An Introduction to Potential Outcomes, DAGs and Single-World Intervention Graphs
summary
Introduces a concrete procedure that turns a DAG into a single-world intervention graph by node-splitting, giving an explicit factorization and a modularity rule that maps graph terms to potential outcomes; derives back-door and front-door formulas by reading d-separations on SWIGs; replaces Pearl’s third do-calculus rule with a simpler “no directed path” check; shows the method avoids making untestable cross‑world independences that nonparametric structural equations implicitly impose.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 4:20Potential Outcomes
- 6:29Splitting
- 14:15SingleWorld Intervention
- 19:30The Nature of SingleWorld Intervention
- 22:23SingleWorld Intervention Example
- 26:09SingleWorld Intervention Summary
- 26:56Modularity Assumption
- 27:33Dynamic Interventions
- 31:19Key Point
- 32:34Confounding
- 38:52Labeling
- 44:34Allow 3 transformations
- 53:11Simplifying
- 54:36Concrete Example
related talks
12mTowards an Equational Calculus of Interventions
29mSome Staged Tree Models For Learning From Interventions
24mA Friendly Introduction To Causality
31mIntroduction to Causal Inference using pgmpy
44mMarginal DAG Models Equivalent to Another DAG
41mOn the Path to Causal Inference
59mA Welcome to Causal Probabilistic Programming
01h 00mLogic and Probabilistic Circuits 3