Timing of social feedback shapes observational learning in strategic interaction

Joshua Zonca,Alexander Vostroknutov,Giorgio Coricelli,Luca Polonio

doi:10.1038/s41598-021-01466-1

Abstract

Many types of social interaction require the ability to anticipate others' behavior, which is commonly referred to as strategic sophistication. In this context, observational learning can represent a decisive tool for behavioral adaptation. However, little is known on whether and when individuals learn from observation in interactive settings. In the current study, 321 participants played one-shot interactive games and, at a given time along the experiment, they could observe the choices of an overtly efficient player. This social feedback could be provided before or after the participant’s choice in each game. Results reveal that players with a sufficient level of strategic skills increased their level of sophistication only when the social feedback was provided after their choices, whereas they relied on blind imitation when they received feedback before their decision. Conversely, less sophisticated players did not increase their level of sophistication, regardless of the type of social feedback. Our findings disclose the interplay between endogenous and exogenous factors modulating observational learning in strategic interaction.

Highlights

Have the possibility to select between these two learning strategies, they do so depending on their capabilities, cognitive costs and contextual g oals[20]
Our study aims at investigating if individuals can improve their level of strategic sophistication by observing a successful agent and whether the timing of social feedback can shape the emergence of sophisticated and generalized learning
These results indicate that the Nash equilibrium model alone is a poor predictor of participants’ strategic behavior in this type of games

Summary

Introduction

Have the possibility to select between these two learning strategies, they do so depending on their capabilities, cognitive costs and contextual g oals[20]. The strategy used by the artificial agent is the one of an unsophisticated level-1 player: this agent selects the action with the highest average payoff for itself, assuming the counterpart to play randomly (See “Games and artificial agent behavior” paragraph in the “Methods” section for a detailed description of the games used in the experiment and the artificial agent’s behavior) Another important aspect of the Assessment phase is that no feedback on the game outcomes was provided to participants: they could not learn from the behaviour of the artificial agent. Participants in the Postfeedback treatment received the feedback about the model’s decision only after they made their decision and could not modify their choice based on the model’s feedback

Methods

Results

Conclusion