Causal Analysis with Observational Data

About

Workshop Contents and Objectives

Does smoking cause bad health? Does income inequality increase political extremism? Do schools increase inequality? Many questions of interest to social scientists are causal. This course provides an introduction to modern methods of causal inference using observational data. Building on the potential outcomes framework to causality the course discusses natural experiments, instrumental variables, difference-in-differences (DID), different types of fixed effects models, and regression discontinuity designs (RDD). All these methods allow researchers to control for unobserved variables and therefore to identify causal effects using observational data. The course also provides an introduction to Directed Acyclic Graphs (DAG), which allows us to graphically depict causal relationships.

Workshop design

The course provides both a sound understanding of each method as well as practical exercises to implement these methods using R and Stata.
There will also be plenty of time to discuss research projects and ideas related to the methods of the course by the participants. Participants are very much encouraged to apply the methods taught in the course to their own research questions.

The workshop has three aims:

To introduce you to each method.
To learn how to implement each method in R and/ or Stata and how to interpret its results for a research paper.
To discuss how to justify and criticize the use of each method in the analysis of your own research and published journal articles of other researchers.
Each of these three aims is covered each day by one of three sessions. We start each day in the morning (8:30–10:00) with a lecture about the methodological background to each method. After a short break, we continue with the implementation of each method in R and/ or Stata and the interpretation of the output (10:30–12:00). After lunch, we discuss how to justify and to criticize the use of each method (13:00–15:30/16:00). This is done by the analysis of both published articles and your own research. You’re very welcome to think about how to implement each method within your own research in this part of the course!

Detailed lecture plan (daily schedule)

Day 1.
The counterfactual framework of causality (including the potential outcomes framework) and Directed Acyclic Graphs (DAGs).

Day 2.
Sibling and twin fixed effects.

Day 3.
Difference-in-differences.

Day 4.
Instrumental variable estimation.

Day 5.
The regression discontinuity design.

On the first day the idea is that you present your own research (a paper, a research question from your thesis, etc.) in the form of a DAG to the other participants in the afternoon. Do not worry about it now, you’ll have plenty of time to think about how to do this on the first day of the course.

Prerequisites

Some elementary knowledge of regression analysis, in particular linear regression, will be necessary to be able to fully follow the content of the course. Statistical analyses will be conducted with R and Stata. A general knowledge of one of these languages will be necessary to implement the practical exercises, as there won’t be the time to learn basic commands. Participants can conduct all exercises in R or Stata, according to their own preferences.

Course leader

Michael Grätz is a SNSF professor in sociology at the University of Lausanne. He is also an associate professor (docent) at the Swedish Institute for Social Research (SOFI), Stockholm University.

Target group

graduate student, doctoral researchers, early career researchers, experienced researchers

Fee info