Is Executive Function a Universal Mind or a Schooled-World Skill?

Precise question

Standard executive-function (EF) tasks — the Dimensional Change Card Sort, digit span, Stroop-type inhibition tasks, verbal fluency — are treated in psychology and education as windows onto a domain-general cognitive control capacity that all human beings share, merely expressed to different degrees. Is that treatment defensible, or do these instruments return a score that fuses a real control capacity with a culturally learned disposition to engage with arbitrary, decontextualized, rule-governed material — a disposition that develops inside "schooled worlds" and therefore differs systematically across populations?

A second question follows and is easier to get wrong: if the instruments are partly culture-specific, what follows for practice? Not that EF programs are worthless, that cross-cultural comparison is impossible, or that the capacities are absent in people who score low. It follows that a score must be decomposed before it can be interpreted, and that the burden of proof sits with the person using the instrument.

Why it matters

EF is not a marginal construct. It is the workhorse of developmental psychology, it underpins large literatures on school readiness and self-regulated learning, and it feeds directly into assessment practice for learning disability and attention disorders. It is also the kind of thing that travels. Batteries developed and normed in North America and Europe are now routinely administered where children have little or no exposure to formal schooling, and the results are used to rank populations, allocate educational resources, and evaluate interventions.

The exposure problem is not a footnote. As a 2025 review in Psychological Review put it, "most of what we know about the development of self-regulation comes from research conducted among communities that are unrepresentative of most of the global population" [11]. When the sampling frame is this skewed, a construct can be culturally elaborated in the places that study it and merely assumed elsewhere. That risk is general to science but acute here, because the construct is defined by the tests that measure it. There is no measurement of EF that does not involve asking someone to switch sorting rules, repeat digits backwards, or suppress a prepotent response — that is, to perform exactly the acts that schooling drills.

Best current answer

The evidence supports a decomposition, not a verdict. Standard EF tasks appear to measure a genuine control capacity plus a culturally specific skill for handling arbitrary, decontextualized demands. Both components matter, they are difficult to separate with current designs, and the field has not yet built the instruments that separate them.

Three findings pin this down, and they point in compatible directions.

First, leaving schooled worlds changes performance qualitatively, not just quantitatively. A 2025 study in PNAS compared 283 children from four groups: the UK (n = 76, ages 3–15), rural Bolivia with limited schooling (n = 73), the Kunene region of Namibia/Angola with high-quality schooling (n = 70), and the same Kunene populations with no schooling at all (n = 64) [1]. Every task showed large gaps against the UK reference: on the card sort, the odds of a canonical switching response fell by coefficients of −2.07 (schooled Kunene), −3.76 (Bolivia) and −4.12 (unschooled Kunene), and verbal fluency gaps ran to 13–15 points.

The backward span result is the striking one. Asked to repeat two words in reverse order, after up to four repetitions of the instruction with examples: 100% of UK children and 91% of schooled Kunene children succeeded. 50% of Bolivian children and 22% of unschooled Kunene children could not manage two items at any age [1]. A floor that does not lift with development is not a quantitative difference in capacity. It is a difference in what counts as a legitimate way to answer the question.

Second, this is not only a problem of distant populations. In 2025 a preregistered registered report on nationally representative US data examined whether the Numbers Reversed, DCCS and Flanker tasks measure the same construct in fourth-graders. The analytic sample was 5,711 White, 3,373 Hispanic/Latine, 1,280 Black and 1,047 Asian American children. Configural invariance could not be established between any two racial or ethnic groups [5]. The weakest link was Black students, for whom even the single-factor measurement model failed to fit (RMSEA = 0.144, 90% CI [0.135, 0.152] in the pooled model). Whatever is driving these non-invariances, it operates inside a single national education system with universal schooling. "It is just a cross-cultural problem" is not an available defence.

Third, the field largely does not test. A 2026 systematic review of 14 language- or culture-adapted EF rating-scale measures across 27 studies found 93% reported high internal consistency and 64% retained the original factor structure — but only 30% evaluated measurement invariance [9]. Internal consistency is what a scale manages without leaving its own culture. So most of the adaptation literature cannot distinguish "this instrument works cross-culturally" from "this instrument is internally tidy," because most of it never ran the test that would separate them.

Against this, the invariance record is genuinely not uniform, and that matters for the conclusion.

Where the batteries do hold

Two large multisite studies, both on populations with substantial schooling exposure, found that the latent structure of EF tasks does transfer.

A study of 1,311 children aged around 12 — 453 in mainland China, 371 in Hong Kong, 487 in the UK — established partial scalar invariance for a computerized EF battery before comparing latent means [6]. Latent factor means were substantially lower for British children than for either Chinese group, and self-reported computer use in mainland China was unrelated to EF performance, which weakens the obvious confound that online tasks advantage children with more screen exposure.

A 2025 study of 1,002 preschoolers in England, Hong Kong and mainland China, tested at home rather than in a school testing session, again found partial scalar invariance, no clear site differences, and equivalence of the EF-to-externalizing association across all three sites (standardized paths between −0.37 and −0.35; constraining them equal did not degrade fit) [7]. EF was unrelated to internalizing problems in all three sites (B = 0.04, SE = 0.12, p = .761).

Read carefully, these are the most important results in the brief, because they are the ones a reader is least likely to encounter. They say that the factor structure of these tasks is transportable across culturally distinct, schooling-rich populations — the construct hangs together the same way — and that the relations between EF and behaviour are stable. That is a genuine, replicated result, and it is why "EF tasks are invalid cross-culturally" is too strong.

There is also older, less comfortable evidence. A battery administered to 786 children aged 7–18 in Bangladesh, Ghana and Tanzania reached only configural invariance — equal factor structure, unequal loadings [8]. Non-schooled and schooling-limited populations are exactly where the weaker results cluster.

The 2026 counterattack, from the same dataset

The 2025 paper triggered two commentaries in the same journal issue, both on 14 April 2026, putting the field's disagreement on the record.

Mazzaferro and colleagues argued that the 2025 authors had forced an unnecessary dichotomy [2]. Their case is measurement theory: under classical test theory a good assessment maximizes construct-relevant variance and minimizes error, and when the testing context changes the ratio between the two shifts — so the same score stops meaning the same thing. They drew the analogy to false-belief research, where decades of work establishing that the Sally-Anne test yields mostly familiarity with names and narrative conventions in unfamiliar settings did not lead anyone to conclude that theory of mind is either universal or culturally specific. Both are true, and the resolution is adaptation, not relabelling. Their formulation: "culture-agnostic" EF assessments are "neither possible nor desirable," because "whereas an American child may engage EF to learn long division, a Himba child may engage EF to learn herding."

Niebaum reanalysed the original dataset, using the reading, math and years-of- schooling data the original paper left unused [3]. Age-related gains appeared in all three non-UK samples and for some tasks survived adjustment for years of schooling. And EF scores correlated with math and reading within the schooled samples: 17 of 18 coefficients were positive, and a substantial minority remained significant after controlling for years of school — backward span with math in schooled Kunene children, 0.84 to 0.50; forward span with math in schooled Kunene children, 0.60 to 0.59; backward span with reading in Bolivian children, 0.74 to 0.41.

Two honest qualifications on that reanalysis. Sign-preservation across 18 coefficients is a weak criterion: with samples of 64–73 children, most non-significant coefficients are consistent with a small positive effect. And a control task in the same battery — Luria's imitation trial, the same motor routine without the inhibition demand — showed no association with either academic outcome in either group, which is what a genuinely task-specific predictor should look like.

The authors' reply [4] accepted the substance and rejected the remedy. They conceded the dichotomy was wrong and adopted "inclusive EFs": EF names both the universal capacity and the culturally evolved skills layered onto it. Their objection was to "special alignment" — measuring each population in culturally appropriate circumstances — on the grounds that it flattens real data. They cited the practice, common in LMIC battery development, of reading non-completion of a task as evidence the child "did not comprehend" it and dropping the task, and argued that doing so with the card sort discards a finding rather than cleaning it. Their counter-principle: task choice is not "is this aligned with the participant's current abilities?" but "does this task present demands that overlap with demands the participant's current or future environments will present?"

That last move is the substantive disagreement. It is not about statistics. It is about whether adapting a test to a population's existing dispositions destroys information about how that population will function in environments it has not yet entered.

Mechanisms: what the tasks are actually asking

The strongest candidate mechanism is not exotic. These tasks all require arbitrary, decontextualized, rule-governed processing: to abandon a rule that made sense a moment ago because an experimenter says so; to treat unrelated nouns as a sequence whose position matters, when the words carry no relational meaning; to enumerate an abstract category; to produce the opposite of a gesture on command. None of this is regulated by the logic of the material itself. All of it is exactly what a classroom drills and a pastoralist herder does not.

The adaptation literature has mapped the incompatibilities [10]: at least six cultural preferences that clash with task assumptions — motivation and compliance with adult instructions; preference for group rather than one-to-one assessment; contextualized rather than abstract-academic thinking; cultural notions of speed and time; willingness to be silly, to be wrong in front of an adult, or to do the opposite of what was asked; and subject-matter familiarity. The recommended adaptations follow — naturalistic tasks, group administration, instructions that normalize silliness and error, accuracy rather than reaction time, avoidance of inhibition tasks premised on a prepotent response being the norm, locally familiar stimuli.

A second mechanism is conceptual rather than cultural, and it cuts against the whole framing. The dominant measurement model treats EF as a small set of latent components with unitary and special properties, a structure whose status the field has itself revised [14, 15]. An influential line of work goes further: EF is not a set of components at all but control skills exercised in particular social contexts [13], and treating it as components with fixed measurement properties is what forces the invariance question into an unwinnable shape [12]. Asking for cultural invariance of a component is then like asking for cultural invariance of a skill at swimming. This objection needs no new data to stand.

A third mechanism is mundane and underappreciated: the tests encode normative expectations about how a reasonable person should treat an adult examiner. That expectation is itself culturally learned, and it is the same expectation the field's own critics sometimes treat as a constant.

Competing explanations

Language and comprehension failure. The strongest evidence against this is internal to the 2025 study. If children simply failed to understand, forward span — which the same children largely could do — should have been fine. Instead forward span also showed large gaps (coefficients of −1.24, −0.91, −1.38), and the backward span floor persisted through four repetitions of instruction with examples.

Response-time artifact. A real concern in this literature: the original UK–Hong Kong comparison found a two-year EF lag much stronger in reaction time than accuracy, suggesting computer familiarity rather than control. The later reanalysis addressed it directly, showing self-reported computer use unrelated to EF performance in the mainland China sample [6]. The 2025 field data is largely accuracy-based, a design strength.

Non-randomized schooling exposure. This is the biggest threat to the causal story. Schooled and unschooled children in the same region are not exchangeable; attendance is plausibly related to distance, family structure, health, and local economics. The 2025 authors are careful about this, describing their data as "consistent with, but certainly do not prove," experience with formal schooling being causal. The defensible reading is that this is a well-controlled natural comparison that identifies a large gap, not an experiment that identifies a mechanism.

Uneven age distributions. The UK sample's mean age was 7.7 against 9.7 to 11.7 elsewhere, over an age range of 3–18. Age adjustment in a cross-sectional design with such shifts does substantial work: the backward-span finding is strengthened by being an age-independent floor, while the continuous task comparisons are more exposed.

Genuinely different regulatory strategies. A herding or foraging ecology requires sustained planning, monitoring and impulse control of a different kind. Control capacities may be universal while their trained expressions are not — the position the 2026 commentaries arrived at, and which the 2025 authors subsequently adopted.

Evidence quality

Ranked, honestly.

Strongest. The 2025–2026 PNAS exchange, because it is primary data plus a structured disagreement plus an independent reanalysis by a researcher with no stake in the original conclusion, all in one open-access issue. The preregistered Miller-Cotto registered report, because the analysis was fixed in advance on nationally representative data and the negative result is the kind that preregistration makes credible. The Xu and Chen multisite studies, because partial scalar invariance is a positive finding requiring adequate power rather than a null one.

Middle. The Kantor systematic review, which documents a clear and consequential practice gap but has small underlying subgroup counts, which its own authors flag. The Jukes adaptation framework, which is a selective review rather than a systematic one, and whose prescriptions are plausible rather than tested.

Weakest. The 2025 field comparisons in causal terms — small per-group samples, no randomization of schooling, uneven age distributions, multiple outcomes without a stated correction strategy. The backward-span floor survives these concerns well; the continuous task comparisons survive them less well.

Not established in either direction. Whether EF task performance predicts anything outside a schooled world. The 2025 authors state plainly that they know of no studies relating EF task performance to life outcomes outside schooled worlds. That gap is the largest hole in the literature.

Limitations of this brief

The strongest counter-design — a preregistered, multi-site study achieving full scalar invariance across both schooled and unschooled populations with adapted tasks — does not exist, and its absence is doing a lot of work in every position here. Configural invariance failure is itself model-dependent: a one-factor partially tau-equivalent model failing is as much a statement about the specification as about the children, and a one-factor model across three tasks is a coarse instrument for this question. The invariance literature is young relative to the construct's popularity, and negative invariance findings are systematically under-published relative to positive ones. Almost all of this concerns children, and because the cross-cultural EF literature is dominated by schooled samples, the alternative hypothesis is also thinly evidenced: the standard-battery critics have far less data than the standard-battery defenders, and that asymmetry should temper both.

Calibrated confidence

• ~90% that standard EF batteries require cultural adaptation before their scores are interpretable for populations outside schooled worlds. This rests on a comprehension-independent floor effect, half a century of cross-cultural findings on decontextualized processing, an established adaptation literature, and a documented practice gap in which 93% of studies report internal consistency while 30% test invariance.
• ~85% that EF task scores conflate a universal control capacity with a culturally learned disposition to comply with arbitrary, decontextualized demands. The phrase "conflate" is doing real work: it is a claim about decomposition, and decomposition is not directly observed anywhere in this literature.
• ~80% that within schooled populations the latent structure of EF tasks is at least partially invariant across cultures, based on replicated partial scalar invariance in two large multisite studies.
• ~75% that "inclusive EF" is the right frame and that the 2026 commentaries' measurement-theory argument was the more sophisticated contribution. This is a judgement about conceptual adequacy, not an empirical finding, and it is the claim in this brief most exposed to being wrong.
• ~60% that the schooling story is the correct mechanism behind the Kunene and Bolivian gaps. The study is consistent with it and does not identify it.
• ~45% that any single study in this dispute settles the decomposition. None does, and the field is currently moving too fast to pretend otherwise.
• ~90% that a low EF score from a child with an unfamiliar educational background should be treated as a statement about task familiarity until invariance has been demonstrated for that population. This is a decision recommendation, and it is close to costless.

Reversal conditions

Any one of these would materially change the reading:

• A preregistered multisite study reporting full scalar invariance across schooled and unschooled samples using culturally adapted tasks, with adequate power per group.
• Longitudinal follow-up of the Kunene cohorts showing that schooled and unschooled children converge on standard EF tasks after comparable years of schooling, which would identify the gap as an acquired skill with a clear acquisition function — or showing they never converge, which would refute the skill story.
• Any study linking EF task performance to non-academic outcomes — health, income, conflict resolution, foraging or herding competence — in unschooled populations. This is the decisive missing evidence, and its absence is currently doing more argumentative work than its absence warrants.
• A preregistered replication of Niebaum's reanalysis with pre-specified models, larger samples, and correction for multiple comparisons, that finds the academic associations do not survive.
• Evidence that configural invariance failures are artifacts of coarse latent models — for instance, two- or three-factor models restoring invariance in the Miller-Cotto data.
• Convergence on the "control skills" position to the point where EF is no longer modelled as latent components, which would dissolve the invariance question rather than answer it.

Decision implications

For anyone using an EF score to make a decision about a child — screening, diagnosis, placement, eligibility — the operating rule should be: an EF score from an unfamiliar educational background is uninterpretable until invariance has been shown for that population, and the burden of demonstrating it sits with the test's user, not the child. Numbers Reversed and DCCS scores from populations not shown to produce comparable item functioning should not be used to rank children across educational backgrounds at all.

For researchers, invariance testing should be a default reported result rather than a reflex reserved for questionnaire studies, and adapted-instrument papers that report internal consistency without invariance have not established that the instrument travels. Reporting within-group predictive relations alongside between-group means costs nothing and is the practice the dispute most directly rewards.

For the intervention industry selling EF training, the cross-cultural question compounds an already weak case: distal academic transfer is contested on contextual grounds even in well-studied populations [17], and early-childhood EF predicts later outcomes with effect sizes the field should report precisely [16]. A program that raises a score has not thereby raised a capacity that travels.

The productive move suggested by the whole exchange is not to pick a side on whether EF is universal or cultural, but to stop asking that question about a label and start asking it about demands: what regulation does this environment require, does this person have practice meeting those demands, and does the score reflect the second thing rather than the first?

References

1. Kroupin, I., Davis, H. E., Burdett, E., Bani Cuata, A., Hartley, V., & Henrich, J. (2025). The cultural construction of "executive function." Proceedings of the National Academy of Sciences, 122(27), e2407955122. https://doi.org/10.1073/pnas.2407955122
2. Mazzaferro, M. M., Obradović, J., Doebel, S., Wolf, S., Munakata, Y., Thorell, L. B., Willoughby, M., & Howard, S. J. (2026). The case for contextually grounded executive function measurement. Proceedings of the National Academy of Sciences, 123(15), e2533899123. https://doi.org/10.1073/pnas.2533899123
3. Niebaum, J. C. (2026). Standard EF tasks can still have predictive validity within diverse cultural contexts. Proceedings of the National Academy of Sciences, 123(15), e2536826123. https://doi.org/10.1073/pnas.2536826123
4. Kroupin, I., Davis, H. E., Burdett, E., & Henrich, J. (2026). Reply to Mazzaferro et al. and Niebaum: The limits of alignment: Grounding executive functions without assuming special contexts. Proceedings of the National Academy of Sciences, 123(15), e2603106123. https://doi.org/10.1073/pnas.2603106123
5. Miller-Cotto, D., Ribner, A., Ahmed, S., Ellis, A., & Czerwiński, S. K. (2026). Examining ethnic/racial measurement invariance in fourth-grade executive function: A registered report of data from the Early Childhood Longitudinal Study—Kindergarten. Journal of Educational Psychology, 118(1), 1–11. https://doi.org/10.1037/edu0000985
6. Xu, C., Ellefson, M. R., Ng, F. F.-Y., Wang, Q., & Hughes, C. (2020). An East–West contrast in executive function: Measurement invariance of computerized tasks in school-aged children and adolescents. Journal of Experimental Child Psychology, 199, 104929. https://doi.org/10.1016/j.jecp.2020.104929
7. Chen, L. L., Heng, J. A., Xu, C., Ellefson, M. R., Edwards, M., D'Souza, H., Fink, E., & Jess, M. (2025). Links between child executive function and adjustment: A three-site study. Child Development, 96(5), 1590–1604. https://doi.org/10.1111/cdev.14264
8. Holding, P., Anum, A., van de Vijver, F. J. R., Vokhiwa, M., Bugase, N., Hossen, T., Makasi, C., & Baiden, F. (2018). Can we measure cognitive constructs consistently within and across cultures? Evidence from a test battery in Bangladesh, Ghana, and Tanzania. Applied Neuropsychology: Child, 7(1), 1–13. https://doi.org/10.1080/21622965.2016.1206823
9. Kantor, C., Dong, Y., & Baral, K. (2026). A systematic review of language- or culture-adapted executive functioning measures. Applied Cognitive Psychology, 40(3), e70230. https://doi.org/10.1002/acp.70230
10. Jukes, M. C. H., Ahmed, I., Baker, S., Draper, C. E., Howard, S. J., McCoy, D. C., Obradović, J., & Wolf, S. (2024). Principles for adapting assessments of executive function across cultural contexts. Brain Sciences, 14(4), 318. https://doi.org/10.3390/brainsci14040318
11. Messer, E. J. E., Roome, H. E., & Legare, C. H. (2025). Learning to control through culture: Explaining variation in the development of self-regulation. Psychological Review, 132(4), 956–972. https://doi.org/10.1037/rev0000554
12. Doebel, S. (2020). Rethinking executive function and its development. Perspectives on Psychological Science, 15(4), 942–956. https://doi.org/10.1177/1745691620904771
13. Munakata, Y., & Michaelson, L. E. (2021). Executive functions in social context: Implications for conceptualizing, measuring, and supporting developmental trajectories. Annual Review of Developmental Psychology, 3, 139–163. https://doi.org/10.1146/annurev-devpsych-121318-085005
14. Miyake, A., & Friedman, N. P. (2012). The nature and organization of individual differences in executive functions. Current Directions in Psychological Science, 21(1), 8–14. https://doi.org/10.1177/0963721411429458
15. 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
16. Stucke, N. J., & Doebel, S. (2024). Early childhood executive function predicts concurrent and later social and behavioral outcomes: A review and meta-analysis. Psychological Bulletin, 150(10), 1178–1206. https://doi.org/10.1037/bul0000445
17. Niebaum, J. C., & Munakata, Y. (2023). Why doesn't executive function training improve academic achievement? Rethinking individual differences, relevance, and engagement from a contextual framework. Journal of Cognition and Development, 24(2), 241–259. https://doi.org/10.1080/15248372.2022.2160723