Causal Inference

31 May - 1 June 2021

Elisa Volpi

Florian Foos, London School of Economics & Political Science

Information: "Correlation is not causation". You must have heard this warning many times. But, what thenis causation, and how can we identify the causal eect of political messages, policies or programmes? Thismodule introduces students to the ideal of the randomized experiment, and other common methods of causal
inference using quasi-experimental data.

Aims: The module provides an introduction to the design-based approach to causal inference. Instead ofusing fancy modelling approaches to correct post-hoc for potential biases, the module encourages students tothink about challenges to causal inference at the design stage of a study. Published work will be evaluated
based on how it addresses three key assumptions underlying causal inference: independence, excludability, and
non-interference. The module will cover identication strategies for a variety of causal research questions in
Political Science. After introducing participants to the design-based approach to causal inference in general, the
module will cover the Regression Discontinuity Design, the Dierence-in-Dierences Design and the Synthetic
Control Method, three methods that allow design-based inference with quasi-experimental data, in detail. The
module is taught as a combination of lectures and applied computer labs in R (Stata code will be provided).

Prerequisites: The only pre-requisite is any course covering linear regression. There is relatively little assumed knowledge, and the aim is to build the statistical foundations from the ground up. If you have conducted a hypothesis test of any kind, you probably have the requisite skills.

Learning Outcomes: Participants will understand the potential outcomes framework, and the key assump-tions underlying causal inference, and will be able to choose appropriate methods for a variety of researchquestions posing dierent identication challenges. Moreover, they will gain the practical skills of applying
these insights and the statistical knowledge to their own research ideas.

Course Dropbox: Some readings and code les are available from the course dropbox.

Required textbook: Cunningham, Scott. Causal Inference: The Mixtape, Yale University Press, 2020.

Recommended textbooks:

• Angrist, Joshua and Joern-Steen Pischke. Mostly Harmless Econometrics: An Empiricist's Companion, Princeton: Princeton University Press, 2009.
• Bueno de Mesquita, Eitan, and Anthony Fowler. Thinking Clearly in a Data-Driven Age, 2019: available via course dropbox.
• Dunning, Thad. Natural Experiments in the Social Sciences. A Design-Based Approach, Cambridge: Cambridge University Press, 2012.
• Gerber, Alan and Donald P. Green. Field Experiments: Design, Analysis, and Interpretation, New York: W.W. Norton, 2012.
• Imbens, Guido W. and Donald B. Rubin. Causal Inference for Statistics, Social and Biomedical Sciences: An Introduction. Princeton: Princeton University Press, 2015.

Software: Participants will have a choice between using R (demonstration in class) and Stata (code will be
provided).

Workshop Outline

1. Introduction to causal inference
2. Potential Outcomes Framework and ATE
3. Regression Discontinuity Design (RDD): Foundations
4. Regression Discontinuity Design (RDD): Application
5. Dierence-in-Dierences Design (DiD): Foundations
6. Dierence-in-Dierences Design (DiD): Application
7. Synthetic Control Method (SCM): Foundations
8. Synthetic Control Method (SCM): Application
9. Advice on students' research designs

ZOOM !

15