4.1.2 - The working memory model

An active, multicomponent memory system

The working memory model (WMM) explains what happens when you are mentally holding information and using it at the same time. It was proposed by Baddeley and Hitch (1974) as a more detailed account of short-term memory than the older idea of one simple short-term store.

Think about calculating $27 \times 6$ in your head. One route is $20 \times 6 = 120$, then $7 \times 6 = 42$, then $120 + 42 = 162$. You may rehearse “120” while calculating 42, retrieve multiplication facts from long-term memory, and direct attention back to adding the partial answers. Picturing the written columns would add visual-spatial processing, but not everyone must use that strategy. The WMM says these are not all done by one passive box called "STM". They are handled by a small system of specialised components.

AQA names four components:

• The central executive directs attention and coordinates the system.
• The phonological loop deals with speech-based and sound-based information.
• The visuo-spatial sketchpad deals with visual and spatial information.
• The episodic buffer binds information together and links working memory with long-term memory.

The 2025 AQA wording asks for the features of the model in terms of coding and capacity. Duration belongs much more directly to the multi-store model topic, so do not make "duration of working memory" the centre of this lesson.

[DIAGRAM: asset_name: Working memory model component map; asset_slug: 4_1_2_02_the_working_memory_model__diagram_01; description: The central executive coordinates the phonological loop, visuo-spatial sketchpad and episodic buffer; the buffer links working memory with long-term memory.]

This model is useful because it turns a vague phrase like "short-term memory" into testable claims. For example, if two tasks both need the phonological loop, they should interfere with one another more than two tasks using different components.

The central executive

The central executive coordinates attention when information must be held and used together.

The central executive is not best thought of as a storage box. It is more like the manager of a very small workspace. It decides what gets attention, switches attention between tasks, and coordinates the phonological loop, visuo-spatial sketchpad and episodic buffer.

If you are doing mental arithmetic, the central executive helps you keep the goal in mind, decide which operation to do next, and stop irrelevant information taking over. If you are listening to a teacher while checking a diagram, the central executive helps you divide attention between verbal and visual material.

For coding, the central executive is not tied to one sensory code. It can work with verbal, visual and spatial information because its main job is control rather than storage. That means an exam answer should not say the central executive "stores sounds" or "stores images"; those claims belong to the phonological loop and visuo-spatial sketchpad.

For capacity, the central executive is limited. When two tasks both require heavy attention or decision-making, performance usually suffers. The model therefore predicts that a person can combine some simple tasks, but not unlimited demanding tasks.

For example, Theo can walk to class while rehearsing a definition because walking a familiar route takes little central-executive control. He finds it much harder to rehearse that same definition while solving a new algebra problem, because both tasks demand attention and monitoring.

The phonological loop

The phonological loop keeps speech-based information active through temporary storage and rehearsal.

The phonological loop is the "sound and speech" part of working memory. It is useful when you repeat a phone number, hold a phrase in mind, learn new vocabulary, or remember the exact wording of a short instruction.

It is usually divided into two subcomponents:

• The phonological store holds speech sounds. This is sometimes called the "inner ear".
• The articulatory control process rehearses information silently or aloud. This is sometimes called the "inner voice".

For coding, the phonological loop mainly uses acoustic or phonological coding. This is why a list of similar-sounding letters, such as B, C, D, G, P, T, can be harder to recall than a less similar list. Even if the letters are printed on a screen, many people silently name them, converting the visual input into a sound-based code.

For capacity, the safest exam answer is that the phonological loop is limited and is strongly affected by rehearsal. A classic finding is the word-length effect: people tend to recall more short words than long words in immediate serial recall. Baddeley, Thomson and Buchanan (1975) argued that this fits a time-limited rehearsal system, because shorter words can be rehearsed more quickly. The classic account describes a capacity of roughly what can be articulated in about two seconds, rather than a fixed number of words; this is an estimate that depends on the task.

Another useful finding is articulatory suppression. If someone has to repeat an irrelevant sound such as "the, the, the" while trying to remember a word list, their rehearsal process is blocked, so verbal recall is usually worse.

For example, if Sara is trying to remember "evaporation, condensation, precipitation" while repeatedly saying "la-la-la", the WMM predicts poorer recall because the articulatory control process is occupied.

The sketchpad and episodic buffer

Visual-spatial storage and integration across sources are different functions in working memory.

The visuo-spatial sketchpad is the "mind's eye" part of the model. It helps you imagine a shape, follow a map, judge where objects are, or picture how a graph should look. Its coding is visual and spatial. "Visual" means appearance, such as colour, shape and pattern. "Spatial" means location, movement and arrangement.

For capacity, the visuo-spatial sketchpad is limited. You may be able to hold a simple route or diagram in mind, but a crowded map, a rotating 3D shape and a moving pointer can overload it quickly. This is why a spatial secondary task, such as tapping a pattern, can disrupt a visual-spatial memory task more than a simple verbal task does.

The fourth component, the episodic buffer, was added by Baddeley in 2000 because the original model did not fully explain how verbal, visual and long-term memory information are bound together into one conscious episode.

For coding, the episodic buffer is multimodal. It can combine different forms of information, such as a sentence you hear, a face you see and a memory of where you met the person before.

For capacity, the episodic buffer is limited. Do not treat it as a huge store or as long-term memory itself. A safe AQA answer is that it has a limited capacity for integrated episodes or chunks, and that it acts as a temporary link between working memory and long-term memory.

For example, Imran is revising a labelled brain diagram. Silently naming the written label "hippocampus" uses the phonological loop, the shape and position use the visuo-spatial sketchpad, and his stored knowledge about memory uses long-term memory. The episodic buffer helps bind these into one meaningful study episode.

Evidence for separate components

Patterns of interference allow researchers to test whether different tasks depend on separate components.

A key prediction of the WMM is selective interference. Two tasks using the same component should interfere with one another more than two tasks using different components. This is why dual-task evidence is so important for AO3.

Baddeley and Hitch (1974) used concurrent memory-load tasks while participants carried out reasoning, comprehension and free-recall tasks. Their evidence suggested a trade-off between storage and processing: larger memory loads, especially around memory span, could slow or disrupt reasoning. This supported the idea of a limited-capacity working space rather than a passive short-term store.

The phonological loop is supported by word-length and articulatory-suppression evidence. If verbal recall depends on quick rehearsal, then short words should be easier than long words and irrelevant speech should disrupt rehearsal. Baddeley, Thomson and Buchanan (1975) found a clear short-word advantage in memory span tasks, which fits this prediction.

The visuo-spatial sketchpad is supported by studies where spatial interference disrupts visual-spatial tasks more than verbal interference does. Robbins et al. (1996) tested working memory in chess. Participants remembered or solved chess positions while completing secondary tasks intended to block different WMM components. Blocking the visuo-spatial sketchpad and central executive disrupted chess performance more than blocking the articulatory loop, which makes sense because chess positions are strongly visual-spatial and strategic.

Neuropsychological evidence also matters. Patient KF, reported in Warrington and Shallice's work, had severely impaired auditory-verbal short-term memory but did not show a general collapse of all learning, comprehension or long-term memory. This kind of dissociation challenges a single-store view and supports the idea that verbal short-term memory can be selectively impaired.

However, the evidence needs qualification. Dual-task findings support the WMM, but they do not prove that the exact four-part diagram is the only possible model. Case studies such as KF are rich but based on unusual brain damage, so they may not represent normal memory in all people. The strongest explanation treats evidence as support for the model's usefulness, not as final proof that every component is perfectly understood.

A fair dual-task comparison must consider task difficulty. If the spatial secondary task is much harder than the verbal one, greater disruption might reflect general attentional demand rather than shared storage. Researchers can measure performance on each task alone, match difficulty as far as possible, and compare the additional cost when tasks are combined. A larger drop for two spatial tasks than for matched verbal-plus-spatial tasks supports selective interference. A uniform drop across every pairing is weaker evidence for separate codes.

For example, in an original demonstration, route recall is 90% alone, 50% while tapping a spatial pattern and 80% while repeating a simple word. The spatial task costs 40 percentage points, compared with 10 for the verbal task. This pattern fits competition within the sketchpad. It does not mean separate components never interfere: both tasks can still demand central-executive attention. If tapping was substantially harder than repeating the word, general difficulty would remain a rival explanation.

Explanatory strengths and limits

The working memory model is useful because it gives a more active and detailed account of short-term memory than a single short-term store. It explains why you can sometimes do two things at once, why verbal and visual tasks can interfere differently, and why some patients can have a specific short-term memory impairment without all memory being damaged.

One strength is that the WMM has research support. Dual-task studies, word-length research and neuropsychological cases all fit the claim that working memory contains specialised components. This makes the model more precise than simply saying "STM has limited capacity".

A second strength is application. The model helps explain everyday multitasking and learning. A teacher might reduce working-memory overload by not asking students to copy a complex diagram while also listening to dense verbal instructions. That is not just a memory fact; it is a practical implication of different coding systems and limited capacity.

There are also limitations. The central executive is less clearly specified than the phonological loop. It is often described as the component that allocates attention, switches tasks and coordinates other systems, but these are broad functions. This makes it harder to test exactly than the phonological loop, where predictions about word length and articulatory suppression are more precise.

The episodic buffer can also be evaluated. It made the model better at explaining how information is integrated with long-term memory, but because it was added after the original model, it can look like a patch for things the earlier model could not explain. A balanced answer should say this is a weakness only if the addition is vague or hard to test; scientific models are allowed to develop when new evidence appears.

Finally, the WMM is a cognitive and partly reductionist model. It breaks memory into components and uses controlled laboratory tasks to test them. That is useful for precision, but it may simplify messy real-life memory, where emotion, motivation, expertise and long-term knowledge also shape performance.

Common mistakes to avoid:

• Do not call the phonological loop "the part that stores all sounds"; it is mainly speech-based/verbal information.
• Do not say the episodic buffer is long-term memory. It links with long-term memory.
• Do not make duration a main WMM feature for this AQA bullet. The named features are coding and capacity.
• Do not evaluate with empty phrases such as "low validity" unless you explain exactly why.

Quick recap

The central executive controls limited attention; the phonological loop holds and rehearses speech-based information; the sketchpad processes visual-spatial material; and the episodic buffer integrates sources and links with LTM. Selective interference and patient evidence support specialised components, but task difficulty and the precision of the proposed mechanisms matter when judging that evidence.