In economic models, it is usually assumed that agents aggregate their private information with all available public information correctly and completely. In this experiment, we identify subjects’ updating procedures and analyse the consequences for the aggregation process. Decisions can be based on private information with known quality and on the observed decisions of other participants. In this setting with random ordering, information cascades are observable and agents’ overconfidence has a positive effect on avoiding a non-revealing aggregation process. However, overconfidence reduces welfare in general.