When explaining behavioral decisions, it is important to distinguish between different explanatory levels, especially between proximate (mechanistic) and ultimate (evolutionary) explanations (Tinbergen Reference Tinbergen1963). Proximate explanations of a given behavior deal with questions about its ontogeny (e.g., how does the behavior change with age and experience) or about its causation, that is, the physiological, molecular, and cognitive mechanisms underlying the behavior and the stimuli that elicit it. Ultimate explanations either deal with questions about the phylogeny of the behavior (e.g., how does it compare with similar behaviors in related species) or its adaptive value (e.g., what is its impact on the individual's survival and lifetime reproductive success).

The concept of weak reciprocity, as defined in the target article, is an attempt to explain the adaptive value of cooperation and punishment because it concentrates on the fitness benefits one could get from cooperating, defecting, or punishing (Alexander Reference Alexander1974; Trivers Reference Trivers1971). This concept is restricted to one explanatory level only. In contrast, strong reciprocity mixes different explanatory levels: it uses proximate arguments to explain ultimate problems (Bowles & Gintis Reference Bowles and Gintis2004; Fehr & Fischbacher Reference Fehr and Fischbacher2003; Reference Fehr and Fischbacher2004; Fehr & Gächter Reference Fehr and Gächter2002; Fehr & Rockenbach Reference Fehr and Rockenbach2003; Gintis et al. Reference Gintis, Bowles, Boyd and Fehr2003). Strong reciprocity has been called, for example, a “predisposition to reward others for cooperative, norm-abiding behaviours” and “a propensity to impose sanctions on others for norm violations” (Fehr & Fischbacher Reference Fehr and Fischbacher2003, p. 785). Such a definition clearly relates to the causal mechanisms of cooperation and punishment. But the concept is then frequently used as to answer ultimate (evolutionary) questions, for example, in Bowles and Gintis (Reference Bowles and Gintis2004, p. 17): “cooperation is maintained because many humans have a predisposition to punish those who violate group-beneficial norms.” Such a mixing up of different explanatory levels can, from an evolutionary point of view, easily lead to over-interpretations of proximate patterns (Hagen & Hammerstein Reference Hagen and Hammerstein2006; Rankin et al. Reference Rankin, dos and Wedekind2009; Sigmund Reference Sigmund2007; West et al. Reference West, Griffin and Gardner2007a; Reference West, El and Gardner2011). For example, punishment that can be observed in anonymous one-shot interactions seems truly altruistic and was interpreted as such in Fehr and Gächter (Reference Fehr and Gächter2002). However, until very recently, humans lived in groups where anonymous one-shot interactions were probably very rare, that is, such interactions are most probably not the context in which human punishment has evolved. If studied within a more natural social context, human punishment may ultimately be self-interested.

As discussed in the target article, explaining punishment from an evolutionary point of view requires determining the costs and benefits of punishment. In line with weak reciprocity models, recent studies have shown that punishment can lead to long-term net benefits and hence be evolutionarily stable when punitive actions contribute to a punishment reputation (dos Santos et al. Reference dos Santos, Rankin and Wedekind2011; Hilbe & Sigmund Reference Hilbe and Sigmund2010). Under such conditions, the immediate costs of punishment can be outweighed by the benefits a punisher receives later because of his or her punishment reputation. Experimental studies that ignore the possible effects of a punishment reputation can therefore easily produce artifacts (Hagen & Hammerstein Reference Hagen and Hammerstein2006).

We also believe that the term “punishment” is currently used too broadly in the literature on cooperation. If punishment is the subtraction of resources from free- riders in order to reduce the frequency of further free-riding, there are at least three different kinds of punishment that may need to be distinguished both for ultimate and proximate analyses. Many of these analyses deal with what could be called “simple costly punishment”; that is, punishers pay a cost to induce a cost on the punished (Dreber et al. Reference Dreber, Rand, Fudenberg and Nowak2008; Fehr & Gächter Reference Fehr and Gächter2000a; Rand et al. Reference Rand, Dreber, Ellingsen, Fudenberg and Nowak2009a; Rockenbach & Milinski Reference Rockenbach and Milinski2006; Wu et al. Reference Wu, Zhang, Zhou, He, Zheng, Cressman and Tao2009).

Another form of punishment could be called “punishment by taking something away” (e.g., as in Cephu's story, described in the target article). Here, the punisher takes something from the punished in order to induce a cost to the punished. Regardless of whether or not the punisher thereby experiences an immediate reduction of his or her own welfare, “punishment by taking something away” and the above-mentioned “simple costly punishment” are likely to differ in their cost-benefit ratios (relevant for ultimate analyses) and may involve, for example, different kinds of emotions (relevant for proximate analyses).

A third category could be called “punishment by refusal.” The punisher then punishes by refusing to cooperate with the punished in a repeated game like, for example, an iterated Prisoner's Dilemma (Fudenberg et al. Reference Fudenberg, Levine and Maskin1994). The examples of ostracism discussed by Guala relate to this kind of punishment. Such defection may typically be a reaction to non-provoked defection and could be called “punishment” if it reduces the income of the punished (i.e., his or her benefits from what would otherwise be cooperative interactions) in order to possibly improve the punisher's long-term benefits from future cooperative interactions with a refined punished or with others.

This third kind of punishment could be immediately costly for the punisher, for example, if it delays the resumption of beneficial mutual cooperation. Such immediate costs would have to be compensated in the long run in order to maintain “punishment by refusal” as an evolutionary successful behavioral strategy. However, a possible alternative function of defection in response to defection may be to simply avoid the losses of anticipated further defection (e.g., avoiding the sucker's payoff in the Prisoner's Dilemma). It is probably not useful to call this latter form of defection “punishment” if it usually does not ultimately increase the level of cooperation within a group or directly with the defector (from an ultimate point of view), or if it is just a precautionary measure to avoid further losses (from a proximate point of view). Therefore, purely punitive actions may not always be easy to identify. Multidisciplinary approaches that carefully exploit the specific advantages of proximate and ultimate analyses are therefore often necessary to better understand human behavior.