The Polytope of Optimal Approximate Designs: Extending the Selection of Efficient Experiments
share
Consider the problem of constructing an experimental design that is optimal for a given statistical model with respect to a chosen criterion. To address this problem, the literature typically provides a single solution; sometimes, however, there exists a rich set of optimal designs. The knowledge of this set allows the experimenter to select an optimal design based on its secondary properties. In this paper, we show that the set of all optimal approximate designs often corresponds to a polytope. The polytope can be fully represented by the set of its vertices, which we call vertex optimal designs. We prove that these optimal designs possess unique characteristics, such as small supports, and suggest their potential applications. We also demonstrate that for a variety of models, the vertex optimal designs can be computed using rational arithmetic with perfect accuracy. Consequently, we enumerate all vertex optimal designs for several standard multifactor regression models.
READ FULL TEXT
