Computer-Aided Experimentation for Human Behaviour Analysis


Loading...

Author / Producer

Date

2022

Publication Type

Doctoral Thesis

ETH Bibliography

yes

Citations

Altmetric

Data

Abstract

The implementation of behavioural experiments in various disciplines has relied on computers for the last five decades. Interestingly, the use of computers has both facilitated and hindered different aspects of the experimental process. Over time, frameworks have been developed that make it easier to design experiments and conduct them. At the same time, the multiplicity of frameworks and the lack of documentation have made the replication of experiments a difficult endeavour. The reproducibility crisis is not unique to computer-aided experiments but is severely aggravated by it. Most experimental research focuses on translating a research question into an actionable hypothesis that can be falsified through experiments. However, the process of conducting the experiment—experimentation—is often left to the researchers' whims. While method sections in papers are meant to ameliorate this by giving an overview of how the research was conducted, it often omits several steps supposedly for clarity. This approach is following in the footsteps of Karl Popper to "omit with advantage" but the reproducibility crisis has shown that it is not easy to tell a priori what actually can be omitted with advantage. In this dissertation, I address this issue by formalising Computer-Aided Experimentation (CAE). The focus of previous research has been on reproducing experimental outcomes instead of creating a theoretical foundation for reproducible experimentation. Here, I rely on concepts from industrial research such as "Digital Twins", cloud research such as the "_ as Code" revolution, and insights from behavioural research on reproducibility to define Computer-Aided Experimentation for Human Behaviour Analysis. This thesis proposes a three-fold approach of theories, systems and applications to define CAE. First, I outline my theoretical framework behind CAE that relies on the three concepts: the "Experiments as Digital Twins" perspective, the "Experiments as Code" paradigm, and the "Design, Experiment, Analyse, and Reproduce" (DEAR) principle. Subsequently, I present five systems that employ the theoretical concepts of CAE. Each system addresses human behaviour analysis across a wide range of disciplines. Lastly, I report how these five systems are put to a test in domain science applications that have advanced their respective research fields. I conclude this dissertation with a discussion on the importance of these new theoretisations of experimentation and point towards future tasks in Computer-Aided Experimentation for Human Behaviour Analysis.

Publication status

published

Editor

Contributors

Examiner : Sumner, Robert W.
Examiner : Hölscher, Christoph
Examiner : Giannopoulos, Ioannis

Book title

Journal / series

Volume

I & II

Pages / Article No.

Publisher

ETH Zurich

Event

Edition / version

Methods

Software

Geographic location

Date collected

Date created

Subject

Computer-Aided Experimentation; Computer-Aided Experiments; Human Behaviour Analysis; Digital Twin; Reproducibility

Organisational unit

03987 - Hölscher, Christoph / Hölscher, Christoph check_circle
08698 - Game Technology Center (GTC)

Notes

Funding

ETH-15 16-2 - A Cognitive Science Approach to Evidence-Based Design Using Virtual Reality (ETHZ)

Related publications and datasets