Bayesian Causal Discovery Thresholds in Linear Gaussian Networks

Latent confounding triggers spurious edge detection in Bayesian causal discovery when correlation exceeds a specific threshold. This threshold decreases as sample size increases, causing the posterior to favor incorrect directed acyclic graphs in linear Gaussian models.