Wisdom of the crowd

Wisdom of the crowd is the capacity of a group’s aggregated judgment to outperform individual judgments, especially when members bring diverse information, think independently, draw on local knowledge, and contribute through a mechanism that combines their views. The idea is neither that crowds are invariably right nor that simply assembling more people guarantees intelligence: the group must contain relevant information, and its members must not all be led into the same error. The book develops the concept through prediction markets, problem-solving experiments, financial bubbles, and the limits of applying collective judgment to politics.

Conditions for Collective Accuracy

The account begins with four conditions: diversity of opinion, independence, decentralization, and aggregation. Diversity means that each person contributes some private information, even if it is only an unusual interpretation of known facts. Independence means that judgments are not determined by the views of those nearby. Decentralization lets people specialize and use local knowledge; aggregation provides a way to turn separate judgments into a collective decision. When these conditions hold, a large enough group’s errors can cancel: each estimate contains both information and error, and averaging can diminish the error while retaining the information.

That mathematical advantage is not enough by itself. A group cannot produce a well-informed answer if its members lack relevant knowledge; the example of children trading stocks after the Challenger disaster makes the point. But where people do possess fragments of useful information, the collective verdict can assemble a strikingly complete picture—even when no individual has the whole explanation. In the market reaction to the explosion, traders collectively judged Thiokol responsible, although no one trader needed to be certain of it. The claim is therefore about the power of combining information, not about crowds possessing an unexplained instinct for truth.

Markets as Tests of the Idea

Sports betting illustrates how an aggregated judgment can change as information arrives. At the Mirage, bookmaker Robert Walker adjusted the point spread after early bets: when a couple of $3,000 wagers backed the Baltimore Ravens, he moved the opening line from Baltimore by two and a half points to three. The line was not simply a prediction fixed by the bookmaker. Once betting began, it reflected the collective response of bettors, much as a stock price responds to investor demand.

The resulting forecast was informative without being infallible. Point spreads only weakly predicted the final scores of most NFL games, and favorites covered the spread in about half of them while underdogs covered in the other half. Yet the book’s review of the evidence found no obvious, persistent mispricing such as home teams regularly winning more often than predicted. Roughly three-quarters of the time, the Mirage’s final line was the most reliable forecast available for NFL outcomes. Across other major sports, betting markets also tended to incorporate available information; horse-racing odds, for example, reliably ordered horses by their likelihood of winning and offered reasonable estimates of winning probability.

The Iowa Electronic Markets offered a more direct comparison with election polling. Its contracts could track a candidate’s share of the popular vote: a George W. Bush contract bought for election prediction would pay 48 cents if he received 48 percent of the vote. A study of forty-nine elections from 1988 to 2000 found election-eve errors averaging 1.37 percent for presidential elections, 3.43 percent for other U.S. elections, and 2.12 percent for foreign elections. The market generally outperformed major national polls, and in the 1988–2000 presidential cycles its price was more accurate than three-quarters of the 596 polls released on the same days. Its forecasts also moved less erratically than poll results, tending to shift sharply in response to new information rather than swinging constantly.

That performance did not depend on a representative sample of voters. The market had never more than about eight hundred traders; most were men, and a disproportionately large, though shrinking, share lived in Iowa. They were not predicting their own votes. Their collective estimate nevertheless proved better than asking voters directly what they intended to do. The example separates a crowd’s accuracy from its demographic resemblance to the population it forecasts: what matters is how information and judgments are combined.

Diversity in Smaller Groups

Markets often acquire some diversity through their size and open access: participants differ in risk tolerance, time horizons, investing styles, and information. Small teams and formal organizations cannot rely on scale to produce the same variety. A few biased members can dominate a small group, so cognitive diversity must be deliberately created rather than assumed.

Political scientist Scott Page tested the value of diversity with computer-simulated groups of ten or twenty agents, each possessing a different set of skills. Some agents were individually strong problem-solvers and others less effective. Yet groups mixing strong and weaker agents almost always outperformed groups made up only of the strongest agents. Randomly selecting a group and letting it work could do as well as, or better than, investing substantial effort in identifying the best individual agents and placing them together.

The result does not make intelligence irrelevant: successful groups still included high-performing agents, and none of the agents was ignorant. It shows instead that individual ability alone cannot guarantee a range of approaches. If highly capable people tend to possess similar skills, selecting only the strongest can leave a group’s collective toolkit narrower than it needs to be. Participants with different, even less developed, skills can add perspectives that the strongest members do not supply, improving the group’s performance as a whole.

How Crowds Lose Their Wisdom

The same tendency to learn from others that can make collective judgment useful can also undermine it. Piggybacking on other people’s judgments is often sensible when the group has genuine information. But if everyone follows the group’s apparent wisdom, nobody contributes fresh judgment to it. The group’s price or consensus may initially be accurate; once members copy it rather than independently assessing the evidence, the process that produced its accuracy begins to disappear.

The Caltech asset-market experiment makes that reversal concrete. Students traded shares in an imaginary company over fifteen five-minute periods. Each share paid a 24-cent dividend at the end of every period, making its total possible return unambiguous: $3.60 if held throughout. Its value should fall by 24 cents after each dividend, from $3.60 before the experiment to $3.36 after the first period, then $3.12 after the second. Instead, the trading price immediately rose to $3.50 and stayed there almost to the end. Participants kept paying that price even as the shares’ remaining value fell below $3, then $2, and eventually $1. Many knew they were overpaying but expected to collect a dividend or two and resell to someone else at the same price. Their conduct relied on the greater fool, not on the shares’ known value.

Real bubbles are less transparent from within. In the Caltech experiment, students could admit they were hoping to pass an overpriced share to another buyer. During a real market boom, some investors believe not only that prices will keep rising but that the assets may deserve their extraordinary valuations. A transformative technology or business trend can provide a rational starting point; the Internet was transformative, and biotechnology changed the pharmaceutical industry. The danger is that trading on expectations can take over from assessing the underlying trend. As investors mirror one another, collective judgment loses the independence on which its earlier strength depended.

Information itself can contribute to that loss when it pushes everyone in the same direction. In the account of Jack Treynor’s jelly-bean experiment, a true piece of information skewed students’ estimates in a shared way; the way it was presented effectively subtracted from their ability to judge the jar. The broader principle is that collective intelligence requires a balance between public information everyone shares and private information held by individual members. During a bubble, expectations converge; during a panic, they can converge in the opposite direction. Media coverage can intensify either feedback loop by making dissenting voices scarce, although it does not create the loop by itself. A market that once combined independent judgments can become a mob.

Scope and Limits

The concept applies most securely where a question has an answer that can be checked against an outcome. Election forecasting, sports predictions, and other estimation tasks allow collective judgment to be evaluated in this way. Choosing political candidates or deciding public policy is different: people can disagree about what the common good means, and there may be no objective answer against which the crowd’s decision can be measured. The book therefore cautions against assuming that voting will display the same kind of accuracy as a prediction market or an estimate of a measurable quantity.

That limitation does not establish that dictatorship, aristocracy, or elite rule would make better policy. Those systems carry a greater risk of unchecked and unaccountable power. Nor are all collective problems alike: coordination and cooperation produce less definite solutions than cognition problems, often emerging gradually through convention and practice rather than through one decisive group verdict. Such arrangements can be fragile and exposed to exploitation by free riders and tax evaders. The wisdom of the crowd is thus not a universal formula for collective life, but one form of collective intelligence whose usefulness depends on the kind of problem and on the conditions under which people make and combine their judgments.

The discussion also distinguishes aggregation from decentralization. Friedrich Hayek emphasized how markets coordinate dispersed local knowledge without central direction, transmitting information through prices; he distrusted the idea that this knowledge could be condensed into a single, centrally produced answer. The book accepts the importance of that decentralized process but argues that aggregation can add value too. Pari-mutuel betting, in which a central authority takes the bets and produces a common set of odds, is not Hayekian in this respect, yet it can yield an unusually accurate picture of the future. The central insight is not that one organizational form always wins, but that crowds become wise when relevant knowledge remains diverse and independent while some process makes its combination possible.