Testing the Hypothesis of Independence Between Two Sets of Variates.

Area of Science:

• Statistics
• Statistical Hypothesis Testing
• Data Analysis

Background:

• Assessing independence between sets of variables is crucial in statistical analysis.
• Existing methods, like the likelihood ratio test, have known limitations in controlling Type I errors.
• Robust statistical measures offer potential improvements for hypothesis testing.

Purpose of the Study:

• To compare five methods for testing the hypothesis of independence between two sets of variates.
• To introduce and evaluate two new robust covariance-based methods.
• To assess the performance of these methods, particularly concerning Type I error control.

Main Methods:

• Comparison of five distinct statistical tests for independence.
• Development of two new methods based on robust covariance measures.
• Monte Carlo simulations to evaluate method performance under various conditions (sample size, normality).

Main Results:

• The two new robust methods demonstrate accurate control of Type I error rates, even with small sample sizes and non-normal data.
• The traditional likelihood ratio test performs unsatisfactorily, with Type I error control potentially worse than previously reported.
• The new procedures are effective in reflecting linear association, offering an alternative to rank-based methods.

Conclusions:

• The novel robust covariance-based methods are reliable alternatives for testing independence.
• These new methods provide superior Type I error control compared to the likelihood ratio test.
• The findings highlight the utility of robust statistics in hypothesis testing scenarios.