Leading questions and the broken glass that was never there
The broken glass was absent from the film but present in some later recollections. In their 1974 experiments, Elizabeth Loftus and John Palmer examined how leading questions affected descriptions of traffic collisions. Their second experiment divided 150 students into three groups. After the same film, one group estimated speed following wording that used smashed; another received hit; the third received no speed prompt. A week later, without another viewing, participants reported whether they had seen broken glass.
Sixteen of the 50 students in the smashed group reported glass, compared with seven in the hit group and six in the control group. The researchers had changed an interviewing detail rather than the collision footage. They proposed that information supplied after an event could combine with information acquired during it, producing a more severe remembered accident. The results did not mean that every participant accepted the suggestion: most people in each group reported no glass.
The FBI Law Enforcement Bulletin later connected leading questions with an especially short interval between experience and interview. It described a 2002 experiment in which children saw a pirate wearing blue-and-white trousers but received prompts suggesting red-and-white trousers. Their responses to misleading prompts were accurate 60% of the time, even immediately after the event. The bulletin interpreted the errors as acceptance of the adult interviewer's suggestion rather than forgetting over a long delay. Its discussion is a policing synthesis of published research, distinct from the original experimental reports it cites.
Elizabeth Loftus and Hunter Hoffman addressed the mechanisms in their 1989 paper on misinformation and memory. They used a hypothetical hardware-store robbery to separate several routes to the same erroneous report. A witness might never notice a stolen hammer, then adopt another person's mention of a screwdriver. Alternatively, the witness might remember both descriptions but trust the other person more. Guessing supplies another possibility. So does interference with a memory that originally included the hammer. An identical final description can conceal different histories of perception and later influence.
With leading questions, accepting misinformation does not always require an existing memory to be erased. Someone who never noticed the original object can acquire a confident recollection from a suggestion alone. Loftus and Hoffman discussed experimental findings in which suggested details were reported quickly and confidently, and they cautioned against attributing all such responses to guessing. They also emphasized that conditions of initial exposure, delay and testing affect the relative contributions of these mechanisms. Memory performance is observable; the precise condition of an underlying memory trace is harder to establish.
The National Institute of Justice translated concerns about leading questions into concrete interview procedures in its 2003 trainer's manual. It recommends beginning with open invitations for witnesses to describe an event in their own words, then using focused prompts when necessary. Its vehicle example distinguishes requesting a description from suggesting that a vehicle was red. Information supplied by another witness can otherwise slip into the investigator's wording and reach someone who never independently reported that feature. The procedure seeks to preserve the origin of each detail while still allowing useful clarification.
The manual also instructs interviewers to caution witnesses against guessing, allow uncertainty and avoid interrupting a recollection. A pause can mean that someone is still searching memory. Investigators are advised to document statements accurately and preserve the witness's own words. Separating witnesses and discouraging discussion of incident details address another route of contamination: people exchanging descriptions before formal questioning. These are official collection safeguards, rather than findings that every departure necessarily produces a false memory. They aim to give volunteered information room to emerge before an investigator supplies assumptions.
Repeated interviewing requires a further distinction. In a 2013 experiment published in PLOS One, Geralda Odinot and colleagues recruited 80 university students and employees to watch a two-minute video of a woman being stalked. Participants received two Cognitive Interviews, with schedules combining immediate interviewing, two-day delays and week-long delays. Across both interviews, people supplied additional correct details. Cumulative correct recall rose from an average of 63.35 details initially to 77.97 when newly recalled details were added. Later interviews could therefore recover information that had not appeared in the first session.
That benefit came with limits. Errors also increased, and newly reported details were less accurate than details repeated across sessions. Groups whose first interview occurred after seven days supplied fewer correct details than groups interviewed immediately. The researchers tested interview timing and interviewer identity within a structured recall method, giving participants another opportunity to describe what they had seen. Their findings distinguish another opportunity to remember from another opportunity to receive suggestions.
- FBI Law Enforcement Bulletin, Malleability of Deception — leb.fbi.gov
- Elizabeth F. Loftus and John C. Palmer, Reconstruction of Automobile Destruction, 1974, original research paper, pp. 585–589 — gwern.net
- Elizabeth F. Loftus and Hunter G. Hoffman, Misinformation and Memory: The Creation of New Memories, 1989, author-hosted text — staff.washington.edu
- National Institute of Justice, Eyewitness Evidence: A Trainer’s Manual for Law Enforcement, 2003, interview procedures — www.ojp.gov
- Geralda Odinot and colleagues, Are Two Interviews Better Than One? Eyewitness Memory across Repeated Cognitive Interviews, 2013, PLOS One — journals.plos.org