Join our Open Science lunch with Balazs Aczel and Dominik Dianovics for a discussion about how hidden assumptions shape the link between theories, models, and data in empirical research.
In their talk titled “The Double Underdetermination Problem: When data don’t determine theories and theories don’t determine models”, Balazs Aczel and Dominik Dianovics from ELTE University in Budapest will discuss the underdetermination of models by theory and provide recommendations on how to manage this uncertainty in research practice.
Event Details
Registration: https://forms.office.com/e/HL0uTQqtEe
Date: 18 November 2025, 11:00 – 13:00
Location: REC GS.11, Roeterseiland Campus (UvA), Nieuwe Achtergracht 129-B
This event is free to attend and includes a free lunch at 12:00, but registration is required since we have a limited number of spots.
About the speakers
Balazs Aczel is a Professor and Vice-Dean for Science at Eötvös Loránd University (ELTE), Budapest, Hungary, where he leads the Metascience and Cognitive Research Lab. His research explores the psychology of science and research transparency, with a particular focus on how scientific practices can be made more credible and reproducible. He has served as Chair of the Program Committee for the Society for the Improvement of Psychological Science (SIPS) and is an active contributor to large-scale, collaborative projects, including transparency checklists, many-analysts studies, and metascientific investigations of research workflows.
Dominik Dianovics is a PhD candidate in the Metascience and Cognitive Research Lab at Eötvös Loránd University (ELTE), Budapest. As a metaresearcher, his doctoral work focuses on the psychology of science, including topics such as academic burnout and researchers’ emotional experiencesthroughout the writing and publishing process.
Abstract
What to blame if my test fails? It would be simple to say ‘the theory’, but in reality, we never test theories in isolation but in a bundle, together with other hypotheses. These auxiliary hypotheses (such as ‘the experiment was coded correctly’, or ‘the participants understood the instructions’) often remain implicit. Nevertheless, when the test fails, they can be brought up to take the blame. Quine’s underdetermination of theory by data (1951) claims that the data can never compel us to accept or reject a specific theory, as any piece of evidence can always be made consistent with multiple theoretical frameworks if we are willing to adjust our auxiliary assumptions. In this talk, I will argue that the problem arises not only when we try to link data back to theory, but also when moving in the opposite direction—from theory to evidence—as each step from theoretical question to experimental design and then to statistical analysis depends on auxiliary hypotheses. Since statistical models always require more specifications than the theory alone can provide, there will always be multiple models compatible with the theory. I refer to this as the underdetermination of model by theory, a problem directly relevant to multiverse and multi-analyst analyses, and to confirmatory statistics more broadly. To end the talk on a more optimistic note, I’ll bring into the discussion an approach for managing this inherent uncertainty in research practice.
