1. Introduction

Altruism is defined by evolutionary biologists as an act that increases the fitness of others while decreasing the fitness of the actor. This definition makes no reference to the psychological mechanisms that motivate the act, which might or might not count as altruistic as psychologists and philosophers use the term. Fitness effects and psychological mechanisms can be understood and related to each other in terms of the evolutionary distinction between ultimate and proximate causation, as reviewed by Sober and Wilson (1998).

Altruism has traditionally been studied in the context of helping, but punishment can also qualify as altruistic in terms of its fitness effects. Consider the simultaneous evolution of two traits: (1) an altruistic act such as providing a public good versus free-riding, and (2) punishing free riders versus failing to punish. Nonpunishers benefit from the policing efforts of the punishers in just the same way that free riders benefit from the public good provided by the altruists. Theoretical models show that altruistic punishment can favor the evolution of other altruistic acts that would not evolve in the absence of punishment (Boyd et al., 2003, Boyd & Richerson, 1992, Gintis, 2000, Henrich & Boyd, 2001). Nevertheless, altruism supported by punishment does not necessarily evolve to fixation. Instead, a behavioral polymorphism often results, which includes public good providers, free riders, punishers, and nonpunishers.

Strong experimental evidence exists for this behavioral mix in humans, thanks largely to the efforts of economists (Fehr et al., 2002, Fehr & Gächter, 2000, Fehr & Gächter, 2002, Ostrom et al., 1994). In games where a public good can be provided at private expense, individuals differ in their tendency to free ride. Once the more generous members of the group realize that they are being exploited by free riders, they tend to withhold their altruism, resulting in the complete absence of public good provision. When the opportunity to punish free riders at private expense is added, some individuals elect to punish, in addition to providing the public good. A sufficient number of punishers make free-riding disadvantageous, and public good provision rises to near maximum levels. Punishment can be regarded as self-interested, despite its private cost, if the punisher benefits from increased public good provision over the long term. However, numerous experiments have been performed in which the return benefits of punishment are rigorously excluded (Fehr & Gächter, 2000, Fehr & Gächter, 2002). These experiments show that some individuals punish only when it is in their perceived self-interest, but a sizeable fraction continues to punish in the complete absence of return benefits. These experiments unequivocally demonstrate the existence of altruistic punishment in humans at the behavioral level. The pronounced individual differences are also in accord with the aforementioned theoretical models.

While the existence of altruistic punishment in humans is well established, its interpretation has become the subject of a vigorous debate. One possibility is that altruistic punishment is an adaptation that evolved because “groups with a high fraction of altruistic punishers would have sustained cooperation more successfully than groups with fewer punishers, and so would have prevailed over them” (Bowles & Gintis, 2002, p. 128). Another possibility is that altruistic punishment is maladaptive in a modern context and only makes sense in relation to the human ancestral environment. According to this mismatch hypothesis, people lack the cognitive adaptations for behaving appropriately in the context of the experiments because social interactions invariably took place among genetic relatives or nonrelatives with a high likelihood of future interactions (Johnson, Stopka, & Knights, 2003).

In this paper, we use fictional scenarios to explore the effects of information regarding genetic relatedness, friendship, and potential for future interactions in situations that invoke altruistic punishment and altruistic helping. We conducted two experiments, using fictional scenarios that were similar to the aforementioned games involving actual interactions. The first experiment was designed to test the effects of information regarding genetic relatedness, friendship, and potential for future interactions on the desire to punish norm violations. The second experiment simplified and changed some details of the transgression scenario and added an altruistic helping scenario. The advantage of this methodology is that the elements of the scenarios can be systematically varied and presented to large numbers of people in a way that would be difficult or impossible to stage with actual interactions. Responses to fictional scenarios provide important insights into psychological mechanisms, even when they do not correspond directly to responses to actual interactions. In our case, we demonstrate a remarkable difference between altruistic helping and altruistic punishment in their sensitivity to information regarding genetic relatedness, friendship, and potential for future interactions. This difference is relevant to the debate over altruistic punishment and more generally to the evolution of psychological mechanisms associated with morality.

2. Experiment 1

2.2. Results

Considering all treatments together, participants responded to the scenario with a high average degree of anger (M=7.4 on a nine-point scale, S.D.=1.7) and desire to punish the transgressor (M=6.1, S.D.=2.3). On average, they felt that the transgressor should pay back approximately what he stole (US$2089.4, S.D.=2576.8), and they were prepared to pay a mean of US$421.3 (S.D.=2140.1) to punish the transgressor. However, there were no statistically significant differences among the treatments (p>.2). The variables of genetic relatedness, friendship, and potential for future interactions appeared to be psychologically irrelevant. There was also no main effect or interaction effects for sex, although it approached significance for a main effect on desire to punish the transgressor (p=.083; MMales=6.4, S.D.Males=2.2; MFemales=5.9, S.D.Females=2.4).

Another important result concerns the distribution of scores (see Fig. 1), in addition to their mean values. The self-report of anger and desire to punish were so strong that the distributions were truncated at the high end. In fact, the most common response for anger was the maximum value of nine. The distribution for the amount that the transgressor should pay includes a large spike corresponding to the amount that he stole, with other participants indicating that he should pay more (retribution) or paradoxically less than he stole. The distribution for the amount that the participant would be willing to pay to punish the transgressor shows that a large fraction of participants are willing to pay nothing or very little, while others are willing to pay a great deal. Taken together, these distributions suggest that, while most participants are very angry and would like to see the transgressor punished, only some are willing to altruistically punish.

Fig. 1. Histograms of participants' responses to the fictional scenarios in Experiment 1 [n is 463, 463, 417 and 431 for (A), (B), (C), and (D), respectively]. Panel (A) shows the distribution of responses for reported levels of anger, showing that a majority of participants reported a high level of anger. Panel (B) shows that a substantial number of participants also had a high desire to punish the transgressor. Panel (C) shows the amount of money that participants thought the cheater should repay, with intervals of US$500 (the x axis labels show the upper limit of the corresponding interval). Panel (D) shows the amount of money that participants reported being willing to pay to punish the transgressor, with intervals of US$2000 (the x axis labels show the upper limit of the corresponding interval). The data show a dissociation between participants' reported emotions and their willingness to punish.

3. Experiment 2

3.2. Results

The results of the second experiment confirmed the results of the first experiment in all essential details: Participants in the punishment treatments indicated strong anger at the transgression (M=6.4, S.D.=2.2) and a moderate to high desire to punish (M=5.4, S.D.=2.4) but were insensitive to information regarding genetic relatedness or potential for future interactions (p>.16 for both main effects). Most wanted the transgressor to pay back what had been stolen (52.1%), and many were unwilling to pay to punish the transgressor (48.1%), despite their anger (see Fig. 2). Providing a socially sanctioned mechanism for punishment (small claims court) significantly decreased the average level of anger [M=5.9, S.D.=2.2 for the treatments with small claims court option vs. M=6.7, S.D.=2.2; F(1,247)=8.58, p=.004] but did not increase the desire of participants to punish (p=.26), the amount they thought the transgressor should pay (p=.83), or the amount that they were willing to pay to punish (p=.99).

Fig. 2. Histograms of participants' responses to the fictional scenarios in Experiment 2 [n is 249, 249, 240, and 233 for (A), (B), (C), and (D), respectively]. Panel (A) shows the distribution of responses for reported levels of anger, showing that a majority of participants reported a high level of anger. Panel (B) shows that a substantial number of participants also had a high desire to punish the transgressor. Panel (C) shows the amount of money that participants thought the cheater should repay, with intervals of US$250 (the x axis labels show the upper limit of the corresponding interval). Panel (D) shows the amount of money that participants reported being willing to pay to punish the transgressor, with intervals of US$250 (the x axis labels show the upper limit of the corresponding interval). The data support the findings in Experiment 1. The less skewed data for Panel (A) contrast with the distribution for Experiment 1 [Panel (A)], which is due to the manipulation of the presence or absence of a normative option to punish the transgressor by going to small claims court.

There was a main effect for sex on desire to punish the transgressor [F(2,238)=5.74, p=.004; MMales=6.0, S.D.Males=2.4, MFemales=5.0, S.D.Females=2.4] and near-significant effect on level of anger [F(2,238)=2.97, p=.05; MMales=6.6, S.D.Males=2.2, MFemales=6.2, S.D.Females=2.2]. This suggests that females are emotionally more moderate in their responses to the transgression, but this did not produce a sex effect for the amount that participants thought the transgressor should pay (p=.92) nor the amount that they were willing to pay to punish (p=.78). Furthermore, the stronger sex effect (on desire to punish the transgressor) did not produce notably different distributions for males and females for responses to that question.

In contrast to the punishment scenario, participants were highly sensitive to information regarding genetic relatedness and potential for future interactions in the helping scenario. Participants showed an overall moderately positive level of emotional sympathy for the individual needing help (M=6.7, S.D.=2.2), a moderate willingness to help (M=6.6, S.D.=2.2), and a mean offer of financial assistance of US$553.1 (S.D.=886.0). Table 1 presents a summary of participants' responses for each of the three dependent measures by perceived relatedness and potential for future interactions.

Table 1. Mean values (S.D., n for cell) for responses to questions regarding the helping scenario

	Dependent measure		Cousin	Stranger
	Sympathy for protagonist	Likely future interactions	7.5 (1.7, 32)	7.1 (1.6, 31)
		Unlikely future interactions	6.8 (2.2, 33)	5.6 (2.6, 33)
	Willingness to help	Likely future interactions	7.3 (1.8, 32)	7.1 (1.6, 31)
		Unlikely future interactions	7.1 (1.7, 33)	4.8 (2.6, 33)
	Amount willing to give	Likely future interactions	768.97 (915.09, 29)	434.22 (291.3, 32)
		Unlikely future interactions	896.36 (1458.14, 28)	175.94 (184.06, 32)

The questions on sympathy for the protagonist and willingness to help the protagonist had a response scale of 1 to 9. The question asking how much help would participants provide to the protagonist, quantified as dollars, was open; that is, participants could suggest any amount that they wished. Note that while all participants were presented with the three questions, the cells present data for independent subjects, not repeated measures.

An analysis of variance showed that participants had greater sympathy for relatives versus strangers [F(1,125)=4.51, p=.036] and for protagonists with an expected future interaction versus no future interactions [F(1,125)=8.8, p=.004]. For willingness to help, participants made similar distinctions between relatives and strangers [F(1,125)=12.87, p<.0005] and between future and no-future situations [F(1,125)=12.37, p=.001]. There was also an interaction effect between those two variables [F(1,125)=9.37, p=.003], with the stranger/no-future interactions treatment significantly differing in post hoc analysis from the other three treatments. Finally, participants in the altruistic helping treatment made significant distinctions depending upon relatedness [F(1,117)=11.61, p=.001] for the amount that they were willing to provide as assistance (participants were willing to provide a mean of US$831.50 to a cousin but only US$305.10 to a stranger), but not depending upon the likelihood of future interactions [F(1,117)=0.18, p=.67]. There was a near-significant effect for sex on level of sympathy [F(2,118)=3.0, p=.054; MMales=6.4, S.D.Males=2.1, MFemales=7.2, S.D.Females=2.0]. This suggests that females report feeling more sorry for the protagonist, but this did not produce a sex effect for how much participants thought they should help (p=.19) or how much they would contribute to help (p=.96).

4. Discussion

Our fictional scenarios were designed to resemble a large body of research involving actual interactions and gave comparable results. In particular, some but not all participants indicated a willingness to punish at their own expense despite the absence of return benefits. Our experiments went beyond previous research by manipulating information regarding genetic relatedness, friendship, and potential for future interactions in a complete factorial design, which is more feasible for fictional scenarios than actual interactions. These are the most important variables influencing the evolution of cooperation and altruism according to kin selection theory (Hamilton, 1964, Hamilton, 1975) and evolutionary game theory (Axelrod & Hamilton, 1981, Maynard Smith, 1982). We therefore expected them to have a strong effect on the desire to altruistically punish and were surprised when they turned out to be statistically insignificant, despite large sample sizes. It is important to emphasize that the altruistic punishment scenario strongly engaged the interest of the participants, as indicated by their reported degree of anger, their desire to see the transgressor punished, and the willingness of at least some of the participants to punish at their own expense. The variables of genetic relatedness, friendship, and potential for future interactions had no effect despite a strong overall psychological response.

These results were so surprising to us that we decided to replicate the experiment with minor changes and to add an altruistic helping scenario. The second experiment confirmed the results of the first experiment with respect to altruistic punishment and demonstrated a strong sensitivity to information regarding genetic relatedness and potential for future interactions with respect to altruistic helping, as we originally expected for all forms of altruism. Thus, insensitivity to these major variables is not an artifact of our experimental procedure but indicates an important psychological difference between altruistic punishment and altruistic helping, at least in response to fictional scenarios.

We think that this is an important empirical result that demands an explanation, even if the correct explanation is not immediately apparent. We conclude with a brief outline of the major possibilities in hope of stimulating additional research.

First, our result might be an artifact of using fictional scenarios that have no relevance to real-world interactions. By itself, this possibility is not explanatory because it criticizes an entire method rather than our particular result does. Fictional scenarios are widely used in traditional and evolutionary psychological research and are closely related to questionnaire studies, which are even more widely used. If the method is to be used and trusted at all, it cannot be invoked to distrust a particular result. Moreover, it does not explain any of the details of our results, such as the difference between altruistic punishment and altruistic helping. It is clear that all research involving fictional scenarios and questionnaires is most instructive in conjunction with studies of real-world interactions and often reveal important psychological mechanisms, even when they do not correspond directly to behavior in actual interactions. In our case, our fictional scenarios are closely related to experiments involving actual interactions and give comparable results with respect to altruistic punishment, suggesting that the difference between altruistic punishment and altruistic helping is more than a meaningless artifact. See Wilson and O'Gorman (2003) for a more detailed discussion of fictional scenarios as a research method and the importance of narrative in psychological and cultural processes.

Second, our result might reflect a mismatch between the modern environment and psychological mechanisms adapted to the ancestral environment, as suggested for altruistic punishment by Johnson et al. (2003). One version of the mismatch hypothesis maintains that all social interactions in the ancestral environment were among genetic relatives and nonrelatives with a high probability of future interactions. Because these were environmental constants, humans did not evolve the ability to discriminate and behave appropriately toward strangers with no possibility of future interactions. Not only is this claim implausible, based on what we know about the ancestral social environment (Fehr et al., 2002, Henrich, 2004), but it is undermined by our experimental results, which show that people are fully capable of discriminating information regarding genetic relatedness and potential for future interactions in the context of helping. Moreover, this particular version of the mismatch hypothesis cannot explain the pronounced individual differences observed in our experiments and previous research involving actual interactions. The mismatch hypothesis cannot be used in its general form to cast doubt on a specific result. A specific version of the mismatch hypothesis must be proposed that accounts for the experimental results in a testable fashion.

Third, our result might be explained in terms of adaptation and natural selection. Moral systems include some rules that are regarded as inviolable and others that are regarded as voluntary. The Ten Commandments provide examples of rules that are regarded as inviolable. The commandment “Thou shalt not steal” is not stated with a list of qualifiers such as “except from a stranger” or “except if you are about to leave the group.” It has an unconditional quality similar to the response of our participants to the cheating scenario. Giving to charity provides an example of a voluntary rule, which is regarded as morally praiseworthy but is not required. It is understood that such voluntary acts can be made at the discretion of the actor, similar to the conditional response of our participants to the helping scenario.

Obviously, the so-called inviolable rules are broken in ways that can even be morally justified, such as to avoid a greater moral evil. Nevertheless, the distinction is important, especially at the psychological level, and probably has functional significance as well. Groups whose members conceptualize at least some rules as unconditional and therefore punishable when broken are likely to function better than other groups. To proceed further, it is necessary to identify the suite of psychological traits that is required for the conceptualization and enforcement of such rules and to see if they can plausibly evolve by natural selection. Theoretical models devoted to this subject are more recent than the earlier theories of kin selection and reciprocal altruism (although the original paper of Trivers, 1971, on reciprocal altruism includes relevant verbal discussion), but they are just as important for understanding human evolution (Boyd et al., 2003, Boyd & Richerson, 1992, Gintis, 2000, Henrich & Boyd, 2001, Sober & Wilson, 1998, Wilson, 2002).

Empirically, Boehm, 1993, Boehm, 1999 has argued that modern hunter–gatherer societies are first and foremost moral communities, with a strong sense of right and wrong that applies to everyone, regardless of their status. This normative social environment enforced by punishment creates a degree of behavioral uniformity within groups and differences among groups (when they abide by different norms) that is highly favorable for the genetic and cultural evolution of within-group cooperation. It is precisely the traits associated with the conceptualization and enforcement of moral rules that expand the circle of cooperation beyond genetic relatives and narrow reciprocators, causing human evolution to embark upon a different trajectory than all other primate species.

To summarize, altruism can take place in the context of punishment in addition to helping behaviors. Altruistic punishment has been empirically demonstrated in humans in experiments involving actual interactions. Our experiments reveal a remarkable insensitivity to information regarding genetic relatedness, friendship, and potential for future interactions in the impulse to altruistically punish, in contrast to the impulse to altruistically help, at least in terms of the psychological response to fictional scenarios. This result is difficult to explain in terms of kin selection and narrow reciprocal altruism, but it is plausible in terms of the psychological traits associated with moral systems. Any human trait revealed by an experiment can potentially be explained as a methodological artifact, as a nonadaptive byproduct, or as an adaptation. We think that the phenomenon of altruistic punishment and its unconditional nature is most likely to be explained as part of a suite of psychological traits associated with moral systems that is very much a product of natural selection.