Multistructure Statistical Model Applied To Factor Analysis

Area of Science:

• Multivariate Statistical Analysis
• Psychometrics
• Econometrics

Background:

• Existing statistical models for mean and covariance structures (e.g., Bock and Bargmann, Joreskog, Wiley, Schmidt, and Bramble) are often specialized.
• Factor analytic models, including Thurstone's multiple-factor model, have limitations regarding variable scaling and unique variances.
• The distinction between principal components analysis and factor analysis can be blurred in traditional models.

Purpose of the Study:

• To describe a general statistical model for the multivariate analysis of mean and covariance structures.
• To present a new class of factor analytic models as a specialization of the general model.
• To introduce a model with factor loadings invariant to variable scaling and positive unique variances.

Main Methods:

• Development of a general statistical model encompassing various existing multivariate analysis techniques.
• Specialization of the general model to derive a novel class of factor analytic models.
• Utilizing matrix calculus for the statistical development of the model.
• Parameter estimation via maximum likelihood with Newton-Raphson iterations.

Main Results:

• The general model provides a unified framework for diverse statistical analyses of mean and covariance structures.
• A novel factor analytic model is derived, offering an alternative to Thurstone's multiple-factor model.
• The proposed factor analytic model ensures common-factor loadings are invariant to variable scaling.
• Unique variances in the new model are constrained to be positive, avoiding confounding with principal components analysis.

Conclusions:

• The general statistical model offers a flexible and comprehensive approach to multivariate analysis.
• The specialized factor analytic model provides a theoretically sound alternative with desirable properties like scale invariance.
• The model clearly distinguishes factor analysis from principal components analysis, enhancing clarity in multivariate research.