Component-Based Item Response Theory Utilizing Generalized Structured Component Analysis

• In Event: 25.082 - Fairmont Royal York Roundtable Session Three
  In Roundtable Session: 25.082-5 - Dimensionality and Estimation: Advancements and Applications in (Multidimensional) Item Response Theory and Factor Analysis

Abstract

Estimations of item and person parameters in item response theory (IRT) generally utilize maximum likelihood estimation (MLE) methods. Although it is popular, MLE depends upon distributional assumptions such as multivariate normality and therefore require large amounts of data to achieve stable estimation. In this paper, we will incorporate generalized structured component analysis (GSCA) utilizing least square estimation into the common IRT model, to help address this small sample and estimation issue. The proposed IRT applying GSCA determines a composite component score rather than a factor score for ability, which is called component-based IRT (CB-IRT). Capability and efficiency will be examined via a simulation study. In addition, we also demonstrate CB-IRT using an empirical data.

Authors

• Ji Hoon Ryoo, University of Southern California

• Kwanghee Jung, Texas Tech University

• Heungsun Hwang, McGill University

• Jaehwa Choi, The George Washington University