Bifactor models and rotations: exploring the extent to which multidimensional data yield univocal scale scores

Area of Science:

• Psychometrics
• Psychological Measurement
• Structural Equation Modeling

Background:

• Psychological measures often yield item response data interpretable as both unidimensional and multidimensional.
• Structural ambiguity in latent structures leads to extensive confirmatory factor analyses (CFAs) to validate single-trait interpretations.
• Bifactor models, an alternative to unidimensional, correlated, or second-order models, are underutilized in personality assessment due to limited understanding.

Purpose of the Study:

• To address the underutilization of bifactor models in personality assessment.
• To clarify conceptual and modeling issues related to multidimensionality.
• To differentiate bifactor models from second-order models and highlight bifactor analysis applications.

Main Methods:

• Description of conceptual and modeling challenges in multidimensionality.
• Explanation of exploratory bifactor modeling techniques, including Schmid-Leiman and target bifactor rotations.
• Detailed explanation of confirmatory bifactor modeling and its distinction from second-order models.

Main Results:

• Bifactor analysis provides a framework for evaluating subscale plausibility.
• It helps determine the extent to which scores reflect a single variable despite data multidimensionality.
• Bifactor analysis aids in assessing the feasibility of applying unidimensional item response theory (IRT) models.

Conclusions:

• Dimensionality determination is distinct from assessing score interpretation as a single variable or the impact on IRT parameters.
• Multidimensional data can yield interpretable scale scores.
• Unidimensional IRT models can be appropriately fitted to multidimensional data in many contexts.