Billet publié sur Bluesky le 10/09/2026 11:25

Résumé

Abstract Diffusion decision models (DDMs) offer the potential to go beyond standard accuracy and response-time indices to better understand perceptual decision-making in autism. One unanswered question is whether autistic participants can flexibly adjust the speed and accuracy of their decision-making according to task demands. Across two preregistered studies, 50 autistic and 50 non-autistic children aged 6–14 years completed a visual orientation task with no explicit instructions to be fast or accurate, and a visual motion task under both speed-emphasis and accuracy-emphasis instructions. These studies allowed us to investigate the influence of task, task instruction and modelling approach on group differences. We fit Bayesian hierarchical DDMs using a rigorous blind modelling approach, and follow-up two-step non-hierarchical analyses. Irrespective of task, instructions, and modelling approach, we found no conclusive evidence for group differences in DDM parameters. For the first time, we showed that autistic children can flexibly adjust their decision-making strategies according to speed–accuracy instructions. These results show that cognitive flexibility is not uniformly reduced in autism. To better understand within-participants variability, we investigated relationships between decision-making parameters and ADHD-related traits, reading ability, sensory processing and coordination skills. We found task-specific evidence for relationships between DDM parameters and sensory underresponsivity, sight-word reading and hyperactivity/impulsivity. Resolving inconsistent results when applying DDMs to autism will require contrasting modelling approaches, clear reporting of task instructions, and considering dimensions that co-occur with autism. Open practices statement De-identified data are available in a safeguarded repository on the UK Data Service (https://doi.org/10.5255/UKDA-SN-858223 ). Experimental code, analysis scripts and output files are available for Study 1 (https://osf.io/vfqwx/ ) and Study 2 (https://osf.io/2gyc4/ ). Both Study 1 and Study 2 were preregistered (https://osf.io/gjksu, https://osf.io/2hdpg/ ). At the point of preregistration, we had collected data from one autistic participant, but we did not look at the data or complete any analyses before preregistration.