Comparison of Emergent Latent Class Analysis Techniques and Distal Outcomes Using the Three-Step Method

• In Event: Methodological Advances in Longitudinal Mixture Modeling

Abstract

Latent class analysis (LCA) is not a new technique. However, fresh implementations of LCA have emerged in the literature which aim to detect unobserved heterogeneity in a population. While this is one goal of LCA, another is the examination of latent classes with respect to distal outcomes. While primary and secondary schools strive to meet accountability standards set by the Every Student Succeeds Act (ESSA), the relationship between their ability to meet these benchmarks and distal outcomes for students who exit these institutions should also be evaluated. In other words, we should evaluate the potential impact that latent class membership has on future, or longitudinal, outcomes for students.
Another concern is the practical application of latent class models. While many models exist, practitioners may be unsure as to whether there are practically significant differences between these approaches. Models should be compared to evaluate whether they engender varying results with regard to class formations and temporal distal outcomes. To do this, the proposed study will evaluate the performance of three approaches to mixture modeling: a general latent class analysis (LCA), latent class tree analysis (hereafter, LCTA; van den Bergh, Schmittmann, & Vermunt, 2017; van den Bergh & Vermunt, 2019) and LCA using individual case residuals (hereafter, LCA-ICR; see Marcoulides & Trinchera, 2019). Because LCA-ICR is a newly-introduced method, a comparison of LCA-ICR to other LCA approaches is needed to determine if any practically significant differences exist regarding latent class formations, distal outcomes, and potential decisions by researchers in the field.
These approaches will be compared in an applied study of school climate data. In South Carolina, school climate surveys are administered to gauge public schools’ yearly progress and meet requirements of the state’s accountability legislation. While schools can be classified as having a positive or poor school climate, schools could also be evaluated at a deeper level to determine if latent subclasses of schools exist and whether they share important characteristics.
The three-step approach is recommended to address the inclusion of distal outcomes in mixture models (Nylund-Gibson, Grimm, & Masyn, 2019). This approach has been documented in the literature with respect to a range of mixture models (e.g., LCA, LCTA), but it has not yet been applied to LCA-ICR. While Marcoulides and Trinchera (2019) suggest that this procedure can be extended to a conditional model with covariates, they make no mention of the inclusion of distal outcomes.
Student and teacher school climate data collected in 2018 by the South Carolina State Department of Education (SCDE) in the spring of 2018 will be analyzed. Data for distal outcomes of interest (e.g., career placement, college performance) will be extracted from the SCDE website. Given the chosen methods and population of interest, the proposed study aims to determine: (a) How schools’ latent class assignments vary across the three methods; (b) how the three methods vary regarding distal outcomes across schools; and (c) the extent to which class assignment and temporal distal outcomes vary across surveyed groups (students, teachers) and the three methods.

Author

• Kathleen McGrath, Charlotte-Mecklenburg Schools