2 Watching Steps of a

Bartkowiak and Szustalewicz, 1997 (B&Sz), have proposed an implementation of the grand tour algorithm designed specially to detect outliers in multivariate data. The algorithm works by performing a sequence of rotations and projections onto a speciic 2D-plane ((proj). This is equivalent to perform a series of rotations yielding transformed coordinates of the data: X (tr) = XA, with A being the rotation matrix, while X and X (tr) denote the data matrices before and after the rotation appropriately. A superimposed concentration ellipse indicates the outstanding points. We complement the paper of B&Sz by considering some details and variants of the implementation of the grand tour algorithm. In particular we watch the trajectories of the projected points. Our concern is the denseness of the watched projections. We look at the trajectories of the projected points visible in the proj plane and the frequencies of their appearing in various sectors of that plane. In the Appendix we present the derivation of the formula for the rotation matrix A employed in each step of the algorithm for obtaining the transformed data coordinates.