The Role of Memory in Belief Formation

A growing body of research has documented both systematic and motivated belief biases in decision making such as correlation neglect and wishful thinking. A better understanding of how people aggregate signals and form beliefs in the first place is crucial for understanding the origin of these biases and potential policy interventions. In this paper, we explore the role of memory in belief formation. We design a simple experiment where subjects read a set of news sources which report naturalistic positive and negative facts about an artificial company. We provide enough information to subjects about the fact-generating process so that they can form a Bayesian posterior on whether the company is 'good' or 'bad'. We then (a) elicit subjects' beliefs about the quality of company based on these narratives and (b) present subjects with a superset of facts to determine whether they recognize previously seen facts. Just like real newspapers, facts can be repeated across several news sources. Unlike most of the existing literature on correlation neglect our experiment introduces repetition through actually repeated signals which we argue is more natural than merely correlated signals. Our design therefore allows us to explore the role of recognition and recall in belief formation which are two fundamental concepts in cognitive psychology. We find that subjects exhibit good recognition which is only slightly increasing by repetition of the signal. Subjects slightly overweigh repeated signals (beyond what is accounted for by the greater recognition) but do not fully double-count. Our framework can help to understand how individual (single-agent) learning differs from social learning.