Why the group’s estimate beats the individual’s

Eight hundred fairgoers guess the weight of an ox and the average lands within a pound of the true figure. Eight colleagues discuss the same quantity for ten minutes and converge on a number worse than four of them started with.

What everyone sees

Crowd accuracy is treated as a property of groups, which suggests that getting people together and talking should improve estimates. The property belongs to the aggregation, not the gathering. Averaging works because individual errors are independent, and conversation is the mechanism that destroys independence.

What is actually happening

Galton’s report on a weight-judging competition found the aggregate of independent guesses remarkably close to the true value despite wide individual dispersion — the errors cancelling because they were unrelated. Lorenz, Rauhut, Schweitzer and Helbing tested what happens when that condition is relaxed: giving participants information about others’ estimates narrowed the spread of answers while leaving accuracy unimproved or worse, and raised participants’ confidence in the converged figure. That combination is the trap. The group ends with a tighter distribution and more conviction, and reads both as evidence of having got closer, when the only thing that changed was how correlated the errors became.

Why it stays hidden

The degradation hides because convergence looks like progress. Disagreement narrowing is the visible sign that a meeting is working, and the resulting confidence is shared, so no participant has a private doubt to raise. The independent estimates that would have averaged well were never written down.