2.2.1. A function for revenge: Changing other individuals' incentives toward the self

We think revenge solves an adaptive problem that is faced by many species: how to change other organisms' incentives to emit benefits and to avoid imposing costs upon oneself (Clutton-Brock & Parker Reference Clutton-Brock and Parker1995; Daly & Wilson Reference Daly and Wilson1988; Tooby & Cosmides Reference Tooby, Cosmides, Runciman, Smith and Dunbar1996). Specifically, we hypothesize that cognitive mechanisms for revenge evolved because their behavioral outputs (i.e., retaliatory impositions of costs or withholdings of benefits) caused individuals to revise downward the net returns they expect to receive by engaging in exploitive behaviors against the vengeful individual in the future, which in turn (a) deters them from efforts to exploit the retaliator (Bshary & Grutter Reference Bshary and Grutter2005) or (b) induces them to emit benefits for the sake of the retaliator. Cognitive systems that motivate organisms to provide these sorts of contingent punishments and rewards (Seymour et al. Reference Seymour, Singer and Dolan2007) may boost their bearers' lifetime reproductive fitness (Chagnon Reference Chagnon1988; but cf. Beckerman et al. Reference Beckerman, Erickson, Yost, Regalado, Jaramillo, Sparks, Iromenga and Long2009) and thereby evolve for their ability to change other organisms' incentives toward the self (Tooby & Cosmides Reference Tooby, Cosmides, Runciman, Smith and Dunbar1996). Our approach to revenge has an affinity with Sell et al.'s (2009) recalibrational theory of anger, which claims that anger is an evolved motivational program designed to motivate behavior in the angry individual that will cause the individual at whom the anger is directed to revise upward the value he or she places on the angry individual's welfare.

2.2.2. Welfare tradeoff ratios

To further develop our proposal that revenge is designed to raise individuals' incentives toward the self, we use Tooby et al.'s (2008) Welfare Tradeoff Ratio (WTR) concept. Tooby et al. (Reference Tooby, Cosmides, Sell, Lieberman, Sznycer and Elliott2008) conceptualize WTRs as internal regulatory variables, stored in memory and continually updated, that humans use to guide social decision making according to appropriate criteria. An individual's WTR for a target individual reflects how much the bearer values the target's welfare – operationally, how large a benefit to the self the bearer would be willing to take if doing so required imposing a given cost of upon the target. WTR values of 1:1 indicate that the actor values the welfare of the target individual equally to his or her own: The actor would impose a 1-unit cost upon the target individual if and only if it, in turn, led to at least a 1-unit benefit for the self. WTRs less than 1 – say, 1:2 – indicate that the actor values the target individual's welfare one-half as much as one's own, and therefore, that the actor would willingly impose a cost of up to 2 units upon the target to obtain one unit of benefit or more for himself or herself (Petersen et al. Reference Petersen, Sell, Tooby, Cosmides and Høgh-Olesen2010; Sell et al. Reference Sell, Tooby and Cosmides2009; Tooby et al. Reference Tooby, Cosmides, Sell, Lieberman, Sznycer and Elliott2008). A WTR of 0 indicates that an actor would willingly impose a cost of any size upon the target to obtain even an infinitesimally small benefit: such a target's welfare is completely ignored. Moreover, individuals will impose costs on enemies even if doing so is a cost, rather than a benefit, to themselves: Reflecting this reality of social behavior, WTRs can be negative (Petersen et al. Reference Petersen, Sell, Tooby, Cosmides and Høgh-Olesen2010).

WTRs, according to Petersen et al. (Reference Petersen, Sell, Tooby, Cosmides and Høgh-Olesen2010) and Tooby et al. (Reference Tooby, Cosmides, Sell, Lieberman, Sznycer and Elliott2008), exist in two natural kinds: monitored and intrinsic. An actor's monitored WTR for a target individual expresses the regard an actor should have for the target's welfare in light of the social consequences that the actor's behavior might trigger if the target (or that individual's kin, friends, or allies) were to discover the actor's behavior and then attempt to harm or help the actor on the basis of that information. Monitored WTRs are responsive to factors that modify a target's ability to respond retributively (i.e., with rewarding or punishing), such as physical strength, coalition size, and access to resources (Petersen et al. Reference Petersen, Sell, Tooby, Cosmides and Høgh-Olesen2010). Monitored WTRs regulate actors' behaviors by way of systems that estimate the probability of contingent reprisals and systems that estimate the probability of contingent rewards from the targets of those behaviors.

In contrast to monitored WTRs, intrinsic WTRs express the regard that an actor possesses for a target individual's welfare solely due to the indirect effects of the actor's behavior toward the target on the actor's own welfare (rather than because of the target's ability to monitor and impose rewards or sanctions). Intrinsic WTRs are cognitive representations of the interdependence of the welfare of an actor and a target irrespective of any retributive responses that the actor's behavior might elicit from the target of that behavior. The claim that people can have different monitored and intrinsic WTRs for the same target individual explains how it is possible for humans to treat a feared coworker with kid gloves in the workplace, but then derogate the coworker in private. To know whether individual A's behavior toward individual B is being guided by A's monitored WTR toward individual B, one must observe how A's regard for B (as revealed through behaviors that influence B's welfare) changes as a function of whether B is able to learn of A's welfare-relevant action. This claim can also explain why humans behave with regard for their children's, mates', and friends' welfare even in situations in which the target individual could not possibly observe the actor's behavior.

Surrounding oneself with individuals who have high monitored WTRs toward the self (if monitoring is feasible) and high intrinsic WTRs toward the self (under all circumstances) is beneficial, as is distancing oneself from individuals who do not. In modern environments, however, many individuals in relatively close proximity consistently fail to provide benefits that they plausibly could provide. For example, people could spontaneously offer strangers the contents of their wallets, but typically they do not. Generally, however, such non-deliveries of benefits neither cause humans to feel entitled to better treatment, nor to attack those that do not provide such benefits. Instead, humans form expectations for how other individuals should treat them based on relationship categories such as friend, father, mother, ally, stranger, enemy, dominant, and subordinate. Expectations also take into account other actors' behavioral histories, including direct experience and reputational information, as well as their perceived ability to harm and help (Petersen et al. Reference Petersen, Sell, Tooby, Cosmides and Høgh-Olesen2010).

Assuming, then, that humans have computational processes for generating baseline WTRs (of both the monitored and intrinsic varieties) for other individuals – and estimates of those other individuals' baseline WTRs toward them – we also posit the existence of cognitive routines for registering that an actor has treated the self with less regard (i.e., that an actor has committed an action that connotes a lower WTR toward the self) than one would have expected based on one's previous estimate of the actor's WTR toward the self. At issue here is not simply whether an individual imposed a cost upon oneself, but whether that cost imposition was permissible given the victim's understanding of the harmdoer's WTR for the self. Parents, for example, readily abide their children's imposing many costs upon them over the life course, and few of those costs are met with revenge because those costs are perceived as permissible given not only the parents' high WTRs for their children, but also the parents' acceptance of their children's relatively low WTRs toward their parents. (Children's behavior, we would argue, is adaptively organized by cognitive mechanisms that motivate children regularly, and largely with impunity, to impose all sorts of costs – both large and small – on their parents. Breastfeeding, begging for food or attention, and tantrums are examples. Parents tacitly accept such costs even if they are inconvenient and annoying.) Understanding a harmdoer's intentions is important because accidental harm does not provide information about the actor's WTR. Intentional harm, however, implies that the harm was caused by the harmdoer's low WTR for the victim (Petersen et al. Reference Petersen, Sell, Tooby, Cosmides and Høgh-Olesen2010; Sell Reference Sell2011).

2.2.3. Revenge and WTRs

The introduction of the WTR concept here enables us to put our proposal concisely: The revenge system, by motivating retaliatory harm, was selected because it caused other individuals to raise their WTRs for avengers (see also Petersen et al. Reference Petersen, Sell, Tooby, Cosmides and Høgh-Olesen2010; Sell et al. Reference Sell, Tooby and Cosmides2009) so that those individuals would refrain from imposing costs upon the avengers in the future. Imposing a large harm on a victim to obtain a small benefit for the self indicates that the aggressor does not highly regard the victim's welfare relative to the aggressor's own welfare. Revenge is an effort to compel an aggressor to increase his or her regard for the victim's welfare – essentially, to teach the aggressor that imposing costs of the same size upon the victim in the future (should they be detected) will be met with retaliatory costs. Again, the central logic is the logic of deterrence: If an aggressor learns that the victim will impose large retaliatory costs, then the aggressor can be made less likely to perform such acts in the future.

By “regard for the victim's welfare,” we do not mean that revenge causes increased feelings of care or concern. Instead, we are arguing that revenge produces vigilance about imposing future costs upon the avenger. Thus, we define revenge as a targeted imposition of costs or withholding of benefits, in response to a cost-inflicting (or benefit withholding) event, that results from a cognitive system designed (i.e., selected) for deterring other organisms from imposing costs (or inducing other organisms to confer benefits) upon oneself or other individuals in whom one has fitness interests.

2.2.4. Other definitional considerations

Our definition of revenge is similar to many evolutionary biologists' definition of punishment. Many biologists have been heavily influenced by Clutton-Brock and Parker (Reference Clutton-Brock and Parker1995), who defined punishment as a costly (in the currency of fitness) imposition of costs (in the currency of fitness) on another individual that results in delayed benefits (in the currency of fitness) for the punisher (see also Jensen Reference Jensen2010). Although our definition of revenge broadly matches Clutton-Brock and Parker's definition of punishment, we depart slightly from their approach: Our claim is that the imposition of costs at a cost is only revenge when it is a response to a harm-imposition or a benefit-withholding that was caused by a mechanism designed to deter cost-impositions or benefit-withholdings in the future. Revenge, by our definition, therefore differs from other forms of punishment such as those administered by individuals acting as representatives of social institutions (e.g., judges, school principals), and in the context of precautions or shaping (e.g., the adult who scolds a child for swearing, or the behaviorist who wants her laboratory rats to learn to avoid the first left turn in a maze).

Our attempt here is to model interactions among individuals rather than among groups of individuals. We take no position on whether the psychology that governs the operation of revenge systems (and the reconciliation systems we will discuss later) also evolved to regulate behavior in intergroup contexts. The computations required to make adaptive decisions about revenge (and forgiveness) in the context of intergroup conflict (e.g., see Lickel et al. Reference Lickel, Miller, Stenstrom, Denson and Schmader2006) might be different enough from those required to make adaptive decisions in the context of conflicts between individuals within a single living group as to require a distinct cognitive architecture (Petersen et al. Reference Petersen, Sell, Tooby, Cosmides and Høgh-Olesen2010).

Before continuing, we clarify four points. First, when one's kin, friends, allies, mates, or offspring are harmed, one suffers indirect harm (in the currency of fitness), and the revenge system should reflect this fact (Lieberman & Linke Reference Lieberman and Linke2007). Similarly, one can impose retaliatory harm by imposing costs on a provoker's kin, friends, allies, mates, and offspring (Aureli et al. Reference Aureli, Cozzolino, Cordischi and Scucchi1992; Engh et al. Reference Engh, Siebert, Greenberg and Holekamp2005; Gould Reference Gould2000). Our analysis of revenge applies to direct impositions of cost, as well as indirect costs that accrue via their effects on one's kin, mates, and affiliates (Aureli et al. Reference Aureli, Cozzolino, Cordischi and Scucchi1992; Bernhard et al. Reference Bernhard, Fischbacher and Fehr2006; Hamilton Reference Hamilton1964; Shinada et al. Reference Shinada, Yamagishi and Ohmura2004). In the remainder of this article, we do not always specifically add these indirect considerations in the interest of brevity. Second, and related, we intend for costs to be in the currency of fitness, even though the relevant psychology is not tracking fitness costs per se, but rather, appropriate proxies (e.g., somatic damage, damage to one's property, reputation, social relationships, etc.).

Third, we include as revenge many behaviors that do not, as a matter of fact, manage to deter anyone (as when people behave aggressively toward a driver whom they perceive to have mistreated them on the road, and with whom they will never interact again). Because the modern world consists of many one-shot interactions, in (possible) contrast to ancestral human environments (Hagen & Hammerstein Reference Hagen and Hammerstein2006), systems designed to deter do not always implement their proper functions (Sperber Reference Sperber, Hirschfeld and Gelman1994) – that is, the effects for which they were selected. Evolved mechanisms are not expected to perform the jobs for which they were naturally selected with perfect fitness-optimizing performance in every possible environment (West et al. Reference West, Mouden and Gardner2011).

Fourth, harming a provoker is only revenge, we propose, when the system that motivated the harmful behavior was designed by selection pressures for deterrence. So, for instance, avoiding a provoker to avert a second harm is not revenge (instead, the harm imposed by avoidance, if any, might be a by-product of a mechanism designed to reduce the probability of future harm to the self), but avoiding a provoker for the purpose of limiting his or her access to a valued resource (i.e., to withhold benefits) might constitute revenge (Barnes et al. Reference Barnes, Brown and Osterman2009). In subsequent sections, we situate avoidance as a strategy per se among a larger suite of strategies that includes avenging the harm, forgiving it, and ignoring it (and perhaps others).

3.1.1. Direct deterrence of cost impositions

By direct deterrence, we mean that revenge discourages an aggressor (who imposed a cost upon an avenger) from imposing other costs upon the avenger in the future. The logic here, identical to the logic underlying deterrence theories of punishment (Bentham Reference Bentham and Bowring1962), is that if a potential aggressor can choose whether to take an action that imposes costs on a potential victim to acquire some benefit, then the potential victim is better off if he or she changes the potential aggressor's incentives so that the expected value of the cost-imposing action for the potential aggressor is lowered. Revenge transforms expected value in this fashion by causing an aggressor to learn that the retaliatory infliction of fitness costs in the future is likely to exceed (or, at least, substantially reduce) the potential benefits to be gained by harming the potential victim.

To work as a deterrent, however, avengers must act in such a way that the aggressors update their expectations of the avenger's future behavior (and, in so doing, revise their monitored WTR toward the avenger, with the consequence that the harmdoer becomes more vigilant in avoiding imposing costs upon the avenger in the future). If an aggressor imposes a cost of 1 on a victim to obtain a benefit of 2, and the victim proceeds to impose a retaliatory cost greater than 2, this retaliatory response only acts as a deterrent if the aggressors updates his or her representation of future options to reflect the possibility that the victim will again impose (similarly sufficiently large) offsetting costs upon the aggressor in the future. Effective updating, of course, requires an array of computational steps: The target of revenge must, among other things, (a) store an internal regulatory variable (Tooby et al. Reference Tooby, Cosmides, Sell, Lieberman, Sznycer and Elliott2008) that represents his or her level of regard for the avenger's welfare (this is Tooby et al.'s [2008] monitored welfare tradeoff ratio), (b) represent the magnitude of the costs and benefits, (c) represent the causal relationship between the initial aggressive act and the retaliatory act, and crucially, (d) infer that retaliation at one time point is diagnostic of how the avenger is likely to behave in similar future situations. As Petersen et al. (Reference Petersen, Sell, Tooby, Cosmides and Høgh-Olesen2010) point out, targets of revenge must also be able to make generalizations about the ranges of behaviors toward the avenger that are likely to be met with vengeance in the future.

The logic of revenge gives rise to strategic complications. For example, though revenge at time 1 might predict revenge at time 2, nothing forces this to be true. An organism could be, for example, intermittently vengeful. This leads to well-known problems associated with inducing others to learn that one's vengeful dispositions are stable over time (Frank Reference Frank1988; Hirshleifer Reference Hirshleifer and Dupré1987). This idea is of interest in the field of international relations because nations must signal their willingness to take revenge, even during the end game of an armed conflict, if they wish to deter conflict in the first place (Schelling Reference Schelling1960).

Experimental evidence in support of the proposition that revenge is well-suited to deterring the imposition of costs comes from studies of human behavior in economic games such as the sequential and iterated Prisoner's Dilemma (Axelrod Reference Axelrod1980; Reference Axelrod1984). This literature is too large to summarize exhaustively, but several stylized findings are worth considering. In the sequential Prisoner's Dilemma Game there is one round of play, but the second mover chooses only after seeing the first player's choice. In such games, the second player is much more likely to cooperate after a cooperative move than after a defecting move. More relevant to our present point, defection is almost always met with retaliatory defection (see Table 6 in Clark & Sefton Reference Clark and Sefton2001; cf. Hayashi et al. Reference Hayashi, Ostrom, Walker and Yamagishi1999).

In the iterated Prisoner's Dilemma, subjects play multiple rounds of the simultaneous move game with either the same partner or different ones. For the present purpose, key issues are whether people respond to defection with defection – moves plausibly interpretable as revenge (though clearly open to other interpretations, such as loss prevention) – and whether such moves elicit subsequent cooperation from one's partner. Experiments using large numbers of trials in Prisoner's Dilemma Games suggest that people do respond to defection with defection (Bixenstine & Wilson Reference Bixenstine and Wilson1963) though the details vary across studies (Rapoport & Chammah Reference Rapoport and Chammah1965). Reciprocal strategies such as tit-for-tat tend to elicit cooperation (e.g., Axelrod Reference Axelrod1984; W. Wilson Reference Wilson1971), hinting at the effectiveness of meeting defection with defection for eliciting subsequent cooperation (Gardner & West Reference Gardner and West2004).

Moreover, in an analysis of data from 5 different laboratory studies of dyadic negotiation in which partners played 250 consecutive trials in which they could either punish, reward, or withhold reward (and punishment) from each other, Molm (Reference Molm1997, see especially pp. 126–34) found that the frequency with which actors used retaliatory punishment (i.e., after one's negotiation partner had punished the actor in a previous move) was positively associated with the frequency with which they rewarded each other. Further, the partner of each dyad who used contingent punishment more frequently in response to his or her partner's previous cost imposition was the partner who received more benefit from it. In contrast, the frequency with which dyads punished non-contingently (that is, independently of whether the punishment was a retaliatory response to punishment or the withholding of benefits) was associated with lower rates of rewarding: It is only when punishment is contingent on previous punishment (or the withholding of benefits, as we elaborate in section 2.2.3) that it promotes cooperation.

In some situations, one can benefit from revenge's efficacy as a deterrent simply by advertising one's ability to retaliate; it is not always necessary to do so. The difference in how people play the Dictator Game, as opposed to the Ultimatum Game (Güth et al. Reference Güth, Schmittberger and Schwarze1982), illustrates this point. In both games, some amount of money, say $10, is to be divided between two people. In the Dictator Game, one person unilaterally decides how to split the money. In the Ultimatum Game, one person – the Proposer – proposes a split, and the other person – the Responder – can either accept the proposal or reject it, in which case both players receive nothing. When the Responder rejects the Proposer's offer, then, the Responder is penalizing the Proposer by the amount that the Proposer set aside for himself or herself, but to impose this penalty, the Responder also pays a price: the amount that the Proposer had allocated to the Responder.

Unsurprisingly, typical proposals in the Ultimatum Game, in which punishment is possible, are larger (roughly 40% of the stake; Oosterbeek et al. Reference Oosterbeek, Sloof and Van de Kuilen2004) than they are in the Dictator Game (roughly 28%; Engel, in press), in which punishment is not possible. Along similar lines, Andreoni et al. (Reference Andreoni, Harbaugh and Vesterlund2003) compared the results of a Dictator Game to those of three other games in which the Responders could either (a) punish, (b) reward, or (c) both punish and reward. In the punishment condition, the second player could pay one unit to impose a five-unit cost upon the first player. Under such conditions, players specify larger transfers to the second player, though transfers are still higher when the second player has access to (using the authors' metaphor) both a carrot (i.e., increasing the first player's payoff) and a stick (i.e., the capacity to inflict retaliatory costs).

The Trust Game (Berg et al. Reference Berg, Dickhaut and McCabe1995) shows a similar deterrent effect for the ability to punish. In the Trust Game, Player 1 – also sometimes called the “investor” – starts with an endowment of money and is given the opportunity to transfer some of it to Player 2, also known as the “trustee.” Transferred money is multiplied (often tripled), and the trustee can then return some, none, or all if it back to the investor. Money sent by the investor is commonly interpreted as trust, and money returned is commonly interpreted as trustworthy behavior on the part of the trustee. Hopfensitz and Reuben (Reference Hopfensitz and Reuben2009) used a game like the trust game in which the investors had only a binary choice – trust or not trust – and the trustees had only 3 choices – low, medium, and high levels of trustworthiness. When punishment is added to this game (by allowing investors and trustees alike to punish their partners, subject to certain conditions), investors are more likely to make the trusting choice, and trustees on average return more money back to investors. Here, the possibility of punishment changes behavior in the desired way: it increases trusting and trustworthy behavior. Fehr and List (Reference Fehr and List2004) reported similar results with students and CEOs.

At least one laboratory experiment also shows how the risk of retaliation deters aggressors from harming the prospective avenger. Diamond (Reference Diamond1977) had undergraduate men write an essay that a confederate proceeded to harshly criticize. Participants came back to the laboratory 24 hours later and were given the opportunity to give ten (bogus) shocks of varying intensities to the person who wrote the insulting reviews. Half were led to believe that after they administered shocks, they would then switch roles and receive shocks themselves. People who believed that they could harm the insulting evaluators without the threat of retaliation gave stronger shocks to the evaluators. Thus, the fear of retaliation deterred aggression. More generally, the possibility of retaliation has been used to explain why defections within cooperative systems (e.g., the relationships between cleaners and clients in cleaner-client mutualisms), and thus, punishment as well, is rare in many cooperative animal systems (Cant Reference Cant2011).

3.1.2. Indirect deterrence of cost impositions

Psychological mechanisms for revenge might be designed to deter would-be aggressors, including those who have not yet exploited the avenger. The logic behind deterring third parties is parallel to that of deterring second parties. If a third party must decide whether to impose a cost on an individual, the prospective victim can change the third party's incentives if he or she previously demonstrated that when a previous provocateur took a similar course of action, he or she inflicted costs upon that provocateur. To the extent that the third party believes that his or her prospective victim will be consistent in his/her propensity for revenge – an important limitation – the third party will alter his or her choices accordingly. In other words, by knowing that an individual is prone to avenging costs that others have imposed upon him or her, third parties will learn to treat the avenger with greater care, and with less willingness to extract costs from the avenger, in the future (dos Santos et al. Reference dos Santos, Rankin and Wedekind2011).

The computational demands here are non-trivial. For revenge to induce learning in third parties, third parties must be able to categorize actions adequately to determine whether the aggressive acts they are contemplating against a prospective victim are sufficiently similar to previous aggressors' actions against the prospective victim to draw the prospective victim's retaliation. Third parties must also be able to track the causal structure of their prospective victims' vengeful acts. Finally, third parties must be able to compute, for any range of benefits to be extracted from prospective victims, the retaliatory costs those prospective victims are likely to subsequently impose. Despite the computationally intensive nature of third-party deterrence, it is plausible that revenge functions to deter both potential repeat offenders and would-be first-time offenders in a way that parallels the legal distinction between special deterrence (directed at recidivism of a criminal offender, in particular) and general deterrence (directed at other possible violators). Psychological revenge systems can do double duty in the same way that criminal justice systems do.

Reputation, then, might play an important role in third-party deterrence (dos Santos et al. Reference dos Santos, Rankin and Wedekind2011). To the extent that ancestral humans lived in small, close-knit groups (Boehm Reference Boehm2008) without police, courts, and prisons for protecting individual rights, a readiness to retaliate might have been an important component of people's reputations that would not only benefit them within a single living group, but, more importantly, as they transferred from one small living group to another (Marlowe et al. Reference Marlowe, Berbesque, Barrett, Bolyanatz, Gurven and Tracer2011). Researchers have documented the importance of defense of honor, and the revenge that it stimulates, as a major cause of violence among people from many societies, including Mediterranean herding societies (Black-Michaud Reference Black-Michaud1975), tribal Montenegro (Boehm Reference Boehm1987), urban white males in the Southern United States (Cohen et al. Reference Cohen, Nisbett, Bowdle and Schwarz1996; Nisbett & Cohen Reference Nisbett and Cohen1996), and disadvantaged urban African-Americans (Anderson Reference Anderson1999). A longitudinal study of roughly 900 adolescent boys also revealed that boys who endorsed street-code beliefs (e.g., that violence is an appropriate response to insults and violations of honor) went on one year later to engage in more violence, including greater participation in gang fights and attacks in which their goal was to seriously injure or kill someone (Brezina et al. Reference Brezina, Agnew, Cullen and Wright2004).

Laboratory research supports the notion that the psychological mechanisms that cause revenge are sensitive to the presence of third parties, which is consistent with the idea that revenge is enacted partly out of reputational concerns. Victims retaliate more when an audience has witnessed the provocation – especially if the audience communicates to the victim that he or she looks weak because of the harm suffered, or if the victim knows that the audience is aware that he or she has suffered particularly unjust treatment (B. R. Brown Reference Brown1968; Kim et al. Reference Kim, Smith and Brigham1998). Moreover, when two men have an argument on the street, the presence of a third person doubles the likelihood that the encounter will escalate from a verbal altercation to one that involves violence (Felson Reference Felson1982). Not all studies, however, find that the presence of observers increases victims' likelihood of punishing (e.g., Bolton & Zwick Reference Bolton and Zwick1995).

3.1.3. Direct deterrence of the withholding of benefits

The logic of revenge applies as much to changing others' incentives to deliver benefits to the self as to simply refraining from imposing costs on the self: From the standpoint of natural selection, there is no principled difference between the two, even if it turns out to be the case that cognitive systems that track the delivery and omissions of desired or anticipated benefits are distinct from systems that track the imposition of costs. Obtaining benefits and avoiding costs are functionally identical (although registering the omission of benefits requires certain computations that registering the imposition of harms might not; see sect. 5.1).

In a study described in section 3.1.1 above, Molm (Reference Molm1997, pp. 126–34) found that the frequency with which actors used punishment in response to a partner's previous withholding of rewards was positively associated with the rate at which partners rewarded each other over the series of trials. In other words, a willingness to punish one's interaction partner in response to his or her withholding of benefits increased the partner's delivery of benefits.

Public Goods Games also illustrate how revenge can deter the withholding of benefits. In these games, a few (e.g., 4–6) participants receive initial endowments of money that they each can divide between two pools. One pool is private; only the subject benefits from money kept in his or her own pool. The other pool is shared; money placed in this pool is multiplied by a number greater than 1 and the product is then divided evenly among all group members. Money maximizers keep everything in their private pools; aggregate group wealth is maximized when everyone contributes their entire earnings to the public pool. These games are social dilemmas (Kollock Reference Kollock1998; Ostrom Reference Ostrom1990) because they create a tension between individual and group outcomes and provide an assay of cooperation (for reviews, see Camerer Reference Camerer2003; Ledyard Reference Ledyard, Kagel and Roth1995). Several Public Goods Game experiments are particularly important to review here because they add an important dose of realism: The avenger must pay a cost for the opportunity to punish (Clutton-Brock & Parker Reference Clutton-Brock and Parker1995; Jensen Reference Jensen2010).

Yamagishi (Reference Yamagishi1986) had subjects play 12 rounds of a Public Goods Game in 4-person groups. He varied whether participants could punish other members of the group, and varied the cost to reduce another player's payoff by one unit. Players used the system for administering punishment when the opportunity to punish was made available to them. When it was available, players contributed greater amounts to the public good (see also, Carpenter & Matthews Reference Carpenter and Matthews2004; Ostrom et al. Reference Ostrom, Walker and Gardner1992; Yamagishi Reference Yamagishi1988). These results imply that punishment reduces the withholding of benefits in these games.

From this and subsequent work, we can draw some tentative conclusions. Punishment seems to be particularly effective in eliciting contributions when (a) the punishing technology makes the cost of punishment relatively low in comparison to the cost imposed on the individual being punished (Egas & Riedl Reference Egas and Riedl2008; Nikiforakis & Normann Reference Nikiforakis and Normann2008); (b) there are many repeated interactions (Egas & Riedl Reference Egas and Riedl2008; Gürerk et al. Reference Gürerk, Irlenbusch and Rockenbach2006; Walker & Halloran Reference Walker and Halloran2004); (c) people can communicate their intentions regarding investment levels and the use of punishment (Ostrom et al. Reference Ostrom, Walker and Gardner1992); and (d) people make choices about their preferred group partners on the basis of those prospective group members' contributions on previous rounds of play (Page et al. Reference Page, Putterman and Unel2005), or can migrate into or out of groups that have the capacity to punish (Gürerk et al. Reference Gürerk, Irlenbusch and Rockenbach2006).

When a participant in such experiments punishes a group member who has withheld contributions to the public good, such punishment behavior could be caused by revenge systems, but it could arguably result from other systems – for example, one designed to produce behaviors that induce the targeted individual to emit benefits for other people in the future, or one designed for the enforcement of social norms (Clavien & Klein Reference Clavien and Klein2010).

Fehr and Gächter (2002) ran an experiment in which players changed groups after every round so that punishment could not be used strategically to induce group members who were uncooperative in round r to cooperate with the punisher in round r + 1. Even with this methodological alteration, Fehr and Gächter (2002) obtained similar results as in Public Goods Game experiments in which participants played with the same partners in each game, or with individuals with whom they might be randomly paired in successive rounds (Fehr & Gächter Reference Fehr and Gächter2000). Participants punished uncooperative group members, and group members cooperated more when the option of punishing was available to the group (see also Anderson & Putterman Reference Anderson and Putterman2006). Fehr and Gächter (2002) interpreted these results as evidence for altruism rather than for revenge because punishers could not use punishment to help themselves directly: All players were regrouped after each round of play. We would argue, however, that a revenge interpretation of Fehr and Gächter's (2002) results is plausible (Clavien & Klein Reference Clavien and Klein2010; Kurzban & DeScioli, submitted). First, Fehr and Gächter (2002) describe subsidiary results based on participants' self-reported responses to hypothetical scenarios (see also Fehr & Fischbacher Reference Fehr and Fischbacher2004) to make a case that third parties' anger (and other group members' fear of those third parties' anger) shapes cooperation and punishment decisions in Public Goods and Third-Party Punishment games. However, these data are consistent with a revenge interpretation also because anger is a common response to personal exploitation (Fessler Reference Fessler, Potegal, Stemmler and Spielberger2010; Sell Reference Sell2011; Sell et al. Reference Sell, Tooby and Cosmides2009; Srivastava et al. Reference Srivastava, Espinoza and Fedorikhin2009). That is, we believe that anger is the motivational system that brings about revenge, though here it is not discharging its proper function (Sperber Reference Sperber, Hirschfeld and Gelman1994): Our view is that punishment in the Public Goods Game is caused by a proximate psychology designed for deterrence of personal harms in a world in which interactions were generally repeated with a relatively small number of interactants (e.g., Hagen & Hammerstein Reference Hagen and Hammerstein2006).

Moreover, other research traditions have associated empathy for victims, rather than anger toward perpetrators, as the proximate emotional cause of action whose goal is to deliver benefits to others (Batson Reference Batson2011). Indeed, some work from this latter tradition suggests that individuals do not naturally become angry upon observing the mistreatment of one stranger by another stranger unless empathy for the victim was experimentally manipulated beforehand (Batson et al. Reference Batson, Kennedy, Nord, Stocks, Fleming, Marzette, Lishner, Hayes, Kolchinsky and Zerger2007). Further, Shinada et al. (Reference Shinada, Yamagishi and Ohmura2004) found that (a) self-reported anger toward low contributors, and (b) judgments of the unfairness of low contributors' behavior are correlated with the extent to which one punishes low contributors within one's own groups, but they are not correlated with the extent to which one punishes low contributors in other people's groups. According to these latter experiments, anger evidently makes people punish individuals who have harmed them directly or who have disrupted cooperation in their groups, but in some circumstances, might not naturally motivate third-party punishment in response to witnessing someone receiving unfair treatment from a third party.

Fehr and Gächter (2002) construe their results as altruism (as opposed to revenge) because of the increase in contributions that punishment elicits from previously low contributors. They articulate the sense in which they mean that punishment is altruistic, writing that punishment “may well benefit the future group members of a punished subject, if that subject responds to the punishment by raising investments in the following periods. In this sense, punishment is altruistic” (p. 137, emphasis added). So, if punishment is being used in this sense, then punishment ought to decline as participants approach the end of the experiment, after which no one will engage in any more rounds of investment, and in one-shot games. This is contradicted by multiple studies (Anderson & Putterman Reference Anderson and Putterman2006, see Footnote 8; Carpenter & Matthews Reference Carpenter and Matthews2004; Fehr & Gächter 2002; Page et al. Reference Page, Putterman and Unel2005).

Our view doesn't commit to revenge instrumentally benefiting the vengeful individual in all instances. We take harm (or withholding of benefits) to be the eliciting factor, anger to be the proximate motivating system (Sell et al. Reference Sell, Tooby and Cosmides2009), and imposition of costs to be the behavioral output. So, although our view does commit to what will elicit revenge, the associated emotion of anger, and the behavioral output, we are not committed to the view that revenge will be absent when there is no chance of repeat interaction or that it will implement its proper function (e.g., Hagen & Hammerstein Reference Hagen and Hammerstein2006; West et al. Reference West, Mouden and Gardner2011). This argument could, of course, be applied equally if one were to argue that third-party punishment psychology (i.e., altruistic punishment) were also “misfiring” because it was designed for a world of repeat interactions (Anderson & Putterman Reference Anderson and Putterman2006, see Footnote 8; Carpenter & Matthews Reference Carpenter and Matthews2004; Fehr & Gächter 2002; Page et al. Reference Page, Putterman and Unel2005). Because of the way that payoffs are structured in Public Goods Games with punishment, revenge can generate benefits for others as a side effect (as in Fehr & Gächter 2002). The fact that some people benefit as a side effect of revenge should not necessarily cause one to infer that revenge systems are designed for altruism (Burnham & Johnson Reference Burnham and Johnson2005; Kurzban & DeScioli, submitted; Price et al. Reference Price, Cosmides and Tooby2002; West et al. Reference West, Mouden and Gardner2011).

Some studies do show that some third parties punish individuals who fail to provide benefits to others even when those third-party punishers themselves have not been harmed, which suggests caution regarding a revenge interpretation. For example, Carpenter and Matthews (Reference Carpenter and Matthews2012) ran a one-shot Public Goods Game and varied whether participants could punish members of their own groups (which could potentially influence the punishers' outcomes) or members of other people's groups (where the punishers' own welfare would be unaffected). In the key treatment, the “one-way TPP” (third-party punishment) condition (in which participants could punish individuals within their own groups and in other people's groups, but could not be punished by the individuals in other groups), they found that 90% of subjects did not punish outside their group (p. 12). The average amount participants used to punish third parties was roughly ten cents, and these same participants used approximately seven times as much money to punish low contributors within their own groups. Moreover, in a condition in which participants could only punish contributors within their own groups, the total expenditure for punishment was ten times as high as the expenditure for third-party punishment in the one-way TPP treatment.

Also, Carpenter and Matthews (Reference Carpenter and Matthews2009) found in a repeated game that second-party punishment and third-party punishment did not differ greatly in magnitude. Further, some evidence indicates that people do engage in third-party punishment in games other than Public Goods Games (Fehr & Fischbacher Reference Fehr and Fischbacher2004), though the expenditures for third-party punishment are, as in Carpenter and Matthews (Reference Carpenter and Matthews2012), lower than for second-party punishment. Nevertheless, in such games, the amounts of punishment that third parties mete out toward uncooperative or ungenerous individuals is not strong enough to deter individuals from withholding benefits from others, whereas second-party punishment is (Fehr & Fischbacher Reference Fehr and Fischbacher2004). In any case, third-party punishment systems might or might not be designed to incentivize the targets of such punishment to benefit others in the future (DeScioli & Kurzban Reference DeScioli and Kurzban2009b); for example, advertising one's willingness to engage in third-party punishment appears to produce reputational advantages (Barclay Reference Barclay2006; Nelissen Reference Nelissen2008), which might suggest an alternative function (Yamagishi et al. Reference Yamagishi, Horita, Takagishi, Shinada, Tanida and Cook2009).

3.2.1. Cross cultural universality in the operation of the revenge system

Cross-cultural research suggests that revenge is a universal response (Brown Reference Brown1991) to the imposition of costs – especially in the extreme case, the homicide of a kinsman. Ericksen and Horton (Reference Ericksen and Horton1992) found that 90% of the 186 societies in the Standard Cross-Cultural Sample (SCCS) showed clear evidence that blood feuds or individual self-redress were either actively used or formally outlawed in favor of formal adjudication procedures (the latter implying that revenge must have historically been a problem) “if a consanguineal kin group member is killed, injured, or insulted by a member of another kin group” (p. 60). For 18 of the 186 societies in the SCCS – most of the remaining 10% – data for making such a determination were either missing or conflicting. Relatedly, Daly and Wilson (Reference Daly and Wilson1988) concluded that 57 (95%) of the 60 societies in the Human Relations Area Files probability sample had “some reference to blood feud or capital punishment as an institutionalized practice, or specific accounts of particular cases, or at the least, the articulate expression of the desire for blood revenge” (p. 226, italics in original).

More recently, Boehm (Reference Boehm2008) reported results from an ethnographic survey of data for 10 “Pleistocene-appropriate” societies (i.e., economically independent pure hunter-gatherers that do not live in permanent year-round settlements). Boehm concluded that punishments of various types in response to prior harms were widespread, including public and private gossip about a violator (100% each); and physical punishment (90%).

A final line of cross-cultural evidence supporting the universality of revenge comes from Henrich et al.'s (2006) study of the Ultimatum Game in 15 small-scale societies (including two groups from North America, three groups from South America, six groups from Africa, a group from Asia, and three groups from Oceania) that differed in language, climate, and economic base. In all 15 societies, as proposers' offers tilted away from a 50/50 split in the proposers' favor, the recipients in all 15 societies became more likely to reject those offers. As Henrich et al. (Reference Henrich, McElreath, Barr, Ensminger, Barrett, Bolyanatz, Cardenas, Gurven, Gwako, Henrich, Lesorogol, Marlowe, Tracer and Ziker2006) wrote, “In every population, less-equal offers were punished more frequently” (p. 1770). Taken together, the results from Ericksen and Horton (Reference Ericksen and Horton1992), Daly and Wilson (Reference Daly and Wilson1988), Boehm (Reference Boehm2008), and Henrich et al. (Reference Henrich, McElreath, Barr, Ensminger, Barrett, Bolyanatz, Cardenas, Gurven, Gwako, Henrich, Lesorogol, Marlowe, Tracer and Ziker2006) suggest that revenge is widespread cross-culturally, if not indeed universal (Brown Reference Brown1991).

3.2.2. Cross-cultural variation in the operation of the revenge system

Nevertheless, there is substantial cross-cultural and temporal variation in the overall amounts and forms of revenge. Across societies, the percentages of homicides attributable to revenge (for instance) range from as low as 8% to as high as 45% (e.g., Cardona et al. Reference Cardona, Garcia, Giraldo, Lopez, Suarez, Corcho, Posada and Florez2005; Daly & Wilson Reference Daly and Wilson1988; Gaylord & Galligher Reference Gaylord and Galligher1994; Kubrin & Weitzer Reference Kubrin and Weitzer2003). If revenge is for deterrence, it might be more prevalent in cultural ecologies in which there are few or no institutions (police, courts) that deter interpersonal aggression and other forms of harm. (We note that we take the question of why and how these institutions emerge to be a separate issue.)

Ericksen and Horton (Reference Ericksen and Horton1992) coded 186 societies in the Standard Cross-Cultural Sample to determine which of three approaches was favored for settling grievances when a member of one's own kin group is killed, injured, or insulted by a member of another group: (a) kin group feuding (i.e., classic blood feuding), (b) individual self-redress (individuals avenging their grievances on their own), and (c) formal adjudication. They found that individual revenge was the primary means of redress in traditional foraging societies. When people organize into tribes – especially tribal societies that emphasize manly honor (e.g., Boehm Reference Boehm1987) – kin groups take up relatives' grievances. Second, kin feuding evidently gives way to formal adjudication as external political forces implement the rule of law. Third, revenge by individuals or extended kin groups is replaced by formal adjudication in societies with favorable resource bases and vertical inheritance – that is, societies in which individuals would stand to benefit from the social stability that comes from replacing revenge with other means of sanctioning. In brief, then, these findings suggest that when institutions arise to administer third-party punishment, individual acts of revenge are crowded out. (One could equally conceptualize this as institutions raising the price of revenge, thereby reducing the demand for it.)

Anderson (Reference Anderson1999) has suggested that concentrated neighborhood disadvantage (e.g., low socioeconomic status [SES] and median family income) precipitates a widespread lack of trust in formal legal methods for settling differences, and instead encourages the adoption of a “code of the street” that prescribes the use of personal revenge, rather than appeals to law enforcement authorities for settling one's interpersonal grievances (Brezina et al. Reference Brezina, Agnew, Cullen and Wright2004). Research on contemporary geographic distributions of revenge homicides supports this view. Kubrin and Weitzer (Reference Kubrin and Weitzer2003) found that concentrated neighborhood disadvantage was an excellent predictor of retaliatory homicide (i.e., homicides motivated by a desire for revenge in response to a previous perceived slight or injury): Indeed, the census tracts in St. Louis, MO, with high levels of poverty, high unemployment rates, and high percentages of children not living with both parents also had the highest rates of retaliatory homicide. Conversely, in census tracts with very low concentrated neighborhood disadvantage, retaliatory homicide was negligible. (These data leave open the possibility that violence, generally, is more common in such environments, as opposed to revenge, narrowly, being more common.) This insight is echoed in many other ethnographic studies that find that revenge is more frequently used in societies in which social institutions for settling grievances are generally viewed as weak, and in which individuals are socialized to defend their honor with retaliatory violence at even the most trivial of interpersonal slights (Black-Michaud Reference Black-Michaud1975; Boehm Reference Boehm1987; Nisbett & Cohen Reference Nisbett and Cohen1996).

3.3.1. Genetic sources of individual differences

Efforts to isolate genetic markers associated with individual differences in revenge have begun (McDermott et al. Reference McDermott, Tingley, Cowden, Frazzetto and Johnson2009), although the source of variation (e.g., random, unselected noise, frequency dependent selection for individual differences in vengefulness; environmental heterogeneity in fitness optima; see Buss Reference Buss2009) remains unknown. One possibility is that these genetic individual differences can be thought of as contingent shifts in social strategy as a function of other heritable phenotypic characteristics – a phenomenon that Tooby and Cosmides (Reference Tooby and Cosmides1990a) called reactive heritability.

Suppose the costs of revenge are higher, on average, for people with low upper body strength or body size – traits that are highly heritable (Carmichael & McGue Reference Carmichael and McGue1995; Silventoinen et al. Reference Silventoinen, Magnusson, Tynelius, Kaprio and Rasmussen2008) – because the vengeful efforts of weak individuals are more likely to be answered with counter-revenge (Sell et al. Reference Sell, Tooby and Cosmides2009). The same logic explains why men with greater body weight and height (both of which are indicative of physical formidability) are more likely to be diagnosed with antisocial personality disorder (Ishikawa et al. Reference Ishikawa, Raine, Lencz, Bihrle and LaCasse2001), why large football players are perceived as more aggressive, less friendly, and less cooperative than are smaller players (Koenig & Ketelaar Reference Koenig and Ketelaar2006), and why men's physical strength is positively associated with their aggressiveness, anger-proneness, their histories of fighting and success in conflict, and their beliefs about the utility of personal aggression (Gallup et al. Reference Gallup, White and Gallup2007; Sell et al. Reference Sell, Tooby and Cosmides2009). If this supposition is correct, then statistically equating men on physical strength should substantially reduce the contribution of additive genetic factors to individual differences in the propensity for revenge.

The same logic builds a causal bridge between the fact that 99.9% of women have less upper body strength than does the average man (Lassek & Gaulin Reference Lassek and Gaulin2009) and the facts that (a) women are nearly a standard deviation less vengeful than men are (Archer Reference Archer2009; Miller et al. Reference Miller, Worthington and McDaniel2008); (b) women are half a standard deviation less physically aggressive than men are, despite being no less anger-prone than men are (Archer Reference Archer2004); and (c) sex differences in provoked aggression are strongest in experimental situations in which females are at greater risk than males of becoming the targets of counter-aggression (Bettencourt & Miller Reference Bettencourt and Miller1996). For women, the costs of revenge may best be lowered not through taking revenge via their own physical strength, but through either (a) indirect aggression such as reputational damage (Hess & Hagen Reference Hess and Hagen2006), or, as Sell et al. (Reference Sell, Tooby and Cosmides2009) suggest, (b) recruiting coalitional support from others who can effect revenge on their behalf and (with their physical strength) deter counter-retaliation.

Phenotypic factors associated with women's success in recruiting male coalitional support might include factors such as physical attractiveness or waist-hip ratio that relate to mate value – some of which are highly heritable in women (Olson et al. Reference Olson, Vernon, Harris and Jang2001; Zillikens et al. Reference Zillikens, Yazdanpanah, Pardo, Rivadeneira, Aulchenko, Oostra, Uitterlinden, Pols and van Duijn2008). Indeed, Sell et al.'s (2009) findings that women's physical attractiveness, but not physical strength, predicts their anger-proneness, beliefs about the utility of personal aggression, sense of entitlement, and self-reported history of success in resolving interpersonal conflicts in their own favor suggests that women's ability to leverage male coalitional support lowers the costs of revenge for women. If this explanation is correct, then heritable individual differences in women's vengefulness should shrink after statistically controlling for measures of mate value.

3.3.2. Environmental sources of individual differences

As with other social strategies, the propensity for revenge might vary across individuals because of the costs and benefits of using the strategy (Buss Reference Buss2009). Cultural influences such as concentrated neighborhood disadvantage (see section 3.2.2 above), civic trust, policing efficiency, and even parental support for retaliation as a way of handling grievances (Black-Michaud Reference Black-Michaud1975; Brezina et al. Reference Brezina, Agnew, Cullen and Wright2004; Copeland-Linder et al. Reference Copeland-Linder, Jones, Haynie, Simons-Morton, Wright and Cheng2007; Ericksen & Horton Reference Ericksen and Horton1992; Herrmann et al. Reference Herrmann, Thöni and Gächter2008; Kubrin & Weitzer Reference Kubrin and Weitzer2003; Nisbett & Cohen Reference Nisbett and Cohen1996; Solomon et al. Reference Solomon, Bradshaw, Wright and Cheng2008) are good candidates for explaining variance in revenge. Variables such as endorsement of a street code of conduct (Brezina et al. Reference Brezina, Agnew, Cullen and Wright2004; Stewart et al. Reference Stewart, Schreck and Simons2006), and (mis)trust in the police (Kääriäinen Reference Kääriäinen2007) may also be useful for capturing important aspects of the proximate psychology through which such environmental effects influence the propensity for revenge.

Finally, there is substantial individual variance in revenge that is attributable to non-shared environmental factors – that is, factors that monozygotic twins do not share in common. To the extent that revenge produces deterrence or other beneficial effects such as an improved reputation, increased social status, or increased attractiveness to prospective mates (Anderson Reference Anderson1999; Boehm Reference Boehm1987), it necessarily produces these effects with respect to specific bullies, despots, friends, peers, and potential mates, and these unique effects will calibrate individuals' propensities to engage in revenge in unique ways. Non-shared environmental effects have been notoriously difficult to identify in behavioral-genetic research (Turkheimer & Waldron Reference Turkheimer and Waldron2000) because they are virtually infinite in number and generally small in magnitude. Nevertheless, as these unique social experiences accumulate over the life course, they will make even identical twins increasingly different from each other (Harris Reference Harris2006). The cumulative effect of these unique social experiences will be to alter people's computations of revenge's costs and benefits, thereby yielding different propensities for revenge over the life course.

4.4.1. Acceptance

First, after an individual has imposed a cost upon the self, one might simply tolerate, or refrain from responding to, the harm (Gardner & West Reference Gardner and West2004) – essentially accepting the aggressor's WTRs for the victim as implied by the cost-imposing action. In such cases, victims simply absorb the costs they have incurred at the hands of the aggressor, avoid the costs associated with enacting revenge, and continue interacting with the aggressor without attempting to modify the aggressor's WTR toward the self. In such cases, social interaction between the aggressor and the victim continues, possibly with the victim lowering his or her estimate of the aggressor's WTR toward him or her. Such a course of action might occur when the costs of the harm are lower than the expected costs associated with attempting to up-regulate the aggressor's WTR toward the self. Specifically, we predict that individuals tend to ignore harms whose costs to the self (discounted by the benefits to be obtained from deterring future similar harms by the aggressor or third parties who might be deterred indirectly) are lower than the costs associated with attempting to adjust the aggressor's WTR toward the self. Acceptance may be signaled using language or by appeasement gestures that communicate one's willingness to accept certain costs imposed by the aggressor and an absence of any residual motivation to engage in retaliatory aggression.

4.4.2. Revenge

Second, one might take revenge – that is, attempt to impose a cost upon the aggressor with the goal of altering the aggressor's monitored WTR toward the self and to obtain the benefits of deterrence more generally. As a result of revenge, social relations might end (e.g., social relations might be completely terminated, or revenge might incapacitate or kill the aggressor), or counter-revenge might ensue. Alternatively, relations between aggressor and victim might be restored under renegotiated and mutually tolerable WTRs. We expect revenge when the benefits of deterring future harms and adjusting the aggressor's WTR toward the self outweigh the costs associated with imposing a retaliatory response.

4.4.3. Avoidance

Third, one might reduce or terminate one's interactions with the aggressor – that is, render it more difficult for the aggressor to impose costs upon, or obtain benefits from, the victim. Avoidance reduces the likelihood that the aggressor will be in a position to impose costs upon the victim again in the future. Avoidance might be more likely to be chosen when (1) the likely effectiveness of revenge is low (for any of the reasons we have already described) and (2) the estimate of the residual value in the relationship is low. When avoidance evolves for its efficacy in deterring harmdoers by depriving them of benefits they could have acquired through cooperative interaction with the individual who conditionally avoids harmdoers – rather than solely for its self-protective effects – then avoidance is better classified as an exit-based form of revenge (Barnes et al. Reference Barnes, Brown and Osterman2009; Cant & Johnstone Reference Cant and Johnstone2006).

4.4.4. Forgiveness

Fourth, the victim might forgive – that is, attempt to raise the aggressor's WTR (particularly the intrinsic WTR) toward the victim without imposing costs or withholding benefits from the aggressor in a retaliatory fashion. Putative forgiveness systems coordinate several tasks: They inhibit (i.e., down-regulate the activity of) systems that motivate revenge and avoidance, and they motivate “reparative behaviors” (Petersen et al. Reference Petersen, Sell, Tooby, Cosmides and Høgh-Olesen2010) which signal that (a) the aggressor's behavior damaged the victim, and that (b) despite that harm, there is the possibility of future gains from interaction if the aggressor is willing to refrain from similar aggressive actions in the future (i.e., increases his or her WTR toward the victim). To cause aggressors to recalibrate their WTRs toward their victims, forgivers might (among other things) attempt to remind aggressors of previous benefits that they have provided to the aggressors (Petersen et al. Reference Petersen, Sell, Tooby, Cosmides and Høgh-Olesen2010). Forgiveness should be more likely as the value of the relationship to the transgressor goes up and as the value of any deterrence to be obtained through acts of revenge or avoidance goes down (e.g., when a transgressor has indicated a disinclination to impose similar of costs on the victim in the future; see sect. 5.3 further on).