38.2 Rule of Thumb for Causal Inference

To validly estimate the causal effect XY:

  • Close all backdoor paths between X and Y (paths that start with an arrow into X). This removes confounding.
  • Do not block any part of the direct or indirect causal path from X to Y (especially via mediators).

This is known as the backdoor criterion—a formal condition from Pearl’s Causal Inference framework.