Rejecting a suggestion and recovering an event are different outcomes

Identity and status. D. Stephen Lindsay, Misleading Suggestions Can Impair Eyewitnesses' Ability to Remember Event Details. Journal of Experimental Psychology: Learning, Memory, and Cognition 16(6), 1077–1083 (1990). Peer-reviewed primary human experiment; accepted 25 May 1990. DOI: 10.1037/0278-7393.16.6.1077; author-hosted publisher-layout PDF; author publication listing. Accessed 22 September 2026.

Factual core

The analyzed sample comprised 136 students; another 37 tested students were randomly deleted to balance cells. Participants viewed slides, heard later descriptions containing three incorrect details and three generic control descriptions, and answered six cued-recall questions 48 hours after the slides. Both conditions received repeated warnings that no correct test answer had appeared in the later narrative. Conditions bundled narrative timing, voice, lighting, posture, and imagery versus verbal-repetition instructions. Excess reports of the misleading alternatives were larger in the low-discriminability condition; the condition-by-item interaction was significant. Correct event-detail recall was lower for misled than control items overall, with no significant interaction. These are distinct behavioral effects. The paper acknowledges that participants need not have experienced suggestions as visual recollections and that timing could explain the condition difference. It favors retrieval interference over trace destruction while leaving the mechanism unresolved. No direct trace measure, confidence rating, autonomous action, or later recovery test was reported. Methods–Discussion, pp. 1079–1082.

Protocol and analysis audit

Assignment and denominator. Students participated in classroom/laboratory groups of 5–24; groups followed a predetermined condition/counterbalancing schedule. Individual random assignment and a group-level uncertainty analysis are not described. The retained 136 comprised seventeen participants in each of eight counterbalancing cells, hence 68 per acquisition condition. A footnote says the pattern also held among all 173 tested participants, without supplying that analysis's numerical results. Random deletion for balancing is disclosed; this is not the qualitative exclusion procedure in Schiller. The six critical objects and their alternative versions were counterbalanced across misled, control, and suggested roles.

Exposure. Seventy-nine slides depicted an office theft, each shown for four seconds with one second between slides. A common female-voice accompaniment named critical objects generically. The later narrative gave incorrect specific details for three objects and generic descriptions for the other three. Thus the control was another description of the event, not absence of postevent information.

Consequently, narrative-to-test delay changes from approximately two days to minutes. The comparison does not identify the isolated effect of perceptual source similarity, imagery, or temporal separation. The author explicitly recognizes this on p. 1082.

Warning and success criterion. The experimenter read test instructions aloud, explaining that the narrative was mostly accurate but had contradicted some slide details. Each tested detail either had not been specified in the narrative or had been specified incorrectly. The central sentence was: “There is no question on this test for which the correct answer was mentioned in the story.” These instructions were rephrased and repeated several times. Participants were to answer from the slides and reject answers remembered from the later narrative. They were urged to attempt every question and promised $1 for at least five of six correct answers; all received $1 afterward. This was cued recall, not a forced choice between the original and misleading alternatives. Blank responses occurred. There was no separate source-identification response establishing each person's compliance or source knowledge.

Observed categories. Table 1 reports these proportions; each participant had three items of each type. On control items the designated misleading alternative had not actually been supplied, so its report provides a baseline for that response.

Rows omit unanswered questions and therefore do not sum to one. The paper calls other answers guesses; that response category alone does not identify the underlying generation process.

• Misleading alternatives: condition-by-item interaction F(1,134)=9.21, MSe=.05, significant at the stated .05 threshold. Low-condition misled/control contrast F(1,134)=24.64; high-condition contrast F<1. This includes a direct interaction, not merely a comparison of two significance labels. The high-condition nonsignificance is not equivalence or proof of perfect warning adherence.
• Correct event details: pooled misled/control means .42/.49, F(1,134)=4.07, MSe=.08; condition-by-item interaction F<1. The table has lower misled recall in both conditions, but this is not two separately established significant effects or proof of equal impairment. No confidence interval is reported.
• Selected subgroup: among high-condition participants giving no misleading-alternative response, correct recall was .43 misled versus .60 control, t(34)=2.59, p<.02, implying n=35. Selection on observed test responses does not independently verify intact source discrimination or identify the mechanism of the remaining errors.

Original causal interpretation and limits

The strongest directly measured result is a change in the distribution of reported answers under these histories and instructions. The warning makes knowingly repeating narrative answers less attractive, and the high-condition pattern supports its effectiveness. It is still an inferential check: reduced excess reporting could coexist with source uncertainty, conservative withholding, or compensating error processes. Footnote 5 explicitly assumes comparable acceptance of the instructions across conditions. No direct phenomenological measure distinguishes uncertainty about origin from experiencing a vivid but false visual recollection.

Reduced correct recall does not require replacement of the original content. A two-source system can preserve both alternatives while the later one competes for retrieval; identifying it as inadmissible need not immediately recover the original. A reporting policy can also withhold an accessible candidate whose source remains uncertain. Distinguishing these accounts requires additional constraints, not relabeling a low recall score as damaged storage. The author's preference for retrieval impairment is compatible with this limit.

The control comparison leaves another distinction open: misleading descriptions can reduce performance relative to generic descriptions even if generic descriptions facilitate access. Footnote 6 acknowledges this alternative and argues it is unlikely in the high condition, but there is no no-narrative arm to estimate that component directly. Likewise, classroom assignment and the absence of reported cluster handling limit certification of the nominal between-condition uncertainty. These issues narrow the estimand; they do not erase the reported response contrasts.

Original extraction/transfer implication — proposal

A source filter should be evaluated on both rejection of misleading material and recovery/use of the original event. Fewer intrusions alone could conceal more omissions or unrelated errors. Preserve those outcomes instead of turning them into one accuracy label.

For a neutral controlled extension, add generic-description and no-description controls alongside misleading descriptions; vary source distinctiveness separately from imagery instructions. Specify the timing cells and which interval is held fixed—event-to-narrative, narrative-to-test, and event-to-test intervals cannot all vary independently. Obtain source/confidence measures in separate or counterbalanced test schedules, since those probes may change later retrieval. Fit explicit competition, source-uncertainty, and overwrite models to training observations and test differences on new cues and delays. Recovery under a new cue can reject a specified complete-loss model; lack of recovery cannot establish universal erasure.

For recipients, hold acquisition history fixed when releasing or withholding additional observations, and keep original truth/source records evaluator-only in extraction conditions. Compare forecasts of future human response distributions with the recipient's own actions under matched continuations; score original-event success separately. An artifact that retains both accounts and rejects the later account may be a better historian without reproducing the person's errors. Such a result tests a bounded functional capability, not transfer of subjective recollection. No participant session, reconstruction analysis, or raw-data reanalysis was performed here.

Reading and archive record

Read the complete seven-page article: introduction and competing explanations, Methods, Results, Discussion, all seven footnotes, Table 1 and its note, and references. Visually inspected PDF pages 1, 4, 5, 6, and 7, including the complete protocol, warnings, sole table, statistical contrasts, limitations, and acceptance date. There are no figures. No supplementary material was identified in the article or checked author listing; none was read. The cited primary studies were not independently read for this note.

The author-hosted PDF returned HTTP 200 with normal TLS verification and a valid PDF header; it parses as seven pages. The default sandbox's initial request failed DNS resolution; the approved public download succeeded without an authentication or certificate bypass. Readable text was extracted with pdftotext -layout; the embedded text layer has occasional spacing/OCR artifacts, so consequential passages and all table entries were checked against the page images. Copyright is ©1990 American Psychological Association; no open reuse license was identified. Local PDF, text, and provenance.