摘要:The aim of this paper is to give an extended front door criterion and a mixed door criterion as new identifiability criteria for causal effects of a joint intervention when causal diagrams for empirical research have some unobserved variables. The criteria enable us to recognize from the structure of the graph whether the causal effect can be expressed by a subset of observed variables. If the answer is affirmative, the criteria provide a closed-form expression for the effect of the external joint intervention.
关键词:causal diagram;d-separation;inference rule;mixed door criterion;recursive factorization