Postgraduate Course: Targeted Causal Learning (MATH11238)
Course Outline
| School | School of Mathematics |
College | College of Science and Engineering |
| Credit level (Normal year taken) | SCQF Level 11 (Postgraduate) |
Availability | Available to all students |
| SCQF Credits | 10 |
ECTS Credits | 5 |
| Summary | You might hear the statement: My headache went away because I took an aspirin. But was it really the aspirin? In everyday life, causal language is often used in an informal way. However, extracting precise answers to causal questions from data requires careful consideration. First, one should assess if the data, whether observational or generated by a designed randomised experiment, is rich enough to answer the causal question. Second, one should develop and use statistical techniques with mathematical guarantees to accurately estimate the quantity of interest.
In the first part of the course, we consider causal questions based on various data structures: From completely randomised experimental designs to various types of observational studies. We look at the role randomisation, design and confounding play in our ability to address these questions. For example, how should one design a population cohort, such as the Universitys Generation Scotland, to deal with confounding, maximise efficiency, and minimise bias in addressing national health policy? Causal questions, such as Did the aspirin cure my headache?, can be expressed using Neyman-Rubins theory of potential outcomes. Throughout, the material is illustrated by examples and case studies from fields such as biomedicine and healthcare policy, economics, and climate science.
In the second part of the course, we turn to a general estimation strategy to answer (causal) questions from observational data. In real-world applications, learning the datas ground truth distribution is often challenging or near impossible. At the same time, seemingly helpful simplifications may render estimates biased and statistical inference invalid. Rather than relying on parametric assumptions, we introduce a model-independent approach to estimation called Targeted Learning (TL). TL integrates causal inference with machine learning, leading to valid statistical inference of (causal) quantities of interest, such as the average causal effect. The relevant statistical theory will be developed, and the methods will be applied to relevant examples and case studies using the available TL packages (tlverse) in R |
| Course description |
In this course, we study causal inference from designed experiments and observational
data via targeted estimation. Topics may include:
- Designed experiments vs observational data, role of confounding
- Basics of causal inference, e.g., potential outcomes, matching
- Case study: Design of population cohorts
- Model-independent definition of a statistical parameter,
- Ensemble learning and cross-validation,
- Targeted Maximum Likelihood Estimation for statistical inference
|
Information for Visiting Students
| Pre-requisites | None |
| High Demand Course? |
Yes |
Course Delivery Information
|
| Academic year 2022/23, Available to all students (SV1)
|
Quota: None |
| Course Start |
Semester 2 |
Timetable |
Timetable |
| Learning and Teaching activities (Further Info) |
Total Hours:
100
(
Lecture Hours 22,
Seminar/Tutorial Hours 5,
Summative Assessment Hours 2,
Programme Level Learning and Teaching Hours 2,
Directed Learning and Independent Learning Hours
69 )
|
| Assessment (Further Info) |
Written Exam
80 %,
Coursework
20 %,
Practical Exam
0 %
|
| Additional Information (Assessment) |
Coursework: 20%
Examination: 80% |
| Feedback |
Not entered |
| Exam Information |
| Exam Diet |
Paper Name |
Hours & Minutes |
|
| Main Exam Diet S2 (April/May) | Targeted Causal Learning (MATH11238) | 2:00 | |
Learning Outcomes
On completion of this course, the student will be able to:
- Explain the nature of randomisation and the problem of confounding
- Recognise and apply various experimental designs
- Decide which methods to apply to extract causality from experimental and observational data
- Provide model-independent definitions of (causal) quantities of interest and apply Ensemble Learning and cross-validation estimation strategies
- Apply Targeted Maximum Likelihood Estimation to case studies
|
Reading List
Main textbooks:
- Lecture notes by the Course Organiser
- Imbens, G.W. and Rubin, D.B. (2015). Causal inference in statistics, social, and biomedical sciences. Cambridge University Press.
- Van der Laan, M.J. and Rose, S. (2011). Targeted Learning: Causal Inference
for Observational and Experimental Data. Springer, New York.
Further reading:
- Box, G.E.P. and Hunter, J.S. and Hunter, W.G. (2005). Statistics for experimenters: design, innovation and discovery. Wiley.
- Gelman, A. and Hill, J. and Vehtari, A. (2021). Regression and other stories. Cambridge University Press.
- Benkeser, D. and Chambaz, A. (2020). A Ride in Targeted Learning Territory. https://achambaz.github.io/tlride/
- Van der Laan, M.J. et al (2021). Targeted Learning in R: Causal Data Science with the tlverse Software Ecosystem. https://tlverse.org/tlverse-handbook/ |
Additional Information
| Graduate Attributes and Skills |
Not entered |
| Keywords | TCL,Designed Experiments,Observational Data,Causal Inference,Targeted Learning,Model-independent Est |
Contacts
| Course organiser | Dr Sjoerd Viktor Beentjes
Tel:
Email: Sjoerd.Beentjes@ed.ac.uk |
Course secretary | Miss Gemma Aitchison
Tel: (0131 6)50 9268
Email: Gemma.Aitchison@ed.ac.uk |
|
|
University Homepage