On Bayes and Nash experimental designs for hypothesis testing problems (Q2209838)

This paper deals with the problem of designed experiments from three different point of views: classical inference, Bayesian inference, and game theory. When the focus of the statistical analysis is not parameter estimation but hypothesis testing, the alphabetical optimality criteria may not the appropriate selection strategies. In such a case maxi-min strategies are usually preferred. Focusing in particular on the ANOVA setting, the authors analyze the relationships between maxi-min designs and Bayesian designs. Under the perspective of game theory, the statistical experiment is viewed as a two-player game, where the first player (the Nature) chooses the value of the unknown parameters, and the second player (the Statistician) chooses the design. The Nash equilibrium coincides with the maxi-min design.