Concepts / Public opinion
Information cascades.
People often learn from what others do, and that is usually sensible. But when each person copies the choices made before them, their own information stops being shared. A whole group can then settle quickly on a choice that rests on very little evidence, and sometimes on the wrong one.
The idea was set out in 1992 by the economists Sushil Bikhchandani, David Hirshleifer and Ivo Welch, and, in a closely related model of herd behaviour, by Abhijit Banerjee. Banerjee's example was two restaurants: a diner who sees one busy and one empty may choose the busy one even if their own information favours the other.
How it works
Sensible choices, a fragile result.
No one in a cascade needs to be foolish or under pressure. Each person reasonably weighs what they know against what others appear to know.
- 01
A private hint
Each person has some information of their own, which is often right but sometimes wrong.
- 02
Watching others
They also see what the people before them chose, and treat those choices as evidence.
- 03
The tipping point
Once earlier choices lean clearly one way, it becomes sensible to follow them, even against one's own hint.
- 04
Information stops flowing
Later choices copy rather than inform, so a growing crowd adds no new evidence.
Why it matters
Agreement can be shallow.
A cascade can look like a broad consensus, yet rest on the private information of the first few people. Because so little information supports it, it is also fragile: a small amount of reliable public information can overturn it. Try it in the graphic above. Press Run again to see how often a cascade forms, and raise how often hints are right to see how much it helps.
Where it can show up
Wherever people decide in sequence.
These are general patterns, not findings about any particular party, community or election.
Momentum
Early support for an idea or proposal can draw in more support, partly because people read others' choices as evidence.
Sharing online
A claim can spread because others have shared it, and shares can be read as checks that nobody actually made.
Meetings and committees
When people speak in turn, later speakers may defer to earlier ones, so the order of speaking can shape the outcome.
Analysts and forecasts
Analysts may lean towards the prevailing view rather than report their own reading, so forecasts can cluster.
Reading agreement more carefully
Keep independent information flowing.
The remedy is not to ignore others, but to make sure that each person's own information still counts.
- Gather views independently first, for example in writing, before people hear each other.
- In meetings, vary who speaks first, and invite less senior people to speak before more senior ones.
- Count evidence, not endorsements: ask how many people checked something for themselves.
- Publish reliable public information; it can correct a cascade that has gone wrong.
- Treat fast, near-unanimous agreement with caution when it grew from a few early choices.
Origins and further reading
Where the idea comes from.
Cascades differ from preference falsification. In a cascade, people sincerely change their choice because they trust others' information. In preference falsification, they keep their view but hide it.
- Bikhchandani, S., Hirshleifer, D. and Welch, I. (1992). A theory of fads, fashion, custom, and cultural change as informational cascades. Journal of Political Economy, 100(5), 992–1026.
- Banerjee, A. V. (1992). A simple model of herd behavior. Quarterly Journal of Economics, 107(3), 797–817.
- Anderson, L. R. and Holt, C. A. (1997). Information cascades in the laboratory. American Economic Review, 87(5), 847–862.
About this concept
A general idea, not a verdict.
This page explains a general concept from economics and political science. It does not describe or assess any particular party, community, government or election. Related: Preference falsification and Consistency is not accuracy.