The Utility of a Mixture Modeling Approach to Understanding Heterogeneity in Dual Language Learners

• In Event: Using Innovative Methodological Approaches to Explore the Heterogeneity of Latinx Dual Language Learners Within Early Childhood Classrooms

Abstract

Mixture modeling is a statistical modeling approach that allows for the identification of subgroups in a population that are hypothesized to exist but are not directly visible. This statistical modeling approach provides a unique context to study profiles of students that combines an innovative statistical approach and developmental theory. Included in this modeling framework are cross-sectional models like latent class analysis (LCA) and latent profile analysis (LPA) as well as longitudinal mixture models such as latent transition analysis (LTA) and growth mixture modeling (GMM), among others. These models are commonly used in areas of educational research to help identify heterogeneous groups.
The present study focuses on applying these methods to identify heterogeneity in dual language learners (DLLs). While previous research has suggested that DLLs are heterogeneous in nature (Gonzalez et al., 2016; Lonigan et al., 2018), there is little research on the application of mixture models to identify subgroups of Spanish-English DLLs and their implications. For example, López and Foster (2021) used LPA to identify a group of heterogeneous Latino DLLs at the end of preschool: Balanced average bilinguals, Spanish dominant, English dominant, and Balanced low. Balanced average students outperformed the other profiles in cognitive assessments and the implications are that administrators should not make assumptions about students abilities based on their DLL label. Bell et al. (2016) used LPA and latent growth curve analysis (LGCA) to test invariance of latent profiles of peer play behaviors across ethnicity and DLL status. Results show that children who engaged in behaviors that facilitated peer interactions had the highest academic skills across the preschool year and differences in academic skills among the profiles remained stable across the school year.
This paper is intended to highlight the use of mixtures as a statistical approach that provides a glimpse into the subpopulations of DLLs and contextualizes the promise of this modeling approach to address important research questions. We will present modeling possibilities for use with DLLs and highlight some of the current research that is available using mixtures models with DLLs. We will highlight modeling steps for the use of mixture modeling and provide resources for future inquiry.

Authors

• Dina Naji Arch, University of California - Santa Barbara
  Presenting Author

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