Strong reciprocity is the behavioral predisposition to cooperate conditionally on others' cooperation and to punish violations of cooperative norms even at a net cost to the punisher (Fehr & Gintis Reference Fehr and Gintis2007). The phenomenon has been the subject of considerable research in the last few decades (e.g., Camerer Reference Camerer2003; Fehr & Gächter Reference Fehr and Gächter2000b; Rabin Reference Rabin1993), and its existence is well established.

In the target article, Guala points to a discrepancy between the strong negative reciprocity that is observed in the lab and the way cooperation is sustained “in the wild.” Specifically, he suggests that there is no indication for costly punishment in the wild. This claim gives rise to the question as to what extent one can predict actual behavior in real-life situations from behavior in the very artificial and contrived laboratory setting. The author suggests that behavior in the laboratory with respect to strong negative reciprocity does not really reflect behavior in real life. However, the matter may actually be somewhat more complex than it seems. Even if there is no strong negative reciprocity in the real world, this may still be in line with the results from laboratory studies. One simply has to make sure that the laboratory studies capture crucial characteristics of the real world.

One of the main properties of real-world situations is some degree of uncertainty (which does not usually exist in laboratory studies and particularly in those the author referred to). In many real-life social dilemmas people face uncertainty of two types: (1) environmental uncertainty, which is uncertainty regarding aspects of the dilemma (e.g., the size of the common resource); and (2) social uncertainty, which is uncertainty regarding the other group members' choices (Messick et al. 1998). Moreover, outcomes may be determined probabilistically.

For negative reciprocity to occur, accurate knowledge regarding the actions and the intentions of the players is important. If uncertainty exists, it will be difficult to determine whether the action and the outcome were the result of violations of cooperative norms. While most laboratory experiments have dealt with situations that are certain, a number of studies have introduced some degree of uncertainty into situations in which negative reciprocity is possible. These studies consistently show that uncertainty lowers the tendency towards negative reciprocity.

Most of the evidence for strong negative reciprocity was observed in the Ultimatum Game (UG). In research on the UG, responders are very likely to reject offers that are less than 30% of the cake (e.g., Güth et al. Reference Güth, Schmittberger and Schwarze1982). By rejecting the offer, the responder gives up a possible gain; and thus the finding is interpreted as evidence for responders' willingness to pay a cost to punish the proposers they perceive as acting unfairly, even if they will never meet the proposers again. In the classic experiment and in most following ones, the size of the pie to be divided is common knowledge. As was noted by Croson (Reference Croson1996), this assumption is unrealistic.

Several experiments investigated UGs with one-sided uncertainty on the part of the responder. Typically, in these experiments proposers know the exact amount of money to be divided, and responders either know nothing at all or know the probability distribution of possible amounts.

In most studies responders accepted lower offers when they did not know the size of the pie and when the lack of information was common knowledge. Proposers, in turn, did not hesitate to exploit this behavior and offered little when the amount to divide was large (e.g., Croson Reference Croson1996; Mitzkewitz & Nagel Reference Mitzkewitz and Nagel1993; Rapoport et al. Reference Rapoport, Sundali and Seale1996). Recently Gehrig et al. (Reference Gehrig, Güth, Levati, Levinsky, Ockenfels and Uske2007) studied a UG with a different source of uncertainty. In their game the responder knows the pie size but not the offer when deciding whether to accept or reject (i.e., has imperfect information). Responders never reject in this game, even when they anticipate low offers.

Under both types of uncertainty responders seem to give proposers the benefit of the doubt: Because a low offer could be fair if the pie is small or the yet-unknown offer could eventually be fair, rejecting the offer would mean punishing the proposer unfairly. Consequently, with uncertainty lower offers are more likely to be accepted. This behavior is strong evidence that rejections in the UG are an expression of preference when responders do know the proposer's payoff (Camerer Reference Camerer2003); and therefore the ability to generalize these preferences to situations with uncertainty is limited.

Uncertainty also affects reciprocity in repeated interactions. The ability of reciprocity to sustain cooperation in the long run, and specifically in the iterated Prisoner's Dilemma, was demonstrated by Axelrod's (Reference Axelrod1984) well-known computer tournaments. Later it has been shown that cooperation is much more difficult to maintain if there is uncertainty regarding players' actions, that is, if there is random error either in choosing actions or in monitoring others' actions (see, e.g., Axelrod & Dion Reference Axelrod and Dion1988; Bendor Reference Bendor1993; Green & Porter Reference Green and Porter1984; Sainty Reference Sainty1999). That is, if actions are noisy, a player does not know whether another player's defection was an error or an intended choice, and strategies involving reciprocation (e.g., tit for tat) can break down. But even if players can monitor others' past actions perfectly in a repeated Prisoner's Dilemma game, if payoffs are noisy, players learn to cooperate much less (e.g., Bereby-Meyer & Roth Reference Bereby-Meyer and Roth2006; Kunreuther et al. Reference Kunreuther, Silvasi, Bradlow and Small2009).

Hence, the fact that Guala in his analysis of real-world situations did not find evidence for strong negative reciprocity does not necessarily imply that results from laboratory studies cannot predict reciprocity behavior in the real world. Instead, one can perhaps conclude that in situations with uncertainty, costly punishment is less likely to occur, and therefore in these situations probably other mechanisms are needed to sustain cooperation.