Pattern/Decision Theory/No. 1090
Wisdom of Crowds
Wisdom of crowds is the pattern in which combined judgments outperform many individual judgments. Studied by Francis Galton and popularized by James Surowiecki, it depends on useful knowledge, partly independent errors, and how judgments are combined; shared errors can limit accuracy.
- Evidence
- Well established
- Read
- 6 min
- Links
- 10 connections
01You've seen this when…
- in life
At a fundraiser, you and your friends separately guess how many beans fill a jar. Your estimates look wildly different, so you submit their average.
- at work
Five colleagues estimate next month’s orders before seeing each other’s numbers. You keep their average alongside the forecast from the team’s most confident member.
- out in the world
Election forecasters disagree about how close a race will be. A combined forecast draws on several estimates rather than making one forecaster’s judgment carry everything.
02The idea
Nobody on your team knows next month’s sales. One person knows the customer pipeline, another watches cancellations, and another remembers seasonal changes. Their estimates disagree. That disagreement can be useful: each estimate contains something the others miss.
The wisdom of crowds is the pattern in which combining judgments produces a better answer than many of the individual judgments. The group needs enough useful information and mistakes that vary in direction. It can work with contributors of ordinary ability.
Numerical averaging has a precise advantage: the average’s squared error cannot exceed the average squared error of the individual estimates. Squared error measures misses while penalizing larger ones more heavily. This does not mean the average always beats the best person, or even that it is close to the truth.
Forecast aggregation is one way to put the pattern to work. It combines predictions using an explicit rule. A room reaching agreement is different: groupthink can make everyone agree while stripping away the information that made the group useful.
03Why it happens
- Different people hold different pieces of the answer. A sales representative sees pending deals. A support agent sees unhappy customers. Combining their judgments can incorporate information that neither person has alone.
- Some mistakes offset other mistakes. One person estimates too high, another too low. Averaging reduces some of the noise in judgment, even when the mistakes remain unidentified.
- Shared mistakes survive aggregation. If everyone builds a forecast from the same inflated market report, the average remains inflated. A common false assumption persists as more contributors join. The correlation between their errors matters as much as the head count.
- A combining rule limits dominance. A recorded average gives each included estimate its specified weight. In an open discussion, the most senior person may dominate, or someone may gain most of the influence by speaking fluently or insistently.
The useful diversity is diversity of relevant information and reasoning. Different backgrounds can help produce it, depending on what information and reasoning people bring. Ten people repeating the same analysis are closer to one source than ten.
04A worked example
In a 2011 experiment, Jan Lorenz and colleagues asked participants to give initial estimates of factual quantities and then make further estimates. Before revising their estimates, some participants could see the group’s average or the other participants’ answers.
What it looks like Useful feedback. After discovering what others think, people reconsider extreme guesses and move toward a more settled answer.
What’s actually going on Social information made estimates more similar. Participants also became more confident. The researchers identified a range-reduction effect: as answers clustered more tightly, the truth could become less well covered by the crowd’s estimates. The crowd gained agreement, creating a misleading impression of improved accuracy.
What would have helped Retain the private first estimates as a separate aggregate and preserve the spread of answers while asking what new evidence justified revisions. The mechanism suggests these practical safeguards, whose effectiveness remains unproven by this study. A revision based on a newly discovered fact is different from a revision made simply because everyone else gives a different number.
05How to spot it
06What to do about it
- Collect answers privately first. Ask everyone to give an estimate with a brief reason and identify their main uncertainty before sharing results. This reduces the chance of an information cascade, where earlier answers displace later contributors’ own evidence.
- Recruit different sources of knowledge. Include people who observe different parts of the problem. Adding another person with the same spreadsheet may add little.
- Choose the combining rule in advance. For comparable numerical estimates, a simple mean is a reasonable starting point. A median is less sensitive to extreme answers. A bias everyone shares persists under either rule.
- Base weights on demonstrated skill. If you give some contributors more influence, use performance on comparable past questions. Check calibration as well as whether their answers sound convincing.
- Discuss reasons after recording estimates. Let people exchange evidence and revise. Keep the original aggregate so you can test whether discussion actually helped.
- Track repeated performance. Compare the aggregate with individuals and a simple baseline over many questions. Wait for repeated results before choosing the method that looks best.
07When it isn’t wisdom
Even a large audience needs relevant knowledge to form an informed crowd. If almost everyone lacks the needed knowledge, averaging can give you a precise-looking expression of ignorance. A small panel of informed contributors may be more useful.
Crowd wisdom extracts information from judgments. Popularity measures how widely something is favored. The bandwagon effect concerns following what others favor. Millions of people repeating one claim may provide less independent evidence than three people checking it separately.
Some tasks also resist simple aggregation. A simple average of conflicting medical diagnoses leaves the diagnostic question unresolved. A vote about park design may reveal public preferences while leaving the question of an objectively correct design open. The question and combining rule must fit each other.
Discussion can help by revealing missing facts and correcting misunderstandings. The danger is replacing useful heterogeneity with conformity, then mistaking the resulting agreement for new evidence.
08Roots
In 1906, Francis Galton visited a livestock exhibition in Plymouth, England. A competition asked entrants to estimate the weight of an ox after it had been slaughtered and dressed. The live animal stood before them; the eventual weight was still unknown. Farmers and butchers participated alongside people with much less relevant experience.
Galton, a statistician interested in how ordinary people judged things, analyzed 787 usable entries. In his 1907 Nature article, Vox populi, he reported a median estimate of 1,207 pounds. The actual dressed weight was 1,198 pounds: a difference of less than one percent. The collection of uneven individual judgments had produced a remarkably accurate middle answer.
James Surowiecki’s 2004 book The Wisdom of Crowds gave the broader pattern its familiar modern label. He connected cases like Galton’s to markets, forecasting, and collective problem-solving. The lesson was that crowd accuracy depends on the process: the way judgments are gathered and combined can matter more than finding one supposedly exceptional judge.
09How solid is this?
Numerical averaging has established mathematical advantages, and experiments demonstrate benefits under suitable conditions. Performance depends on relevant knowledge, aggregation rules, and shared errors; social influence can reduce diversity without improving accuracy.
10Connections
- Often confused with Bandwagon Effect, Groupthink
- Helps counterNoise in Judgment
- See also Information Cascade, Social Proof, Forecast Aggregation, Independence, Correlation, Heterogeneity, Calibration
11Origin and sources
Francis Galton documented a striking crowd-estimation result in Vox populi (1907). James Surowiecki popularized the broader idea and its familiar name in The Wisdom of Crowds (2004).
- [1]Galton, F. (1907). Vox populi. Nature, 75, 450–451.
- [2]Surowiecki, J. (2004). The Wisdom of Crowds. Doubleday.
- [3]Lorenz, J., Rauhut, H., Schweitzer, F., & Helbing, D. (2011). How social influence can undermine the wisdom of crowd effect. Proceedings of the National Academy of Sciences, 108(22), 9020–9025.
Suggest an edit· Updated 2026-10-02