In explaining costly punishment in general, Guala's article focuses on (cultural) group selection, and subsequently “strong reciprocity.” But other mechanisms can also lead to punishment and which might only appear to be altruistic. These mechanisms include: kin selection; mutualism; pseudo-reciprocity, and direct reciprocity (see overview in Bshary & Bergmüller Reference Bshary and Bergmüller2008). There is also general reputation-based cooperation, where “helpful and harmful acts [are returned] in kind” – even by third-parties (Nowak & Sigmund Reference Nowak and Sigmund2005), and/or where net-benefits accrue to both sender and receiver via the logic of costly signaling theory (see Barclay & Willer Reference Barclay and Willer2007; Nelissen Reference Nelissen2008). In this commentary, I concentrate on such general reputation-based cooperation.

Within a reputation-based account, other-benefiting costs are reimbursed by an increase in that individual's reputation – which is beneficial in the long run. Imagine an agent acts as an altruistic punisher. In the long run, she may benefit since her punishing will increase her reputation, which in turn will render it more likely that she will benefit through the actions and choices of others.

In general, a good reputation can increase the likelihood of receiving future help (Wedekind & Braithwaite Reference Wedekind and Braithwaite2002) and of gaining effective partners (Sylwester & Roberts Reference Sylwester and Roberts2010) – while it decreases the likelihood of punishment (“bad” free-riders are punished) and shunning (Panchanathan & Boyd Reference Panchanathan and Boyd2004). Push and pull.

Importantly, reputation might be established through several routes (Russell et al. Reference Russell, Call and Dunbar2008): by direct interaction, by eavesdropping (e.g., A observes B and C interact), and by reporting (gossip, e.g., A later reports to D). It is possible that reputation-based cooperation may be sensitive only to the amount of the cost paid (Nelissen Reference Nelissen2008) but not to the form of the cost (e.g., helping vs. punishing; cf. Semmann et al. Reference Semmann, Krambeck and Milinski2004). However, a reputation for being a punisher may also bring special (or additional) reputational benefits based on the threat developed through past punishment (dos Santos et al. Reference dos Santos, Rankin and Wedekind2011).

If the lab setup allows for reputation to be formed or passed on, then this (positively) influences general group beneficial behavior (Milinski et al. Reference Milinski, Semmann and Krambeck2002; Wedekind & Milinski Reference Wedekind and Milinski2000). Reputation then also allows for individual benefits gained for punishment (Barclay Reference Barclay2006; Nelissen Reference Nelissen2008); and punishing also increases with larger audiences (Kurzban et al. Reference Kurzban, DeScioli and O'Brien2007). Thus, reputation is an important factor in such games. But how does reputation-based cooperation hold up in the light of punishing in anonymous lab situations? Following and extending others' work, I suggest that reputational concerns may still be key here.

Any perception of anonymity can be wrong: some audience may be observing after all (Frank Reference Frank1988). Thus, either the perceived anonymity is real (“factual anonymity,” i.e., no audience) or not (“false anonymity,” i.e., some audience). Benefits in false anonymity are standard benefits in the reputation system: increased reputation (e.g., Barclay Reference Barclay2006). An actor should thus estimate the likelihood of the perceived anonymity being factually false, and should subsequently be more likely to punish the likelier it seems that some audience is present after all. As long as there is an overall net-gain across situations, this approach should work. But, of course, such an estimate will sometimes be wrong, and the actor may thus sometimes punish even in factual anonymity. Some have proposed that this explains punishment in anonymous lab settings, that is, that it may be ultimately due to a somewhat fixed, but outdated, adaptation of the past (and possibly helped by a general rarity of factual anonymous situations in the past). All this then falls under the label of the “Big Mistake hypothesis” (Richerson & Boyd Reference Richerson and Boyd2005) – an evolutionary logic embraced by several authors (Burnham & Johnson Reference Burnham and Johnson2005; Hilbe & Sigmund Reference Hilbe and Sigmund2010; Tennie et al. Reference Tennie, Frith and Frith2010; Trivers Reference Trivers2004). In addition, subjects may still come to regard the test situation as a falsely anonymous one today (a “nagging suspicion” may remain; see Nowak & Sigmund Reference Nowak and Sigmund2005). In both cases, punishment in factual anonymity may still be due to intended self-interest in a reputation based system.

These explanations may still assume that any act of punishment in factual anonymity must always be associated with a net-cost for a reputation-based cooperator. But there might even be a net-benefit. As already mentioned, reputation can be passed on in the absence of audiences – through gossip. In this way even a punisher observed by no one may later benefit – namely, if the act of punishment enables him to later convince others that he indeed punished. This may be possible: his “bragging” will likely be more convincing if in fact he did punish, because (at least some) listeners can detect lies (Ekman & O'Sullivan Reference Ekman and O'Sullivan1991). Additionally, it is cognitively less costly to tell the truth (Paglieri Reference Paglieri, Hansen, Tindale, Johnson and Blair2007), and these costs may help outweigh the costs of the punishment. Finally (depending on the sort of punishment), others may observe the effects of his punishment and may therefore be able to independently validate his reports. Granted, in modern economic games these effects may be negligible, but the general Big Mistake logic might apply also here.

The scenario just described may explain the general presence of punishment in anonymous lab situations, but what about the apparently excessive frequency of punishment in the lab reported by Guala? Guala points out that punishment is often the main way in which participants can influence others' behavior in such economic games. The reason why we do not see equally high levels of punishment under field conditions may therefore simply be that real life offers more behavioral options that can influence others' behavior than the lab usually does. In addition, punishment seems to be a less preferred option in general (maybe even a “last resort”) – Guala mentions the fear and cost of retaliation (see also Barclay Reference Barclay2006; Hill et al. Reference Hill, Barton and Hurtado2009). In sum, special conditions in the lab (lack of choice; regardless of preference) in comparison to the field (availability and preference for other behavioral options) could explain the discrepant findings of punishment frequencies reported in Guala's article.