The segment method as an alternative to minimax in hypothesis testing

Abstract An alternative approach to the design of hypothesis tests within modeling uncertainty is proposed. This approach is based on a design technique known as the segment method, which has been established previously as a favorable alternative to the minimax concept in the context of optimal control. The general segment concept is discussed here as it applies to binary decisions made in the presence of unknown prior distribution or parametric uncertainty. In the latter context, several aspects of the design of hypothesis tests on Gaussian data are considered in detail.