The question
Witnesses to accidents are routinely asked about speed. Elizabeth Loftus and John Palmer, at the University of Washington, suspected that the answer might depend partly on how the question was asked. They wanted to know two things. Does the wording of a question shift a witness’s estimate? And if it does, is that only a matter of how people answer, or does the question actually change what they later remember?
How it was done
Experiment 1. Forty-five students watched seven short films of traffic accidents, each between 5 and 30 seconds long, cut from driver-education films. After each film they described what they had seen and then answered a set of questions. The key one asked roughly how fast the cars were going when they “hit” each other. For different groups of nine students, the verb was swapped for smashed, collided, bumped or contacted. Nothing else changed.
Four of the films showed staged crashes at known speeds (20, 30, 40 and 40 mph), which let the researchers compare estimates with reality.
Experiment 2. A new group of 150 students watched a single film, under a minute long, showing a multiple-car accident that lasted about four seconds. Afterwards they described the accident and answered questions. Fifty were asked how fast the cars were going when they smashed into each other, fifty when they hit each other, and fifty were not asked about speed at all.
A week later, the students came back and, without seeing the film again, answered ten more questions. Hidden among them was a simple one: did you see any broken glass? The film showed none.
What came out
The verb moved the numbers. Average speed estimates in Experiment 1 rose with the strength of the verb:
- smashed: 40.8 mph
- collided: 39.3 mph
- bumped: 38.1 mph
- hit: 34.0 mph
- contacted: 31.8 mph
The difference was statistically significant. People were also poor at judging real speed: the crashes filmed at 20 mph and at 40 mph drew similar estimates, between about 36 and 40 mph. The authors concluded that the wording of the question explained much more of the variation than the actual speed did.
The verb planted glass. In Experiment 2, the “smashed” group again gave higher speed estimates than the “hit” group (10.46 vs 8.00 mph). A week later, 16 of 50 people in the “smashed” group said they had seen broken glass, compared with 7 of 50 in the “hit” group and 6 of 50 in the group not asked about speed. In proportions, that is 32% against 14%.
This was the more important result. Loftus and Palmer had noted that the speed findings could simply reflect a response bias: unsure people leaning towards the number the verb suggested. The glass question was about something else entirely, asked a week later. Even when they compared people who had given similar speed estimates, those who heard “smashed” were still more likely to report glass.
Their explanation was that memory combines two kinds of information: what a person perceived at the time, and what they learn afterwards. Over time the two merge, and people can no longer tell which detail came from where. As they put it, “All we have is one ‘memory.’” A question that says the cars smashed adds the idea of a violent crash, and in a violent crash broken glass seems likely.
What really happened (replications and critiques)
The verb effect is less solid than its fame suggests. The 1974 study is very widely cited, but the speed effect has not been easy to reproduce. A 2024 paper by Kevin Holmes, Lena Kassin and Stephen Flusberg lists earlier attempts: some found no difference between “smashed”, “hit” and “bumped” (Read and colleagues, 1978; McAllister and colleagues, 1988; Raghunath and colleagues, 2021), while others found the effect only under certain conditions, such as unfamiliar accident settings or with 7th and 8th graders but not younger children or college students.
Holmes and colleagues then ran three preregistered online studies with 1,204 US participants, who watched a 9-second clip of a two-car accident. They found no overall effect of the verb on speed estimates, even when contrasting the most extreme pair, “smashed” and “contacted”. In exploratory analyses, the effect was weaker among people who wrote longer descriptions of the accident first, and also weaker among people who saw a bigger difference in severity between the two verbs. These hints still need testing. Their studies did not look at broken glass, so the second, more important finding of 1974 has not been tested at a comparable scale.
A large, multi-lab direct replication has been proposed by Zwaan and Eerland, with Loftus’s support; in September 2025 it was still awaiting funding.
The misinformation effect holds up better. The 1974 study became one of the founding experiments of a much larger research programme. In 1978, Loftus, D. G. Miller and H. J. Burns showed 195 students slides of a car stopping at either a stop sign or a yield sign, then asked a question that mentioned the wrong sign for some of them. Given a choice between two slides later, people who had heard accurate information picked the right one 75% of the time; those who had been misled, only 41%. Loftus later defined the misinformation effect as a change in memory reports, usually for the worse, after misleading information. A recent replication by Sera Wiechert and colleagues, with Loftus as a co-author, again found that misled participants picked the correct sign less often.
Memories of whole events. In 1995, Loftus and Jacqueline Pickrell went further. With help from a relative of each participant, they presented 24 people with three true childhood events and one invented one: getting lost in a shopping mall as a child. Six of them (25%) went on to report a full or partial memory of the invented event. The authors presented this as proof that such memories can form, not as an estimate of how often.
A preregistered replication in Ireland by Gillian Murphy and colleagues (2023), with 123 participants, judged that 35% had formed a false memory by the final interview, most of them partial. Yet only 14% of participants themselves said they remembered the event, while a further 52% said they believed it had happened.
That gap is now at the centre of a live dispute. Reanalysing the transcripts, Bernice Andrews and Chris Brewin (2024) found that people judged to have false memories recalled on average fewer than two of six suggested details, many never explicitly recalled being lost, and about half described experiences that could have been real and that could be told apart from the invented one. By their stricter criteria, the false memory rate fell to about 4%. Wade, Loftus and colleagues (2025) replied that memory is reconstructive, so a false memory will naturally mix real fragments with suggestion, and that even a low rate matters in practice. Others (Zsofia Szojka and colleagues, 2025) argue that the coding of false memories is too unreliable and that a lost-in-the-mall story is too plausible to say much about false memories of rarer, more serious events.
Without any suggestion at all. Also in 1995, Roediger and McDermott revived a 1959 method by Deese. People heard lists like table, sit, legs, seat, soft, desk…, all linked to a word that was never said: chair. They later recalled the missing word 40% of the time, and on a recognition test they “recognised” it 84% of the time, almost as often as words they had really heard. This DRM task shows that memory can produce confident errors from associations alone, with no misleading question needed.
Ethics
The 1974 study involved mild deception: as far as the paper shows, participants were not told that the wording of one question was the real object of study, and the broken glass question was a test of memory for something that never appeared. The paper does not describe a debriefing. The films were standard driver-education material, and there is no sign of distress. A similar design would very likely be approved today, with a debrief at the end.
The lost-in-the-mall method raises more serious questions, because it relies on relatives helping to convince participants of something false about their own childhood. In the 2023 replication, participants were debriefed, and when told that one story was invented, 85% of them correctly picked out the lost-in-the-mall event.
What it means today
The best-supported lesson is the broad one: memory is rebuilt each time we recall something, and information that arrives later, from questions, conversations or news coverage, can blend into it. That is why the 1978 sign study and the many studies that followed it shaped how researchers think about police interviews and eyewitness evidence.
The narrower claim, that one verb reliably adds about 9 mph to a witness’s estimate, is now uncertain.
A few practical habits follow, though they go beyond what these studies tested:
- If you saw something important, write down what you remember before talking to others or reading about it.
- When you ask someone what happened, use neutral questions (“what did you see?”) rather than ones that suggest the answer.
- Feeling sure is not proof. Confident memories can still contain details that came from somewhere else.
Would you have passed?
Coming soon. Watch a short animated collision and estimate the speed; the question’s verb is chosen at random. Later you’ll be asked what you saw, and at the end you’ll see your answers next to the 1974 results and the 2024 replications. One caveat: once you know what the experiment is about, its effect is likely to be much weaker.
Questions people ask
What did the Loftus and Palmer car crash study find?
People watched films of traffic accidents and were asked how fast the cars were going when they 'smashed', 'collided', 'bumped', 'hit' or 'contacted' each other. Estimates rose with the strength of the verb, from 31.8 mph for 'contacted' to 40.8 mph for 'smashed'. In a second experiment, people asked the 'smashed' question were more than twice as likely to say a week later that they had seen broken glass, though there was none.
Has the Loftus and Palmer study been replicated?
Results are mixed. Several later studies found no effect of the verb on speed estimates, or found it only in some groups or settings. In 2024, three preregistered online studies with 1,204 participants found no overall effect. The broader finding that misleading information after an event can change memory reports has held up much better.
What is the misinformation effect?
It is the change in what people report remembering, usually for the worse, after they are exposed to misleading information about an event. In a classic 1978 study, people who were told about the wrong road sign picked the correct sign 41% of the time, against 75% for people given accurate information.
What is the 'lost in the mall' study?
In 1995 Elizabeth Loftus and Jacqueline Pickrell, with help from relatives, presented 24 people with a made-up story of getting lost in a shopping mall as a child; 6 of them came to report a full or partial memory of it. A 2023 preregistered replication in Ireland reported 35%, but a 2024 reanalysis argued that the true rate of genuine false memories was far lower, and the debate continues.
Sources
- Loftus, E. F., & Palmer, J. C. (1974). Reconstruction of automobile destruction: An example of the interaction between language and memory. Journal of Verbal Learning and Verbal Behavior, 13(5), 585–589. doi:10.1016/S0022-5371(74)80011-3
- Loftus, E. F., Miller, D. G., & Burns, H. J. (1978). Semantic integration of verbal information into a visual memory. Journal of Experimental Psychology: Human Learning and Memory, 4(1), 19–31. doi:10.1037/0278-7393.4.1.19
- Loftus, E. F., & Pickrell, J. E. (1995). The formation of false memories. Psychiatric Annals, 25(12), 720–725. doi:10.3928/0048-5713-19951201-07
- Roediger, H. L., III, & McDermott, K. B. (1995). Creating false memories: Remembering words not presented in lists. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21(4), 803–814. doi:10.1037/0278-7393.21.4.803
- Loftus, E. F. (2005). Planting misinformation in the human mind: A 30-year investigation of the malleability of memory. Learning & Memory, 12, 361–366. doi:10.1101/lm.94705
- Holmes, K. J., Kassin, L., & Flusberg, S. J. (2024). When does suggestive language shape memory for car accidents? Assessing the role of elaboration and pragmatics in a classic framing effect. Proceedings of the Annual Meeting of the Cognitive Science Society, 46. escholarship.org/uc/item/0wk430xs
- Wiechert, S., Verschueren, A., Ben-Shakhar, G., Pertzov, Y., Loftus, E. F., & Verschuere, B. (2026). The misinformation effect: A contemporary replication and extension of Loftus et al. (1978) to investigate its underlying mechanisms. Journal of Experimental Psychology: Learning, Memory, and Cognition, 52(7), 1225–1241. doi:10.1037/xlm0001529
- Zwaan, R. (2025, September 8). Memory, misinformation, and the need to replicate. Substack. rolfzwaan.substack.com/p/memory-misinformation-and-the-need
- Murphy, G., Dawson, C. A., Huston, C., Ballantyne, L., Barrett, E., Cowman, C. S., Fitzsimons, C., Maher, J., Ryan, K. M., & Greene, C. M. (2023). Lost in the mall again: A preregistered replication and extension of Loftus & Pickrell (1995). Memory, 31(6), 818–830. doi:10.1080/09658211.2023.2198327
- Andrews, B., & Brewin, C. R. (2024). Lost in the Mall? Interrogating judgements of false memory. Applied Cognitive Psychology, 38(6), e70012. doi:10.1002/acp.70012
- Wade, K. A., Riesthuis, P., Bücken, C., Otgaar, H., & Loftus, E. F. (2025). Still lost in the mall: False memories happen and that's what matters. Applied Cognitive Psychology, 39(1), e70028. doi:10.1002/acp.70028
- Szojka, Z. A., Block, S., & La Rooy, D. (2025). Is it time to leave the shopping mall behind? Measurement flaws, plausibility, and external validity of false memory research. Applied Cognitive Psychology, 39(3), e70083. doi:10.1002/acp.70083
- Andrews, B., & Brewin, C. R. People likely aren't as susceptible to false memories as researchers thought. Scientific American, 333(1). www.scientificamerican.com/article/people-likely-arent-as-susceptible-to-false-memories-as-researchers-thought/