Nonparametric inference in factorial designs with censored data

Area of Science:

• Statistics
• Nonparametric Methods
• Experimental Design

Background:

• Factorial designs are crucial for analyzing complex relationships between multiple factors.
• Testing hypotheses of no main effects and no interaction is fundamental in factorial analysis.
• Existing methods can be limited, especially when dealing with censored data.

Purpose of the Study:

• To propose a novel nonparametric method for testing main effects and interactions in factorial designs.
• To develop an extension of the Hodges-Lehmann estimator for censored data within this framework.
• To evaluate the performance and applicability of the proposed method.

Main Methods:

• Formulating hypotheses of no main effects and no interaction as a vector of contrasts.
• Utilizing nonparametric estimation of these contrasts, analogous to two-sample location difference estimation.
• Introducing an extended Hodges-Lehmann estimator for censored data, focusing on computational simplicity and variance evaluation.
• Conducting a simulation study for a two-by-two design and analyzing a real-world three-way layout with censoring.

Main Results:

• The proposed method offers a viable nonparametric approach to hypothesis testing in factorial designs.
• The extended Hodges-Lehmann estimator is computationally efficient and allows for straightforward variance assessment.
• Simulation results and real data analysis demonstrate the method's performance, particularly with censored observations.

Conclusions:

• The developed nonparametric method provides a flexible and robust alternative for analyzing factorial designs, especially under censoring.
• The extended Hodges-Lehmann estimator is a valuable tool for nonparametric inference with censored data in complex experimental designs.
• This approach enhances the ability to test key hypotheses in various scientific fields utilizing factorial designs.