Confidence procedures for two-sample problems

In the usual two-sample problem one is estimating the constant additive effect of a treat ment or the additive constant by which two random variables differ. However, if the treatment-effect may depend on the response level, then a more general approach to two-sample problems seems appropriate. We deflr a treatment-effect function t and characterize distribution-free confidence bounds for the function t both in the case where t has specified parametric forms and in the case where t is not parameterized.