Reward Learning Positively Biases Widespread Impressions and Social Choice

People learn about their social world by inferring traits others possess (e.g., “generous”), allowing them to flexibly predict others’ behavior in new settings. However, people also experience rewarding outcomes in social interactions (e.g., “I got a gift”). Rewards feel good and may cast a positive glow on others, apart from specific traits their behavior implies. We tested whether people apply these two types of social learning differently to new circumstances. Specifically, we hypothesized that rewards lead to general positivity toward others across distinct judgments and choices, whereas trait inferences exert more flexible influence. In four experiments, participants learned about others in a sharing game. These individuals independently varied in how rewarding they were to the participant (the absolute amount of money they provided) and how generous they were (the proportion of available money they shared). Participants made decisions across many games, hypothetical scenarios, and trait ratings that varied in relevance to the sharing game. Participants used trait feedback selectively: they preferred generous partners more when settings were more relevant to generosity. In contrast, participants preferred previously rewarding partners to a similar degree across a range of scenarios and rated them positively to a similar degree across various traits. These results suggest a small, consistent influence of rewarding outcomes on representations of others, compared to trait learning’s more flexible influence. Beyond reinforcing behaviors, rewards lead people to consistently ascribe positive attributes to and feel drawn toward rewarding individuals, whereas traits convey specific meaning that people apply based on relevance.