A Reanalysis of Cohn Et Al. 2014, Nature, ‘‘Business Culture and Dishonesty in the Banking Industry’’: The use of flawed statistics methods, used routinely in so-called “evidence-based” science, led the authors to distort the “evidence”

Hupé, Jean-Michel, 2018. “Shortcomings of Experimental Economics to Study Human Behavior: A Reanalysis of Cohn Et Al. 2014, Nature 516, 86–89, ‘‘business Culture and Dishonesty in the Banking Industry’’”. SocArXiv. March 20. osf.io/preprints/socarxiv/nt6xk

Abstract: In the wake of financial scandals, Cohn and collaborators published a headline-grabber study in the field of behavioral economics. M.C. Villeval (2014), in the News and Views of the Nature issue where this papers was published, summarized the main message: the “experiment shows that although bank employees behave honestly on average, their dishonesty increases when they make decisions after having been primed to think about their professional identity.” Cohn et al. thus provide evidence that “the incentives and the business culture developed in the financial sector may undermine the honesty norms of ordinary employees.” This study may have important consequences for policy, since, Villeval continues, “it is crucial to ensure a business culture of honesty in this industry to restore trust in it.” Villeval also argues that “from a scientific perspective, this study […] supports the economic theory of social identity […], links this theory with the economic analysis of lying behavior [… and] shows how behavioural economists can contribute to a broader reflection in science about how people manage their 'multiple selves' ”. Here I show that the use of flawed statistics methods, yet used routinely in so-called “evidence-based” science, led the authors to distort the “evidence”. Should we therefore question the contribution of behavioral economics to the understanding of human behavior? I am also using this data-set as an interesting example to explore how we can use modeling and simulations to provide a fair account of the information and uncertainty conveyed by the data, based on Confidence Intervals. I provide the R-code. I conclude with considerations on honesty and science.