Capacity varies with item complexity, so simple and complex items can't be compared on one number

What it is

The number of items working memory can hold at once is not a fixed value independent of content — it varies with how much information each item itself carries. Items defined by a single feature, like color, can be held in noticeably greater numbers than complex items combining several features at once, like shape, color, and orientation together. This means any specific capacity number is only meaningful once you specify how complex each item is; quoting a number without reference to item complexity carries no information.

Why it happens

Capacity drops with complexity because the processing effort needed to maintain an item is proportional to how much information that item carries: a single simple feature needs relatively little resource to stay in a retrievable state, while a complex, multi-feature item requires simultaneously maintaining more interrelated details, consuming more resource accordingly. Given a roughly fixed total processing budget, the more any single item consumes, the fewer items can be held at once — this differs from the intuitive picture of a fixed number of slots that doesn't care about content, and is closer to a mechanism where a shared resource is allocated across items as needed.

Studying it

The typical paradigm is a visual working memory change-detection task: capacity is first estimated using items defined by a single feature (solid color patches, for instance), then the same task is repeated with items varying on several features at once (composite shapes that vary in color, orientation, and form simultaneously), and the two capacity estimates are compared. If the estimate from complex material is noticeably lower, that confirms capacity really does scale with information load rather than being a material-independent constant. Researchers also systematically vary the complexity gradient — increasing the number of varying features step by step — to check whether the capacity number declines monotonically, confirming a continuous resource-allocation effect rather than a threshold effect.

Where it stops holding

Complexity's effect on capacity is discounted when items share structure the observer is already familiar with: if several features tend to combine in a regular, recognizable pattern that can be treated as one unit, the resource actually consumed is lower than treating each feature as independent information would predict. This is where the complexity effect and chunking interact rather than contradict each other — though working out that interaction in detail is not what this particular boundary is meant to cover.

Applying it

• When designing any interface that requires users to hold or compare multiple items at once, don't rely on item count alone as the load metric — also account for how much variable information each item carries. Five icons that differ only in color impose a different memory load than five icons that combine color, shape, and positional meaning at once, and the latter should be designed to a stricter count ceiling.
• For scenarios requiring users to track multiple complex object states at once (a multi-variable dashboard, several progress indicators needing simultaneous attention), prioritize reducing the number of independent features carried by each object rather than simply cutting the total number of objects. Merging features that can be expressed as one visual whole reduces load without reducing the underlying information.
• How to check: run a comparison test at the same total item count but different complexity levels (single-feature versus multi-feature items) and compare memory or verification error rates between the two conditions, to determine whether the interface's current item complexity already exceeds what users can actually carry at that count.

Related

• Same group: A6.02.1 The number of items working memory can hold at once is limited · A6.02.2 Retention is brief and vulnerable to interference · A6.02.3 Requiring users to remember content across pages is itself a design flaw · A6.02.4 The classic capacity estimate is closer to four chunks than the earlier figure of seven · A6.02.5 Without rehearsal, working memory content decays naturally within roughly ten to twenty seconds · A6.02.6 Rehearsal extends retention but draws on the same limited processing resource · A6.02.8 Capacity limits and time-based decay are two independent mechanisms, and easing one doesn't substitute for the other
• Nearby: A6.03 Chunking · A9.06 Intrinsic, extraneous, and germane load
• Search terms: visual working memory · chunk complexity · information load per item