What the Cambridge meta-analysis of 59 studies actually found about autistic cognitive flexibility
The largest meta-analysis of cognitive flexibility in autism to date pooled 59 studies — 2,122 autistic individuals and 2,036 neurotypical controls. The headline finding: a small-to-moderate effect size (g = 0.44) for cognitive flexibility differences in autism. That’s measurable, but not the catastrophic deficit the clinical literature sometimes implies.
Perseverative errors — the tendency to repeat a previously correct response after the rules have changed — emerged as the most consistent finding. This maps onto the Wisconsin Card Sorting Test, the workhorse of cognitive flexibility research, where participants must infer rule changes from feedback. Neurodivergent participants, on average, made more perseverative errors than neurotypical controls.
But the heterogeneity matters more than the average. The effect sizes varied substantially across studies, across tasks, across age groups. There is no uniform autistic cognitive flexibility profile.
Some studies found large effects. Others found none. The meta-analysis couldn’t identify age as a significant moderator — meaning childhood, adolescence, and adulthood showed similar patterns. Whatever cognitive flexibility differences exist, they don’t simply resolve with development.
The Cambridge team also flagged a persistent methodological problem: most studies used clinical samples with co-occurring conditions, and few controlled adequately for IQ differences. The “autistic cognitive inflexibility” finding is real in aggregate — but it’s far messier, and more nuanced, than the textbooks suggest.
Why laboratory findings on autism and real-world, lived neurodivergent experience persistently diverge
Here’s the paradox that haunts autistic cognitive flexibility research: laboratory tests show mixed results, but autistic people and those around them consistently report real-world flexibility challenges. The Behavioral Rating Inventory of Executive Function shows robust flexibility difficulties when parents or teachers rate autistic children. Self-report measures show the same pattern in autistic adults. Yet the controlled lab tasks — card sorting, task switching, reversal learning — produce inconsistent findings.
The standard explanation is ecological validity. Laboratory tasks strip context.
A card-sorting task presents explicit rules, immediate feedback, and no social stakes. Real life is saturated with implicit rules, delayed consequences, and relationships that matter. The flexibility required to navigate an unexpected change at work — where your manager’s mood, your colleagues’ expectations, and your own sensory state all interact — bears little resemblance to inferring that “colour” has replaced “shape” as the sorting criterion.
But there’s a deeper issue. The paradox assumes that “cognitive flexibility” is a single thing that should show up consistently across contexts. If it doesn’t, we call that a measurement problem. The alternative interpretation: flexibility isn’t a single thing. It’s a family of context-dependent processes that share a label but not necessarily a mechanism. This matters because the “paradox” framing protects the construct. If lab results and real-world reports diverge, we blame the lab tasks for lacking ecological validity — rather than questioning whether “cognitive flexibility” as a unified trait was ever the right model.
How cognitive flexibility is learned and conditioned rather than a fixed executive trait
The traditional model places cognitive flexibility in the executive function hierarchy — a top-down control process that supervises “lower” associative learning. Flexibility, in this view, is the capacity to override learned associations when they’re no longer useful. It sits above habit and below goal-directed reasoning.
Recent theoretical work collapses this hierarchy. Cognitive flexibility isn’t a separate module that governs associative learning — it’s grounded in associative learning. The same mechanisms that create habits can create flexibility. The difference is what gets conditioned.
Flexibility can be triggered by contextual cues — even subliminal ones. It can be enhanced or suppressed by reward history. It can be bound to specific stimuli rather than generalising across tasks. This means flexibility is learned, not fixed. The environments you’ve navigated, the feedback you’ve received, and the contexts you’ve encountered all shape how flexibly you respond in new situations.
The implications for autism are substantial. If flexibility is conditioned by environment and reward history, then autistic cognitive inflexibility might reflect different learning histories rather than a broken executive module. An autistic child who receives inconsistent feedback in chaotic sensory environments learns different flexibility patterns than a neurotypical child in a predictable, well-scaffolded context. The “deficit” might be a reasonable adaptation to unreasonable conditions.
This doesn’t mean autistic people don’t experience genuine flexibility challenges. It means those challenges might not originate where the clinical model assumes — in a malfunctioning prefrontal cortex — but in the accumulated weight of environments that punished exploration and rewarded rigidity.
Why the rigidity frame fails autistic people and what should replace it
The “rigidity” frame does two things. First, it locates the problem inside the autistic person — a fixed trait, an executive dysfunction, a brain that can’t shift sets. Second, it implies the solution is remediation — intervene at the best developmental periods to do so, teach the autistic person to be more flexible, train them out of their perseverative tendencies, provide “cognitive flexibility” corrections.
This misses the structural picture. If flexibility is learned and context-dependent, then the intervention target isn’t the autistic brain — it’s the environment. Predictable contexts, explicit rule changes, sensory-appropriate feedback, and sufficient processing time all support flexible responding. Chaotic, implicit, and sensorially overwhelming contexts suppress it.
The meta-analysis finding — small-to-moderate effect size, substantial heterogeneity — is consistent with this reframe. Autistic people aren’t uniformly inflexible. They show variable flexibility depending on task, context, and conditions. The research literature has spent decades trying to pin down “the” autistic flexibility deficit when the variability itself was the finding.
What should replace the rigidity frame? A structural account that asks: under what conditions does flexibility emerge, and under what conditions does it collapse? This shifts the question from “what’s wrong with autistic cognition?” to “what environmental features support or undermine flexible responding?” The answers will be different for different autistic people — because flexibility was never a single thing in the first place.
The “rigidity” label persists because it’s clinically convenient. It justifies intervention. It locates the problem in a way that doesn’t implicate institutions, workplaces, or educational systems. But it fails the people it claims to describe — and the research, read carefully, has been showing this for years.
Citations
Lage, C., Smith, E. S., & Lawson, R. P. — A meta-analysis of cognitive flexibility in autism spectrum disorder
Braem, S., & Egner, T. — Getting a grip on cognitive flexibility
