Causal Machine Learning

About

Participants of this course will learn and apply recent Causal Machine Learning methods to analyse effects of either experimental or observational interventions. Causal Machine Learning combines two mature fields in data analytics. On the one hand, the field of Machine Learning advanced our ability to detect correlational pattern in data, which is important to form high-quality predictions. On the other hand, the field of Causal Inference advanced our knowledge about how to assess the effects of interventions, which is essential for high-quality decision making. The promise of Causal Machine Learning is to deliver the best of both worlds to draw (more) reliable and more informative causal inference.
This course will focus on tools that are already mature in the sense that they are easy to implement for practitioners in the software R and covers three major topics:
Estimation of heterogeneous effects for experimental data
Estimation of average and heterogeneous effects for observational data
Policy learning from experimental or observational data
The final day will also discuss how these methods extend to other research designs and questions like difference-in-differences, instrumental variable and mediation analysis.
The course will be based on three pillars to teach the new methods: (i) lecture based introduction of the theoretical concepts, (ii) getting to know the methods with toy synthetic data in R notebooks that are provided by the lecturer, (iii) supervised application to provided or own datasets.

Course leader

Michael Knaus is Assistant Professor of Econometrics at the University of St. Gallen, Switzerland. Gabriel Okasa is a PhD candidate in econometrics at the University of St.Gallen, Switzerland.

Target group

Participants will find the course useful if they
- are familiar with the basics of causal inference and regression analysis and are curious how machine learning methods could enter their empirical toolbox.
- work with experimental and observational data in social science or related fields.
- want to be relieved from the decision whether age should enter linearly, as a quadratic, or as categorical variable in their regression models.

Prerequisites:
- Basic understanding of probability theory (conditional expectations) and regression analysis (OLS)
- Basic understanding of causal research designs, in particular randomized experiments and observational designs that control for confounding factors
- Basic experience with the software R
- (not required, but an advantage) Basic understanding of Machine Learning methods, in particular shrinkage methods (e.g., Lasso, Ridge) and tree based methods (regression trees, random forest)

Course aim

By the end of the course participants will:
- Understand popular methods that are likely to appear in future studies they consume.
- Know in which settings and for which research questions the current state of Causal Machine Learning provides attractive alternatives to standard tools.
- Be able to apply Causal Machine Learning in basic settings.
- Have the background knowledge to learn about Causal Machine Learning methods for more complex settings that are not covered in the course.

Fee info