Social Learning in Groups: an Experimental StudyMarina Agranov
10. Juni 2021
Marina Agranov (California Institute of Technology, https://agranov.caltech.edu/) presents her project "Social Learning in Groups: an Experimental Study" (joint with Gabriel Lopez-Moctezuma, Philipp Strack, and Omer Tamuz) in the Microeconomics Research Seminar via Zoom on Thursday, June 10, 17:15-18:45. You may write to huyen.nguyen-1"AT"uni-hamburg.de or gero.henseler"AT"uni-hamburg.de to register for the seminar (getting the Zoom access data).
Abstract: A large literature in economics and finance has shown the dangers associated with — and the inefficiencies that arise from — the imitation of others’ actions, and from herd behavior in particular. We contribute to this literature by providing empirical evidence of the benefits of imitation in repeated social learning environments. We show experimentally that observing the actions of others improves the quality of decisions and leads to higher information aggregation rates, even in settings where the actions of others do not provide additional information, beyond the private information available to agents.
Specifically, we conduct a series of controlled laboratory experiments, in which subjects repeatedly try to estimate an unknown state of nature while observing private information that arrives sequentially, and, depending on the treatment, additional observations of others. There are no direct externalities in this setting, and each subject is rewarded independently of others for estimating correctly. Between treatments, we vary the information available to subjects at the time they make their guesses. This simple setting allows us to address two questions. First, we document whether at all and to what extent people are capable of extracting the information contained in their peers’ private signals by observing their actions, in a repeated setting. Second, we study the usefulness of observing other people’s decisions when private signals of others are also available, and, thus, others' actions should be irrelevant.