Analysis of Covariance in General Factorial Designs Through Multiple Contrast Tests Under Variance Heteroscedasticity

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Statistical Inference

Background:

• Analysis of Covariance (ANCOVA) is commonly used in clinical trials to test treatment effects adjusted for covariates.
• The standard ANCOVA F-test has limitations, including lack of information on individual hypotheses and strict assumptions like variance homoscedasticity.
• Existing methods may not adequately address variance heterogeneity or provide detailed insights into specific treatment effects.

Purpose of the Study:

• To extend existing methods for Analysis of Covariance (ANCOVA) to a multiple contrast test procedure (MCTP).
• To enable testing of arbitrary linear hypotheses, providing insights into both global and individual null hypotheses.
• To develop methods for calculating simultaneous confidence intervals for individual effects, even under variance heterogeneity.

Main Methods:

• Extension of Konietschke et al.'s method to a multiple contrast test procedure (MCTP).
• Derivation of a small sample size approximation using a multivariate t-distribution for the test statistic.
• Introduction of a Wild-bootstrap method as an alternative for statistical inference.

Main Results:

• The developed MCTP allows for testing of global and individual null hypotheses in ANCOVA.
• Compatible simultaneous confidence intervals can be calculated for individual treatment effects.
• Simulations demonstrate the applicability and robustness of the proposed methods, particularly in small sample size scenarios.

Conclusions:

• The proposed multiple contrast test procedure (MCTP) offers a flexible and powerful alternative to traditional ANCOVA methods.
• The methods are suitable for clinical trial settings, especially when dealing with variance heterogeneity and small sample sizes.
• The approach provides valuable information on individual treatment effects and their confidence intervals.