Causality: Interventions | Part A

Tutorial on causal inference, covering the basics of interventional reasoning. Topics: causal effect; Simpson's paradox; interventional distributions; causal effect rule; Markovian and semi-Markovian causal models; identifiability criteria such as backdoor, frontdoor and the do-calculus; and why the causal effect may not be identifiable. The computation/estimation of causal effect is discussed in both an idealized setting (complete model) and a practical setting (causal graph + observational data). 00:00 The Causal Hierarchy 03:37 Agenda: Idealized and Practical Settings For Causal Inference 07:17 Causal Effect: Interventional Probability 12:14 Computing Causal Effect Using Surgery 15:43 Why Bayesian Networks Are Not Enough To Compute Causal Effect 18:03 Why Observational Data Is Not Enough To Compute Causal Effect 19:58 Computing Causal Effect Using Do-Node 23:12 Truncated Formula of Interventional Distribution 27:22 Notation for Causal Effect 29:12 Computing Causal Effect Using Causal Effect Rule 31:57 Identifiability Of Causal Effect: Input-Output --- Slides available at: http://web.cs.ucla.edu/~darwiche/caus...