ABSTRACT
As has been repeatedly emphasized in this book, all statistical analyses depend on
choices made by the statistician. At a minimum, a model { fθ : θ ∈ Θ} is chosen. It has been argued that more than this is needed to be unambiguous about how to
measure statistical evidence concerning various questions of interest. In particular, a
proper prior pi on Θ needs to be chosen. A natural question is then: how are these ingredients to be selected? Certainly this shouldn’t be done arbitrarily, as that would
undermine the validity of the subsequent analysis. So part of this chapter is concerned
with this aspect of a statistical analysis.