al., 2012).

Of the 100 street fight videos sampled from the Internet, 52 were simple unilateral assaults in which only winners struck a blow (Potegal et al., 2025). Of these, 16 showed an apparent “sneak attack,” e.g., a strike from behind, in which there was no apparent pre-fight signaling and the target appeared to not be attending to the other fighter before being struck for the first time. We are unaware of any documentation of this extreme form of first-strike strategy in other species.

5.11 Assessments in humans

Stronger men are more aggressive, indicating self-assessment of physical strength (see Section 3.5.2). Humans also judge the strength and fighting ability in others (e.g. Sell et al., 2009) and thus show mutual assessment. Other examples of mutual assessment in antagonistic human encounters are the initial feints and jabs that fencers, boxers, and other combat sports competitors use to initially gauge and “feel out” an opponent’s reactions (Czajkowski, 2009). Cognitive expectations of opponents’ actions are likely to be mediated by the D/MN, as noted in Section 4.7.2. Indeed, studies eliciting “hostile attribution bias,” i.e., expectations of antagonism, very typical of chronically angry and reactively aggressive individuals, consistently reveal the activation of both the salience network and the D/MN (Wagels and Hernandez-Pena, 2024).

The majority of attacks in street fight videos were preceded by threats and other signals of aggressive intent that would have been one source of information for pre-fight assessment (Potegal et al., 2025). However, majority of the fights were too short and too violent to permit intra-fight assessment. In particular, hard blows to the head rapidly incapacitate losers who then have no opportunity to make assessments or quit decisions.

6 Conclusion

BE, HSP, and BN have each gained their own insights into the characteristics and processes of aggression. BE has elucidated the resource benefits, fight costs, and RHP factors that determine the probability of fighting and the outcomes of fights. For example, BE modeling predicts that the value of the resource should influence the commitment to fighting: virtually all empirical studies show that fights are longer and/or fiercer when that value is high (Arnott and Elwood, 2008). BE modeling has also been clear about how RHP affects contest decisions, with empirical studies consistently finding that larger losers persist for longer (Arnott and Elwood, 2009a; Pinto et al., 2019). HSP has focused on humans, largely ignoring commonalities with other animals. BN has elucidated some of the proximal neurobiological mechanisms involved in the instigation of threat and its escalation to fighting, a prime example being the neuronal activity within the VMHvl that instantiates offense motivation. BN studies of people have shown that the amygdala is involved in threat detection, as in other animals. So, too, is the PAG involved in defensive responding across species. Happily, there is current general agreement on the existence of transient states of increased aggressive motivation that contribute to fight probability and outcome, although there are differences in their measurement. These three disciplines should contribute to each other to broaden and deepen their mutual understanding of aggression in animals in general and the social psychology of human aggression in particular.

BE primarily aims to understand the function and evolution of behavior, but many studies on aggression implicitly use ideas linked to causation. Indeed, the widely used BE term “assessment” relates to the acquisition of information that is then used in decision-making (Parker and Stuart, 1976). The different assessment models noted in Section 5 simply differ in the nature of the information that is gathered. Although each functional model explores the maintenance of strategies in the population, they each have elements that are similar to the models of motivational state (Parker and Stuart, 1976; Enquist and Leimar, 1983). This has led to the development of explicit two-dimensional motivational models (Maynard Smith and Riechert, 1984; Enquist and Leimar, 1987; Huntingford and Turner, 1987). However, these motivational models tended not to explore variation specifically in the resource value and the contest cost. For example, Huntingford and Turner (1987) described the two key factors in aggressive motivation as aggression and fear. This use of terms relating to emotion has been a feature of HSP and subsequently developed within BE as factors that mediate the changes in aggression motivational state (Crump et al., 2020). However, because of the vast evidence showing that the resource value and contest cost (actual or predicted) affect contest decisions, then these measurable features should be used when considering motivational state (Elwood and Arnott, 2012). This more intuitive approach can account for the contest structure, the activities employed, and the durations noted in empirical studies.

One practical gap between fields arises from studying different species. The broad view of BE encompasses many phyla studied in both the laboratory and the field and attempts to generate general models that promote understanding of aggression across these diverse species. Neuroscience has been more restricted, with some work on fruit flies, but much more on rodents and, since the advent of optogenetics, on mice. Greater crosstalk between these two disciplines might be facilitated if both focused on some amenable species. Immediate/early gene mapping has identified activity within the social decision-making brain network in zebrafish, suggesting that they might be especially suitable for research into the genetic, neural, and hormonal control of aggression. HSP, by definition, focuses on a single species. Placing human data in a broad BE perspective, coupled with guidance from the proximate mechanism from BN, should provide new insights.

BN has elucidated the neural pathways of mutual assessment in rodents in terms of olfactory/vomeronasal cues to the age, sex, and dominance status of conspecifics. It remains to be determined whether rodents also switch from mutual to self-assessment as fights progress. Relinquishing the asymmetric resident–intruder paradigm in favor of fights between animals of equal status could initiate the neuroscience of rodent quit decisions. Recording from individual neurons during unconstrained fights may be more challenging, but other neuromethodologies could be applied. As noted, the effects of perceived opponent RHP has been documented in human formidability studies. However, these studies have only rarely assessed the role of self-assessment in subjects of different RHPs, such as height, muscularity, and martial arts training, and have not done so in the context of an actual confrontation (cf. Fessler et al., 2014; Raine et al., 2018).

Shared methods of assessing aggressive motivation and behavior are also needed. It would be of considerable interest to compare startle pause recovery measurements of aggressive motivation in rodents to other measures currently in use, e.g., operant behaviors. The comparison could be extended across taxa. Siamese fighting fish have swum through rings to access a display challenge in multiple studies (Wirth et al., 2003). Zebrafish also learn operants (Kuroda et al., 2017). Finer-grained neuroimaging of the core subcortical structures will determine whether the human hypothalamus instantiates central motivational states corresponding to anger and/or dominance-related aggressive arousal. It would be of great interest to determine what, if any, neural activity correlates with subjects’ self-reports of their angry or non-angry aggressive motivation.

Issues within BE also need to be addressed. The characteristics of fights in some species do not fit any standard BE assessment model (Kareklas et al., 2019). Even when they do fit one model, there is often considerable variance in, e.g., the fight durations as a function of loser and winner RHP. One reason may be the choice of proxies for RHP. For example, weight, length, or flank surface area has been used as RHP measure for fish. While fish can visually inspect other fish, their eye positions do not permit self-inspection. They must track self-RHP through other signals (Elwood and Arnott, 2012). Both signals could independently modulate the central aggressive state, as suggested. Thus, the scope and accuracy of BE modeling could be improved by incorporating what BN is learning about inputs that modify central aggressive states.

There are two further challenges to the quantitative modeling of the assessment strategies involving central states that have not been fully considered. One of these is the possibility that performing threat displays may increase the displayer’s own aggressive arousal (Bond, 1989). There is certainly evidence that autonomic and sensory feedbacks from the physiological concomitants of anger serve to intensify it, e.g., in cardiac “excitation-transfer” effects (Zillmann, 1994). Anger intensifying feedback from angry facial expressions is particularly robust (Coles et al., 2019). The second challenge is the even clearer evidence that offense is rewarding. This challenge relates firstly to mutual assessment, BE modeling of which should arguably incorporate the phenomena of “social instigation” and “attack priming,” both of which increase the initial aggressive motivation. Furthermore, self-reinforcing behavior might be expected to continue indefinitely. However, satiation of aggression does occur, as noted in Sections 2.3.6 and 4.5. Accordingly, BE modeling of aggressive interactions should take into account both the initial reward value of aggression and the possibility of its eventual satiation.

Aggression has been crucial in the evolution of animal species because the resulting unequal division of resources has driven natural selection. This must also have occurred during human evolution. Offense and defense are natural, genetically determined, and experientially modified behaviors that are pervasive across taxa. A greater understanding would be useful in attempts to reduce the negative impacts of aggression for farm animals that are exacerbated by captive conditions that do not allow escape from aggressors. For humans, the socially and culturally exacerbated aggression of modern times, typically involving manufactured weapons of great lethality, has devastating impacts on the victims, their families, and the communities and society in general. A deeper understanding of developments in BE and BN may provide insights leading to the social psychological amelioration of aggression.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Aras Petrulis, Georgia State University, United States

Reviewed by: Kim Huhman, Georgia State University, United States

Meghan Sosnowski, University of California, Davis, United States