Confounding Equivalence in Causal Inference
This work addresses covariate selection and model testing in causal inference, offering a practical tool for researchers, but it is incremental as it builds on existing back-door criterion and Markov boundary concepts.
The paper tackles the problem of determining whether two sets of variables are equally effective at reducing bias in causal inference, providing a simple test based on causal diagrams that requires either both sets to be admissible or have identical Markov boundaries around manipulated variables.
The paper provides a simple test for deciding, from a given causal diagram, whether two sets of variables have the same bias-reducing potential under adjustment. The test requires that one of the following two conditions holds: either (1) both sets are admissible (i.e., satisfy the back-door criterion) or (2) the Markov boundaries surrounding the manipulated variable(s) are identical in both sets. Applications to covariate selection and model testing are discussed.