One Label, Different Errors: What a Memory Test Can Predict

Someone reports remembering a word that was never shown. In another task, the same person describes a detail learned afterward as part of an earlier event. Both answers are wrong. Does that resemblance mean the person is equally vulnerable to every kind of memory error? The label false memory can gather different outcomes under one heading. Establishing whether performance in one task actually predicts performance in another requires more than a shared name.

A 2013 study by James Ost, Hartmut Blank, Joanna Davies, Georgina Jones, Katie Lambert and Kelly Salmon addresses that predictive question. The same participants complete a task involving related word lists and another involving misleading information supplied after an event. The researchers can consequently distinguish two questions: do both tasks produce errors on average, and do the people who make more errors in one also make more in the other? A broad label becomes a testable relationship between measured performances. [1]

The word that fits but was not there

The word-list procedure is known as DRM, after Deese, Roediger and McDermott. Participants encounter words connected in meaning. Those words are selected to evoke another word, a critical lure, that is not actually presented. During a later recall or recognition test, someone may report the lure as belonging to the original list. This is not an arbitrary addition. The erroneous answer fits the semantic organization of what was studied.

The experiment uses five lists of fifteen words each. Each word appears on a computer screen for one second, followed by a one-second blank interval. Participants write down the words they remember after each list. A later recognition test presents studied words, the related but unpresented lures, and unrelated new words. Keeping these categories separate prevents every false positive from becoming the same undifferentiated error count.

A participant can accept many related lures without accepting unrelated new words at the same rate. Semantic fit can make one kind of answer especially compelling. The task therefore cannot be read only as a verdict that someone's memory is good or bad. Researchers can examine what is recognized, how readily the person accepts an item, and how well old and new information are discriminated. These different measures make different aspects of performance visible.

A second account after the event

The event task begins with a two-minute, silent security-camera recording of an incident in a jewelry store. Participants later receive misleading information about selected details. In one condition, it arrives through a written statement presented as another witness's account. In another, an experimental collaborator inserts matching, rehearsed statements into a discussion after the event. Instead of semantic connections within a list, this task introduces a subsequent source of event information.

Six critical details are used. Each participant receives misleading information about three, with the other three serving as comparison items. Their assignment is counterbalanced across participants. A particular detail's inherent difficulty should therefore not automatically account for the difference between misled and comparison responses. The researchers examine whether the supplied misinformation enters what people report about the original recording.

The order of the procedures matters. Participants first watch the recording and complete the presentation and free-recall portion of the word task. Misinformation is then introduced. Afterward they complete word recognition and both free-recall and multiple-choice recognition tests for the event. The event tests are repeated a week later. The word task is presented to participants as an unrelated intervening activity, although the analysis reported here uses it to examine the relationship between the two kinds of performance.

That arrangement does not mean every detail of the larger experiment is fully reported in this paper. The authors explain that the data came from a broader investigation and present the information relevant to the relationship under study. The appropriate focus is the comparison they actually analyze, not an invented reconstruction of every question or interaction participants experienced.

Two effects do not imply one ranking of people

The sample includes one hundred and twenty University of Portsmouth undergraduates, mostly psychology students. Both tasks produce the expected patterns of error. In the word-recognition test, participants correctly recognize 85 percent of studied words. They falsely recognize 68 percent of related lures, compared with 3 percent of unrelated new words. These are average response proportions for item categories. They do not say that 68 percent of participants possess an entirely false memory system.

The event task also produces misinformation effects. Within this arrangement, misleading information has a stronger influence on recognition than on free recall, and the written statement has a stronger effect than the collaborator's discussion. That finding belongs to the particular materials and procedures used. It is not sufficient to declare every real conversation safer than every written account.

The decisive analysis concerns covariation across individuals. When the researchers compare their word-task measures with their misinformation measures, none of the relevant correlations is statistically significant. The average of thirty comparisons is minus 0.01, extremely close to zero. This is not an average drawn from thirty separate experiments. It summarizes correlations between different performance measures within the reported study.

Showing that a group produces both word errors and event errors is different from showing that those errors concentrate to similar degrees in the same individuals. The experiment establishes the first pattern without supporting the second. Accepting an unpresented but related word does not operate in these data as a general indicator of accepting subsequently supplied event misinformation. The inference that two error-producing tasks must measure the same susceptibility lacks the predictive relationship it needs.

What remains when a correlation is absent?

The researchers also consider why a genuine relationship might fail to appear. Unreliable measurements can weaken an observed correlation, and insufficient statistical power can prevent detection. They examine internal consistency for some word measures and the relationship between immediate and delayed event scores. Not every measure permits the same reliability calculation; the paper explicitly notes quantities for which that calculation could not be made meaningfully. The authors judge that measurement unreliability and inadequate power are not plausible explanations for the near-zero overall pattern they obtained.

This assessment is important, but it does not prove that no relationship could ever occur under other conditions. An absent behavioral correlation is not a direct demonstration of two completely separate brain mechanisms. The discussion considers alternatives involving semantic association, external suggestions, and monitoring the source of information. A task constructed to elicit a particular error may not reveal a stable susceptibility applying equally across contexts. That possibility remains a question for further investigation rather than a measured inner trait.

There is also a practical distinction between measuring an error and using that measurement as a substitute for another judgment. A laboratory task can reliably produce interesting behavior without becoming a universal assessment of a person's memory. In this study, the researchers do not merely observe an illusion and then assume its wider diagnostic reach. They test whether the needed relationship exists in the same participants, using several ways of scoring performance.

The evidence-based conclusion is not that memory should always be trusted or always be doubted. It is that one test should not acquire an unrestricted meaning merely because its errors share a familiar label with errors elsewhere. In these data, falsely recognizing a semantically related word and incorporating post-event misinformation do not provide interchangeable rankings of individuals. Understanding a memory error begins by asking how information entered the task, what response was required, and which relationship was actually demonstrated. The wrong answer matters, but the route to it matters too.

Sources

1. Ost J, Blank H, Davies J, Jones G, Lambert K, Salmon K. (2013). False Memory ≠ False Memory: DRM Errors Are Unrelated to the Misinformation Effect. PLOS ONE 8(4): e57939. Open primary research article.
2. PLOS. Licensing, attribution, and reuse policy.

Reviewed October 14, 2026. The research, © 2013 Ost and colleagues, is credited under its Creative Commons Attribution license. This independently written English explanation reorganizes the material and adds interpretive discussion. No direct quotations, video, or tables are reproduced. Raw data were not reanalyzed; the article does not recommend a personal memory assessment or a witness-reliability test.