D03: Causal Mediation Analysis

Causal mechanisms are a fundamental component of social science theories. Researchers are not only interested in whether one variable causally affects another, but how such a causal relationship arises. Unfortunately, most existing research designs and empirical strategies merely aim at establishing causal effects between variables, resulting in a ‘black box’ approach to causality. As the black box approach has been criticized for being atheoretical and even unscientific, the primary goal of this course is to introduce participants to a set of concepts, methods, and algorithms that helps researchers empirically investigate the causal mechanisms underlying social science theories.

The course is divided into three parts. The first part involves a thorough discussion of the concepts and frameworks that are essential for understanding and conducting causal mediation analysis. The second part focuses on the research designs and methods that allow researchers to identify and estimate causal mediation effects. The final part of the course focuses on a wide variety of applications of these concepts and empirical strategies in the context of substantive research. Participants get to explore first-hand how the different methods covered in this course can be used to provide answers to research questions about causal mechanisms.

Day 1
On the first day, the class will begin with an introduction to causal inference under the Rubin causal framework, or potential outcome framework. Participants are introduced to concepts such as potential outcome, independence and average treatment effect (ATE) that are essential to understand causal inference and causal mediation. With these foundations, we will then define and conceptualize causal mechanism as a decomposition of the causal effect into a direct effect and an indirect effect, or causal mediation effect. We then move on with the assumptions required for the identification of average causal mediation effect (ACME). We will discuss in detail the idea of Sequential Ignorability, the key assumption to identify ACME. With the understanding of Sequential ignorability, we shall see why the causal chain approach – despite being intuitively appealing – is fallacious for causal mediation analysis.

Day 2
Focuses on process tracing, a qualitative method for studying causal mechanisms within a single case. Since social scientists often complement quantitative or formal analyses with case studies, our goal here is to learn how to conduct case studies more scientifically for the purpose of testing proposed causal mechanisms. We will begin with the definition of equifinality, or the existence of alternative paths through which the same outcome could have occurred. We then introduce different variants of process tracing. Participants will see that this method has its foundation rooted in the Bayesian logic, which gives rise to two key concepts, uniqueness and certainty. Different levels of uniqueness and certainty are associated with four types of tests for possible causal mechanisms. We then move on to discuss case selection strategies for the different variants of process tracing. If possible, a guest speaker is invited to present examples of process tracing, allowing participants to see how the method works out in real research contexts.

Day 3
Focuses on the experimental approach to causal mediation analysis. Participants will see why the standard experimental design is insufficient for identifying the ACME. We will then discuss several designs such as the parallel design that make Sequential Ignorability plausible. The key difference here is that experimentalists need to manipulate both the treatment and the mediator for the identification of ACME. Examples of social science experiments are used to illustrate the key ideas that underline each of the designs. Participants will also have the opportunity to design their own experiments.

Day 4
Introduces the estimation strategies. We begin by introducing the Baron and Kenny procedure and the structural equation framework for estimating causal mediation effects. We will discuss the assumptions of linear structural equation modeling (LSEM) and see that a major limitation of the LSEM is that it cannot be directly applied to mediators and outcomes that are measured with discrete (binary or ordered) variables. To overcome this limitation, we will study a recently developed general algorithm based on nonparametric identification that can be applied to any type of data. Finally, considering that Sequential Ignorability is an irrefutable assumption, we will introduce sensitivity analysis that formally quantifies the extent to which the assumption has to be violated before the original conclusion is reversed. 

Day 5
On the last day, we will put what we have learned into practice in a practical session, whereby participants will learn to use the “mediation” package in R to estimate ACME and conduct sensitivity analysis. We will conclude the course with a “Last-day Seminar” where participants who take the course for credits will present their research or research proposals and receive feedback from the instructor and their peers on research design and data analysis. Participants who are not taking the course for credits may also volunteer to present their work. 

Readings (indicative only)

Imai, Kosuke, Luke Keele, Dustin Tingley, and Teppei Yamamoto. “Unpacking the Black Box of Causality: Learning about Causal Mechanisms from Experimental and Observational Studies.” American Political Science Review 105, no. 4 (2011): 765-789.

MacKinnon, David. Introduction to Statistical Mediation Analysis. Taylor & Francis Group. 2008.

Beach, Derek, and Rasmus Brun Pedersen. Process-Tracing Methods: Foundations and Guidelines. University of Michigan Press, 2013.

Imai, Kosuke, Luke Keele, and Dustin Tingley. “A General Approach to Causal Mediation Analysis.” Psychological Methods 15, no. 4 (2010): 309-334.

Pearl, Judea. “Interpretation and Identification of Causal Mediation.” Psychological Methods 19, no. 4 (2014): 459-481.

Prerequisite knowledge

This course builds on and makes use of several fields and techniques in social science methodology. Students should possess basic knowledge and skills in some – if not all – of the following areas: 1) causal inference and the Rubin causal framework; 2) experimental methods; 3) regression analysis (and preferably structural equation modeling), 4) instrumental variable method, 5) case study and process tracing, and 6) basic skills in R. Students are not discouraged from signing up for this course if they do not fully meet the above criteria. However, they are strongly encouraged to take the relevant courses offered by the Summer School before or concurrently with this course.

This description is subject to change at the discretion of the Instructor

2 Credits

For completion of all work before and during the course, as outlined by the Instructor, and 90% participation and attendance of the course.

2 Additional Credits

Course specific extra assignments. These can include submitting assignments before the course, daily assignments, and/or a final assignment to be completed after the course as decided by the Instructor.