Simple Structure Rotation

 

In component analysis, the components resulting immediately from an analysis are often hard to interpret. Exploiting rotational freedom of the components, however, one can rotate these components to simple structure (for instance, by “varimax”) and thus considerably simplify the interpretation of the components. This is well-known for principal components analysis, but it also holds for various other models. I have developed varimax rotation procedures for Multiple correspondence analysis and variants thereof (see Kiers, 1991), and am currently working with Michel van de Velden on varimax rotation for Correspondence Analysis.

In three-way analysis, the results of Tucker3 analysis have rotational freedom in various directions: All component matrices as well as the core array can be rotated. It is very important in practical applications to have techniques for simplifying the component matrices and the core, so as to simplify interpretation. After some first attempts to rotate the core over all three directions simultaneously to a simple form (see Kiers, 1992), rotation to optimal simple forms was proposed by Kiers (1997, 1998), in the form of three-way generalizations of varimax and simplimax rotation. A procedure which, additionally, aims at simplicity of the component matrices for the three modes is provided in Kiers (1998).

The above involvement in simple structure rotations led to some developments for simple structure rotation of ordinary two-way components as well. Properties of existing simple structure rotation techniques were described (Kiers & Ten Berge, 1994), and a new procedure called SIMPLIMAX, leading to optimally simple loading matrices, was developed (Kiers, 1994).