Nonparametric tests for testing the cotrol versus effects of treatment levels

Abstract

Two tests are introduced to test whether or not at least one of the different levels of a treatment cause a significant increase in the mean or variance over the control. The two new tests are compared with two existing tests on the basis of estimated powers for three different types of underlying distributions (symmetric and skewed). It is found that the proposed tests do have higher powers than the existing tests when there is a change in both the scale and location parameters. The results as to which of the proposed tests has higher powers depend on whether the underlying distributions are symmetric or skewed.