Independence of Judgment

Independence of Judgment is the relative freedom of individuals to form opinions without simply adopting those of the people around them. It is one of the conditions that allows a crowd to become more than a collection of followers: together with diversity, decentralization, and aggregation, independence helps preserve distinct information so that individual errors can cancel rather than reinforce one another. The book’s examples show that independence is not isolation or a guarantee of personal accuracy. It is a design condition for collective intelligence—and one that institutions can either protect or undermine.

Independence and the Wisdom of Crowds

The account begins with four conditions for a wise crowd: diversity of opinion, independence, decentralization, and aggregation. Each person should contribute some private information, however modest or eccentric; judgments should not be determined by those nearby; people should be able to draw on specialized or local knowledge; and some mechanism must combine their separate views. When a sufficiently large, diverse, independent group estimates a probability or makes a prediction, individual guesses contain both information and error. If the errors do not point systematically in the same direction, averaging can cancel them and leave a collective judgment that retains much of the information.

Independence does not make every crowd right. The information in its members’ judgments still has to be worth combining; a group with no relevant knowledge will not become wise merely by making its members independent. What is striking, the book argues, is how much information crowds can sometimes hold collectively, as in Francis Galton’s ox-weight experiment and the market’s response to the Space Shuttle Challenger disaster. In the latter case, the market’s collective verdict could reflect the scattered knowledge traders carried, even if no single trader knew for certain that Thiokol was responsible.

The NFL betting market offers another illustration of collective judgment without a promise of perfect prediction. Bookmaker Robert Walker shifted the Mirage’s point spread after early $3,000 bets favored the Baltimore Ravens, since a one-sided flow of wagers suggested that the opening line might be wrong. The final line, shaped by bettors’ activity, was difficult for even informed gamblers to beat consistently; in roughly half of games favorites covered and in the other half underdogs did. The market was not flawless, but the Mirage’s final line was the most reliable available forecast about three-quarters of the time. The example depends on bettors contributing judgments to a shared market rather than on any one person possessing a crystal ball.

The Iowa Electronic Markets likewise forecast elections without relying on a representative sample of voters. A study of 49 elections from 1988 to 2000 found election-eve errors averaging 1.37 percent in presidential elections, 3.43 percent in other U.S. elections, and 2.12 percent in foreign elections. Across 596 polls released during the presidential elections in that period, the market price was more accurate three-quarters of the time. Its traders—never more than about 800, mostly men, and disproportionately from Iowa—were not a miniature electorate. Their judgments could nevertheless aggregate information about what voters would do. These cases show independence at work within a larger system of diversity and aggregation, not as a stand-alone virtue.

Why Independent Judgment Matters

The book describes independence as important for two related reasons. First, it prevents individual errors from becoming correlated. A mistake made by one person need not damage a group’s answer if the others are not led to make the same mistake; when people rely on one another for information, their judgments can instead become systematically biased in the same direction. Second, independent people are more likely to contribute information that others do not already possess. The most capable groups therefore combine different perspectives with enough autonomy for those perspectives to remain distinct.

Autonomy is not the same as rationality or impartiality. An individual can be biased or irrational and still contribute to a group’s intelligence if that person’s errors are independent of everyone else’s. This is why independence is a relative condition, not a claim that people stand entirely apart from society. People learn from one another, and the neighborhoods, schools, and corporations in which they live shape what they know, believe, notice, and fear. The challenge is to benefit from social learning without making every person’s judgment a copy of the same signal.

The contrast with army ants gives the danger a stark form. When ants become separated from their colony, they may follow the ant in front of them and form a circular mill, which usually ends only when a few ants stray off by chance and others follow. The same simple responsiveness that helps an ant colony find food and carry out its work can trap these ants in a fatal loop. Humans, the book argues, need not march in a circle merely because those ahead are doing so: their judgments can remain their own even while they participate in a group.

A group’s decisions are especially vulnerable when judgments are made one after another. Information cascades are a form of aggregation, and in classroom experiments cascading groups selected the better alternative about 80 percent of the time—better than any individual could do. But people in a sequence can discount their private evidence after observing earlier choices. In an experiment by Angela Hung and Charles Plott, students drew marbles from one of two urns: one contained twice as many light marbles as dark ones, the other twice as many dark as light. Students used the color they privately drew and the guesses announced by those before them to identify the urn; they could not disclose the marble’s color. When three earlier students had chosen one urn, a student who drew evidence favoring the other often followed the group instead. Cascades began in 78 percent of trials.

Hung and Plott then changed what counted as success. Instead of paying students for their own correct answer, they rewarded them according to whether the group’s majority answer was correct. Students consequently gave more weight to their imperfect private information and less to the choices already made. Individual guesses could become less reliable while the group became more accurate, because the answers preserved more distinct evidence. By reducing the influence of earlier choices, the experiment weakened the sequential element that had encouraged cascading. The book draws an organizational lesson: groups should, where possible, collect people’s judgments simultaneously rather than letting the first answers steer everyone who follows.

Decentralization, Information, and Coordination

Independence depends partly on how knowledge and authority are organized. Decentralization allows people to specialize and draw on local knowledge, including tacit knowledge tied to a particular place, job, or experience. Its strength is that people can work from what they know directly while still coordinating with others. Its weakness is the reverse side of that local focus: valuable information found in one part of a system may never reach the rest. A crowd needs both room for individual knowledge to develop and a means of making that knowledge collectively useful.

The failure of U.S. intelligence agencies to share information before the September 11 attacks illustrates the problem. Congressional criticism focused on fragmented holdings and the absence of a coherent picture assembled from localized snapshots. Senator Richard Shelby argued that the FBI’s decentralized structure had left information in largely independent fiefdoms; his proposed remedy was a central agency above the disputatious bureaucracies. In this case, decentralization did not ensure the useful independence described elsewhere. Without adequate sharing and aggregation, separate judgments and information could remain stranded instead of contributing to a collective assessment.

The tension also appears in the television ratings system. Replacing the Nielsen sweeps diaries with people meters might have improved the accuracy of local ratings, but installing the technology across the country’s 210 TV markets could cost at least hundreds of millions of dollars. Although networks, stations, and advertisers might collectively benefit, no single participant stood to gain enough to justify paying for the alternative. Local advertisers in Sioux Falls wanted accurate numbers for the CBS affiliate, but their spending could not warrant financing meters for the town; ABC, meanwhile, had little direct economic incentive to stop stunt programming. Each could wait for someone else to act and still share in any eventual improvement. The result was an institutional obstacle to collective reform, even where a better shared information system might serve the participants as a whole.

A different kind of coordination is visible in Toyota’s production system. Frontline workers are trained in a wide range of skills and in how the production process works as a whole; teams, rather than isolated workers each responsible for one piece, take charge of their production process. Any worker can pull a cord to stop the line when something needs fixing. The cord is rarely pulled, but its availability gives workers a recognized means to act on local knowledge rather than waiting for constant orders. Decentralization here supports both independent judgment and coordination: authority is placed close to the work, while workers’ decisions remain connected to a shared process.

Collaboration Without Uniformity

Independence does not mean that people must work alone or keep knowledge to themselves. The WHO’s global effort to identify the cause of SARS shows how separate lines of inquiry can become mutually useful. After receiving reports of the disease, the WHO contacted 11 laboratories in countries and regions including France, Germany, Japan, the United States, Hong Kong, Singapore, Canada, the United Kingdom, and China. The labs began work together on March 17, 2003, held daily teleconferences, posted images, analyses, and test results to a WHO website, and exchanged virus samples so that they could check and learn from one another.

The labs could investigate the same samples simultaneously. They considered and eliminated possible causes; by March 21, scientists in Hong Kong had isolated a likely candidate, while scientists at the U.S. Centers for Disease Control separately identified a coronavirus. Over the following week, researchers detected the coronavirus in samples from many people diagnosed with SARS, and laboratories in Germany, the Netherlands, and Hong Kong sequenced it. In early April, infected monkeys in the Netherlands laboratory developed full-blown SARS. On April 16, only a month after the collaboration began, the laboratories announced that the coronavirus caused the disease.

The achievement cannot be assigned to one discoverer. Electron microscopist Cynthia Goldsmith first spotted the coronavirus, but that sighting alone did not establish that it caused SARS. The conclusion depended on weeks of work across laboratories, including tests that ruled out other viruses and evidence that linked the coronavirus to the illness. The WHO’s account therefore described the group as having discovered the coronavirus collectively. The network had no director dictating each laboratory’s samples, tasks, or exchanges. The WHO convened it; the labs agreed to share relevant data and speak each morning, then worked out how to divide the investigation. Their independence made parallel inquiry possible, while collaboration allowed separate findings to be tested and combined.

That model also reflects the wider practice of open science. Scientists publish results and make data available so others can inspect evidence, challenge conclusions, build hypotheses, and conduct new experiments. Even failed hypotheses can help colleagues by showing where not to look. Scientific knowledge is therefore not merely accumulated from predecessors; it is produced through reciprocal dependence among contemporaries. Independence of Judgment does not require intellectual isolation. Its deeper demand is that cooperation preserve enough distinct inquiry for a group to discover what no single member could establish alone.