Belief revision in a micro-social network: Modeling sensitivity to statistical dependencies in social learning

Both in professional domains and everyday life, people frequently rely on their 'social network neighbors' to form and update their beliefs. Here, it is important to understand how people deal with statistical dependencies underlying correlated beliefs in their social environment. Using an interface allowing us to elicit full probabilistic beliefs from people, we investigated people's ability to distinguish between the evidential value of social information across three conditions: integrating independent beliefs, dependent beliefs formed on the basis of the shared evidence, and dependent beliefs that result from sequential communication between sources. Comparing participants' judgments to a normative Bayesian model, we found that they distinguished dependent from independent sources but treated social sources as much weaker sources of evidence than direct experience. The value of eliciting and visualising beliefs as full probability distributions and potential implications for modeling belief revision in social networks (e.g., using agent-based models of echo chambers) are discussed.