Extending the Bifactor Model to Account for Multiple Sources of Measurement Error

• In Event: Advances in Estimating Reliability: Innovative Approaches and Applications

Abstract

Nearly all applications of bifactor models to date use single-occasion data that fail to account for all relevant sources of measurement error and confound excluded components of measurement error with systematic global and group-specific trait effects. In this study, we eliminated these problems using confirmatory factor analysis techniques to extend the bifactor model to account for multiple sources of measurement error. The extended models partition systematic trait variance into general and group components, and residual variance into transient, specific-factor, and random-response measurement error. Results at score composite and subcomponent levels can be derived in a single computer run to identify the most appropriate scores to report in practice.

Authors

• Walter Peter Vispoel, University of Iowa

• Murat Kilinc, University of Iowa