Context matters when machine learning meets multilevel data

RC Trust is looking forward to welcoming Linus Hany from the University of Basel for a one-month research stay in Dortmund. At the beginning of his visit, he will offer insights into his current methodological research at the Statistical Colloquium at TU Dortmund University.
His talk, Barking up the right tree: Diagnosing and correcting split-selection bias in multilevel decision trees, addresses a challenge that arises when machine-learning methods are applied to data with a hierarchical structure – for example, when students are nested within classes.
Existing adaptations of decision trees can account for dependencies between observations in such multilevel data. However, they may not sufficiently consider the level at which individual predictors operate. Previous research by Linus Hany and collaborators has shown that this can bias split selection, causing predictors at the higher level to be selected too often. His talk will present a methodological response currently under development: a level-aware multilevel decision tree that uses different, level-appropriate tests for predictors at different levels.

A research visit with Marjolein Fokkema

The talk marks the beginning of Linus Hany’s research stay with Prof. Marjolein Fokkema and her Chair of Computational Statistics with Applications in Psychology. Linus Hany and Marjolein Fokkema collaborate on multilevel decision trees, connecting his work in Basel with the group’s research on statistical and machine-learning methods for scientific inference.
Linus Hany is a PhD candidate and research assistant in Statistics and Data Science at the Faculty of Psychology at the University of Basel, where his doctoral research is supervised by Prof. Mirka Henninger. He studied psychology and philosophy at the University of Zurich.
A recurring question in his research is when and why particular statistical or machine-learning methods are appropriate for particular data. His work aims to better understand existing methods and develop approaches that help researchers avoid misleading conclusions–an issue that becomes especially important when analyzing complex, context-dependent psychological data. Alongside his methodological work, he conducts applied research on disability sport, studying physical activity, mental health and substance use among people with disabilities and para-athletes.

During his stay in Dortmund, Linus Hany will be available for academic exchange with researchers interested in his work.

About his talk

Talk: Barking up the right tree: Diagnosing and correcting split-selection bias in multilevel decision trees
Speaker: Linus Hany, University of Basel
Date: September 30, 2026
Time: 2:15 p.m.
Location: Room E 21, Mathetower, Campus Nord, TU Dortmund University
Online: Participation via Zoom is also possible.

[ https://tu-dortmund.zoom.us/j/91555260271?pwd=2Tu1PafyWJ1NSb6Nm8wiwFudiF61c6.1#success ]