The Impact of Attribute Hierarchies’ Distribution on Diagnostic Classification Accuracy

Chia-Ling Hsu, Yi-Jhen Wu

DOI: 10.64028/bdhc769116

Keywords: cognitive diagnostic model, attribute structure, classification accuracy, misspecification, diagnostic measurement

Abstract: Cognitive diagnostic models (CDMs) give detailed information about how well examinees’ grasp a set of fine-grained, discrete, latent skills/attributes. This information allows researchers and teachers to tailor instruction and craft cost-effective interventions to improve student learning. While learning, a student typically masters lower-level skills before a higher-level skill, which suggests four hierarchical attribute structures: linear, convergent, divergent, and unstructured (Leighton et al., 2024). Past studies assumed that all students within a sample have the same hierarchical structure (e.g., linear structure for primary school students’ learning of arithmetic). However, students’ learning processes can vary widely and yield different hierarchical attribute structures. Recognizing this possibility, this study ran simulations to test how well cognitive diagnostic models classified students across hierarchical attribute structures. The findings revealed that the distribution of these structures impacted classification accuracy. More candidate hierarchical attribute structures for classifying students yielded greater accuracy.

Submission Number: 28