Textbook of Usability

Introduction

Perception delivers information to the mind, but it is memory that holds information in place long enough for the user to act on it. A nurse reading a patient's vital signs must hold the numbers in mind while navigating to the correct entry field. A pilot reading a checklist must remember which items have been completed. A visitor in a building must remember the directions given at the reception desk while walking through corridors. The capacity of human memory is finite and fragile. Understanding its structure — the distinct systems of sensory memory, working memory, and long-term memory — reveals why some designs support human performance while others overwhelm it.

Sensory Memory

When a stimulus reaches the senses, a brief, high-fidelity trace is retained in sensory memory. For vision, this is called iconic memory; for hearing, echoic memory. Iconic memory lasts approximately 250–500 milliseconds and can hold a large amount of visual information, but it fades rapidly and is overwritten by new visual input. Sperling's classic experiments Sperling, 1960 demonstrated that although subjects could report only 4–5 items from a briefly flashed array of 12 letters, they had momentary access to the entire array — iconic memory held it all, but the trace decayed before it could be fully reported.

Working Memory

Working memory is the system that holds and manipulates information currently in use Baddeley, 1974. It is the cognitive workspace where reasoning, comprehension, and decision-making take place. Its defining characteristic is severely limited capacity.

Miller's Magical Number

In 1956, George Miller published "The Magical Number Seven, Plus or Minus Two," arguing that working memory can hold approximately 7 (± 2) items simultaneously Miller, 1956. Subsequent research, particularly by Cowan Cowan, 2001, has revised this estimate downward to approximately 4 ± 1 chunks for novel information when chunking is controlled for. Cowan's figure, not Miller's original, is the value taken as the working capacity of the modern designer's budget.

Duration and Decay

Without rehearsal (the mental repetition of information), working memory contents decay within approximately 15–30 seconds. Peterson and Peterson Peterson, 1959 demonstrated this by having subjects remember three consonants while counting backward (to prevent rehearsal); recall dropped to near zero within 18 seconds.

Interference

Working memory is vulnerable to interference from similar information. Trying to hold two phone numbers in mind simultaneously is much harder than holding a phone number and a colour name, because the two phone numbers compete for the same representational resources. In interface design, requiring users to compare two similar data sets from memory (rather than displaying them side by side) invites interference errors.

Long-Term Memory

Long-term memory has no known capacity limit and can retain information for a lifetime. However, it is not a passive archive; storing and retrieving information are active processes that can succeed or fail.

Encoding

Information transfers from working memory to long-term memory through encoding. Deeper processing — relating new information to existing knowledge, generating mental images, or elaborating on meaning — produces stronger memory traces. This is Craik and Lockhart's levels-of-processing framework Craik, 1972. For interface design, this means that features learned through meaningful interaction (exploring, making errors, solving problems) are better remembered than features learned by rote memorisation of menus or keyboard shortcuts. Interfaces that provide meaningful feedback during learning support deeper encoding.

Retrieval: Recognition vs. Recall

Retrieval from long-term memory takes two forms. Recognition is the ability to identify previously encountered information when it is presented again — seeing a menu item and knowing it is the right one. Recall is the ability to produce information from memory without an external cue — remembering a keyboard shortcut without seeing it listed anywhere. Recognition is dramatically easier than recall. In classic memory experiments, recognition accuracy for previously seen items often exceeds 90%, while free recall of the same items may be only 30–40%.

Mental Models

Users build mental models — internal representations of how a system works — that guide their expectations and actions [Norman, 1983; Johnson-Laird, 1983]. These models are often incomplete, sometimes inaccurate, but always influential. A user whose mental model of a filing system is based on physical folders will expect documents to exist in one location at a time; the concept of a shortcut or alias may violate their model. Good design aligns with common mental models or makes the system's actual behaviour transparent enough for accurate models to form. When the system's behaviour contradicts the user's mental model, the result is confusion, errors, and frustration.

Cognitive Load Theory

Cognitive load theory, developed by John Sweller and colleagues beginning in the 1980s Sweller, 1988, provides a framework for understanding how the demands on working memory affect learning and performance Sweller, 1998. The theory distinguishes three types of cognitive load.

Intrinsic Load

Intrinsic load arises from the inherent complexity of the task or material. It depends on the number of elements that must be processed simultaneously and the interactions between them. Intrinsic load cannot be reduced without simplifying the task itself.

Extraneous Load

Extraneous load is imposed by the design of the interface or instructional material rather than by the task itself. Poorly organised layouts, unclear navigation, inconsistent terminology, and unnecessary visual clutter all increase extraneous load. This is the load that designers can and should minimise.

Germane Load

Germane load is the cognitive effort devoted to building mental models and integrating new information with existing knowledge. Unlike extraneous load, germane load is productive — it contributes to learning and understanding.

The Split-Attention Effect

When related information is presented in separate locations — a diagram on one page and its explanation on another, or a form field in one part of the screen and its validation message in another — the user must mentally integrate the two sources, consuming working memory. This split-attention effect increases extraneous load Sweller, 1998.

The Redundancy Effect

Conversely, presenting the same information in multiple formats simultaneously can increase load rather than reduce it. If a diagram is self-explanatory, adding a text description that repeats the same information forces the user to process both and verify their consistency. The redundancy effect reminds designers that "more information" is not always better.

Design Patterns That Reduce Cognitive Load

Progressive Disclosure

Rather than presenting all information and options at once, progressive disclosure reveals them in stages as the user needs them. A setup wizard that shows one step at a time imposes less working memory load than a single form with thirty fields. The trade-off is increased interaction cost (more clicks or steps), but for complex tasks the reduction in cognitive load typically outweighs the navigation overhead.

Chunking and Grouping

Organising information into meaningful chunks — using visual grouping, headings, whitespace, and consistent structure — reduces the number of independent items the user must process. A list of 20 items is cognitively expensive; the same 20 items grouped into 4 categories of 5 becomes manageable.

External Memory Aids

Any design element that externalises information the user would otherwise have to remember reduces working memory load. Breadcrumb trails show where the user is in a hierarchy. Progress indicators show how many steps remain. Shopping carts display selected items. These external memory aids are among the most reliable strategies for supporting human performance.

Consistency

Consistent interfaces reduce cognitive load by allowing users to apply learned patterns rather than processing each interaction from scratch. When buttons, labels, layouts, and interaction patterns follow predictable conventions, users can allocate working memory to the task rather than to the interface.

Defaults and Suggestions

Pre-populated form fields, smart defaults, and autocomplete suggestions reduce the recall demands on the user. Instead of requiring the user to remember and type a value, the system suggests or provides it, converting a recall task into a recognition task.

Memory and Expertise

The relationship between memory and design changes as users gain expertise. Novices rely heavily on recognition, need explicit guidance, and benefit from progressive disclosure. Experts develop chunked representations of common patterns, can recall commands and shortcuts from long-term memory, and may find progressive disclosure frustratingly slow.