Evidence-based statistics and decision-making with e-values
Time and place
Time: 3 - 7 June 2027
Place: University of Turku, Finland
Lecturer: Nick Koning (Erasmus University)
Level and language: Graduate and Doctoral studies, taught in English
NOTE! The following course information is preliminary, and changes are possible.
Course overview
What is evidence? And how can we use it to support decision making?
In this course, we study state-of-the-art developments in statistical evidence and evidence-based decision-making. The course will center around the e-value, a recently proposed quantification of< evidence.
The course starts by studying general principles of data-informed decision-making. We study how e-values arise in such problems, and how they relate to other evidence variables such as tests and p-values. We continue by studying how e-values can be constructed, combined and sequentially accumulated. We then move to using e-values to support decision making, showing how valid evidence can be translated into valid decision-making. We finally study how a statistical inference problem can be abstractly formulated, and show how e-values can be used to answer them.
Target audience and prerequisites
The course is aimed at doctoral students in statistics or related quantitative fields, as well as doctoral students and researchers from fields of application.
Participants should know university-level probability and statistics, including expectations, conditional expectations, likelihoods, and basic hypothesis testing, and be comfortable following mathematical arguments. The course will be concept-heavy but light on mathematical proofs. Prior knowledge of e-values is not required.
Learning outcomes
After completing the course, participants:
- Know the general principles of data-informed decision-making
- Understand the relationship between e-values, p-values and tests
- Can apply the principles of decision-making to construct good e-values
- Can use e-values to combine evidence and sequentially gather evidence
- Mathematically formulate an inference problem in terms of a class of hypotheses
- Use e-values to support decision-making
Schedule
Provisional timetable. The lunch break is from 12:00 to 13:00 each day.
3 June Models and data-informed decision-making
09:00–12:00 Lecture
13:00–14:00 Discussion
14:00–15:00 Exercise session
15:00–17:00 Independent work on exercises
4 June Evidence variables
09:00–12:00 Lecture
13:00–14:00 Discussion
14:00–15:00 Exercise session
15:00–17:00 Independent work on exercises
5 June Constructing, merging and sequentially updating e-values
09:00–12:00 Lecture
13:00–14:00 Discussion
14:00–15:00 Exercise session
15:00–17:00 Independent work on exercises
6 June Using e-values to make decisions
09:00–12:00 Lecture
13:00–14:00 Discussion
14:00–15:00 Exercise session
15:00–17:00 Independent work on exercises
7 June Using e-values for statistical inference
09:00–12:00 Lecture
13:00–14:00 Discussion
14:00–15:00 Exercise session
15:00–17:00 Independent work on exercises
Teaching and course materials
Teaching takes place over five days through lectures, discussion, and exercises. Digital lecture notes and assignments form the main course materials, supplemented by selected readings.
Credits
5 ECTS
Assessment
The final grade is determined by a written 2-hour exam, with mostly open questions. Exam dates: 2 dates (to be decided).
Selected reading
The following sources support the course. Specific reading selections will accompany the lecture notes.
- Koning, N. W. (2026). Axioms for validity: e-values, p-values and data-dependent levels. arXiv:2605.28429
- Koning, N. W. (2024). How e-values generalize hypothesis testing: a Neyman-Pearson lemma for the e-value. arXiv:2409.05654
- Koning, N. W. (2026). The E-measure. arXiv:2604.20788
- Ramdas, A., and Wang, R. (2025). Hypothesis testing with e-values. Foundations and Trends in Statistics, 1(1–2), 1–390. doi:10.1561/3600000002
Xu, Z., Solari, A., Fischer, L., de Heide, R., Ramdas, A., and Goeman, J. (2025). Bringing closure to false discovery rate control: A general principle for multiple testing. arXiv:2509.02517