A High-Dimensional Nonparametric Multivariate Test for Mean Vector.

Area of Science:

• Statistics
• Multivariate Analysis
• Nonparametric Statistics

Background:

• Classical Hotelling T^2 tests are inadequate for nonnormal, high-dimensional data, especially with heavy tails.
• Existing modified tests show poor power performance in genomics and quantitative finance applications.
• There is a need for robust statistical tests for high-dimensional mean vectors with nonnormal distributions.

Purpose of the Study:

• To propose a novel high-dimensional nonparametric test for the population mean vector.
• To analyze the test's theoretical properties, including its limiting null distribution and local power.
• To compare the proposed test's efficiency against modified Hotelling T^2 tests.

Main Methods:

• Development of a new nonparametric test for high-dimensional mean vectors.
• Theoretical analysis using modern probability theory to establish the limiting null distribution (normal).
• Comparison of local power and relative efficiency with existing modified Hotelling T^2 tests.

Main Results:

• The proposed test has a limiting null distribution that is normal under mild conditions, even when p >> n.
• The new test demonstrates substantial power gains compared to traditional nonparametric tests, especially for large p.
• Monte Carlo simulations confirm the finite sample performance, and an application to genomics data is shown.

Conclusions:

• The novel nonparametric test is effective for nonnormal, high-tailed high-dimensional data.
• It offers significant power advantages over existing methods in relevant fields like genomics.
• The test provides a valuable tool for analyzing complex, real-world datasets.