Power consideration and caveats of Causal Inference Test (CIT) for mediation analysis

Abstract

Causal Inference Test (CIT) has been proposed to infer the causality mediation from a genotype to a trait through a candidate mediator, for example certain gene expression. This thesis aims to study the performance of CIT under diverse settings including partial mediation where the temporal order of the candidate mediator and the trait cannot be determined. Through algebraic derivation and simulation, I show that there are scenarios of partial mediation that CIT will yield false negative results even for a large sample size, and a scenario where CIT will lead to false positives. In conclusion, CIT may not reliably detect the direction of the mediation.