Information Cascades.

01Overview

An information cascade describes a specific and disturbing form of herding: individuals deciding in sequence abandon their own private signals and follow the observed choices of predecessors, because the crowd's choices seem to encode information they lack. The first few decisions tip a fragile threshold, and thereafter behaviour propagates not because it is correct but because it is observed. The cascade is rational at each step and potentially catastrophic in aggregate: the group can end up in a state no individual would have chosen with full information, and the private knowledge that could have corrected it is never expressed. The theory shows that correct individual inference and correct collective outcomes are not the same thing — a fact central to markets, committees, and every institution that reads popularity as proof.

02Key theorists

Bikhchandani, Hirshleifer and Welch's 1992 paper in the Journal of Political Economy formalised the cascade, showing with a simple urn model that rational Bayesian agents will rationally discard their own evidence after observing enough predecessors. Banerjee's companion paper in the Quarterly Journal of Economics the same year modelled the herding equilibrium. Anderson and Holt's 1997 laboratory experiments confirmed the phenomenon under controlled conditions — human subjects cascade exactly as predicted, and cascade fragility was demonstrated too: a single public, credible signal can shatter a cascade built on dozens of private ones. The framework now underwrites analyses of financial bubbles, technology adoption, voting, and medical decision-making.

03How it works

1. 01Private signal: each individual receives noisy but useful evidence about the right choice
2. 02Sequential observation: choices are public, reasons are not — the crowd sees actions, never the evidence behind them
3. 03Bayesian discard: after enough prior choices, copying beats the individual's own fragment of evidence
4. 04Threshold fragility: cascades are often built on very few decisions — typically a handful — and can reverse on a single public signal
5. 05Information suppression: private information contradicting the cascade is never revealed, so the group cannot self-correct
6. 06Cascade collapse: a credible public correction can destroy in minutes what imitation took months to build
7. 07False consensus feedback: observers interpret the cascade as the crowd's knowledge, not as congealed silence

04Where it shows up

Restaurants and platforms where queues decide what everyone tries; financial markets where momentum trading detaches price from fundamentals; doctors adopting a treatment because colleagues did; parliamentary votes that swing behind whichever way the wind starts blowing; viral misinformation that spreads because each sharer assumes prior sharers checked.

05The propaganda link

Oceania's economy ran on cascades by design: everyone had seen the enemy's face, and no one needed to have verified it. Manufactured consensus maps onto the cascade's weak point — the moment of ignition — so controlling the early actors controls the direction of the entire crowd without a single false statement.

06How to resist

• Harvest private information before it is suppressed: collect independent estimates in writing before any discussion
• Run decisions in parallel, not sequence, where possible
• Track the sequence's origin — the fourth follower is evidence of nothing if the third was merely following
• Reward revealed dissent with status, since silence is the cascade's fuel
• Rely on outcomes where available: experimental evidence dissolves cascades faster than argument