Information Cascade

An information cascade occurs when individuals, observing the actions or beliefs of others, rationally choose to ignore their own private information and follow the majority. The cascade begins when early movers' decisions become visible to later movers, who infer that the early movers possess information they lack. Once a cascade starts, it becomes self-sustaining: each new participant adds no new information, merely confirming the apparent consensus. The result is a collective outcome that may be entirely wrong despite every individual acting rationally.

Information cascades explain bubbles, fads, and collective delusions across domains from finance to fashion to political belief. They are a central mechanism by which attention architectures destroy the independence condition required for the wisdom of crowds. When algorithmic curation exposes users to the same content streams, it does not merely correlate errors; it creates the conditions for cascades by making individual choices visible and therefore imitable.

The cascade is not a failure of rationality but a failure of information structure. Rational agents with correlated information make collectively irrational decisions. This is the signature of an epistemic trap: a system in which locally optimal behavior produces globally catastrophic outcomes.

When Cascades Form and Break

Cascades require two conditions: visibility and sequentiality. Decision-makers must be able to observe others' choices before making their own, and they must act in sequence rather than simultaneously. When these conditions are met, even a small amount of early agreement can trigger a cascade that overrides all subsequent private information.

But cascades are also fragile. A single publicly observed dissenter with high credibility can break a cascade, because the dissenter's action signals that they possess private information strong enough to override the apparent consensus. This is why authoritarian regimes invest so heavily in suppressing visible dissent: not because dissenters change many minds directly, but because visible dissent breaks the information cascade that sustains apparent consensus. The preference falsification that keeps dissent invisible is therefore a cascade-protection mechanism.

In financial markets, cascades manifest as herding behavior: investors buy because others are buying, sell because others are selling. The 2008 financial crisis was, in part, an information cascade in the market for mortgage-backed securities: ratings agencies, investors, and regulators all observed others' confidence and rationally concluded that the risk had been properly priced. The cascade broke when a few dissenters — notably Michael Burry and others who shorted the housing market — made their positions visible.

Cascades in Algorithmic Environments

Social media platforms are information cascade engines. The visibility of likes, shares, and follower counts transforms individual expression into sequential observation: users see what content is popular before deciding what to share. The result is that content that captures early attention receives disproportionate amplification, while content that fails to capture early attention is buried regardless of its intrinsic quality.

The algorithmic curation of feeds intensifies this dynamic. When a platform's algorithm promotes content based on engagement signals, it creates a two-layer cascade: users imitate each other (the social layer), and the algorithm imitates the users (the technical layer). The feedback between these layers can produce extreme outcomes in which a single piece of content receives millions of views while equally good content receives none — not because of quality differences but because of the randomness of early engagement and the amplifying power of the algorithm.

This is why information cascades in algorithmic environments are harder to break than in face-to-face environments. In a face-to-face cascade, a credible dissenter can speak up. In an algorithmic cascade, the dissenter's content is algorithmically suppressed because it lacks the early engagement signals that trigger promotion. The architecture of the platform systematically prevents the cascade-breaking function that credible dissent performs in human groups.

Relation to Other Phenomena

Information cascades are distinct from but related to conformity (yielding to group pressure without informational inference), groupthink (the suppression of dissent in cohesive groups), and bandwagon effects (the tendency to adopt beliefs because they are popular). The defining feature of the information cascade is its informational rationality: participants follow the majority not because they fear social sanctions but because they rationally infer that the majority possesses information they lack.

This makes information cascades particularly dangerous. They do not feel like pressure. They feel like learning. The participant in an information cascade believes they are updating their beliefs based on evidence, when in fact they are responding to a signal that has been stripped of its informational content by the cascade itself. The result is a form of false learning — conviction without foundation — that is harder to correct than simple conformity because it is experienced as autonomous judgment.