Optimal Designs for Model Discrimination and Estimation in Binary Response

Abstract

In this talk, optimal design problems for binary response experiments are discussed. Different kinds of optimal designs have been investigated under an assumed model in the literatures, which include some for discrimination between rival models. The main goal of this work is to find an optimal design which concerns about model discrimination and estimation at the same time. More explicitly we look for designs which maximize the ordinary sum of squares of deviations at support points for two binary response models and show that it also possesses the model robustness property which minimizes the maximum bias between the true and assumed models. Some numerical results concerning the D-, A-efficiency as well as comparisons with other types of model discrimination optimal designs are presented for the optimal designs obtained here. (based on a joint work with Shi-Hau Hwang and Wei-Shan Hsieh.)