UCPH Statistics Seminar: Dominik Janzing
Speaker: Dominik Janzing, Principal Research Scientist at Amazon Tübingen and Privatdozent at Karlsruhe Institute of Technology
Title: Toward quality standards for causal explanations
Abstract:
Causal explanations of why an unexpected event happened dominate our discussions about politics, psychology, economics, and medicine. Unfortunately, it often seems that people's willingness to accept or reject an explanation is driven by ideological preconceptions or simply by whether they like the narrative (including how entertaining it is). As scientists, we should instead insist on formal criteria for assessing the plausibility of causal explanations. After all, far-fetched causal explanations do a great deal of harm. Among others, I would propose the following criteria:
- Generalization: An explanation should come with some commitment about which other scenarios it applies to.
- Falsifiability: The existence of potential observations that would let us reject or revise a hypothesis is what distinguishes science from pseudo-science, according to Karl Popper.
- Empirical Evidence: Real observations or data points that would have falsified the hypothesis had they turned out differently.
I believe our ICML paper [1] already moves a bit in this direction, since it quantitatively evaluates plausibility—but I'm sure there is much more to say. Let's brainstorm together!
[1] Anahita Haghighat, Dominik Janzing: Formalizing and Falsifying Causal Pathways of Rare Events, ICML 2026, arXiv:2605.31254