Latent-free equivalent mDAGs

Robin J. Evans
DOI: https://doi.org/10.2140/astat.2023.14.3
2023-06-12
Abstract:We show that the marginal model for a discrete directed acyclic graph (DAG) with hidden variables is distributionally equivalent to another fully observable DAG model if and only if it does not induce any non-trivial inequality constraints.
Statistics Theory
What problem does this paper attempt to address?