Permutation Tests for Two-Sample Means of High-dimensional Data

The generation of high-dimensional data has led people to research permutation tests. Permutation tests are widely used in practice, but unduly strong conditions are often required for its validity. In this paper, we use a modified statistic which calls for only marginal standardization, and calculate the test statistics using pseudo samples that are generated through the Bootstrap method. We show that the corresponding permutation test is consistent under mild conditions. By comparing with some existing methods are made by simulation, the use of permutation tests for two sample means in high dimensional settings has better performance.