Introduction

Abstract

In August 2017, the Swiss Federal Railways (SBB) was launching a campaign by means of which passengers should gain more attention for additional trains during rush hour. Between 6 a.m. and 8 a.m. employees dressed with a fox tail and fox ears had to walk around on platforms with posters that indicated when and where the additional trains leave. The idea of the SBB was that passengers who use these additional trains are sly foxes and vixens because these trains are less crowded than the normal ones.

In August 2017, the Swiss Federal Railways (SBB) was launching a campaign by means of which passengers should gain more attention for additional trains during rush hour. Between 6 a.m. and 8 a.m. employees dressed with a fox tail and fox ears had to walk around on platforms with posters that indicated when and where the additional trains leave. The idea of the SBB was that passengers who use these additional trains are sly foxes and vixens because these trains are less crowded than the normal ones. Now, the SBB was particularly looking for female employees who would disguise as a vixen and stand on the platform.¹ They advertised this job on the online platform of the University of Zurich and ETH Zurich as follows: “Affable young women wanted for SBB-campaign! For our “commute cleverly” campaign we are looking for affable and confident young women. Your job is to bring commuters attention to alternative rail connections. In order to do so, you, disguised as a “sly vixen”, hold up a poster and walk around on the platform. That’s it—all it needs is a little confidence. For two hours of effort, employees get paid CHF 150.” (Heininger & Hartmann, 2017).

This employment ad triggered a little shitstorm. The fact the SBB was particularly looking for sly vixens and not also sly foxes was perceived as sexist. While the women’s organisation Terre des Femmes Switzerland described the campaign as questionable, Tamara Funiciello, leader of the social democratic youth party (JUSO) used more drastic words. For her, the campaign was beneath contempt and a demonstration of how sexist Swiss society still is. Moreover, Funiciello stated that such campaigns would foster gender stereotypes. The whole debate received quite some media response. Several Swiss newspapers such as Blick (Heininger & Hartmann, 2017), Tagesanzeiger (Lehmann, 2017), Handelszeitung (Iseli, 2017), or Watson (sda, 2017) reported on it. The SBB replied that sly foxes can of course apply for the job as well. The reason why they particularly addressed women was that the sly foxes and vixens have to wear a hairband (on which the fox ears are mounted) and they thought that women can wear these better (Iseli, 2017; Heininger & Hartmann, 2017).

While Funiciello and Terre des Femmes Switzerland perceived the campaign and its application procedure as sexist, many readers of the abovementioned newspapers apparently had a different opinion. The comments of the articles reveal quite a nuanced analysis of the topic. First of all, against whom was this campaign discriminatory? On one hand, it could be argued that it discriminated against women because only they would walk around on the platform, wearing these rather silly costumes. So, the campaign particularly ridiculed women. Moreover, at the moment of the debate, the precise costumes had not yet been presented to the public. Therefore, it was unclear whether the costumes would emphasis women’s sexiness. If that had been the case, the campaign would not only have ridiculed but also objectivised women. Yet, because the costumes were very benign the reproach of objectivisation of women got more or less obsolete. The only argument that remained was that fox ears could remind people of playboy bunnies that normally wear bunny ears and are a symbol of female objectivisation. Furthermore, the mere fact that women disguise as vixens and walk around on platforms is evocative of older sexist ad campaigns, which makes this campaign at least questionable. On the other hand, it could also be argued that it discriminated against men because they were excluded from the recruitment process. CHF 150 for two hours of work is a very fair hourly wage, especially for students. So, it is discriminatory that only women got the chance to apply for the job.

Then, Funiciello said that such campaigns would foster gender stereotypes. Yet, she did not articulate which gender stereotypes the campaign fostered. In German, sly vixens is positively connoted. Thus, to describe a woman as a sly vixen is in most contexts not a depreciation but rather a compliment. Terre des Femmes confirms this impression and notes that vixens stand for astuteness, which they perceive as a positive characteristic (sda, 2017).² So, if any, sly vixens would establish a positive female stereotype. Maybe, Funiciello wanted to indicate that holding a poster for two hours in a vixen costume is an undemanding/ridiculous job. Consequently, if only women would do it, the picture could be portrayed that, compared to men, women are more willing to do such undemanding/ridiculous jobs. This would lead to a negative female stereotype. Yet, the fact that only women do the job could also suggest that men consider themselves too good for it. Such an interpretation comprises a rather negative stereotype of men: They are arrogant (the job is beneath their dignity) and lazy (the job is too inconvenient).

Ultimately, the SBB not only wanted to particularly recruit young women but young academic women. To be fair, it is unknown whether they exclusively advertised the job on the online platform of the University of Zurich and ETH Zurich. However, all newspapers wrote that the SBB are looking for young female academics. Astonishingly, no reporter or politician said that this recruitment strategy discriminates against non-academics. In order to do this job, you certainly do not need an academic background. So, there is no reason why the SBB particularly looked for academics. It might be objected that such a job is typically done by students because often they are short of money and therefore happy about some additional income. So, you probably find more applicants among students than among some other group as for example bankers. Nevertheless, there certainly are non-academics that are as happy about the additional income the job provides as academics. So, if the SBB really exclusively recruited academics, they would have systematically excluded all non-academics, which can be seen as an act of discrimination against non-academics.

We see that the whole topic is rather complex. It is unclear whether the campaign discriminates against women, who might be ridiculed and objectivised because of the vixen costume, against men because the employment ad particularly asks for women, or against non-academics because the SBB published the ad on the platforms of two universities. Moreover, it cannot be said with certainty whether the costume that the sly vixens had to wear and the job they had to do reinforced female stereotypes. And even if they did, we do not know whether these stereotypes would be positive or negative. Nonetheless, this sly vixen incident clearly demonstrates two circumstances: (1) Our society is very sensitised for possibly discriminatory acts, which would immediately receive harsh critique; (2) At least sometimes there is no consensus about whether an act truly is discriminatory and about who the victims of discrimination are.

These two circumstances are not only observable in case of humans that behave in a possibly discriminatory way but also in case of machines or algorithms that “behave” in a possibly discriminatory way. Only recently, several news and scientific articles have been published that cover that topic (e.g. Wolfangel, 2018; Gratwohl, 2018; Steinharter & Maisch, 2018; Frisse, 2019; Kleinberg et al., 2019; Williams et al., 2018). For example, the title of Eva Wolfangel’s article is “programmed racism”, which implies that, just like humans, algorithms (or more precisely their output) can be racist and thus discriminatory. In order to illustrate this, she uses among others the example of an algorithm that screens job applicants and in so doing is programmed not to consider skin colour. Now, through machine learning, the algorithm has found a positive correlation between fluctuation rate and how far away someone lives from her workplace. Consequently, the algorithm recommends applicants who live close to their workplace.³ According to the article, this disadvantages black people because they are more likely to live in suburbs (and thereby further away from the workplace) than those of other skin colour. This is why the author writes that the algorithm’s output is racist and thus discriminates against black people.

The message of Wolfangel is unambiguous: Such algorithms are not only discriminatory but also dangerous and illegitimate. She finishes the article by writing that at the end of the debate, we might come to the conclusion that after all it is still better if not machines but humans pass judgements on humans; mistakes included. Yet, as in case of the sly vixen campaign, not all readers agree with her argument of racist algorithms. Regarding the applicant screening algorithm mentioned above, some write in their comments that it simply found a correlation between two variables. The fact that one of these two variables also correlates with skin colour does not make the algorithm’s output racist and thus discriminatory.

This whole debate about whether algorithms can be discriminatory and should be legally examined and forbidden if necessary is gaining more and more momentum. This is particularly true for the U.S., where algorithms are already used for several years in different areas such as to assess a criminal defendant’s likelihood of becoming a recidivist (Larson et al., 2016) or to identify potential criminal activity (Kartheuser, 2018). Yet, also in European countries such as Germany or Switzerland, the debate about discriminatory algorithms has been launched as multiple news articles demonstrate (Wolfangel, 2018; Gratwohl, 2018; Steinharter & Maisch, 2018; Frisse, 2019). So, it can be expected that beside our already existing sensitivity of human discrimination we will also develop (or amplify) a sensitivity of algorithmic discrimination.

The fact that our society is very sensitised (and seems to get even more sensitised) for possible discriminatory acts is an incredibly great achievement. 200 years ago, in the U.S., slavery and thereby inconceivable discrimination against black people was part of everyday life. Then, although slavery was forbidden after the civil war in 1865, it took another 99 years until segregation and discriminating election tests became illegal. Ultimately, in 2008 and 2012, Barak Obama got elected as the first black President of the United States of America. Hence, much has changed in the last 200 years. Nevertheless, up until today, racism is common in the U.S. and as for instance the Black Lives Matter movement demonstrates, African Americans still have to fight against discrimination. Western Europe had a long history of terrible discrimination against Jews. In the Middle Ages, Christians accused them to have poisoned wells and therefore to be responsible for the plague (Cohn, 2007). Shortly after, pogroms were introduced and many Jews killed. Centuries later, discrimination against Jews culminated in the Holocaust. Nowadays, even though antisemitism has still not disappeared completely, Jews are no longer a threatened but a protected minority in Western Europe. In Switzerland, official discrimination against women survived until 1990. On a national level, women got the right to vote after a referendum in 1971. Yet, on a cantonal level the Federal Supreme Court of Switzerland had to ultimately force the Canton of Appenzell Innerrhoden to finally introduce women’s right to vote in 1990. Today, there certainly is much less discrimination against women as there used to be. Nonetheless, in Switzerland, women still earn 20% less than men, whereof 39.1% are not explainable by means of structural factors (BFS, 2016).

These are just three examples where discrimination against a certain group decreased over the last decades and there certainly are many more. Yet, there are also groups against which discrimination is not decreasing but increasing. For example, in 2016, the European Union Agency for Fundamental Rights conducted a survey, including more than 10’500 people that described themselves as Muslims. Overall, 15 EU-member states were involved. The results revealed that in the last five years 17% of participants felt discriminated against because of their religion. In 2008, this number was 10% (Reimann & van Hove, 2017). This increase in hostility against Muslims is not silently accepted but thematised within the public discourse. Thus, although there is more discrimination against Muslims, people should also become more sensitised for discriminatory acts against Muslims. But of course, only because people are sensitised does not automatically imply that they stop to discriminate. As the examples of the last paragraph demonstrate, this often takes decades, if not centuries.

As desirable as this overall sensitisation for discrimination and especially its positive impact on the reduction of discrimination is, it also has a side-effect. Talking about differences between social groups has become a normative minefield. If you say that immigrants are more likely to be convicted for a crime (Schmidli et al., 2016) or to be unemployed (Rütti, 2017), you might be called a xenophobe. If you say that a 30-year-old woman who wants to have children is potentially costlier for an employer than a 30-year-old man who wants to have children, you might be called a sexist. If you say that women are still underpaid and that careerwise there is a glass ceiling for them, you might be called a feminist. If you say that some moral values of immigrants who come from conservative Islamic countries are unwanted in the Western World, you might be called an islamophobe. If you say that a 22-year-old man who still has to do exercises for the reserves is costlier for an employer than a 22-year-old man who is unfit for the military, you might be called a traitor of your country. And if you say that there appear to be significant biological differences among human populations, you might be called a racist (Reich, 2018).

While some proudly call themselves feminists, few people want to be called a xenophobe, sexist, islamophobe, traitor of your country, or racist. Therefore, in our discrimination-sensitised society you have to be cautious when you talk about group differences because the accusation of discrimination is always just around the corner. And even if only a few critics call you a discriminator, this can already afflict substantial damage to your reputation. For example, although the public was divided on whether the sly vixens campaign of the SBB was discriminatory against women, it triggered quite some negative press.

The word discrimination has such power because of the following circumstance: In everyday use, it carries a heavy normative load. To discriminate is always morally reprehensible. As a consequence, discrimination is a moral offence and discriminators are bad people. Now, since most of us want to be a good person it naturally is of great importance to know which actions or statements are discriminatory and which are not. The Cambridge Dictionary (2018) provides an answer. It defines discrimination as “treating a person or particular group of people differently, especially in a worse way from the way in which you treat other people, because of their skin colour, sex, sexuality, etc.”. This definition is perfectly compatible with the abovementioned examples. When only men had the right to vote the law treated women worse than men and therefore discriminated against women. When only white people were allowed to study at a certain university, the law treated black people worse than white people and therefore discriminated against black people. And when there were exclusive employment bans for Jews, the law treated Jews worse than non-Jews and therefore discriminated against Jews.

Of course, discrimination not only occurs on a state level but also on an individual one. If a landlord does not rent his apartments to Muslims, he treats Muslims worse than all others and therefore discriminates against Muslims. If an employer is solely looking for male applicants, he treats women worse than men and therefore discriminates against women. And if a ticket collector only controls those passengers that seem to be immigrants, he treats immigrants worse than natives and therefore discriminates against immigrants.

All these examples are hardly controversial and most of us would agree that the described acts are discriminatory and therefore also morally reprehensible. Yet, if we for instance go back to the sly vixens campaign of the SBB, we see that, unlike in the examples of the last two paragraphs, it is unclear which (if any) group was treated unfavourably and therefore was the victim of discrimination. Did the campaign discriminate against men because the ad exclusively addressed women and therefore treated men worse than women since men were not intended to apply for the job? Or did the campaign discriminate against women precisely because the ad exclusively addressed women and therefore treated women worse than men since ultimately only women would do this rather ridiculous job? Or were the true victims of discrimination non-academics because the ad might have only been visible on the university platform which treated non-academics worse than academics since the former had no chance to see it? Or did the campaign discriminate against both non-academics and men? Could it also have discriminated against both non-academics and women although this would imply that both the exclusion of a certain group and the inclusion of another one were acts of discrimination? And who would have been the reference group compared to which the discriminated groups were treated worse? Lastly, did the campaign discriminate against any group at all?

The reason why these questions are hard to answer is as follows: We examine them through a normative lens. Without any doubt, men and women (and probably also academics and non-academics) had been treated differently in case of the sly vixens campaign. So, following the definition of the Cambridge Dictionary, we have a clear case of discrimination. The only aspect that might remain unclear is whether there was discrimination against a group or simply discrimination between groups. Nevertheless, at least for some people the sly vixens campaign obviously did not feel discriminatory. And this is true although they would agree that the involved groups were treated differently. Now, it might be objected that this precisely is because they think that groups were merely treated differently and not worse. If they realised that it was not only a different but also a worse treatment, they would describe the treatment as discriminatory too. Yet, there are situations where we appear to mistreat a certain group, yet, do not perceive this mistreatment as a discriminatory act. So, neither different nor worse treatment by itself seems to always make us perceive an act as morally reprehensible and thus discriminatory.

In order to illustrate this argument, let us go back to the Cambridge Dictionary (2018). Here, we find the following example sentence for discrimination: “She felt she had been discriminated against because of her age.” In a situation where a 60-year-old woman never gets invited to a job interview although she is perfectly qualified for the advertised jobs we in all likelihood agree with the example sentence. This woman does not get invited to job interviews because of her age, which most people would describe as discriminatory and morally reprehensible. Now, let us consider the following two situations. (1) A 55-year-old woman has to pay CHF 3860 for her train abonnement, whereas a 64-year-old woman only has to pay CHF 2880 for the exact same abonnement because of a senior discount. Clearly, the younger woman is treated worse than the older one. So, the sentence “she felt she had been discriminated against because of her age” should be applicable to this situation as well. Yet, while there was much critique of the SBB’s sly vixens campaign, there has hardly ever been critique of their senior discounts (Stauffacher, 2018). Apparently, senior discounts are not perceived as being discriminatory against younger people.⁴ (2) A 30-year-old and an 85-year-old woman enter a crowded bus. While someone immediately offers his seat to the older woman, the younger woman does not get offered a seat and has to stand the whole ride. Again, the younger woman is therefore treated worse than the older one and should have been discriminated against because of her age. Yet, probably no one would label this act of offering your seat to old but not young people as discriminatory and morally reprehensible. On the contrary, such behaviour is not only considered to be morally right but also socially desirable.

The evident reason why these last two examples are not perceived as being discriminatory is that these mistreatments are socially legitimised and accepted. But this poses a problem because according to the definition of the Cambridge Dictionary, different (and particularly worse) treatment is sufficient for discrimination, regardless whether it is legitimate or not. There are two solutions to this problem: (1) We adjust the definition of discrimination in a way that an act does not only have to involve dissimilar treatment but illegitimate dissimilar treatment. (2) We acknowledge that discrimination is not always morally reprehensible and thus separate the two concepts. Let’s scrutinise the implications of the first solution. It suggests that the legitimacy of a dissimilar treatment makes it non-discriminatory. Now, what if 300 years ago, the fact that Africans could be enslaved was socially legitimised and accepted? Or what if 120 years ago, the fact that only men could vote was socially legitimised and accepted? If this applied, there would have been no discrimination against black people or women (only in hindsight). Thus, we see that the first solution can lead to dissatisfying outcomes. Moreover, it simply shifts the question of which acts are discriminatory to which dissimilar treatments are illegitimate. What about the second solution? It suggests that a dissimilar treatment always involves discrimination, regardless whether it is socially legitimised or not. Due to that there can also be discrimination that is socially desirable. If we accept this argument, the two legitimate and thus seemingly non-discriminatory situations mentioned in the last paragraph would still be discriminatory.

This outcome and the possibility of legitimate discrimination in general might be counterintuitive at the first moment. Yet, by allowing this option we free ourselves from a normative bottleneck when analysing discrimination: We do no longer have to care about the legitimacy of an act and can completely confine ourselves to different treatment. Comparing how we treat several individuals or groups and analysing whether these treatments differ is a rather descriptive task. And this is precisely what we want to do in this dissertation: dissecting discrimination from a descriptive perspective. Thereby, we ask how and why the identity of people involved in a possible treatment influences the specifications of this treatment. The method we use so as to answer these questions is decision theory enriched with behavioural economic, social psychological, evolutionary biological, sociological, and epistemological insights. Therefore, this dissertation has a broad interdisciplinary approach.

What is the benefit of such a perspective? The example of the sly vixens campaign or the applicant screening algorithm has demonstrated that people normally consider discrimination from an entirely normative angle. This quickly produces hardened fronts between different normative views. Moreover, it leads to a very one-dimensional conception of discrimination although the phenomenon actually has multiple facets. And these facets can be thoroughly worked out by means of a descriptive approach. Through examining the mechanisms and functions of these different facets, we are able to assess the significance of discrimination in everyday life. The results of this descriptive analysis can then be used as a basis for a normative theory of discrimination. For example, if a descriptive analysis leads to the result that discrimination is an essential ability of a functioning human being, a normative theory should acknowledge that and cannot simply condemn discrimination in general. Furthermore, it is also possible that the different facets of discrimination might differ from each other regarding their legitimacy. In this way, a descriptive analysis of discrimination might already provide a clear line between what a normative theory later defines as legitimate and illegitimate discrimination. Finally, the challenges we face when analysing discrimination from a descriptive perspective are also aspects that a normative theory of discrimination should consider.

The dissertation is subdivided into four major parts. The first part introduces decision theory. By use of decision theory, we again define discrimination. This leads to a comprehensive definition that is broader than the one used in this introduction since it involves different treatment of both things and people/groups. In a next step, we limit our analysis on social discrimination which comprises different treatment of people/groups. Then, we analyse the possible manifestations of social discrimination in two different decision situational settings: decision-making under certainty and decision-making under uncertainty. This leads to the following findings: (1) In case of decision-making under certainty, social discrimination always implies taste-based discrimination. (2) In case of decision-making under uncertainty, social discrimination can be revealed in form of taste-based discrimination and/or statistical discrimination.

The second part of the dissertation is about taste-based discrimination. We first investigate one of the most central social categorisations in taste-based discrimination, namely that of the ingroup and outgroups. Here, we will delve into social identity theory, which provides the best-known explanation for our different treatment of the ingroup and outgroups, called ingroup favouritism. Moreover, we will analyse the precise manifestations of ingroup favouritism and discuss whether it is mainly due to ingroup love or outgroup derogation. The next chapter compares taste-based discrimination with statistical discrimination and sheds light on the question whether taste-based discrimination is actually always statistical discrimination in disguise. This seems not to be the case but requires that people have a certain type of preferences. Thus, in the final chapter, we will examine whether humans truly have such preferences and in so doing reveal how they could have evolved.

The third part of the dissertation examines how we get our beliefs on which statistical discrimination is ultimately based. There are three superordinate chapters. The first chapter covers the idea of whether there are inherent beliefs that humans “learned” during the course of evolution. In the second chapter, we investigate how people truly update their beliefs and how much this differs from what economists describe as a rational updating process, namely Bayes’ law. The third chapter explores how historical and societal circumstances influence our beliefs, why it is difficult to overcome these beliefs, and why the way societies are structured leads to group inequalities.

The fourth part of the dissertation reassembles the dissected components of discrimination and puts them into a descriptive model of discrimination. This model depicts the centrepiece of the dissertation. In a next step, we look at what implications for a normative theory of discrimination we can derive from the descriptive model. This leads to five aspects that a normative theory of discrimination should consider.

Finally, conclusions are drawn. Here, we will shortly summarise the main findings of the dissertation, show how they improve our understanding of discrimination, and state where future research has to shed light on.