10.6 — Lie Detection, and Why Almost Everyone Is at Chance

Charles Bond and Bella DePaulo pooled hundreds of studies in which people judged whether someone was lying or telling the truth.

Average accuracy: 54 per cent. Chance is 50.

And it did not improve much with expertise. Police officers, customs officials, judges and interrogators performed close to the same level as students. What did differ was confidence — professionals were considerably more sure of themselves, and no more accurate.

That is the starting point for this page, and it is one of the most robustly established findings in applied psychology.

Why the cues do not work

DePaulo and colleagues examined every proposed behavioural cue to deception across the literature. Most had no relationship to lying. The few that did had relationships so weak as to be useless for any individual case.

The specific cues people believe in, and what the evidence says:

Gaze aversion. No reliable relationship. This is the most widely believed cue in the world — surveys across dozens of countries find it is the top-ranked signal nearly everywhere — and it does not work. Some studies find liars make more eye contact, because they know the stereotype and compensate.

Fidgeting. No reliable relationship. If anything, liars sometimes move less, because they are controlling themselves.

Touching the nose or face. No.

Hesitation and pauses. Weak and ambiguous. People pause when thinking honestly.

Nervousness. The core problem with the whole enterprise. An honest person who is frightened of being disbelieved looks exactly like a liar, and the situations where lie detection matters most — an interrogation, an accusation, a job at stake — are precisely the situations that make innocent people nervous. This is called the Othello error, after the man who read his wife's genuine fear as evidence of guilt.

Micro-expressions. Covered in 10.2. Not supported for this purpose, and a large government programme built partly on them was found by auditors to lack evidence of effectiveness.

And the polygraph. Measures arousal, not truthfulness. A scientific review by the United States National Academies concluded its accuracy is insufficient for security screening and that it can be defeated with countermeasures. It is inadmissible as evidence in most courts for good reason, and its main documented effect is that people confess because they believe it works.

Why it is so hard

Three structural reasons, and they explain why better training has produced such modest gains.

There is no behaviour unique to lying. Every proposed cue is also produced by nervousness, fatigue, cultural norm, personality or the mere fact of being scrutinised.

People differ enormously in baseline. Some are naturally fluent and calm; some are hesitant and awkward. Without knowing the individual's normal, no deviation can be read (10.3), and in most real detection situations you do not know it.

And there is a truth bias. People judge most statements as true, which is adaptive — most statements are true — and means that accuracy on lies specifically is usually below chance while accuracy on truths is above it. The 54 per cent average conceals this asymmetry.

What does work, somewhat

The research has moved away from cue-spotting towards changing the interaction, and this line of work is more promising.

Increase the cognitive load. Lying is harder work than telling the truth — you must invent, keep it consistent, monitor the listener, and suppress the truth. Make the task harder and the liar degrades more than the truth-teller.

Aldert Vrij's work has found improvements from techniques such as:

Asking for the account in reverse order. Reconstructing a real memory backwards is difficult and possible; reconstructing an invented story backwards is much harder.

Asking unanticipated questions. A prepared story covers the expected ground. Asking about spatial detail, or about something peripheral, exceeds the preparation.

Asking for more detail, repeatedly. True accounts have more verifiable detail and can be extended; invented ones tend to be thinner and to resist elaboration.

And keeping the person talking. The longer the account, the more opportunity for inconsistency with checkable facts.

These improve accuracy meaningfully in studies and they do not make anybody a reliable lie detector. They shift the odds.

The thing that actually works

Checkable facts.

Not demeanour. Not tone. Whether what they said corresponds to something you can verify.

Was he where he said he was. Does the number match the invoice. Did the person he named confirm it. Is the timeline consistent with the messages.

This is unglamorous and it is what actually resolves the question, and it is why investigators who are good at their jobs spend their time on records rather than on faces.

And the related version for everyday life: consistency over time. What someone said in March against what they said in August. Nobody keeps an invented account perfectly consistent over months, and this is the most accessible verification available to an ordinary person.

The special problem of people you know

Accuracy with close relationships is not better and may be worse, which surprises people.

Why. You have a strong model of them and a strong motivation not to detect. Motivated reasoning (1.4) is at its strongest precisely where the answer matters most.

And there is a documented pattern in long relationships where accuracy at detecting a partner's deception declines over time, apparently because confidence in knowing them replaces attention.

The practical consequence: if you are worried about being deceived by someone close, your instinct is not the instrument. Verifiable facts are.

What your gut is actually doing

There is a persistent belief in unconscious detection — that some part of you knows even when you cannot articulate it. The evidence for it is weak and the studies claiming it have not replicated well.

But something real sits underneath the belief, and it is worth naming precisely.

Your discomfort around a person is real data about the interaction. It is frequently detecting something — inconsistency, evasion, a mismatch between warmth and behaviour, a change from their baseline.

What it is not doing is identifying the cause. The feeling says something is off. It does not say what, and "he is lying" is one candidate among several — including that he is anxious, that he dislikes you, that he is embarrassed, or that he reminds you of somebody.

So the correct use of the gut signal is as a prompt to investigate, never as a verdict. Something feels wrong here, so I will check the checkable things is a sound procedure. Something feels wrong, therefore he is lying is 54 per cent accurate.

The costs of getting it wrong

Worth stating, because the popular framing treats lie detection as a pure win.

False accusation is extremely damaging, to the relationship and to the person, and it is difficult to withdraw.

Suspicion changes behaviour. Treat someone as though they are lying and they become defensive, inconsistent and nervous — which you then read as confirmation (2.6's self-fulfilling loop).

And in institutional settings, unreliable detection methods produce systematic injustice, falling hardest on people whose demeanour differs from the expected norm — the anxious, the autistic, the culturally different, the frightened.

What to do with this

Stop trying to read whether someone is lying from how they look. You cannot, nobody can, and the confidence is manufactured.

Ask more and better questions instead. Unanticipated detail, the account again from a different angle, more elaboration. Not as an interrogation — a genuinely curious "tell me more about that" does most of the work.

Verify what is verifiable. One check beats a hundred impressions.

Watch consistency over months, which is the version of verification available in ordinary life.

Treat your discomfort as a prompt, not a verdict. Something is off, so I will look rather than something is off, so I know.

And be careful with the accusation. The base rate of your being right is much lower than it feels, and the cost of being wrong is high. Ask about the thing rather than about their honesty — "help me understand the timeline" rather than "I think you're lying to me" — and let the facts do the work.

Next: 10.7 covers the judgements formed before any of this — the first two seconds, what they actually get right, and why they are so hard to overturn.