Course catalogue doctoral education - VT22

  • Application can be done between 2021-10-15 and 2021-11-15
Application closed
Title Advanced Causal Inference
Course number 5298
Programme 0-Not part of doctoral programme
Language English
Credits 3.0
Date 2021-09-01 -- 2021-10-27
Responsible KI department Department of Medical Epidemiology and Biostatistics
Specific entry requirements The specific entry requirements are one of the following: 1) At least 60 higher education (HEC) credits from 1st and/or 2nd cycle courses in the subjects: statistics, probability, mathematics and data science of which at least 20 HEC credits must be in statistics or probability. 2) Course on Probability theory at least 7.5 HEC on 2nd cycle level; and Theory of statistical inference at least 7.5 HEC on 1st cycle level. In addition, the student should have taken courses in epidemiology (Epi I and Epi II) and in basic causal inference on 3rd cycle level.
Purpose of the course The purpose of this course is to give doctoral students with a previous degree in mathematics, statistics or a related area an introduction to the rigorous foundation of modern causal inference. Emphasis will be put on mathematical concepts, derivations and proofs.
Intended learning outcomes After having successfully completed the course the students will be able to:
- Use potential outcomes and counterfactuals to define causal effects.
- Briefly account for the philosophical controversies around counterfactuals.
- Show how exchangeability makes causal effects identifiable.
- Use d-separation on causal diagrams, to determine sufficient covariate adjustments for hypothesis testing.
- Use d-separation on twin networks, to determine sufficient covariate adjustments for effect estimation.
- Show how exposure regression (i.e. G-estimation) and outcome regression can be used to estimate conditional causal effects, and use M-estimation theory for asymptotic inference on these effects.
- Show how exposure regression (i.e. IPW) and outcome regression (standardization) can be used to estimate marginal causal effects, and use M-estimation theory for asymptotic inference on these effects.
- Use the concept of (non-)collapsibility to show the relation between conditional and marginal causal effects.
- Use causal diagrams to show how hypothesis testing works in instrumental variable (IV) settings.
- Show how two-stage estimation and G-estimation can be used for effect estimation in IV settings, and use M-estimation theory for asymptotic inference on these effects.
- Outline how linear programming techniques can be used to bound causal effects in IV settings.

Contents of the course The course content is defined by the learning outcomes.
Teaching and learning activities The course will use a flipped classroom model, where the learning is student-driven and focused on understanding and problem solving. All materials will be provided to the students in advance and during the first bi-weekly meeting with the teacher, all topics will be briefly introduced. One specific topic for the following class will be identified in advance for a longer, but still brief, discussion and any questions that the students have will be answered in a discussion with the teacher. After the first meeting, at each meeting, one or two of the students will give an oral presentation on the previously discussed topic. After the presentation, the students and the teacher will engage in a further discussion of the topic. Finally, at the end of each meeting the teacher will briefly introduce the topic for next meeting and allow for a brief discussion and assign students to present that topic. The students who are presenting will have two weeks until the next meeting to prepare their presentation. The teacher will be available via email or in person (as COVID-19 allows) to answer any questions the presenting students might have. All students, not only those assigned to present, are expected to study the material to be presented carefully so that they are able to actively engage in the discussion at all meetings. All students will be required to write a summary of each presentation demonstrating their understanding of the material. These will not be graded, but they will be read and feedback will be provided as needed to individual students to aid in mastery of the material.
Compulsory elements All seminars are compulsory. If a student miss a seminar, then the student will be given a chance to compensate for this by presenting the topic individually for the teacher, at a separate occasion.
Examination Each student will give a graded oral presentation where they demonstrate mastery of a set of the learning outcomes specific to the topic they cover, this presentation will be graded by the teacher for accuracy and understanding. In addition, all learning outcomes will be examined as an oral exam, which will involve a problem-solving exercise specific to a given data source and a scientific question that requires mastery of all learning outcomes. The problem will be provided in advance. The students will demonstrate their understanding by explaining to the teacher their understanding of the problem and how and which causal methods to apply to solve the problem. This solution and therefore the mastery of all learning outcomes will be graded. Students must pass both the presentation assessment and the oral exam to pass the course. Students that fail either piece can demonstrate mastery of the learning outcomes by providing an adequate written summary of the topic their presentation, which they had previously not demonstrated mastery of, or, in the case of failure of the oral exam, student can provide in writing an updated and now adequate solution to the problem posed to them. Students will have two additional opportunities for each requirement to demonstrate mastery, after which students will be asked to retake the course. Students can always select to provide a written solution to the problem posed to them, rather than an oral exam, but the same timeframe will be applied in both cases.
Literature and other teaching material The course is loosely based on the mandatory course literature:
- Pearl, J. (2009). Causality. Cambridge university press.
- Hernán MA, Robins JM (2020). Causal Inference: What If. Boca Raton: Chapman & Hall/CRC.
Additional material, mainly in the form of published scientific papers, will be handed out during the course.
Number of students 10 - 30
Selection of students Selection will be based on 1) the relevance of the course syllabus for the applicant's doctoral project (according to written motivation), 2) start date of doctoral studies (priority given to earlier start date)
More information The course is given 9.00-12.00 on every second Wednesday within the time span, i.e. on 1, 15 and 29 September as well as 13 and 27 October. The remaining time is devoted to self-studies. The last course occasion is reserved for the exam.
Additional course leader
Latest course evaluation Not available
Course responsible Arvid Sjölander
Department of Medical Epidemiology and Biostatistics
Contact person -