Consider what a child has to do to complete the Wisconsin Card Sorting Test. They sort cards by one rule, colour perhaps, until the examiner tells them they are wrong, at which point they must work out that the rule has changed and find the new one. The test is filed under cognitive flexibility, and a low score gets read as a flexibility problem. But look at what else is required. The child must inhibit attention to the stimulus dimensions that are no longer relevant. They must hold the examiner’s feedback in working memory and apply it. They must sustain attention across a long and repetitive procedure. They must understand the instructions, which arrive verbally. And they must tolerate being told, repeatedly, that they are wrong.

A low score could reflect any of those. That problem has a name in the assessment literature, the task impurity issue, and it is the organising concern of a review published in August in Research in Autism by Francesca Anderle of the University of Trento and the University of Oxford, working with Paola Venuti at Trento.

The review examines the neuropsychological tools used to assess executive functioning in autistic children and adolescents aged 3 to 18, drawing on empirical studies published between January 2013 and December 2025. Its conclusion is that findings across this literature vary substantially, and that much of the variation traces to age, task characteristics, and the linguistic and fine motor demands the tasks impose rather than to executive function itself.

What the Review Covered, and What It Did Not

The scope is specific and the boundaries matter.

Included studies needed at least 20 autistic participants, a neurotypical comparison group, and participants aged 3 to 18 with an IQ of 70 or above who were verbal. Studies of children with intellectual disability or significant language impairment were deliberately excluded, on the reasoning that broader cognitive differences would confound interpretation. Studies of bilingual children only were also excluded, given the documented relationship between bilingualism and executive function. Neuroimaging-only studies and intervention studies were left out.

So this describes autistic children with average intellectual profiles and functional language, and does not extend to those with intellectual disability or high support needs. That is a real limit on generalisation and the authors state it.

They are equally direct about the review’s other constraints. It is a narrative review, not a systematic review or scoping review, and was not conducted under a formal registered protocol. The search was restricted to open-access publications and articles available through institutional subscriptions, which means relevant studies behind paywalls were excluded. Some commonly used tasks were left out because no studies employing standardised versions could be found. These are meaningful caveats, volunteered rather than extracted.

Language Is Inside Every Task

The strongest thread running through the review concerns language, and the evidence is more striking than the general framing suggests.

Ellis Weismer and colleagues examined executive function in autistic children aged 8 to 12 across a sorting task, Corsi Blocks, an n-back task, a Go/No-Go task, and a Flanker paradigm. Differences between autistic and non-autistic children were present. When the samples were matched on social communication abilities, those performance differences disappeared.

That result covers inhibition, working memory, updating, and shifting, four of the domains this field treats as separable executive components. It does not mean the difficulties were unreal. It means what the tasks detected was not cleanly attributable to executive function.

Other findings point the same way. Yeung and Chan found that autistic children and adolescents without language delays performed comparably to non-autistic peers on fluency tasks, while those with language delays performed worse, which suggests the fluency measure was partly indexing language. Anderle and colleagues found no group differences on an enumeration task used to measure verbal working memory once linguistic fluency was removed from the composite score. And in verbal fluency testing, autistic children recalled fewer items in the fruit and object categories than in the animal category, a pattern that looks like a property of categorisation rather than of generativity.

The review also notes that verbal working memory tasks show more consistent group differences than spatial ones, which is what you would expect if verbal demand is doing some of the work.

The Motor Demand Nobody Counts

The second confound gets less attention than language and is arguably more concrete.

Cancellation tasks, used to assess visual attention, and graphical fluency tasks like the Five-Point Test and the Trail Making Test, used to assess generativity and flexibility, all impose substantial fine motor demands. The child has to draw, connect, or mark, quickly, with a pen. Motor coordination difficulties are common in autism.

Autistic children show longer completion times on motor-intensive measures including the Trail Making Test and the Stroop Test. When fine motor skills were removed from the composite score of a visual search task, the group difference disappeared.

The implication is uncomfortable for anyone interpreting a timed, pen-based test. A child who is slower with a pencil will produce a lower score on an instrument named for cognitive flexibility, and nothing in the score itself will tell you which one you measured.

Slower Is Not the Same as Worse

A pattern the review returns to repeatedly deserves separating out, because it changes what a score means.

Across multiple studies and multiple tasks, autistic children showed comparable accuracy but took significantly longer. On planning tasks, some studies found autistic children aged 8 to 18 did not differ from peers in accuracy, but required more time to make the first move.

The authors read this as a difference in processing efficiency rather than a core executive deficit, and they draw the obvious practical conclusion: allow enough time to capture what the child can actually do. A timed test administered under time pressure to a child who thinks accurately but slowly will report a deficit that a longer administration would not find.

Age Changes What You Are Measuring

Executive function is not a fixed set of components that children get better at. The components themselves separate out over development, and the review lays out the sequence.

In preschoolers, executive function is best described as a single unified construct rather than distinct abilities. By around age 4, inhibition and updating can be differentiated. Cognitive flexibility does not fully emerge as a separate dimension until roughly ages 8 to 9. By ages 10 to 11, all three core dimensions are fully differentiated from one another.

This has a sharp consequence. Administering a cognitive flexibility measure to a five-year-old means measuring something that has not yet separated from the rest of their executive functioning. A score can still be produced. What it represents is a different question.

The age effect also shows in which difficulties get reported. Studies of children aged 6 to 12 most often identify planning as the hardest domain. Studies including adolescents most often identify shifting. The authors suggest this partly reflects rising demands as children progress through school, and partly reflects that compensatory strategies like calendars and structured planning systems can mitigate organisational difficulty while shifting remains harder to work around.

A methodological point follows. Studies using wide age ranges frequently report no group differences, while studies using narrower ranges report clear ones. Pooling a 3-year-old and a 17-year-old into one analysis obscures both.

The Testing Situation Is Part of the Test

Several findings in the review concern the administration itself rather than the task content, and they are the ones most likely to surprise.

Negative feedback matters. On card sorting tests, positive feedback improved performance while negative feedback increased perseveration errors, and removing negative feedback enhanced performance outright. Since telling the child they are wrong is built into how the Wisconsin Card Sorting Test works, part of what the test measures is how the child responds to being corrected.

Instruction modality matters. Performance improves when instructions are delivered both verbally and visually rather than verbally alone.

Stimulus familiarity matters. Preschoolers performed worse on an inhibition task using numerical stimuli than on a version using a familiar boy and girl distinction. Simple recall outperformed tasks requiring simultaneous processing of letters and numbers.

Task framing matters. Autistic children find open-ended requests more difficult than structured, well-defined ones, in laboratory tasks as in daily life.

And some tasks may recruit systems they were not designed to test. The authors note that n-back tasks and implicit rule shifts may inadvertently engage emotional regulation, and that changes in arousal can worsen performance in ways that have nothing to do with the executive process under examination.

On format, the review finds no clear advantage either way between computerised and paper-based versions for school-age children. Computerised versions may reduce fine motor and motor speed involvement, which is useful for younger children, and autistic preschoolers do perform better on computerised inhibition tasks than on examiner-administered ones. That difference disappears by school age.

Laboratory and Life Disagree

The review documents a divergence that will be familiar to anyone following the assessment literature.

Laboratory-based executive function measures yield mixed findings in autism research, while ecological assessment consistently reveals difficulties. In one study, autistic and non-autistic children showed no significant difference in planning on laboratory tasks, while parent-report measures showed notable discrepancies in the same age group.

Nor do informants agree with each other. Teachers more often report difficulties with self-monitoring and emotional control; parents more often highlight difficulties with initiation. The authors attribute this to the genuinely different demands of home and school. A parent is watching whether homework gets started and finished. A teacher is watching whether a child can monitor and regulate their behaviour among thirty others.

This is the same divergence that recent work has documented in preschoolers, where performance tasks and parent ratings correlated at close to zero, and in adults undergoing ADHD evaluation, where rating scales and objective tests failed to converge even after screening for over-reporting. Three literatures, three populations, one pattern.

What the Authors Propose Instead

The review does not stop at diagnosis of the problem, and its recommendations are specific.

The central proposal is a shift away from what the authors call fractionated approaches, which assess executive domains in isolation, toward a broader executive attention framework. Under that view, executive function is an integrated control system managing attentional resources and coordinating goal-directed activity, rather than a set of separable modules. A low Wisconsin Card Sorting Test score would then be interpreted as a global weakness in executive attention with a probable primary difficulty in flexibility, rather than as a flexibility deficit full stop.

They also call for tools designed to isolate executive ability from linguistic and motor demands, including subtractive methods that remove the contribution of language or fine motor speed from a composite score. Several findings in the review come from exactly that approach, and they are the ones where group differences vanished.

Three further recommendations follow. Assess language and motor skills alongside executive function, so their contribution can be accounted for rather than assumed away. Combine laboratory, home, and school information rather than relying on one context. And adopt strength-based approaches that identify what a child can do rather than cataloguing deficits.

What This Means for Anyone Reading a Score

For clinicians, educators, and anyone who receives a report containing executive function test results, several things follow.

A low score is a question, not an answer. Before concluding that a child has a flexibility difficulty, the reasonable questions are how much language the task required, how much fine motor work, how much time pressure, and whether the child was being told they were wrong throughout.

Language and motor ability belong in the interpretation. If those were not assessed, the executive result cannot be cleanly interpreted, because the confounds are documented and substantial.

Slow is not the same as impaired. Comparable accuracy with longer completion time is a specific finding with specific implications, and it is not what most score reports communicate.

One measure is not a profile. Given how much results move with task design, age, and context, a single instrument produces a single data point rather than a picture.

None of this argues for abandoning these tests. They remain the best structured tools available, and the review is written by researchers who use them. The argument is narrower and harder to dismiss: a score from an executive function task is a measurement of a child performing a specific task under specific conditions, and the distance between that and a statement about their executive functioning is larger than the label on the test implies.