Social norms are ubiquitous in human life. Their role is essential in allowing cooperation to prevail, despite the presence of incentives to free ride. As far as norm enforcement devices are concerned, it would be impossible to have widespread social norms if second parties only enforced them. However, both the quantitative relevance and the motivations underlying altruistic punishment on the part of ‘unaffected’ third parties are still largely unexplored. This paper contributes to shed light on the issue, by means of an experimental design consisting of three treatments: a Dictator Game Treatment, a Third-Party Punishment Game Treatment (Fehr and Fischbacher, 2004) and a Metanorm Treatment, that is a variant of the Third-party Punishment Game where the Recipient can punish the third party. We find that third parties are willing to punish dictators (Fehr and Fischbacher, 2004; Ottone, 2008) and, in doing so, they are affected by ‘reference-dependent fairness’, rather than by the ‘egalitarian distribution norm’. By eliciting players’ normative expectations, it turns out that all of them expect a Dictator to transfer something – not half of the endowment. Consequently, the Observers’ levels of punishment are sensitive to their subjective sense of fairness. A positive relation between the level of punishment and the degree of negative subjective unfairness emerges. Subjective unfairness also affects Dictators’ behaviour: their actual transfers and their ideal transfer are not significantly different. Finally, we interestingly find that third parties are also sensitive to the receivers’ (credible) threat to punish them: as the Dictator’s transfer becomes lower and lower than the Observer’s ideal transfer, the Observer’s reaction is – other things being equal – significantly stronger in the Metanorm Treatment than in the Third-Party Punishment Game Treatment. Hence, despite their being to some extent genuinely nonstrategically motivated, also third parties – like second parties – are sensitive to the costs of punishing.
