Optimal designs for comparing curves in regression models with asymmetric errors

Abstract

This paper devotes the optimal design problem for comparing two regression curves estimated from two samples with asymmetric errors. The μp-optimality criterion is developed under the framework of second-order least squares estimation, and the associated design theory, including the general equivalence theorems and efficiency bounds, is established. The particle swarm optimization (PSO) is used to generate μ∞-optimal designs. A simulation study shows that the use of the μp-optimal designs yields a substantial reduction on part of the design space in terms of the width of confidence bands. Some extensions of the μp-optimal designs are also addressed.