SELECTION OF POST-TREATMENT VARIABLES FOR ESTIMATING TOTAL EFFECT FROM EMPIRICAL RESEARCH

抄録

The selection of post-treatment variables for estimating a total effect is discussed when a causal diagram with a linear structural equation model for empirical research has unobserved variables. The front door criterion is utilized as a selection criterion of post-treatment variables for identifying a total effect. It is known that sets of post-treatment variables which satisfy the front door criterion are generally not unique. In this paper, the exact variance of a total effect using a set of post-treatment variables is obtained and the influence of a selection of post-treatment variables on the variance of the total effect is investigated.