What the data actually shows
The classic illustration of crowd accuracy is Francis Galton's observation at a 1906 county fair, where the average of hundreds of independent guesses of an ox's weight came out almost exactly right — closer than any individual expert. James Surowiecki's book The Wisdom of Crowds popularised the pattern and named its preconditions: the crowd does well when individual judgments are diverse, largely independent, and then aggregated. Remove any of those — especially independence — and the effect breaks down.
The failure mode has a name too. Irving Janis's work on groupthink described how cohesive groups under pressure to agree can suppress dissent, ignore warning signs, and converge on poor decisions while feeling highly confident. The mechanism is social, not intellectual: when agreement is rewarded and disagreement feels disloyal, the group stops surfacing the information that would have corrected it.
A related finding is group polarization: rather than landing in a sensible middle, discussion often pushes a group toward a more extreme version of its starting lean. People who began cautious tend to get more cautious together, and people who began bold get bolder — partly through exposure to one-sided arguments and partly through the desire to fit the group's perceived position.
Why this feels different from how it actually is
Groups often feel more reliable than they are because confidence rises with consensus. When everyone in the room agrees, it feels like confirmation — but agreement can come from people influencing each other rather than from independent signals converging. A unanimous group can be unanimously wrong, and it usually feels more certain than a lone thinker who is actually right.
It also feels different because the social experience of deciding together is pleasant and cohesive, while the experience of being the lone dissenter is uncomfortable. That asymmetry quietly biases groups toward whoever speaks first, loudest, or with most status — which is not the same as toward the best judgment. The smoothness of consensus is easy to mistake for the quality of the decision.
And we tend to remember the crowd's hits — the eerily accurate average, the team that nailed it — more than the times a committee diluted a good idea or talked itself into a bad one. The wisdom-of-crowds story is genuinely real, but it is conditional, and the conditions are exactly the parts that get lost when a group simply gathers and chats its way to agreement.
A unanimous group can be unanimously wrong, and it usually feels more certain than a lone thinker who is actually right.
What the research says to do about it
The most consistent practical lesson is to protect independence before pooling judgments. Having people form their own view privately first — writing down an estimate, a vote, or a position before discussion — preserves the diversity that makes aggregation powerful, and limits the early-anchoring and conformity that discussion introduces. Gather independent inputs, then combine them, rather than letting the group converge in real time.
Building in real diversity and structured dissent also helps. Janis's own remedies for groupthink centred on legitimising disagreement: assigning someone to argue the other side, inviting outside views, and having leaders withhold their preference so people don't simply align to it. The goal is to keep genuinely different perspectives in the room long enough to count.
For judgments that can be expressed as estimates, simple aggregation is often the highest-value move. Averaging several independent guesses tends to cancel out individual errors, which is the engine behind the crowd's accuracy. A group's advantage comes far more from combining diverse independent inputs than from discussing its way to one shared answer.
What the research says does not help
Simply adding more people does not reliably improve decisions, and can make them worse. Without independence and diversity, a bigger group mainly amplifies whatever bias or social pressure is already present. Size is not the lever; structure is.
Open, unstructured discussion until everyone agrees is one of the weaker approaches the research identifies. It maximises mutual influence, anchors people to whoever spoke first, and produces the consensus that feels like accuracy but often reflects conformity. Talking until the group is comfortable optimises for comfort, not correctness.
Deferring to the most confident or highest-status voice is also a poor heuristic. Confidence and status correlate weakly with accuracy, and groups that let them dominate effectively discard the independent information everyone else brought — collapsing back to a single, unchecked judgment that happens to feel collective.
Size is not the lever; structure is. Without independence and diversity, a bigger group mainly amplifies whatever bias is already present.
What this looks like in real life
Averaging ~800 independent guesses
At a 1906 county fair, hundreds of people each guessed an ox's weight without conferring. No single expert was reliably right, yet the average of all those independent guesses came within about one percent of the true weight. The power came precisely from the fact that people didn't talk first — their individual errors, pointing in different directions, cancelled out when pooled.
The meeting that agrees too smoothly
A team gathers, the most confident person speaks first, and within minutes everyone is nodding. It feels efficient and certain — but that certainty came from people anchoring to whoever spoke first, not from independent signals converging. The comfortable, unanimous room is exactly the setting where dissent feels disloyal and the correcting information never surfaces.
Real numbers in context
The vivid origin story is Galton's roughly 800 fairground guesses of an ox's weight in 1906: the average came within about one percent of the true weight, beating the individual entries. It is a single striking demonstration rather than a precise law, but the underlying pattern — that pooling many independent, diverse estimates cancels individual error — has held up broadly across later research.
The contrast is just as important and harder to quantify cleanly: groupthink and group polarization are documented tendencies, not fixed effects with a tidy percentage. The research is best read as a set of conditions rather than a verdict — independent and diverse plus good aggregation tends to beat individuals, while cohesive and consensus-pressured can underperform a single careful thinker. Treat any blanket claim that 'groups are smarter' or 'committees ruin everything' as overstated.