38.4 Causal Discovery

Causal discovery involves algorithmically identifying causal relationships from data under a set of assumptions (like faithfulness and causal sufficiency). Key algorithms include:

  • PC algorithm: Constraint-based, uses conditional independence testing
  • GES (Greedy Equivalence Search): Score-based method
  • FCI (Fast Causal Inference): Extends PC to handle latent confounders

See (Eberhardt, Kaynar, and Siddiq 2024) for a comprehensive discussion on the assumptions and limitations of discovery algorithms in practice.

References

Eberhardt, Frederick, Nur Kaynar, and Auyon Siddiq. 2024. “Discovering Causal Models with Optimization: Confounders, Cycles, and Instrument Validity.” Management Science.