Abstract
Understanding growth and development is a central objective in educational and psychological research, particularly in contexts where individual differences inform instruction, intervention, and policy decisions. Traditional variable-centered approaches, such as latent growth curve models, primarily estimate average trajectories of change and may obscure meaningful heterogeneity in developmental patterns. In contrast, person-centered methods—including growth mixture models (GMMs) and diagnostic classification models (DCMs)—offer frameworks for identifying latent subpopulations and classifying individuals based on distinct developmental trajectories or patterns of skill mastery. However, the application of these models introduces important methodological challenges related to model specification, scoring, and classification consistency across time and groups.
This dissertation examines key methodological issues associated with model-based classification in longitudinal and cross-sectional educational contexts. The first study investigates the role of covariate-informed scoring in growth mixture models, evaluating how the inclusion of covariates during the scoring process influences latent class enumeration and the interpretation of developmental trajectories. The second study explores the integration of diagnostic classification models with vertical scaling approaches, assessing how combining continuous growth measurement with categorical skill classification affects model fit, score comparability, and diagnostic inference across grade levels. The third study examines the role of anchor items in maintaining classification consistency in diagnostic classification models, focusing on how item linking strategies influence the comparability of classifications across assessments administered at different grade levels.
Across these studies, the findings highlight the sensitivity of latent classification results to methodological decisions and underscore the importance of aligning modeling approaches with the underlying structure of the data and intended use of results. This work contributes to the methodological literature by clarifying conditions under which classification-based models produce stable and interpretable results and by offering practical guidance for researchers and practitioners applying these models in educational and psychological settings. Ultimately, this dissertation aims to support the development of more accurate, interpretable, and actionable assessment practices for understanding individual differences in development.
