What is the extent of a frequency-dependent social learning strategy space?

Abstract

Models of frequency-dependent social learning posit that individuals respond to the commonality of behaviours without additional variables modifying this. Such strategies bring important trade-offs, e.g. conformity is beneficial when observing people facing the same task but harmful when observing those facing a different task. Instead of rigidly responding to frequencies, however, social learners might modulate their response given additional information. To see, we ran an incentivised experiment where participants played either a game against nature or a coordination game. There were three types of information: (a) choice frequencies in a group of demonstrators; (b) an indication of whether these demonstrators learned in a similar or different environment; and (c) an indication about the reliability of this similarity information. Similarity information was either reliably correct, uninformative or reliably incorrect, where reliably correct and reliably incorrect treatments provided participants with equivalent earning opportunities. Participants adjusted their decision-making to all three types of information. Adjustments, however, were asymmetric, with participants doing especially well when conforming to demonstrators who were reliably similar to them. The overall response, however, was more fluid and complex than this one case. This flexibility should attenuate the trade-offs commonly assumed to shape the evolution of frequency-dependent social learning strategies.

Keywords: Social learning; conformity; cultural evolution; frequency-dependent social learning; social norms.

Graphical abstract

Figure 1.

The payoff matrix shown to participants for the game against nature. Text in bold represents the expected payoffs for the focal participant's choices.

Figure 2.

The payoff matrix shown to participants for the coordination game. Text in bold represents the expected payoffs from the focal participant's choices.

Figure 3.

A typical round for the social learners. Type A participants were demonstrators and Type B participants were social learners. We avoided the term ‘demonstrator’ or ‘social learner’ in case it led the participants to respond in certain ways. The top half of the screen reminds the participants of the expected pay-offs from Game Left and Game Right (for the coordination game in this case). The bottom half of the screen contains frequency-dependent information (i.e. the number of demonstrators who chose @ or %), similarity information (i.e. whether the demonstrators were identified as playing the same or different game version to the social learners) and the reliability information.

Figure 4.

The proportion of social learners choosing % based on the number of demonstrators choosing %. The panels show the social learners’ frequency-dependent social learning strategies for each level of the second- and third-order social information, for both the game against nature (learning skills, in red) and the coordination game (learning social norms, in blue). The error bars give the 95% bootstrapped confidence interval clustered on social learners, to reflect the multiple observations gathered per learner. The regions shaded in grey depict where the social learners’ data would fall if they conformed, whilst the dashed lines give points of reference for proportions of learners choosing % at 0, 0.5, and 1.