Executive function training: what children keep.

Every classroom task has a hidden second task. To follow a story, a child must hold the last sentence in mind while decoding the next one. To finish a sum, she must ignore the noise beside her. To switch from adding to subtracting, she must drop one rule and load another. Psychologists call these skills executive functions: working memory, inhibitory control, and cognitive flexibility. In plain terms — holding things in mind, stopping yourself, and switching tracks.

Schools care about these skills for a simple reason: they track achievement. Across 67 studies, children’s executive function scores correlate with reading at roughly r = .30 and with maths at roughly .31 — both at a single point in time and years ahead (Jacob & Parkinson, 2015). A correlation of .30 is large by the standards of child development. Children who start school low on these skills tend to stay behind, and teachers can usually name them within a fortnight.

From that fact grew a seductive idea. If executive function predicts achievement, then training executive function should raise achievement. Curricula were built on it. Apps were sold on it. Ministries funded it. This article follows that idea through the randomized evidence — the trials where children were assigned by lottery to train or not. The story it tells is not a simple debunking. Something real happens when children train these skills. It is just much narrower than the sales pitch.

The promise

The scientific case peaked in 2011, when Adele Diamond and Kathleen Lee reviewed the interventions shown to aid executive function development in children aged 4 to 12 (Diamond & Lee, 2011). Published in Science, the review found encouraging signals almost everywhere it looked: computerized training, certain games, aerobic exercise, martial arts, mindfulness practices, and whole classroom curricula.

Two of its conclusions shaped everything that followed. First, executive functions need continual challenge — children improve when the demand keeps rising, not when they repeat what is easy (Diamond & Lee, 2011). Second, the children with the weakest executive function gain the most from any programme (Diamond & Lee, 2011). That second point made the field a matter of equity, not enrichment. If disadvantage shows up partly as depleted self-regulation, then training it looked like a lever for closing gaps.

The review was honest about the thinness of the trial base. But the market did not wait. Within a few years, executive function had become a product category — and one curriculum became its flagship.

Tools of the Mind on trial

Tools of the Mind is a preschool and kindergarten curriculum built on the ideas of the Soviet psychologist Lev Vygotsky. Children write “play plans” before entering extended pretend play. They regulate each other in paired tasks — one child reads, the other holds a drawing of an ear and checks the listening. Self-regulation is not a lesson; it is the design of the whole day. As a theory of embedding executive demands in real activity, it is exactly what the science recommends. So its trial record matters.

The first randomized trial, in a low-income New Jersey district, assigned 274 preschoolers to Tools or to the district’s balanced curriculum (Barnett et al., 2008). Classroom quality rose and problem behaviour fell in Tools classrooms — clear, measured wins. But on academic and most cognitive outcomes, the gains were small and mostly not statistically reliable (Barnett et al., 2008).

The strongest result came six years later. Clancy Blair and Cybele Raver ran a cluster randomized trial — schools, not children, were the units assigned — across 29 schools, 79 kindergarten classrooms and 759 children (Blair & Raver, 2014). Tools children improved on working memory, attention control, and reasoning, and on vocabulary, reading and maths. The maths effect in the full sample was modest, roughly 0.13 standard deviations. But in high-poverty schools the effects grew several-fold — on the order of 0.4 to 0.5 standard deviations for vocabulary and reasoning — and several gains persisted into first grade (Blair & Raver, 2014). Here was the equity promise, apparently delivered.

The Tennessee null

Then came the largest and longest test. Dale Farran’s team at Vanderbilt ran a field trial across roughly 60 pre-K classrooms — 32 assigned to Tools of the Mind — in Tennessee and North Carolina, following children from pre-K into first grade (Nesbitt & Farran, 2021). The result, published as a full research monograph, was flat. Tools children gained no more than control children on academics, on executive function, or on self-regulation. And how faithfully teachers implemented the curriculum showed no consistent relationship with outcomes (Nesbitt & Farran, 2021).

That last detail stings the most. When a programme fails, believers usually reach for the implementation excuse — it was not done properly. Here, doing it properly did not help. The honest summary of the flagship’s record is now: one early trial that moved behaviour but not learning, one strong result concentrated in high-poverty schools, and one careful, well-powered null (Nesbitt & Farran, 2021). Same curriculum. Different places, ages and comparison classrooms. The spread itself is the finding — whatever Tools does, it is not a reliable, portable effect that a ministry can order at scale.

The near-transfer problem

To read the wider literature, one pair of terms is needed. Near transfer means improvement on tasks very like the ones trained — practise holding number strings in mind, get better at holding number strings in mind. Far transfer means improvement on something genuinely different: another executive skill, reading, maths, behaviour. Every parent and minister who funds this training is paying for far transfer. Near transfer is what the money usually buys.

The cleanest accounting comes from a meta-analysis — a study pooling many trials into one estimate — of executive function training in typically developing children (Kassai et al., 2019). Across 43 controlled comparisons, training produced a solid near-transfer effect of g ≈ 0.44 on the component trained. Across the 17 comparisons that measured far transfer to untrained components, the pooled effect was g ≈ 0.11 — small, and not statistically distinguishable from zero (Kassai et al., 2019).

Think about what that pattern means in a school. A child spends twenty minutes a day on a working-memory game. Her scores on that game climb steadily — real learning, honestly measured. Meanwhile her inhibition, her flexibility, her reading and her maths sit where they were. The benefit stops at the boundary of the practised task. The authors’ conclusion was blunt: the absence of generalized benefits questions the practical relevance of training executive function components in isolation (Kassai et al., 2019).

Working memory training, the sharpest test

If any single skill should reward direct training, it is working memory. It is the most studied executive function, the best predictor of classroom learning, and the target of the most polished commercial programmes. Computerized packages drill children daily on remembering longer and longer sequences, with the difficulty adapting as they improve. The theory is continual challenge, exactly as prescribed.

Monica Melby-Lervåg and Charles Hulme pooled 23 randomized and controlled studies of such training in children and adults (Melby-Lervåg & Hulme, 2013). Immediately after training, the gains looked impressive: roughly 0.79 standard deviations on verbal working memory tasks and roughly 0.52 on visuospatial ones. Then the picture dimmed. The verbal gains were not sustained at follow-up months later. And on everything a school actually cares about — nonverbal reasoning, verbal ability, word decoding, arithmetic — there was no convincing evidence of improvement against treated control groups (Melby-Lervåg & Hulme, 2013).

Their verdict has become the standard citation of the field: these programmes produce short-term, specific training effects that do not generalize (Melby-Lervåg & Hulme, 2013). The same pattern later sank the adult brain-training industry — a story we tell separately in our review of the commercial brain-training evidence. The child version of the lesson is stricter, because the buyer is a school and the currency is instructional time. Twenty minutes a day of memory drills is twenty minutes not spent reading — and reading, unlike an abstract span task, carries its gains with it.

The correlation that started it all

Why did so many smart people expect more? Because the correlation is real and keeps replicating. Jacob and Parkinson’s review of 67 studies found executive function and achievement travelling together at roughly r = .30, in the same year and across years (Jacob & Parkinson, 2015). The mistake was the next step — reading that association as a causal lever. Their review looked hard for evidence that changing executive function changes achievement, and concluded there was no compelling demonstration of it (Jacob & Parkinson, 2015). Once background factors and IQ are accounted for, the association itself also shrinks substantially.

The deepest review of the intervention literature reaches a compatible verdict from the other side. Diamond and Ling examined the full menu of approaches and sorted the justified conclusions from the hype (Diamond & Ling, 2016). Narrow computerized training came out worst: reliable gains on the practised task, little beyond it. The most promising signals came from whole-day approaches — curricula, and activities that combine cognitive challenge with physical, social and emotional engagement (Diamond & Ling, 2016). Notably, plain aerobic exercise with no cognitive demand showed weaker executive-function benefits than the folk wisdom expects (Diamond & Ling, 2016). The pattern is consistent: context-rich challenge, not isolated drill.

What the evidence doesn’t show

The trial record supports narrow claims and forbids broad ones. Six limits belong in any honest briefing.

• Trained gains are real but narrow. Children genuinely improve on trained components — g ≈ 0.44 — so studies showing “executive function can be trained” are correct as far as they go. The far-transfer pool is what fails (Kassai et al., 2019).
• No causal chain to achievement. The correlation with reading and maths is solid; a demonstration that raising executive function raises achievement is still missing (Jacob & Parkinson, 2015).
• Durability is mostly unmeasured. Where follow-ups exist, trained gains often fade — verbal working memory gains were not sustained months later (Melby-Lervåg & Hulme, 2013).
• The subgroup promise needs replication. The high-poverty amplification is the most hopeful result in the record — and it comes chiefly from one trial (Blair & Raver, 2014).
• Fidelity did not rescue the null. In the largest pre-K trial, how well teachers implemented the curriculum showed no consistent link to outcomes — the implementation excuse is unavailable (Nesbitt & Farran, 2021).
• Publication bias is plausible. This is a market-adjacent literature with small early studies and enthusiastic framing; the deepest review devotes itself to separating justified conclusions from hype (Diamond & Ling, 2016).

Where the evidence stops

1. 1Trained gains are real but narrow
2. 2No causal chain to achievement
3. 3Durability is mostly unmeasured
4. 4The subgroup promise needs replication
5. 5Fidelity did not rescue the null
6. 6Publication bias is plausible

© 2026 FUTURE PROOF™

Executive function, by the evidence

Read together, the trials write a short manual for schools, parents and ministries — and it is not “do nothing”.

Do not buy stand-alone drills with instructional time. Isolated component training buys near transfer and little else (Kassai et al., 2019). A memory app that shows climbing in-app scores is showing you near transfer, which was never in doubt.

Embed the challenge in real content. The promising signals come from approaches that raise executive demands inside ordinary activity — challenging play, paired work, and lessons that make children hold, plan and switch while learning something worth knowing (Diamond & Ling, 2016). Ask any vendor: what curriculum content carries your training?

Aim programmes where the gains concentrate. The children with the weakest starting skills, and the highest-poverty schools, show the largest measured benefits (Diamond & Lee, 2011). In the strongest trial, effects several times the average appeared precisely there (Blair & Raver, 2014). Universal rollouts dilute what targeted ones deliver.

Use executive function as a signal, not a curriculum. The .30 correlation makes these skills an excellent early-warning indicator for children who will struggle (Jacob & Parkinson, 2015). Screen with them, then respond with instruction in the actual subject — which is where the achievement evidence lives.

Judge any programme on pre-agreed academic and behaviour outcomes. The flagship’s record swung from positive to null across settings (Nesbitt & Farran, 2021). A school piloting an executive-function programme should name its outcome measures before term starts, and keep the ones the vendor did not choose.