Overcoming Model Uncertainty - How Equivalence Tests Can Benefit From Model Averaging.

Area of Science:

• Biostatistics
• Clinical Trials
• Statistical Modeling

Background:

• Equivalence testing is crucial for comparing effects across patient groups in clinical trials.
• Classical methods focus on single quantities, which can be inaccurate when covariate-dependent differences exist.
• Existing approaches often assume known regression models, leading to potential errors under misspecification.

Purpose of the Study:

• To develop a flexible equivalence testing method that overcomes the assumption of known regression models.
• To introduce model averaging using smooth Bayesian information criterion weights for robust statistical inference.
• To propose a hypothesis testing procedure leveraging the duality with confidence intervals.

Main Methods:

• Utilized model averaging with smooth Bayesian information criterion (BIC) weights to handle model uncertainty.
• Developed a testing procedure based on the duality between confidence intervals and hypothesis testing.
• Employed a simulation study to validate the proposed methodology.

Main Results:

• The model averaging approach demonstrated improved accuracy and reliability in equivalence testing.
• The proposed method effectively addresses issues arising from regression model misspecification.
• The approach was validated through simulations and a practical case study.

Conclusions:

• Model averaging provides a flexible and robust solution for equivalence testing under model uncertainty.
• The proposed method enhances the applicability and accuracy of comparing regression curves across different groups.
• This approach offers practical relevance for analyzing complex biological and clinical data, such as toxicological gene expression.