The Fisher Information Function and Design of Experiments for Estimation in Non-Linear Statistical Models

摘要

The problem of designing experiments for estimating parameters of nonlinear models is studied in a Bayesian framework, with the objective of maximizing the anticipated Fisher information. The theoretical set-up for optimal two-stage designs is formulated. Optimal designs for reliability attribute life testing experiments are derived. A non-Bayesian measure of efficiency of the designs is defined and computed. Sequential group testing experiments which are epsilon-most efficient are presented. (Author)