54 percent — barely better than chance

Asked the open question “How do you tell that someone is lying?”, people in 75 countries and 43 languages give the same first answer: he looks away. A research consortium publishing as the Global Deception Research Team put the question to samples around the world in 2006 and found a single dominant stereotype across all of them — the averted gaze. It is probably the most widely shared belief about lying there is.

Three years earlier, Bella DePaulo and colleagues had assembled what behaviour reveals about deception at all: 1,338 individual estimates covering 158 behavioural cues. The result is a disappointment with a structure. Many of the cues examined were unrelated to deception, or related only weakly. What did show up was narrow and indirect. Liars disclose less, their accounts are less compelling and less often contain the ordinary unevenness and the unusual particulars that mark genuine reports; they leave a more negative impression and seem more tense. These traces grow clearer when much is at stake — especially when a person’s own standing is at issue — and when the lie covers a transgression. No single telltale signal is among them, and certainly not the averted gaze.

What 54 percent means

The meta-analysis by Charles Bond and Bella DePaulo pooled 206 studies with judgements from 24,483 raters. Their abstract names both at once — the rate and its imbalance:

That is barely above the chance level of 50 percent, and it is lopsided. Anyone who assumes truth is right in most cases — and misses the deception for exactly that reason.

Timothy Levine has built a theory on this. The normal state of interpersonal communication is the assumption of truth. We believe first, and it is to that default setting that conversation owes its ability to work at all. It is not a thinking error but a well-calibrated bet — most statements by most people are accurate. The price is a systematic blind spot: on average, lies are noticed less often than truths are confirmed.

Why there are no gifted lie catchers

The obvious hope is that at least some individuals can do it. Bond and DePaulo tested that as well, across 247 samples in which people judged strangers in real time and without aids. Once measurement error is corrected for, almost nothing remains of the differences between judges: the spread in ability lies under one percentage point, and the best judges are no more accurate than a chance mechanism would produce. What people do vary in markedly is how credulous they are — and how credible they appear. The outcome of a credibility judgement depends more on the person speaking than on the person listening.

That places professional experience in a different light too. In research on interrogation, Christian Meissner and Saul Kassin separated two quantities that everyday language merges: the ability to discriminate, and the tendency to decide in one direction. Neither training nor years on the job improved discrimination. Both shifted the tendency, however — trained and experienced investigators answered “deception” more often. They do not become more accurate; they become more suspicious and more certain.

No muscle knows the truth — nor does nervousness

Microexpressions are very brief facial displays. As a means of detection, the evidence for them is thin. Stephen Porter and Leanne ten Brinke found them in around two percent of all expressions examined and in about a fifth of participants — and where they did occur, they did not separate true statements from false ones. That emotions sometimes leave suppressed traces does not turn those traces into a universal lie detector.

The basic problem is not one of measurement precision but of attribution. A person may show fear because they are lying, because they are not being believed, or because much is at stake. The same expression has several causes, and no camera decides which one applies.

What the polygraph measures

The polygraph sharpens the problem technically. It records breathing, pulse, blood pressure and skin conductance. These quantities respond to arousal and attention, not specifically to deception. Control questions, expectations and countermeasures all affect the result. The National Research Council’s report of 2003 evaluated 57 studies and reached a double verdict: in tests concerning a specific incident, the procedure separates lie from truth clearly above chance but clearly below perfection; for security screening, by contrast, the panel found the evidence inadequate. Where offenders are rare, even an apparently good test produces many false suspicions — one would have to cast suspicion on hundreds of loyal employees in order to catch a single real case.

One variant asks a different question, and it performs better. The guilty knowledge test puts no question of guilt but a set of particulars: among several equally plausible statements about a crime scene, one is the right one. A person who does not know it responds to all of them alike; a person who does know it responds differently to one. Gershon Ben-Shakhar and Eitan Elaad evaluated 169 conditions from 80 laboratory studies and found, via skin conductance, a mean effect size of 1.55; in staged crime scenarios 2.09, and in the ten studies that came closest to real conditions of use, 3.12. The catch lies not in the statistics but in the preconditions: non-public details of the crime have to exist, they must not have leaked through the press or the interview, and the findings come almost entirely from the laboratory. The test measures recognition, not lying — it can establish knowledge, not guilt.

Brain imaging and “AI lie detectors” do not close the logical gap either. An algorithm can learn patterns in training data. Without an unambiguous link between pattern and lie, what it learns may be the camera, the language, stress, or the markers of a social group.

The quarrel over the better interview

If behaviour yields little, the question that remains is whether more behaviour can be produced. The cognitive approach attempts that: it raises the mental demand without threatening — an unexpected order of questions, spatial detail, sketches, open follow-ups. The idea is plausible, because lying is more work than remembering. Whether it holds up is one of the more open questions in the field.

From 14 studies with a direct comparison, Aldert Vrij, Ronald Fisher and Hartmut Blank reported a clear advantage: 71 percent overall accuracy against 56 percent, and for lies 67 against 47 percent. Timothy Levine and colleagues contested that analysis in the same journal. Erik Mac Giolla and Timothy Luke recalculated and arrived at 60 percent raw, and 55 percent after correction for publication bias; where a single study set both conditions against each other, the cognitive and the standard interview barely differed — 54.24 against 54.06 percent. The argument is not settled, but it shows how narrow the margin being fought over is.

What helps in every variant is unspectacular. What decides the matter is not whether someone “seems suspicious” but whether their statements fit together and fit independent evidence. Contradictions are instructions to check, not proofs of guilt; honest memory has gaps too.

Fear is not a property of lying. An honest person may be nervous because much is at stake, because they fear authority, or because traumatic experience is being activated. Conversely, a practised deceiver may appear calm. Cultural rules also shape eye contact, pauses and expression. Reading such signals as moral evidence risks a feedback loop: suspicion creates pressure, pressure creates nervousness, and nervousness confirms the suspicion. In the end the interview has produced precisely the behaviour that then counts as proof.

What remains

Of the three candidates in the subtitle, none carries the weight it is advertised with. Eye contact is a prejudice believed worldwide with no finding behind it. Microexpressions are rare and do not discriminate. The polygraph does discriminate better than chance on a concrete question about a specific incident, but it is no instrument for screening entire workforces; and the one physiological method with respectable numbers, the guilty knowledge test, does not answer the question of guilt at all but the question of knowledge — and for that it needs details of the crime that nobody has let slip.

How to recognise it in everyday life: by the direction in which the checking runs. Someone who reads a person collects impressions that cannot be refuted, and grows steadily more certain in the process. Someone who checks a statement collects claims that can be true or false outside the conversation, and stays uncertain until they have looked them up. The second stance feels worse and yields more. The certainty that comes from the other person’s face is almost always one’s own expectation looking back.