An Introduction to Potential Outcomes, DAGs and Single-World Intervention Graphs

on YouTube ↗01h 08m

" summary

Shows how a randomized encouragement design with noncompliance (binary Z, X, Y) becomes multiple observational studies, builds one polytope per Z level and takes their intersection of polytopes to get tight bounds on the average causal effect (example produces ACE bounds ≈ 0.39–0.78), derives the linear falsification inequalities that detect direct Z→Y effects, and contrasts a naive Dirichlet/MCMC approach (extremely prior‑sensitive) with a reparameterization on the observed distribution for more stable Bayesian or frequentist inference.

AI-written summary · about

" chapters (12)

1. 0:00Start
2. 0:17Interactive Plots
3. 4:07Summary
4. 5:14Instrumental Variables
5. 22:37Encouragement Design
6. 25:27Exclusion Assumption
7. 31:06Previous Results
8. 44:40Falsification Test of the Binary Iv Model
9. 50:33Two Approaches to Bayesian Inference
10. 59:13Average Causal Effect
11. 1:02:34Bayesian Approaches
12. 1:03:53Inference for the Observed Distribution

" related talks

• 24mA Friendly Introduction To CausalityAlex Dimakis · MLconf SF 2016
• 41mOn the Path to Causal InferenceMark Farragher · PyData London 2019
• 49mCausal Inference in Complex Systems: Network Interference, Strategic Agents, and BeyondPanos Toulis · Quantifying Uncertainty: Stochastic, Adversarial, and Beyond
• 30mCausal Estimands And InterventionsVanessa Didelez · Learning from Interventions 2022
• 44mCausal ReductionsJoris Mooij · Quantum Physics and Statistical Causal Models (Simons Institute workshop, 2022)
• 31mIntroduction to Causal Inference using pgmpyAnkur Ankan · PyData Amsterdam 2024
• 31mA Gentle Introduction To Causal InferenceEyal Kazin · PyData Global 2022
• 41mKeynote: Personalization, Bandits, and Causal inferenceMaurits Kaptein · PyData Eindhoven 2019