Analysis of covariance and correlation structures (Q1059965)

The analysis of covariance and correlation structure models consists of the development of a model for the covariance or correlation matrix, estimation of the parameters and establishment of statistical properties of the estimation method. The earliest covariance structure model was the factor analytic model. Recently, it has been extended to more sophisticated models. These generalizations enable one to test relatively complex, substantive theories. The main purpose of this paper is to consider different approaches in analyzing covariance or correlation structures with parameters subject to general nonlinear constraints. An attractive covariance structure model is defined. This model is of the most general form; it is scale-invariant and includes all practical covariance structure models as special cases. It can be used in assessing the covariance structures as well as the correlation structures, and provides clear and precise interpretation of the results. The possibility of imposing constraints enables the model to achieve extremely wide applications. A new estimation method, the two-stage constrained maximum likelihood procedure, is also developed. In a large sample this new procedure is shown to have similar statistical properties as the classical constrained maximum likelihood approach. But it requires less computer time and storage space. Also, either the sample covariance matrix or the sample correlation matrix can be used. Therefore, it represents an attractive estimation method for the present general model. Two examples are given to illustrate the theory discussed, and technical details are presented in the Appendix.