Conventions are customary and arbitrary rules of behaviour that coordinate our interactions with others if all parties share an interest to conform (Lewis 1969). When conformity runs counter to individual self-interest, interactions may be regulated by social norms, which are endowed with prescriptive pressure and supported by sanctions (Cialdini et al 1990, Horne, 2001, Bicchieri 2006). Conventions and social norms have been kept distinct along these lines. However, both can emerge tacitly, as unintended consequences of individual actions (Sugden 1986, Young 2015, Centola and Baronchelli 2015), and what originates as a mere convention may, over time, acquire a prescriptive force and turn into a social norm (Opp 2004, Tummolini et al 2013). Although there have been first attempts (Guala 2013, Diekmann and Przepiorka 2016), no study so far has systematically measured the normativity of emerging conventions.

Here we induce and measure the emergence of social norms from conventions in the repeated volunteer’s dilemma (Diekmann 1985, 1993). The volunteer’s dilemma (VOD) is a binary choice, n-person game in which a single actor’s cooperation is necessary and sufficient to provide the collective good for the entire group. Previous research has shown that an ‘egalitarian’ convention, in which each person sequentially incurs the cost of volunteering (turn taking), can emerge in the symmetric VOD, in which the cost of volunteering is the same for everyone. In an asymmetric VOD, in which one person has a lower cost of volunteering, an ‘exploitative’ regularity often emerges; the person with the lower cost takes a disproportionately large share of volunteering (strongest-always-volunteer) (Diekmann & Przepiorka, 2016).

Building on these results, we conduct a two-part experiment in which subjects start in either the symmetric or asymmetric VOD and then move to the other VOD in the second part. Drawing on the work of Bicchieri and colleagues (Bicchieri, Lindemans, & Jiang, 2014), we employ a set of incentivised measures to capture when conventions turn into social norms. Our design allows us to explore (1) whether conventions such as turn-taking or the strongest-always-volunteer turn into social norms and (2) whether these social norms make the prescribed behaviour ‘sticky’ in face of structural changes (i.e. when subjects move from the symmetric to the asymmetric VOD or vice versa). We also measure social approval providing us with additional information on the normativity of the conventions that emerge.

Our results show that groups with a higher consensus on which convention should be followed, turn-taking or strongest-always-volunteers, take longer to change their behaviour when they move to the VOD with an alternative payoff structure. However, the exploitative norm is less stable than the egalitarian one. Although the stability of the egalitarian norm is also affected by changes in incentives, it is more likely to persist in an environment that would be otherwise conducive to unequal division and exploitation. Our analysis of approval scores reveals that once established even exploitative norms are perceived as legitimate, suggesting that normative judgments can swiftly shift in support of the social order that emerges due to changes in incentives.