Information Aggregation and Social Networks: Responsiveness and Overturning
TL;DR
It is shown that a network is uniquely optimal for some informational environment if and only if the focal agent observes every predecessor, and that which network performs best depends critically on the informational environment.
Abstract
This paper studies how network structure affects information aggregation in social learning. Agents sequentially choose actions based on private signals and observations of neighbors'actions. Comparing a focal agent's expected payoff across networks at a finite period, we show that a network is uniquely optimal for some informational environment if and only if the focal agent observes every predecessor. This characterization reveals that which network performs best depends critically on the informational environment. We then revisit two important implications of the characterization through transparent constructions: the star network can uniquely outperform every alternative under binary signals, while the complete network can do so with richer signals. These constructions highlight a trade-off between the responsiveness effect and the overturning effect: sparse networks facilitate information aggregation by preserving the responsiveness of actions to private signals, whereas dense networks facilitate information aggregation by revealing extreme information that overturns existing public beliefs.