Selection of Identifiability Criteria for Total Effects by using Path Diagrams
This work is incremental, as it applies known methods to improve accuracy in causal inference for researchers in statistics and machine learning.
The paper compares three existing identifiability criteria for total effects—back door, front door, and conditional instrumental variable—in terms of asymptotic variance to determine which is superior based on graph structure, aiming to increase estimating accuracy.
Pearl has provided the back door criterion, the front door criterion and the conditional instrumental variable (IV) method as identifiability criteria for total effects. In some situations, these three criteria can be applied to identifying total effects simultaneously. For the purpose of increasing estimating accuracy, this paper compares the three ways of identifying total effects in terms of the asymptotic variance, and concludes that in some situations the superior of them can be recognized directly from the graph structure.