Psychology/FACS & Observation/Limits of inference

Language:EnglishEspañolPortuguês (BR)

FACS & Observation

The limits of facial inference

This is the page most likely to cost us a sale, and it is the reason the section exists. The commercial market around reading faces substantially over-states what the evidence supports.

What the evidence reasonably supports

  • Facial movements can be recorded reliably by trained coders — that is, coders can agree about what moved.
  • Some facial movements are associated, at group level and above chance, with some reported states in some contexts.
  • Context contributes at least as much to an observer’s judgement of an expression as the face itself does.
  • Observers are consistently more confident in their reading of a face than their accuracy justifies, and training in short courses tends to raise confidence more than accuracy.

What it does not support

Claim Why it fails
A specific expression reliably indicates a specific emotion in an individual. Association at group level does not license confident inference about one person on one occasion. Variation between individuals, contexts and cultures is large.
Micro-expressions reveal concealed feelings. The reliability of detecting brief movements outside laboratory conditions is poor, and the inference from movement to concealed content is a further, unsupported step.
Facial cues detect deception. Meta-analytic work finds most behavioural cues to deception close to chance. Training programmes typically improve confidence, not accuracy.
Automated facial analysis infers emotion at scale. Such systems classify movement, then apply a mapping to emotion labels that inherits every limitation above — and adds sensitivity to lighting, skin tone, head pose and training-set composition.

Four claims to refuse

  • “Ninety-three per cent of communication is non-verbal.” A misreading of two small laboratory studies on single spoken words. It has never meant what it is used to mean.
  • “This method is used by intelligence and law-enforcement agencies.” Adoption is not validation. Ask for the validation evidence, the population, and the purpose it was validated for.
  • “You can be trained to spot a liar in two days.” No. You can be trained to feel certain in two days.
  • “The face does not lie.” The face moves. The interpretation is where the error lives.

Why we are this blunt

Confident misreading of faces has real consequences: in interviews, in classrooms, in disciplinary processes, in custody and immigration decisions, and in clinical settings. The cost of the error falls on the person being read, not on the reader.

If we sold certainty here, we would sell more. We would also be doing harm.

What to do instead

  • Record movement, context and timing — and record it repeatedly.
  • Ask the person. Their account outranks your inference.
  • Use patterns across many observations, never a single expression.
  • Where a decision affects somebody materially, do not base it on facial inference at all.
  • Use the observation method instead — it answers a narrower question and answers it far more defensibly.

Educational scope. Everything published on PMA Science Psychology is provided for education, study and professional documentation. Nothing here is a diagnostic instrument, a psychometric test, or a substitute for assessment, treatment, supervision or emergency care by a qualified professional.

If you are at risk right now, contact your local emergency number or a crisis line immediately. Read our boundaries and safety statement.