The Confident Witness

For most of the twentieth century, courts treated eyewitness testimony as among the strongest evidence available. A witness who had seen the event, who identified the defendant without hesitation and who expressed certainty was regarded as close to conclusive. The subsequent history of DNA exoneration has made that position untenable. Of the first several hundred convictions overturned by DNA evidence in the United States, roughly seven in ten involved a mistaken eyewitness identification, and in a substantial proportion of those cases the witness had been highly confident.

The psychological explanation rests on a distinction that intuition resists. Memory is not a recording that is played back. It is reconstructed at the moment of recall, assembled from fragments, and each reconstruction can incorporate material that was not present in the original experience. Elizabeth Loftus demonstrated in a celebrated series of experiments that the wording of a question could alter what participants later reported seeing. Those asked how fast cars were going when they "smashed into" each other estimated higher speeds than those asked about cars that "hit" each other, and were more likely, a week later, to report broken glass at a scene where there had been none.

Confidence complicates matters further, because confidence is itself malleable. In a witness's initial identification, confidence and accuracy do correlate reasonably well: a witness who says "I am certain" at the first viewing of a line-up is meaningfully more likely to be right than one who hesitates. What destroys this relationship is everything that happens afterwards. A witness who is told, casually, that they picked the same person as another witness becomes markedly more confident — and this inflation occurs without any change in the underlying accuracy of the memory. By the time the case reaches trial, months later, the witness may express total certainty about an identification they made tentatively.

This finding has produced an unusual convergence between researchers and courts, because it points to a remedy rather than merely a problem. If confidence is informative at first and corrupted later, then the measure that matters is the confidence recorded immediately, in the witness's own words, at the moment of identification, before any feedback of any kind. Several jurisdictions now require exactly this.

Other procedural reforms follow the same logic. Line-ups administered by an officer who does not know which person is the suspect prevent unintentional cues. Presenting the members of a line-up one at a time, rather than together, discourages the witness from choosing whoever most resembles their memory — what researchers call relative judgement. Instructing the witness explicitly that the offender may not be present removes the assumption that an answer is required.

These reforms are modest, cheap, and supported by evidence, which raises the question of why adoption has been so uneven. Part of the answer is institutional inertia. But part of it, researchers suggest, is that the reforms reduce the number of identifications made, and a system under pressure to secure convictions experiences that as a cost. The reforms do not make witnesses remember better. They make the record of what a witness remembered harder to contaminate, which is a more modest aim and a more achievable one.