The use of generalizability theory to inform sampling of educator language used with preschoolers with autism spectrum disorder

Purpose: A myriad features can impact the nature, frequency, and length of adult–child interactions important for language learning. Empirical investigations of language learning opportunities for young children with autism spectrum disorder (ASD) provide limited generalizable insight, with inferences more constrained to the sample than is often considered. The aim of this study was to explore a multidimensional understanding of reliability and define optimal measurement procedures for a measurement approach used to examine the language of preschool educators interacting with children with ASD. Method: We employed the logic of Generalizability Theory to differentiate sources of error for two measurement facets, occasion and observer. We video-recorded four 15-min occasions of educator–child interactions for 11 participants with ASD during free-play in their respective inclusive preschool classrooms. Two trained observers coded all videos for six educator language variables: open-ended questions/statements, choice questions, yes/no questions, imitation prompts, statements, and other talk. Results: The generalizability studies illustrated that, across all variables measured, observer accounted for little to no error. Occasion, however, accounted for much of the error for all language variables. To determine the number of occasions needed to achieve stable estimates of the variables, we manipulated occasion in the decision study. Five to more than 15 occasions were needed to achieve stability in educator language variables. Conclusion: To advance our understanding of the language learning environments of preschool classrooms that serve children with ASD, researchers must understand how aspects of the measurement design in those environments, such as occasion, impact the inferences they make.