Data-Driven Rank Tests for Classes of Tail Alternatives

摘要: Tail alternatives describe the frequent occurrence of a non-constant shift in two-sample problem with function increasing tail. The classes functions can be built up using Legendre polynomials. It is important to rightly choose number polynomials involved. Here this choice based on data, modification Schwarz's selection rule. Given data driven model, appropriate rank tests are applied. Simulations show that new work very well. While other for detecting as Wilcoxon's test may completely break down tail alternatives, have high and stable power. also higher power than unconstrained problem. Theoretical support obtained by proving consistency against large including all common alternatives. A simple but accurate approximation null distribution makes application easy.