Associations between the pairs of the autism subphenotypes were tested. Chi-square tests were used for the pairs of categorical variables; nonparametric Kruskal-Wallis tests were used for the pairs of categorical and continuous variables, while Spearman rank correlation tests were used for the pairs of continuous variables. The effects of the six potential covariates on SOC and BEH were tested using mixed linear models with family as a random effect (SAS v 9.1, SAS Institute, Cary, North Carolina). There were four categorical covariates: AGP site, gender, best estimate IQ (in three categories, <50, 50–69, and ≥70), and verbal/nonverbal status (in three categories, verbal, nonverbal with ≥5 words, and nonverbal with <5 words). There were two continuous covariates: the age and calendar year of ADI-R completion. These covariates were chosen based on the literature and our preliminary review of the data. Due to nonnormality, rank transformation was applied to SOC and Box-Cox transformation to age of ADI-R completion.
