When the Correction Arrives in Another Room

Learning that an event's first explanation was wrong does not necessarily end the explanation's influence. Someone may have read the correction and still draw inferences that rely on the earlier account. Is the correction insufficiently integrated into the event representation, or is it present in memory but not retrieved when needed? Changing where the information is encountered offers one way to investigate those possibilities. Yet a change of context need not produce the predicted effect. Its consequences also depend on what happens between the correction and the later test. [1]

In a 2022 study, Jasmyne Sanderson, Simon Farrell and Ullrich Ecker placed a room change between misinformation and its retraction. The first experiment used physical rooms; the second used virtual reality. Their prediction was that separating the two pieces of information with a spatial event boundary would hinder integration and increase subsequent misinformation reliance. The findings did not support that prediction. In the virtual experiment, the numerical pattern pointed in the other direction, but the evidence's strength and the design's remaining ambiguities need to accompany that observation. [1]

Two possible points of failure

The study began from an account in which people construct a mental model of an event and update it as information arrives. If the correction and original explanation belong to different event segments, bringing them together might become difficult. This explanation concerns integration when the correction is encountered. A retrieval account instead emphasizes which information becomes accessible when an inference is made. Both the original claim and its correction may be stored, while only one guides a particular response. [1]

The room manipulation was designed to test a behavioral prediction arising from these accounts. It was not a test of whether a doorway literally erases a person's entire memory. The outcome concerned later reasoning after information had been learned in the same or different spatial contexts. Inference scores served as an indirect indicator. Neither successful integration nor moment-to-moment retrieval of the correction was directly observed in the experiments. [1]

A fictional report in physical rooms

Experiment 1 randomly assigned 112 University of Western Australia students to shifting or nonshifting conditions. Each participant read one of four fictional event reports. The first six of twelve messages supplied the initial account; the later six included its retraction and alternative information. One scenario concerned an emergency airplane landing initially attributed to extreme weather and later explained by mechanical failure. This was experimental material, not a verified report about an actual aviation incident. [1]

Participants in the nonshifting condition read both parts in one room. Those in the shifting condition moved before reading the second part. Lighting, decoration and the computer presentation differed between rooms. A word-finding distractor lasted longer in the nonshifting condition to allow comparable overall intervals once travel time was considered. Both laboratory groups subsequently moved to a third room for a distractor and the final questions. The test was therefore not simply administered wherever the initial information had been learned. [1]

A no-retraction comparison came from a separate online MTurk sample, rather than a third randomly assigned group from the same student population. The authors described it as a rough baseline. Comparing that online sample with the laboratory groups cannot isolate retraction perfectly from differences in population and testing environment. After exclusions, there were 55 participants per condition, for a total of 165. Retaining the separate origin of the control is essential to interpreting what the comparison can show. [1]

Remembering a statement is not endorsing it

Four multiple-choice questions assessed memory for report details. Misinformation reliance was measured with an open-ended summary and six rating items. Higher combined inference scores meant greater reliance on the original incorrect account. In coding the summary, mentioning the retraction or alternative explanation counted against misinformation reliance even if the original explanation was also mentioned. Someone could therefore remember the first claim without treating it as the valid explanation. [1]

Experiment 1 did not support the predicted increase in reliance after moving rooms. The Bayes factor for the comparison between laboratory conditions was 3.02 in favor of no difference. Numerically, the shifting group had lower reliance, but that numerical direction was not presented as strong evidence for a reversed effect. Memory for report details was also similar. The experiment showed that the proposed integration obstacle did not emerge in the measured reasoning outcome; it did not determine precisely which processing stage explained the absence. [1]

A virtual detective and corrective emails

Experiment 2 made context changes more central to the task. Participants wore a virtual-reality headset and acted as detectives investigating a fictional burglary. Eight clues in a living-room crime scene included information implicating the homeowner's son. Eight subsequent emails retracted both critical and noncritical clues; the critical correction indicated that the son was out of town. These messages described an invented case, not evidence concerning a real suspect. [1]

One group read the emails at the crime scene, while another read them in a police-station foyer. Participants remained standing in the laboratory while using the virtual environment; they were not walking between real laboratory rooms. Both groups eventually answered questions in a virtual interrogation room. The article reports that the design and analysis plan were preregistered. Of 127 participants, two were excluded for low memory performance, leaving 125. The primary analyses nevertheless used the first 100 completions as specified in the plan, with full-sample analyses additionally reported. [1]

For those primary analyses, the mean inference score was 4.24 when the correction appeared in the same location and 3.43 when it appeared after a shift. This direction was opposite to prediction. However, the two-sided Bayes factor of 1.95 provided only weak evidence for any difference. The stronger reported factor of 14.36 concerned evidence against the specifically predicted direction. Strong evidence against a directional prediction is not automatically strong evidence for an unrestricted effect in the opposite direction. [1]

A separate decision did not follow the rating pattern

An exploratory question asked whom participants would charge: the son, another suspect, or nobody because evidence was insufficient. The study reported support for no difference between conditions in this choice. It would therefore be misleading to infer that changing rooms prevented wrongful accusations. Related outcome measures can capture different responses. Memory-detail scores were also similar between the two groups, so the rating pattern should not be redescribed as a general improvement in remembering every part of the report. [1]

The path from correction to test also differed. Participants who read the emails at the crime scene subsequently moved to the foyer and then the interrogation room, crossing two spatial boundaries after the retraction. Those who read the correction in the foyer crossed only one boundary before the test. This arrangement complicates the attempt to distinguish integration of the original information and correction from accessibility of the correction during later reasoning. [1]

The researchers discuss the possibility that fewer postcorrection boundaries helped retrieval of the retraction relative to the misinformation. That interpretation fits the pattern, but the relevant process was not directly measured. Moving between the two experiments also changed reports, environments and some outcome measures. Treating every difference as a single isolated room effect would remove distinctions the study itself identifies. [1]

The conclusion the evidence permits

The authors additionally note that a condition receiving no misinformation at all would help establish continued influence more formally. A rating score above zero is not a perfect natural baseline by itself. Neither that limitation nor the unexpected pattern supports claims that corrections never work or that incorrect information is impossible to overcome. The experiments examined a particular prediction about context, not the effectiveness of every correction strategy. [1]

Their defensible conclusion is narrower and useful: a spatial boundary did not support the prediction that separating misinformation from retraction would impair integration and increase reliance on the old account. The findings instead leave room for explanations involving relative accessibility at test, while the unequal number of postcorrection transitions prevents a clean mechanistic verdict. Where a correction is encountered, what intervenes afterwards, and how its use is measured are separate questions. Keeping them separate makes the study informative even when the original prediction fails. [1]

Sources

[1] Sanderson, J. A., Farrell, S. and Ecker, U. K. H. (2022). Examining the role of information integration in the continued influence effect using an event segmentation approach. PLOS ONE, 17(7), e0271566. Open access, Creative Commons Attribution; the authors' work is credited in independently written explanations and interpretations. CC BY 4.0. No quotations were used; supplementary data, preregistration files and analysis code were not independently audited or rerun.