Discussion: posterior probabilities and experimental designs for discriminating between regression models

In Section 1 the relationship between the design criterion of Box and Hill (1967) and the T-optimum designs of Atkinson and Fedorov (1975a) is explored for competing pairs of models. The extension to more than two models is considered in Section 2. This extension requires the calculation of the posterior probabilities of individual regression models, which is not as straightforward as the article implies. Section 3 considers the use of prior information in inference and experimental design. Sections 4 and 5 clarify some potentially misleading comments in the article on optimum design. In Section 6 a brief mention is made of extensions of T optimality to nonnormal models.