|
Abstract:
|
Mediation analysis provides an attractive tool for investigating causal pathways, specifically, assessing whether an exposure influences an outcome through an intermediate variable. However, testing the mediation effect is challenged by the fact that the null hypothesis of interest is composite. As a result, the classic and common mediation tests can be overly conservative under certain cases, which may yield invalid inference conclusions; this problem has gained increasing attention in recent genome-wide mediation studies. To tackle this issue, this work develops an adaptive test that can adapt to the various types of the composite null hypothesis. This new method would achieve the correct type I error control and also improve the statistical power.
|
