Projection indices and multimodality for mixtures of normal distributions

Projection pursuit indices are intended to give 'interesting' one- or two-dimensional projections of data points in higher-dimensional spaces. This paper compares the performance of seven such indices. They are all minimum for a normal distribution, and are standardized to be zero for that case. They are compared by considering projections of a mixture of four trivariate spherical normal distributions with means at the vertices of a regular tetrahedron.