The Factor Structure of Cognitive School Readiness Skills: Conceptual Versus Statistical Distinctions

• In Event: Poster Session 14
  In Poster Session: PS 14 Topic - School Readiness/Childcare

Abstract

School readiness refers to the broad set of skills (including executive function [EF], language and literacy skills, and mathematical skills) that children develop in the early years of their education (e.g., preschool) and that predict later academic success (Duncan et al., 2007). Given the complexity of these skills, researchers often use multiple instruments to assess them (e.g., Schmitt et al., 2017). Latent factors are commonly used to model constructs in statistical analyses to reconcile the shared variance among tasks and to eliminate measurement error. Recent research, however, has suggested that the use of latent variable models may be inappropriate in some cases, mitigating the benefits of these models (Rhemtulla et al., 2020). Theoretically, we expect that (a) measures within the same construct (e.g., EF) would be more closely correlated with each other than they would with measures of a different construct; and (b) skills at a given time point (e.g., fall of preschool) would be more highly correlated than scores across time points. The purpose of this study was to examine whether there is statistical evidence to support construct-specific latent factors of EF, language/literacy, and mathematics, and to support time-specific latent factors in the fall and spring of preschool.

This study examined data from three cohorts of preschool-aged children who were enrolled in state-subsidized programs. Our sample included 684 students (52.34% male; Mage = 4.80 years) who were nested in 180 classrooms and 85 providers. In the fall and spring semesters, trained research assistants administered a battery of assessments that included three measures of EF, four measures of language/literacy skills, and six measures of early mathematical skills. Using Mplus 8 (Muthen & Muthen, 2017), we ran nine a priori models to examine to what extent the data were consistent with different specifications of construct- and time-specific latent factors.

As shown in Table 1, the model that had the most statistical support (Model 7) included a single random intercept latent factor across the two timepoints and two time-specific latent factors of the residual variation. That is, all of the tasks, regardless of their theoretical construct or timepoint, shared substantial variation. A post-hoc version of this model was run that removed non-significant loadings from the time-specific factors and is presented in Figure 1. These time-specific latent factors primarily represented early literacy and mathematics skills (i.e., factor loadings > 0.10), but not EF or language skills.

The results of this study underscore the importance of considering the statistical interpretations of latent variable models and testing multiple hypothetical models. Contrary to the typical conceptualization of school readiness skills, the best-fitting model in this study did not provide evidence to suggest that EF, literacy/language skills, and mathematical skills load onto distinct latent constructs despite their conceptual distinctions. These results suggest that latent variables of construct-specific factors may not be appropriate when studying these constructs in early childhood, or better measurement is needed to identify distinct statistical latent factors. Additionally, the considerable shared variation across time-points is necessary to consider in any latent variable analyses.

Authors

• Kirsten Lee Anderson, Purdue University, West Lafayette
  Presenting Author

• Robert Duncan, Purdue University, West Lafayette
  Non-Presenting Author

• Jennifer Kristine Finders, Purdue University, West Lafayette
  Non-Presenting Author

• David James Purpura, Purdue University, West Lafayette
  Non-Presenting Author

• James G Elicker, Purdue University
  Non-Presenting Author

• Sara Anne Schmitt, University of Oregon
  Non-Presenting Author