The seer-sucker theory

The seer-sucker theory names a durable bargain in the world of prediction: people keep paying for the supposed foresight of exceptional experts even when the evidence offers little reason to believe such seers exist. The book uses the phrase to challenge the search for a single authority and to argue instead for combining judgments—while making clear that a crowd is useful only when it contains information and its members can contribute independently.

The Seer and the Sucker

J. Scott Armstrong, a researcher on expertise and forecasting, summarizes the problem in a deliberately sharp formulation: “No matter how much evidence exists that seers do not exist, suckers will pay for the existence of seers.” The phrase does not merely say that forecasters sometimes get things wrong. It points to the persistent market for people who claim to predict change reliably, despite evidence that confidence and status do not guarantee accuracy.

The research discussed alongside Armstrong’s formulation gives the warning substance. Experts in fields including stock picking, livestock judging, and clinical psychology often disagree with one another; in one study, medical pathologists reached internally inconsistent judgments, giving a different opinion half the time when presented with the same evidence. Experts also tend to overestimate the likelihood that they are right. A study of foreign-exchange traders found that they overstated the accuracy of their predictions 70 percent of the time. The problem, then, is not simply error, but poor calibration: forecasters may not know how wrong they are.

Why Expertise Does Not Settle the Forecast

Armstrong’s conclusion is narrower than a rejection of expertise. Sophisticated analysts can contribute useful information, but their advice and predictions should be pooled with those of others. The book argues that the effort to “chase the expert”—to find the one person who will solve an organization’s problem—is wasted when performance varies sharply from problem to problem. A group’s decision can be better than most of its members, and can outperform even its best member; over repeated decisions, no single forecaster is likely to beat the group consistently.

The relevant distinction is between knowing a great deal and being able to forecast change. Armstrong writes that expertise beyond a minimal level has little value for that task, and that claims of accuracy by a lone expert have no practical value. The theory’s compact insult captures an institutional habit: when uncertainty is uncomfortable, organizations and audiences may buy the reassurance of a celebrated predictor instead of arranging to compare different judgments.

What Makes a Crowd Informative

The alternative is not simply to replace experts with an undifferentiated mass. A crowd’s judgment is likely to improve when its members have diverse opinions, judge independently, draw on decentralized or local knowledge, and contribute to a mechanism that aggregates their estimates. Under those conditions, averaging can cancel individual errors, leaving information that no one participant possesses in full. But the book stresses a necessary qualification: if the estimates contain little information to begin with, averaging errors cannot make the result wise.

Several examples show what this qualification means in practice. When investors reacted to the Challenger explosion, Morton Thiokol’s stock fell much more sharply than the shares of the other contractors, and later investigation confirmed that defective O-ring seals in Thiokol’s booster rockets had contributed to the disaster. The early market response was striking, though the book preserves the limits of the inference: possible explanations included luck, the company’s exposure to a downturn in the space program, or a signal from the trading halt, and the researchers could not establish exactly how investors reached the correct judgment. The case supports the possibility that scattered information, aggregated through trading, can produce a verdict no individual trader can explain; it does not prove that markets always do so.

Other evidence supplies a more systematic contrast between individual prediction and aggregation. Across forty-nine elections from 1988 to 2000, the Iowa Electronic Markets’ election-eve forecasts were off by an average of 1.37 percent in presidential elections, 3.43 percent in other U.S. elections, and 2.12 percent in foreign elections. The market was small—never more than about eight hundred traders—and did not resemble the electorate, yet it generally outperformed major national polls, including three-quarters of the time when compared with 596 polls released during presidential elections from 1988 to 2000. In professional football, the betting crowd is difficult for informed gamblers to beat consistently, and the Mirage’s final line was the most reliable forecast available roughly three-quarters of the time. These cases do not vindicate every crowd or market; they show why an expert’s reputation alone is not a sound substitute for measured performance.

Diversity, Independence, and the Limits of the Formula

Experiments on group decisions extend the argument beyond forecasting markets. In computer simulations by political scientist Scott Page, groups of ten or twenty problem-solving agents with varied skills—including some less effective agents—almost always outperformed groups composed only of the strongest agents. Diversity added approaches that a uniformly high-performing group might lack. The point was not that ability is irrelevant: successful groups still included high-performing agents, but a collection of similar strengths could leave the group knowing less than a more varied team.

A separate experiment by Angela Hung and Charles Plott showed why independence matters. Students used private evidence—the color of a marble drawn from one of two urns—to infer which urn was in use, while observing the guesses made by those before them. Under the original rules, information cascades began in 78 percent of trials: participants often followed earlier guesses even when their own draw pointed elsewhere. When payment instead depended on whether the group’s majority answer was correct, students paid more attention to their private information, and the group’s collective judgment became significantly more accurate than the cascading groups’ judgments. The lesson complements the seer-sucker theory: better decisions do not come from identifying one privileged mind, but from structuring a process that preserves independent information and makes it count.