A group decides worse than its members when everyone talks first

Where a group’s advantage actually comes from

Aggregating several judgements beats a typical single judgement for a mundane statistical reason: individual errors point in different directions and partly cancel, while whatever signal each person has points the same way. This is why an average of independent estimates is often strikingly close to the truth even when no individual estimate is.

Everything in that sentence depends on the word independent. The cancellation requires that the errors be unrelated, and the moment people hear each other, their errors stop being unrelated. A group that discusses before estimating doesn’t get a better average; it gets one estimate held by several people, with the confidence of a crowd and the information of whoever spoke first.

What discussion does to the distribution

Two things happen when a meeting talks before it decides. Opinions converge, which feels like progress and is usually just anchoring on the first articulate contribution. And the spread of stated views narrows faster than the underlying uncertainty does, so the group ends up more confident than the evidence supports, having heard the same reasoning from several mouths.

There is also a documented tendency for groups to spend their discussion on information that members already share, rather than on the facts held by only one person — which are precisely the facts that could change the answer. Shared information is easier to raise, gets validated when mentioned, and crowds out the unique contribution that the meeting existed to surface.

The seniority problem is not about politeness

When the person with the most authority speaks first, everything afterwards is a response to their position rather than an independent reading. This does not require anybody to be intimidated. It is enough that their statement defines what the question is about, and that disagreement now costs more effort than agreement.

The corrective is procedural and slightly awkward: the most senior person speaks last, and ideally after everyone has committed a view in writing. Almost every organisation knows this rule and almost none follows it consistently, because the natural rhythm of a meeting is for the person who called it to open by saying what they think.

The same logic applies to any strong signal that arrives early, not only to rank. A circulated recommendation, a slide with a preferred option already highlighted, or a colleague’s confident remark in the corridor beforehand will all correlate the room’s judgements before anybody has spoken formally. Independence has to be protected upstream of the meeting, not only inside it.

Writing before talking

The cheapest intervention available is to have each participant record their judgement, with a confidence and a one-line reason, before any discussion. It takes a few minutes. It preserves the independence that makes aggregation work, it surfaces the range of views honestly, and it makes visible the cases where an apparent consensus was manufactured in the first ninety seconds.

The written record also changes what disagreement means. When three people wrote down three different numbers before speaking, the disagreement is data about the uncertainty in the problem, and the discussion has somewhere useful to go: what does each person know that the others do not. When the same disagreement emerges from debate, it tends to be read as a contest to be resolved.

Where consensus is doing something else

It is worth distinguishing decisions where a group is trying to estimate something from decisions where a group is trying to commit to something. For the first, independence is everything and discussion is a threat. For the second, the discussion is the point, because a decision nobody argued about is one nobody will defend when it becomes inconvenient.

Most real meetings mix the two, which is why they go badly. Separating them explicitly — estimate first, in writing, then discuss the commitment — costs little and stops the commitment phase from quietly overwriting the estimate.

A note on how strong this evidence is

The statistical case for averaging independent judgements is solid and follows from the arithmetic rather than from any contested psychology. The findings on discussion effects, information pooling and the shift of group positions after debate come from a large experimental literature of varying quality, and effect sizes differ considerably with task and group composition.

So the confident part of the advice is narrow: independence before aggregation is genuinely valuable, and it is cheap enough to be worth doing even where the surrounding evidence is mixed. The rest — the exact procedures, the ideal group size, the ordering rules — is reasonable practice rather than established fact, and should be described that way.