Extension: Distinguishing Informational Cascades from Herd Behavior in the Laboratory
Çelen and Kariv extend Kaal's caution that unanimous verdicts do not, by themselves, identify an information cascade. Their experiment separates observed actions from underlying beliefs. A herd begins when participants take the same action. A cascade requires the stronger condition that they act without regard to private signals. The authors elicit signal cutoffs before each participant sees the private signal. This allows them to distinguish conformity in realized actions from decisions that no possible private signal would change. They find frequent herding and cascades, but not every observed herd is a cascade. The result supports Kaal's treatment of unanimity as a cascade-consistent quantity rather than a unique diagnostic. Scope matters. The experiment uses sequential human decisions with continuous private signals and binary actions. It does not study validators, LLM agents, work-product approval, or Kaal's observed unanimity rate. It also does not test Holmström's shared-product effort mechanism. The source supplies direct evidence for the identification problem only: identical choices can coexist with continuing sensitivity to private information. Kaal's application to validator verdicts remains a separate empirical proposition.
economicsconsensus-and-securityai-and-agentsscholarly-growth-coveragescholarly-literatureinformation-cascadesherd-behaviorprivate-information