Appearance
11.3 — How Groups Think
A group of intelligent, experienced people, none of whom individually believes a plan will work, unanimously approves it.
This is not rare and it is not stupidity. Groups have failure modes that individuals do not, and they are produced by the structure of the interaction rather than by the quality of the people.
Groupthink
Irving Janis coined the term after studying disastrous foreign-policy decisions, and identified a set of symptoms: an illusion of invulnerability, unquestioned belief in the group's rightness, stereotyping of outsiders, direct pressure on dissenters, self-censorship, an illusion of unanimity, and self-appointed guardians who filter out inconvenient information.
The honest status: the concept is enormously influential and the empirical support for the full model is weaker than its fame suggests. Laboratory tests have produced mixed results, and the original evidence was case studies chosen after the outcome was known — which is 1.4's problem exactly.
What does hold up is the component mechanisms, each of which has independent support: self-censorship, the illusion of unanimity, and pressure on dissenters. Those are real, they are the ones you can do something about, and they are largely the Asch effect (11.1) operating inside an organisation.
Group polarisation
Better established and less well known.
Discussion makes a group's average position more extreme, in the direction it already leaned.
A group of people mildly in favour of a risk becomes, after discussion, strongly in favour. A group mildly opposed becomes strongly opposed. The discussion does not moderate — it amplifies.
Two mechanisms. People hear arguments they had not thought of, all pointing the same way, because the group leans that way. And people adjust towards what they perceive as the group's position, frequently overshooting in order to be a good member.
The practical consequence is large. A committee of like-minded people does not average out to a moderate position; it ends up further out than any individual started. This is one of the better explanations for what happens in online communities, in political factions, and in any organisation that has hired for cultural fit.
Diffusion of responsibility
From 11.2, applied to work rather than to emergencies.
When several people are responsible, nobody is. The task that is everybody's job does not get done, and each person reasonably assumes somebody else has it.
And it applies to objections as well as to tasks. In a room of eight people who all have doubts, each assumes somebody else will raise them.
The fix is structural and simple: a named person per item, and it is not "the team".
Social loafing
Individual effort declines as group size increases, on tasks where individual contribution is not visible.
Demonstrated originally with people pulling on a rope — the total force of a group is reliably less than the sum of what each produced alone.
When it happens: contribution is invisible, the task is not personally meaningful, and nobody expects to be identified with the outcome.
When it does not: individual contributions are identifiable, the task matters to the person, or the group is small enough that everybody sees everything.
The fix follows directly: make contributions visible. Not surveillance — attribution. Who did what, said out loud.
Deindividuation
The loss of individual identity in a crowd, associated with behaviour people would not engage in alone.
What is supported: anonymity, being part of a large group, and high arousal all reduce individual restraint.
What has been revised: the old idea that crowds simply strip away civilisation and release a primitive mob has not held up. Modern work finds that people in crowds do not lose norms — they shift to the norms of the group they are identifying with. A crowd at a religious festival and a crowd at a riot both behave according to a shared identity, and the content of that identity predicts the behaviour.
Which is more useful, because it says the intervention is about identity rather than about control.
What actually makes a group decide well
The research on this converges on a short list, and it is mostly about process rather than about people.
Diversity of information, not only of opinion. Groups whose members hold different information outperform homogeneous ones — provided the different information actually gets into the room, which is the hard part. There is a well-documented tendency for groups to spend their discussion time on what everybody already knows, and to under-discuss what only one person knows.
Independent judgements first, discussion second. This is the single highest-value procedural change available. It preserves the information in individual views before it is contaminated by the group's emerging position.
A designated critic. Formal, rotating, and explicitly assigned so that objecting is a role rather than a personal act (1.8).
Leader speaks last. Otherwise everything after is anchoring.
Psychological safety. Amy Edmondson's construct — the shared belief that speaking up will not be punished — has decent evidence as a predictor of team performance and of error reporting. The key finding is counter-intuitive: teams with better safety report more errors, not fewer, because the errors are being surfaced rather than hidden.
And a decision rule agreed in advance. Who decides, on what basis, by when. A great many bad group decisions are the result of nobody knowing whether the meeting was a consultation or a vote.
The two-meeting rule
A practical structure that addresses most of the above.
First meeting: information and options only. No decision permitted, explicitly. People bring what they know, questions are asked, alternatives are listed. The prohibition on deciding is what makes people willing to raise doubts, because raising a doubt is not blocking anything.
Then a gap. Overnight at least. This is where the objections that nobody could formulate in the room arrive.
Second meeting: decide. With written positions collected beforehand.
The gap is doing most of the work, and almost nobody builds it in.
The specific failure to watch for
When a decision is going smoothly and nobody has objected, that is the warning sign, not the good sign.
Alfred Sloan of General Motors is supposed to have adjourned a meeting on hearing complete agreement, saying he would postpone the decision until there was some disagreement, so the group would have time to understand what it was deciding. Whether or not the anecdote is accurate, the instinct is right.
A genuinely good decision usually has at least one person who is uneasy, and the absence of that person means either the decision is trivial or the doubts are not being said.
What to do with this
Collect written positions before discussion, on anything that matters. This one change addresses conformity, self-censorship, polarisation and the illusion of unanimity simultaneously, and it takes five minutes.
Speak last if you are the most senior person in the room.
Assign one name to each action, and never "the team".
Run the pre-mortem (1.8) — it is a year from now and this failed, why? — before committing to anything significant. It gives the doubters a legitimate format.
And treat unanimity as a signal to look harder. "Is there anyone who thinks this is a bad idea, or who would be uneasy if it went ahead?" asked directly, with a pause long enough to be uncomfortable, will produce something surprisingly often.
Next: 11.4 turns from group decisions to individual influence — the levers people use to move you, how each one works, and how to defend against them without becoming impossible to talk to.