FCI theory

The FCI algorithm (Spirtes et al., 2000, Ch. 6) drops the causal sufficiency assumption required by PC, returning a partial ancestral graph (PAG) over the observed variables. The PAG is a compact representation of the Markov equivalence class of maximal ancestral graphs (MAGs) compatible with the observed CI relations.

Note

This page is currently a stub. The full Background / Assumptions / Algorithm / References treatment, parallel to PC theory, will land in v0.x.x.

References

• Spirtes, P., Glymour, C., & Scheines, R. (2000). Causation, Prediction, and Search (2nd ed.). MIT Press, Chapter 6.

• Zhang, J. (2008). On the completeness of orientation rules for causal discovery in the presence of latent confounders and selection bias. Artificial Intelligence, 172(16-17), 1873–1896.