Contribution
Extends the location updating effect of Radvansky & Copeland (LIT-1392) to a less immersive display (17-inch monitors) and to a real three-room laboratory, and tests an encoding-specificity account by sending people back into the room where they picked the object up. It also states the event horizon model’s account of the effect: an object carried into a new room is held in two event models, which compete at retrieval.
Key insight
Going back does not undo the cost. “In Experiment 3, the original encoding context was reinstated by having a person return to the original room in which objects were first encoded. However, inconsistent with an encoding specificity account, memory did not improve by reinstating this context” (abstract, p. 1632). What predicted errors was “not the number of spatial shifts that disrupt memory, but the number of new areas entered, consistent with the competitive retrieval aspects of the event horizon model” (p. 1640).
Assumptions
- The event horizon model (p. 1633), five components, three used here: events are segmented and “different event models are created with people processing one at a time”; information in the current event is foregrounded; and retrieval that must pick one of several models holding the same element suffers interference, “a kind of a fan effect”. “Note that, for simplicity, we assume that there can only be one event model active at one time.”
- Rooms individuate event models. Each room entered is a new model; returning makes the original room one of the models again rather than reinstating it, on the authors’ reading.
- The double-shift and return conditions are matched in distance and number of shifts, differing only in whether the final room is the original (p. 1638).
Key results
- Experiment 1: smaller displays (N = 55; 48 probes, half after a shift). Errors rose after a shift for positive probes (.12 to .19) and negative probes (.07 to .22), both p < .001; A′ .97 against .93, p < .001. No response-time effect for positives, marginal for negatives (Table 1).
- Experiment 2: real rooms (N = 60, 15 in each of four path orders). Three lab rooms, six trials; on each, six blocks were boxed, carried to the next table across a room or into the next room, and tested after two minutes of arithmetic as a distractor. Errors rose after a shift for positives (.23 to .28, p = .02), marginally for negatives (.18 to .21, p = .06); A′ .86 against .82, p = .03. Response times n.s. (Table 2). “Information associated with a prior location became less available even though it continued to be task relevant” (p. 1638).
- Experiment 3: return and double shift (N = 48; 66-inch display; 64 probes). A′: no shift .94, shift .92, return .90, double shift .86. No shift was better than each of the others (shift p = .03); shift and return did not differ (p = .20); return was better than double shift (p = .02). For error rates, no shift against shift was only marginal for positives (p = .08) and n.s. for negatives (p = .23) (Table 3). “Of most importance here was that there was no improvement by returning to the original context” (p. 1641).
- Table 4: pooled pattern. Error rates for five experiments (2006 Experiment 2, both 2010 experiments, Experiments 1 and 3 here), averaged: associated/no shift .09, associated/shift .16, dissociated/no shift .17, dissociated/shift .20. The model’s predicted ordering holds on average; in this paper’s Experiment 3 the dissociated/shift cell is the lowest of the dissociated pair, which the authors “tentatively attribute … to random variation” (p. 1644).
- A processing-load account (visual–spatial working memory clogged by successive rooms) is set aside by citing their 2010 study, where verbal word pairs unrelated to the objects were also disrupted (p. 1641).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | The location updating effect survives a smaller display | moderate | Experiment 1, errors and A′; no RT effect |
| C2 | It occurs in real rooms | weak to moderate | Experiment 2: a modified task (six boxed objects, two minutes of arithmetic), p = .02 to .06 |
| C3 | Returning to the encoding room does not restore memory | moderate as a null | Experiment 3: return no better than shift, numerically worse |
| C4 | Memory worsens with the number of new rooms involved, not the number of doorways crossed | weak to moderate | Experiment 3: double shift worse than return (p = .02) |
| C5 | The effect is retrieval interference between event models that share the probed object | interpretive | event horizon model; Table 4’s pooled ordering |
Concepts
- Location updating effect: poorer memory for recently handled objects after moving from one location to another than after moving as far within one.
- Event horizon model: Radvansky’s account of event cognition; here segmentation, foregrounding, and competitive retrieval among event models.
- Competitive retrieval: a recognition probe must select one trace; an object present in two event models (where picked up, where carried) yields interference “even though in this case they would all support the same response” (p. 1634).
- Return / double-shift conditions: two doorways each, ending in the original room or in a third room.
Connections
It builds on LIT-1392 and on Radvansky, Tamplin & Krawietz (2010), which this record does not hold, and draws on the fan-effect work of Radvansky & Zacks (1991) and Anderson (1974). It cites event segmentation theory (Kurby & Zacks 2008; Swallow et al. 2009; Zacks et al. 2009), not Zacks et al. 2007 (LIT-1382). The interference account is tested further by Pettijohn & Radvansky (LIT-1385). McFadyen et al. (LIT-1387) describe Experiment 3 as showing that returning to the original location “impairs memory further”; the paper reports return as no different from shift (p = .20), and better than double shift.
Bearing on the record
- CASE-136. The case’s bullet is right: the effect held with smaller displays and in a real environment, and returning did not improve memory. Two things it leaves out matter for what the case is used for. First, the authors’ explanation is retrieval interference: the carried object sits in two event models. Second, in Experiment 3 the plain shift effect on errors was only marginal.
- A37’s inference from the return result (“our theory shouldn’t reduce perspective to location … an organized, goal-sensitive relationship to the environment”). The first half is supported: the encoding location does not reinstate availability. The second half is not tested. No goal, task or intention was varied. On the authors’ account, the relevant unit is still room-bound: each room entered makes another event model, so returning to a room adds a model rather than reinstating one. The paper shows that availability tracks the number of rooms entered, which is more location-bound than A37’s reading, not less. A37’s proposed discriminating variants (a doorway crossed while a salient task continues; a change of task with no doorway) are the experiments that would test the goal-sensitive reading. Neither is here.
- CLAIM-421 (activation interruption). The authors’ mechanism is not the one CLAIM-421 attributes to the case. CLAIM-421 says the earlier “contextual supports are no longer active”. The event horizon account says the probed object is in too many event models, and retrieval has to choose among them. That is a crowding of retrieval routes, not a loss of activation. It still fits the claim’s general distinction, since the item is held and harder to deploy (TERM-055). But the doorway effect, as its authors explain it, is a weak illustration of activation interruption in particular.
- CLAIM-287, TERM-071 (reinstatable persistence). Relevant by contrast: returning to the original room is the obvious reinstatement cue, and it did nothing. Reinstating a context did not reinstate access. That is a finding about object recognition in a probe task, and it should not be generalized to the reinstatement of an organization.
- Proposed THEORY-888 takes this paper as a primary source.
- No instruction for machine-learning practice.
Limitations
- Experiment 2 changes the task (six objects boxed per trial, no associated/dissociated contrast, a two-minute arithmetic distractor) and has six trials per person. Its effect is p = .02 and .06.
- In Experiment 3 the four conditions share one set of 64 probes, so each cell holds few trials. The A′ simple effects report F(1, 59) for an N of 48, an inconsistency in the paper.
- Room changes again co-occur with wall-pattern changes in the virtual experiments.
- Table 4 pools four of five experiments from this lab. The deviation in Experiment 3’s dissociated/shift cell is set aside as noise.
- The event horizon model is the authors’ own and is fitted here after the fact. No condition was designed in which interference and deactivation predict opposite results, except return against double shift, which either account can absorb.
Open questions
- Would a return condition with fewer intervening models (for example, a brief step out and back) restore memory, as deactivation predicts and interference does not?
- Does the double-shift cost depend on the object being carried through both rooms, as interference requires?
- Do the published raw data (posted on the lab page) support the per-cell reliability the pooled table suggests?