Evaluating Autism Communication Theories

Following literature review around causal theories of key features of ASD, the investigator team reviewed data items collected in this release. We determined that we had sufficient items to evaluate some of the claims regarding communication theories of autism. We used items from the following tables of the SPARK release: Basic Medical Screening, Background History–Child (ASD only), the Background History–Sibling (Sib only), the Repetitive Behaviors Scale–Revised (Lam and Aman, 2007 (link); ASD only), and the SCQ (Rutter et al., 2003 ). Items related to each theoretical construct, as they were discussed in the literature, were grouped, and we tested construct fit using a confirmatory factor analysis (CFA). We set our criteria as a root mean square of error approximation (RMSEA) < 0.05 (Kim and Mueller, 1978 ; Brown, 2015 ) and standardized root mean squared residuals (SRMR) < 0.08 (Kim and Mueller, 1978 ; Brown, 2015 ). Any construct that did not initially meet these criteria was split into two factors, and the factor with the more theoretically-relevant items was selected and refit. Each individual participant’s value for each construct was derived from Multivariate Item Response Theory (MIRT), and a normal distribution was imposed on each group. For a few constructs with few items, quantification was obtained via Principal Components Analysis (PCA) or as the value of a single item, in cases where only a single item pertained to the construct. Because the ASD and sibling samples included somewhat different items (i.e., the Sibs’ data did not include the RBS-R and some items in the background history), some constructs in the ASD-Only analysis differed in item composition from their corresponding partners in the ASD + Sibs analyzes. All data items present in the Sibs cohort were available in the ASD cohort, but not vice versa. Therefore the constructs in the ASD-Only cohort in some cases derived from a greater number of items than items than corresponding constructs in the Sibs (and therefore ASD + Sibs) cohort. We examined correlations between the “finer” constructs in the ASD-Only sample and the corresponding “coarser” constructs in the ASD + Sibs sample.