Introduction to Causal Inference

Identification, potential outcomes, and causal graphs

Chad Hazlett · Academia Sinica ISSM 2026 · July 16–17

This two-day course concentrates on the foundational idea behind all causal inference: causal identification, and how it differs from an ordinary statistical problem. A causal claim always concerns the relationship between quantities we can observe and quantities we can never observe, and for this reason it can never be settled by data alone — we must always grapple with assumptions. The course introduces two complementary notational systems that let us state these assumptions and challenges with mathematical precision: the potential outcomes model and graphical/structural causal models. We put both to work in the context of two central strategies for making causal claims: randomized experiments, and the attempt to recover causal claims from observational data through controlled comparisons under the assumption of no unobserved confounding. Throughout, students rehearse and consolidate these ideas with pencil-and-paper worksheets completed in class.

Materials

Session	Topic	Slides	In-class worksheets
Day 1 July 16	Potential outcomes and ignorability	Slides (PDF)	WS1: Potential Outcomes (PDF) WS2: Experiments & Ignorability (PDF)
Day 2 July 17	Causal graphs and what they reveal	posted after the session	posted after the session

Slides are posted as handouts (one page per slide). Worksheets are meant to be worked by hand in class; blank copies are linked here for reference.