partial ancestral graph (PAG)
Definition
Partial ancestral graph (PAG) is a graphical representation used in constraint-based causal discovery to depict causal structures in cyclic causal models when the presence of unobserved confounding variables prevents the determination of complete causal relationships from observational data. PAGs employ specific edge markings to indicate different types of causal relationships and uncertainties, allowing researchers to represent both directed causal edges and ambiguous relationships. In cyclic causal discovery contexts—particularly relevant for psychological phenomena characterized by feedback loops between variables—PAGs provide a way to visualize the output of discovery algorithms that cannot fully resolve causal direction or the presence of latent confounders from observational data.
Sources: Park et al. (2024)
Related Terms
Applications
Partial Ancestral Graph (PAG) and Cyclic Causal Discovery
PAGs are generated as output from constraint-based cyclic causal discovery algorithms, which infer causal structures from patterns of statistical (in)dependence in observational data while accommodating feedback loops between variables.
Sources: Park et al. (2024)
Partial Ancestral Graph (PAG) and Unobserved Confounding
PAGs explicitly represent uncertainty arising from unobserved confounding variables by using specific edge markings to distinguish between different types of relationships. This graphical approach allows researchers to communicate findings about causal structure when analyzing observational data in psychological research.
Sources: Park et al. (2024)
