Research

Research in the ECS group is guided by the question of how we can learn from empirical data. In consequence, much of our work is centered around Statistics & Research methods. The main content-area we work on is biological information processing in humans, but we also have worked on applying statistical and machine-learning methods to technical systems (e.g., to solar thermal systems). Below find a sketch of some typical questions we have been working on. For a more fine-grained and complete view, please visit our publication pages.

Main funding source are currently the DFG collaborative research center Robust Vision, the research unit Modal and amodal cognition, and the excellence cluster Machine Learning: New Perspectives for Science.

Biological information processing

Two visual systems or one?

Comparing the results of all studies on Garner interference in classic perceptual tasks, grasping, and manual size estimation. Quite surprisingly, manual size estimation behaves differently than is often assumed. See also Bhatia et al. (2025).

A famous issue in the Neurosciences is the question of how we can define functional sub-processes within the human cognitive system. Current theories often assume that the neuronal areas and processes that are relevant for visual object recognition are fundamentally different from those that guide visual actions. We found that these processes seem to be more similar than is often thought—which was very unexpected and surprising for many scientists. Sample publications:

• Bhatia, K., Osenberg, A., Janczyk, M., & Franz, V. H. (2025). Reviewing evidence for the perception-action model from Garner interference. Journal of Experimental Psychology: Human Perception and Performance, 51(2), 217-242. [ DOI | Website | PDF | Pre-registrations: 1 | 2 | 3 | Open materials & data | Abstract ]

• Bhatia, K., Löwenkamp, C., & Franz, V. H. (2022). Grasping follows Weber's law: How to use response variability as a proxy to JND. Journal of Vision, 22(12), 1-27. [ DOI | Website | PDF | Open materials & data | Abstract ]

• Kopiske, K. K., Bruno, N., Hesse, C., Schenk, T., & Franz, V. H. (2016). The functional subdivision of the visual brain: Is there a real illusion effect on action? A multi-lab replication study. Cortex, 79, 130-152. [ DOI | Website | PDF | Explanation of this preregistered report | Open materials & data | Abstract ]

• Franz, V. H., & Gegenfurtner, K. R. (2008). Grasping visual illusions: Consistent data and no dissociation. Cognitive Neuropsychology, 25(7), 920-950. [ DOI | PDF | Abstract ]

• Franz, V. H., Gegenfurtner, K. R., Bülthoff, H. H., & Fahle, M. (2000). Grasping visual illusions: No evidence for a dissociation between perception and action. Psychological Science, 11(1), 20-25. [ PDF | Abstract ]

Can eyes be surprised?

Responses of the human pupil to sudden changes in the environment. See also Basgol et al. (2025).

Humans constantly track their environment for statistical regularities and use internal models to make predictions about the world. When such predictions are violated, a general arousal signal seems to be generated by the locus coeruleus in the brainstem that triggers new learning of the changed environmental situation. This arousal signal can be measured as a dilation of the pupil in the eye -- which seems suprising in itself. Sample publications:

• Basgol, H., Dayan, P., & Franz, V. H. (2025). Violation of auditory regularities is reflected in pupil dynamics. Cortex, 183, 66-86. [ DOI | Website | PDF | Pre-registration | Open materials & data | Abstract ]

• Basgol, H., Dayan, P., & Franz, V. H. (2022). Is pupil-linked arousal a marker of model violation but not model update? (Poster at the 44th European Conference on Visual Perception (ECVP)) [ DOI ]

Can high-level cognitive functions be unconscious?

The standard reasoning employed in research on unconscious processing infers from a significant effect in the indirect task (on reaction times, EEG, fMRI, or similar) that there is an underlying good sensitivity to the stimulus. We show that this is a fallacy. See also Meyen et al. (2022).

It is often assumed that even complex mental processes can be performed outside consiousness. If true, this would have far-reaching implications for the understanding of the functional role of consiousness. However, we found a serious methodological limitation that affects a large number of studies in this realm. In consequence, it is necessary to reassess the empirical evidence for many of the stronger claims about unconscious vs. consious processing. Sample publications:

• Meyen, S., Vadillo, M. A., von Luxburg, U., & Franz, V. H. (2024). No Evidence for Contextual Cueing Beyond Explicit Recognition. Psychonomic Bulletin & Review, 31, 907-930. [ DOI | Website | PDF | Pre-registration | Open materials & data | Abstract ]

• Meyen, S., Zerweck, I. A., Amado, C., von Luxburg, U., & Franz, V. H. (2022). Advancing research on unconscious priming: When can scientists claim an indirect task advantage? Journal of Experimental Psychology: General, 151(1), 65-81. [ DOI | Ask for copy | Preprint | Online ITA-calculator | Open materials & data | Abstract ]

• Schnepf (nee Zerweck), I. A., Friedrich, F., Hepting, C., Meyen, S., & Franz, V. H. (2022). Neural mechanisms of response priming do not support veridical unconscious processing. Consciousness & Cognition, 102, 103348. [ DOI | Website | Ask for copy | Abstract ]

Can humans feel when somebody is lying?

Using a simple toy-example based on the weight of babies, we show that a significant difference does not imply an underlying accurate classification. See also Franz & von Luxburg (2014) and Meyen et al. (2022).

Researchers sometimes report that humans have an intuitive feeling of when somebody is lying — superior to their conscious assessment. If true this would have far-reaching practical consequences (for example for the legal system and for jurors at court). However, we found for studies on lie detection a similar methodological problem as for unconsciousness research (see above). This calls into question whether humans really have intuitive lie-detection capabilities that go beyond what they can consciously report. Sample publications:

• Franz, V. H., Meyen, S., & von Luxburg, U. (2024). Technical comment on Gunderson, ten Brinke, and Sokol-Hessner (2023). When the body knows: Interoceptive accuracy enhances physiological but not explicit differentiation between liars and truth-tellers. Personality & Individual Differences, 204, 112039. Personality & Individual Differences, 217, 112439. [ DOI | Website | Ask for copy | Preprint | Open materials & data | Abstract ]

• Meyen, S., Zerweck, I. A., Amado, C., von Luxburg, U., & Franz, V. H. (2022). Advancing research on unconscious priming: When can scientists claim an indirect task advantage? Journal of Experimental Psychology: General, 151(1), 65-81. [ DOI | Ask for copy | Preprint | Online ITA-calculator | Open materials & data | Abstract ]

• Franz, V. H., & von Luxburg, U. (2015). No evidence for unconscious lie detection: A significant difference does not imply accurate classification. Psychological Science, 26(10), 1646-1648. [ DOI | Website | PDF | Preprint | Open materials | Abstract ]

• Franz, V. H., & von Luxburg, U. (2014). Unconscious lie detection as an example of a widespread fallacy in the Neurosciences. (preprint at arXiv:1407.4240; this preprint is more comprehensive than the corresponding Psychological Science article) [ DOI | Website | Abstract ]

What is the representational format of human cognition?

How is a dog represented in the human cognitive system? Figure from Kaup et al. (2023). Modal and Amodal Cognition. Psychological Research.

A longstanding question in the cognitive sciences is the question of the representational format of human cognition. Is, for example, a dog represented in the brain by abstract concepts? Or is it represented by the sensory quality when interacting with the dog (smell, sound, etc.)? Sample publications:

• Kaup, B., Ulrich, R., Bausenhart, K. M., Bryce, D., Butz, M. V., Dignath, D., Dudschig, C., Franz, V. H., Friedrich, C., Gawrilow, C., Heller, J., Huff, M., Hütter, M., Janczyk, M., Leuthold, H., Mallot, H., Nürk, H.-C., Ramscar, M., Said, N., Svaldi, J., & Wong, H. Y. (2023). Modal and Amodal Cognition: An Overarching Principle in Various Domains of Psychology. Psychological Research, 1-31. [ DOI | Website | PDF | Abstract ]

• Janczyk, M., Eichfelder, L., Liesefeld, H. R., & Franz, V. H. (2024). Learning and transfer of response-effect relations. Quarterly Journal of Experimental Psychology. [ DOI | Website | PDF | Pre-registration | Open data | Abstract ]

• Eichfelder, L. A., Franz, V. H., & Janczyk, M. (2023). Is there hierarchical generalization in response-effect learning? Experimental Brain Research, 241, 135-144. [ DOI | Website | PDF | Pre-registration | Open materials & data | Abstract ]

• Bhatia, K., Osenberg, A., Janczyk, M., & Franz, V. H. (2025). Reviewing evidence for the perception-action model from Garner interference. Journal of Experimental Psychology: Human Perception and Performance, 51(2), 217-242. [ DOI | Website | PDF | Pre-registrations: 1 | 2 | 3 | Open materials & data | Abstract ]

• Bhatia, K., Löwenkamp, C., & Franz, V. H. (2022). Grasping follows Weber's law: How to use response variability as a proxy to JND. Journal of Vision, 22(12), 1-27. [ DOI | Website | PDF | Open materials & data | Abstract ]

Statistics & Research methods

Information theory for cue combination in biological and technical systems

Combining two independent classifiers each with an accuracy of 70% (colored symbols with numbers 1) can result in very different accuracies (colored symbols with numbers 2). We use classic information theory to disentangle those possibilities and to derive a measure of how 'confident' a classifier (be it a human or a technical system) should be in its classifications. See also Meyen et al. (arXiv preprint 2021).

We investigated classic information theory for the combination of different cues. This has surprisingly general applications for biological as well as technical systems, ranging from assessing the performance of groups, to combining multiple classifiers in machine learning, to the assessment of meta-cognitive abilities in humans. Sample publications:

• Meyen, S., Sigg, D. M. B., von Luxburg, U., & Franz, V. H. (2021). Group decisions based on confidence weighted majority voting. Cognitive Research: Principles and Implications, 6(18), 1-13. [ DOI | PDF | Open materials & data | Abstract ]

• Meyen, S., Göppert, F., Alber, H., Luxburg, U. von, & Franz, V. H. (2021). Specialists Outperform Generalists in Ensemble Classification. (preprint at arXiv:2107.04381) [ Website ]

Confidence intervals for within-subject designs, ratios, replications, etc.

Confidence intervals for the ratio of two random variables are suprisingly difficult. We derived a geometric construction method for those confidence intervals and used it to device a new Bootstrap procedure. See also von Luxburg & Franz (2009) and Franz (2007).

Estimating the size of empirical effects lies at the heart of the scientific method. This includes to quantify the uncertainty of these estimates. Confidence intervals are a traditional way to achieve such an uncertainty assessment. In a number of projects, we have worked on confidence intervals for specific situations and research designs. Sample publications:

• Göppert, F., Bhatia, K., Meyen, S., & Franz, V. H. (in press). Realistic expectations for replications: Expecting too little is just as bad as expecting too much. Advances in Methods and Practices in Psychological Science. [ Ask for copy | Abstract ]

• Franz, V. H., & Loftus, G. R. (2012). Standard errors and confidence intervals in within-subjects designs: Generalizing Loftus & Masson (1994) and avoiding biases of alternative accounts. Psychonomic Bulletin & Review, 19(3), 395-404. [ DOI | Website | PDF | Abstract ]

• von Luxburg, U., & Franz, V. H. (2009). A geometric approach to confidence sets for ratios: Fieller's theorem, generalizations, and bootstrap. Statistica Sinica, 19(3), 1095-1117. [ Website | PDF | Abstract ]

• Franz, V. H. (2007). Ratios: A short guide to confidence limits and proper use. (preprint at arXiv:0710.2024) [ Website | Abstract ]

Frequentist vs. Bayesian tests: Which should be used?

Regions of decision allow to compare different variants of Frequentist and Bayesian tests. See also Göppert et al. (TeaP-conference 2025).

While estimating the size of empirical effects should be of supreme importance in science (see our project on confidence intervals above), statistical tests can be helpful when they supplement (but not supplant) those estimates. In recent years (and following the replication crisis) there has been renewed discussion as to whether scientists should better use Bayesian or Frequentist tests. We contribute to this discussion by comparing those tests in a general way. Sample publications:

• Göppert, F., Szillat, L., Meyen, S., & Franz, V. H. (2025). Is optional stopping really no problem for Bayesians? (Poster at the 67th “Tagung experimentell arbeitender Psychologen” (TeaP), 9.3.-12.3.2025, Frankfurt, Germany) [ Abstract ]

• Göppert, F., Meyen, S., & Franz, V. H. (2024). Relating Frequentist and Bayesian hypothesis tests using regions of support. (Poster at the 66th “Tagung experimentell arbeitender Psychologen” (TeaP), 17.03.-20.03.2024, Regensburg, Germany) [ Abstract ]

What can we expect from replications?

We compared different methods to assess which effects to expect from the replication of an original study. See also Göppert et al. (in press).

It can be frustratingly difficult to replicate certain scientific claims (replication crisis). We contributed to the ensuing discussion by showing that we should expect even better replication rates than is sometimes assumed. Sample publications:

• Göppert, F., Bhatia, K., Meyen, S., & Franz, V. H. (in press). Realistic expectations for replications: Expecting too little is just as bad as expecting too much. Advances in Methods and Practices in Psychological Science. [ Ask for copy | Abstract ]

Applications to technical systems

Using statistics & machine learning to improve solar thermal systems (and heatpumps etc.)

We use machine learning to reconstruct the time series data of solar thermal systems. Deviations from those reconstructions allow to detect faults and improve performance. See also Ebmeier et al. (2022).

We use statistical methods and machine learning to improve environmentally friendly heating systems (like solar thermal systems, heat-pumps, etc). Although those systems are becoming increasingly complex (and therefore lend themselves to such an analysis), they are still much less complex than the human brain. They therefore provide a good baseline and comparison for the other work we are doing. We also published one of the first detailed and openly available data sets of solar thermal systems under real-world conditions (see Ebmeier et al. 2024 below). Sample publications:

• Ebmeier, F., Ludwig, N., Martius, G., & Franz, V. H. (2024). PaSTS: An operational dataset for domestic solar thermal systems. In Proceedings of the 15th ACM International Conference on Future and Sustainable Energy Systems (pp. 529-534). New York, NY, USA: Association for Computing Machinery. [ DOI | Website | PDF | Open materials & data | Abstract ]

• Ebmeier, F., Ludwig, N., Martius, G., & Franz, V. H. (2022). Fault Detection in Solar Thermal Systems Using Machine Learning. (Poster presented at the ISES & IEA SHC International Conference on Solar Energy for Buildings and Industry; EuroSun2022; 25-29 September 2022; Kassel, Germany). [ Abstract ]

Open Science

Open data, open methods, pre-registrations, registered reports, etc.

The review process of a Registered Report has two stages: First the research plan is assessed by reviewers (before data have been collected) and the journal issues an 'in-principle acceptance'. Only then the researchers are allowed to start collecting data and submit their results in a second review process to the journal. See also Kopiske et al. (2016).

Science needs to be transparent: Only when scientists can vet the claims of their fellow scientist can we even hope to approach something like objective, reliable knowledge about the world. In the wake of the replication crisis, many positive reforms have been implemented and we actively take part in this process: Whenever feasible, we make our data and methods openly available and pre-register our experiments (see corresponding links in our publication list). We also conducted one of the first fully registered reports (which is much more than a simple pre-registration, see explanation here) in the journal Cortex (Kopiske et al, 2016, see below), which was a 3-4 year long endeavor that comprised four different labs distributed all over Europe performing the same experiment in parallel. Sample publications & public outreach:

• Kopiske, K. K., Bruno, N., Hesse, C., Schenk, T., & Franz, V. H. (2016). The functional subdivision of the visual brain: Is there a real illusion effect on action? A multi-lab replication study. Cortex, 79, 130-152. [ DOI | Website | PDF | Explanation of this preregistered report | Open materials & data | Abstract ]

• Franz, V. H. (2021, February 10). Doping in der Wissenschaft: Gegenmaßnahmen & prä-registrierte Studien (Doping in Science: Countermeasures & Registered Reports). Center for Interdisciplinary and Intercultural Studies (CIIS), Tübingen. [ PDF | Abstract ]

• Franz, V. H. (2020, November 13). Doping in Science: Countermeasures & Registered Reports in Psychology and the Neurosciences. Universität Graz, Austria. [ PDF | Abstract ]

• Himmelbach, M., & Franz, V. H. (2019, January 22). ... und bin so klug als wie zuvor - Denkfallen in der Wissenschaft. Science Pub, Tübingen, Germany. [ Website | PDF | Abstract ]