Causal inference at the interface of statistics, study design, and epistemology, is the science of learning effects from data and background knowledge. Fundamentally important across multiple domains, including epidemiology, economics, political science, psychology, clinical science, engineering, and social sciences, there is increasing interest in the teaching of causal inference theory and statistical methods within many academic units.

Different from more traditional statistics, causal inference fundamentally relies on non-statistical assumptions in order to make inferences. Entire systems of notation and graphical conventions have been developed to produce the framework within which statistical analysis can be planned. In addition, a vast literature of statistical (or so called “causal”) methods have been developed to address the purely quantitative components of causal inference. Therefore, relaying the basic ideas of causal inference in relatively simple terms may seem like a daunting task.

In this workshop, we will outline and explain the elements of causal inference that we teach and have found to be the most relevant for an advanced undergraduate or graduate-level course, and the exercises that accompany them. We will focus on explaining how these elements are interconnected and give a global view on how causality can be addressed in study planning and analysis. These elements include

- Counterfactual theory, notation, and parameters,
- Identifiability of causal/counterfactual parameters via statistical estimands,
- Directed acyclic graphs and their role in identifiability,
- The interface between study design (e.g. randomized controlled trials, pseudo-experimental studies, observational studies, target trial emulation) and parameter identifiability,
- The targeted learning roadmap,
- Statistical estimation for counterfactual parameters, including marginal structural models (Regression, inverse probability of treatment weighting, G-computation),
- Semiparametric frameworks (including targeted maximum likelihood estimation) and machine learning integration,
- Alternative identifiability through instrumental variable methods,
- Overview of more advanced topics such as mediation analysis and longitudinal treatments.

We will also include discussion on how to target your course material to your audience and some approaches to evaluation.

Prerequisites of workshop: Interest in causal inference, understanding of basic statistical theory and methods, generalized linear regression.