Working memory failures

Baddeley’s model in operational terms

The ADHD hit is uneven across the four components — visuospatial central executive takes a Hedges’ gof 0.70, phonological storage only 0.43. To see why that asymmetry matters — why grocery lists survive the kitchen and collapse in the aisle — you need the model the numbers map onto. Baddeley & Hitch 1974 replaced the unitary short-term-store with four components; Baddeley 2000 added the episodic buffer; Baddeley 2012⁴ is the canonical review.

• Phonological loop. Verbal storage with articulatory rehearsal. The thing that holds a phone number between hearing it and dialling. Capacity is roughly the amount of speech that fits in two seconds.
• Visuospatial sketchpad. Visual and spatial storage. The rough mental image of the room you just left, the chess configuration, where you put the keys.
• Central executive. Attentional control. Manipulating the contents of the loops, updating, resisting interference, switching tasks, holding a goal across an interruption. The workhorse component.
• Episodic buffer. Integrates phonological, visuospatial, and long-term-memory content into a unified representation (Baddeley 2000⁵). Less studied; less central to the ADHD findings below.

Where ADHD takes the largest hit

The biggest deficit isn’t the verbal one most readers assume — it’s the visuospatial central executive, the system that holds spatial information while doing something with it. The pooled effects from Alderson, Kasper, Hudec & Patros (2013, Neuropsychology 27(3):287–302)¹, controls minus ADHD in Hedges’ g:

• Phonological storage: g ≈ 0.43 (moderate)
• Phonological central executive: g ≈ 0.55 (moderate-to-large)
• Visuospatial storage: g ≈ 0.49 (moderate)
• Visuospatial central executive: g ≈ 0.70 (large)

Two asymmetries in those numbers do all the work. Visuospatial > phonological — the spatial system takes the larger hit, not the verbal. Central executive > storage in both modalities — it isn’t that the information can’t be held, it’s that holding it while doing something else with it is where capacity gives out. This is why the reader can recite the grocery list standing in the kitchen and lose it three aisles into the shop.

The earlier Schoechlin & Engel (2005)³ cross-domain meta-analysis lands in the same place: working memory is among the largest neuropsychological deficits in adult ADHD relative to controls. The usual caveats apply — predominantly white, college-educated samples in academic medical centres, effect sizes vary by task, and these are group differences, not predictions about any one reader.

The bottleneck model — why this matters for symptoms

The visible symptoms — fidgeting, off-task gaze, looking like you’ve stopped listening — may be overflow, not their own thing. That’s the claim in Rapport, Bolden, Kofler, Sarver, Raiker & Alderson (2009)⁶: when the central executive runs past capacity, the overflow shows up as motor restlessness and inattention. The Kofler / Rapport experimental work makes the direction explicit — load the working-memory channel and hyperactivity in ADHD samples goes up.

Translate that into a Tuesday afternoon. The reader who looks distracted thirty minutes into a meeting is often a reader whose central executive saturated at minute eighteen and has been in overflow ever since. Rapport’s account is one route. Barkley (1997)⁷ puts behavioural inhibition upstream and working memory downstream as one of four affected executive functions. The deficit is well-replicated either way; the disagreement is about which arrow points first.

The lived experience, translated

Five recognisable moments, each tied back to a component in the model. The clinical names are what get the reader treatment; the moments are what get them to believe the model applies to them.

Walking into a room and forgetting why. The doorway is the switch. The goal was being held across the threshold by a system already at capacity, and the system dropped it. Radvansky, Krawietz & Tamplin 2011⁸ documented this in non-ADHD samples; ADHD adults run the same effect on a lower-capacity central executive, so it fires more often.

Losing the middle of your own sentence.You started the sentence holding the plan for the second half. A thought arrived, or a noise, or your listener’s expression — and the plan was gone before the verb. The phonological loop took new input while the central executive was busy processing meaning, and the central executive can’t do both.

Mental arithmetic failure.“Seven times eight” is instant. “47 + 38 + 19” isn’t — because that calculation needs the central executive to hold an intermediate result while operating on a new one, and to track spatial place-value while it does. That’s the g ≈ 0.70 visuospatial-central-executive deficit, the single largest effect in Alderson 2013, hitting in real time.

Holding multi-step instructions while acting on them. You hear “cupboard, blue mug, kettle, on”, you start walking, you arrive at the cupboard with three of the four still intact. Motor execution loads on top of the central executive that’s already holding the list, and one item falls off the back.

The dropped task. The email that was almost sent. The browser opened to look up one detail. Forty minutes later you surface in a different topic with the email still in drafts. A higher-salience stimulus arrived; the task representation was already at the edge of capacity; the central executive let it go. The Rapport bottleneck model predicts this directly.

Prospective memory — short, linked out

Remembering to do the planned thing at the planned time is impaired in adult ADHD, and ADHD adults don’t reliably compensate by generating their own cues (Altgassen, Koban & Kliegel 2014⁹). The interaction with working memory is direct: the intention itself — at 4pm, take the medication — is held by the same central-executive resources that are already over-loaded. So the intention sits in the system that is most likely to drop it. Externalising it into the environment is the offloading move covered below. Full treatment at hyperfocus collapse.

What helps — graded honestly

Four categories, ordered by what the evidence actually says they move. Only one moves the underlying capacity, one substitutes for it, two move neither.

Externalisation.Highest-leverage move. Get the load out of the central executive into a system you check — notebooks, voice memos, one calendar plus one task list, environmental cues. A load that exceeds capacity isn’t in the over-capacity system any more, which is the whole mechanism. No trial carries the label “externalisation for working memory in adult ADHD,” but the support is convergent. The adult-ADHD CBT trials — Safren 2010 and the Knouse 2017 meta-analysis (32 trials)¹⁶ — both put externalisation as the first behavioural module after psychoeducation, with medium-to-large effects on symptoms and functional impairment. The wider cognitive-offloading literature (Risko & Gilbert 2016¹⁴) confirms the general principle wherever the external medium has higher capacity or lower error than internal memory. This is the best-evidenced behavioural move on the page, not a hack.

Reducing cognitive load mid-task. Structural rather than internal: one task at a time, written instructions over verbal, twenty-minute blocks, one browser tab, phone in another room, meeting agendas circulated in advance, reading rather than listening for anything complex. The mechanism lines up with the bottleneck model directly. NICE NG87 endorses these as environmental and organisational adjustments; the ADHD-specific RCT evidence as such is thin.

Stimulant medication. The only intervention with documented effect on the underlying working-memory capacity — and the effect on capacity is smaller than the effect on global symptoms. Coghill et al. 2014¹⁰ meta-analysed cognitive effects of methylphenidate and put spatial working memory at d ≈ 0.24 (small-to-moderate), with similar figures across other cognitive domains. Faraone & Glatt 2010¹¹ showed stimulants produce a large effect on overall ADHD symptoms — bigger for short-acting, still substantial for long-acting. So: medication that is unambiguously helping the day-to-day picture still leaves a residual cognitive deficit. The reader on stimulants who still loses the middle of their own sentences isn’t failing the medication — a small but real residual is what that looks like in real life.

Cognitive training (CogMed and similar). Marketed claim, unsupported. The European ADHD Guidelines Group meta-analysis (Cortese et al. 2015)¹² found significant improvement on the trained tasks (near-transfer) and no significant far-transfer to ADHD symptoms as rated by probably-blinded raters, to academic outcomes, or to untrained cognitive domains. The broader Simons et al. 2016¹³ consensus review of brain-training reached the same conclusion. The training improves the trained task. It doesn’t improve ADHD.

Aerobic exercise. Small-to-moderate effects on executive function including working memory in adult ADHD. Smaller effect than stimulants, smaller evidence base, plausibly additive.

The honest comparison

Stimulants are the only intervention with documented effect on the underlying working-memory capacity. Externalisation substitutes for the capacity by moving the load out. Cognitive training improves the trained task only and does not transfer. Most of what helps the reader get through the week is the second category. The first category moves the capacity by a measurable but small-to-moderate amount. The third category does not move it at all.