Distinguishability of causal structures under latent confounding and selection
This research introduces a Selected-Marginalized Directed Graph (smDG) to classify causal structures when data has latent confounders and selection bias. It proves that two causal graphs are indistinguishable if they share the same smDG, meaning they yield the same results under both intervention and passive observation. This provides a new tool for determining structural equivalence in complex observational data.