An Application of Mixture Modeling in Reading Disabilities Research

• In Event: A Novel Methodological Approach to Understanding Disabilities

Abstract

Mixture modeling has begun to garner more attention as a useful technique to identify early readers who may be at-risk for later reading difficulties. Compton, Fuchs, Fuchs, Elleman, and Gilbert (2008) used latent transition analysis (LTA) to compare subgroups of children with reading disabilities to readers with typical development from first through fourth grades. The subgroups were based on measures of word identification, sight word efficiency and passage comprehension. Their findings demonstrated reading disability and typically developing classifications were fairly stable over time. However, there was a small group of students who transitioned from the typically developing group to the reading disability subgroup. Thus, LTA was able to identify students with late-emerging reading disability. McIntyre et al. (2017) utilized latent profile analysis (LPA) to identify subgroups of children with autism spectrum disorder based on multivariate profiles of reading skills. They found four distinct profiles. One profile was identified as average readers, one was characterized by specific comprehension difficulties, and the remaining two performed below average to varying degrees. Further, these profiles also differed with respect to autism symptomatology. Thus, the authors were able to identify associations between autism symptomatology and reading performance that could be used to inform intervention.

This paper provides a demonstration of how mixture modeling can be used to identify student profiles in early elementary that may indicate risk for reading difficulties. Reading skills data were collected from N = 965 students in the fall of first grade. Teachers were asked to identify students they considered at-risk and not at-risk. Students were also assessed to empirically identify risk status.

LPA was conducted using multiple measures of phonological awareness, word reading, and listening comprehension. Five profiles were identified, which were primarily differentiated by degrees of achievement; three demonstrated achievement in the average range or higher, while two performed below average across multiple measures. Multiple findings emerged, but two were novel and notable. First, the lowest performing profile exhibited substantial difficulties with listening comprehension, which is a skill not often targeted in early elementary but is critical for reading comprehension in later grades. Second, a profile consisting of students considered not at-risk by teachers demonstrated substantial overlap with a profile consisting of students considered at-risk by teachers.

These findings illuminate nuances that might not have been found by traditional analyses. Moreover, while risk for reading difficulties exists along a continuum, it is not uncommon for teachers to group students according to achievement levels. Mixture modeling aligns well with these practices as students are empirically subgrouped instead of relying on cutoff scores. The profiles identified in this study can assist in providing educational practitioners with heuristics to better group children by achievement levels across multiple reading skills and subsequently provide targeted intervention. For instance, it appears the lowest performing profile would benefit from direct listening comprehension instruction that complements phonological awareness and word reading instruction. By modeling heterogeneity, mixture models can provide useful and nuanced information about students who have special needs that vary from typically developing student populations.

Author

• Ryan Grimm, SRI International