Pattern/Sociological Concept/No. 0512
Information Cascade
An information cascade is a form of herd behavior in which people follow earlier choices regardless of their own private information. Studied in economics, it can cause agreement to grow without adding independent evidence, even when the shared choice is wrong.
Also called Informational Cascade
- Evidence
- Well established
- Read
- 6 min
- Links
- 12 connections
01You've seen this when…
- in life
You pass an empty restaurant for the crowded one. The people ahead of you may also have chosen it because it was crowded, without knowing anything about the food.
- at work
The first two managers back a new supplier. You have a worrying test result, but assume they have better information and add your approval.
- out in the world
At a town meeting, the first speakers endorse a flood-defense plan. Later speakers set aside their doubts, assuming someone ahead of them has checked the engineering.
02The idea
Three approvals can look like three separate checks. But perhaps the first person checked, the second trusted the first, and the third trusted both. The visible agreement grows while the underlying evidence barely changes.
An information cascade starts when earlier people’s choices seem informative enough that someone follows them regardless of what their own private information says. That choice then gives the next person little or no new information because it reflects the visible history regardless of the person’s private clue.
This can happen without stupidity, pressure or a desire to belong. If other people might know more than you, their actions are evidence. Following them can be individually reasonable even when the resulting crowd is wrong.
The problem is the difference between many choices and many independent reasons. A hundred people repeating one person’s judgment share a single source for their endorsements. The wisdom of crowds depends on preserving information that this process can hide.
03Why it happens
- Actions are easier to see than reasons. You can see which supplier won, which restaurant filled up or which investment attracted buyers. You usually cannot see the evidence each person had. Learning from actions means trying to work backward from a choice to its cause.
- Early choices can outweigh a private clue. If two earlier people appear to have received favorable evidence, your one unfavorable clue may not be enough to overturn them. This is compatible with Bayesian updating: revising a belief as evidence arrives.
- Following stops revealing. Once people choose the same option whatever their private clue says, later observers cannot recover those clues from their actions. Useful information remains scattered across people instead of entering the public record.
- Order gives some evidence disproportionate influence. Early decision-makers act before there is much to copy. Their choices reveal more about their own information and can set the direction for everyone behind them. Chance can therefore help determine which option wins.
04A worked example
Lisa Anderson and Charles Holt tested cascades in laboratory experiments published in 1997. Participants earned money for choosing correctly after receiving private information and observing earlier choices.
Here is an illustrative version of that setup, using a hypothetical sequence. One container holds two red balls and one blue; another holds two blue and one red. Either container is equally likely to have been selected. On each turn, a person draws a ball in private and replaces it before guessing which container is in use. They see earlier guesses, while each draw stays private. For this illustration, ties are broken by following one’s own draw.
What it looks like The first two people choose the red-heavy container. The third draws blue but also chooses red-heavy. A fourth draws blue and follows them. Agreement appears to strengthen with every turn.
What’s actually going on The first two choices reveal two red draws. Those support the red-heavy container by odds of four to one. The third person’s blue draw reduces those odds to two to one, so choosing red-heavy still makes sense. But that person would also choose red-heavy after drawing red. Their choice therefore reveals nothing about their draw. The fourth person sees three matching choices, with fresh information coming from only the first two. The private blue evidence never gets pooled.
What would have helped Reveal the private draws behind everyone’s guesses. Then later participants could combine the actual evidence. Anderson and Holt observed cascades, including incorrect ones, showing that this information blockage can occur with financial incentives.
05How to spot it
These clues are inconclusive. Matching choices can also reflect independent evidence.
06What to do about it
- Collect private judgments before showing the group’s choices. Ask people to record a recommendation, confidence level and main reason before discussion. Preserve those initial answers after the group reaches its final consensus.
- Share the evidence behind endorsements. A supplier approval should come with the test results and assumptions behind it. Let the next person evaluate that material directly, so they can see what the check mark rests on.
- Trace agreement back to its sources. Find out whether five recommendations rest on five separate checks or one shared report. Count the separate information sources behind the signatures.
- Keep contrary information visible. Record unresolved objections alongside the decision. Preserve the warning that accompanied an approval.
- Separate collection from aggregation. Gather forecasts independently, then combine them using a clear forecast aggregation method. Publicly announcing the first estimate can turn later estimates into echoes.
- Check what the action actually signals. Someone may buy because of a discount, endorse because of a contract or join because friends are there. Their choice need not reveal the same judgment you are trying to make.
07When it isn’t a cascade
Learning from others is often sensible. If their evidence is strong and independent, agreement deserves weight. In the strict model, a cascade means someone’s action no longer changes with their private signal—not merely that others influenced them.
The motive also matters. A bandwagon effect is broader than an information cascade. Groupthink concerns defective group decision processes, often involving pressure for agreement. Preference falsification means publicly misrepresenting a private preference. Cascades can coexist with these, but need none of them.
Cascades are not necessarily wrong or permanent. In classic models, a modest amount of new public evidence can overturn one because its apparent strength exceeds the information supporting it. Real-world commitments, switching costs and incentives may make reversal harder.
08Roots
In 1992, Sushil Bikhchandani, David Hirshleifer and Ivo Welch published a theory with fashion and fads in its title. Its surprising move was to explain conformity without giving people any desire to conform. Each person tried to choose well using a private clue they received and the earlier choices they watched. From that spare setup, a crowd could form around the wrong answer.
Abhijit Banerjee published a closely related model of herd behavior that same year. Both approaches made the sequence of decisions central: people near the front reveal information; people farther back may simply inherit the pattern. The crowd can make a mistake even when each person’s reasoning is sound.
Anderson and Holt later brought this logic into the laboratory, where private clues and visible choices could be controlled. The experiments helped establish the mechanism beyond the mathematical models. Applying it to markets, institutions and online popularity requires more care: outside the lab, people’s motives differ, and so do the evidence they have and what they can observe.
09How solid is this?
The mechanism is established in formal models and controlled laboratory experiments, including incorrect cascades. Identifying it in real-world crowds is harder: shared evidence, social pressure and incentives can produce similar behavior.
10Connections
- Often confused with Bandwagon Effect, Pluralistic Ignorance, Groupthink
- Countered by Bayesian Updating, Forecast Aggregation
- In tension with Independence
- Can follow from Conformity, Informational Social Influence, Social Proof
- See also Illusory Truth Effect, Wisdom of Crowds, Preference Falsification
+ 2 more in the list
11Origin and sources
Sushil Bikhchandani, David Hirshleifer and Ivo Welch named and modeled informational cascades in 1992. Abhijit Banerjee independently published a closely related model of herd behavior the same year.
- [1]Bikhchandani, S., Hirshleifer, D., & Welch, I. (1992). A Theory of Fads, Fashion, Custom, and Cultural Change as Informational Cascades. Journal of Political Economy, 100(5), 992–1026.
- [2]Banerjee, A. V. (1992). A Simple Model of Herd Behavior. The Quarterly Journal of Economics, 107(3), 797–817.
- [3]Anderson, L. R., & Holt, C. A. (1997). Information Cascades in the Laboratory. The American Economic Review, 87(5), 847–862.
- [4]Bikhchandani, S., Hirshleifer, D., & Welch, I. (1998). Learning from the Behavior of Others: Conformity, Fads, and Informational Cascades. Journal of Economic Perspectives, 12(3), 151–170.
Suggest an edit· Updated 2026-10-02