A two-stage strategy for the construction of d-optimal experimental designs

Abstract

Under the classical framework of optimal designs, it is usually assumed that the model is known for which an optimal design is derived under a given optimality criterion. In real-life applications, however, models are rarely known in advance especially when new processes are being investigated. Therefore, the optimal design derived from the classical framework may be the best design under the assumed but wrong model. In this paper, we propose and study a two-stage strategy to tackle the problem. We also introduce the concept of expected relative efficiency in order to compare the performance of the proposed two-stage designs with the classical uni-stage designs. Some open problems for future research are discussed

Keywords:

• Model discrimination
• Model augmentation
• Expected relative efficiency