Contribution
The first multi-experiment attempt outside the originating lab cited in the record. Experiments 1 and 2 rebuild the Radvansky paradigm in head-mounted virtual reality with identical rooms, a door placed unpredictably on the left or right of each room, and 30 shift and 30 no-shift transitions per block. Experiments 3 and 4 move to a real hallway divided by curtains. In the authors’ words: “Across this series of experiments, we observed no significant effect of doorways on forgetting” (Abstract, Results).
Key insight
The effect is not robust to small changes of paradigm, and where a doorway mattered here it showed as confusion rather than loss: “the reduction in d’ after a shift was predominantly due to there being a higher false alarm rate for negative probes, rather than a reduced hit rate for associated and dissociated probes” (Results, Experiment 2). “Thus, under working memory load, memory was more susceptible to interference after moving through doorways” (Abstract, Results).
Assumptions
- Conceptual replication. The design keeps the original’s probe logic (colour–shape names for the carried object, the set-down object, or a recombination) but changes immersion (headset), movement (passive: the participant is moved, seated), room appearance (all identical) and trial balance.
- Identical rooms isolate the doorway. “Any forgetting could only be attributed to boundaries rather than salient changes in context or visual processing” (Experiments 1 and 2, Aim).
- A power analysis from a pooled prior effect of d = 0.66 called for 27 participants (Methods, Participants). Bayes factors (JASP, default Cauchy prior) are used to support nulls.
- Curtains make event boundaries in Experiments 3 and 4, by analogy with an imagined-navigation study (Lawrence & Peterson 2016).
Key results
- Experiment 1: VR, no load (40 recruited, 9 stopped for motion sickness, 2 excluded; N = 29). Hits about 95% in every cell; false alarms about 8% (Table 1). No effect of shift on d′ (p = .354) or bias (p = .720); Bayes factors for no shift effect 3.87 (associated) and 3.59 (dissociated). Accuracy was at ceiling; among the lower-performing half (N = 14) still no effect, though the null for dissociated probes was inconclusive (BF01 = 1.77). Responses were faster to associated than dissociated probes (1.689 against 1.813 s), so the associated/dissociated effect replicated.
- Experiment 2: VR, counting backwards aloud during movement (63 recruited, 14 motion sickness, 4 excluded; N = 45). d′ lower after a shift (1.315 against 1.569), F(1,43) = 8.37, p = .006, and a bias toward “yes” (p = .004). Sign tests: the shift effect was significant only for negative probes (33 of 45 participants worse after a shift, p < .001, BF10 = 68.6), not for associated (p = .136, BF01 = 4.32) or dissociated (p = 1, BF01 = 6.12) probes. No response-time effect of shift.
- Experiment 3: video of a hallway walk (N = 26) and Experiment 4: walking it (N = 26), with two curtain boundaries crossed twice, a counting task, and recall of the positions of 16 butterfly photographs in a grid. No shift effect: mean differences 0.41% and 0.40%, BF10 = 0.281 and 0.290. In Experiment 4 accuracy was higher in whichever condition came first. About 64% of participants said the shift condition felt harder.
- On earlier studies. Hit rates of 80–99% with few, unbalanced trials (“6 shift trials and 12 no shift trials per probe” in Pettijohn & Radvansky’s environmental-effects study) mean a 5–10% difference is one or two trials: “Behavioural patterns generated from few observations are susceptible to spurious artefacts, especially when the same experimental procedure is used for all participants” (Discussion).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Passive VR movement through identical rooms with a door does not lower recognition hits for carried or set-down objects | moderate (Bayes factors 3–6 for the null) | Experiments 1 and 2 |
| C2 | Under a concurrent counting load, a doorway raises false alarms to recombined probes and lowers d′ | moderate, single experiment | Experiment 2 |
| C3 | A curtained boundary, watched or walked, does not affect memory for a spatial arrangement of 16 items | moderate as a null for that task | Experiments 3 and 4 |
| C4 | The doorway effect is not robust to small changes in paradigm | moderate as stated; the changes are several at once | the four experiments against the prior literature |
| C5 | Active navigation and visual context change probably augment the effect | weak: speculation | Discussion, citing Pettijohn & Radvansky 2018 on passive interaction halving the effect |
Concepts
- Location updating effect / doorway effect / event horizon effect: three names for one finding, per the Background.
- Hits and false alarms; d′ and C: signal-detection measures replacing the original error rates; a negative probe recombines the colour of one remembered object with the shape of the other.
- Event boundary: here a door in a virtual room, or a curtain in a hallway.
Connections
It cites the whole doorway series from the Radvansky lab (LIT-1392, LIT-1384, LIT-1385 and five others) and two independent positive studies not held here: Lawrence & Peterson (2016, imagined walking) and Seel et al. (2019, recall versus familiarity). It describes LIT-1384 as showing that returning to the original location “impairs memory further”, which that paper does not report (return was no different from shift). For the mechanism it cites the event horizon model and Radvansky & Zacks (2017); it does not cite Zacks et al. 2007 (LIT-1382).
Bearing on the record
- CASE-136. The case’s summary of this paper (“found no significant
doorway effect; under load, false alarms to mismatched probes rose”) is
accurate to the abstract. Three qualifications belong in it:
- Experiment 2 did find a significant doorway effect on d′ and on bias. It was not forgetting in the original’s sense (fewer hits), but more confusion.
- These are conceptual, not direct, replications. Experiments 1 and 2 moved participants passively; Experiments 3 and 4 tested a different memory (spatial arrangement), and the authors offer that and the curtains as possible reasons for the null.
- The record’s view of the replication record is thin. The case’s title says “some replications find no effect”. This paper is the only non-replication the record holds, and it is cited against a literature that also has independent positive studies (Lawrence & Peterson 2016; Seel et al. 2019) the record does not hold. Fairly put: the effect is well replicated within one lab and its paradigm, mixed outside it, and at least sensitive to passivity, load and boundary salience.
- CLAIM-421. Its defeater already allows the doorway case to fall without harm to the claim, and the case’s “What it does not show” says the effect is not shown robust. This paper bears that out. Its one positive finding, interference under load, fits the competition account of LIT-1385 better than the deactivation account CLAIM-421 attaches to the case. Increased false alarms mean the two remembered objects became harder to tell apart, not that one became unavailable. That is the authors’ reading too (“sufficient cognitive interference in the memory system to effectively reduce the discriminability across encoded objects”, Discussion).
- Proposed THEORY-888 cites this paper as its main evidence on the limits of the effect.
- THEORY-886 (event segmentation). The Background puts the puzzle that matters for a prediction-error account: with all events so similar, “it seems somewhat surprising that the doorway effect is so robust across the literature”. If boundaries arise from prediction error, identical rooms should produce weak ones. The authors’ null in identical rooms is consistent with that, though they do not test it.
- No instruction for machine-learning practice.
Limitations
- Movement in Experiments 1 and 2 was passive, which the authors themselves say may weaken the effect. 23 of 103 VR participants stopped for motion sickness.
- Experiments 3 and 4 change both the boundary (curtains) and the memory test (placing 16 photographs), so their null cannot be assigned to either.
- Experiment 1 was at ceiling. The lower-half analysis has N = 14.
- Experiment 2’s text gives N = 45 and its figure caption N = 40.
- No preregistration is mentioned; the counting decrement was adjusted for 13 participants between blocks.
Open questions
- Is the effect a product of active navigation and salient context change, as the authors suspect? An independent test with active movement and identical rooms against distinct rooms would say.
- Do false alarms under load and fewer hits without it reflect one process at two strengths, as the Conclusions suggest?