Informational cascade

Informational cascade is a process in which people infer what to do from the choices of others and may follow those choices instead of acting on their own private information. The book treats this as a central danger in collective judgment: a sequence of individually plausible decisions can conceal information rather than pool it. Its examples distinguish imitation from genuine aggregation, and show why groups need both independent knowledge and channels that make that knowledge collectively useful.

How a Cascade Forms

A cascade develops when people can observe earlier decisions and take those decisions as evidence. As each person follows the visible pattern, later observers encounter an increasingly strong appearance of agreement. Yet that agreement may not represent many independent judgments: people may be copying the same earlier choice, while their own information goes unexpressed. The result is a social signal that can become more influential than the evidence held by the individuals who make up the group.

The book places informational cascades within a broader discussion of herding behavior, conformity, and social influence. It notes that formal models examine both how cascades arise and how networks function, while real-world accounts trace how the mechanism can shape events. The explanation is not that people are simply irrational for learning from others; rather, the danger arises when observed behavior replaces independent assessment, so that apparent consensus grows without incorporating the full range of available information.

Independent Information and Collective Judgment

The distinction between following and aggregating matters because groups can be more informative than their members only when their members’ judgments contribute something distinct. In the account of the lost submarine, John Craven assembled people with varied expertise and asked them to estimate the likelihood of different scenarios separately. Their wagers concerned such questions as what had gone wrong, the submarine’s speed as it descended, and the steepness of its descent. He then used Bayes’s theorem to combine the estimates into a collective picture of the submarine’s final location. The method made dispersed judgments useful without first asking the participants to converge on a shared discussion or answer.

The account of decentralization makes the same structural point from another direction. Specialized people acquire local knowledge, but that knowledge can remain isolated unless a system allows it to travel and be combined. Before September 11, the U.S. intelligence community had divided responsibilities among agencies, gaining specialization and different analytical perspectives, but lacked a real means of aggregating judgments. The proposed remedy is not simply to place all interpretation in a central authority: information-sharing mechanisms, including linked agency databases, could support coordination while preserving the independence and expertise that decentralization affords.

Cascades in Institutions and Markets

The book’s institutional examples show how hierarchy can suppress independent information even when people close to a problem possess it. During NASA’s investigation of a foam strike on the Columbia, engineers on the Debris Assessment Team had concerns and wanted images of the shuttle’s wing to make an informed assessment. In the Mission Management Team’s discussion, those concerns did not receive a serious hearing. The book connects the failure to a hierarchical culture and a meeting structure that gave limited opportunity for team members to contribute. When a group starts with a conclusion, new information can be reinterpreted to fit it; the ensuing agreement need not show that the evidence was independently examined.

The 1998 crisis at Long-Term Capital Management supplies a contrasting case of concentrated confidence. The fund’s expert team included prominent figures in modern finance and had produced impressive returns for four years. After Russia defaulted on its debt, a flight to quality left the fund holding billions of dollars in assets that buyers no longer wanted. It lost $4.5 billion in a matter of months, and its forced selling pushed prices down further, contributing to losses at Wall Street banks. The case warns that expertise and a strong record do not by themselves protect a decision-making system from shared assumptions or a changing environment.

The Limits of the Concept

Informational cascades are not presented as the only explanation for herding. Robert Shiller is described as skeptical that cascades are ubiquitous, while still emphasizing social influence as a way to explain herding behavior. That qualification keeps the concept from becoming a blanket account of every instance in which people make similar choices: similarity may arise through influence, but the evidence does not warrant treating every convergence as a cascade.

The practical issue is therefore not whether people should ever learn from others, but whether a group can tell imitation from informed agreement. The book’s examples repeatedly return to the need to protect independent judgment while making dispersed knowledge available to the whole. A crowd becomes collectively intelligent not merely by bringing many people together, but by preventing the first visible answers from drowning out the information that later participants could add.