A0289
Title: A robust combined nonparametric method for comparing two locations
Authors: Marco Marozzi - Ca' Foscari University of Venice (Italy) (Italy) [presenting]
Abstract: Traditional methods for comparing two samples are based on sample means, which are very non-robust estimators for population means. Non-normality and different variances adversely affect the size and power of traditional tests. Trimmed means are robust and have high efficiency relative to population means. The Yuen test is a familiar method based on trimmed means. There is no agreement about the preferable trimming rate. The power of several Yuen tests with different trimming rates is studied. In general, there is no test being uniformly best. Different tests are sensitive to different features of the data. Therefore, no test can be expected to perform well in all possible situations. In fact, the various Yuen tests show very different power for different distributions because the best trimming rate depends on distribution tail weight. In this context, combined testing is appealing because it aims at providing a test, based on the combination of several tests -each with at least one good feature- that inherits the good features of the single tests being combined. The combined test is expected to perform well in many different situations, with a power that is always larger than the least powerful single test and very often similar to the most powerful single test. Therefore, a bootstrap test based on the combination of several Yuen tests is presented. It is shown that the combined test is powerful irrespective to the underlying distribution.
