Seminar in Statistics: Causal Inference from Observational Data

Short Description:

We are used to study symmetric measures between random variables like correlation. In many problems, however, we are interested in the underlying causal mechanism, which is something inherently asymmetric. In this seminar, we learn under which assumptions we can infer causal relations from observational data and derive predictions about interventional experiments.

Prerequisites:

As prerequisites, we require:

• Introductory course to statistics
• The willingness to engage with a new topic

Program:

The schedule can be found here (UPDATED 25.02.13!).

Miniscript:

There is a miniscript containing definitions and results without proof (written by V. Lapteva). The current version (28.05.13) can be found here. Please send me an email if you find typos or mistakes.

Terms and conditions:

Each student is required to hold a presentation (see schedule, UPDATED 25.02.13! for available topics) about a given subject. Depending on the number of participants we allow 2 to 3 students per class who present 80 minutes in total (including questions). Each group will coordinate their oral contributions. Additionally, the presence in the seminar classes is mandatory.

Two weeks prior to the presentation date each student attends a meeting with C. Nowzohour or J. Peters (see schedule, UPDATED 25.02.13!) to discuss the topic and discuss remaining questions. It is expected that the student has already read the material and is familiar with the topic. One week prior to the presentation each student meets again with the same assistant. It is expected that presentation and handout are more or less finished. These meetings will take place either on Monday or Tuesday (time and place have to be fixed with the assistant directly).

We regard feedback as an important part of such a seminar. Rules for giving and receiving feedback can be found here, criteria for feedback can be found here. One week after the presentation one speaker is responsible for giving feedback, another one for asking questions.

Literature:

The seminar introduces the participants to causal inference, which is still a young and vibrant field of research. Therefore, there is no standard literature available. The following three books cover parts of the seminar:

• J. Pearl: "Causality: Models, reasoning, and inference''. Cambridge University Press, 2nd edition, 2009.
• P. Spirtes, C. Glymour and R. Scheines: "Causation, Prediction, and Search''. MIT Press, 2nd edition, 2000.
• S. Lauritzen: "Graphical Models''. Oxford University Press, New York, 1996.

If the books are not available in the library, you can come in my office (J. Peters) and copy the relevant pages.

Guidelines for the Presentation:

• The main goal of a talk is to impart (=vermitteln) knowledge.
• Therefore: use examples and pictures.
  
• You are responsible for scheduling your presentation. Including a discussion at the end (5 minutes), you should not go over 80 minutes.
• At each presentation one participant is responsible for asking questions.
  
• After the presentation there will be feedback (5 minutes).
• Look at other sources (books, papers, internet) to gain even more knowledge about your topic.