A Meta-Analysis of Gaze Differences to Social and Nonsocial Information Between Individuals With and Without Autism

Abstract

Objective: Numerous studies have identified abnormal gaze in individuals with autism. However, only some findings have been replicated, the magnitude of effects is unclear, and the pattern of gaze differences across stimuli remains poorly understood. To address these gaps, a comprehensive meta-analysis of autism eye-tracking studies was conducted.

Method: PubMed and a manual search of 1,132 publications were used to identify studies comparing looking behavior to social and/or nonsocial stimuli between individuals with autism and controls. Sample characteristics, eye-tracking methods, stimulus features, and regions of interest (ROIs) were coded for each comparison within each study. Multivariate mixed-effects meta-regression analyses examined the impact of study methodology, stimulus features, and ROI on effect sizes derived from comparisons using gaze-fixation metrics.

Results: The search yielded 122 independent studies with 1,155 comparisons. Estimated effect sizes tended to be small to medium but varied substantially across stimuli and ROIs. Overall, nonsocial ROIs yielded larger effect sizes than social ROIs; however, eye and whole-face regions from stimuli with human interaction produced the largest effects (Hedges g = 0.47 and 0.50, respectively). Studies with weaker study designs or reporting yielded larger effects, but key effects remained significant and medium in size, even for high-rigor designs.

Conclusion: Individuals with autism show a reliable pattern of gaze abnormalities that suggests a basic problem with selecting socially relevant versus irrelevant information for attention and that persists across ages and worsens during perception of human interactions. Aggregation of gaze abnormalities across stimuli and ROIs could yield clinically useful risk assessment and quantitative, objective outcome measurements.

Keywords: autism spectrum disorder; eye tracking; meta-analysis; meta-regression; social information processing.

Figure 1

Flow diagram of search strategy, inclusion/exclusion, and final sample.

Figure 2

Stronger methodology produces lower effect estimates (A–C), but effects remain significant with medium effect sizes under high rigor designs (D). Note: DD = developmental disability.

Figure 3

Nonsocial regions-of-interest (ROI) show larger effect sizes than social ROIs, and the effect is strongest for human interactive stimuli (A); eyes and whole face ROIs show the largest effect sizes (B). Note. P-value represents significantly large effect sizes for social and nonsocial ROIs contained within human/interactive stimuli.