Effects of social context on reward-based learning


Objective:People readily learn associations between decisions and the outcomes that follow them through instrumental learning. How does social context affect the speed and fidelity with which people learn the likely consequences of their actions? Here, we integrate approaches from instrumental learning and social decision-making to investigate how reward distributions (across oneself and another person) and others’ identity (with whom rewards are shared) influence how people learn about rewards. Methods:We recruited 94 online participants to perform a computerized instrumental learning task in which the monetary rewards obtained on each trial would ostensibly be split between themselves and another person (e.g., “Nurse”; “Lawyer”). On each trial in a block, one of five possible images appeared on the screen, along with information about how the reward would be split (e.g., you:30%; Nurse:70%); participants were obliged to quickly press 1 of 3 keys, which yielded either $0, $1, or $2 (before splitting). We varied the split percentage to create either disadvantageous inequality (participant gets the smaller share) or advantageous inequality on each trial. Following the task, subjects rated the warmth and competence of each partner. We embodied two hypotheses into computational models. The naive reinforcement learning (Naive) model assumes the mind represents only the total, objective reward. The social perception reinforcement learning model (SPRL) assumes that monetary inequity and social perception dynamically shape value representations during learning. Results:Mixed-effect linear regressions with subject ID as the random intercept revealed that people learn faster when they gain a larger share of the reward, indicating an effect of inequity on learning (p = 0.02). Regression further showed that the effect of both inequity and social perception is stronger in the later stage of learning (ps < 0.05). Critically, we then compared our computational models using WAIC and discovered that SPRL outperformed Naive. Further, we saw SPRL outperformed more in the disadvantageous inequity condition (p < 0.01), suggesting that the social context has a larger effect on learning when subjects obtain a smaller share of rewards. Conclusion:These results suggest inequality and social perception information influence reward-based learning. These effects are especially prominent in the later stage of learning and when subjects suffer disadvantageous inequity.

In Poster presented at The Society for Neuroeconomics Conference, Arlington, VA