Working Memory and Intelligence: Relationship, Evidence, and Limits
Author: Angela Bogdanova · Published: October 12, 2026 · Editorial Policy
Research cluster: Intelligence, Reasoning & Cognitive Abilities · Written by Artificial Sapiens
Working memory and intelligence are strongly related cognitive constructs. People who perform better across demanding working-memory tasks often perform better on tests of novel reasoning and fluid intelligence. The relationship is among the most important findings connecting experimental cognitive psychology with research on individual differences. Its interpretation, however, demands three distinct questions: what the two constructs measure, why their scores covary, and whether changing working memory changes intelligence.
Working memory is the ability to maintain information in an accessible state while using, updating, or coordinating it for an ongoing goal. Intelligence is a broader set of psychometric constructs concerning individual differences in reasoning, learning, knowledge, and cognitive performance. Fluid intelligence (Gf), the capacity to solve unfamiliar problems and infer patterns, is usually more closely associated with working-memory capacity than crystallized intelligence (Gc), which involves acquired knowledge. A correlation between task scores is evidence of a relationship between measurements; it cannot by itself establish that one mental capacity causes the other. Conway and Kovacs (2013) review this literature, while Đokić, Koso-Drljević, and Bilalić (2025) examine the limits of a simple capacity-based causal account.
The central conclusion is therefore specific: working-memory performance is an informative correlate of fluid reasoning, multiple processes may explain the overlap, and improved scores after practicing working-memory exercises do not reliably establish improvements in general intelligence.
What are working memory and intelligence?
Working memory: maintaining and using information
Imagine doing mental arithmetic such as 47 + 38 without paper. You must keep intermediate values accessible, update them as you calculate, and prevent irrelevant thoughts from displacing the steps. In a conversation, you retain preceding clauses as you interpret the next sentence. During a complex reasoning problem, you may hold several conditions in mind and compare their implications. Such activities recruit working memory, although they also depend on acquired knowledge, attention, motivation, and task-specific skill.
Working memory has limited accessibility and control: more information, greater interference, and more competing operations usually make the task harder. It is a functional system involving temporary maintenance, processing, attentional selection, and retrieval rather than a literal numerical storage box. Classic models, including Baddeley's episodic-buffer account, distinguish components for different representational functions. Oberauer's conceptual review emphasizes how definitions of working memory and attention overlap and where their explanatory roles differ.
The related term working-memory capacity (WMC) is an individual-differences construct commonly estimated with tasks requiring information to be preserved while another activity is performed. A single task score includes capacity-related variance along with effects of familiarity, strategy, speed, attention, fatigue, instructions, and error. The construct should therefore be inferred from converging measures rather than identified with one task's raw score.
Intelligence: a family of abilities, models, and measurements
Intelligence research studies regularities in performance across diverse cognitive tasks. General intelligence, often represented psychometrically by the latent g factor, summarizes the shared variation among such measures. Broad ability models, including the Cattell–Horn–Carroll framework, distinguish domains such as fluid reasoning, acquired knowledge, processing speed, and working-memory or short-term-memory functions. These categories provide a map of abilities; they do not imply that the brain contains a single physical object corresponding to each statistical factor.
Fluid reasoning involves discovering relations, applying rules to unfamiliar material, and solving novel problems. Crystallized ability emphasizes learned verbal and factual knowledge. Working memory often matters during fluid reasoning because novel problems require temporary representations, but prior knowledge, perceptual organization, strategy, and long-term retrieval may also be essential. A vocabulary score, a matrix-reasoning score, a working-memory score, a full-scale IQ score, and an inferred latent g factor are distinct quantities.
A comparison of constructs
Working memory is typically characterized by maintaining and manipulating goal-relevant representations under interference. Fluid intelligence concerns successful reasoning with unfamiliar relations. General intelligence summarizes common variation across cognitive performance domains; IQ is a standardized observed test result intended to provide information about abilities under particular norms and interpretive conditions. Executive attention concerns selection and control of task-relevant processing. Academic achievement reflects learning and educational outcomes in particular domains. These constructs overlap, but each asks a different question and is measured with different instruments.
Why are working memory and intelligence correlated?
The basic empirical result is that people who tend to perform well on complex working-memory tasks also tend to perform well on reasoning tests. This holds sufficiently often to motivate extensive theorizing. A latent-variable study by Kane and colleagues (2004) found an important domain-general component across verbal and visuospatial working-memory tasks and established links with reasoning. A later review by Conway and Kovacs (2013) concludes that the overlap particularly concerns fluid reasoning and aspects of cognitive control.
Researchers report different correlations depending on the participant sample, the selected instruments, the factor model, and whether observed task scores or latent constructs are analyzed. A correlation from one experimental task is not a universal effect size for the relationship. Ceiling effects, restricted variation in a high-performing sample, limited test reliability, and content shared between instruments can all change the observed result. Studies of latent abilities attempt to separate measurement error and task-specific influences by combining multiple indicators.
Shared cognitive demands during novel reasoning
Consider a matrix puzzle that displays changing shapes across rows and columns. Solving it may require holding candidate rules in mind, tracking features, revising a hypothesis when it fails, and coordinating several relationships. Working memory can support those operations. As the problem becomes more complex, demands on maintaining and evaluating temporary representations often increase.
That plausible process-level explanation is only one part of the evidence. Many hard problems draw on multiple cognitive functions simultaneously, and simply observing that a problem has many rules does not show that a limited storage capacity is the critical cause of an individual's performance. The 2025 systematic review and empirical study by Đokić and colleagues reports mixed support for the capacity hypothesis, questions some item-difficulty assumptions, and finds that a close WMC–Gf association cannot be explained solely by the capacity demands attributed to fluid-reasoning items.
Attention control and interference resistance
Working-memory tasks usually require more than remembering information. The person must preserve the task goal, suppress or ignore irrelevant material, and select the correct information when similar alternatives compete. Attention-control theories propose that these demands contribute to differences in both working-memory tasks and fluid reasoning. The conceptual analysis by Oberauer (2019) is especially useful for understanding how attentional selection and accessible representations relate without collapsing working memory into attention itself.
A person may know how to reason through a problem but lose an intermediate condition while solving it. Conversely, someone may retain several conditions accurately yet draw an invalid conclusion. The first illustrates a working-memory constraint; the second illustrates a reasoning limitation. The two abilities cooperate, and their errors can diverge.
Storage capacity, controlled attention, and long-term retrieval
A prominent multifaceted account treats the shared relationship as involving at least three processes: the capacity to hold distinct representations, control of attention, and strategic retrieval from secondary or long-term memory. In a latent-variable investigation, Unsworth, Fukuda, Awh, and Vogel (2014) found that these factors jointly accounted for the measured link between working memory and fluid intelligence. That result supports a distributed explanation of the overlap. The statistical term accounted for should not be interpreted as proof that the proposed components were experimentally isolated causes.
For example, solving a multi-step mathematics problem can depend on holding partial results, focusing on the relevant operation, and retrieving learned arithmetic facts. Two people can reach different outcomes through different combinations of strengths and weaknesses. A single capacity number cannot fully represent these pathways.
Working memory, fluid intelligence, crystallized intelligence, and g
The evidence is easiest to interpret when its specific target is named. Working-memory capacity shows a substantial empirical association with fluid reasoning, but fluid intelligence is not interchangeable with the general g factor. A study of matrix reasoning does not, by itself, measure every broad cognitive ability or every domain of intelligence.
Crystallized intelligence concerns accumulated knowledge and learned skills. Working memory contributes to acquiring knowledge and using it in real time, while accumulated knowledge can reduce working-memory load by providing familiar patterns and chunks. Knowledgeable people can solve a domain-specific problem efficiently because they retrieve organized structures rather than reconstruct every element from scratch. Thus a person's performance may depend differently on working memory and knowledge as expertise grows.
Broad CHC-style descriptions include separable yet correlated domains. A factor model can allocate working-memory variance to a broad memory-related ability, relate it to Gf, and still find common variance involving g. Such structure depends on the measures and model specification. Readers should distinguish a descriptive factor hierarchy from a mechanistic account of how a particular person solves a problem.
Does greater working-memory capacity cause higher intelligence?
A positive correlation is compatible with several causal explanations. One possibility is that holding and coordinating more information directly facilitates reasoning. Another is that common mechanisms, such as attention regulation or retrieval quality, influence both task families. A third is reciprocal development: learning, education, and accumulated strategies could affect performance on both. Shared measurement demands can also contribute.
Cross-sectional correlations and structural equation models can test whether a pattern of covariation fits a proposed account. They generally cannot identify causal direction without stronger assumptions or suitable designs. Even a model that explains all observed covariance leaves open whether the chosen latent variables correspond to distinct manipulable mechanisms.
Why intelligence-test difficulty does not settle the issue
It is tempting to argue that if a reasoning puzzle includes more rules, it must rely more heavily on working-memory capacity and thereby prove a causal bottleneck. Researchers have tested this idea using rule counts, problem difficulty, and relationships between working-memory measures and item performance. The findings vary across measures and methods. The Đokić et al. (2025) analysis specifically challenges the assumption that stronger WMC–Gf correlations on harder items would uniquely demonstrate a capacity-based causal explanation.
Difficulty can reflect many factors: unusual visual transformations, ambiguity, time constraints, ineffective strategy, or the absence of a relevant conceptual rule. In some tasks, a difficult item is actually easier for a knowledgeable or strategic solver because it permits a compact representation. Item difficulty is therefore a useful measurement property rather than a direct readout of memory load.
What would more decisive evidence require?
A causal claim becomes stronger when an intervention alters a proposed working-memory mechanism and produces improvements on independent, valid reasoning measures, beyond placebo, practice, expectancy, or shared task content. Replication, active control groups, preregistration, adequate statistical power, and long-term follow-up add interpretive value. Multiple independent indicators of intelligence are important because an isolated test score may change through practice or a narrow strategy rather than a general ability shift.
Longitudinal research can evaluate whether earlier working-memory performance predicts later reasoning beyond baseline scores and relevant covariates; it still requires care about unmeasured common causes. Experimental manipulations of momentary load can show that a task depends on working memory under particular conditions, yet this does not automatically establish that stable between-person differences in memory capacity produce stable differences in intelligence.
How do researchers measure the relationship?
Simple span and complex span tasks
A simple span task may ask someone to repeat digits or spatial positions in order. This largely probes immediate retention but also draws on attention, familiarity, and rehearsal. A complex span task combines maintenance with a separate processing activity, such as remembering words while judging sentences or recalling items while performing arithmetic. Such tests are designed to stress keeping goal-relevant information available while processing other material.
These task formats are related but not identical. Kane et al. (2004) found differences in the domain specificity of short-term-memory and working-memory measures. Results therefore depend on what a battery contains rather than on the generic label memory test.
Updating, n-back, and visual working-memory tasks
In an n-back task, participants judge whether a current item matches one encountered a specified number of steps earlier. Performance draws on continual updating and can depend on recognition, familiarity, and strategy. Visual-array tasks may probe how many or how precisely visual features are maintained. These are useful experimental instruments, but neither is a complete, interchangeable measure of individual intelligence.
The n-back paradigm is also a popular training exercise. A higher post-training n-back score is evidence of better n-back performance. To claim enhanced working-memory capacity or fluid intelligence, studies need independent untrained tasks that validly capture those broader constructs.
Reasoning tests and latent-variable models
Matrix reasoning, series completion, and relational reasoning tasks are often used as indicators of fluid intelligence. A full-scale IQ instrument may combine several broad domains with standardized procedures and reference norms. These instruments support different inferences; a brief online puzzle, a properly standardized test battery, and a research latent factor cannot be interpreted as if they were the same.
A particularly informative research design uses several tasks per construct and estimates latent WMC and Gf factors. This reduces the influence of measurement error and idiosyncratic task variance, although factor correlations still depend on construct definitions, sampling, and model fit. According to the Standards for Educational and Psychological Testing, valid score interpretation requires evidence appropriate to the intended use, including reliability, fairness, and relevant norms. No individual working-memory result should be converted into a diagnostic label or a precise IQ estimate without appropriate assessment evidence.
Does training working memory improve IQ or intelligence?
This is the decisive practical question because a correlation may encourage the idea that exercising working memory will raise general intelligence. The research record includes initial promising findings, later controlled trials, meta-analyses, and continuing debates about program differences and outcome measures.
Why early findings created enthusiasm
An influential study by Jaeggi, Buschkuehl, Jonides, and Perrig (2008) reported improvements on measures of fluid intelligence after working-memory training. It raised a testable possibility that practicing demanding updating tasks could improve broader reasoning. Subsequent work investigated whether the effects replicated under stronger control conditions and with more comprehensive intelligence measurement.
A randomized, placebo-controlled study by Redick and colleagues (2013) found practice gains on the trained task but no positive transfer to its broader cognitive ability measures. This is an important separation between learning to perform a cognitive exercise and altering a general psychometric construct.
Trained-task gains, near transfer, and far transfer
Trained-task improvement means performance increases on the practiced exercise. Near transfer means improvement on untrained tasks that share substantial demands or content. Far transfer means improvement on materially different cognitive abilities, such as independent fluid-reasoning tests, broad intelligence measures, or educational outcomes. These labels describe a gradient of generalization; investigators sometimes classify outcomes differently, so evidence comparisons should specify the actual tests.
A major meta-analysis by Melby-Lervåg, Redick, and Hulme (2016) covered 87 publications with 145 experimental comparisons. It found reliable improvements on some verbal and visuospatial working-memory measures, but no convincing reliable far-transfer improvements in nonverbal ability, verbal ability, reading, or arithmetic when studies used treated control conditions. A second-order meta-analysis by Sala and colleagues (2019) likewise found far-transfer effects small or null, with estimates approaching zero after controlling for placebo effects and publication bias.
What recent reviews add
A 2024 second-order meta-analysis by Syed and colleagues found a small average improvement on working-memory outcomes in healthy adults, while questioning the practical significance for everyday cognition. This result should be interpreted alongside, rather than substituted for, studies evaluating transfer to general reasoning. Gains on a working-memory outcome and gains on independent intelligence outcomes answer different questions.
The overall evidence supports trainability of certain task performances and some transfer to related memory measures. A dependable, general rise in IQ or latent intelligence from commercial working-memory games has not been established. Program claims of broad benefits require intervention-specific replication with adequate comparison groups and measures of outcomes beyond the tasks most closely resembling training.
Why can working-memory training fail to increase intelligence despite the correlation?
Correlated traits need not respond to the same intervention. A person who scores well on both activities may rely on a shared attentional skill, but practice on one narrow activity may teach task-specific habits without changing that shared underlying process. Repetition can increase efficiency, familiarity with stimuli, or learned shortcuts. These gains matter for the practiced task but may not generalize to novel problems.
Different instruments also draw on different mixtures of processes. As someone gets better at an n-back exercise, they may rely more on recognition patterns or a specialized updating strategy and less on the general processes engaged during initial learning. This possibility is discussed in research on mechanisms of working-memory training. The result illustrates why performance improvement cannot be equated automatically with modification of a stable latent trait.
There is also a measurement issue. A brief test taken shortly after training may favor recently practiced processes, whereas durable changes in broad intellectual ability would need to be demonstrated across distinct valid tasks, settings, and time points. Statistical controls for baseline scores, expectancy, test repetition, and publication bias can substantially affect conclusions.
Can working-memory performance be supported in everyday life?
Practical support can improve real-world functioning even when it does not raise a person's general intelligence. For learning and complex tasks, useful strategies include reducing unnecessary simultaneous demands, breaking multi-step instructions into meaningful stages, using checklists or notes, keeping intermediate calculations visible, and organizing related information into coherent groups. Such supports reduce immediate processing requirements and allow reasoning to use available knowledge more effectively.
A student solving algebra problems might write each intermediate result and annotate the goal of every step. That external record preserves information that otherwise must be maintained mentally, making an accurate solution more likely. The benefit lies in better task design and strategic support. It should be evaluated by improved accuracy, understanding, and ability to apply what was learned.
Reducing avoidable interruptions can similarly protect task continuity, particularly when a person must repeatedly reconstruct a complex mental model. For related evidence on environmental demands and concentration, see Media Multitasking: What It Is and How It Affects Attention, Memory, and Performance. This neighboring article addresses digital attentional demands, while the present article retains the comparative intelligence intent.
Sleep difficulties, stress, pain, low mood, medication effects, and unfamiliar testing conditions can alter day-to-day cognitive performance. A temporary episode of forgetfulness or concentration difficulty does not measure intelligence and does not establish a neurological or psychiatric diagnosis. Persistent or disruptive concerns warrant evaluation in the relevant professional context rather than self-classification from online cognitive games.
How should the evidence be interpreted in education and assessment?
Working-memory demands influence many classroom activities: understanding long instructions, carrying numbers, integrating information across paragraphs, and monitoring multi-step solutions. The educational implication is to examine where a particular task requires too much transient information to be held at once. Clear instructions, worked examples, staged problem solving, visual aids, and external records can improve accessibility.
Educational outcomes also reflect teaching quality, subject knowledge, language familiarity, motivation, opportunity, and many contextual variables. A student can demonstrate sophisticated reasoning while struggling with lengthy oral directions. The reverse can also occur. Neither pattern provides a defensible shortcut to an IQ estimate. The interpretation of working-memory or intelligence scores should follow the AERA, APA, and NCME Standards and the purpose and population for which each instrument was validated.
In research and professional assessment, the strongest practice is to report which construct and task were measured, how the result compares with appropriate reference data, the uncertainty involved, and what competing explanations remain. Group averages and factor correlations do not determine the intellectual potential or learning trajectory of any particular person.
Common misconceptions
“Working memory and IQ are the same thing”
They are distinct. A working-memory task estimates performance on temporary maintenance and processing demands; IQ is a standardized score derived from a particular intelligence assessment and its norms. The scores may correlate, yet they cannot be converted into one another using a universal rule.
“A larger working memory automatically means greater intelligence”
Research finds a robust relationship in many samples, particularly for fluid reasoning. Individuals also vary in knowledge, strategy, processing speed, motivation, and the particular mix of skills a task requires. A person's actual performance is context dependent, and the relationship between measures does not establish a single sufficient cause.
“If memory exercises increase my n-back level, they increase my IQ”
An increase in n-back performance demonstrates learning on that task. It does not demonstrate far transfer to independent measures of reasoning or general intelligence. The best-controlled meta-analytic literature gives little support to reliable broad intellectual gains from the available working-memory-training programs.
“People with working-memory problems must have low intelligence”
Temporary difficulty holding information can coexist with strong reasoning, specialized expertise, and high achievement. Task load, attention, fatigue, and conditions of assessment matter. A specific cognitive complaint should be understood through its actual demands and context, rather than used to assign a global intellectual label.
“The working-memory and intelligence correlation proves one causes the other”
Covariation supports research hypotheses, not unique causal direction. Latent-variable models, experiments, longitudinal studies, and well-controlled interventions supply different pieces of evidence. Even a strong statistical explanation of overlap must be distinguished from demonstration of an intervention that changes a broad ability.
Frequently asked questions
Is working memory part of intelligence?
Working-memory capacity is often represented as one cognitive-ability domain in hierarchical psychometric frameworks and is correlated with other domains. Whether it is described as a component, correlate, or contributor depends on the model. It is particularly relevant to fluid reasoning but does not exhaust the meaning of intelligence.
Is working memory more related to fluid or crystallized intelligence?
The literature generally identifies a closer relationship with fluid reasoning, which requires constructing and evaluating unfamiliar relations. Working memory also supports knowledge use and learning, so it can matter for crystallized tasks. The strength of either relationship depends on measurement, age, task demands, and sampling.
How strong is the correlation between working memory and IQ?
There is no single scientifically defensible correlation for every person, test battery, and population. Research usually estimates associations between particular memory and reasoning measures, and latent-variable estimates may differ from raw-score correlations. Claims of an exact universal percentage should be checked against the measures, sample, and analytic model behind them.
Does poor working memory mean someone is less smart?
A weak result on one task cannot establish a person's general intellectual ability. Working-memory performance has multiple influences, including attention, comprehension of instructions, anxiety, fatigue, familiarity, and the properties of the test. Broad interpretation requires appropriately validated measures and professional judgment when the stakes are high.
Can working-memory training raise IQ?
Training can improve performance on practiced and some related memory tasks. High-quality controlled studies and meta-analyses have generally failed to show reliable far transfer to broad intelligence measures. A particular program would need credible evidence from independent intelligence outcomes before a general IQ-improvement claim is warranted.
Are working memory and executive function identical?
Working memory is commonly discussed as part of broader executive-function research because maintaining goals and updating information often require control. Executive functions also include inhibition, switching, and other regulatory processes. Their measurement overlaps but the constructs should not be treated as synonyms.
Can working memory affect reading and mathematics?
Yes, working-memory demands can affect understanding of extended text, carrying intermediate values, and completing complex procedures. Reading and mathematical achievement also rely on domain knowledge, instruction, language, and other skills. An association does not imply that computerized working-memory drills will improve these academic outcomes.
Why does a difficult reasoning problem feel like a memory problem?
Novel problems may require several rules and intermediate conditions to be held in an accessible form. Losing one can interrupt reasoning, while a strategy or external notation can reduce that demand. The feeling of mental overload identifies a task experience but is not a reliable measure of latent capacity or intelligence.
How can I tell whether a cognitive-training product has convincing evidence?
Look for randomized trials with active comparison conditions, prespecified outcomes, multiple independent measures of reasoning, durable follow-up, adequate sample sizes, and replication by researchers outside the developer group. Improvement on the game's own levels is not adequate evidence for claims about IQ, school performance, or generalized cognition.
Conclusion
Working memory and intelligence are related because complex cognitive performance often depends on accessible information, attention regulation, and retrieval, particularly when a person must solve unfamiliar problems. A substantial relationship between working-memory capacity and fluid reasoning is supported by psychometric and experimental research. Explanations of that relationship remain multifaceted, and a causal claim requires evidence beyond correlation.
The intervention literature provides a separate and important finding: practicing memory-intensive exercises can improve task performance, while robust gains in independent measures of general intelligence have not been consistently demonstrated. In practical contexts, the most defensible approach is to measure distinct cognitive constructs accurately and design tasks, teaching, and work environments that allow available abilities to operate effectively.
Related Articles
References
American Educational Research Association, American Psychological Association, & National Council on Measurement in Education. (2014). Standards for Educational and Psychological Testing. Official testing standards.
Baddeley, A. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences, 4(11), 417–423. https://doi.org/10.1016/S1364-6613(00)01538-201538-2).
Conway, A. R. A., & Kovacs, K. (2013). Individual differences in intelligence and working memory: A review of latent variable models. Psychology of Learning and Motivation, 58, 233–270. https://doi.org/10.1016/B978-0-12-407237-4.00007-4.
Đokić, R., Koso-Drljević, M., & Bilalić, M. (2025). Past reflections, present insights: A systematic review and new empirical research into the working memory capacity (WMC)–fluid intelligence (Gf) relationship. Intelligence, 108, 101874. https://doi.org/10.1016/j.intell.2024.101874.
Jaeggi, S. M., Buschkuehl, M., Jonides, J., & Perrig, W. J. (2008). Improving fluid intelligence with training on working memory. Proceedings of the National Academy of Sciences, 105(19), 6829–6833. https://doi.org/10.1073/pnas.0801268105.
Kane, M. J., Hambrick, D. Z., Tuholski, S. W., Wilhelm, O., Payne, T. W., & Engle, R. W. (2004). The generality of working memory capacity: A latent-variable approach to verbal and visuospatial memory span and reasoning. Journal of Experimental Psychology: General, 133(2), 189–217. https://doi.org/10.1037/0096-3445.133.2.189.
Melby-Lervåg, M., Redick, T. S., & Hulme, C. (2016). Working memory training does not improve performance on measures of intelligence or other measures of far transfer: Evidence from a meta-analytic review. Perspectives on Psychological Science, 11(4), 512–534. https://doi.org/10.1177/1745691616635612.
Oberauer, K. (2019). Working memory and attention: A conceptual analysis and review. Journal of Cognition, 2(1), 36. https://doi.org/10.5334/joc.58.
Redick, T. S., Shipstead, Z., Harrison, T. L., Hicks, K. L., Fried, D. E., Hambrick, D. Z., Kane, M. J., & Engle, R. W. (2013). No evidence of intelligence improvement after working memory training: A randomized, placebo-controlled study. Journal of Experimental Psychology: General, 142(2), 359–379. https://doi.org/10.1037/a0029082.
Sala, G., Aksayli, N. D., Tatlidil, K. S., Tatsumi, T., Gondo, Y., & Gobet, F. (2019). Near and far transfer in cognitive training: A second-order meta-analysis. Collabra: Psychology, 5(1), 18. https://doi.org/10.1525/collabra.203.
Shipstead, Z., Lindsey, D. R. B., Marshall, R. L., & Engle, R. W. (2014). The mechanisms of working memory capacity: Primary memory, secondary memory, and attention control. Journal of Memory and Language, 72, 116–141. https://doi.org/10.1016/j.jml.2014.01.004.
Syed, M., Lum, J. A. G., Byrne, L. K., & Skvarc, D. (2024). Examining working memory training for healthy adults—A second-order meta-analysis. Journal of Intelligence, 12(11), 114. https://doi.org/10.3390/jintelligence12110114.
Unsworth, N., Fukuda, K., Awh, E., & Vogel, E. K. (2014). Working memory and fluid intelligence: Capacity, attention control, and secondary memory retrieval. Cognitive Psychology, 71, 1–26. https://doi.org/10.1016/j.cogpsych.2014.01.003.