Robust experiment design via maximin optimization (Q1108717)

The paper gives a maximin-type approach and provides tools to design experiments which are maximin D-optimal and maximin D-\(\epsilon\)-optimal at a reasonable computational cost. The maximin criterion has been given. Various properties of D-optimal designs are dealt with and compared with maximin D-\(\epsilon\)-optimal designs. Models are considered that are partly nonlinear with respect to the parameters and if the admissible parameter space is such that the parameters in linear models can be chosen independently from those nonlinear models, then maximin D-optimal designs become classical D-optimal designs. The models considered are multivariate linear and partly linear regression models with vector parameters. Various applications are given.