A comparative evaluation of factor- and component-based structural equation modelling approaches under (in)correct...

Area of Science:

• Statistics
• Quantitative Psychology
• Econometrics

Background:

• Structural equation modeling (SEM) encompasses factor-based and component-based domains, distinguished by how constructs are statistically represented.
• Current evaluations often compare SEM approaches solely under factor models, potentially yielding misleading performance conclusions.
• A lack of clear formulation for population component models and their relationships hinders comprehensive SEM approach evaluation.

Purpose of the Study:

• To clarify population component models and their interrelationships.
• To comprehensively evaluate four SEM approaches: maximum likelihood, factor score regression, generalized structured component analysis (GSCA), and partial least squares path modeling (PLSPM).
• To assess the robustness of SEM approaches to construct misrepresentation.

Main Methods:

• Clarification of population component models and their relationships.
• Comprehensive evaluation of four SEM approaches under diverse experimental conditions.
• Comparison of factor-based (maximum likelihood, factor score regression) and component-based (GSCA, PLSPM) SEM methods.

Main Results:

• Factor-based SEM approaches are optimal for estimating factor models, while component-based approaches are suitable for component models.
• Component-based SEM approaches demonstrate greater robustness against construct misrepresentation compared to factor-based approaches.
• GSCA is recommended over PLSPM for component-based SEM, irrespective of construct representation accuracy.

Conclusions:

• The choice of SEM approach should align with the underlying construct representation (factor or component model).
• Component-based SEM offers superior robustness when constructs are potentially misrepresented.
• GSCA emerges as the preferred component-based SEM method due to its performance and robustness.