Measuring Skill Growth and Change: Latent Growth Cognitive Diagnostic Models

During the past decade, Cognitive Diagnostic Models (CDMs) have drawn escalating attention in educational assessment and measurement for it holds the potential to provide a fine-grained diagnosis for the respondents engaging in assessment or questionnaire items (Rupp & Templin, 2008). Generally, CDMs focus on single cross-sectional time points to provide diagnostic information for learning. However, longitudinal assessments have been commonly used in education to assess students’ learning trajectories as well as to evaluate intervention effects. Thus, it becomes necessary to identify longitudinal growth in skills profile mastery and conduct more studies on the CDMs extended to longitudinal settings. To address this need, I proposed a family of latent growth CDMs (LG-CDMs) that incorporate latent growth curve modeling, covariate, and mixture extensions, to measure the growth of skills mastery, evaluate attribute-level intervention effects, and compare latent subgroups in terms of different learning trajectories. I conducted simulation and empirical studies to examine the parameter recovery, classification accuracy, and the practical application of the proposed unconditional LG-CDM, conditional LG-CDM, mixture unconditional LG-CDM, and mixture conditional LG-CDM. The results indicated stable parameter recovery and classification of latent classes for different assessment conditions as well as demonstrated practicality in providing diagnostic information about learning and growth. Findings from my dissertation suggest that building on the well-established growth curve modeling, the application of covariate and extension-based LG-CDMs can help understand the effect of the intervention and explanatory factors on the change of attribute mastery concerning various latent populations, which can yield meaningful inferences on learning and teaching.