Analysis of Unique Factor Structures in Factor Analysis

Abstract

We often modify the number of factors, or factor patterns, in order to improve models of factor analysis. However, these techniques have the disadvantage of skewing the meaning of factors. Moreover, although there could be correlations among unique factors used in Structural Equation Modeling (SEM), the criterion for adding correlations to unique factors is not clear. We developed a new approach to SEM by using covariance selection for graphical modeling. We have verified the effectiveness of this method by describing and analyzing three examples. It is concluded that we can modify models of factor analysis by using this technique, while maintaining the meaning of common factors without violating their normal distribution.