Abstract
Recent theoretical and experimental investigations of altruistic behavior in intergroup conflict in humans frequently make use of the assumption that warfare can be modeled as a symmetrical n-person prisoner's dilemma, abstracting away the strategic differences between attack and defense. In contrast, some empirical studies on intergroup conflict in hunter-gatherer societies and chimpanzees indicate that fitness relevant risks and potential benefits of attacks and defenses might have differed substantially under ancestral conditions. Drawing on these studies, it is hypothesized that the success of defenses was much more important for individual and kin survival and that a disposition to act altruistically during intergroup conflict is thus more likely to evolve for the strategic situation of defense. It is then investigated empirically if such asymmetries in the occurrence of altruistic behavior during intergroup conflict can be found. Analyzing detailed historical case data from 20th century wars, this study finds that altruistic behavior towards members of the in-group indeed seems to occur more frequently when soldiers are defending themselves and their comrades against enemy attacks. It is proposed that this asymmetry reflects adaptive behavioral responses to the materially different strategic character of attacks and defenses under ancestral conditions. If true, this would call for a refinement of theories of the evolutionary interaction of intergroup conflict and altruism.
Introduction
Violent intergroup conflict in humans (“war”) has received renewed attention within evolutionary biology and neighboring disciplines recently. Many of these new studies have focused on mathematical modeling of the evolutionary dynamics of violent intergroup conflict (Bowles, 2009; Choi and Bowles, 2007; Konrad and Morath, 2012; Lehmann and Feldman, 2008; Smirnov, Arrow, Kennett, and Orbell, 2007), frequently in order to explain how human altruism towards the in-group, i.e., a readiness to incur fitness costs to the benefit of others, could have evolved. In addition, recent field and laboratory studies have gathered instructive new evidence (Fry and Söderberg, 2013; Ginges and Atran, 2011; Gneezy and Fessler, 2011; Halevy, Weisel, and Bornstein, 2012; Koopmans and Rebers, 2009; Mathew and Boyd, 2011; Rebers and Koopmans, 2012). One prominent assumption frequently used in modeling and experimental investigation of intergroup conflict is that participation in war poses an n-person prisoner's dilemma and that war efforts can thus be modeled as a public good (Bowles, 2009; Choi and Bowles, 2007). A problematic consequence of this assumption, however, is that all war efforts would be altruistic by definition: If participation in war posed an n-person prisoner's dilemma, game theory would force us to accept that non-participation would be every individual's best choice under all circumstances. Actually, we do observe many instances of violent intergroup conflict — and this is very likely not a historically new phenomenon (see, e.g., Keeley, 1996). These observed deviations from the individually optimal strategy, then, would have to be explained with reference to some sort of altruistic motivation of the participants. This line of argument has led to the theory of “parochial altruism” (Bowles, 2009; Choi and Bowles, 2007): the idea that altruism towards the in-group (“in-group love”) and the readiness to engage in violent intergroup conflict (“out-group hate”) might have co-evolved throughout human pre-history.
By describing participation in war as necessarily altruistic, however, the n-person prisoner's dilemma view of war obscures the fact that there are other potential motivations for engaging in war that are not altruistic at all, like loot, status, and access to additional women (see, e.g., Glowacki and Wrangham, 2013; McDonald, Navarrete, and van Vugt, 2012). It is therefore problematic to draw inferences on altruistic motivations of war participants only from the fact that they are fighting.
In this study, evidence from real violent intergroup conflict (20th century wars) is presented which could call simple symmetrical n-person prisoner's dilemma models of war into question. The evidence suggests that altruistic behavior in war — in forms other than fighting — is conditional on the respective strategic situation, i.e., attack or defense. Specifically, it seems that these forms of altruistic behavior occur more frequently during defenses. This finding is in line with theoretical predictions derived from a more detailed model of the strategic structure of ancestral intergroup conflict (see Rusch, in press) and with the results of previous laboratory studies that found that intergroup conflict fosters “in-group love,” but not necessarily “out-group hate” (de Dreu et al., 2010; Halevy et al., 2012). Before the evidence is described, however, I will present a more detailed theoretical argument for why we might expect asymmetries in altruistic behavior between attacks and defenses in the following section.
Theoretical Background and Working Hypothesis
Some studies of violent intergroup conflict in chimpanzees (pan troglodytes) and human hunter-gatherers suggest significant differences in the strategic character of attacks and defenses (Kelly, 2005; Wrangham, 1999; Wrangham and Glowacki, 2012). According to these studies, coordinated group attacks in both species seem much more likely to be carried out when the perceived superiority of the aggressors is strong enough to induce low mortality risks for the attackers. Additionally, in a very recent study Glowacki and Wrangham (2013) also present evidence that the prospect of individual benefits leads to increased intensities of simple warfare among forager groups. In a recent paper (Rusch, in press), I present a simulation model that accommodates these observations. One of the predictions derived from that model is that altruism towards in-group members should occur more frequently in defenses than in attacks. The basic rationale of this prediction is laid out in the following section.
The asymmetry hypothesis
Successful attacks produce divisible goods — loot — which can be exclusively owned, i.e., private goods. The most important one among these goods is probably additional access to women (Kelly, 2005; McDonald et al., 2012), both of the out-group (see, e.g., Kohler and Kramer Turner, 2006) and the in-group (Chagnon, 1988), making intergroup conflict an arena of male reproductive interests (van Vugt, de Cremer, and Janssen, 2007; but also see Beckerman et al., 2009). Attacks might also produce public goods, namely deterrence and additional territory, but both these goods are of double-edged character. Aggressively trying to attain them can easily yield results opposite to the intended, i.e., higher risks of retaliatory counterattacks and net territory loss through the formation of perilous buffer zones between enemy groups' core territories — see Kelly (2005) for territory and Beckerman et al. (2009) for deterrence. Thus, compared to the direct fitness benefits of additional resources and reproduction, deterrence and territorial gains might eventually be better considered occasional by-products rather than primary goals of ancestral offensive warfare.
The primary goal of defenses, on the other hand, is not the production of additional goods but rather the protection of the available resources as well as individual and kin survival against threats from the outside — be they predators, other natural hazards, or human enemies. Protection, however, is a much less divisible good than loot or additional access to mates, i.e., protection is more of a public than a private good.
Moreover, once the critical threshold required to overpower the enemy combatants is reached, the individual share of loot acquired through raiding out-groups is likely to decrease with the number of warriors participating in the raid, while the individual chances of survival for defenders are likely to increase with every additional man participating in the defense; see Rusch (in press) for further discussion and additional references. Thus, in-group defense is probably best characterized as a threshold public good that is produced only if a certain critical number of people participate. In threshold public goods games, altruistic forms of cooperation have much better chances to evolve than in other public goods games (Bach, Helvik, and Christiansen, 2006).
Accordingly, if this reasoning is sound, the fitness relevant cost/benefit structures of attacks and defenses lead to quite different expectations about the behavior that should be observed in each of the strategic situations. First, obviously, if we can safely generalize from these assumptions to a tentative model scenario of ancestral human intergroup conflict, we should expect psychological and behavioral responses to have evolved that differentiate between attacks and defenses. More specifically, it can be hypothesized that evolution might have favored a greater readiness to behave altruistically during defenses, because under ancestral conditions, compared to the failure of attacks, failure of the collective action in defenses presumably was much more likely to result in detrimental consequences for the individual combatant and his kin. Behaving altruistically, i.e., supporting in-group members at a personal cost, in order to prevent these consequences can be adaptive, as theoretical studies show (e.g., Bach et al., 2006). I will refer to this conjecture as the ‘asymmetry hypothesis’: Altruistic behavior towards in-group members during violent intergroup conflict should be observed more frequently in defenses. Also see, e.g., Durham (1976), Kelly (2005), Rusch (in press), and Tooby and Cosmides (2010) for further discussions of these points.
Testing the asymmetry hypothesis
Testing the asymmetry hypothesis, however, is not a straightforward task. Many previous studies have used economic experiments to measure altruism in situations of intergroup conflict (e.g., Ginges and Atran, 2011; Gneezy and Fessler, 2011; Halevy et al., 2012; Koopmans and Rebers, 2009; Rebers and Koopmans, 2012). These studies face the problem of interpreting subjects' behavior in economic experiments (Burton-Chellew and West, 2013; Kümmerli, Burton-Chellew, Ross-Gillespie, and West, 2010). More importantly in our context, they also fail to distinguish clearly between offensive and defensive situations. In most cases, a situation of intergroup conflict is simply forced upon the subjects by the experimental design, making it unclear if they will perceive themselves as “under attack” or “attacking.”
A recent study by Halevy et al. (2012) is a rare exception. There, subjects in a modified public goods game were given the choice between an in-group beneficial allocation with negative external effects on an out-group, i.e., an “offensive” option, and an in-group beneficial allocation without these effects. Although this option was introduced only after a phase of inevitable intergroup conflict in which only the offensive option was available, the result was that subjects quickly opted out of the conflict with the out-group while continuing cooperation with the in-group.
While this experiment is not a direct test of the asymmetry hypothesis, it yields a first indication that “in-group love” and “out-group hate” might not be linked as rigidly as suggested by theories of “parochial altruism.” In a similar vein, in an experiment using the same game, de Dreu et al. (2010) were able to establish that oxytocin, a neurotransmitter which has been found to promote cooperation and trust, increases in-group cooperation but does not promote aggression towards the out-group.
The current study takes an alternative approach. It analyzes historical case data from real-world intergroup conflicts of the 20th century (see materials and methods). While these data allow for a clearer distinction of behavior in attacks from behavior in defenses, the analysis is burdened with the intricate problem of having to define and to operationalize sufficiently precise measures of altruistic behavior in real war. If this can be achieved, though, the ecological validity of the resulting findings is guaranteed.
Thus, in order to capture different characteristic aspects of altruism among the members of a group in wartime, four interrelated proxies were defined and systematically coded from the historical case data. Multiple proxies are used in order to capture several behavioral expressions of an altruistic motivation. All of them involve taking great individual risks, up to actually sacrificing one's own life, in order to directly help others. To increase the chances of actually capturing behavior motivated by altruistic intentions, the focus is on proxies that do not involve extensive fighting by the target individual.
A word of caution is in order here: There is no doubt that brave fighting against the out-group is costly behavior that benefits one's in-group, at least in the short-run. In this sense all behavior during war except for desertion can be called altruistic. However, as, e.g., Kerr and Godfrey-Smith (2002) argue, there is an important difference between behavior that is in the interest of the individual but also benefits its group as a by-product (which they label “weakly altruistic” behavior), and behavior that inevitably causes a net loss of fitness relevant resources or chances of survival for the individual while benefiting the in-group (labeled “strongly altruistic” behavior by Kerr and Godfrey-Smith). Using this distinction, the asymmetry hypothesis can be restated like this: Instances of “strong altruism” are more likely to be observed during defenses, because, under ancestral conditions, the majority of attacks presumably were cooperative raids motivated by the prospect of individual gains (Glowacki and Wrangham, 2013), while defenses posed n-person threshold public good problems (see above).
As already discussed, fighting behavior cannot be clearly classified as weakly or strongly altruistic. Therefore this study tries to focus on behaviors that represent somewhat less questionable instances of strong altruism (with all due caution, see discussion). These behaviors are instances of direct help, like, e.g., protecting comrades from the impact of explosions using the own body as a shield or running to the rescue of casualties under enemy fire. However, the data also allow for a contingency analysis of the risks induced by the acts of brave fighting depending on strategic situation. A skewed distribution of these risks could yield at least a tentative indication that there are asymmetries in weak altruism between attacks and defenses as well.
In the other statistical analyses, the null-hypothesis is that the instances of brave fighting and instances of (more) directly altruistic behavior — as defined in the following section — are equally probable in attacks and in defenses. This is the symmetry assumed by the theory of “parochial altruism.” It rests on the assumption that, in general, attacks and defenses offer occasions for all kinds of altruistic behavior with the same probability, and that these occasions are used by the protagonists to the same degree. The alternative hypothesis is the asymmetry hypothesis, which assumes that, while the occasions for altruistic behavior are equally probable in both kinds of situations, they are used more often in defenses.
It should be noted, though, that if the following analysis reveals differences in the frequencies of altruistic behavior between attacks and defenses, this does not allow us to jump to the conclusion that these differences are actually caused by a conditional motivational psychology of intergroup conflict. These differing frequencies may still either be caused by individual reactions to the strategic situation (“individuals are more altruistic in defenses”) or due to structural differences of attacks and defenses (“defenses offer more occasions for altruistic behavior”). In any case, if asymmetries are found, this could challenge the symmetry assumption made by current theories of “parochial altruism.”
Materials and Methods
Data source
The publicly available official citations of 966 US-American Medal of Honor recipients from WWI, WWII, and the Korean and Vietnam wars were systematically analyzed — see, e.g., http://history.army.mil/. These documents were used because they include detailed descriptions of the actions for which the medals were awarded and are thus a quite reliable source of information on individual behavior during war; see Appendix 1 for an example citation. Compared to those from later wars, the 119 citations from World War I are rather brief, however, thus making it impossible to extract reliable information on injuries from these cases. The particular events studied here presumably are not representative of all behavior during war because they represent acts of “outstanding gallantry” – as is the official wording of the citations — i.e., heroic deeds. They are, however, well suited to investigate conditions triggering altruistic behavior in war.
Variables coded
There are six main variables of interest in this study; their possible values are given in square brackets and their meaning is explained in the subsequent paragraph. Please also see Appendix 2, the coding instructions, for additional details. Variables: (1) strategic situation [attack/defense]; (2) action category [leadership/altruism/fighting]; (3) acts of direct help involved in action [yes/no]; (4) self-sacrifice by covering explosive [yes/no]; (5) severe injury of soldier [yes/no]; (6) soldier killed in action [yes/no].
First, it was determined whether the action described took place during a defense or in an attack. Defenses were defined as situations in which the soldier is under the attack of enemy forces and his strategic goal is to either hold his position or to safely retreat. An attack, on the other hand, was defined as a situation in which the soldier is aggressively engaging the enemy. His strategic goal usually is to (re-)conquer enemy ground, to try the enemy's strength or to destroy selected targets.
Second, it was determined whether the particular action described was an example of either skillful military leadership, brave fighting against the enemy, or outstanding direct altruism towards comrades. In the following sections the respective groups of soldiers will be referred to as “leaders,” “fighters,” and “altruists.” An action was classified as (directly) “altruistic,” and will be referred to as such in the following, when it involved only very little or no fighting by the soldier and when its predominant purpose was to help comrades instead of fighting the enemy. Examples are: carrying wounded comrades from danger zones or protecting them by diverting an incoming threat to oneself.
Third, it was noted separately if an action included direct help to comrades — instead of just fighting together. The main difference between “acts including help” and acts of “direct altruism” is that helping is the principal goal of the latter, while it occurs more incidentally in the former. Note that the categorizations (2) and (3) are not independent. Altruistic actions virtually always (270 of 271 actions) included direct helping but not vice versa: 99 of 568 (17%) acts of brave fighting and 32 of 127 (25%) acts of skillful leadership also include helping.
Fourth, starting with WWII, it is frequently the case that soldiers were awarded the Medal of Honor for protecting comrades nearby by covering live grenades or other explosive devices with their own body. A special count was made of these instances, called “self-sacrifice” in the following. Self-sacrifice (variable 4) always counted as an altruistic action (variable 2) and as helping (variable 3). It should also be noted that, with regard to the asymmetry hypothesis, this measure of self-sacrifice is susceptible to bias caused by possibly differing frequencies of the use of grenades in modern warfare depending on the strategic situation. If grenades were thrown mostly by attackers, this kind of self-sacrificial behavior will, obviously, be more frequent on the defending side. No reliable source of information on this question could be found. Therefore, findings which rely on the “self-sacrifice” proxy have to be interpreted with special caution.
Fifth, it was noted if the soldier was severely wounded during his action. Sixth, it was determined whether the soldier died in the course — or as a direct consequence — of his action. Finally, the military branch of the soldier and whether he served as a medic or a chaplain was also noted.
Reliability
Coding was carried out by the author of the study. The full list of citations and coded variables is available upon request. To test the reliability of the coding procedure, 30 citations were also coded by two colleagues, who were given the coding instructions for the six variables, see Appendix 2, and a briefing on the military vocabulary of the citations. They were not informed about the particular research question of the study. Twenty-five initial disagreements (out of 3 × 180 decisions overall) could be resolved in discussion. Eventually, only a small number of decisions remained ambiguous. Overall inter-rater reliability was very high (three raters, 180 decisions each, Fleiss' κ = .94, or 95% agreement, after having resolved 25 disagreements in discussion). Within categories, i.e., 30 decisions per rater each, agreement and reliability also were high: (1) strategic situation: κ = .87, 90% agreement; (2) action category: κ = .92, 93% agreement; (3) direct help involved: κ = .87, 90% agreement; (4) self-sacrifice: κ = 1.0, 100% agreement; (5) severe injury of soldier: κ = .85, 97% agreement; (6) soldier killed in action: κ = 1.0, 100% agreement.
Pre-analyses
The data on Medal of Honor recipients from World War II is also part of a separate study on heroism in war for which more information was gathered than for the study at hand. In particular, information about rank (coded as pay grade), age, race, education, time in military service, and enemies killed is (partially) available for these 464 soldiers. Table 1 gives an overview of this information.
Additional characteristics of Medal of Honor recipients of WWII and tests for differences between attacks and defenses
Notes: Values are mean ± SD; all characteristics are measured at the time of the individual heroic action.
There is a noteworthy difference in ranks between attacks and defenses, which entails a difference in education levels, because rank and education are correlated (Pearson's r = .47, p < .001, n = 216). Closer analysis reveals, though, that this difference is only significant within the group of fighters (U = 10,622.5, p < .001, n = 305), but not the altruists and leaders (U = 950.5, p = .411 for altruists, n = 92; U = 464.5, p = .389 for leaders, n = 62). This finding, which might be quite interesting for military historians, should therefore not interfere with the analysis of altruism as coded here. Finally, no differences between attacks and defenses exist regarding race (Fisher's exact test, two-sided, p = .931).
The following section presents the results obtained when analyzing the full set of 966 cases from all four wars. It could be argued, though, that the research question of this particular study requires the exclusion of a number of soldiers. First, those heroes who distinguished themselves by skillful tactical leadership likely were not in the position to display the forms of altruism investigated here as frequently because they mostly did not take part in combat as directly as the other soldiers. This is also reflected by their lower mortality rates (see results: Table 3). Second, soldiers who were ordered to help, i.e., chaplains and medics, might bias altruism statistics. Third, including members of Air Force and Navy — i.e., in this case mostly pilots and seamen — in the analysis might appear problematic because their situation is very different from the conditions of ancestral warfare. Analyzing only the cases of Army and Marine Corps members and excluding medics and chaplains might yield a better approximation of those conditions (this subset of soldiers is referred to as “land forces” in the following). Appendix 3 therefore reports the results of repetitions of the central analyses on respectively modified case selections and shows that the central results of this study also hold for these subsets.
Contingency statistics for attributes of heroic acts of war and strategic situation (attack or defense)
Notes: Two-sided χ2-tests
p < 0.01;
p ≤ .001) and p values of more conservative, two-sided exact Fisher tests are stated.
Contingency statistics for different heroic acts and mortality
Note: Two-sided χ2-tests (***p ≤ .001) and p values of more conservative, two-sided exact Fisher tests are stated.
Results
Main results are summarized in Table 2 and depicted in Figure 1. The central finding is that all three proxies – “altruism,” “help,” and “self-sacrifice” – indicate significant differences between attacks and defenses. More specifically, during defenses the heroic actions included more unambiguous instances of direct altruism, involved more acts of direct helping, more frequently were acts of self-sacrifice, and additionally entailed an increased risk of being seriously wounded.
Overview of the distribution of the characteristics discussed
The overall mortality risk of the heroic actions, however, is not significantly different between attacks and defenses. Note that this finding does not contradict the assumption made in the derivation of the asymmetry hypotheses that risks between attacks and defenses in intergroup conflict might have differed throughout our evolutionary past. This assumption cannot be tested with the available data because the cases analyzed here exclusively comprise outstandingly heroic deeds that, obviously, are not representative of war in general. Furthermore, the data at hand stem from modern warfare in which the actual risks presumably differ from those of ancestral intergroup conflict. Finally, and most importantly, the asymmetry hypothesis does not even require any particular distribution of the risks that altruistic acts entail other than that acting altruistically must be more risky than not doing so, all else being equal. An analysis of the risks induced by the different forms of altruistic acts in this data set clearly shows that they were costly for the individual soldier in terms of chances of survival, as is shown in Table 3.
Remember, though, (i) that this is a comparison of the risks induced by different kinds of heroic acts which all are presumably much more risky than non-heroic acts and (ii) that the proxies are not independent. When excluding self-sacrificial acts, altruism remains significantly more costly — i.e., compared to other heroic actions — only when restricting the analysis to land forces (61.5% mortality risk for altruists, n = 39, as compared to 48.7% for fighters, n = 507, and 31.0% for leaders, n = 87; χ2 = 12.80, p = .001, Fisher's exact test: p = .001). The risk induced by helping also remains higher (50.7% of 142 helpers died compared to 46.0% of 491 the other heroes) but this difference is no longer significant. This shows that the differences in mortality risks between altruistic and non-altruistic heroic acts are to a large extent driven by the acts of self-sacrifice. As argued above, however, this does not conflict with the assumptions leading to the asymmetry hypothesis. It is probably safe to assume that the heroic altruistic acts analyzed here induced higher mortality risks than non-heroic behavior as the Medal of Honor is officially awarded to soldiers who “distinguished themselves conspicuously by gallantry and intrepidity at the risk of their life above and beyond the call of duty” as is the official wording. The asymmetry hypothesis, in fact, receives additional support by the contingency analysis of action characteristics and strategic situation with self-sacrificial actions excluded, as is shown in Table 4.
Contingency statistics for attributes of heroic acts of war and strategic situation with self-sacrificial actions excluded
Note: Two-sided χ2-tests (***p ≤ .001; *p ≤ .05) and p values of more conservative, two-sided exact Fisher tests are stated.
As mentioned above, a closer analysis of the risks induced by fighting behavior is in order, because the behavioral category “brave fighting” in this study is a residual category which consists of acts of “indirect altruism.” Table 5 shows the distribution of injury and mortality risks between attacks and defenses, when only those actions are included in the analysis that were classified as brave fighting.
Contingency statistics for risks induced by brave fighting and strategic situation
Note: Two-sided χ2-tests (***p ≤ .001; *p ≤ .05) and p values of more conservative, two-sided exact Fisher tests are stated.
The results suggest that brave fighting during defenses may have induced higher risks of being injured and of being killed. If it can be presumed that there are no structural differences in the general risks of brave fighting between attacks and defenses, which is suggested — but certainly not established — by the lack of a significant difference in mortality risks between attacks and defenses when analyzing the full data set (see Table 2), this could indicate that brave fighters were willing to take greater risks during defenses. It could, however, also be the case that medals for brave fighting were preferentially awarded for offensive actions. This, then, might require somewhat “braver” fighting during defenses in order for such an action to be recognized as commendable. In fact, this is not unlikely, because it can be suspected that military commendation is mostly used to reinforce unhesitating offensive bravery. This is a potential source of bias in the category “brave fighting” which is not as likely to be effective for acts of direct altruism, however, because altruistic actions are presumably more likely to be commended independent of the strategic situation (but see discussion).
Finally, for the subset of cases drawing on WWII, an additional proxy for directly altruistic heroic behavior can be analyzed because information on the number of enemies killed is available here. Using this more objective criterion, this proxy can be used to crosscheck the findings reported so far, which all rely on variables which required more interpretation by the coder. This fourth proxy is based on the following assumption: An action which did not involve the killing of enemies but was regarded heroic enough to be honored with the highest US-American military award probably consisted of directly helping others. (Indeed, 26 of the 31 of such actions were classified as directly altruistic acts when considering the full information available from the citations.) Significantly more acts of this kind occurred during defenses, as is shown in Table 6, which again confirms the asymmetry hypotheses. This is especially noteworthy because in WWII on average more enemies were killed by Medal of Honor recipients during defenses (see Table 1).
Contingency statistics for heroic actions involving the killing of enemies and strategic situation during WWII
Notes: Two-sided χ2-test (***p ≤ .001) and p value of more conservative, two-sided exact Fisher test are stated;
one of these actions (the one by James K. Okubo) could not be classified as either attack or defense, because the citation included too little information.
This finding is robust against the exclusion of self-sacrificial actions, which represent eight of the actions not involving the killing of enemies. Of the remaining 22 actions 14 took place during a defense (representing 19.7% of the 71 defenses) and eight during attack (representing 4.0% of the 199 attacks) which is a significant difference (χ2 = 17.23, p < .001; Fishers exact test: p < .001).
Discussion
Analyzing data from 20th century wars, this study finds noteworthy asymmetries in directly altruistic behavior as well as indications of asymmetries in indirectly altruistic behavior during intergroup conflict. All four proxies used here to capture behavioral expressions of altruistic motivations consistently indicate that acts of altruism are more frequently observed during defenses. This asymmetry was predicted by an earlier theoretical analysis of the fitness relevant costs and benefits of ancestral intergroup conflict (Rusch, in press) and is in line with recent anthropological findings regarding motives for going to war (Glowacki and Wrangham, 2013) and some findings from experimental economic games (de Dreu et al., 2010; Halevy, Weisel, and Bornstein, 2012).
This study is limited in a number of ways, however. First, it relies on historical descriptions of heroic actions. It is difficult to assess which biases might have influenced the process of selecting heroic actions for being awarded the Medal of Honor. This does not interfere with the analyses at hand, though, as long as there is no reason to assume that altruistic heroic deeds during attacks were systematically disregarded in the selection process or altruistic acts during defenses systematically favored.
Second, this study exclusively analyzes actions of “outstanding gallantry” which are, obviously, unlikely to be representative of all behavior in war. This should, however, also not pose a critical problem, since the aim of this study is to determine conditions promoting one particular kind of outstanding behavior, i.e., altruism.
Third, critics might take issue with the fact that some of the measures of altruism used here are not independent and were obtained in a somewhat subjective manner by systematically screening Medal of Honor citations using a fixed coding scheme. Although overall inter-rater reliability was high this remains true, but mostly for the “direct altruism” proxy. The criteria for self-sacrifice (“Soldier covered an explosive with his own body in order to protect comrades nearby”) and the WWII actions not involving any killing of enemies can hopefully count as objective. A potential problem for the “self-sacrifice” proxy lies in missing information about the base rate of the use of grenades in attacks and defenses. The results relying on this proxy should thus be regarded as tentative. All other proxies, including the second objective proxy that uses the number of enemies killed to distinguish acts of direct altruism from acts of heroic fighting, however, still indicate asymmetries when acts of self-sacrifice are excluded.
Another potential problem of the coding system used in this study is the decision to focus on displays of “direct altruism” and “direct help” and separating these from “indirect” forms of altruism, particularly brave fighting for the benefit of the in-group. I have tried to argue in favor of this distinction by pointing out that brave fighting alone is an ambiguous indicator of an altruistic motivation while directly helping comrades without fighting the enemy is not, or at least not as much. It is undoubtedly true, however, that fighting the enemy, e.g., forcefully breaking through an encirclement or staying back to secure the retreat of comrades by providing covering fire, promotes the well-being of the in-group. Brave behavior of this kind is not captured by the “direct altruism” proxy of this study. The contingency analysis of the risks induced by brave fighting depending on strategic situation, however, has indicated that asymmetries seem to exist in this category as well. An implicit assumption of this study is that by treating fighting behavior in this way no systematic bias is introduced to the “direct altruism” proxy. But even if this proxy is biased because of the way the coding scheme was designed, this study was able to present three other proxies for altruistic motivations, which still indicate differences between attacks and defenses and are less likely to be biased in such a way — although they do have their own problems, of course.
More importantly, however, it cannot be ruled out that there is an unthought-of structural difference between attacks and defenses which reduces the frequency of occasions on which direct altruism can be displayed during attacks. It therefore remains to be experimentally tested if a defensive strategic situation actually causes more altruistic behavior or not. This could not be determined by the current study because it solely relies on contingency analyses.
In summary: The evidence presented here suggests that readiness to incur fitness costs in order to benefit others during war depends on the combatants' strategic situation. If this conclusion is accepted, current models of the evolution of bellicose behavior and especially altruism in war need to be refined. This study's findings do certainly not rule out the possibility that some features of intergroup conflict can be modeled as public goods problems. They raise the question, though, of why there seem to be asymmetries in the frequencies with which altruism is observed in attacks and defenses. One speculative suggestion is that this finding reflects different payoff structures of the two situations throughout our evolutionary past. Particularly, attacks during ancestral warfare might mostly have consisted of raids by superior aggressors with rather low risks for the individual attacker. Defenses, in contrast, might have been more risky; at the same time, their success might also have been much more important for individual survival and the survival of kin. This might have caused selection for greater readiness to act altruistically when defending the in-group. In game theoretical terms: attack might be better described as a problem of team production, while defense might form a threshold public good, both of which are scenarios in which interests are partially aligned and not as strictly conflicting as in n-person prisoner's dilemmas.
Footnotes
Acknowledgements
I thank Karina Kraft, Carl Salk, and Charlotte Störmer for help with coding, and Max Albert, Matthias Greiff, Eckart Voland, and EP's anonymous reviewers for valuable criticism. This publication represents a component of my doctoral thesis (Dr. rer. nat.) in the Faculty of Biology at the Justus-Liebig-University Giessen, Germany.
An Example Medal of Honor Citation
| Soldier's name: | BAUGH, WILLIAM B. | |
| Rank and organization: | Private First Class, U.S. Marine Corps, Company G, 3rd Battalion, 1st Marine, 1st Marine Division (Rein). | |
| Place and date: | Along road from Koto-ri to Hagaru-ri, Korea, 29 Nov. 1950. | |
| Citation: | For conspicuous gallantry and
intrepidity at the risk of his life above and beyond the
call of duty while serving as a member of an antitank
assault squad attached to Company G, |
|
| Source: | http://history.army.mil/html/moh/koreanwar.html [accessed July 22, 2013] | |
|
|
||
| Coded as: | Category: |
|
| Strategic situation: |
|
|
| Direct helping: |
|
|
| Self-sacrifice: |
|
|
| Wounded: |
|
|
| Killed in action: |
|
|
|
|
||
Note: Underlines were added by the author to mark the phrases which led to the respective codings.
Coding Instructions
Please read through the each of the following 30 official Medal of Honor citation. Then answer the following questions after you have carefully read through the additional specific instructions and notes.