Conformity

Why will a person state an answer they can see is wrong once everyone else has given it?

Asch's task was chosen to be insultingly easy. A card shows one line; a second card shows three lines, one of which plainly matches. People judging alone get it right essentially always — the error rate in the control condition is under one per cent.

Now seat the participant last in a row of seven, and have every person before them give the same wrong answer, confidently, without hesitation. Roughly three-quarters of participants will go along with the group at least once, and across the critical trials people give the group's wrong answer about a third of the time.

The obvious reading is that people are weak. But hold that thought, because the same experiment contains a variation that makes almost all of it disappear, and weakness is not a thing you can switch off.

Start by asking what the participant is actually facing. Two things they normally expect to agree have come apart: what their eyes report, and what six apparently sane people report. Both are ordinarily good evidence. So there is a genuine puzzle here even before any social pressure arrives — six independent observers being wrong about a line is, on the face of it, very unlikely.

That suggests one route to the wrong answer: the group is information. If you take their reports as evidence, you should update, and if you update far enough you may sincerely see the line differently. Call that informational influence.

But there is a second route, which has nothing to do with belief. You might know perfectly well that the group is wrong and say their answer anyway, because contradicting six people out loud is a thing you would rather not do. Call that normative influence — conforming to be accepted rather than to be right.

Can we separate them? Yes, and the experiments do. Have people write their answer privately instead of announcing it, and conformity falls sharply. That is diagnostic: if the group had genuinely changed what you believed, privacy would not help, because you would still believe the changed thing. Asch's own interviews agree — most participants who conformed said afterwards that they had known the majority was wrong. A minority reported real doubt about their own eyes, so both routes are live, but the public-versus-private result says the normative one is doing most of the work.

Now vary the group. How many opponents does it take? Fewer than you might think, and then it stops mattering. One confederate giving the wrong answer produces almost nothing — around three per cent. Two produce noticeably more. Three produce roughly the full effect, and adding a fourth, or an eighth, does not push it meaningfully higher.

That plateau is a clue. If conformity were pressure accumulating person by person, more people should mean more pressure. It does not. What three people supply that two do not is the appearance of a consensus — something you can be the lone dissenter from.

Which sets up the most useful finding in the whole line of work. Give the participant a single ally who answers correctly, and conformity collapses — to around five per cent, a small fraction of the unanimous rate. One person out of seven. And here is the part that settles the interpretation: an ally who gives a different wrong answer also cuts conformity substantially. Someone who is no help at all as evidence is enormous help as company. It was never the number. It was the unanimity.

The tempting summary is that people cannot trust their own eyes. That is not what the data show. About a quarter of participants never conformed once, most conformers reported seeing the lines correctly throughout, and the whole effect can be dismantled by a single dissenter. What is fragile is not perception. It is the willingness to be the only one saying so.

Two refinements worth keeping. First, the two routes have different signatures, and confusing them leads to bad predictions: normative conformity is undone by privacy and by allies, informational conformity is not, and grows instead with genuine ambiguity in the task. Real situations mix them, and the mix shifts as the question gets harder — Asch's line was deliberately unambiguous precisely to strip informational influence out.

Second, the numbers are not constants of human nature. A meta-analysis by Bond and Smith across many countries and decades found conformity rates in this paradigm varying substantially — higher in more collectivist societies, and declining in the United States since the 1950s. The mechanism travels; the rate does not.

And one thing the study does not show: that groups make people stupid. The participant who deferred to six observers was, in most ordinary circumstances, reasoning well. Treating agreement as evidence is a good policy. It fails here because the observers were not independent, and independence is the assumption the whole inference rests on — which is also why an echo of the same opinion arriving from six directions online should count for rather less than it feels like it does.

• How the same mechanism produces pluralistic ignorance, where nearly everyone privately disagrees with a norm they publicly uphold.
• What happens when the task is genuinely ambiguous, as in Sherif's autokinetic studies, and informational influence takes over.

Every term the collection defines is gathered in the glossary.