T-optimal designs for discrimination between rational and polynomial models

Abstract

In the current article the problem of constructing analytically experimental designs, optimal according to the popular criterion of T-optimality introduced by Atkinson and Fedorov in 1975, for discrimination between simple rational and polynomial regression models is considered. It is shown how the classical results of approximation theory can be utilized to achieve explicit formulas describing the behavior of support points and weights of T-optimal designs for different fixed prior parameter values. An example of a practical problem with rational and polynomial regression models is provided. Then the numerical calculation of the experimental designs, optimal according to the robust analogs of T-criterion, for the models in the example is briefly discussed. Refs 10. Fig. 1. Tables 3.