bi-factor model , confirmatory factor analysis , S-1 bi-factor model , psychological measurement
Abstract:
To represent complex psychological constructs such as multifaceted personality traits, general intelligence, or mental disorders, the bi-factor model is frequently used. Its main advantage over competing models is its clear and often insightful distinction between diff erent parts of multidimensional constructs. It defi nes a general trait
across all observed variables and specific traits representing the various facets of the construct.
The unique characteristics of the bi-factor model’s structure come with several challenges that currently need more attention. In this thesis, I tackle three of these
issues in three articles. The first article investigates the frequent occurrence of weak specifi factors in bi-factor model applications. It explains why the characteristics of the bi-factor model in combination with typical measurement design in psychology should be expected to produce weak specific factors. The meta-analysis shows the pattern of problematic parameter estimates. Using simulations, the article analyses the statistical power and the parameter recovery under realistic conditions and provides guidelines for applied research. The second article investigates the flexibility of bi-factor model variants and their relationships to one another. Whereas previous research has noted the excessive flexibility of the bi-factor model compared to other models, the current work shows in simulations that its different variants can flexibly imitate each other. The most important consequence is that even some of the most basic claims derived from the model need to be questioned, because they may entirely depend on the choice between two equally well-fitting representations of the data. It is discussed that this issue cannot be resolved from a statistical perspective alone and a detailed account of the infl uence on parameter and trait estimates is provided. The third
article proposes an alternative modeling approach for cases in which the underlying assumptions of a full, symmetrical bi-factor structure are violated. On a large example dataset, a set of replications and a multiverse analysis highlight the key strengths and limitations of this proposed approach.
The current work aims to expand the statistical bi-factor model toolbox and to guide the application and interpretation of previously suggested models. For this
purpose, I combine statistical insights with a meta-scientific perspective on the bi-factor model’s application. In this way, it became clear that an improved understanding of the discussed problems is key to their solution.
Dieser Eintrag ist Teil der Universitätsbibliographie.
Das Dokument wird vom Publikationsserver der Universitätsbibliothek Mannheim bereitgestellt.