Causal inference
Contents
Causal inference is the goal of many empirical studies in the health and social sciences. In a causal analysis, a causal parameter is formally defined and the underlying assumptions are explicitly stated. These assumptions clarifies the differences between experimental data and observational data for making causal conclusions. In the course we focus on two dominating approaches for causal analyses, i) the potential outcomes framework and ii) structural analyses (graphical models). We learn the foundations for these two frameworks as well as methods for estimation of causal effects in experimental and observational studies. Data-examples are given throughout the course and practical problem solving and data analyses are carried out in parallel with the theoretical material.
Expected learning outcomes
Knowledge and understanding
After a completed course the student should be able to show:
- an in-depth knowledge of the potential outcomes framework and use of directed acyclic graphs for causal inference;
- an in-depth understanding of assumptions underlying causal analyses with experimental and observational data;
- an in-depth knowledge of the presented methods to draw causal inference.
Skills and abilities
After a completed course the student should be able to:
- apply parametric and non/semi-parametric estimators (including using machine learning to estimate nuisance functions) of causal effects and understand the corresponding underlying assumptions;
- perform causal analyses with existing software.
Ability to evaluate and approach
After a completed course the student should be able to:
- evaluate the plausibility of underlying assumptions in experimental and observational studies;
- evaluate the sensitivity of a causal effect estimate.