KurisutinaIdentity and access. D. Stephen Lindsay and Marcia K. Johnson, The eyewitness suggestibility effect and memory for source. Memory & Cognition 17(3), 349–358. Peer-reviewed primary human experiments. DOI; public publisher PDF, linked by the author publication archive. This is distinct from the authors’ other 1989 paper on reversing exposure order. Accessed 2026-09-22. Local PDF, text, provenance.
Factual core
Two experiments retained 108/117 and 132/136 students after random deletions to balance four cells. Participants were individually randomized within classroom groups. They viewed one office slide, read an accurate or misleading narrative, and completed either picture yes/no recognition or four-category source judgments. Misleading text added eight absent objects. Both experiments showed significant narrative-by-test interactions: misleading-text increases in picture endorsements were much smaller under source testing. Source testing also reduced endorsements of objects genuinely pictured, especially objects appearing in both sources. Experiment2 changed the encoding cover story and test instructions, including asking what participants remembered noticing rather than simply what was present. The authors considered criterion changes, retrieval, and demand characteristics; genuine subjective confusion remained unresolved. These were between-person test comparisons, not demonstrations that the same person’s earlier false claim was corrected. No confidence measure, later recovery test, or independent trace measure was reported. Methods–Discussion.
Measurement and causal audit — original analysis
The intervention changes the task as well as its emphasis. Source-test responses were picture only, text only, both, and neither. Picture attribution combines picture-only and both, providing a reasonable common endorsement outcome with yes/no. However, the richer format also makes remembered text a positive reportable answer. It can redirect retrieval, change the evidential standard for asserting visual experience, or change compliance with perceived demands. The experiment does not isolate any one of these mechanisms. Assigning half of each format’s response categories to picture endorsement equates chance only under uniform category guessing; it does not establish equal guessing behavior or decision criteria.
Do not carry over the paper’s “new/miss” comparison as a matched measure. On the yes/no test, a negative answer means the item was not in the picture—or, in Experiment2, was not remembered there. It does not mean the item was absent from both sources. On the source test, neither explicitly refers to both sources. Calling both responses new collapses different questions. In particular, correctly rejecting a text-only item on yes/no can be labeled a miss by this coding even when the person remembers it from the text. Lower neither responses therefore cannot, by themselves, establish preserved old/new recognition or rule out a response-policy explanation.
Reduced false attribution has a cost here. Rejecting a suggested object is not enough to certify improved historical reporting: the same test also reduces correct picture endorsements. Evaluate the full response distribution, especially correct text-only attribution, false picture attribution, correct picture attribution, and neither. Near-zero misled/control excess under source testing is not zero false attribution, perfect source knowledge, or demonstrated equivalence. The direct interactions nevertheless establish more than a difference between significant and nonsignificant simple effects.
Selection and stimulus scope. Individual randomization within the classroom groups supports a test-format effect for this procedure. Random deletion for cell balance wastes information but is not the same as selecting participants for the desired outcome. One scene and fixed item-role assignments limit stimulus generalization; the authors explicitly identify an item-set baseline difference. A new scene, longer delay, or a contradictory replacement detail could change the result. Changing encoding instructions and test wording together across experiments also prevents attributing their different effect sizes solely to recognition wording.
Local reporting limits. Experiment2’s control/source mean for misleading-item picture endorsement is 1.36 in Table4 but 1.33 in the prose. Experiment1 calls F(1,104)=4.76 and 4.63 significant although its stated threshold is .01; my approximate F-tail calculations give .0314 and .0337. These concern the already problematic supplementary new-response analysis, not the primary interactions. The interactions F(1,104)=13.30 and F(1,128)=8.06 do meet .01. No raw responses were obtained to resolve the discrepancies.
Comparison with the fully read 1990 warning study
The 1990 source note addresses a different limitation. Its participants all received explicit warnings excluding narrative details as correct answers; testing used cued recall after forty-eight hours. Its suggestions contradicted specific original details, permitting separate measurement of misleading alternatives and correct original recall. In 1989, the added absent objects had no corresponding original alternative whose recovery could be scored. Correctly denying that an added object was pictured therefore does not demonstrate recovery of a displaced event detail.
The studies are compatible: changing the reporting task can reduce false picture endorsements, while correctly excluding later information need not recover original content. Neither design uniquely distinguishes lost storage from unavailable retrieval or conservative reporting. The 1990 comparison additionally bundles exposure timing and several source-distinctiveness manipulations; the 1989 within-experiment randomized test contrast more directly manipulates the assessment procedure, but still combines instructions with response options.
Original extraction/transfer proposal
Record the exact question and permitted answers with every memory report. A transferred “no” without its original response contract can acquire the wrong meaning. Evaluate a source filter jointly on intrusion reduction and access to genuine event information; retaining two source records is useful only if the recipient can select and use them appropriately.
A controlled extension could cross instruction emphasis with a common response format, and separately vary retention delay. Distinguish endorsements from confidence and source attribution, accounting for earlier probes changing later retrieval. Hold source acquisition history fixed for paired recipient releases and keep event truth evaluator-only. Test personal response forecasts separately from historically correct recipient actions. Success under a richer test need not establish that an earlier respondent consciously knew the source, nor that a recipient has recovered subjective recollection. These are proposed tests, not results from this paper.
Read and archive audit
Read all substantive main text, both experiments’ complete methods/results, General Discussion, both endnotes, and the complete narrative Appendix. Visually verified all six tables, test instructions, notes, and Appendix against the PDF; there are no figures. References were retained but cited studies were not independently followed. The final-page conference notice is not study content. No supplement, raw data, code, or registration was identified on the checked primary sources. The ten-page publisher scan has an existing text layer; OCR occasionally drops table cells, so consequential values were checked visually. Title/authors/DOI/journal/pages agree with Crossref, PubMed, and the author listing. No open reuse license was verified; the Crossref TDM link is not an open reuse license. No participant sessions or model-transfer experiments were run.