Integrating Omission Behaviors in Computerized Adaptive Testing: A Comparative Analysis of the IRTree and Unidimensional IRT Models (Poster 8)

• In Event: Division D Graduate Student Committee In-Progress Research Gala Poster Session

Abstract

Computerized adaptive testing (CAT) enhances assessment precision and efficiency by tailoring item difficulty to individual ability levels. However, traditional CAT algorithms, such as those based on the unidimensional 2PL IRT model, often overlook test-taking behaviors like omissions, potentially biasing ability estimates and compromising validity. This study introduces an innovative CAT algorithm using the Item Response Tree (IRTree) model to jointly estimate ability and omission propensity traits. A pilot simulation with 1,000 test-takers demonstrated that the IRTree-based CAT maintains comparable precision while modeling omissions, enhancing fairness and interpretability. Findings highlight the potential of IRTree models for advancing equitable high-stakes assessments.

Authors

• Lizzy Wu, University of Illinois at Urbana-Champaign

• Justin L. Kern, University of Illinois at Urbana-Champaign

• Huan Kuang, Florida State University