Typologies of Autism: The Utility of Latent Class Analysis to Understand Heterogeneity in Social Responsiveness

• In Event: A Novel Methodological Approach to Understanding Disabilities

Abstract

The Social Responsiveness Scale (SRS; Constantino et al, 2003) is a widely used parent report measure of traits for youth with ASD that varies from low to high on a continuum. This assumption of a single continuum and the validation methods used, however, may mask the complexity of social responsiveness heterogeneity. This paper contributes a nuanced understanding of variance by taking an alternative lens to how expressions of social responsiveness differ for youth with ASD.

The purpose of this presentation is to introduce the potential utility of LCA for applied special education researchers. We will illustrate the use of LCA by addressing if patterns of social responsiveness for children diagnosed with autism can identify distinct sub-groups. The mixture modeling framework provides a way to empirically investigate whether variation in presentations of autism are best understood as a single continuum or via a framework of typologies (Ingram, Takahashi, Miles, 2007). LCA is considered a “person-centered” modeling approach and uses as the unit of analysis an individual's response to a set of items (Masyn, 2013). This method considers the possibility that heterogeneous groups, called latent classes, may be differentiated by response patterns for a set of items.

To illustrate the use of latent class analysis, a dataset is used from the National Database for Autism Research, 2019; NDAR) comprising data across 54 studies. This paper focuses on the SRS scale composed of 65 items that can be used to understand social responsiveness and reciprocal expressions for youth with ASD. This study provides an alternative perspective, highlighting heterogeneity in expression of social responsiveness for individuals with ASD.

Choosing which items from the SRS to use in the LCA model is an important research decision and has received little attention in the methodological literature. To contextualize the use of mixture in this paper, we will use the SRS scale to do the following, 1) choose a reduced number of SRS items with the highest discriminatory power (Marbec & Sedki, 2017), 2) run a LCA on the reduced set of items, 3) contextualize the results of the LCA by comparison to other commonly used latent variable models (e.g., factor analysis). Throughout the presentation of the LCA framework, we focus our discussion on the connection of the methodological approach and its use in special education research.

Results of the analytic procedure resulted in the selection of the 12 items most likely to distinguish sub-groups, falling in-line with our theoretical objectives focused on highlighting diversity in the expression of social responsiveness. Next, we present the procedure for deciding between models when different numbers of latent classes are present utilizing theoretical considerations and model-based fit criteria (Nylund, Asparouhov, Muthen, 2007). Emphasis will be placed on correct interpretation and challenges commonly encountered when using this approach. Finally, the presentation will include a detailed overview of the mixture modeling results for the SRS scale using a sample of youth diagnosed with ASD. Particularly, the interpretation and real-world significance of the results will be considered.

Authors

• Adam Garber, University of California - Santa Barbara

• Mian Wang, University of California - Santa Barbara

• Karen L. Nylund-Gibson, University of California - Santa Barbara