Correlation does not imply causation. This insight is crucial for any data analyst who aims to inform decisions. This course introduces two frameworks to think about causality: (i) potential outcomes that are more prevalent in economics and statistics, and (ii) directed acyclic graphs (DAGs) that are more prevalent in computer science and industry. These frameworks are applied to understand how causal effects can be identified in experiments (A/B testing), natural experiments (instrumental variables, difference-in-differences, regression discontinuity), and more complex causal structures as well as how to discover causal structures. Application are run in R. In combination with some general machine learning module like "DS405 Machine Learning Applications in Business and Economics", it is the perfect preparation for E463 Causal Machine Learning (see below).
ECTS:
6
Language:
English
Time:
Tuesdays from 10 am to 12 pm (c.t.) and Thursdays from 2 pm to 4 pm (c.t.)
Prerequisites:
Knowledge of probability theory and ordinary least squares
This module introduces recent methods at the intersection of causal inference and machine learning. This includes (i) methods that leverage supervised machine learning for the estimation of average or heterogeneous treatment effects, like Double Machine Learning and Causal Forests, and (ii) policy learning that recommends ads/policies/treatments/… in a data-driven way, like bandit algorithms. The methods are applied to simulated and real datasets using the programming language R.
ECTS:
9
Language:
English
Prerequisites:
Successful participation in either "S411 Advanced Time Series Analysis" or "S422 Advanced Microeconometrics" or comparable courses (please consult Michael Knaus).
The combination of E464 Causal Inference (see above) with general machine learning modules like "DS405 Machine Learning Applications in Business and Economics" is the perfect preparation for this module. However, you can also take this module without prior participation in these courses.