Lectures on Causal Panel Analysis
Yiqing Xu, Stanford University
Slides for six lectures of 100 to 120 minutes each on causal inference with panel data. The lectures are for applied researchers and graduate students in the social sciences who work with panel data. Each method is introduced through an example and a picture, and formulas appear only where the ideas need them.
| # | Lecture | Slides | Source |
|---|---|---|---|
| 1 | The Parametric Approach | 52 pp | code |
| 2 | Difference-in-Differences | 47 pp | code |
| 3 | Factorial Difference-in-Differences | 39 pp | code |
| 4 | Two-Way Fixed Effects Revisited | 33 pp | code |
| 5 | Modern DID | 48 pp | code |
| 6 | Synthetic Control and Extensions | 56 pp | code |
The page counts link to the PDFs, which have one page per slide. To present with the step-by-step reveals, build that version from the LaTeX source (make animated). “Code” opens the lecture’s folder on GitHub: the LaTeX source, the figures, and the R scripts that draw them.
The arc
Lectures 1 and 2 cover the parametric panel toolkit and the canonical difference-in-differences design. Lecture 3 extends that design to events that affect everyone (factorial difference-in-differences). Lecture 4 diagnoses three problems with two-way fixed effects: the design problem (what strict exogeneity implies for treatment assignment), violations of parallel trends, and the weighting problem when effects differ across units. Lecture 5 repairs the weighting problem with estimators that allow effects to differ, and works through three applications from a large reanalysis study. Lecture 6 turns to synthetic control, factor models, and doubly and triply robust estimators, which handle some violations of parallel trends.
The field is moving fast, and these slides lean toward the author’s own research and understanding of it. Read them as one path through the material rather than as a survey.
Software
The R packages used in the lectures, with tutorials: