4.1.2 - Eyewitness accuracy and the cognitive interview

From event to testimony

An eyewitness does not play back an event like a recording. A witness first encodes information during the event, then stores it, then retrieves it later when they give an account. At each stage, accuracy can be affected by the way the event is experienced and by information encountered after the event.

The updated AQA requirement focuses on four linked ideas:

The theme across the topic is reconstructive memory: people combine the original memory trace with later cues, expectations and social information. That does not mean all eyewitness testimony is useless. It means testimony is evidence that needs careful collection and careful evaluation.

For AO1, know the named factors and the main research procedures. For AO2, apply them to realistic police or courtroom scenarios: who said what, when, and how that could change the witness's answer. For AO3, evaluate whether the evidence comes from artificial laboratory tasks, naturalistic field cases, or meta-analyses, and whether it supports a clear practical recommendation.

Leading questions

Question wording can introduce details that the witness did not independently recall.

Leading questions affect accuracy because they add information after the event. Sometimes they produce a response-bias effect: the witness gives the answer implied by the wording without their underlying memory changing. Sometimes they may produce a memory-substitution effect: the misleading wording becomes integrated into the memory, so the witness later experiences the suggested detail as part of what happened.

Loftus and Palmer (1974) is the core study for this factor. In their first experiment, 45 students watched short films of traffic accidents and then estimated speed. The critical question used different verbs, such as "hit", "bumped", "collided", "contacted" or "smashed". More intense verbs produced higher speed estimates. A separate second experiment used a different sample; participants returned one week later and were asked whether they had seen broken glass, although there was none in the film. Participants originally asked with "smashed" were more likely to report broken glass than those asked with "hit". The first experiment therefore establishes an effect on speed reports. The delayed false-detail result is also consistent with reconstruction, but is not direct access to a memory trace; participants’ interpretation of the later question remains relevant.

Loftus and Zanni (1975) gives a second, simpler wording example. After seeing a film, participants were asked either "Did you see a broken headlight?" or "Did you see the broken headlight?" The definite article made the broken headlight sound as if it existed, and this increased false recognition of a detail that was not present.

For example, imagine a witness is asked, "How fast was the getaway car going when it crashed into the van?" That question assumes there was a crash. A less leading alternative would be, "Describe what happened between the car and the van." The first wording may inflate speed or damage estimates; the second invites recall without planting a detail.

Loftus and Palmer's work has strong control because the wording of the critical question was manipulated while the film was held constant. That makes cause and effect easier to infer. However, the study used filmed accidents and student participants, so the anxiety, personal involvement and legal consequences were unlike many real crimes. There is also a demand-characteristics issue: participants may have guessed that the verb mattered and adjusted their answer without genuinely changing memory. The best evaluation is balanced: the research demonstrates that wording can distort reports, but the size and mechanism of the effect may differ in real investigations.

Post-event discussion

Co-witness conversations can alter a later account by supplying another person's details.

Post-event discussion matters because real witnesses are not always isolated. They may talk at the scene, message each other, watch media reports, or hear another person's confident version of events. The key process is memory conformity: one witness's report shifts towards another person's account.

Gabbert, Memon and Allan (2003) tested this with a realistic co-witness design. Participants watched a video of the same event from different perspectives, so each person saw some details that their partner could not see. Some pairs discussed the event before later recalling it individually; control participants did not discuss it. A large proportion of those who discussed the event later reported details they could only have acquired from the co-witness discussion.

Two explanations are especially useful for AQA:

For example, if Aisha saw the suspect wearing a dark jacket but Leo says, confidently, "It was definitely a green hoodie," Aisha may later include a green hoodie in her statement. She may not be lying. She may have misattributed Leo's detail to her own perception, or she may have relied on Leo because he sounded more certain.

Gabbert et al. improved ecological validity by using actual discussion between paired witnesses, rather than only written misleading information. That makes the study especially relevant to real post-event contamination. However, the event was still a video and the stakes were low, so it cannot fully reproduce the stress and consequences of a serious crime. A useful practical implication is that police should separate witnesses before formal interviews and warn them not to discuss the event.

Anxiety

Anxiety affects how attention is distributed during an event, which can change later recall.

Anxiety is the most debatable factor because it can affect accuracy in different directions. High anxiety may reduce accuracy if attention narrows onto the most threatening or central feature of the scene. This is often called weapon focus.

One classic teaching example is Johnson and Scott’s reported weapon-focus demonstration: participants in a waiting-room scenario saw a man leave either with a harmless object or with a blood-stained knife after a more alarming exchange. Those exposed to the weapon condition were less accurate at identifying the man. The important limitation is that the weapon condition changed more than one thing at once: it involved a weapon, apparent violence, surprise and higher arousal, so it is difficult to isolate anxiety as the only cause.

Loftus, Loftus and Messo (1987) developed weapon-focus evidence by tracking eye movements while participants viewed slides of a customer pointing either a gun or a chequebook. Participants looked more at the weapon and were less accurate in identifying the customer. This supports the attention-narrowing explanation.

However, anxiety does not always reduce accuracy. Yuille and Cutshall (1986) studied real witnesses to a shooting in Vancouver. Thirteen witnesses were re-interviewed four to five months later, and their accounts remained generally accurate. Witnesses also resisted misleading questions, and reported stress did not show a negative effect on subsequent memory. This is a valuable counterpoint because it used a real event with genuine emotional involvement. However, stress was reported after the event and not experimentally assigned. Witness distance, attention and involvement may differ, so the association cannot establish that greater anxiety caused better accuracy.

One proposed explanation is an inverted-U relationship: performance may improve with arousal up to an optimum and then decline at very high arousal. This is a useful hypothesis, not a rule established for every eyewitness task. The relationship is not simple. Moderate arousal may increase alertness and improve memory for central details, while extreme stress may impair recall, especially for peripheral details and facial identification. Deffenbacher, Bornstein, Penrod and McGorty (2004) meta-analysed studies of high stress and found overall support for the view that high stress negatively affects both identification and recall accuracy. This gives broader evidential weight than relying on one isolated study.

For example, if a victim remembers the colour and shape of the knife very clearly but cannot identify the offender's face, that is consistent with weapon focus. If a witness remembers the sequence of a shocking event clearly months later, that may fit Yuille and Cutshall's field evidence that real-life involvement can sometimes preserve central details.

The cognitive interview

The cognitive interview uses supported retrieval techniques to obtain a fuller eyewitness account.

The cognitive interview was developed by Geiselman and Fisher and colleagues from research on memory retrieval. Its aim is not to make witnesses more confident. Its aim is to help them access more accurate information without contaminating the memory.

The original cognitive interview uses four main techniques:

Geiselman, Fisher, MacKinnon and Holland (1985) compared the cognitive interview with standard police interviewing and hypnosis. Participants watched police training films of simulated violent crimes and were interviewed 48 hours later by experienced law-enforcement personnel. The cognitive interview produced more correct information than a standard interview without increasing incorrect information.

Later reviews are also useful. Memon, Meissner and Fraser's (2010) meta-analysis and study-space review found that cognitive interview procedures generally increase correct details, although there can be a small increase in incorrect details. That gives a strong AO3 judgement: the cognitive interview is effective, but it requires training, time and careful questioning.

The enhanced cognitive interview adds social and communication elements such as building rapport, reducing interruptions, transferring control to the witness, and using open questions. This links to the broader investigative principle that good interviews should help witnesses retrieve memories rather than pressure them to please the interviewer.

For example, if a witness says, "I cannot remember anything else," the interviewer might ask them to close their eyes and mentally recreate the street, the lighting and the sounds before reporting every detail. A poor interview would ask, "Was the man wearing a red jacket?", because that risks planting information.

The cognitive interview has clear practical value because it translates cognitive psychology into police procedure. Its limits are practical: it takes longer than a standard interview, requires training, and may be difficult with very distressed witnesses or young children unless adapted carefully. Also, gaining more information is not the same as gaining only accurate information, so statements still need corroboration.

Judging eyewitness evidence

Judging eyewitness research requires distinguishing the amount recalled, its accuracy and the strength of a causal explanation.

Evaluate what was measured before judging a finding. The number of correct details is not the same as the proportion of details that are correct. In an illustrative interview comparison, a standard interview yields 20 correct and 5 incorrect details; a cognitive interview yields 30 correct and 10 incorrect details. The second obtains 10 more correct details, but also 5 more incorrect details. Its accuracy proportion is lower: $30/40 \times 100 = 75\%$ compared with $20/25 \times 100 = 80\%$. Neither measure alone establishes that every part of the interview is better. Check correct and incorrect details separately and corroborate claims independently.

To assess cause and effect, identify what was manipulated and what else changed. Keeping the filmed event and recall delay constant while varying question wording helps isolate a leading-question effect. If different films are used as well, differences in visibility could explain the result. By contrast, field studies of anxious witnesses have realism but cannot normally separate anxiety from distance, involvement, rehearsal or attention through random allocation.

An improvement should address the identified problem: use the same event across wording conditions to control event differences, or have independent raters compare reports with verified event details using the same scoring rules. Neutral prompts and independent witness statements reduce contamination; they do not guarantee accuracy.

Evidence about reports does not automatically demonstrate a changed underlying memory. Speed estimates may reflect response bias, while later false-detail reports offer additional evidence consistent with reconstruction. Even then, competing explanations remain possible. A careful judgement specifies the effect, explains the mechanism and qualifies the inference rather than declaring all eyewitness memory reliable or unreliable.

Quick recap

Leading questions and co-witness discussion can alter reports through suggestion, social influence and source confusion. Anxiety can affect attention, with high stress often impairing identification and recall, but the outcome depends on what is measured. Cognitive-interview techniques support retrieval without inviting invention. More correct details and a higher accuracy proportion are different outcomes.