Module 4: Causal Mediation Analysis


Dates - Venue - Description - Target audience - Exam
Course prerequisites - Teachers - Course material - Fees - Enrol


Monday and Thursday evenings: November 16, 19, 23, 26 and 30, December 3, 2015, from 5.30 pm to 9 pm (each lecture is followed by a hands-on practical session).


Faculty of Science , Campus Sterre, Krijgslaan 281, Building S9, Ghent.


This course aims to provide tools for estimating direct and indirect effects of exposure on outcome. It will introduce standard structural equation models as well as more flexible causal inference methods.

In observational epidemiology, psychology and sociology, there is substantive interest in separating direct exposure effects from indirect effects that are mediated through given intermediate variables. The use of mediation analysis has thus become quite common, especially in the social and psychological sciences, where a method based on regression analysis advocated by Baron and Kenny is now utilized routinely.

More recently, an approach to mediation arising from the causal inference literature and based on the notion of counterfactuals has been proposed. This has lead to an improved understanding of the conditions under which the standard regression approach to mediation analysis is valid and to novel techniques that enable effect decomposition of a total effect into a direct and indirect effect in much more general settings than those commonly considered. The newer techniques can handle nonlinear models, interactions, … and more appropriate adjustment for confounding.

In this course, we will first give an introduction to traditional mediation analysis under linear structural equation models with a single or multiple mediators, and we will discuss extensions to multilevel designs.

We will next introduce modern mediation analysis approaches based on so-called natural direct and indirect effects. Flexible state-of-the-art estimation techniques based on the mediation formula will be introduced, along with imputation strategies for a general class of natural effect models for mediation analysis.

Emphasis will be on concrete methods in an up-to-date survey of current status in the area.

Computer demonstrations in the freeware statistical software package R will be included, without assuming prior familiarity with the software.

Target audience

This course will benefit medical investigators, sociologists, psychologists, research scientists, clinical research associates, … who need to use statistical methods for mediation analysis or wish to develop a better understanding of techniques for confounder control.


Participants can, if they wish, take part in an exam. Upon succeeding in this test a certificate from Ghent University will be issued to participants with a university degree at the bachelor level or an equivalent degree.

To qualify for reimbursement from the UGent Doctoral Schools one must attend all classes and pass the exam. Visit the ICES website and your DS website for further information. Additional conditions and procedure.

Course prerequisites

Participants are expected to be familiar with the basic principles of statistical inference, linear and logistic regression analysis. Some familiarity with R is an asset.


Foto lesgeverProf. dr. Tom Loeys is Professor at Ghent University, Department of Data Analysis, Faculty of Psychology and Educational Sciences. His methodological interests include causal mediation analysis and dyadic data analysis. He also worked in the pharmaceutical industry, where he was involved in the design and analysis of clinical trials.


Foto lesgeverProf. dr. Beatrijs Moerkerke is Professor at Ghent University, Department of Data Analysis, Faculty of Psychology and Educational Sciences. She teaches courses in statistics and methodology to students in psychology and educational sciences. Her current research interests include causal mediation analysis and the analysis of brain imaging data.


Foto lesgeverProf. dr. Stijn Vansteelandt is Professor at Ghent University, Department of Applied Mathematics and Computer Sciences, Faculty of Science. He teaches courses in statistics to students in the Faculty of Science, the Faculty of Pharmaceutical Science and the Masters program in Statistical Data Analysis. He did postdoctoral research at the Harvard School of Public Health and Ghent University. His current research focuses mainly on mediation analysis and estimation of the causal effect of time-varying exposures in longitudinal studies.

Course material

  • A full set of slides will be provided.
  • The course will be based on current, accessible papers, and a full reading list will be available ahead of the course.
  • Recommended book (optional):
    VanderWeele, T.J. (2015). Explanation in Causal Inference: Methods for Mediation and Interaction, OUP, ISBN 978-0199325870.


    A different price applies, depending on your main type of employment.

    EmploymentModule 4BookExam
    Private sector/industry1 800 70 30
    Non-profit, government, university outside AUGent2 360 70 30
    (Doctoral)student outside AUGent2 280 70 30

    1 If three or more employees from the same company enrol simultaneously for this course a reduction of 10% on the module price is taken into account.

    2 AUGent-staff and AUGent doctoral students who pay through use of an SAP internal order/invoice can participate at these special rates.

    Enrol for this course