VisszaAbstract:
We seed individuals in a real-world social network with information about quiz questions to
experimentally measure the effectiveness of social learning. Using data on both the pre-existing social network and the actual conversation network, we find strong evidence that people learn from direct and indirect friends, but also that—unlike in standard models—information transmission is imperfect. We then compare two theories of social learning: a DeGroot-style model in which people double-count signals that reach them through multiple paths, and a “Streams” model in which people tag the source of information and hence do not double-count. In the conversation network, the weight a decision maker attaches to the signal of a person increases in the number of paths between them when the person is an indirect friend, but not when she is a direct friend. This fact is consistent with the Streams model in which multiple paths only increase the weight by increasing the transmission probability when—like with an indirect informer— transmission is imperfect. Structural and reduced-form estimates which exploit this and other variation generally support the Streams model combined with probabilistic transmission in our setting.