On Demonstrating the Bifactor Model with 2015 PISA Collaborative Problem-Solving Items

Abstract

New assessments built on theoretical frameworks require effective measurement tools to evaluate their psychometric properties. However, conventional unidimensional model assumes that the latent construct is unidimensional, without accounting for interdependencies among sub-factors. In contrast, the bifactor model offers an alternative framework that incorporates a general factor and orthogonal skill-specific factors, enabling a comprehensive evaluation of constructs with underlying subfactors. This project demonstrates the use of the bifactor model in evaluating assessments for constructs with multiple dimensions, with a focus on the collaborative problem-solving (CPS) competency assessment items from the 2015 Programme for International Student Assessment (PISA). The study focuses specifically on 15-year-old students from four Chinese provinces in Mainland China, with a sample size of N = 1,675 participants. EFA determines the optimal number of factors, followed by a comprehensive CFA using empirically driven and theory driven models. Findings demonstrate that the empirically driven bifactor model outperforms other theory driven MIRT models, exhibiting the best model fit. The implications identify problematic items requiring further investigation. This paper contributes to the relatively unexplored realm of CPS assessment design using the bifactor model as an analytic tool. The analysis and findings contribute to the scholarly understanding and offer practical implications for practitioners and researchers in the field.