6 ideas
In Loftus and Palmer's experiment, viewers watched the same car-crash film, then were asked how fast the cars were going when they 'smashed' versus 'hit' each other. The 'smashed' group gave higher speed estimates and, a week later, were more than twice as likely to falsely report seeing broken glass. A single verb in a question changed both the judgment and the stored memory of the event.
Information supplied after an event, such as a suggestive question, a news report or another witness's account, gets absorbed into the original memory and is later reported as something actually seen. Witnesses cannot reliably tell which details came from the event and which came from later sources, so contaminated memories feel just as authentic as genuine ones.
Eyewitness memory can be analyzed in three stages, and each has its own distortions. Acquisition is limited by stress, exposure time, lighting and expectation. Retention decays over time and absorbs new information. Retrieval is shaped by how questions are worded and by the social pressure of the setting, so errors can come in at any point before testimony.
A witness's certainty is only weakly related to whether their identification is correct, yet jurors treat confident testimony as highly persuasive. Confidence can be inflated by repetition, feedback and rehearsal while accuracy stays the same, so courts that weigh certainty heavily systematically overbelieve mistaken witnesses.
When a weapon is present during a crime, the witness's attention locks onto it, and their encoding of the perpetrator's face and other peripheral details gets worse. High arousal narrows perception, so the most frightening crimes often produce the least reliable facial identifications.
Memory works less like a videotape that is replayed and more like a story rebuilt each time it is recalled, drawing on fragments, expectations and later suggestions. Seen this way, even vivid, detailed and emotional recollections are evidence that needs corroboration, not direct access to the past.