Contrarian Incentives and Costly Social Learning

We study sequential social learning when agents pay a fixed cost for private information and prefer less popular actions. Actions taken without new information leave beliefs unchanged but alter popularity and subsequent decision cutoffs, potentially restarting acquisition. In a binary-signal benchmark, we characterize the restart region and show that public log odds at information dates form a stopped random walk. Contrarian incentives initially expand this region and weakly improve terminal beliefs and action accuracy in discrete steps. After the region reaches an intrinsic information-cost frontier, beliefs stop improving; beyond a second threshold, the long-run frequency of correct actions declines toward one half. For general experiment menus, any positive fixed fee uniformly bounds expected purchases and, with full-support signals, implies incomplete learning. The restart mechanism extends to recency-weighted popularity indices and to endogenous Gaussian precision.