Objectives: To identify good endophenotypic indicators of autism susceptibility via admixture analysis of cognitive-style measures in a large, genetically informative sample.
Methods: The sample comprised adults with a clinical diagnosis of autism (n=232) and typical adults with (n=138) or without (n=808) a child with autism. Mean response times for the Eyes test, EFT, and MRT administered online were subject to multivariate admixture analysis. Models of 1–4 latent components were evaluated using the Bayesian information criterion and posterior predictive checks; posterior probabilities were used to estimate each participant's component membership, which served as a grouping variable in subsequent inferential analysis.
Results: Mixture modeling revealed evidence of ternary structure underlying the joint normal distribution of these performance data. The major latent component showed typical performance and comprised 70% of the sample. Remaining taxa with discrete shifts in performance contained unaffected parents or adults with autism in significant frequency. Mental rotation was undifferentiated in this model, whilst performance of the Eyes test was the most efficient endophenotypic marker. Further analysis found no artifacts in distributional characteristics or participants' clinical features.
Conclusions: Multivariate admixture analysis suggests that the Eyes test and EFT are candidate autism endophenotypes with broad relevance to research as objective criteria for the classification of individuals with autism.