← all systems · see it in The System
Working memory
Working memory is a cognitive system with a limited capacity that can hold information temporarily. It is important for reasoning and the guidance of decision-making and behavior.
Properties published scienceour proposal
Sourced fields, and our own readings.
| born at | person: infancy; kind: not dated |
| born precision | kind undated |
| boundary | Inside: the few items or chunks held in a readily accessible form over seconds, the stores or activated representations that hold them (phonological loop, visuospatial sketchpad, the activated part of long-term memory, the focus of attention), the processes that keep and use them (rehearsal, attentional refreshing, updating, binding in the episodic or 'awareness' buffer), and the neural states that carry them (delay activity, short-term synaptic traces, dynamic population codes). Outside: the senses and sensory memory that deliver items; long-term memory, which supplies chunks and receives what is learned; attention in general; the behaviour, speech and eye movements it guides. |
| nested level | A part of cognition (its attention-and-working-memory floor) and of memory (the working store), carried by networks of prefrontal, parietal and sensory cortex with the basal ganglia and thalamus; in the human brain it sits under the thoughts floor. |
| scale | organism |
Measures (33) published science
| central capacity, young adults | 3 to 5 chunks |
| capacity across vision and hearing, sensory memory masked | about 4 items |
| capacity with rehearsal blocked | about 3 units (chunks) |
| digit span, most adults | about 7 (trained: up to 80) digits |
| silent rehearsal span | about 2 seconds of speech |
| decay limit of activated long-term memory without refreshing | up to about 30 s |
| short-term synaptic plasticity window in PFC | about 2 s |
| serial scan of a held list | 20-30 ms per item |
| brain share of body energy; action potentials and glutamate shares of it | about 20%; 47% and 34% percent |
| neurons active at once | less than 1% percent |
| capacity at 7 years vs older children and adults | about 1.5 vs about 3.0 items (k) |
| sentences accessed, grade 1 vs grade 6 | about 2.5 vs 4 sentences |
| refreshing appears | after about 4 years of age |
| growth across childhood | approximately linear 4-14 years, levels off 14-15 |
| adult course (Betula, cross-sectional) | stable 20-50; linear decline from 55-60 to 75-80 years |
| crow capacity (change detection) | K = 3.05 ± 0.038, peaking at 4 items |
| relational reasoning limit | chance at 5 interacting variables (about 4 in parallel) |
| most efficient search group size (simulation) | about 3.5 items |
| precision against set size | a power law relation |
| heritability of capacity | about half of the variation fraction |
| children with working-memory deficits (screening) | one in ten fraction of mainstream pupils |
| training meta-analysis size | 87 publications, 145 comparisons count |
| far transfer by study size (treated controls) | g = 0.01 (k = 34 adequately sized) vs g = 0.26 (k = 25 smaller) Hedges' g |
| prediction from age 5 | working memory at 5 the best predictor of literacy and numeracy 6 years later |
| an LLM's n-back capacity | about 3, like people n (n-back) |
| CDA capacity range across individuals | 1.5 to about 5 objects |
| contralateral delay activity against set size | rises, then approaches a limit at each individual's capacity relation |
| rat tactile delay | 2 s |
| human S1 TMS window for touch | 300 or 600 ms into the delay |
| school training trial size | 580 children consented (of 599) |
| academic-track entry after training | +16 percentage points, three years later |
| n-back training to Gf | small but significant, 20 studies effect |
| MTL conjunction delay | impaired at 8 s; objects and locations normal |
Open contentions (13) open contention
Where sources or readings disagree. Both stay on the map.
- C-413 Where do thoughts come from?
booked: No theory is proven, so the map lists the main ones side by side and doesn't pick one. It shows where thinking is handled in the brain, not where thoughts start.
the other reading: the positions listed under C-413 (thought origin) - C-1105 How many things can we hold in mind at once?
booked: The map uses 3-5 chunks, and keeps the classic 'about seven' alongside.
the other reading: about seven (Miller 1956) - C-3570 Is working memory a separate store, or the active part of long-term memory?
booked: The map keeps the competing readings; none is ruled.
the other reading: Cowan: working memory is the activated subset of long-term memory, not a separate system (src:wp:working-memory); Baddeley, Hitch and Allen 2018 question replacing the short-term store by activated long-term memory (abstract); both models tested together in an adversarial collaboration (WoMAAC; src:logie-2023) - C-3571 Is working memory a few fixed slots or a shared resource?
booked: The map keeps the competing readings; none is ruled.
the other reading: slots: all-or-none items, about 3-4 (Cowan 2010; Buschman 2011 per hemifield); resource: precision shared flexibly, a power law with set size (Ma, Husain and Bays 2014); imaging signals saturate rather than plateau (Bays 2018); hybrid: discrete hemifield resources each shared among objects (Buschman 2011); slots plus averaging: three cups and a bottle of juice; fixed resolution over a narrow range (Zhang and Luck 2008, read in sections); variable precision: continuous and variable resource (van den Berg et al. 2012, abstract); neural marker: the CDA plateaus at each person's capacity (Vogel and Machizawa 2004, abstract), challenged as saturation (Bays 2018) - C-3728 Did a late boost to working memory make the modern human mind?
booked: The map keeps the competing readings; none is ruled.
the other reading: a recent enhancement of working memory was the final piece of modern cognition (Wynn and Coolidge); the usual emphasis on language and symbolism - C-3801 Why can people hold only three to five things in mind at once?
booked: The map keeps the competing readings; none is ruled.
the other reading: weakness: biological expense, interference, too few neurons at a locale; strength: search most efficient near 3.5 items; small capacity helps learn rules like grammar; the number of gamma cycles per theta cycle (Lisman and Jensen); the theta-gamma account is written against 7±2 in Lisman and Jensen's review, so the same mechanism is fitted to 7 and to 4 (src:lisman-jensen-2013) - C-3802 What exactly is working memory?
booked: The map keeps the competing readings; none is ruled.
the other reading: multicomponent (Definition 3); storage plus processing, complex span (Definition 5); generic: components holding information in heightened availability (Definition 6); Olton's recent-event memory in rat mazes (Definition 4) - C-3804 Is there a single 'executive' controlling working memory?
booked: The map keeps the competing readings; none is ruled.
the other reading: working-memory capacity reflects executive attention (the encyclopedia's account of Engle and colleagues); executive control did not form a coherent latent factor, and its link to capacity and fluid intelligence is questioned (Rey-Mermet et al. 2019, abstract); unity and diversity: updating, shifting and inhibition correlated but separable; operation span tied to updating (Miyake et al. 2000, abstract); bifactor: no inhibition-specific factor, common EF absorbs inhibition (Friedman and Miyake 2017, sections); so Rey-Mermet's null bears mostly on inhibition (specialist's reading) - C-3805 Do things slip from working memory because they fade or because other things push them out?
booked: The map keeps the competing readings; none is ruled.
the other reading: decay unless refreshed: cognitive load sets forgetting (time-based resource sharing); interference between similar contents; delaying recall has little effect (encyclopedia); none explains all the data; interference, no temporal decay (Lewandowsky, Oberauer and Brown 2009, abstract); TBRS in its own paper: traces decay when attention is switched away (Barrouillet et al. 2004, abstract) - C-3806 Does the hippocampus take part in holding things in mind for a few seconds?
booked: The map keeps the competing readings; none is ruled.
the other reading: needed only when material exceeds working-memory capacity, resists rehearsal or attention is diverted (Jeneson and Squire 2011, abstract); needed for conjunctions even over 1-8 s delays (Olson et al. 2006, abstract) - C-3807 Is a held thought kept by neurons firing continuously, or stored silently in their connections?
booked: The map keeps the competing readings; none is ruled.
the other reading: persistent spiking underlies working memory; rhythmic and synaptic-only models are inconsistent with the data (Constantinidis et al. 2018, abstract); delay activity yes, persistent no: sparse bursts and dynamic coding (Lundqvist, Herman and Miller 2018, abstract); activity-silent synaptic storage (Stokes 2015); dynamic population activity in high time-constant neurons (Cavanagh et al. 2018) - C-3808 Does brain training make people smarter in general?
booked: The map keeps the competing readings; none is ruled.
the other reading: training raises fluid intelligence (Jaeggi et al. 2008, via the encyclopedia); a randomized study of 580 six-year-olds reporting transfer to reading, maths and a 16-point higher academic-track rate (via the encyclopedia, not read in the original); Jaeggi et al. 2008 (abstract): transfer to Gf, dose-dependent; Au et al. 2015 (read in sections): small significant n-back effect on Gf across 20 studies; Berger et al. (SSRN 2020, read in sections; JPE 2024 abstract): 580 children, spillover to geometry, Raven's and inhibitory control, +16 points academic track three years later; reading not robust; Simons et al. 2016 (abstract): little evidence for distant transfer - C-3809 Where is working memory held: in the front of the brain, or in the sensory areas that first saw the thing?
booked: The map keeps the competing readings; none is ruled.
the other reading: the PFC holds the content (Goldman-Rakic tradition; Constantinidis et al. 2018); posterior sensory regions hold it and the PFC controls it (D'Esposito and Postle 2015; TMS meta-analysis, Phylactou et al. 2022); sensory cortex is nonessential for storage (Xu 2018, title only: record)
Cross-scale patterns seen here (20) our proposal
Shapes that recur when the same kind of problem is solved at different scales, and how each shows in Working memory. Each links to every other system that shows it.
- nested ladder (structural)
Embedded levels: activated long-term memory (no fixed limit) contains the focus of attention (up to about four) which contains Oberauer's one-chunk focus; an item at each rung is also on the rungs below. - same role, different builds (structural)
Holding about three or four items over seconds is done by the macaque's prefrontal cortex and by the crow's nidopallium caudolaterale, a non-cortical pallial area, 'despite a different architecture of the critical forebrain structures'; and, read with the same n-back tasks, by a transformer language - The carry (structural)
All three faces. Overflow: 'Once this capacity is reached, an additional item can only be maintained if it replaces a previous item' (Hahn et al.), or chunks are 'off-loaded to long-term memory' while the focus takes in more (Cowan 2014). The wait: a held list is 'serially and exhaustively scanned a - The Payment to Entropy (structural)
Two forms. Signal: errors accumulate with the delay ('a gradual accumulation of error resulting from noise in memory') and each item's precision falls as more items share the resource (a power law). Thermal/metabolic: spikes are 'energetically costly', and at large set sizes the extra spikes cost mo - the whole sets the local reference (structural)
Divisive normalization: each item's population activity 'is divided by the grand sum of the activities of neurons in all populations encoding items', so the whole held set fixes the gain each item is read against. Shared by crows and monkeys. - trade one property for another (structural)
Quantity for quality: the resource reading says precision, not the number held, is the currency, and a cued item's precision is bought with the others' ('a corresponding cost to other stimuli in memory'). A second trade: maintenance against concurrent processing (reading span; cognitive load), weake - Umwelt: each system carves its own slice of one shared medium (structural)
Working memory's slice of the shared world: of everything the senses register it keeps 3 to 5 chunks, modality-specific traces only while sensory memory lasts, a central abstract code after; in monkeys each visual hemifield has its own capacity, so the slice is cut twice. Touch adds its own slice: a - conservation law plus closure (partial)
Resource models: a fixed total of resource shared among items is the balance law; it predicts nothing about one item's precision until a second rule closes it (divisive normalization, or allocation by priority). - holding / entrainment (partial)
The theta-gamma code: the slower theta cycle sets when each item's assembly fires, one per gamma subcycle, and the number of subcycles 'may be what sets the capacity'. Partial: shown most clearly in the hippocampus, a hypothesis for working memory. - namespace completeness (the map predicts its gaps) (partial)
The multicomponent taxonomy left gaps it could name: 'Where would spatial information about sound go? Where would touch information go?' These questions 'may have helped motivate the episodic buffer', a component added (2000) to fill the slots the arrangement exposed. Partial: the gaps were named, n - same shape, opposite sign (partial)
One gate, two directions: striatal dopamine gating of prefrontal contents; with less dopamine, updating fails and contents become more rigid and more distractor-resistant, so the same mechanism traded from 'update' to 'maintain'. Too little or too much dopamine both impair (an inverted U). - same toolkit, independent builds (partial)
Crows and monkeys share divisive normalization as the computation behind the limit, built in forebrains of different architecture; Hahn et al. say this 'would further suggest' convergent evolution, so the independent-build reading is the authors' suggestion, not shown. - symmetry breaking (name the cut) (partial)
Before a cue the resource is spread over all items; a cue (even a retrospective one, after the display is gone) names one item and breaks the even spread, setting which item is recalled precisely. - templating (partial)
The same module reused unchanged: divisive normalization, a mammalian-cortex model, also fits crow pallium; the same change-detection task built for monkeys works in crows, and the n-back task in people and in a language model. - who opens the door (partial)
The receiver opens the door ahead of time: goals and cues set before the display decide what enters (attention 'to fill working memory with the items one should be remembering'), and the striatum gates maintain-or-update for the prefrontal contents. Human evidence: prefrontal and globus pallidus act - borrowed-then-held (surface)
Infant tasks provide supports that adult tasks do not, so infants look as if they have adult capacity; refreshing by attention appears only after about age 4. A borrowed phase (task support) and a held phase (own refreshing) are both described, but not as one job moving inward. - causal ceiling (surface)
A capacity of 3 to 5 chunks that holds across materials, ages after childhood, crows and monkeys; its cause is open (biological cost, or an optimum near 3.5 items for search), so it may be a practical limit rather than a structural one. - lightning (surface)
On single trials, delay activity 'often occurs in sparse transient bursts' rather than steady firing; whether each burst is a discharge that resets the state is not shown in the abstract read. - runaway from an unstable balance (surface)
Recurrent excitation keeps pyramidal cells firing through the delay once started, tuned by inhibition; the balance point and the runaway are not shown in the sources read. - seed (surface)
A chunk is a small packet that unpacks into the items it points to (trained spans of 80 digits by hierarchies of chunks); and the contested enhanced-working-memory hypothesis reads a late rise in capacity as the last piece of modern cognition. Neither shown to unpack new combinatorics.
Valence: what it trades, and with whom (15) published science
Each exchange with a neighbour, read from both sides: who gains, who pays, and what carries it.
- Basal ganglia via sys:electrochemical-gradient · Working memory gain
- Human (Homo sapiens) via sys:sensory-environment · Working memory gain
- Language (human language) via sys:acoustic-wave · Working memory even
- Prefrontal cortex via sys:electrochemical-gradient · Working memory gain
- Hippocampus (memory) via sys:electrochemical-gradient · Working memory gain
- Consciousness via sys:mental-representation · Working memory gain
- Education via sys:sensory-environment · Working memory even
- Posterior parietal cortex via sys:electrochemical-gradient · Working memory gain
- Memory via sys:mental-representation · Working memory gain
- Attention via sys:electrochemical-gradient · Working memory gain
- Cognition via sys:mental-representation · Working memory gain
- Sensory memory via sys:sensory-environment · Working memory gain
- Stress (biology) via sys:electrochemical-gradient · Working memory loss
- Intelligence via sys:mental-representation · Working memory even
- Numeracy via sys:mental-representation · Working memory even
Partners and parts published science
Built from / builds (20)
- Whole-number strategy against fraction magnitude for the answer in a fraction comparison
- Thalamus
- Prefrontal cortex
- Medial temporal lobe
- Posterior parietal cortex
- Phonological loop
- Visuospatial sketchpad
- Central executive
- Episodic buffer (renamed the awareness buffer, 2026)
- Focus of attention
- Absolute count against proportion for the choice of a population to draw from
- The proportional (linear) model against area and volume scaling for the answer to an enlargement problem
- The base rate against the streak for the next prediction
- Two tasks competing for the single central stage
- Psychology (the mind and behaviour as psychology describes them)
- Learning a second language (second-language acquisition)
- Competition for working memory during calculation: mathematics anxiety against the problem
- Intelligence
- Information competing for limited processing
- Held items compete for a limited working-memory resource
Sits inside / holds (3)
expressed by (11)
- Recombination/Composition
- Encapsulation & Modularity (Abstraction)
- Internal Coordination/Regulation
- Hypothesis/Goal Selection
- Mapping/Modeling
- Uncertainty Reduction/Learning
- Offline Simulation & Replay-Guided Planning
- Triage & Prioritized Allocation
- Quality Control & Error Correction
- Bridging Supply Gaps (Store & Dormancy)
- Scheduled Build-and-Dismantle (On-Demand Structure)
floor filled by (24)
- Prefrontal cortex
- Recursion: defining or computing a thing in terms of smaller versions of itself, as computer science and mathematics describe it, with its base cases, call stack, recursive data, induction, and its sightings in language and nature
- Consciousness
- Synapse
- Basal ganglia
- Human (Homo sapiens)
- Thalamus
- The world through the senses (sensory environment)
- Hippocampus (memory)
- Visual cortex
- Posterior parietal cortex
- Memory
- Attention
- Sensory memory
- Phonological loop
- Visuospatial sketchpad
- Central executive
- Episodic buffer (renamed the awareness buffer, 2026)
- Focus of attention
- Theta-gamma code
- Somatosensory cortex
- Psychology (the mind and behaviour as psychology describes them)
- Learning a second language (second-language acquisition)
- Numeracy
represents (1)
Kinds (12)
- Memory
- Executive functions
- Verbal working memory
- Visual working memory
- Spatial working memory
- Long-term working memory
- Short-term memory
- Working memory in crows
- Working memory in monkeys
- Working memory in a large language model (analogue)
- Tactile working memory
- Working memory in rodents (Olton's sense and delayed comparison)
wired (7)
- Iconic memory
- Human brain
- Hippocampus (memory)
- Memory
- Global neuronal workspace (conscious access, as GNW theory names it)
- Sensory memory
- Mathematical anxiety
Notes published science
- Crows hold about three coloured squares in mind, the same as monkeys, and their neurons share the load the same way (each item's signal divided by the total), even though a crow's forebrain has no layered cortex at all (src:hahn-2021).
- Fifty-two years after creating their model of working memory, Baddeley and Hitch, with Allen, proposed renaming its fourth part: the 'episodic buffer' becomes the 'awareness buffer', moved to the centre of the model. It is the authors' own proposal in an opinion piece, not yet the field's usage (src:allen-baddeley-hitch-2026).
- Brain-training for working memory: across studies big enough to trust, the effect on anything beyond the trained tasks was g = 0.01; only the smaller studies showed an effect, g = 0.26 (src:melby-lervag-2016).
- A rat can hold the speed of a whisker vibration for two seconds, and while it does, its whisker cortex and its prefrontal cortex hum together in a theta rhythm; in people, one magnetic pulse to the touch cortex in the first half-second of the wait spoils the comparison (src:esmaeili-diamond-2019).
Layers
The ways it takes in its world, where they meet, and how it acts back, each from L1 (where it meets the world) up.
the boardroom (shared floors)
- L4: The capacity-limited focus: 3 to 5 chunks in young adults (about 1.5 at age 7), or a limited resource whose precision falls with set size; frontal and parietal cortex
- L5: Control: the central executive (directing attention, suppressing the irrelevant, coordinating two tasks), attentional refreshing against decay, striatal gating of maintain versus update
- L6: Binding and manipulation: the episodic (awareness) buffer, a one-chunk focus that selects what is worked on, concepts formed from co-present ideas
- L7: Guidance of reasoning, comprehension, planning and behaviour by what is held
way in: Arousal and stress (neuromodulators)
- L1: A stressor, acute or chronic
- L2: Catecholamine release in the prefrontal cortex
- L3: Dopamine and noradrenaline set PFC network firing, too little or too much impairing it
way in: Cues and task goals (what to keep, what matters)
- L1: A cue or instruction: which item will be tested, even after the display is gone
- L2: Priority given: the cued item gains precision at the others' cost
- L3: Priority maps in the frontal eye field and lateral intraparietal area
way in: Hearing (spoken words and digits)
- L1: Spoken digits and words, sometimes in an unattended channel
- L2: Auditory sensory memory: how a sentence sounded
- L3: The phonological loop's store
way in: Long-term memory (knowledge that fills and chunks what is held)
- L1: Long-term knowledge: the vast amount saved in a lifetime
- L2: Retrieval cues and chunks: a chunk held points to what it contains
- L3: Activated long-term memory: the representations currently active
way in: Sight (visual and spatial items)
- L1: Light from a scene; a brief array of coloured squares or locations
- L2: Visual sensory memory and encoding: how a picture looked; encoding not instantaneous
- L3: The visuospatial sketchpad; in the brain, held features in visual cortex
way in: Thought (inner speech and rehearsal)
- L1: Inner speech: silently saying items over
- L2: Subvocal rehearsal in left frontal cortex (Broca's area)
- L3: The phonological store in left posterior parietal cortex
way in: Touch (felt items)
- L1: Objects and pressure on the skin
- L2: Modality-specific traces of how something feels
- L3: Delay activity holding a vibration: S2, prefrontal and premotor populations (monkey); S1 early in the delay (human); whisker S1 with prelimbic cortex (rat)
way out: Guiding attention (top-down bias)
- L2: Processing in posterior sensory areas biased toward the goal
- L3: Priority maps weight what is looked at next
- L5: Goal-driven top-down signals from prefrontal cortex
way out: Into long-term memory (learning)
- L5: Chunks off-loaded to long-term memory while the focus takes in more
- L6: Working memory as the portal to long-term memory
way out: Memory-guided movement (saccades and reaches)
- L1: The eye or hand moves in the world
- L5: The held saccade target compared with what the eye then sees
- L7: A delayed reach or saccade to a remembered place, its cells active until the response is made
way out: Recall and report (span tasks, answers)
- L1: Speech heard by the tester
- L4: Repeat back in order: about seven digits
- L7: Report what is held: the last words of a list of sentences (reading span)
way out: Rehearsal (inner speech back into the loop)
- L2: Inner speech: about 2 seconds of speech kept by silent rehearsal
- L5: Subvocal rehearsal from left frontal cortex
working memory in the brain: from spikes to the networks that hold them (secondary: the neural view)
- L1: Signals carrying the memorandum in sensory cortex (held features decode from V1 without raised overall activity)
- L2: Synapses: glutamate synapses on dendritic spines; short-term synaptic plasticity holds a trace for about 2 s
- L3: Local recurrent circuits: delay-active pyramidal networks with recurrent glutamate excitation, tuned by GABAergic lateral inhibition
- L4: Population codes: dynamic trajectories, activity-silent synaptic states, divisive normalization dividing gain among items
- L5: The network: reentrant loops between frontal and posterior cortex and subcortical structures; the striatum gates maintain-or-update
- L6: Rhythms: theta-gamma coupling orders several items; theta power rises with load
working memory: from what reaches the mind to action guided by what is held (reader's view)
- L1: What reaches the mind: seen arrays, heard words and digits, felt objects, a task and its cues
- L2: Sensory registration and encoding: modality-specific traces of how things look, sound or feel; encoding takes time
- L3: The holding stores: the phonological loop and visuospatial sketchpad (multicomponent model), or the activated part of long-term memory (embedded processes)
- L4: The capacity-limited focus: 3 to 5 chunks in young adults, or a limited resource spread over all items
- L5: Control: the central executive, attentional refreshing against decay, updating and gating of what is held
- L6: Binding and manipulation: the episodic buffer (renamed the awareness buffer, 2026) binds sources into episodes; ideas co-exist so concepts can form
- L7: Guidance: reasoning, comprehension, planning, decisions and behaviour steered by what is held
Sources (27) published science
Carruthers P 2013. Evolution of working memory. PNAS 110 Suppl 2. doi:10.1073/pnas.1301195110 (behind its links)
Gong D, Wan X, Wang D 2024. Working memory capacity of ChatGPT: an empirical study. Proc AAAI 38(9). doi:10.1609/aaai.v38i9.28868 (sections: abstract, results) (behind its links)
Cowan N 2017. The many faces of working memory and short-term storage. Psychon Bull Rev 24:1158. doi:10.3758/s13423-016-1191-6 (sections: the definitions) (behind its links)
Hunter WS 1913. The delayed reaction in animals and children. Behavior Monographs (Biodiversity Heritage Library). doi:10.5962/bhl.title.44534 (sections: contents) (behind its links)
Lisman JE, Jensen O 2013. The theta-gamma neural code. Neuron 77:1002. doi:10.1016/j.neuron.2013.03.007 (sections: abstract, working-memory capacity passages) (behind its links)
Ma WJ, Husain M, Bays PM 2014. Changing concepts of working memory. Nat Neurosci 17:347. doi:10.1038/nn.3655 (author manuscript) (behind its links)
Vogel EK, Machizawa MG 2004. Neural activity predicts individual differences in visual working memory capacity. Nature 428:748. doi:10.1038/nature02447 (behind its links)
Second-language acquisition (Wikipedia, rev 1380191148, 2026-10-14; via Wikimedia Enterprise snapshot) (behind its links)
Fluid and crystallized intelligence (Wikipedia, via Wikimedia Enterprise snapshot 2026-11-01) (behind its links)
Dowker A (2026) Mathematics anxiety: Effects of age, gender and culture. Br J Educ Psychol. doi:10.1111/bjep.70057 (CC BY) (behind its links)