ABSTRACT
When discussing experimental design, many of the controversies surrounding Bayesian analysis abate. Most statisticians agree that although different approaches to statistical inference engender different design goals, “everyone is a Bayesian in the design phase,” even if only informally. Because no data have been collected, all evaluations are preposterior and require integration over both the data (a frequentist act) and the parameters (a Bayesian act). In the terminology of Rubin (1984), this double integration is a “Bayesianly justifiable frequentist calculation.” As such, experimental design shares a kinship with the other material in this chapter, and so we include it here.
