Wisdom of the Crowds

Crowds can produce remarkably accurate estimates when people make independent judgments and their errors point in different directions. If answers share the same bias or spread through imitation, averaging amplifies error instead of correcting it.

At a 1906 country fair, roughly 800 people tried to guess an ox’s weight. Nobody got it exactly right—but the crowd’s median answer came within 1% of the true weight.

R1

Accuracy made from mismatched mistakes

The crowd does not become wise because its members secretly know the answer. Each estimate combines some useful information with noise: one person guesses high, another low, and others miss by different amounts. When those errors are sufficiently independent and diverse, aggregation cancels much of the noise while preserving the signal they share. That is the causal core captured by the Diversity Prediction Theorem: disagreement helps only when it reflects genuinely different error patterns.

R1

Where it shows up

Pune to Paris

FutureIQ asked ten office colleagues to estimate the unfamiliar straight-line distance from Pune to Paris. The point was not that any respondent knew it, but that combining their scattered guesses could outperform the typical individual answer.

E1

The ox at the fair

The ox contest shows why individual wrongness and collective accuracy can coexist: hundreds of imperfect estimates supplied errors on both sides of the truth, allowing the median to land extraordinarily close.

R1

A crowd can share one mistake

Averaging cannot remove an error that everybody inherits. If participants copy one another, draw on the same assumptions, or converge on a vivid answer through salience, their mistakes become correlated. The mechanism then moves toward Madness of the Crowds: more voices create confidence, not correction.

Collect before people confer

For your next uncertain numerical decision—such as a cost, deadline, or demand forecast—ask several people to submit estimates privately before any discussion. Record the answers, aggregate them, and only then reveal the group result; this protects the independence that makes the crowd useful.

Episodes that teach this