Motion Perception in Autism: A Difference Too Small to Rely On

Two literatures, one task

One set of studies reports that autistic people need more visual information than others before they can tell which way a moving pattern is travelling. Another set, using the same task, reports no difference at all. The first appeared in 2000 and has been replicated many times; so has the second, and neither has displaced the other.

The task is a random dot kinematogram. A field of dots moves across a screen, a proportion of them travelling together in one direction while the rest scatter at random, and the viewer says which way the coherent portion is heading. The measure is the coherence threshold, the smallest proportion of dots moving together at which a person can still read the direction. No single dot carries the answer, so the threshold indexes how well separate local motion signals are combined into one global percept.

Ruth Van der Hallen, working at KU Leuven, set out to count that literature rather than argue with it. With Catherine Manning at the University of Oxford and Kris Evers and Johan Wagemans in Leuven, she searched Web of Science and PubMed for work published between January 1980 and August 2018, screened 1460 records, and kept 48 articles comparing an autistic group with a typically developing one. The meta-analysis appeared in the Journal of Autism and Developmental Disorders in 2019.

Twenty-eight of those articles used random dot kinematograms. The other twenty used point-light displays, in which a walking human figure is represented only by dots at the joints and must be told apart from a scrambled version carrying the same local motions. That second paradigm, biological motion, has generated a disagreement of its own, and the review analysed the two both together and against each other.

Pooled across 227 effect sizes, the group difference reached 0.30 standard deviations, with a confidence interval running from 0.17 to 0.44. Autistic participants were on average less accurate, slower, or in need of more information before the motion resolved. A difference of that size counts as small under the conventional thresholds. The two paradigms produced nearly the same figure, 0.33 for the moving dots and 0.26 for the point-light figures, and the gap between them did not approach significance.

Effect sizes varied significantly both between studies and within them. Eleven moderators were tested against that variability: which paradigm was used, whether the task asked for detection or discrimination, which dependent variable was recorded, how long each stimulus lasted, how fast the dots moved, how many of them there were, the mean age of the sample, verbal, non-verbal and full-scale IQ, and whether the two groups had been matched on IQ. Not one of them accounted for it.

Why it has not settled

The arithmetic of the disagreement is in the sample sizes. Total enrolment per study ran from 18 to 141, averaging 34. Detecting a difference of 0.30 with any consistency requires samples above 200, so a study of ordinary size will find the effect or miss it depending largely on who volunteered that year. Twenty years of studies at that size produce exactly the split the field now has.

The split is not an artefact of what journals agreed to print. A rank correlation test found no relationship between how large a study was and how large its result was, which the authors attribute to a field in which a null finding on motion was as publishable as a positive one.

Motion thresholds also vary widely from person to person inside both groups, and that variation survived every adjustment for age and intellectual ability the review could make. Autistic and non-autistic participants overlap enough that an average difference of 0.30 standard deviations says almost nothing about any particular person. The samples were narrow in ways worth stating plainly: study mean ages ran from 6 to 37 years and averaged 21, mean full-scale IQ across studies was 111, and girls and women made up 11 percent of participants.

The mechanism is contested even among researchers who accept the difference. The usual account holds that local motion signals are integrated less efficiently in autism. Manning’s own earlier work with children cuts across that account, since a paradigm designed to separate the components found typical levels of internal neural noise and better than typical integration in the autistic children, who nonetheless performed like their peers on a standard coherence task. The review’s authors take this as a sign that the difficulty may lie in telling signal dots apart from noise dots, and they present it as a hypothesis.

Their recommendation to their own field amounts to a moratorium: stop running small between-group comparisons, and adopt paradigms capable of identifying the mechanism. Nothing in the meta-analysis licenses a claim about how any individual child sees movement, and nothing in it connects a coherence threshold to reading, catching a ball, or crossing a road, none of which were measured. If a report describes your child as impaired on a motion task, the finding standing behind that phrase is genuine at the level of groups, small, and after two decades still unexplained.