Do people use social information to improve predictions about everyday events?

Following Griffiths and Tenenbaum (2006), we explore whether people use relevant social information to improve their already nearly optimal predictions about quantities in everyday events. We tested this question in two experiments involving quantities in three domains: cake baking times, movie runtimes, and podcast lengths. In Experiment 1, we found that participants were sensitive to the difference between relevant and irrelevant social information. In Experiment 2, we found that people consistently used relevant social information to adjust their predictions in the expected directions. We introduce an optimal social prediction model but find that it does not consistently perform better at accounting for our participants' social predictions than an optimal non-social prediction model. We conclude by discussing whether people use social information for prediction in an optimal way.