Cognition Implicitify Research Team

Change Blindness and the Slot Model: Why a 4-Item Scene Can Tell You About a Brain

One of the more counterintuitive findings in late-twentieth-century cognitive psychology is that you can show a person two photographs that differ in a large, central, theoretically-relevant respect, alternate them with a brief blank in between, and have the person fail to notice the difference for tens of seconds. The phenomenon is called change blindness, and the experimental paradigms that produced it forced the field to revise downward its assumptions about how richly the visual world is represented in conscious awareness. The Scene Change Detection Task (SCDT) is a clinical instrument built directly on that revision.

What change blindness actually shows

Ronald Rensink's To see or not to see: The need for attention to perceive changes in scenes (with O'Regan and Clark, 1997, Psychological Science, 8, 368–373) used a paradigm called flicker: an original scene and a modified scene alternate, separated by a brief blank field. Without the blank, the change is detected immediately — motion transients in the visual periphery do the work. With the blank, the same change can take many seconds to find, and is sometimes missed entirely. The blank disrupts the motion signal that normally captures attention, and what is left is a slow serial process: the observer has to direct attention to the changed object before the change can be consciously registered.

Rensink's Change detection (2002, Annual Review of Psychology, 53, 245–277) generalized the finding across a family of paradigms — mudsplash (small visual disruptions that mimic the eye movement transients), gradual change (slow morphs that produce no transients), and saccade-contingent change (changes timed to the moment of an eye movement). Across all of them, the same picture emerged: changes that occur outside the focus of attention are not consciously detected, even when they are large, even when they are central to the scene, and even when the observer is actively looking for them.

The theoretical implication Rensink and others drew was that conscious visual experience is far less complete than it phenomenologically feels. We do not represent the scene in detail; we represent the focus of attention in detail and rely on the world itself as an external store. When the world is briefly hidden, the contents of detailed representation are limited to whatever was being attended.

Luck and Vogel: the slot model

The complementary finding came from the change-detection paradigm with simple arrays developed by Steven Luck and Edward Vogel. The capacity of visual working memory for features and conjunctions (1997, Nature, 390, 279–281) used arrays of colored squares: an array was presented for 100 ms, followed by a delay, followed by a test array that was either identical or differed by one item. The number of items in the array was varied. Performance was at ceiling for arrays of 1, 2, and 3 items, dropped sharply at 4, and was at chance for 5 and above. The data fit a model in which visual working memory holds approximately four discrete items — the slot model of visual short-term memory.

Nelson Cowan's The magical number 4 in short-term memory (2001, Behavioral and Brain Sciences, 24, 87–114) was the broader theoretical synthesis. Cowan argued that across paradigms — change detection, partial-report, attention-cuing — the core capacity of focal attention/visual working memory converges on three to four items. The earlier Miller "magical number seven" was, on this account, a chunking-and-rehearsal-inflated estimate; the unchunked, unrehearsed core was closer to four.

The Luck-Vogel finding has not been without controversy. Wilken and Ma's A detection theory account of change detection (2004, Journal of Vision, 4, 1120–1135) and Bays and Husain's Dynamic shifts of limited working memory resources in human vision (2008, Science, 321, 851–854) have argued for a continuous-resource alternative in which each item in working memory is represented with finite precision and the precision degrades as more items are stored. The contemporary literature is somewhere between the slot and resource accounts, with substantial evidence that both apply to different aspects of the data. What both models agree on is the practical fact: performance breaks down at array sizes around four, which is the design choice the SCDT inherits.

Why a 4-item scene is the right test stimulus

The SCDT uses naturalistic scenes (a kitchen, an office, a park) rather than abstract arrays. Each scene contains four target objects whose retention is being probed. The trial structure is short — Study (5 s, fixed) → Mask (1 s) → Test (until response) → Response (≤ 30 s) — and the response is a binary judgment: change or no change.

The choice of four objects is not arbitrary. At three or fewer, performance is at ceiling for healthy adults and floor effects for impaired populations are obscured because intact performers are at the maximum and there is no room for normal variation to show. At five or more, performance is at floor for healthy adults and the test cannot discriminate among the impaired. Four is the sweet spot — it sits at the Cowan k ≈ 3–4 capacity limit, where intact adults perform reliably above chance but not at ceiling, and where capacity reductions produced by acquired brain injury, mild cognitive impairment, or attentional disorder produce measurable drops without driving the impaired patient to floor.

The 5-second study window is similarly principled. Long enough that intact participants can encode all four objects with normal scanning behavior, but short enough that compensatory rehearsal strategies do not dominate the score. The 1-second mask is the disruption that produces the change-blindness effect — without it, motion transients would carry the change-detection load and individual differences in working memory would be obscured.

What the score is sensitive to

A score on a 20-item SCDT is, mechanically, the count of correctly-classified trials. What that count is sensitive to, neurologically and clinically, is more interesting:

Visual working memory capacity (Cowan k). The most direct interpretation. Reductions in k due to mild cognitive impairment, early dementia, or normal aging show up as a roughly linear reduction in score. The drop is one of the earliest cognitive markers of MCI, often appearing before episodic memory measures show clinically significant impairment.
Attentional deployment. Even with intact capacity, a person who scans the scene inconsistently or who fails to allocate attention efficiently across the four target locations will miss changes. ADHD patients, particularly the inattentive subtype, often show an error pattern (scattered errors, slower RT, no consistent quadrant clustering) rather than a generalized capacity reduction.
Hemispatial neglect. Right-hemisphere lesions producing left-sided neglect appear as systematic miss patterns concentrated in the left visual hemifield. The SCDT quadrant analysis was designed to make this signal visible without requiring a structured neurological exam.
Posterior cortical involvement. Parkinson's disease and posterior cortical atrophy produce reductions in scene-maintenance capacity that are dissociable from frontal-attentional deficits. The pattern — moderate accuracy loss with substantially prolonged correct-trial reaction time — is suggestive when it appears alongside motor or visuospatial signs.
Diffuse axonal injury. Traumatic brain injury, particularly in the chronic phase, produces a characteristic profile of intact basic perception with degraded multi-object scene maintenance. The SCDT is sensitive to this profile.

The SCDT is not a substitute for a structured neurological evaluation. It is a screening-quality cognitive task whose pattern of errors can flag specific concerns and inform the choice of subsequent assessment. It earns its place in a battery because it is short, it is administered the same way to every patient, it produces an interpretable error pattern in addition to a global score, and it taps a domain — visual working memory under attentional load — that few short tests probe directly.

What the SCDT does not do

The instrument is silent on auditory working memory, on language-based working memory, on episodic memory in the long-term sense, on executive function in the planning sense, and on most of what an IQ test measures. It is not a dementia screen on its own; combining it with a brief verbal memory measure and a measure of orientation produces a much more useful triage. It is not validated in children below 12 or in adults above 85, and the discontinue rule is conservatively set so that profoundly impaired patients are not asked to complete the full set.

These limits are appropriate for what the test is. A short, focused, performance-based probe of one specific cognitive capacity, calibrated to the slot-model literature it descends from, is more useful in a multimethod battery than a longer instrument that tries to do everything badly.