摘要
In this paper, the concept of distribution effect is proposed without the causal diagram. Following the notation of Stone [11], we assume that the exposure treatment X is an unknown deterministic function of the confounder set Pa(X) and a random error ε. We discuss sufficient and necessary conditions for homogeneity, collapsibility and nonconfounding for distribution effects and discuss relations among them.
In this paper, the concept of distribution effect is proposed without the causal diagram. Following the notation of Stone [11], we assume that the exposure treatment X is an unknown deterministic function of the confounder set Pa(X) and a random error ε. We discuss sufficient and necessary conditions for homogeneity, collapsibility and nonconfounding for distribution effects and discuss relations among them.
基金
Supported by the NSFC (10801019)