The Social Shapes Test as a Self-Administered, Online Measure of Social Intelligence: Two Studies with Typically Developing Adults and Adults with Autism Spectrum Disorder.

The test's internal consistency was adequate but not ideal, and the authors state it is not reliable enough for high-stakes diagnostic use. Study 1 did not have a consistent measure of cognitive ability for all participants, so education was used as a rough proxy. The ASD group included adults recruited from SPARK, where diagnosis history was self- or parent-reported rather than directly clinically evaluated, and a smaller clinical subgroup of 25 adults. SPARK adults were described as living independently and not having a record of cognitive impairment, so the sample may not represent more impaired adults. The comparison groups differed in age, education, and race/ethnicity, and the analysis controlled for these differences rather than matching participants. The Frith-Happé categorization items had weak internal consistency, especially for theory-of-mind and goal-directed videos.

The paper names the National Institutes of Health as funder.

Do not use the Social Shapes Test to diagnose autism in a clinic. The paper states that its reliability is not high enough for high-stakes diagnostic use, and the adjusted difference between adults with and without autism was d = -0.27.