Guala raises important questions about the misinterpretation of experimental studies that have found that subjects engage in costly punishment. Instead of positing that punishment is the solution for social dilemmas, earlier research posited that when individuals facing a social dilemma agreed on their own rules and used graduated sanctions, they were more likely to have robust solutions over time.

Claims regarding negative strong reciprocity do indeed rest on experiments lacking established external validity, often without even a small “menu of options.” Guala's review should prompt strong reciprocity proponents to extend the real-world validity of their work, exploring the preferences participants bring to experiments. That said, Guala's approach fails to differentiate among group selection approaches and glosses over cross-cultural variability.

Guala does not go far enough in his critique of the assumption that human decisions about sharing made in the context of experimental game conditions accurately reflect decision-making under real conditions. Sharing of hunted animals is constrained by cultural rules and is not “spontaneous cooperation” as assumed in models of weak and strong reciprocity. Missing in these models is the cultural basis of sharing that makes it a group property rather than an individual one.

In a unified theory of human reciprocity, the strong and weak forms are similar because neither is biologically altruistic and both require normative motivation to support cooperation. However, strong reciprocity is necessary to support cooperation in public goods games. It involves inflicting costs on defectors; and though the costs for punishers are recouped, recouping costs requires complex institutions that would not have emerged if weak reciprocity had been enough.

Guala notes that low-cost punishment is the main mechanism that deters free-riding in small human communities. This mechanism is complemented by unusual human vulnerability to gossip. Defenders of an evolutionary discontinuity supporting human sociality might seize on this as an alternative to enjoyment of moralistic aggression as a special adaptation. However, the more basic adaptation of language likely suffices.

Guala is right to draw attention to the difficulty of extrapolating from the experimental evidence for weak or strong reciprocity to what is observed in the “wild.” However, there may be more strong reciprocity in real-world communities than he allows for, as strikingly illustrated in the example of the Mafia.

Strong reciprocity is not the only account that can explain costly punishment in the lab; it can also be explained by reputation-based accounts. We discuss these two accounts and suggest what kinds of evidence would support the two different alternatives. We conclude that the current evidence favors a reputation-based account of costly punishment.

The discovery of the altruistic punishment mechanism as a replicable experimental result is a genuine achievement of behavioural economics. The hypothesis that cooperation in hunter-gatherer societies is sustained by altruistic punishment is a scientifically legitimate conjecture, but it must be tested against real-world observations. Guala's doubts about the evidential support for this hypothesis are well founded.

Contrary to Guala, I claim that several mechanisms can explain punishment in humans. Here I focus on reputation-based cooperation – and I explore how it can lead to punishment under situations that may or may not be perceived as being anonymous. Additionally, no particular mechanism stands out in predicting an excess of punishment under constrained lab conditions.

This commentary seeks to clarify the potential discrepancy between lab-based and field data in the use and effectiveness of punishment to promote cooperation by recommending theory that outlines key differences between the lab and field, such as the shadow of the future and degree of information availability. We also discuss a recent meta-analysis (Balliet et al. 2011) that does not support all conclusions outlined in Guala's target article.

Lab experiments on punishment are of limited relevance for understanding cooperative behavior in the real world. In real interactions, punishment is not cheap, but the costs of punishment are of a different nature than in experiments. They do not correspond to direct payments or payoff deductions, but they arise from the repercussions punishment has on social networks and future interactions.

In small-scale societies, punishment of adults is infrequent and employed when the anticipated cost-to-benefit ratio is low, such as when punishment is collectively justified and administered. In addition, benefits may exceed costs when punishers have relatively greater physical and social capital and gain more from cooperation. We provide examples from the Tsimane horticulturalists of Bolivia to support our claims.

To add ethnographic perspective to Guala's arguments, I suggest reasons why experimental and ethnographic evidence do not concur and highlight some difficulties in measuring whether positive and negative reciprocity are indeed costly. I suggest that institutions to reduce the costs of maintaining cooperation are not limited to complex societies.

I argue in my target article that field evidence does not support the costly punishment hypothesis. Some commentators object to my reading of the evidence, while others agree that evidence in favour of costly punishment is scant. Most importantly, no rigorous measurement of cost-benefit ratios in the field has been attempted so far. This lack of evidence does not rule out costly punishment as a cause of human cooperation, but it does pre-empt some overconfident claims made in the past. Other commentators have interpreted my article as an anti-experimental pamphlet or as a flat denial of the existence of pro-social motives – which it was not intended to be. While we have enough data to establish the existence (and theoretical relevance) of strong reciprocity motives, I argue in this response that their efficacy (and policy relevance) has not been demonstrated.