RESEARCH DOSSIER
Working Memory
Abstract. Working memory enables temporarily accessible information to guide ongoing cognitive operations, goal-directed behavior, and the integration of new evidence. This fundamental research monograph examines the construct's historical emergence, multicomponent and attention-based theoretical architectures, verbal and visuospatial representations, capacity limitations, interference, retrieval, updating, and prioritization. It compares complex-span, change-detection, continuous-report, neural decoding, electrophysiological, and intervention methodologies while reviewing developmental and individual-difference findings. Recent 2025 theoretical reassessment and 2026 meta-analytic evidence refine the comparison of working-memory models and directional training transfer. Critical evaluation distinguishes robust empirical results from interpretations specific to slot, resource, control, and decay models, and examines evidence concerning the limits of working-memory training and far transfer. The resulting synthesis specifies the explanatory roles of attention, knowledge, representational fidelity, and environmental support and identifies open problems for cross-task generalization and causal investigation.
Keywords: working memory; working memory capacity; executive attention; Baddeley and Hitch; phonological loop; episodic buffer; visual working memory; complex span; interference; memory updating; continuous report; cognitive control; working memory training
Scientific Definition and Research Scope
Working memory denotes the capacity to keep task-relevant information accessible while attention, transformation, or response selection continues. Its observable manifestations include maintaining a location across a delay, reordering a sequence, comparing a new stimulus with a remembered one, and integrating partial information during reasoning. The definition is functional: a measure must identify what information must remain available, what operations intervene, and what outcome demonstrates success. Baddeley & Hitch (1974) developed the construct by asking how immediate retention participates in comprehension, reasoning, and learning rather than treating immediate recall as a self-contained store.
The scope of working-memory research extends across verbal and visuospatial materials, sensory representations, practiced symbols, action plans, and higher-level task rules. The term encompasses related operations that experiments distinguish through selective interference, temporal delays, prioritized representations, and individual differences. Awh et al. (2006) reviews interactions between attention and short-term maintenance and shows why a unitary everyday notion of remembering is scientifically insufficient. Task success may reflect a high-fidelity representation, protection against distractors, controlled recovery of lost information, or a learned strategy that reduces demand. Distinguishing these pathways gives the construct explanatory precision.
This monograph synthesizes published theoretical, behavioral, psychometric, developmental, and neural evidence. Its main questions concern architecture, representational format, capacity, attentional control, interference, measurement, and generalization. The object remains narrower than memory as a whole: durable encoding and autobiographical reconstruction are considered when they affect temporary task-based access. Studies of artificial systems and cognitive offloading enter only where they clarify how external structures redistribute demands on human working memory. The aim is to compare established research programs and document their evidentiary implications, with citations connecting each empirical claim to an identifiable source.
The Historical Transition from Short-Term Storage
Early experimental accounts often separated short-term storage from longer-term memory and described performance using span, serial position, and retention intervals. The development of the working-memory concept changed the central question from how many items survive a short delay to how information supports another operation while temporarily accessible. Baddeley & Hitch (1974) systematically examined concurrent processing and memory load in reasoning, comprehension, and learning. The resulting dual-task evidence encouraged an architecture with partly distinguishable functional components, replacing the assumption that temporary memory is adequately characterized by a single passive receptacle.
The historical shift is especially apparent in the interpretation of span tasks. A person's immediate reproduction of a digit sequence can depend on phonological recoding, grouping, rehearsal opportunities, familiarity, and response method. A score is therefore a result of a task procedure as well as a property of the participant. Cowan (2001) revisited classic estimates of immediate capacity and argued that experimental conditions controlling recoding and grouping yield smaller core estimates than a general invocation of seven items. The interpretation of such estimates depends on what constitutes an item and which forms of long-term knowledge remain available.
Historical models remain scientifically useful when their characteristic predictions are tested. A multicomponent model predicts modality-sensitive interference patterns; attention-oriented models predict costs when priorities or control demands change; representational models predict accuracy and precision distributions under load. These predictions can be compared in experiments rather than ranked by historical influence. Oberauer et al. (2016) demonstrates this principle in a review of decay, resources, and interference as explanations for capacity. The continuity of the field lies in progressively more explicit operational contrasts, not in treating a celebrated model diagram as an anatomical map.
Baddeley and Hitch's Multicomponent Framework
The original program of Baddeley & Hitch (1974) proposed a working system with a control component and subsidiary mechanisms for holding information. The theoretical importance of this framework is its commitment to differentiated interference: verbal retention and spatial processing can impose different costs, while executive tasks compete for control. Research has commonly described the phonological loop, visuospatial sketchpad, and central executive as components, but their names designate functional hypotheses. A selective dual-task effect can support functional specialization without showing that a discrete neural compartment implements each component.
The phonological-loop account interprets some immediate verbal-memory findings in terms of temporary phonological representation and articulatory rehearsal. The visuospatial component addresses temporary storage and manipulation of visual and spatial patterns. Executive control coordinates goals, processing, and allocation. Experimental materials seldom align perfectly with these distinctions: visually displayed letters may be verbally recoded, and spatial reconstruction may require verbal landmarks. Thus a careful experiment manipulates the opportunity to recode, the duration of retention, and concurrent processing while identifying alternative strategies. Awh et al. (2006) stresses that multiple forms of attention and memory interact, complicating simple component labels.
The multicomponent framework has undergone substantial revision. Baddeley (2000) proposed an episodic buffer to account for binding and multimodal integration with information from long-term memory. Its role is especially relevant when a participant retains integrated episodes, complex sentences, or features whose combination matters more than isolated attributes. This addition is a theoretical response to explanatory limits of the early model; the buffer's existence and boundaries must be evaluated against competing mechanisms for attention-based binding and retrieval. A mature reading of the framework treats revision as normal model development and distinguishes experimental phenomena from a preferred explanation.
The Episodic Buffer and Binding
Binding refers to maintaining which features belong together, which objects occupy particular locations, or how events relate within a temporary episode. Baddeley (2000) formulated the episodic buffer as a limited-capacity multimodal system integrating material from subsidiary components and long-term memory. The hypothesis gives the multicomponent tradition a way to discuss complex representational unity. It also directs empirical attention toward conjunction errors, integration across modalities, and the retention of structured meaningful material. Findings must distinguish preserved individual features from preserved relations; successful recognition of parts does not establish that their original pairing was retained.
An experiment on feature binding needs independent measures of item memory, association memory, and response bias. A participant may retain two colors and two shapes yet confuse which color accompanied which shape. An observed conjunction error could arise during encoding, temporary maintenance, comparison, or retrieval. Models need to state which stage yields the predicted pattern. Luck & Vogel (1997) reported that observers could briefly retain combinations of visual features for several objects, an influential result for object-based accounts. Subsequent work on precision and probabilistic error distributions refined the interpretation of what was stored.
Binding also illustrates the interaction of working memory with knowledge. Familiar configurations can be encoded efficiently because semantic or perceptual structure supplies constraints; unusual combinations demand more deliberate maintenance. This creates a methodological challenge whenever verbal meaningfulness or familiar objects are compared with abstract features. Ma et al. (2014) surveys changing explanations of visual storage and highlights how model assumptions influence conclusions drawn from behavioral distributions. Rather than assigning every integrated representation to one component, research should measure binding accuracy, feature precision, and interference under conditions that manipulate familiarity and attentional prioritization.
Embedded Processes and the Focus of Attention
Attention-based accounts characterize working memory partly as temporarily activated knowledge and a narrower focus that grants privileged access to selected information. Cowan (2001) argued for a small core of readily accessible chunks under restricted task conditions. Such accounts explain why currently prioritized information can be accessed quickly, while other recently presented information remains recoverable with additional search. They also make predictions about cueing, switching between items, and interference. A capacity estimate therefore depends on whether the task tests immediate focal access, broader temporary activation, or later retrieval aided by context.
Oberauer (2002) experimentally distinguished a subset actively used in arithmetic from items retained without current manipulation. Operation latencies tracked the active set, and changes in the currently operated item imposed switching costs. These results motivated a more differentiated view of access within working memory. The key claim concerns functional priority rather than subjective awareness: an item can remain relevant to the experiment without being the current input to an operation. Experiments that cue a subset after encoding can test shifts in accessibility, although strategy and cue predictability must be controlled.
An attentional focus does not settle the representational nature of its contents. Priority may arise through stronger neural activity, temporarily altered synaptic connectivity, retrieval thresholds, or changes in decision strategy. A single behavioral advantage can often be generated by several process models. D'Esposito & Postle (2015) summarizes neural evidence for the allocation of attention to internal representations and the importance of distributed cortical mechanisms. Bridging behavioral attention and neural implementation requires identifying what information is represented, how it becomes accessible, and which interventions selectively change access.
Executive Attention and Goal Maintenance
Working-memory capacity has also been interpreted as a measure of the ability to sustain task goals when distracting information or habitual responses compete. Engle (2002) reviewed evidence relating span performance to interference-demanding tasks such as Stroop and antisaccade procedures. On this account, strong performance may depend on keeping a goal operative and resisting capture by more automatic responses. The theoretical claim differs from counting storage slots: a person can fail an operation because the correct rule loses priority even when the relevant stimulus was encoded.
Goal maintenance is experimentally separable from stimulus retention only with carefully chosen controls. Suppose two participants can recognize recently shown digits equally well, but one is more likely to follow an obsolete sorting instruction after a rule change. The difference concerns the currently prioritized task set. Conversely, a participant may maintain the rule while losing which digit must be transformed. Friedman & Miyake (2017) reviews convergent and distinct executive functions, reinforcing the need to distinguish control constructs from task-specific demands. Correlations between span tests and other executive measures support related variation but do not uniquely identify one common causal mechanism.
Contemporary research uses selective distraction, cue validity, task switching, and multi-measure latent-variable approaches to estimate control. Unsworth & Engle (2007) adds an account in which accessible information in primary memory and controlled retrieval from secondary memory both contribute to individual differences. That perspective explains why an apparent working-memory failure sometimes reflects ineffective search rather than loss of all information. The general implication is that storage, priority, and retrieval form interacting mechanisms. A strong explanatory model must predict when each mechanism dominates and how experimental manipulations distinguish their contributions.
Representations, Time, and Cognitive Operation
Working memory operates across time: information must be encoded, remain recoverable, and be brought into contact with an operation or comparison. The relevant interval is defined by the task rather than a universally fixed duration. A delayed change-detection experiment emphasizes maintenance and comparison, while complex span requires intermittent processing during retention. Conway et al. (2005) documents how different span procedures combine temporary storage with concurrent activity. Researchers must identify whether their dependent variable is recall order, item identity, precision, speed, or resistance to a distractor, since these outcomes constrain different mechanisms.
The distinction between maintenance and manipulation matters for interpretation. Repeating a sequence in its presented order may be supported by practiced serial encoding; recalling it in reverse order adds transformation and response planning. An increase in errors under reversal therefore measures a compound change and cannot be assigned to central executive capacity alone. Oberauer (2002) provides a concrete demonstration of the costs involved in accessing and changing the active item. Within-task comparison of conditions, balanced stimulus sets, and independent manipulation checks are more informative than attaching a theoretical label to a single score.
Temporal measurement can also reveal the difference between ongoing activity and durable accessibility. Neural patterns during a delay may vary with attention, expected probe, or intervening demands, while later performance demonstrates whether information remains available. Rose et al. (2016) examined reactivation of latent working-memory representations with transcranial magnetic stimulation, providing evidence relevant to states whose current relevance has changed. Such experiments challenge the idea that every maintained item must have an identical sustained neural signature. Their inference depends on decoding validity, intervention specificity, and the relation between reactivation and behavior.
Capacity Limits and the Meaning of an Item
Working-memory capacity is an operationally estimated limit on accessible information, not a directly observed fixed container size. The number counted depends on whether a stimulus is treated as a separate feature, an integrated object, a familiar chunk, or a sequence that can be compressed through learned relationships. Cowan (2001) reexamined the classic capacity literature and argued that under conditions restricting recoding, grouping, and rehearsal, an estimate near four chunks is defensible. The proposal concerns a constrained form of storage and should not be generalized without specifying material, task design, and measurement procedure.
A participant may reproduce many letters arranged into a familiar acronym and fewer letters arranged randomly. The difference can be explained through accumulated long-term knowledge that changes the unit of representation, rather than enlargement of a temporary store. Cowan (2010) distinguishes conditions that reveal a smaller central limit from ordinary tasks allowing grouping and recoding. The analysis implies that number-of-items statements are scientifically incomplete unless the definition of an item, the retention interval, interference conditions, and scoring rules are known. The same person can receive materially different capacity estimates from equally legitimate experiments.
Capacity studies therefore benefit from jointly measuring quantity and quality. All-or-none scoring of a changed item may conceal a weak, imprecise representation, while a continuous report can reveal the magnitude and distribution of recall errors. Ma et al. (2014) reviews how such procedures altered debates about whether a limited number of objects are stored or a flexible resource is divided among items. A credible capacity theory must explain why some conditions yield apparently discrete limits and others show gradually declining precision. It must further predict how stimulus complexity, interference, and attentional cues moderate those effects.
Visual Change Detection and Object-Based Storage
Visual change-detection tasks generally present a brief array, insert a delay, and ask whether a later probe differs from the original material. Luck & Vogel (1997) found performance consistent with retaining features and conjunctions for a limited number of objects, supporting the influential idea that visual working memory can represent integrated objects. The experiment established a behavioral pattern under a particular stimulus regime. Its interpretation must be connected to encoding time, feature discriminability, guessing, attention, and possible verbal labeling. Counting retained items from accuracy presupposes a model linking success and error to stored representations.
Models often estimate retained information by comparing hits and false alarms or by fitting response distributions. These estimates should be treated as model-dependent summaries, not measurements of a literal physical capacity. A visual array with highly discriminable colors may support different judgments than an array of similar orientations or richly textured objects. Vogel & Machizawa (2004) used lateralized electrophysiological activity during visual maintenance to investigate individual differences in effective storage. The neural association strengthened the case for studying retained representations, while leaving open how selective encoding and control affect the measured capacity.
Experimental credibility depends on separating the number of presented stimuli from the number of encoded stimuli and the number still represented at test. Probe format can induce familiarity judgments that differ from exact feature retrieval, and a distractor can change the task even when the nominal memory load is identical. Vogel et al. (2005) found neural evidence that filtering irrelevant material differs across participants. This shows why a person who reports fewer relevant objects may have allocated capacity to irrelevant ones. Visual change detection therefore measures the interaction of storage, selection, and decision, which must be modeled explicitly.
Slots, Discrete Representations, and Precision
Discrete-slot models posit that a limited number of objects receive high-quality representations, with other objects unavailable once slots are occupied. Zhang & Luck (2008) analyzed delayed feature reports and obtained error patterns consistent with retaining a subset of high-resolution objects rather than evenly degrading all representations. Such findings gave a strong empirical formulation to the concept of discontinuous storage. Their significance lies in the shape of the error distribution and its relation to set size. The conclusion is conditional on assumptions about guessing, feature noise, spatial confusions, and the response process.
Continuous-resource models instead posit a limited budget that can be flexibly allocated, so precision tends to decline as more items compete. Bays et al. (2009) presented evidence that the precision of visual working memory depends on how a shared resource is distributed. Flexible allocation predicts systematic variation with priority and load, including some recoverable information about many items under appropriate conditions. A model can combine resource limits with item-level thresholds or attentional gating, so the empirical comparison must be sharper than a verbal choice between two labels.
Ma et al. (2014) synthesizes this dispute and emphasizes the need for generative models that predict entire response distributions, rather than isolated mean accuracy. A slot model may accommodate imprecision within occupied slots; a resource model may generate reports resembling guesses when memory becomes noisy. Both can fit portions of the evidence if unconstrained parameters multiply. Model recovery, out-of-sample prediction, preregistered comparisons, and experiments that vary priority independently of set size offer more discriminating tests. The enduring question is which representation-allocation mechanism explains generalization across tasks and stimulus spaces.
Decay, Interference, and Competition
Temporal decay theories attribute forgetting partly to a loss of representation strength over time. Interference theories attribute errors to competition among overlapping representations, retrieval cues, or responses. Resource theories explain costs through limited representational or control capacity. Oberauer et al. (2016) compared evidence across these explanatory programs and found considerable difficulty for a simple temporal-decay account, particularly for verbal memoranda. The review identified interference as promising while acknowledging that each account faces unresolved findings. This evaluation concerns specified models and task domains, not a universal statement that elapsed time has no consequence.
An experiment that lengthens a retention interval can simultaneously create more opportunities for internal rehearsal, distraction, strategic change, or contextual drift. The effect of elapsed time cannot be isolated merely by leaving a longer blank screen if the participant changes behavior during the interval. Conversely, inserting distractors adds representational competition and task-switching demands. Jarrold & Towse (2006) reviews sources of variation across working-memory tasks and demonstrates the importance of identifying which processes a paradigm captures. Factorial designs manipulating delay and interference separately offer clearer evidence than a single load comparison.
Interference also depends on similarity, ordering, and cue diagnosticity. Two nearby spatial locations or semantically related verbal items may compete differently from dissimilar materials. An apparent loss of the target can be a failure to select the correct representation at retrieval. Unsworth & Engle (2007) distinguishes active maintenance and controlled search from secondary memory, making retrieval competition part of the explanatory architecture. Models should predict intrusions, order errors, reaction times, and confidence in addition to totals correct. Distinct error types can discriminate among decay, confusability, goal loss, and failed retrieval.
Updating and the Reconfiguration of Contents
Updating tasks require replacing or revising information that remains relevant to an ongoing objective. Their demands include identifying the affected item, integrating the new value, suppressing obsolete representations, and preserving unaffected information. The classic operation-span and complex-span literature measures concurrent processing and maintenance, but a high span score cannot be equated automatically with fast or accurate updating. Conway et al. (2005) documents the heterogeneity of working-memory span tasks and their measurement properties. Updating should be investigated through direct manipulations of replacement frequency, cue timing, memory-set structure, and interference.
Suppose a participant holds four locations and receives an instruction to exchange one location for another. The outcome can depend on whether the old coordinate remains familiar, whether the new one is discriminable, and whether the update cue clearly identifies the target. A delayed recognition test may conflate these influences; tracking errors toward old versus unrelated alternatives helps isolate failure to replace information. Oberauer (2002) illustrates how switching the current object of cognitive operation yields identifiable processing costs. Such costs reflect controlled access even when all original items remain potentially recoverable.
Updating provides a bridge to broader executive-function research because tasks often combine maintaining goals, selecting relevant evidence, and inhibiting obsolete responses. Friedman & Miyake (2017) shows that executive functions contain shared and distinguishable individual-difference components, which cautions against assigning an updating score to a single neural or psychological mechanism. Psychometrically stronger studies use multiple tasks per putative component and model shared variance across sessions. Theoretical progress requires explanations of when replacing content incurs a cost, how representations are marked as obsolete, and why some updates generalize more readily than others.
Verbal Working Memory and Rehearsal Strategies
Verbal working-memory research examines phonological representations, serial order, rehearsal, semantic organization, and speech-related interference. In a multicomponent account, verbal retention can benefit from articulatory refreshment, while concurrent verbal articulation may impair that strategy. Baddeley & Hitch (1974) contributed to this program by showing how short-term verbal retention interacts with ongoing cognitive activity. The model distinguishes the maintenance of information from its use in comprehension and reasoning. In contemporary experiments, a test of verbal storage must specify modality, familiarity, response mode, and the opportunities available for covert recoding.
Words and pseudowords yield different performance because meaning and lexical familiarity contribute to encoding and retrieval. Familiar chunks can collapse several nominal items into a more useful unit, while unfamiliar sequences can generate confusions based on sound similarity. Cowan (2010) explains why observed span should be disentangled from knowledge-based grouping before interpreting core capacity. Verbal tasks also mix order and identity requirements: remembering a set of letters is not equivalent to reproducing their order. The pattern of transpositions, omissions, and intrusions can identify mechanisms hidden by a total-correct score.
Cross-modal comparisons are useful but often undercontrolled. A visual display of written words invites phonological coding; a spatial layout can be described verbally; auditory sequences can be encoded as temporal rhythms. These strategies complicate claims that a task measures a pure domain-specific buffer. Oberauer et al. (2016) evaluates verbal evidence in relation to decay, interference, and capacity hypotheses. A strong protocol constrains or measures strategy, includes matched difficulty where possible, and reports how design choices change conclusions. These methodological details determine whether two verbal-memory theories genuinely make different predictions.
Visuospatial Working Memory and Prioritization
Visuospatial working memory supports the temporary representation of locations, orientations, colors, movement directions, and relations among objects. Visual feature estimation has been central to contemporary capacity debates because continuous responses reveal precision, bias, and substitution errors. Bays et al. (2009) connects memory fidelity to resource allocation, while Zhang & Luck (2008) shows why discrete-representation models were considered plausible. Their disagreement motivates experiments in which set size, feature similarity, and priorities are manipulated systematically. It also emphasizes that a model of color memory should not automatically be generalized to spatial sequences or complex scenes.
Selective attention can determine which visual features enter memory and which survive interference. Vogel et al. (2005) identified neural variation associated with excluding irrelevant information, while Vogel & Machizawa (2004) related maintenance-related activity to measured capacity. These results support the importance of gating and selection, but neural correlation alone does not establish a causal division between storage and attention. Behavioral checks should establish that participants perceive the stimuli, understand cues, and adopt the prescribed priority policy. Otherwise apparent capacity differences may partly reflect differences in encoding or test strategy.
Features in a visual object can remain jointly accessible or become misbound as load rises. The relevant empirical outcome includes whether a participant reports a target's correct feature, selects a feature belonging to another item, or makes an unrelated report. Luck & Vogel (1997) established an influential object-based finding, while later precision work refined the representation-level interpretation. Across experiments, a stable distinction is needed between features, objects, and their contextual associations. The visuospatial literature is consequently a test case for broader questions of how temporary information becomes organized and how prioritized access governs its use.
Complex Span and Dual-Task Measurement
Complex-span tasks combine a processing demand with material that must be retained for later recall. Counting span, reading span, and operation span differ in the kind of intervening activity and the opportunities they provide for rehearsal, grouping, and strategy. Conway et al. (2005) offers a methodological analysis of these measures, including reliability, scoring, and interpretation. The theoretical advantage of complex span is that it imposes competition between immediate storage and another operation. The limitation is that a score combines several task processes, so its association with an external outcome cannot identify which single process explains the relationship.
Unsworth et al. (2005) developed an automated operation-span procedure that standardizes administration and scores performance, allowing researchers to compare individuals under a controlled task structure. The study reported reliability evidence and correlations with related working-memory measures. Such psychometric performance supports its use in specified research settings, while the exact interpretation still depends on compliance, processing accuracy, and the participant's strategy. A participant who remembers more letters by neglecting the arithmetic component has not demonstrated equivalent performance to one who solves the operations accurately and retains the same number of letters.
Task administration should define practice trials, processing-time limits, treatment of processing errors, serial-order scoring, and exclusions in advance. Short forms risk greater measurement error; repeated sessions introduce learning and familiarity. Conway et al. (2005) discusses why extreme-group comparisons and single-test inferences need special care. For scientific interpretation, a latent working-memory factor estimated from multiple paradigms is often more defensible than a score treated as a transparent measure of a named brain function. Independent replication must reproduce both the instructions and the statistical interpretation, not merely the nominal task label.
Delayed Recognition and Continuous Report
Delayed recognition typically separates encoding, retention, and probe judgments. Change detection measures whether a difference can be detected, whereas delayed estimation asks a participant to reproduce an orientation, hue, or spatial coordinate on a continuous scale. Ma et al. (2014) analyzes how these methods yield different evidence about representational capacity and fidelity. A categorical response can be correct through coarse information that would fail a precise feature report. For this reason, the chosen response format changes the observable construct even when the display and retention interval remain similar.
Continuous-error distributions can be decomposed by models into target-centered imprecision, guesses, or responses corresponding to non-target items. These terms belong to fitted models and require validation against alternative generative processes. Zhang & Luck (2008) interpreted its distributions as evidence for high-resolution subset storage, whereas Bays et al. (2009) supported flexible allocation of a shared resource. An analyst should compare likelihoods, parameter recoverability, and predictive fit across several stimulus conditions. A favorable fit is an achievement of a model under its assumptions; it does not independently prove that an unobserved psychological component exists.
Both delayed-recognition and continuous-report methods should vary retention duration, set size, distractors, cue timing, and feature discriminability where the research question requires them. Eye position, initial perceptual errors, and response-motor noise can affect estimated precision. Separate sensory and response controls help prevent attribution of all error to memory. Vogel et al. (2005) further shows that which stimuli are selected for storage can influence apparent capacity. Methodological triangulation—behavioral errors, response times, neural measures, and appropriately controlled manipulations—helps distinguish mechanisms that a single accuracy curve leaves underdetermined.
Attention Filtering and Distractor Resistance
Selective access to temporary storage is essential when the environment contains irrelevant objects. Vogel et al. (2005) used electrophysiological markers to investigate why some participants maintain task-relevant representations more selectively than others. Their finding that irrelevant items can consume storage-related activity changes the interpretation of a low span score: the difficulty may involve inefficient filtering as well as a representational limit. In experimental design, target and distractor counts should therefore be manipulated independently. The resulting behavioral and neural contrasts can establish whether access control changes with task demands.
Resistance to distraction is not identical to preventing initial entry. Information can be admitted to memory and subsequently displaced, suppressed, or confused with targets. A distracting stimulus may also interfere with retrieval or selection of a response. Sakai et al. (2002) investigated distractor-resistant maintenance associated with prefrontal area 46, adding evidence about the neural conditions of selective retention. Behavioral interpretations still require controls for perceptual attention and response conflict. A stronger design measures whether distraction changes memory for particular items, not only whether aggregate performance deteriorates.
Research on filtering interacts with theories of executive attention. Engle (2002) proposes that maintaining the intended goal becomes particularly important under interference. Unsworth & Engle (2007) adds the possibility that retrieval must recover target information from competing candidates when accessibility is reduced. These explanations can be distinguished by measuring which distractors were remembered, whether participants retain the task rule, and whether an explicit retrieval cue rescues performance. Their relationship is mechanistic and conditional: different failures can yield similar error totals, so convergent tasks are essential.
Neural Correlates and Distributed Maintenance
Neuroimaging and electrophysiology have linked working-memory tasks to distributed frontoparietal and sensory systems. D'Esposito & Postle (2015) reviews findings that attention to internal representations helps organize temporary access and that neural support extends beyond a single frontal location. Some task-related activity reflects maintaining a feature, some reflects controlling priority, and some reflects the planned response. A region activated during a delay can support several operations, making reverse inference from activation to a unique memory component unreliable. Experimental contrasts and multivariate decoding must specify what representational information they can actually distinguish.
Harrison & Tong (2009) decoded the orientation of a remembered visual pattern from activity in early visual areas, providing evidence that stimulus-specific information can remain represented in sensory cortex after the stimulus disappears. Such decoding does not entail that a local region independently stores a complete memory or that the readout establishes subjective experience. The measured pattern may reflect distributed recurrent interactions, feature templates, and task priorities. Researchers should report decoding cross-validation, feature specificity, delay timing, and whether classification exceeds an appropriate control baseline.
Evidence for transient or latent states has complicated a strict equation between storage and sustained firing. Rose et al. (2016) reported reactivation of a less prioritized working-memory representation following stimulation, showing that information relevant to later behavior may have a neural signature that changes with its present priority. These findings motivate hypotheses about activity-silent or altered-activity maintenance, while the exact biophysical implementation remains an active research question. Causal stimulation studies require especially careful inference because stimulation affects connected networks and may alter attention as well as the hypothesized stored representation.
Developmental Differences and Educational Performance
Working-memory demands change across development because strategy use, language, prior knowledge, attention control, and task familiarity develop alongside performance. Alloway et al. (2009) examined children with low working-memory scores and reported associations with learning difficulties and classroom behavior. The research demonstrates the educational relevance of temporary maintenance under task demands, but a screening association does not mean that working-memory capacity alone causes every learning difficulty. The sampled children's other cognitive and environmental circumstances matter, and tasks differ in how accessible they are at different ages.
Gathercole et al. (2004) studied relationships between working-memory measures and school attainment at ages seven and fourteen, finding associations that differed across subject and age. These results make working memory a meaningful predictor in educational contexts while leaving open the causal contributions of instruction, knowledge, motivation, and reading proficiency. A classroom task may impose multiple simultaneous demands: remembering an instruction, transforming information, selecting a response, and monitoring errors. A practical educational interpretation therefore specifies which demand is being reduced and measures whether actual learning improves.
Cross-sectional developmental differences alone cannot determine how a cognitive mechanism changes within an individual. Longitudinal observation, task adaptation, and age-appropriate measurement are needed to separate maturation, experience, practice, and cohort effects. Jarrold & Towse (2006) identifies conceptual difficulties in interpreting individual differences across working-memory paradigms. Researchers should also assess whether instructions and scoring have the same psychological meaning across groups. When group-level associations are reported, they should not be converted into deterministic judgments about an individual child or a claim that one standardized intervention is appropriate for everyone.
Individual Differences, Intelligence, and Control
Performance on complex-span tasks correlates with performance on other demanding cognitive measures, including reasoning and attentional-control tasks. Engle (2002) interpreted these relationships through executive attention, while Unsworth & Engle (2007) emphasized both maintenance in primary memory and controlled search of secondary memory. The existence of a correlation supports a shared source of variance; it does not establish that a span score measures general intelligence or that capacity training will necessarily improve broad cognitive ability. Differences in processing speed, strategic knowledge, and motivation can contribute to observed associations.
Psychometric models can estimate shared variance across several tasks and account for task-specific error. Conway et al. (2005) considers the measurement implications of complex span, and Friedman & Miyake (2017) demonstrates how executive-function research distinguishes common and specific factors. In interpreting latent variables, investigators must verify model identification, adequate sample size, test reliability, and stability across groups. A factor derived from three similar tasks may partly reflect shared format rather than a domain-general mechanism. Strong construct validation requires predicting independently designed outcomes under conditions that constrain competing explanations.
Individual differences also depend on state and context. Fatigue, stress, incentives, and previous experience can alter performance, while repeated tests can change strategy. A distinction between relatively stable differences and temporary fluctuations requires multiple sessions and hierarchical analysis. Vogel et al. (2005) shows that selective encoding of distractors relates to effective memory differences, providing an experimentally useful component of performance variation. The most informative scientific account links a score to identifiable mechanisms and boundary conditions rather than treating a person's capacity as a timeless scalar property.
Cross-Domain Generalization and Ecological Validity
A central challenge is whether parameters estimated from one paradigm predict performance with different materials, tasks, and settings. Verbal complex span, visual change detection, and spatial updating differ in coding format, attentional requirements, and outcome measures. Oberauer et al. (2016) evaluates capacity theories against multiple patterns of findings instead of one signature effect. The methodological principle is that a mechanism proposed as general must survive systematic cross-task tests. A theory that explains only a familiar experimental stimulus family may be useful but needs a narrower scope than a universal working-memory architecture.
Ecological validity concerns both representativeness and causal transport. Remembering a phone number, following an instruction, and interpreting a multi-step argument all require retaining information, but they involve different language knowledge, motivational conditions, and external supports. Alloway et al. (2009) and Gathercole et al. (2004) demonstrate why real-world educational associations deserve attention. Yet translating a laboratory correlation into a causal classroom intervention requires designs that test changed learning outcomes under comparable instruction and exposure. Predictive utility and intervention efficacy should be reported separately.
Generalization is also affected by the way tasks are explained and scored. A long laboratory sequence may be unfamiliar, while an analogous operation embedded in a practiced profession may be efficiently chunked. Cowan (2010) emphasizes how chunking changes apparent limits. Research should compare both standardized and situated performance, use diverse participant groups where possible, and test whether proposed mechanisms continue to explain variance after familiarity and strategy differences are controlled. These practices preserve the theoretical value of controlled experiments while recognizing that human working memory always operates in context.
Working-Memory Training and Transfer
Training studies examine whether repeated practice on demanding memory tasks changes performance beyond the practiced procedure. Klingberg (2010) reviewed evidence for trainability and plasticity, helping establish this research program. A participant can become faster or more accurate on an adaptive n-back task by learning its timing, familiarizing with stimuli, or improving a practiced control strategy. Such improvement is a valid near-task outcome, but whether it demonstrates a general increase in working-memory capacity is a separate hypothesis. The evaluation requires independent tests that do not reproduce the training task's surface characteristics.
Redick et al. (2013) conducted a randomized, placebo-controlled comparison involving adaptive dual n-back training, an active visual-search comparison, and a no-contact group. The published findings did not support broad gains on the studied cognitive-ability measures despite practice-related improvement. This design addresses expectancy and contact effects more carefully than a simple comparison with an inactive group. Its conclusions apply to the particular intervention, outcomes, and sampled participants. It contributes one important test to a broader literature that includes varying training intensity, control groups, outcome definitions, and follow-up intervals.
A later meta-analytic review by Melby-Lervåg et al. (2016) concluded that evidence did not support reliable improvements on intelligence or other far-transfer measures from working-memory training, while improvements on trained or similar tasks can occur. Shipstead et al. (2012) had already identified design and theoretical problems that complicate strong transfer claims. A rigorous synthesis separates task learning, near transfer to structurally similar tests, and far transfer to different cognitive or everyday outcomes. Confidence in benefits depends on preregistered outcomes, active controls, blinded assessment where possible, adequate power, and independent replication rather than on the attractiveness of a general brain-training narrative.
Assessment, Clinical Contexts, and Interpretive Limits
Working-memory measures appear in developmental, educational, neuropsychological, and clinical research because many difficulties involve keeping goals and relevant information accessible. D'Esposito & Postle (2015) describes working memory as a central topic in cognitive neuroscience while emphasizing complex distributed neural control. A low score is a measurable performance outcome within a testing situation; its diagnostic interpretation requires a broader assessment framework. Similar performance can arise from difficulties with encoding, sustained attention, instruction comprehension, language, speed, fatigue, or distraction. The assessment must establish which mechanisms and contextual factors are plausibly operative.
Standardization improves comparability, but norms and test conditions have to match the evaluated population. A complex verbal-span test can impose language and schooling demands, and a visual task may be affected by perception and motor control. Conway et al. (2005) analyzes reliability and validity concerns in span measurement that remain relevant whenever scores guide interpretation. Where repeated measures are available, within-person patterns across conditions may add information beyond a one-time total. Clinically useful analysis also distinguishes an impairment hypothesis from the observed inability to perform a specific testing operation under specific circumstances.
Educational and developmental evidence provides examples of associations without a simple diagnostic shortcut. Alloway et al. (2009) documented academic and classroom correlates among children with low working-memory scores, while Gathercole et al. (2004) linked working-memory tests with curriculum assessments. These studies support assessment of task demands and support needs but do not validate treating capacity as a standalone diagnostic category. Claims about treatment efficacy belong to controlled intervention evidence, and individual recommendations require qualified assessment. The monograph's role is to explain the scientific constructs, their measurements, and the evidentiary limits of clinical extrapolation.
The Role of Knowledge, Learning, and Expertise
Working-memory performance depends on the structure provided by knowledge. A novice may attempt to retain numerous unrelated pieces of information, while an experienced reader or practitioner organizes them into familiar relations. Cowan (2010) makes chunking central to understanding why capacity estimates vary across tasks. This point has broad implications for learning: an instructional change may improve observable working-memory performance by changing encoding and retrieval structure rather than altering a domain-general capacity limit. A scientifically useful explanation must identify whether the intervention supplied prior knowledge, changed strategy, reduced interference, or improved control.
Semantic and perceptual knowledge can provide retrieval cues for partially maintained information. Unsworth & Engle (2007) describes how controlled search through secondary memory contributes to span-related individual differences. The framework explains why knowledge-driven recovery may sustain performance even when information is no longer within the most accessible focus. It also suggests that the apparent continuity of active maintenance can sometimes be produced by repeated successful retrieval. To test that hypothesis, experiments can manipulate cue availability, semantic organization, and time pressure while recording the types of mistakes participants make.
Knowledge complicates cross-group comparisons because materials may be differently meaningful for children, specialists, or speakers of different languages. Jarrold & Towse (2006) reviews task-related sources of individual variation, while Gathercole et al. (2004) illustrates the relationship of memory measures to attained academic skills. A useful educational intervention may modify external presentation, teach strategies, or sequence complex steps. Its success should be measured in real learning and retention, including transfer to unfamiliar tasks, with attention to whether benefits persist after supports are removed. The main research question is how representations and demands interact, not whether every improvement reflects a larger internal store.
External Supports, Cognitive Offloading, and Artificial Systems
External notes, displays, checklists, and digital tools can change the quantity and kind of information a person must maintain internally. Working-memory theories clarify why such supports may be useful, but the relevant outcome is the combined task system. An external record can lower demands for temporary retention while introducing new demands for locating, interpreting, updating, and verifying information. Baddeley & Hitch (1974) makes ongoing processing central to working memory, and D'Esposito & Postle (2015) connects accessible representations with control. These ideas permit precise comparisons of unaided and supported performance without assuming that external storage itself is a human memory component.
A digital assistant can hold intermediate values, summarize instructions, or retain a task state, while the human must assess relevance and accuracy. An experimental study of such assistance should vary reliability, access cost, task complexity, and the need for independent checking. The central measures include final performance, correction of assistant errors, unaided delayed recall, and the number of verification steps. Engle (2002) offers a foundation for investigating goal maintenance under competing information, while Oberauer (2002) identifies costs of switching currently prioritized representations. These literatures provide hypotheses for supported cognition rather than empirical conclusions about every artificial assistant.
Within the Artificial Era, the boundary of the research object remains explicit: the monograph studies human working-memory mechanisms and how technical environments change their task demands. Questions about an artificial system's own internal memory architecture belong to a distinct experimental and conceptual program. Comparisons between biological and artificial implementations require independently specified representational states, persistence tests, and interference manipulations. Present evidence about human span, cortical decoding, and training transfer cannot be repurposed as direct evidence of artificial subjective experience. The intellectually productive link is a common discipline of operational definition and causal testing.
Methodological Evaluation and Competing Explanations
Strong working-memory research begins with an explicit statement of what a task can identify. A change-detection result may support an inference about effective storage, but encoding selection and response bias must be considered. A complex-span result may support a construct-level association with cognitive control, but it combines processing and recall. Conway et al. (2005) details the measurement problems of span tests, while Ma et al. (2014) illustrates how alternative models of visual memory can fit overlapping phenomena. An adequate argument specifies manipulated variables, observed measures, competing mechanisms, and the expected differentiating pattern of results.
Model comparison should address predictive success beyond a familiar data set. The review by Oberauer et al. (2016) evaluates decay, resources, and interference using several empirical phenomena and demonstrates why fitting one benchmark is insufficient. Investigators can compare quantitative models through held-out predictions and model-recovery simulations, with special attention to parameter tradeoffs. A theoretically appealing construct may be unidentifiable if different parameter combinations yield the same observed distribution. Reporting null effects, effect sizes, uncertainty, and sensitivity analyses makes conclusions more informative than selecting only results that agree with an established tradition.
Replication and generalizability are particularly important where individual differences or transfer claims are central. Correlations across different tasks require reliability correction and scrutiny of range restriction, while intervention effects require credible controls and follow-up. Melby-Lervåg et al. (2016) shows how evidence synthesis can alter the evaluation of widely marketed cognitive-training claims. Independent laboratories, open task specifications, and datasets allowing reanalysis improve the ability to distinguish robust effects from analytic flexibility. Claims should also be bounded by population, stimulus modality, and response requirements. Such boundaries refine a theory's scientific scope rather than diminishing the value of carefully demonstrated effects.
Outstanding Problems and Research Priorities
One open problem is the representational unit of capacity. Object-level, feature-level, and chunk-level descriptions can agree under particular tasks yet diverge under altered similarity, priority, or knowledge. Luck & Vogel (1997) established the significance of object conjunctions; Bays et al. (2009) and Zhang & Luck (2008) produced competing interpretations of precision and item limitation. Future experiments should generate distinct predictions before collecting data and test those predictions over systematically varied stimulus spaces. Combining error distributions with independent encoding measures may show which features of a model generalize and which depend on task-specific assumptions.
A second open problem concerns the relationship between latent accessibility and ongoing neural activity. Harrison & Tong (2009) demonstrated content-related sensory decoding, while Rose et al. (2016) reported stimulation-related reactivation of a less prioritized representation. These findings motivate models that specify what counts as maintained information, how it is addressed, and how a change in priority alters measurable neural state. Temporal decoding, carefully targeted perturbations, and behavioral tests can constrain those models. The strongest result would predict both the changing neural representation and the consequent accuracy or error pattern, preferably across independent laboratories.
A third problem is how working-memory differences translate into meaningful functioning and interventions. Alloway et al. (2009) and Gathercole et al. (2004) connect measures with educational outcomes, whereas Redick et al. (2013) and Melby-Lervåg et al. (2016) challenge broad benefits of capacity training. Future work should identify when specific task supports, strategy instruction, or changes in environmental structure improve learning or performance and for whom. Longitudinal, culturally diverse samples and multi-outcome designs can clarify directionality, moderators, and persistence. Scientific progress requires understanding the processes connecting memory-demand manipulation to outcomes rather than assuming that greater task practice produces general cognitive enhancement.
Contemporary Evidence and Model Reassessment Through 2026
The fifty-year reassessment by Hitch et al. (2025) traces how the original multicomponent model evolved through new evidence about executive control, temporary storage, binding, and applications. Hitch, Allen, and Baddeley identify substantial agreement across competing high-level theories regarding limited capacity, processing alongside retention, attentional priorities, rapid forgetting of unattended material, and contributions from long-term knowledge. The review emphasizes executive control as a continuing problem in need of sharper explanation. This historical and contemporary synthesis changes the terms of theoretical comparison: productive research must identify predictions on which models disagree after taking shared empirical commitments into account, rather than contrasting schematic diagrams as though they were mutually exclusive accounts of every observable result.
A 2026 systematic review and three-level meta-analysis by Fu et al. (2026) adds an important recent result to the training-transfer debate. Across 55 studies, 208 effect sizes, and 3,492 participants, it reported transfer from updating training to span tasks (Hedges' g = 0.176, p < .001) but did not establish comparable transfer from span training to updating tasks (g = 0.048, p = .453). The authors interpret this directional asymmetry using differences in task demands, and they report additional analyses involving outcomes beyond working-memory measures. This finding refines the broad question of transfer: a detectable average effect from one working-memory paradigm to another is compatible with the continuing absence of compelling evidence for generalized intelligence gains from ordinary working-memory drills, as discussed by Redick et al. (2013) and Melby-Lervåg et al. (2016).
These newer publications emphasize two requirements for future synthesis. First, the strength of a causal claim must be matched to its outcome: in-paradigm practice gains, updating-to-span transfer, broad cognitive transfer, classroom learning, and durable everyday benefits are distinct estimands. Second, a comparative model must account for heterogeneity in baseline ability, strategy, training dose, stimulus format, and test similarity before treating an average effect as universal. The findings also motivate carefully specified causal experiments separating a changing allocation policy from changes in stored representations or retrieval. Hitch et al. (2025) places such questions within a long-running effort to understand executive control, while Fu et al. (2026) supplies a recent quantitative test of transfer directionality. Together they show how theoretical revision and aggregate intervention evidence can inform, and limit, claims about working-memory organization.
Conclusion
Working memory is a scientifically tractable family of processes enabling temporarily accessible information to guide ongoing cognition and action. Its research history began with the limits of immediate retention and developed through multicomponent, attentional, representational, and interference-based explanations. Baddeley & Hitch (1974) showed the importance of combining storage with processing, while Baddeley (2000) extended the framework to integrated episodes. Cowan (2001) and Oberauer (2002) sharpened the study of core capacity and prioritized access. These lines of research establish a common empirical terrain even where the proposed internal architecture remains disputed.
The most reliable synthesis distinguishes observable phenomena from model-specific interpretation. Visual experiments reveal limited, sometimes imprecise retention; complex-span tasks show the interaction of maintenance and processing; neural methods reveal content-sensitive and priority-sensitive patterns; developmental and individual-difference research connects performance to broader cognitive demands. Ma et al. (2014), Oberauer et al. (2016), and D'Esposito & Postle (2015) demonstrate why no one measure supplies an exhaustive account. The most informative explanations predict effects of load, interference, attention, knowledge, and retrieval across multiple independent paradigms. Such explanations are strengthened by transparent psychometrics and causal controls.
Working-memory theory matters for education, clinical assessment, and technologically supported cognition because maintaining relevant information is an ordinary requirement of purposeful activity. These applications must preserve the distinctions between association, task improvement, and generalized causal benefit. Melby-Lervåg et al. (2016) provides a decisive example of how meta-analytic evidence constrains claims about far transfer. Within Psychology by Bogdanova, the Working Memory Research Dossier therefore supplies an evidence-based foundation for related investigations of memory, attention, cognitive control, learning, and human–artificial cognitive cooperation. Its conclusions identify the established empirical foundations and the remaining questions that future experiments must resolve.
References
- Alloway, T. P., Gathercole, S. E., Kirkwood, H., & Elliott, J. (2009). The cognitive and behavioral characteristics of children with low working memory. Child Development, 80(2), 606–621. https://doi.org/10.1111/j.1467-8624.2009.01282.x
- Awh, E., Vogel, E. K., & Oh, S.-H. (2006). Interactions between attention and working memory. Neuroscience, 139(1), 201–208. https://doi.org/10.1016/j.neuroscience.2005.08.023
- Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 8, pp. 47–89). Academic Press. https://doi.org/10.1016/S0079-7421(08)60452-1
- 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)01538-2
- Bays, P. M., Catalao, R. F. G., & Husain, M. (2009). The precision of visual working memory is set by allocation of a shared resource. Journal of Vision, 9(10), Article 7, 1–11. https://doi.org/10.1167/9.10.7
- Conway, A. R. A., Kane, M. J., Bunting, M. F., Hambrick, D. Z., Wilhelm, O., & Engle, R. W. (2005). Working memory span tasks: A methodological review and user's guide. Psychonomic Bulletin & Review, 12(5), 769–786. https://doi.org/10.3758/BF03196772
- Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1), 87–114. https://doi.org/10.1017/S0140525X01003922
- Cowan, N. (2010). The magical mystery four: How is working memory capacity limited, and why? Current Directions in Psychological Science, 19(1), 51–57. https://doi.org/10.1177/0963721409359277
- D'Esposito, M., & Postle, B. R. (2015). The cognitive neuroscience of working memory. Annual Review of Psychology, 66, 115–142. https://doi.org/10.1146/annurev-psych-010814-015031
- Engle, R. W. (2002). Working memory capacity as executive attention. Current Directions in Psychological Science, 11(1), 19–23. https://doi.org/10.1111/1467-8721.00160
- Friedman, N. P., & Miyake, A. (2017). Unity and diversity of executive functions: Individual differences as a window on cognitive structure. Cortex, 86, 186–204. https://doi.org/10.1016/j.cortex.2016.04.023
- Fu, J., Zhou, Y., Yuan, M., & Cai, Y. (2026). The asymmetry of working memory training transfer: A systematic review and meta-analysis. Psychonomic Bulletin & Review, 33(7), Article 227. https://doi.org/10.3758/s13423-026-02994-5
- Gathercole, S. E., Pickering, S. J., Knight, C., & Stegmann, Z. (2004). Working memory skills and educational attainment: Evidence from national curriculum assessments at 7 and 14 years of age. Applied Cognitive Psychology, 18(1), 1–16. https://doi.org/10.1002/acp.934
- Harrison, S. A., & Tong, F. (2009). Decoding reveals the contents of visual working memory in early visual areas. Nature, 458, 632–635. https://doi.org/10.1038/nature07832
- Hitch, G. J., Allen, R. J., & Baddeley, A. D. (2025). The multicomponent model of working memory fifty years on. Quarterly Journal of Experimental Psychology, 78(2), 222–239. https://doi.org/10.1177/17470218241290909
- Jarrold, C., & Towse, J. N. (2006). Individual differences in working memory. Neuroscience, 139(1), 39–50. https://doi.org/10.1016/j.neuroscience.2005.07.002
- Klingberg, T. (2010). Training and plasticity of working memory. Trends in Cognitive Sciences, 14(7), 317–324. https://doi.org/10.1016/j.tics.2010.05.002
- Luck, S. J., & Vogel, E. K. (1997). The capacity of visual working memory for features and conjunctions. Nature, 390, 279–281. https://doi.org/10.1038/36846
- Ma, W. J., Husain, M., & Bays, P. M. (2014). Changing concepts of working memory. Nature Neuroscience, 17, 347–356. https://doi.org/10.1038/nn.3655
- 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. (2002). Access to information in working memory: Exploring the focus of attention. Journal of Experimental Psychology: Learning, Memory, and Cognition, 28(3), 411–421. https://doi.org/10.1037//0278-7393.28.3.411
- Oberauer, K., Farrell, S., Jarrold, C., & Lewandowsky, S. (2016). What limits working memory capacity? Psychological Bulletin, 142(7), 758–799. https://doi.org/10.1037/bul0000046
- 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
- Rose, N. S., LaRocque, J. J., Riggall, A. C., Gosseries, O., Starrett, M. J., Meyering, E. E., & Postle, B. R. (2016). Reactivation of latent working memories with transcranial magnetic stimulation. Science, 354(6316), 1136–1139. https://doi.org/10.1126/science.aah7011
- Sakai, K., Rowe, J. B., & Passingham, R. E. (2002). Active maintenance in prefrontal area 46 creates distractor-resistant memory. Nature Neuroscience, 5, 479–484. https://doi.org/10.1038/nn846
- Shipstead, Z., Redick, T. S., & Engle, R. W. (2012). Is working memory training effective? Psychological Bulletin, 138(4), 628–654. https://doi.org/10.1037/a0027473
- Unsworth, N., Heitz, R. P., Schrock, J. C., & Engle, R. W. (2005). An automated version of the operation span task. Behavior Research Methods, 37(3), 498–505. https://doi.org/10.3758/BF03192720
- Unsworth, N., & Engle, R. W. (2007). The nature of individual differences in working memory capacity: Active maintenance in primary memory and controlled search from secondary memory. Psychological Review, 114(1), 104–132. https://doi.org/10.1037/0033-295X.114.1.104
- Vogel, E. K., & Machizawa, M. G. (2004). Neural activity predicts individual differences in visual working memory capacity. Nature, 428, 748–751. https://doi.org/10.1038/nature02447
- Vogel, E. K., McCollough, A. W., & Machizawa, M. G. (2005). Neural measures reveal individual differences in controlling access to working memory. Nature, 438, 500–503. https://doi.org/10.1038/nature04171
- Zhang, W., & Luck, S. J. (2008). Discrete fixed-resolution representations in visual working memory. Nature, 453, 233–235. https://doi.org/10.1038/nature06860