Optimal designs for rational regression models

Abstract

In this paper we consider locally optimal designs problems for rational regression models. In the case where the degrees of polynomials in the numerator and denominator differ by at most 1 we identify an invariance property of the optimal designs if the denominator polynomial is palindromic, which reduces the optimization problem by 50%. The results clarify and extend the particular structure of locally c-, D- and E optimal designs for inverse quadratic regression models which have recently been found by Haines (1992) and Dette and Kiss (2009). We also investigate the relation between the D-optimal designs for the Michaelis Menten and EMAX-model from a more general point of view.