Ask 1,000 people to guess the weight of an ox and average their answers. The average is more accurate than 95% of individual guesses — often more accurate than any single expert. This is the "wisdom of crowds." But it only works under one critical condition that most organizations systematically violate.
The Framework
Decorrelating errors is Kahneman's term for the statistical principle behind collective intelligence: when independent judges make errors, those errors tend to cancel out in the aggregate — but only if the errors are independent (uncorrelated). If every judge makes the same error in the same direction (correlated errors), the aggregate is just as wrong as any individual. The wisdom of crowds requires that each person's error is their own, not copied from someone else.
This is why Kahneman insists on a specific protocol for group decision-making: each member must form and record their judgment independently before any group discussion occurs. The moment you hear someone else's opinion, your errors become correlated with theirs (anchoring, social proof, authority deference), and the group's collective intelligence collapses to the intelligence of whoever spoke first.
Where It Comes From
Kahneman presents decorrelation in Chapter 7 of Thinking, Fast and Slow as a practical principle for organizational decision-making. The concept draws on Francis Galton's 1907 ox-weighing demonstration and the mathematical principle that the variance of the average of N independent estimates equals the individual variance divided by N. As N increases, the average converges on truth — but only if independence holds.
> "The standard practice of open discussion gives too much weight to the opinions of those who speak early and assertively, causing others to line up behind them." — Thinking, Fast and Slow, Ch 7
Cross-Library Connections
Kahneman's structured interview protocol (Chapter 21) is a decorrelation exercise: scoring six traits independently prevents the halo effect from correlating the errors across dimensions.
Fisher's brainstorming protocol in Getting to Yes (separate inventing from deciding) maintains independence during the creative phase by prohibiting evaluation that would correlate participants' ideas toward the first suggestion.
The Implementation Playbook
Meeting Protocol: Before discussing any important decision, require each participant to write their judgment, estimate, or recommendation independently. Share only after everyone has committed in writing. This eliminates anchoring and social conformity — the two primary error-correlation mechanisms.
Forecasting and Estimation: For any numerical estimate (project timeline, budget, market size), collect independent estimates from 5+ informed individuals. Average them. The average will outperform any individual estimate — but only if the estimates are genuinely independent.
Hiring Panels: Each interviewer should complete their evaluation form before discussing the candidate with other interviewers. The panel discussion happens after independent scoring, not before. Pre-discussion sharing correlates the errors and reduces the panel to a single judgment with artificial confidence.
Investment Committees: Investment analysts should submit their ratings before the committee meeting. The meeting's purpose is to discuss disagreements (which are informative) rather than to build consensus (which is error-correlating).
Key Takeaway
The wisdom of crowds is real but fragile. It requires independence — and every standard organizational practice (open discussion, HiPPO decision-making, brainstorming with real-time evaluation) destroys independence. The fix is simple and counterintuitive: silence before speech, writing before talking, individual judgment before collective discussion. The first person to speak in a meeting doesn't just share their opinion — they correlate everyone else's errors with their own.
Continue Exploring
[[Structured Interview Protocol]] — The hiring application of the decorrelation principle
[[Halo Effect]] — The primary mechanism by which one impression correlates all subsequent evaluations
[[Algorithms vs. Experts]] — Structured methods outperform judgment partly because they decorrelate dimensional assessments
📚 From Thinking, Fast and Slow by Daniel Kahneman — Get the book