The Journal28 June 20268 min read
One Verb Is Enough to Move a Memory
How Loftus showed that memory reconstructs rather than retrieves
In 1974 Elizabeth Loftus and John Palmer showed 45 students films of traffic accidents and asked all of them the same question — except for one word. Those who heard that the cars had “smashed” into each other estimated the speed at an average of 40.8 miles per hour. With “collided” it was 39.3, with “bumped” 38.1, with “hit” 34.0 — and those who heard the cars had “contacted” each other came to 31.8. All of them had watched the same film. Nine miles per hour separated the first verb from the last.
The everyday theory of memory is a video library: what happens is recorded, filed away and played back on demand; forgetting is a lost tape, remembering a replay. Almost everything memory research has established over fifty years contradicts that picture — and no body of work contradicts it more thoroughly than Loftus’s. Her findings sit in file 009 of this archive; this piece places them in the larger question of what remembering is, if it is not playback.
From the question to the remembered detail
A shifted estimate is not yet a shifted memory. So Loftus and Palmer ran a second experiment with 150 participants and waited a week before asking: did you see broken glass? In the “smashed” group 16 of 50 said yes, in the “hit” group 7 of 50, and in the control group, which had been asked no speed question at all, 6 of 50. There was no broken glass in the film.
The question had become a source. Whatever information arrives after the event — a verb, a follow-up question, a newspaper report, a conversation with another witness — is worked into the memory as though it had been part of the event all along. The misinformation effect is today among the best-supported findings in the field, confirmed in hundreds of studies and meta-analyses. It is not a weakness of particular people but the working method of a system that stores meaning rather than tape.
Implanting whole events
The second step was more radical. Can one not only shift details but add whole events? In 1995 Loftus and Jacqueline Pickrell presented 24 participants with four childhood stories: three genuine ones, confirmed by the family — and one invented. Lost in a shopping mall as a child, found in tears, returned to the family by an older person. After two interviews six of the 24 took on the invented episode, some of them with self-generated details that had appeared in no script.
Later work by other groups reproduced the pattern with other events. A pooled analysis of eight implantation studies with 423 participants found false memories in 30.4 percent — and in a further 23 percent or so something that has to be distinguished from them: the firm conviction that the event had taken place, without any memory of it. Belief and memory are two states, and a procedure can produce the one without touching the other. The rates depend heavily on criterion and method; the spread is part of the truth of this finding.
The framework: source monitoring
Why does this work? The most robust explanation comes from Marcia Johnson’s source-monitoring framework. Remembering, on this account, is not a replay but an attribution: memory delivers fragments — images, sentences, feelings of familiarity — and a second process decides where they came from: lived, heard about, dreamt, imagined. That verdict on origin is error-prone, and it grows more so the more often something is imagined. What is vividly imagined accumulates precisely the features — richness of detail, vividness, emotional colour — by which the source check recognises the lived. Picture a scene three times and by the fourth you hold a piece of memory with a forged return address.
This also accounts for the most disquieting side finding in the literature: false memories do not feel false. Confidence, detail and emotional involvement — the three marks by which lay people and courts hope to recognise genuine memories — barely separate genuine memories from implanted ones. Ulric Neisser’s Challenger study, in file 007 of this archive, had shown the same for flashbulb memories: agreement with the protocols written the day after was thin, the confidence of those remembering was high, and the two had nothing to do with each other.
The prehistory of this idea is older than the debate usually knows, and it runs through this archive. Frederic Bartlett had English students retell an unfamiliar story — published in 1932 — and watched it bend, with each retelling, towards the expectations of the tellers. He called it schema; it is the original version of reconstructive memory. His sentence on it stands to this day:
Remembering is not the re-excitation of innumerable fixed, lifeless and fragmentary traces. It is an imaginative reconstruction
Hermann Ebbinghaus had founded the opposing position fifty years earlier: syllables without sense, so that nothing could be reconstructed. Both were right, each for his own material — the forgetting curve holds where meaning is absent; reconstruction governs where meaning is in play. Loftus’s work stands in the Bartlett line and gave it what it had lacked: controlled experiments with countable errors.
The real case: the memory wars
In the 1990s the question left the laboratory. In therapies that used suggestive techniques to search for “buried” memories of abuse — hypnosis, guided imagery, interpretive pressure — clients produced memories that tore families apart and ended up in court; Loftus became the most fiercely attacked expert witness in the field. The dispute was fought as the “memory wars”, and its bitterness makes sense only because both sides were defending something true: that abuse is real and common — and that memories can be manufactured.
The laboriously won middle ground of today is written into the guidelines for assessing testimony: memories reported spontaneously deserve to be taken seriously; memories that arise only under suggestive procedures are worthless as evidence — which says nothing about whether what was reported happened. The decisive difference lies not in the content of the account but in how it came about. It is one of the few debates in the field whose outcome has changed procedure in courtrooms: instructions on witness memory, documented first interviews, double-blind line-ups.
The limits belong in this piece too, and they are well mapped. The misinformation effect shifts details reliably; implanting whole events by no means always succeeds, and not with everyone. Central, highly emotional experiences are more resistant than peripheral details; repetition, the authority of the source and the plausibility of the event all raise the success rate of the distortion.
And the most spectacular claim in the literature was itself cut back. In 2015 Julia Shaw and Stephen Porter reported that in 70 percent of their participants they had produced, within three sessions, the memory of having committed a crime as an adolescent. In 2018 Kimberley Wade, Maryanne Garry and Kathy Pezdek recoded the transcripts against stricter criteria and arrived at 26 to 30 percent. Shaw replied in the same issue and defended her coding scheme; the argument is open. What survives even the strict analysis is substantial — a quarter to a third. Only the headline failed the check. That is the replication crisis in miniature, and it speaks for this literature, not against it.
| Test | Year | Scope | What came of it |
|---|---|---|---|
| Bartlett, retellings | 1932 | an unfamiliar story, retold repeatedly | bending towards the tellers' expectations |
| Loftus & Palmer, the verbs | 1974 | 45 students | 40.8 against 31.8 miles — depending on one word |
| Loftus & Palmer, the glass | 1974 | 150 participants, a week later | 16 of 50 against 6 of 50; there was no glass in the film |
| Neisser & Harsch, Challenger | 1992 | protocol against account 2½ years later | agreement thin, confidence high |
| Loftus & Pickrell, shopping mall | 1995 | 24 participants | 6 of 24 took on the invented episode |
| Scoboria et al., mega-analysis | 2017 | 8 studies, 423 participants | 30.4 % false memories, a further 23 % mere belief |
| Shaw & Porter, then Wade et al. | 2015 / 2018 | the same protocols, coded more strictly | 70 per cent became 26 to 30 |
What remains
The misinformation effect is secure: whatever is added after an event can migrate into the memory of the event, and a single word in a question is enough. It is equally secure that whole episodes can be implanted — in a substantial share of people, not in everyone. The number remains contested: exactly where the line runs between a false belief and a false memory, and how one codes it, decides between rates of a quarter and seventy percent. Anyone quoting a single percentage has settled on a coding scheme without saying so.
In everyday life the pattern shows up as an order effect: when someone mentions a detail only after hearing it elsewhere, the source can no longer be disentangled — and confidence is no substitute for disentangling it. Remembering is a workshop report, not an archival extract: serviceable as a rule, always reconstructed, malleable at the edges. Where wording and detail matter, only the early record helps; investigating authorities have institutionalised this, diarists always knew it. And where someone offers absolute confidence as proof, files 007 and 009 of this archive are worth a look: confidence is a feeling about a memory, not a property of its truth.
Sources, and why they are here
Bartlett, F. C. (1932). Remembering: A Study in Experimental and Social Psychology. Cambridge University Press.
The original version of reconstructive memory, half a century before Loftus — and the source of the quotation on p. 213. What it lacked were countable errors.
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.
Both of the article's experiments: the five verbs with their mileage figures and, a week later, the broken glass that was never in the film.
Neisser, U., & Harsch, N. (1992). Phantom flashbulbs: False recollections of hearing the news about Challenger. In E. Winograd & U. Neisser (eds.), Affect and Accuracy in Recall. Cambridge University Press.
Shows the same for the form of memory held to be especially reliable — and that confidence and accuracy have nothing to do with each other there.
Johnson, M. K., Hashtroudi, S., & Lindsay, D. S. (1993). Source monitoring. Psychological Bulletin, 114(1), 3–28.
The framework that explains why the distortion works at all: remembering is an attribution of origin, and origins can be confused.
Loftus, E. F., & Pickrell, J. E. (1995). The formation of false memories. Psychiatric Annals, 25(12), 720–725.
The step from the shifted detail to the whole event: the invented shopping mall, six out of twenty-four, some with details of their own devising.
Volbert, R. (2004). Beurteilung von Aussagen über Traumata: Erinnerungen und ihre psychologische Bewertung. Huber, Bern.
The German-language line in statement psychology from which the article's middle position comes: what is decisive is not the content of a statement but the history of how it arose.
Shaw, J., & Porter, S. (2015). Constructing rich false memories of committing crime. Psychological Science, 26(3), 291–301.
The most spectacular figure in the literature — 70 per cent — and thereby the occasion for the re-examination that followed three years later.
Scoboria, A., Wade, K. A., Lindsay, D. S., Azad, T., Strange, D., Ost, J., & Hyman, I. E. (2017). A mega-analysis of memory reports from eight peer-reviewed false memory implantation studies. Memory, 25(2), 146–163.
The sturdiest overall view: 423 participants, 30.4 per cent false memories — and the separation of belief from memory that the marginal note is about.
Wade, K. A., Garry, M., & Pezdek, K. (2018). Deconstructing rich false memories of committing crime: Commentary on Shaw and Porter (2015). Psychological Science, 29(3), 471–476.
The same protocols, coded more strictly: 70 per cent becomes 26 to 30. The dispute is not about data but about the coding scheme.
Shaw, J. (2018). How can researchers tell whether someone has a false memory? A reply to Wade, Garry, and Pezdek (2018). Psychological Science, 29(3), 477–480.
The reply in the same issue. It belongs here because the article treats the dispute as open and not as settled.