Discrimination Between Two Binary Data Models. Sequentially Designed Experiments.

In this paper we describe and illustrate a sequential approach for the problem of designing experiments to discriminate between binary data models as efficiently as possible. The models considered ought to belong to the class of binary response models which is conveniently regarded as a subclass of Generalized Linear Models. Further they may be either nested or non-nested, and theirlinear predictor structures must be given. We examine this problem for the case of two rival models. Optimal design theory is utilized so as to derive important properties of and relevant results with respect to optimal experimental designs. Two sequential methods of designing efficient experiments for discrimination between two models are suggested. Furthermore, simulation techniques are used not only to illustrate how the methods can be implemented but also to enable comparisons between sequentially designed experiments and local optimal ones, that are designs which are optimal with respect to specific parameter values and model assumptions. (author's abstract)