Math for Causal Inference
All simulators

Single-World Intervention Graphs

Split the treatment node, relabel potential outcomes, and read ignorability straight off the graph.

Day 60: Phase 4 Capstone: DAGs Meet Potential Outcomes (SWIGs)
ORIGINAL DAG
LAMY
SWIG FOR do(A = a)
LAaM(a)Y(a)
Y(a) ⫫ A | L: holds (d-separation in the SWIG).
A ← L → Y(a) — blocked at L

The SWIG splits the treatment into its natural value A (keeps incoming arrows) and the value we set, a (keeps outgoing arrows); everything downstream becomes a potential outcome. Click nodes to condition on them. Ignorability Y(a) ⫫ A | L holds exactly when every path between A and Y(a) is blocked. Dashed nodes are unobserved.