The Use Of Generalizability Theory To Inform Sampling Of Language Learning Environments For Young Children With Autism Spectrum Disorder

Abstract

Features of autism spectrum disorder (ASD) can impact the nature, frequency, and length of adult-child interactions that are important for language learning. Empirical investigations of these interactions in preschool classrooms are limited and often provide minimal insight into the reliability of the observations beyond inter-rater agreement. To promote a multidimensional understanding of reliability and define optimal measurement procedures, the researcher employed the logic of Generalizability Theory to differentiate sources of error, namely persons (i.e., educator or child participants) and two measurement facets or conditions, occasion and observer. The researcher video-recorded 4, 15-minute occasions of educator-child interactions for each of the 11 participants with ASD during free play in their respective inclusive preschool classrooms. Two trained observers coded all videos for variables of educator proximity, educator language (i.e., open-ended, choice, yes/no, imitation, statement, and other), and child language. The generalizability studies illustrated that, across all variables measured, observer accounted for little to no error. Occasion, however, accounted for the majority of the error for all behaviors except child language. To determine the number of occasions needed to achieve stable estimates of the variables, the researcher manipulated occasion in the decision study. Although researchers would need only three occasions to reliably estimate child language, five to more than 15 occasions were needed for educator language and proximity variables. With a need to balance statistical rigor with pragmatics, refining, eliminating, or identifying new variables may be a necessary step toward making inferences about the language learning environments of young children with ASD.