The evolution of extreme cooperation via shared dysphoric experiences

Willingness to lay down one’s life for a group of non-kin, well documented historically and ethnographically, represents an evolutionary puzzle. Building on research in social psychology, we develop a mathematical model showing how conditioning cooperation on previous shared experience can allow individually costly pro-group behavior to evolve. The model generates a series of predictions that we then test empirically in a range of special sample populations (including military veterans, college fraternity/sorority members, football fans, martial arts practitioners, and twins). Our empirical results show that sharing painful experiences produces “identity fusion” – a visceral sense of oneness – which in turn can motivate self-sacrifice, including willingness to fight and die for the group. Practically, our account of how shared dysphoric experiences produce identity fusion helps us better understand such pressing social issues as suicide terrorism, holy wars, sectarian violence, gang-related violence, and other forms of intergroup conflict.


Results
Mathematical models and theoretical predictions. Our models included many of the standard assumptions of theoretical approaches to within-group cooperation in evolutionary biology. We treated individual willingness to cooperate with group-mates as a genetically controlled trait 1,2 . Individual fitness was determined by an outcome of a collective goods game 35 that the group members participate in. Collective action of group members can be thwarted by free-riding 36 ; this problem can be solved to some extent by kin selection, reciprocity, punishment, or group selection [37][38][39] . Here we offer a novel solution -conditioning cooperation on shared prior experience. In our model, some groups facing a collective action previously had fitness enhancing experiences while others had fitness decreasing experiences. Below we show that conditioning individual efforts in a collective action on these qualities of previously shared experience can evolve by natural selection and can help to solve the free-riding problem. Our model predicts that groups undergoing fitness-decreasing experiences are more likely to contribute substantially to future collective actions. That is, shared past negative experiences can augment future pro-group behavior increasing the overall fitness of both the group and its individual members. Our results however predict a particular evolved social psychology that biases humans to greatly increase cooperation if their groups go through shared negative experiences.
Models. More specifically, we considered a population of individuals living in a large number G of groups of constant size n. Generations are discrete and non-overlapping. We focused on a single collective action 35,40,41 that groups attempt to accomplish. The effort of individual i in group j towards the group's success in the collective action was modeled as a nonnegative continuous variable z ij ; the total group j effort of is = ∑ Z z j i ij . We defined the individual payoff from the collective action as where b and c are constant benefit and cost parameters. Function P j = P j (Z j ) gives the normalized value of the resource produced by group j as a result of collective effort; we normalize P j relative to a maximum possible reward size (0 ≤ P j ≤ 1). Relative individual fertility was proportional to f f / ij j , where f j is the average payoff in group j.
There are two general types of collective actions in which our ancestors were almost certainly engaged. The first includes group activities such as defense from predators, some types of hunting or food collection, use of fire, etc. The success of a particular group in these activities largely does not depend on the actions of neighboring groups. We will refer to such collective actions as "us vs. nature" contests and define the relative success as Here Z 0 is a "half-saturation" constant; the larger Z 0 , the more group effort Z j is needed for the success. The second type of collective action, which we will refer to as "us vs. them" contests, includes direct conflicts and/ or competition with other groups over territory and other resources such as mating. The success of one group in an "us vs. them" contest means failure or reduced success for other groups. In these contests, we defined the relative success as where the sum is over all groups 43,44 . We studied "us vs. nature" and "us vs. them" contests separately. Our formulation implied there was an incentive to free-ride on the efforts of group-mates 36,35 . The collective action models introduced above belong to a general class of the Volunteer dilemmas 45,46 , where individuals would prefer to free-ride on the effort of their group-mates but if nobody else is willing to do it, it may become advantageous to volunteer in spite of the costs involved. It is important to realize that in our models, individuals make contributions to a collective good not because they are "altruistic" but because this increases their fitness. We extended the above standard model by assuming that groups differ in previous experience which both contributes to the overall probability of the group's survival and can potentially condition individual cooperation. Specifically, we posited that a random proportion π of the groups previously had an euphoric (that is, payoff-increasing) experience whereas a proportion 1 − π had a dysphoric (that is, payoff-decreasing) experience. For example, one can think of an "endowment" (e.g. a forest with fruits, or pigs) that the group has initially but may lose because of some random exogenous effects. The loss of the endowment would then represent a dysphoric event experienced by all group members. The previous experience (specified by an indicator variable E j = 1 for euphoric groups and to 0 for dysphoric groups) and the relative success P j in the focal collective action jointly controlled the group's survival probability which is set to be proportional to Here 0 ≤ h ≤ 1 is a constant parameter measuring the importance of the previous experience for the group's survival. The equation above uses a simple linear function to specify how previous experience (shared by all group members by assumption) affects group survival. In our model, given everything else the same, the probability of group survival S in "euphoric" groups is larger by h than that in "dysphoric" groups (S euphoric = h + (1 − h)P, S disphoric = (1 − h)P)). Groups that did not survive were replaced by the offspring of surviving groups (see Methods and the Supplementary Information, SI). In our model, group-level selection favors large individual efforts z ij (which would increase the probability of group survival S j ), while individual-level selection may favor low efforts z ij (which would reduce the individual costs term cz ij ) 47,48 . We assumed that previous experience not only controlled the group's survival probability, but could also potentially influence individual efforts in the collective action, by triggering different behavioral modules. Specifically we postulated two independent (unlinked) loci with allelic effects x ij and y ij , the first of which was expressed in individuals with euphoric experience, so that in such individuals z ij = x ij , and the second was expressed in individuals with dysphoric experience, so that in such individuals z ij = y ij . In each individual, only one gene was expressed, and all individuals from the same group expressed the same gene. Initially, the allelic effects in both genes in all individuals were very close to zero and the individuals did not contribute much to the collective action. We allowed for mutation, recombination, and migration. We were interested in whether gene effects would increase over time and whether the increase would be similar or asymmetric between the two genes, that is, whether individuals would condition their cooperation on shared past experience.
Our model operated on an evolutionary time-scale focusing on genetic changes leading to the evolution of pro-social behavior conditioned on shared past experience. Individuals' feelings underlying the development of identity fusion during the individual's life-time were not modeled explicitly. However, to the extent that identity fusion mediates the relationship between shared past experience and future pro-group actions of an individual (as we show below experimentally), our results also concern the effects of shared experience on identity fusion.

Results and predictions.
To study our models we used both analytical approximations (shown in the SI) and individual-based simulations. To remove the effects of genetic relatedness, groups were formed randomly at the beginning of each generation. We measured the averages of euphoric x and dysphoric y efforts across the whole system at a (stochastic) equilibrium state to which the system evolves. Figure 2 illustrates the effects of five different parameters on the contribution to collective action in the two contests in euphoric and dysphoric groups (see also the SI). Overall, our results lead to the following predictions. First, previous shared experience does affect individual behavior in collective action (x and y depend on parameters π and h specifying past experience). Second, dysphoric experience makes individuals contribute more than euphoric experience ( > y x). Third, more intense experience results in stronger effects on prosociality (ratio y x / increases with h). The intuition behind these results is very simple: groups in poor initial conditions (e.g. with a reduced endowment or fitness) really need to cooperate in order to make it into the future while those in better initial conditions may "afford" somewhat reduced efforts. Fourth, the effect of shared dysphoria on prosocial behavior is much stronger if groups compete directly against other groups ("us vs. them" contests) than if they cooperate against nature ("us vs. nature" contests). Moreover, the effect is stronger in smaller groups (decreasing n increases x and y). The last two predictions are in line with earlier comparisons of "us vs. them" and 'us vs. nature" games 41,42 . The explanations are that "us vs. them" games impose stronger selection on the underlying genes than "us vs. nature" games 41,42 and that free-riding is a more effective strategy in larger populations 36 . We expect that the effects of the above factors on identity fusion will parallel those on individual actions captured explicitly by our model.
Our models were designed to study the effects of previous experience in the absence of genetic relatedness between group members. We can contrast our results with those for the case when group members are genetically related but the effects of previous experience are absent. The corresponding "us vs. nature" and "us vs. them" contests were studied previously 41,42 . With biologically realistic small values of average genetic relatedness 28 , the values predicted by these results can be significantly smaller than those observed in Fig. 2 (see SI). Of course, we do not know realistic values of some important parameters which control the model's predictions. Nevertheless our results suggest that effects of shared dysphoric experience on willingness to perform individually-costly pro-group acts can potentially be stronger than those of genetic relatedness.
In the models studied above, each individual values the group's success equally which implies equal degree of identity fusion. In the SI we use results from refs 40,41 to show that highly-fused individuals will exhibit more pro-group sentiments than low-fused individuals. In particular, under conditions of strong between-group competition the model predicts that the efforts of highly-fused individuals will be so high that their fitness will be almost zero. That is, highly-fused individuals are predicted to effectively sacrifice themselves for their groups. Reference 15 provides complementary experimental evidence on the willingness of such individuals to self-sacrifice for their groups.
Empirical tests. Our models make general predictions concerning cooperation in collective actions. Next, we test five specific predictions focusing on a particularly interesting and extreme type of cooperation -willingness to self-sacrifice for the group. Because we are interested in general behavioral predispositions, we chose a diverse set of samples in eight studies totaling 2,836 individual participants, including citizens of countries, fans of football teams, military veterans, college fraternity/sorority members, martial arts practitioners, and both monozygotic and dyzygotic twins. We ran a total of eight studies, employing correlational (Studies 1, 2, 4, 5, 6), quasi-experimental (Studies 1, 3, 8), and experimental (Study 8) methodologies.
Hypothesis 1: Shared experience promotes willingness to preform extreme pro-group action. We ran two studies to test this hypothesis 15 . Both studies distinguish everyday experiences from self-defining experiences (i.e., those that are vividly remembered and are central to one's self-concept) 49 . In Study 1, American participants were more willing to cooperate (e.g. donate money, volunteer) to solve problems associated with either a natural disaster (N = 97) or a terrorist attack (N = 98) in the United States when they reported sharing more self-defining (r = 0.239, P = 0.001) and everyday experiences (r = 0.187, P = 0.009) with fellow Americans. In Study 2, Americans (N = 122) were asked about their willingness to endorse extreme, self-sacrificial pro-group actions. We also measured participants' levels of identity fusion with their country. Shared experiences increased willingness to endorse extreme pro-group behaviors via increasing identity fusion. This held for both self-defining experiences, = .
. of an ingroup member in the classic trolley dilemma (r = 0.120, P = 0.001) than fans of winning (i.e., euphoria-producing) teams. The effects of team support on self-sacrificial responses and pro-group moral endorsements were both mediated by identity fusion, = . = .
.  is that shared dysphoria motivates extreme cooperation via identity fusion, as high fused individuals of both unsuccessful and successful football teams were found to endorse self-sacrificial behaviour. In earlier studies of group competition involving monetary donations rather than self-reported endorsement of prosocial acts, losing groups increased their contributions while winning groups decreased it 51,52 .

Hypothesis 3: More intense experiences of shared dysphoria produce stronger effects on self-sacrifice for the group.
We ran three studies to test this hypothesis. Military veterans vary widely in exposure to shared dysphoric events 53,54 , thus we surveyed U.S. combat veterans of the Vietnam War (N = 380) in Study 4. As predicted, greater exposure to shared dysphoric combat experiences (e.g. losing a close co-combatant in battle) predicted both identity fusion (r = 0.203, P < 0.0001) and willingness to provide support for veterans in need (r = 0.184, P < 0.0001). Combat experiences increased willingness to provide support to fellow veterans via increasing levels of identity fusion, = .
. . b SE CI 0 1026( 00290), 95% [0 0521, 0 1668]. In Study 5, past and current members (N = 146) of U.S. college fraternities and sororities who had undergone hazing and other such initiation rituals, were asked about the extent to which the initiation ritual was self-defining 24 . Perceived self-definingness of the experience predicted both identity fusion (r = 0.430, P < 0.0001) and expressed willingness to sacrifice self for group (r = 0.429, P < 0.0001). Self-definingness increased pro-group sacrifice by increasing identity fusion, = .
Similarly, in Study 6, we used online advertisements to recruit Brazilian Jiu Jitsu (BJJ) practitioners (N = 564), as BJJ promotion events can involve either a painful belt-whipping gauntlet run or less severe practices. This provided an opportunity to compare a population with a significant degree of variation in the dysphoric arousal of important affiliative events, which practitioners are typically unaware of before joining (62.5% reported having "no idea" about their school's graduation rituals before joining, while a further 16.1% had only "a vague idea"). Despite the significant heterogeneity involved in a worldwide sample, we found that the intensity of belt promotions predicted levels of identity fusion (ρ = 0.135, P = 0.002), and that identity fusion predicted participants' stated willingness to risk their lives fighting for the club (ρ = 0.542, P < 0.0001), as well as their willingness to donate time (ρ = 0.508, P < 0.0001) and make costly donations of potential prize money (ρ = 0.250, P < 0.0001) to the club. These relationships remained when controlling for other relevant factors, including age, sex, years training, group identification, and average time training per week. Mediation analyses also showed that elevated intensity of experiences increased participants' willingness to endorse pro-group behaviors via increasing levels of identity fusion.

Hypothesis 4: The effect of shared dysphoria on prosocial behavior is stronger where groups compete directly against
other groups, rather than if they cooperate against nature. Study 1 (see above) was designed to test hypothesis 4 as well as hypothesis 1. In Study 1, we found dysphoric contexts involving terrorists elicited more cooperation than those involving natural disasters, t(193) = 2.534, P = 0.012, Cohen's d = 0.363.

Hypothesis 5:
The effects of shared dysphoric experience on the willingness to perform pro-group acts can be stronger than those of genetic relatedness. We ran two studies to test this hypothesis. In Study 7,198 participants either wrote about an experience that has shaped them (Experience), genetically transmitted traits (Genes), or the changing seasons (Control). Participants then imagined interacting with someone who shared the same experience, discovered a long lost sibling, or met a stranger, respectively. Both .332]) independently predicted identity fusion. Furthermore, hierarchical regression analyses showed that shared experience continued to predict identity fusion even after controlling for shared genes.

Discussion
Overall our theoretical and empirical studies both suggest that shared dysphoric experiences are a powerful mechanism for promoting pro-group behaviors which under certain conditions can be extremely costly to the individuals concerned. Our ancestors had a common stake in their group's fate, especially when facing existential threats. Under threatening conditions, having a shared evolutionary future likely was a more decisive factor in cooperation and self-sacrifice than shared ancestry (i.e., genetic relatedness). A pervasive source of these threats was highly variable environmental conditions during the Late Pleistocene 56,57 making adaptation and survival difficult. Another potential source was competition with other human groups for resources and mating opportunities [58][59][60] .
Our model captures explicitly how individual efforts in public good games depend on previous group-shared experience. In the model, shared experience has two effects prominent in evolutionary biology and game theory: one on the group survival and another on gene expression. We did not model individual emotions and the sense of identity fusion explicitly. (This would not be possible.) That is, in the triad experience → identity fusion → action, we model explicitly only experience and action. However our experiments as well as earlier work show that identity fusion comes as a "proximate" mediator in the experience → action relationship. Therefore, our results also concern the effects of shared experience on identity fusion.
Previous theoretical research in evolutionary biology has identified a number of mechanisms for the evolution of cooperation [37][38][39] . Our work brings to light an additional mechanism -conditioning cooperation on shared prior experience. In our models, individuals acquire social instincts to contribute to collective actions because this increases their fitness over evolutionary time. However evolved social instincts may comprise relatively open behaviour programmes that are sensitive to cues such as shared dysphoria, leading to high levels of identity fusion and self-sacrificial acts.
Our proximate explanation for self-sacrifice is that dysphoric experiences and the knowledge that they are shared with the group 61 shape personal identity and the perception that one's personal identity is irrevocably tied to the group. The resulting state of identity fusion enables simultaneous activation of group and personal identity. In this light, threats to the group are experienced as threats to self and the drive to defend the group is consequently a form of self-defense 24 . Our empirical findings across study groups suggest a consistently robust trend for dysphoria's role in extreme cooperation, beyond the effects of group performance or kinship on cooperation that have previously been documented.
There has been recent interest in theoretical literature in the effects of variable environment on the evolution of cooperation [62][63][64] with some studies arguing that populations evolving under harsh environments would become more cooperative. Our models are very different in that we consider individual efforts as conditioned on previous group experience. Nevertheless there are some parallels in conclusions: we predict that experiencing an instance of a harsh environment would trigger more cooperative behavior.
Our modeling results naturally have a number of limitations. For example, to isolate the effects of previous experience, we purposely neglected genetic relatedness by randomly forming groups each generation. To simplify analysis, we assumed a simple genetic mechanism underlying instinctive behavior in collective actions while neglecting cultural effects (and transfer of experience between generations). Studying interactions between identity fusion, genetic relatedness, and cultural transmission of behaviors will be an important next step.
Our eight experimental studies provide preliminary empirical evidence for our model, as do other previous studies on the causes and consequences of identity fusion 20,22,65 . Our empirical findings across study groups suggest a consistently robust trend for dysphoria's role in extreme cooperation, beyond the effects of group performance or kinship on cooperation that have previously been documented. However, more experimental and longitudinal research is required to substantiate the causal claims made by the model. It is also necessary to develop experiments directly contrasting the hypotheses advanced here with alternative explanations. Furthermore, our studies have all relied on self-report measures. This is in part because the behavioral variables in which we are primarily interested -costly self-sacrificial behaviors -are difficult to measure directly. The use of more benign and commonplace behavioral measures (e.g., economic games) do not approximate our interests, and are therefore poor proxies. Nevertheless, measuring extreme sacrifice directly is impractical and unethical for obvious reasons. Instead we adopt a variety of plausible proxies for extreme sacrifice including identity fusion which has been shown repeatedly to motivate endorsement of extreme sacrifice (e.g. using trolley problems) as well as actual self-sacrifice in real-world correlational studies (e.g. among insurgent groups in Libya 20 ). Further field-based experiments, in which we can set up realistic scenarios for costly sacrifice, are required. Finally, we have not tested all the predictions of the model. For example, an intuitive prediction of our model, which we have not directly tested here, is that identity fusion will be stronger in small groups than in large groups. This is consistent with the observation that soldiers are more willing to die for each other (their unit comrades) than for abstract group categories or values (e.g., God and country) 66 .
Our models are meant to capture conditions faced by our ancestors tens of thousands years ago. As such they are not directly applicable to modern groups which have much larger sizes and experience different selection regimes. However our argument (which is standard in evolutionary psychology 67 ) is that certain "social instincts" in humans that evolved under ancestral conditions can still be expressed under certain conditions (cf. with "spontaneous altruism" observed in experiments where subjects are forced to make decisions quickly 68 ).
Understanding the causes of self-sacrifice for a group is a high priority not only for the evolutionary and psychological sciences but also for society at large. The spirit of self-sacrifice for the group has been a driving force of many historical events 69,70 . Many of the world's ongoing violent conflicts are fuelled by extreme commitment to groups. Nevertheless, people show variation in the extent of fusion with their groups 15 . This heterogeneity could be caused by differences in life history, cultural environment, or developmental factors. Certain groups have high levels of identity fusion, and certain events and/or experiences can cause higher identity fusion that can be exploited to mobilize extreme pro-group behaviors. Understanding altruistic and cooperative behavior by individuals and groups is notoriously difficult as there are multiple forces and factors underlying it, including kinship, reciprocity, punishment, mutualism, and various cultural beliefs and biases. However if we are to address such pressing social issues as suicide terrorism, holy wars, gangland violence, and other forms of intergroup conflict, we should take into account psychological predispositions conditioning extreme cooperation on shared past experiences.

Methods
Numerical simulations. We treated individuals as sexual haploid. To implement selection, we used the two-level Fisher-Wright framework 40,42,71 . Specifically, group selection is captured by making each group in the new generation to independently descend from a group in the previous generation with probability proportional to S j . Individual selection within each group is implemented by first independently choosing 2n parents from the group members with probabilities proportional to payoffs f ij and then producing n offspring assuming free recombination. [Results with completely linked genes are qualitatively similar]. Offspring production was followed by random mutation and then by random dispersal of nG offspring among G groups.
Scientific RepoRts | 7:44292 | DOI: 10.1038/srep44292 In numerical simulations we considered all possible combinations of the following parameters: benefit of collective action b = 0.5, 1.0, 2.0; cost of collective action c = 0.5, 1.0, 2.0; group size n = 4, 8, 12 (refs 52,53); relative importance of previous experience h = 0.2, 0.5, 0.8, and the proportion of groups with dysphoric experience π = 0.2, 0.5, 0.8. Parameters that did not change are: number of groups G = 1000, mutation rate μ = 0.0001, and the standard deviation of mutational effects σ = 0.5. To simplify the comparison of the two games we set the half-saturation parameter Z 0 = 1 and made the total contested benefit in "us vs. them" games equal to bG, so that that the expected benefit per group is b as in "us vs. nature" games. We ran simulations for 20,000 generations 10 times for each combination of parameters (see the SI for more details).

Experiments.
Individual study methodologies, including scale items, as well as individual study data analyses are detailed in the SI. All studies involving human participants were conducted in accordance with APA guidelines and regulations for conducting psychological research. In addition, methods and experimental protocols were approved by the University of Oxford's Central University Research Ethics Committee, the Murcia University Ethical Committee, or UNSW Human Research Ethics Advisory Panel C. Informed consent was obtained from all participants.