Punishment may have repercussions leading to a less favorable equilibrium with lower payoffs

In evolutionary ecology, one distinguishes between the direct and the ecological costs of a trait. For example, Strauss et al. (Reference Strauss, Rudgers, Lau and Irwin2002) discuss the costs of resistance to herbivory in plants. Direct costs of herbivory resistance can mostly be described in terms of resource allocation; resources allocated to defense mechanisms cannot be allocated to growth or reproduction. The ecological costs of herbivory resistance are more long-term and indirect; examples are decreased attractiveness to pollinators or decreased competitive ability. The costs associated with punishment mechanisms such as ostracism may be distinguishable in a similar way. We agree with Guala that the direct short-term costs associated with ostracizing free-riders will often be low. However, on the longer term, there can be strong negative implications. Ostracized individuals may become desperados, causing a lot of trouble. They may resort to antisocial or criminal behavior, affecting the feeling of safety in their former group and necessitating protection measures. In the worst case, trust and cooperation break down. This way, the presence of ostracized individuals in the environment can lead to a new equilibrium with lower payoff levels than in the original state. Although there have been some experiments that include ostracism as an option (e.g., Maier-Rigaud et al. Reference Maier-Rigaud, Martinsson and Staffiero2009; Masclet Reference Masclet2003), they do not accommodate those “ecological” costs.

Punishment in one type of interaction may have implications for different types of interaction

Economic experiments typically focus on a single type of interaction, such as a public goods game. If punishment is incorporated in these experiments, it can only affect behavior in that specific context. This is not in line with how behavior is structured in humans (and other animals). There is ample evidence that behavioral tendencies in one type of interaction are closely correlated with the behavior in quite different contexts. As shown by evolutionary models from the biological literature (e.g., Wolf et al. Reference Wolf, van Doorn, Leimar and Weissing2007; Reference Wolf, van Doorn and Weissing2008), such correlation structures (called “behavioral syndromes” or “personalities”) can be adaptive, even if the behavior in a particular type of situation may appear maladapted. For example, the tendency to show antisocial behavior in a public goods context may – for good reasons – be correlated with the tendency to actively participate in group defense when the group is facing an external challenge. Ostracizing individuals because of their behavior in a public goods context may therefore have harmful effects later.

Punishment may destroy established hierarchies and role patterns and lead to social unrest

The participants of a typical economic experiment do not know each other well and interact anonymously. In real life, many interactions take place in small communities where individuals do know each other, and are well aware of their place in the group. Individuals differ in relevant aspects (like age, expertise, or authority), and relationships between individuals (like leadership and social rank) have been settled in the past. Such patterning of a group due to well-established relationships between its members is important, because it reduces conflict and facilitates division of labor. Punishing an individual by social exclusion can break down such group structures, leading to social unrest. The re-establishment of stable social relationships can take a long time, and some individuals may end up in a worse position than they had before. Guala himself refers repeatedly to the work of Ostrom (Reference Ostrom1990), who has shown that stable group membership is one of the key predicting features making institutions for collective actions viable.

Punishment may have asymmetric effects, thus leading to tension between group members

Interactions in economic experiments are usually random. In contrast, real-world interactions take place in interaction networks that are often highly structured. This can be important, because group members may differ considerably in the way they are connected to a punished individual. Individuals will differ not only in the degree they suffer from the free-riding behavior of a specific individual, but also in the implications that punishment of that individual may have for them. Ostracizing an individual may have a small effect on group member A, while it severely affects the social network of group member B. The costs and benefits of punishing a particular free-rider can therefore be highly asymmetrical, leading to contrasting preferences between group members and, as a consequence, to social tension within the group.

If punishment were as cheap as Guala suggests, one would expect that individuals would readily punish defectors. In contrast, daily-life experience tells us that individuals are reluctant to punish free-riding group members. Denouncing others is often considered a bad habit, even if these others exhibit antisocial behavior. Groups of students assigned to a joint project, for example, are typically not only reluctant to punish free-riders, but even to call in an authoritative person (such as a professor) to resolve the situation. In fact, whistle-blowing is considered more a vice than a virtue, as young children are already being told by their parents or at school. This reluctance to apply seemingly cheap punishment is an indication that the hidden, long-term costs of punishment may be substantial. Economic experiments focusing exclusively on the direct costs of punishment are valuable, but they do not tell much about how cooperation is stabilized in human societies. For a complete understanding, the social costs of punishment should be taken seriously.