Causal Inference

Lecturer: Jan Beyersmann

Exercises: Alexander Stemke

General Information

Lectures:

Exercises:

Exam (open):

Language:

Audience:

Tuesday, 10:00 - 12:00 Uhr / Room 120 (Heho 18)

Wednesday, 10:00 - 12:00 Uhr / Room E20 (Heho 18)

TBA

English, unless all students have sufficient knowledge of German

Master students from one of the mathematical programs. The course counts as "Fortgeschrittene Methoden der Mathematischen Biometrie B" for students of Mathematical Biometry. Causal Inference is also a key field in statistical Data Science and students with at least a vague interest in data analyses are more than welcome. Students should at least have taken a class on elementary probability theory and statistics in order to follow this course.

Contents:

There are two reasons for a statistical analysis. One is prediction of future data based on what one has learned from past data and accounting for uncertainty. Prediction need not be concerned with understanding cause-effect relationships, but understanding causality is central to our understanding of data and how we use that knowledge. For instance, standard statistical techniques allow to predict the survival probability of a current smoker, perhaps predicting later death compared to non-smokers based on smoking status alone - something we will understand in this course using causal graphs. But there is no standard statistical technique that analyses the causal effect of smoking on mortality. The difficulty is that smoking is not assigned in a randomized experiment, and there are more differences between smokers and non-smokers than just smoking status. In fact, defining a causal effect is not even part of the usual statistical and mathematical formalism. In the last 30 years or so, there has been a statistical revolution of developing causal inference, motivated by practical needs as in the search for effective HIV treatments. The aim of this lecture is to introduce students to this groundbreaking new field which is also currently shifting data science paradigms from "predicting" to "understanding".

Literature: