Can executive function be trained, and does it transfer to learning?

Verdict

no-effect, and it is the same verdict as brain training reached by a different road. That topic tested the software; this one tests the curricula — Tools of the Mind, self-regulation programmes, school mindfulness — which are the versions a school is actually offered. The answer converges.

Be precise about what is denied. Executive-function task performance can be trained. Explicit training gives g = 0.38 at post-test on objective neurodevelopmental tests, and that is a real finding produced by randomised trials. What does not exist is any of the three things that would make it a school decision:

1. Durability. The pooled follow-up estimate across all EF interventions is 0.008 after publication-bias correction, with inhibition at 0.04 and cognitive flexibility at −0.10.
2. Transfer. EF-specific curricula reach g = 0.12 at post-test and −0.03 at follow-up. The field's own most influential advocate writes in her systematic review of 179 studies that "transfer of training is narrow… rarely transfer to untrained skills," and that this "has been the nearly universal conclusion of reviews for the past 25 years."
3. Causality of the underlying link. The EF→achievement correlation that motivates the whole enterprise loses more than half its size, and usually all its significance, once background and IQ are controlled.

What the evidence shows

The training literature, sorted by what it measures

The follow-up column is the finding. Almost every number a school is shown is a post-test number. Takacs and Kassai are the only people who pooled the follow-ups, and the corrected estimate is 0.008. That is not a small effect; it is nothing, measured on the field's own preferred instruments.

The active-control column is the mechanism of the illusion. Dunning et al. give the cleanest demonstration available anywhere in this archive: same meta-analysis, same outcome, d = 0.35 against passive controls and d = 0.04 against active ones. The effect is attention, expectancy and time-on-something, not capacity.

Tools of the Mind: the developer-involvement gradient in one programme

This is a textbook case and it is worth stating as a pattern, not just a result. A developer-involved single-site trial reports success on a teacher-rated scale. The federal review finds the flagship Science paper does not meet evidence standards. Two independent trials — one across five districts and two states with a three-year follow-up, one prospectively registered with an active control — find nothing. When the developers were shown the null and intensified their coaching, the second cohort was null too. And fidelity, the field's standard escape hatch, was not associated with outcomes in either independent trial.

Nesbitt and Farran's incidental finding is the one that generalises furthest: teachers varied as much within treatment and control classrooms as they did between conditions. Adopting a curriculum is not the same event as changing what happens in a room.

Mindfulness as EF training: the definitive trial and the mechanism test

MYRIAD deserves its own note. It is the largest school mental-health trial ever run, it was conducted by a team whose leaders direct mindfulness centres and receive royalties on mindfulness books, and it returned effects of 0.005, 0.02 and 0.02 with confidence intervals the authors say rule out important effects. Five secondary outcomes moved and all five favoured the control arm, including higher self-reported hyperactivity and inattention in the trained group. The companion mechanism trial then found that the training did not even raise mindfulness itself.

The construct: 99–100% heritable, with shared environment at zero

Hereditarian-lens assessment

Risk: low for the verdict, which rests on randomised trials — and the twin evidence is not doing the work of the verdict, it is explaining it.

Two independent twin cohorts, one adolescent and one in middle childhood, put the latent Common Executive Function factor at essentially 100% heritable with shared-environment variance indistinguishable from zero. That is an unusually extreme estimate even by this archive's standards, and it is the most direct statement available of why the training literature looks the way it does: near transfer is routine because tasks are learnable; far transfer is absent because the latent capacity is not an environmental product.

The archive's premise (2) applies exactly. Claims of durably raising general capacity require extraordinary evidence, and executive function is capacity in the most literal sense the field has. The correct inference is not that children cannot improve — it is the inference this archive draws everywhere: what transfers is knowledge and skill, not capacity.

Jacob and Parkinson supply the corresponding correction to the observational literature, and it is a worked example of the archive's default. EF correlates with reading at r = .30 and with maths at r = .31, invariantly, across ages and measures — exactly the sort of robust correlation that gets turned into a programme. Control for background characteristics and IQ, and one of thirteen predictive associations survives, with effect sizes falling by more than half. Two heritable traits correlate; the school-facing causal claim does not follow.

Boundaries & what critics say

• The steelman is Pandey 2018, d = 0.42 across 49 randomised trials, published in JAMA Pediatrics. Its weakness is precisely the variable the rest of this file turns on: it does not separate active from passive controls. Read against Dunning's 0.35-versus-0.04 split, 0.42 is what a passive-control pool looks like.
• Blair & Raver 2014 is the best positive Tools trial and should not be waved away. It is large, randomised, uses objective assessments and biomarkers, and shows a first-grade reading effect of 0.14 that grew rather than faded. Two things bound it: the control was business-as-usual against two years of professional development and coaching, and the large effects are subgroup estimates in high-poverty schools. It is unreplicated, with no independent kindergarten trial and no follow-up past grade 1.
• Diamond 2019 reports striking reading results — 33% versus 10% reading at grade-1 level. It also measures no executive function whatsoever, rests on 18 clusters, and comes from the programme's most prominent advocate. If Tools raises reading through better classroom management rather than through executive function, that is a different and more interesting claim, and nobody has tested it.
• Diamond & Ling 2016 is the optimistic review and it is genuinely optimistic — reporting effects around 0.8 for lower-income children on the all-day Tools curriculum. But note two things: those figures come from the advocate's own programme, and even this review states that "people only improve on what they practise" and that benefits "almost always grow smaller as the time since training increases." The 2020 systematic review by the same authors is markedly more sceptical, and this topic weights the later one.
• Scionti 2020 disagrees with Takacs and Kassai about who benefits — at-risk children in one, typically developing children in the other. The disagreement is live and unresolved. What both agree on is that behavioural and learning-related outcomes do not move.
• Clinical populations are a separate question. As in brain training, EF training for ADHD symptom management is a different decision from EF training for school achievement, and this verdict is about the latter.
• This topic does not contradict brain training; it extends it. That topic covered working-memory software; this one covers curricula, mindfulness and self-regulation programmes. Same conclusion, wider scope, and the near-transfer-is-real caveat is preserved in both.

Practical guidance

• Do not replace curriculum with an EF programme. The one trial that ran the swap across five districts with a three-year follow-up found every significant effect favouring the control group.
• Do not buy school mindfulness for attention or behaviour. The definitive trial is null on all three co-primary outcomes and five of its 28 secondaries went the wrong way.
• Apply two tests to any EF claim you are shown. Was the control group active? Is there a follow-up at six months or more? The effect survives neither.
• Distinguish "the child got better at the task" from "the child got better." A Peg Tapping or digit-span improvement is a genuine result about Peg Tapping.
• If the real goal is a calmer classroom, buy classroom management — small, replicated, cheap (see behaviour management) — rather than a curriculum that claims to install self-regulation in the children.
• If the real goal is achievement, teach the subject. The archive's consistent finding is that content-building interventions persist and capacity-building ones do not; see fadeout and persistence.

Open questions

• Whether Blair and Raver's high-poverty kindergarten result is real is genuinely unresolved, and it is the one place where an independent replication would change this verdict.
• If Tools of the Mind raises early reading (Diamond 2019) without raising executive function, what is the mechanism? Nobody has tested the classroom-management explanation.
• No EF trial in this file has a genetically informative design, so differential trainability — the same gap the archive flags in talent and trainability — remains untested here too.
• Why near transfer decays so lawfully with task distance is still unexplained, and it is the only robust phenomenon this literature has produced.