Tuesday, July 11, 2023 1 p.m. – 3 p.m | Lecture 1: Introduction to Potential Outcomes What distinguishes data from a randomized experiment from that arising in an observational study? This lecture introduces the notion of a potential outcome, explains the importance of randomization and describes the limits to identifying causal estimands. |
Wednesday, July 12, 2023 10 a.m. – 12:00 p.m. | Lecture 2: Non-Compliance and Instrumental Variables What inferences can be drawn from randomized studies in which some subjects do not comply with their assigned treatment? Building on the concepts introduced in the first lecture we show how, given certain assumptions, it is possible to derive bounds on causal effects. We will also outline simple approaches to Bayesian inference via transparent parametrizations. |
Thursday, July 13, 2023 1 p.m. – 3 p.m. | Lecture 3: Single-World Intervention Graphs (SWIGs) How can causal models based on potential outcomes be related to those based on directed acyclic graphs? In this lecture we provide a simple synthesis of the two main approaches to formulating causal models. This then leads to a simple reformulation of Pearl’s celebrated do-calculus, that we call the potential outcome (po) calculus. |