Invariant admissible and optimal designs in cubic regression on the \(\nu\)-ball (Q1193807)

It is for computational reasons that we restrict ourselves to rotatable designs when we discuss design optimality in multiple polynomial regression models. However, a (globally) optimal design needs not to be rotatable even for an orthogonal equivariant criterion. The unrestricted design problem for linear and quadratic models has been solved by the first author [J. Stat. Plann. Inference 31, No. 2, 219-233 (1992; Zbl 0756.62029)]. This paper deals with the cubic set-up on the ball, and the authors find an essentially complete class of invariant designs by combining the concept of admissibility of designs with a less restrictive invariance property (i.e. the invariance of information matrices with respect to the permutations and sign changes on the experimental region).