Using deepfakes for experiments in the social sciences - A pilot study.

The advent of deepfakes - the manipulation of audio records, images and videos based on deep learning techniques - has important implications for science and society. Current studies focus primarily on the detection and dangers of deepfakes. In contrast, less attention is paid to the potential of this technology for substantive research - particularly as an approach for controlled experimental manipulations in the social sciences. In this paper, we aim to fill this research gap and argue that deepfakes can be a valuable tool for conducting social science experiments. To demonstrate some of the potentials and pitfalls of deepfakes, we conducted a pilot study on the effects of physical attractiveness on student evaluations of teachers. To this end, we created a deepfake video varying the physical attractiveness of the instructor as compared to the original video and asked students to rate the presentation and instructor. First, our results show that social scientists without special knowledge in computer science can successfully create a credible deepfake within reasonable time. Student ratings of the quality of the two videos were comparable and students did not detect the deepfake. Second, we use deepfakes to examine a substantive research question: whether there are differences in the ratings of a physically more and a physically less attractive instructor. Our suggestive evidence points toward a beauty penalty. Thus, our study supports the idea that deepfakes can be used to introduce systematic variations into experiments while offering a high degree of experimental control. Finally, we discuss the feasibility of deepfakes as an experimental manipulation and the ethical challenges of using deepfakes in experiments.

COVID-19, normative attitudes and pluralistic ignorance in employer-employee relationships.

Employment relationships are embedded in a network of social norms that provide an implicit framework for desired behaviour, especially if contractual solutions are weak. The COVID-19 pandemic has brought about major changes that have led to situations, such as the scope of short-time work or home-based work in a firm. Against this backdrop, our study addresses three questions: first, are there social norms dealing with these changes; second, are there differences in attitudes between employees and supervisors (misalignment); and third, are there differences between respondents' average attitudes and the attitudes expected to exist in the population (pluralistic ignorance). We find that for the assignment of short-time work and of work at home, there are shared normative attitudes with only small differences between supervisors and nonsupervisors. Moreover, there is evidence for pluralistic ignorance; asked for the perceived opinion of others, respondents over- or underestimated the consensus in the (survey) population. Such pluralistic ignorance can contribute to the upholding of a norm even if individuals do not support the norm, with potentially far-reaching consequences for the quality of the employment relationship and the functioning of the organization. Our results show that, especially in times of change, social norms should be considered for the analysis of labour markets.

How beauty works. Theoretical mechanisms and two empirical applications on students' evaluation of teaching.

Plenty of studies show that the physical appearance of a person affects a variety of outcomes in everyday life. However, due to an incomplete theoretical explication and empirical problems in disentangling different beauty effects, it is unclear which mechanisms are at work. To clarify how beauty works we present explanations from evolutionary theory and expectation states theory and show where both perspectives differ and where interlinkage appears promising. Using students' evaluations of teaching we find observational and experimental evidence for the different causal pathways of physical attractiveness. First, independent raters strongly agree over the physical attractiveness of a person. Second, attractive instructors receive better student ratings. Third, students attend classes of attractive instructors more frequently - even after controlling for teaching quality. Fourth, we find no evidence that attractiveness effects become stronger if rater and ratee are of the opposite sex. Finally, the beauty premium turns into a penalty if an attractive instructor falls short of students' expectations.

Social control, social learning, and cheating: Evidence from lab and online experiments on dishonesty.

Varying the conditions of the decision-making environment we offered participants the opportunity to increase their payoff by undetectable lies. In addition to a baseline treatment, in which subjects rolled a die in private and showed a high extent of dishonest behavior, we increased the degree of social control by a novel treatment in which subjects played in randomly assigned pairs of two. The presence of others proved to substantially, but only temporarily reduce dishonest behavior. Furthermore, one treatment group received feedback on unethical behavior of participants in a similar experiment. Knowing that others betrayed in the experiment facilitated social learning and led to a higher prevalence of cheating. Finally, increasing the degree of anonymity by re-running the experiment online increased the extent of norm transgressions slightly.

Class attendance and students' evaluations of teaching: do no-shows bias course ratings and rankings?

BACKGROUND: Many university departments use students' evaluations of teaching (SET) to compare and rank courses. However, absenteeism from class is often nonrandom and, therefore, SET for different courses might not be comparable. OBJECTIVE: The present study aims to answer two questions. Are SET positively biased due to absenteeism? Do procedures, which adjust for absenteeism, change course rankings? RESEARCH DESIGN: The author discusses the problem from a missing data perspective and present empirical results from regression models to determine which factors are simultaneously associated with students' class attendance and course ratings. In order to determine the extent of these biases, the author then corrects average ratings for students' absenteeism and inspect changes in course rankings resulting from this adjustment. SUBJECTS: The author analyzes SET data on the individual level. One or more course ratings are available for each student. MEASURES: Individual course ratings and absenteeism served as the key outcomes. RESULTS: Absenteeism decreases with rising teaching quality. Furthermore, both factors are systematically related to student and course attributes. Weighting students' ratings by actual absenteeism leads to mostly small changes in ranks, which follow a power law. Only a few, average courses are disproportionally influenced by the adjustment. Weighting by predicted absenteeism leads to very small changes in ranks. Again, average courses are more strongly affected than courses of very high or low in quality. CONCLUSIONS: No-shows bias course ratings and rankings. SET are more appropriate to identify high- and low-quality courses than to determine the exact ranks of average courses.

Fat, muscles, and wages.

Recent studies in health economics have generated two important findings: that as a measure of fatness the body mass index (BMI) is biased; and that, when it comes to analyzing wage correlates, both fat-free mass (FFM) and body fat (BF) are better suited to the task. We validate these findings for Germany using the BIAdata Base Project and the German Socio-Economic Panel. While we find no significant correlation between BMI and wages in any of our models, simple linear regression models featuring both contemporary and time-lagged fatness measures indicate that FFM and, to a lesser extent, BF are associated with hourly wages: more specifically, the relationship between FFM/BF and hourly wages is about two to three times higher for females than for males. In contrast, fixed-effects models indicate that there is no correlation between hourly wages and both FFM and BF with one exception: a significant correlation (and one in line with expectations) is found to be the rule among job changers.