Thirty seconds are enough — for some things

In 1993 Nalini Ambady and Robert Rosenthal cut three ten-second clips from video recordings of 13 teachers, turned off the sound and showed them to strangers. The judges saw thirty seconds of silent behaviour and knew nothing about the people on screen. Their ratings predicted how the students would evaluate those same teachers at the end of term.

The finding is still uncomfortable, because it cuts both ways. It shows that very short slices of expressive behaviour — “thin slices”, in the technical term — can carry real information. But it shows just as clearly that a formal assessment covering an entire term is anticipated by thirty seconds of body language. Which of those two numbers is the measure of truth, the study does not say.

The title of the paper, incidentally, names two sources of judgement side by side: nonverbal behaviour and physical attractiveness. Both were present in the same silent clips, and both were equally available to the judges. Anyone reading the finding as evidence of insight into people has to read the second source along with the first.

What the meta-analysis found

A year earlier the same two researchers had pooled the existing work. Their meta-analysis of 38 results yielded an average hit rate of r = .39. What that figure is worth can be put in numbers with a tool Robert Rosenthal helped devise himself: the hit rate is translated into a proportion of correct assignments. r = .39 then becomes a success rate of roughly 70 against 30 per cent. That is markedly more than chance and markedly less than skill — about the resolution at which thirty seconds of silent film yield a rank order, but not a person.

More striking than its size was a secondary finding:

Slices under half a minute predicted no worse than four- to five-minute observations. More watching did not buy more accuracy.

That equivalence has since been contested. In 2007 Dana Carney, Randall Colvin and Judith Hall found the opposite: in their data accuracy rose with observation time, and roughly a minute gave the best ratio of effort to precision. The dispute is not settled. No published re-analysis correcting the 1992 meta-analysis exists to date. The documented qualification comes from Carney and colleagues: accuracy depends heavily on which trait is being judged and on how long one looks.

Which traits show themselves in seconds

That dependence on the trait is the most usable part of the finding. In five-second slices, negative affect, extraversion, conscientiousness and intelligence were already judged moderately well. Positive affect, neuroticism, openness and agreeableness, by contrast, required longer observation.

The split looks arbitrary at first but follows a logic. Traits that translate directly into visible behaviour — pace, volume, gaze, the orderliness of movement — are there within seconds. Traits that concern inner states and that people actively regulate are not. Anxiousness can be concealed; expressiveness cannot.

Four places where it can break down

In 1995 David Funder described everything that has to succeed between a trait and an accurate judgement about it. His model names four stages. First relevance: the trait must generate behaviour that has something to do with it at all. Second availability: that behaviour must occur in the situation being observed, not merely somewhere else. Third detection: the judge has to notice it. And fourth utilisation: the judge has to read it correctly.

The chain can break at any stage, and the second breaks most often. A job interview reliably shows how someone conducts job interviews — not automatically how they will work in a team six months later. Ten seconds of teaching say something about a lecturer’s effect on students, but little about their care in marking. An application, a quarrel or a camera produces roles rather than neutral samples; the ecological validity of the scene helps determine what can follow from it in the first place.

Funder’s model also explains why a workable criterion of truth is needed. Self-report, informant report, behaviour and success do not measure the same thing. A judgement that matches a person’s self-image may diverge from their behaviour — and neither of the two is automatically the right one.

What the slice invents

Alongside the information, something else travels at the same speed. In 2005 Alexander Todorov and colleagues showed that judgements of competence predict the outcome of American congressional elections better than chance. The judgements rested on the candidates' faces alone. In the 2004 Senate races they were correct in 68.8 per cent of contests, and they were linearly related to the winner’s margin. One second of looking at the face was enough. The judgements were, moreover, specific to competence rather than a matter of general liking.

Note what is being predicted here. Not the competence of those elected, but the election result. An impression that forms in a second and knows nothing about actual fitness for office has a measurable share in a decision regarded as a model case of deliberate weighing-up.

How little time such an impression needs was measured by Janine Willis and Todorov in 2006. Across five experiments, 100 milliseconds of exposure to a face were enough for judgements of attractiveness, likeability, trustworthiness, competence and aggressiveness to agree closely with judgements made without any time limit. Looking longer did not raise the agreement. It raised the confidence with which judges stood by their verdict. Two further findings complete the picture. Between 100 and 500 milliseconds the judgements became more negative and were made faster; with longer exposure the impressions also became more differentiated. The eye keeps working, then — just not on the question of whether it is right.

That is the real trap. Time feels like scrutiny without being any. Accent, clothing, body, gender and origin shape impressions even when they are irrelevant to the question at hand. And when several judges agree, that is still not truth — a distinction Funder insists on explicitly. Consensus measures how similarly people judge, not how well. Many judges can share the same cultural bias and confirm it for one another.

Intuition with a rear-view mirror

A useful first assessment is treated as a hypothesis. What exactly did I observe? What alternative cause is there? Would I read the same behaviour the same way in a different person? What later information would change my mind?

In practice it helps to separate observation from interpretation: “interrupted three times” is an observation, “is disrespectful” is already an explanation. Several scenes, different observers and a question defined in advance narrow the space in which liking, stereotypes and expectations quietly turn into traits.

What remains

The first impression is neither magic nor superstition but an instrument with very uneven resolution. For traits that translate into visible behaviour it is usable and astonishingly quick. For everything turned inward or open to regulation it is not. And for traits people believe they can read off a face it reliably delivers a verdict — one that need have nothing to do with the person.

What tells the difference in everyday life is not a question of intensity but a question of time. When looking longer only makes the impression firmer without changing it, you are measuring your own confidence and no longer the other person. A first impression can open a door. It should not change the lock before the person has even spoken.