Working Memory¶

Learner Variable · All learner variables
Evidence · 93 claims (55 show a difference, 27 mixed, 11 no difference) · 76 studies (20 causal, 19 review, 18 associational, 11 quant-synthesis, 5 theoretical, 3 qualitative), q1–q4 · 13 of 76 report an effect size · 87 claims rest on one study

Description¶

How much a learner can hold and manipulate at once. It is the best-evidenced dimension here and the most directly actionable: it converts into element counts per screen, whether two sources must be read together, and whether a problem is given whole or as a worked example. Capacity is near-fixed, but effective capacity is not — prior knowledge supplies the schemas that let several items be held as one, which is why the same screen overloads a novice and not an expert [~S].

How strongly it relates to achievement¶

Mathematics. These are correlations across meta-analyses (Breit et al. 2025, a systematic review of 30 meta-analyses; overview [+W]): how strongly learners who differ on this characteristic differ in mathematics achievement, not how much changing it would change achievement. Many of these characteristics are correlated with one another, and the review is about mathematics only.

Composite working memory r = .38, working memory .35 (k = 829); numerical .34, visual-spatial .31, verbal .30. Claim [+W]

Implications¶

Context¶

Sets how many elements a screen may hold at once, and settles split-attention and worked-example decisions.
Load is a property of the material and the learner together, never of the material alone — which is why a load rating belongs to a placement rather than to a component.
The remedies reverse with expertise: support that helps a novice becomes redundant and then harmful [~S].

Target Learners¶

Novices in the domain, who lack the schemas to chunk and are most exposed to overload.
Learners with attention or language demands already consuming the same resource.

Target Learning Objectives¶

Affects acquisition of anything with interacting parts — procedures, multi-step problems, integrated diagrams.
Barely affects tasks whose elements can be learned one at a time, which is why load-reduction is not a universal good.

Claims¶

Chunking reduces working memory load [+S] — grouping into fewer, larger units is the central mechanism for effective capacity
Split attention effect degrades learning [-M] — forcing integration of two separated sources spends capacity on the integration
Redundancy effect impairs learning [-M] — duplicated information must be processed and reconciled before it can be discarded
Presenting words as narration rather than on-screen text alongside graphics [+M] — splits input across channels instead of contending for the visual one; instruction changes it
Segmentation benefits shrink with expertise [~M] — the boundary condition — the same segmentation stops paying as schemas form
Adaptive training produced substantial and sustained gains in working memory for children with poor working memory [+M] — instruction changes it
Childhood poverty interferes with adult working memory, with chronic childhood stress a major contributor [+M] — instruction changes it
CLT-aligned e-learning modules improved cognitive load profiles and OSCE performance in simulation-based medical training (review-attributed to Gutierrez et al., 2023) [+W] — instruction changes it
Cognitive Overload Degrades Learning [+M] — learners who differ on it differ in outcomes
Deliberate practice can expand short-term memory capacity beyond the conventional seven-item limit [+W] — instruction changes it
Multisensory literacy strategies help students with low working memory become better at decoding [+W] — an instructional effect differs with it
Part-task practice reduces cognitive load for absolute novices during initial skill acquisition. [+W] — instruction changes it
Revised visuals reduce cognitive load while reading lesson text for low prior knowledge students but increase it for middle prior knowledge students [~M] — instruction changes it
Stereotype threat in collaborative contexts can interfere with working memory and reduce meaningful contribution [+W] — instruction changes it
Studies disagree on whether low or high working memory learners benefit more from retrieval practice, and several find no relationship [~M] — an instructional effect differs with it
Type 1 one-step equations (1 operational, 2 relational lines) are argued to be easier to learn than Type 2 (2 operational, 3 relational lines) due to lower cognitive load [+W] — instruction changes it
Under the components model in ECLS-K, working memory is the most closely associated EF component with math, but only at time point 1 does it show a positive residual correlation beyond latent EF [~W] — learners who differ on it differ in outcomes
Working memory is a strong predictor of decoding skills [+W] — learners who differ on it differ in outcomes
Teachers trying a mock-up of a scrutable concept-map interface expected it to reduce their cognitive load and said it helped them generate content quickly when creating classroom materials with generative AI [+W] — instruction changes it
Identical representational designs produced wide individual variation in what felt helpful versus overwhelming (representational fit) [~W] — an instructional effect differs with it
Role design improves learners' perceived usefulness and ease of use of GAI, but moderately increases collaborative cognitive load [+M] — instruction changes it
Self-reported cognitive load showed no significant differences across either volume-based or content-based offloading profiles [-W] — instruction changes it
Self-reported mental effort increased less across tasks in the ChatGPT condition while perceived difficulty rose equally in both groups [~M] — instruction changes it
Adding a second writing agent increased reported cognitive load, and writing confidence did not increase monotonically with support, with the Single condition reporting the lowest confidence [+M] — instruction changes it
Students rate LLM-generated consolidated reports favourably, with cognitive load rated lowest [+M] — instruction changes it
Training on AI interaction strategies slightly increases cognitive load during AI-assisted design [+W] — instruction changes it
Higher extraneous cognitive load predicted lower post-test scores after controlling for prior knowledge [+M] — learners who differ on it differ in outcomes
The authors conclude real-time AI feedback reduces extraneous cognitive load and mitigates fear of failure with expensive laboratory equipment [+W] — instruction changes it
Segmentation reduces overall cognitive load during multimedia learning [+M] — instruction changes it
The segmenting, cognitive-load, and learning-time effects are confirmed for system-paced segmentation [+M] — instruction changes it
Evidence on working memory capacity as a moderator of the interleaving effect is mixed, but WM does not appear to be a major barrier to classroom use [~M] — an instructional effect differs with it
CCT yields small significant gains in nonverbal memory, working memory, and visuospatial skills, but only weak evidence for verbal memory [+M] — instruction changes it
Working-memory-only training shows no positive evidence of efficacy in healthy older adults [-M] — instruction changes it
Germane cognitive load fully mediates the effect of information-search tool on justification quality (exploratory analysis) [+M] — instruction changes it
Offloading attenuates individual differences linked to unaided ability [~M] — an instructional effect differs with it
Reduction of cognitive load can sometimes impair learning rather than enhance it [~W] — learners who differ on it differ in outcomes
Students using ChatGPT 3.5 experience lower extraneous cognitive load than students using Google during a web research task [+M] — instruction changes it
Students using ChatGPT 3.5 experience lower intrinsic cognitive load than students using Google during a web research task [+M] — instruction changes it
After controlling for age, WM correlates significantly only with the most demanding condition (priming task, noncanonical word order) [~M] — learners who differ on it differ in outcomes
Age predicts decoding efficiency in children with dyslexia while working memory and perceptual organization covariate effects are non-significant [-W] — learners who differ on it differ in outcomes
Most predictor effect sizes showed high heterogeneity; only morphological awareness, working memory, and language comprehension fell in the moderate I² range [~W] — learners who differ on it differ in outcomes
Numerical working memory predicts decimal number line estimation accuracy [+W] — learners who differ on it differ in outcomes
The central executive coordinates WM subsystems and supports higher-level language-learning functions such as attention switching and strategy switching [+M] — learners who differ on it differ in outcomes
The episodic buffer integrates multimodal representations from WM components and long-term memory, which is important for learning [+M] — learners who differ on it differ in outcomes
Verbal and spatial working memory, and verbal attention span, explain unique variance in Coding beyond motor speed, symptom severity, and IQ [+M] — learners who differ on it differ in outcomes
Verbal STM/WM and several executive and sensorimotor measures show no robust relation with math in children with normal reading skill [-M] — learners who differ on it differ in outcomes
Verbal working memory directly and positively predicts math performance in grades 3 and 4 [+W] — learners who differ on it differ in outcomes
Verbal working memory predicts oral reading fluency in children with and without ADHD [+M] — learners who differ on it differ in outcomes
Working memory accounts for the relation between ADHD traits and communicative perspective-taking [+M] — learners who differ on it differ in outcomes
Working memory capacity is a robust indicator of success or failure in language learning [+M] — learners who differ on it differ in outcomes
Working memory is the strongest predictor of adolescents' communicative perspective-taking [+M] — learners who differ on it differ in outcomes
Working-memory capacity correlates positively with accuracy on all four sentence-production conditions [+M] — learners who differ on it differ in outcomes
Children with MLD show pervasive working memory deficits while both LA groups score in the average range on all three working memory components [+M] — learners who differ on it differ in outcomes
CogMed computerized working-memory training improves practiced working-memory skills with narrow transfer, and controls playing the same games without increasing difficulty show no gains [~M] — instruction changes it
Phonological short-term memory (nonword repetition) shows a modest unique contribution to multiplication fact performance, while phonological awareness shows no relation [+S] — learners who differ on it differ in outcomes
Transfer from computer-based working-memory and reasoning training in children is narrow: gains stay within the trained domain [~M] — instruction changes it
Hippocampal malleability from resource-demand mismatch implies students benefit most from challenges matched to individual capacity [+M] — an instructional effect differs with it
Vocabulary fully mediates the effect of working memory on both local and global coherence inference making in 6- to 10-year-olds [+M] — learners who differ on it differ in outcomes
Children's understanding of whole number advances by one developmental level per additional unit of working memory capacity [+W] — learners who differ on it differ in outcomes
Rational-number CCS test performance is predicted by working memory capacity and, to a lesser extent, by intrusion errors in the listening span [+W] — learners who differ on it differ in outcomes
Adding equations to tree diagrams reduced cognitive load but did not improve learning outcomes [~M] — instruction changes it
Extraneous cognitive load arising from limited scope and high distance in the rule-based game pushed the facilitator toward first- and second-order interventions instead of the third-order interventions the learning goal required [+M] — instruction changes it
Higher working memory capacity was associated with better final test performance on advanced materials [+M] — learners who differ on it differ in outcomes
Learning with tree diagrams increased experienced cognitive load across intrinsic, extraneous, germane, and overall measures [+M] — instruction changes it
Participants in the rule-based simulation game perceived it as demanding, and learning and applying its procedures and rules hindered investment in their learning goals [+M] — instruction changes it
Retrieval practice conditions produced significantly higher cognitive load than rereading [+M] — instruction changes it
No difference in extraneous cognitive load between modality conditions in VR, contrary to the modality principle [-M] — instruction changes it
No difference in intrinsic cognitive load between visual-only and audio-visual text conditions in VR [-M] — instruction changes it
Visual-only text produces higher germane cognitive load than audio-visual text in VR [+M] — instruction changes it
Working memory capacity moderated disfluency effects in one study but not another [~M] — an instructional effect differs with it
Some SRSD students, particularly those with poor working memory, struggled to implement the planning strategies they were taught [+M] — learners who differ on it differ in outcomes
Betweenness centrality decreases as working memory capacity increases [+M] — learners who differ on it differ in outcomes
Harmonic centrality and local clustering coefficient show no relation with working memory capacity [-M] — learners who differ on it differ in outcomes
Moderate-to-vigorous intensity active breaks are associated with slightly greater working-memory improvements than moderate intensity, in exploratory meta-regression [~M] — instruction changes it
Readers with higher working memory capacity make more transitions between sentences (higher strength) [+M] — learners who differ on it differ in outcomes
In a visual-spatial integration task, the interaction between visual-spatial working memory capacity and visual design is quadratic, with the inverted-U progression more strongly accentuated for the dynamic design [~W] — an instructional effect differs with it
Failure to obtain a testing effect with immediate tests may be partly due to working memory depletion [~M] — instruction changes it
Working memory training shows short-term near-transfer gains but no compelling evidence of consistent long-term academic achievement gains [~M] — instruction changes it
Extensive Chinese logographic literacy training is associated with higher visuospatial short-term storage that does not generalize to phonological storage [~M] — instruction changes it
Only low-working-memory-capacity students benefited from making pointing gestures when explaining number equations (review attributes this to Marstaller & Burianová, 2013) [~M] — an instructional effect differs with it
High task variability can overload working memory and reduce learning efficiency in some cases [~W] — instruction changes it
In elementary school students, short-term green breaks restore attention and working memory, with effects on inhibition mixed [+M] — instruction changes it
In university students, walks or sitting in natural settings after cognitive fatigue improve some attention and working memory measures, with null results for others [~M] — instruction changes it
Increasing germane cognitive load is not a significant moderator of drawing support effects on comprehension [-M] — an instructional effect differs with it
Reducing extraneous cognitive load is not a significant moderator of drawing support effects on comprehension [-M] — an instructional effect differs with it
Teachers' load-reducing strategies correlate with lower cognitive load and higher motivation, engagement, and achievement, and their motivating styles with lower load and higher motivation and engagement (Evans et al., 2024) [+M] — instruction changes it
Cognitive barriers—insufficient background knowledge or working memory capacity—impede productive engagement in generative learning activities [+W] — learners who differ on it differ in outcomes
Physical activity interventions in childhood improve working memory with a small summary effect, based on very low-certainty evidence [+W] — instruction changes it
The negative working memory span–listening connection may reflect common-effect structure and overload of listening input rather than a direct detrimental role [~W] — learners who differ on it differ in outcomes
Working memory span, updating, and cognitive flexibility positively predict English writing, response inhibition positively predicts reading, and working memory span negatively predicts listening [~M] — learners who differ on it differ in outcomes
Example elaboration training is studied for decreasing cognitive interference and overload (per Stark, Mandl, Gruber, & Renkl) [+W] — instruction changes it
Instructional techniques for studying worked examples can increase germane cognitive load (per van Merriënboer, Schuurman, De Croock, & Paas) [+W] — instruction changes it
Worked examples are studied for their effects on cognitive load in older learners (per Van Gerven, Paas, & Schmidt) [+W] — instruction changes it
Working-memory limitations can be circumvented by schema construction, automation, and multiple presentation modalities [~W] — instruction changes it
Working memory compensatorily moderates the explaining intervention's advantage: supported self-explaining benefits students with weaker working memory most [~M] — an instructional effect differs with it
Instruction based on cognitive load theory can compensate for age-related cognitive declines [+M] — an instructional effect differs with it

The evidence by question¶

Do learners with lower working memory capacity gain more, or less, from retrieval practice, interleaving and other desirable difficulties? Studies disagree on whether low or high working memory learners benefit more from retrieval practice, and several find no relationship [~M]
Can training or practice raise working memory itself, and by how much? CCT yields small significant gains in nonverbal memory, working memory, and visuospatial skills, but only weak evidence for verbal memory [+M]

Prior knowledge — supplies the schemas that raise effective capacity; the two are inseparable in practice.
Attention — a different bottleneck often mistaken for this one: attention is what reaches the store, capacity is what it holds.
Reading and language — decoding that is not automatic spends this budget before comprehension starts.

Examples¶

Cognitive Load Management — the element that names the design response
Worked Examples — substitutes studying a solution for searching for one
Just-in-Time Information — defers detail until the moment it is needed rather than front-loading it
Chunking to Reduce Working Memory Load — the direct application
Segmenting — breaks a continuous presentation into learner-paced parts
Faded Worked Examples — withdraws the support as expertise makes it redundant
4C/ID (Four-Component Instructional Design) — sequences whole tasks simple-to-complex against entry capacity
Cognitive Load Theory — the explanatory frame for all of the above
Expertise Reversal Effect — why every remedy here has an expiry date

Key Sources¶

The evidence is carried by the claim pages linked above, each holding its own ## Evidence entries with q/i codes and DOIs checked against Crossref. Citations are deliberately not duplicated here — a second copy is a second thing to keep correct.

Page metadata

Field	Value
`type`	`learner-variable`
`id`	`working-memory`
`title`	`Working Memory`
`description`	`How much a learner can hold and manipulate at once before the material itself becomes the obstacle.`
`status`	`review`
`canonical`	`True`
`generated`	by `claude/unspecified` on `2026-09-02`
`trust`	unverified — no one has confirmed this page's content

Frontmatter is the page's machine-readable record — see the Schema & Guide.