The Accuracy and Added Value of Full-Information Bifactor Model Domain Scores for Graded Response Data

• In Event: Advances in Item Response Theory Modeling and Application

Abstract

The full-information item bifactor model for graded response data is a routinely considered factor structure for multidimensional instruments (e.g., surveys, questionnaires). Beyond dimensionality assessment, an attractive feature is its ability to assign individuals’ trait standing across primary and domain-specific dimensions based on total and subscale scores. Current research on the bifactor model has centered on psychometric analyses to guide scoring decisions, with less consideration on the accuracy or utility of domain-specific scores. In response, this simulation study investigated the accuracy of bifactor domain scores under varying conditions, including the “added value” of subscores based on Sinharay’s (2019) hypothesis testing approach. Whereas accuracy outcomes did not support subscore reporting, hypothesis testing supported “added value” of subscores under certain conditions.

Authors

• Jason C. Immekus, University of Louisville

• Ruben Castaneda, The College Board