What Eye-Tracking Studies of Autistic Preschoolers Actually Measure

What the eye-tracker recorded

A child under five sits in front of a screen. Two films run side by side, one of people and one of objects, and a camera records which region of the screen the eyes rest on and for how long. The number taken from that recording is a proportion: of all the time the child looked at anything, how much fell on the non-social side.

Twenty-nine such experiments were pooled in 2024 by Julian Hinz and Svein Eikeseth of Oslo Metropolitan University, with Fillip Eikeseth of Oslo University Hospital and Katarzyna Chawarska of the Yale Child Study Center, in the journal Autism Research. Between them the studies had tested 4,125 children, 2,214 with an autism diagnosis and 1,939 without, none older than five, with a mean age in the autism groups of 31.6 months. The question was whether autistic preschoolers look at non-social things more than their peers do.

Pooled across the 37 comparisons the review extracted, they did, by 0.39 of a standard deviation, with a confidence interval running from 0.23 to 0.56. On the scale the authors adopt, 0.2 counts as a small difference, 0.5 as medium and 0.8 as large. Variation between studies was substantial, an I² of 82 percent, which means most of the spread in results came from genuine differences among the experiments rather than from sampling.

The authors then divided the studies by what had actually been displayed. Stimuli matching the interests autistic adolescents and adults commonly report about themselves went in one group: machines and vehicles, geometric and fractal patterns, letters, numbers, signs, construction blocks. Every other non-social image went in the other, meaning toys, household objects, scenery and backgrounds.

Across the 18 comparisons using ordinary objects, the difference was 0.07, with an interval from −0.13 to 0.27 and a p value of 0.49. Across the 19 using interest-related content, it was 0.69, from 0.52 to 0.86, p below 0.001. The combined 0.39 falls between two results rather than summarising either, and heterogeneity remained high within each group, 64 percent and 73 percent.

One name covered a considerable range of procedures as well. Eighteen of the 29 studies used the side-by-side preference design, nine used free viewing of a scene, one a visual search task and one a value-learning task, and the outcome extracted might be proportion of fixation time, mean or peak viewing duration, number of fixations, or saccadic reaction time. Further analyses tested whether age, sex, autism severity on the ADOS or non-verbal developmental quotient explained the variation among the interest-related studies; only 13 and 12 studies reported severity and developmental scores, too few to detect a small association, and none of the tests reached significance. Only the content on the screen moved the result.

What it means for you and your child

The familiar version of this research is that autistic children attend to objects in preference to people. Across studies that showed toys, household items and everyday scenery, no reliable difference in looking was found at all. The preference these tasks register attaches to a particular category of content, and the general claim overstates what the pooled evidence supports.

Four unusually large studies supplied more than half the children, 2,271 of the 4,125, and several of them were built as early-detection work, testing whether looking behaviour could flag autism in the second and third years. A screen made of geometric films may well do that. Detecting a difference and describing how a child attends to the world are separate achievements, and the second does not follow from the first.

Nor is the direction of the preference settled by the design. In the side-by-side tasks the non-social film always competed with a social one, so time spent on the geometry could equally reflect the faces holding less interest. The authors checked this and found that the design did not account for the null result among ordinary objects, which is the more informative half of the check; within the interest-related studies the comparison was limited by how few used any other paradigm.

The interest categories themselves were drawn from what autistic adults and adolescents say about their own preoccupations, then applied to two-year-olds. No standard instrument assesses circumscribed interests in children that young, and the ADI-R does not attempt it below the age of three. Whether a toddler’s looking predicts the intense interests of a later childhood is a question the authors name as the next piece of work, not one this review answers.

Why interest-related content draws the eye is also left open. Reward value, a systemising cognitive style, and interest as a way of managing anxiety have each been proposed in the literature, and this analysis adjudicates none of them. The result that is firm is narrower and more useful: a two-year-old who watches the traffic rather than the actress is showing a preference located in the content, and one that ordinary toys did not produce. What that preference becomes by seven or twelve, no one has yet followed a child long enough to know.