The Confident Witness

Of the first several hundred convictions overturned by DNA evidence in the United States, around seventy per cent involved a mistaken eyewitness identification. Not a corrupt one. A sincere, confident, mistaken one: very often from a witness who had looked at the defendant in court and said, without hesitation, that they were certain.

Everything in the previous chapter applies here, with a jury attached.

The relationship between confidence and accuracy is the piece most people get wrong, and getting it wrong in either direction is a problem. The naive view is that a certain witness is a reliable one. The overcorrection, common after the first wave of research, is that confidence means nothing at all. The evidence supports a more specific claim: confidence measured immediately at the first identification, under clean conditions, is a reasonably good guide. Confidence measured later, in court, after months of interviews and preparation, is nearly worthless, because confidence inflates with rehearsal and feedback while accuracy does not.

The feedback part is the sharpest finding. Witnesses who were told after picking someone out that they had identified the suspect subsequently remembered having a better view, having been paying closer attention, and having been more sure at the time. Nothing about the original observation changed. A single sentence from an officer altered the entire remembered experience of the event, retroactively.

The other findings are about the conditions, and they are largely about how the question is asked. A simultaneous line-up (six faces at once) invites comparison: which of these looks most like the person? That is a relative judgement, and it will always have a winner, including when the actual culprit is absent. Sequential presentation, one face at a time with a decision on each, pushes toward an absolute judgement against memory rather than against the other five, and reduces false identifications. The trade-off is real: it lowers correct identifications somewhat too, and the field argued for two decades about whether the exchange is worth it.

Double-blind administration is less contested. If the officer running the line-up knows which one is the suspect, they will cue it, without any intent to and usually without knowing they have. The fix, an administrator who does not know, costs nothing and is now standard practice in many jurisdictions, which is a rare case of psychological research changing procedure directly.

Then there are the effects nobody can design around, only account for. Weapon focus: when a weapon is present, attention narrows onto it, and memory for the face gets worse, which is exactly backwards from what the justice system needs, since the crimes where identification matters most are the ones where a weapon was present. Cross-race identification is reliably less accurate than same-race. Stress, contrary to a widespread belief that danger sharpens the memory, degrades it. Brief exposure, poor light, and distance do what you would expect.

The system's response has been to treat an identification as evidence with known failure modes rather than as testimony to be believed or disbelieved. Courts increasingly instruct juries about the confidence-accuracy relationship; identification procedures are recorded; the witness's confidence is now documented in their own words at the moment of the first identification, precisely because that number will never be available again.

It is worth being clear about what this research does not say. It does not say witnesses lie, or that identification is useless. It says something narrower and more corrosive: that a person can be completely honest, completely certain, and completely wrong, with no internal signal available to them or to anyone watching that distinguishes that state from being right. Nothing in the courtroom performance carries the information. It has to come from the procedure, or it does not come at all.