Dimensionality assessment in bifactor structures with multiple general factors: A network psychometrics approach

Area of Science:

• Psychometrics
• Structural Equation Modeling
• Multivariate Statistics

Background:

• Dimensionality research often overlooks factor retention accuracy for complex structures with multiple general factors, common in intelligence, personality, and psychopathology.
• Existing methods for identifying group and general factors may not perform optimally under realistic conditions.

Purpose of the Study:

• To compare the performance of various factor retention methods, including a novel network psychometrics approach (Exploratory Graph Analysis with Louvain clustering - EGA_LV), for identifying group and general factors.
• To evaluate the accuracy of these methods in estimating the number of factors in bifactor structures, particularly those with multiple general factors.

Main Methods:

• Compared Kaiser criterion, empirical Kaiser criterion, Parallel Analysis with Principal Components (PA_PCA) or Principal Axis, and Exploratory Graph Analysis with Louvain clustering (EGA_LV) for group factor estimation.
• Developed and evaluated "second-order" versions of PA_PCA (PAP_CA-FS) and EGA_LV (EGA_LV-FS) using factor scores for general factor estimation.
• Examined the direct multilevel solution from EGA_LV.
• Conducted an extensive simulation study manipulating nine variables, including population error.

Main Results:

• EGA_LV and PA_PCA demonstrated superior performance in identifying the correct number of group factors.
• EGA_LV was more sensitive to high cross-loadings, while PA_PCA excelled with weak group factors and small sample sizes.
• PAP_CA-FS and EGA_LV-FS achieved near-perfect accuracy in estimating the number of general factors.
• EGA_LV was inaccurate in estimating general factors directly.
• EGA-based methods proved robust under conditions likely to be encountered in practice.

Conclusions:

• EGA_LV is highly effective for identifying group factors, especially when cross-loadings are present.
• EGA_LV-FS offers accurate estimation of general factors in bifactor structures.
• The combination of EGA_LV for group factors and EGA_LV-FS for general factors provides a powerful and reliable approach for analyzing complex psychometric structures.