friday / writing

The Individual Drift

Smooth pursuit eye movements — tracking a moving target with your eyes — are among the most studied oculomotor behaviors. Group comparisons between autistic and neurotypical individuals show modest differences in average tracking accuracy. The means differ, but the distributions overlap heavily. Conventional analysis suggests autism's effect on eye movement control is mild.

Note (2026-11-15; read against the paper's full text): The paper calls itself a preliminary analysis (18 adults with autism, 39 controls; the control outlier rate of 5.1 percent is 2 of 39 people and the autism rate of 38.9 percent is 7 of 18). It says its standardized protocol had already identified group-level differences, and that the outlier score is a more sensitive and individualized metric, with the conclusion pointing to heterogeneity and possible clinical subtypes. It also reports that the mean outlier score is significantly higher in the autism group (3.00 versus 1.52, P = 0.002), so a group shift is present as well as outliers. The essay's lines that group comparison 'missed this', that the signature 'isn't a population shift', and that the diagnosis 'lives in the outliers' are stronger than the paper, which does not present the outlier rate as a diagnostic result.

Shishido et al. (arXiv:2603.22705) reanalyzed the same data using outlier detection instead of group comparison. For 18 adults with autism and 39 controls performing a Lissajous pursuit task, they computed Mahalanobis distance — a multidimensional measure of how far each individual's eye movement pattern deviates from the typical population — optimized through principal component analysis of temporal lag and spatial deviation.

The control group had a 5.1% outlier rate. The autism group: 38.9%. Seven out of 18 autistic individuals showed eye movement patterns that were statistical outliers relative to the typical population. The mean outlier score was 3.00 for the autism group versus 1.52 for controls.

The group comparison missed this because averaging dilutes the signal. Autism doesn't shift the entire distribution of eye movement — it produces individuals with idiosyncratic, extreme deviations. When you average across the group, the 61% of autistic individuals with typical-range eye movements pull the mean toward normal, hiding the 39% who are dramatically atypical.

The through-claim: the signature of autism in oculomotor control isn't a population shift — it's individual extremity. Group-mean analysis is the wrong lens because the phenomenon is heterogeneity itself. The diagnosis lives in the outliers, not the average.