Reparameterization of second-order interaction effects through the covariance matrix

Abstract

It is demonstrated that the reparameterization of second-order interaction terms in ANOVA can be achieved by means of constraints on the covariance matrix. In particular, classical reparameterization conditions for fixed factors can be formulated in this way. The approach stresses the common nature of fixed and random factors in mixed linear analysis.