Nonlinear experimental design based on the distribution of estimators (Q1205477)

Optimal experimental designs are studied for nonlinear regression models with normal errors. A more sensitive approximation to the distribution of the maximum likelihood estimator of the unknown parameter vector is used than asymptotic normality. Furthermore, various types of constraints for the parameter space can be taken into consideration by adding an adequate penalty function. On the basis of this approach the mean square error is taken as optimality criterion. A stochastic approximation algorithm is proposed for calculating optimal designs with fixed size. Three examples illustrate the theory.