A1005
Title: From linear structural equation modeling to generalized multiple mediation formula
Authors: Sheng-Hsuan Lin - Institute of Statistics (Taiwan) [presenting]
Abstract: Causal mediation analysis is advantageous for mechanism investigation. In settings with multiple causally ordered mediators, path-specific effects (PSEs) have been introduced to specify the effects of certain combinations of mediators. However, most PSEs are unidentifiable. The interventional analogue of PSE (iPSE) is adapted to address the non-identifiability problem. Moreover, previous studies only focused on cases with two or three mediators due to the complexity of the mediation formula in a large number of mediators. We provide a generalized definition of traditional PSEs and iPSEs with a recursive formula, along with the required assumptions for nonparametric identification. The three major contributions are: First, we develop a general approach for causal mediation analysis with an arbitrary number of multiple ordered mediators and with time-varying confounders. Second, we demonstrate identified formula of iPSE is a general form of previous mediation analysis. It is reduced to a linear structural equation model under a linear or log-linear model, to causal mediation formula when only one mediator. Third, a flexible algorithm built based on the g-computation algorithm is proposed along with user-friendly software online. All methods and software contribute to comprehensively decomposing a causal effect confirmed by data science and help to disentangle causal mechanisms when multiple ordered mediators exist, which make the natural pathways complicated.