Visibly Discriminated

Javier G. Polavieja

18 mins - 20 de Abril de 2023, 07:05

The author presents the first large-scale study of employment discrimination in Europe to reveal the importance of phenotype in access to employment
Due to increased migration flows from other continents, the European population today is much more diverse than at any other time in modern history.  It is estimated that in 2017 there were more than 19 million citizens in the European Union with foreign parents, of whom more than 5 million had parents born outside Europe (Eurostat data). This means, among other things, that today many "new Europeans" are "visibly" identifiable as being of non-European descent.  The risk of such visibility being a source of prejudice and discrimination is a real risk and yet surprisingly little studied in Europe.

Existing research on the socio-economic integration of international immigrants in Europe has traditionally considered that coming from (or being descended from) Muslim countries is the main trigger for prejudice and discrimination. It has thus been argued that prejudice and discrimination on our continent has a fundamentally religious-cultural basis, in contrast to the case of the US, where 'race' is seen as the main factor generating inequality of opportunity. This argument has been sustained, however, by a striking lack of empirical research. The majority of European social scientists' rejection of the concept of 'race' as an analytical category coupled with the lack of adequate data and methodological tools, have prevented us from investigating to what extent belonging to a 'visible' minority may be an additional source of discrimination against the (mis)named second generation - that is, Europeans descended from international immigrants, whom it would probably be more appropriate to call 'new Europeans'.

To fill this gap, researchers from the Discrimination and Inequality Lab of the Universidad Carlos III de Madrid (D-Lab), the University of Amsterdam, the Berlin Social Science Centre (WZB) and the German Centre for Integration and Migration Research (DeZIM) have conducted the first large-scale comparative experiment on racial discrimination in access to employment in Europe, recently published in the academic journal Socio-Economic Review. As the lead author of this study, which was entirely funded by public research funds, I would like to share some of our most important findings in order to contribute to its public dissemination, following the recommendations of our funding bodies, the European Research Council and the Spanish State Research Agency.

A pioneering experiment
Field experiments are the most appropriate methodological tool for the study of discrimination in access to employment because they allow us to observe the real behaviour of employers (or their agents) in personnel selection processes - for which it is essential that employers do not know that we are observing them (an issue we will now skip over). The most appropriate field experiments for the study of employment discrimination are the so-called matching tests, which basically consist of sending the CVs of fictitious candidates to real job offers for a number of occupations selected by the researchers. The CVs sent are identical in all the characteristics relevant to the job offered, but differ in those characteristics that we consider hypothetically likely to generate discrimination in a given labour market, characteristics that we call "treatments" (as this methodology originally comes from the field of clinical trials). With a sufficient number of experimental units (firms), the randomisation of treatments allows us to identify the existence of discrimination in access to employment, estimate its intensity and calculate its statistical significance. To do so, we measure the average probability that firms are interested in applicants with the characteristic we want to study, as opposed to being interested in applicants with identical curricula but who do not possess this characteristic. To minimise the damage caused to the companies analysed, the matching test ends at the first unequivocal sign of interest (phone call or e-mail) in the job applicant.

Well, in our experiment on racial discrimination, we recorded the responses of almost thirteen thousand European companies to as many fictitious job applications in three EU countries, Germany, the Netherlands and Spain. We chose these three countries because in all of them it is common practice for CVs to include a photograph of the candidate (this is not the case in other countries, as I will explain below). This allowed us to manipulate two crucial treatments: the ethnic ancestry of the job applicants (which we primarily signalled through their full names) and the phenotype (which we signalled through the photographs). All our simulated applicants were "new Europeans", i.e. young men and women, with nationality of the country of the experiment (i.e. Spanish, German or Dutch, as appropriate) and parents born in more than forty different countries, which we grouped into four major ancestry regions: Europe-US, Middle East-North Africa, Middle East-North Africa, Europe-US, Europe-US and Europe-North Africa. We group them into four main regions of ancestry: Europe-USA, Middle East-North Africa, Latin America-Caribbean and Asia. We created job applications for six different occupations (cook, hairdresser, shop assistant, receptionist, accountant and IT-software developer), which together account for 15-20 per cent of the workforce in each country. The names used to indicate the ancestry of the applicants were carefully selected to avoid religious or socio-economic status connotations and were tested to ensure that they were easily recognisable as coming from the four major regions of interest. In turn, the applicants' photographs were subjected to a thorough validation process to ensure that they were comparable to each other in physical attractiveness, likeability and perceived competence, which is essential to isolate the effect of phenotypic variation on firms' responses - in total, approximately 7,500 respondents participated in the validation of 16 photographs, 8 for each gender. In the experiment we used four phenotypic groups (corresponding to the popular perception of "races") which we called, unoriginally, "Black", "Asian/Amerindian" (ASIN), "Caucasian Brown-Skinned (CPM)" and "White". Finally, we performed additional validation tests to check that the phenotypes used were plausible for people descended from the four major ancestry regions (which does not mean that they are necessarily plausible for each and every country that makes up each region). Figure 1 shows an illustration of these four plausible phenotypes for (fictitious) female job seekers and high physical attractiveness scores.

The experiment was carried out simultaneously and in a fully standardised way in the three countries analysed for an unprecedented number of companies in the field of discrimination studies, which has allowed us to obtain the first cross-country comparable estimates of racial discrimination (based on physical appearance) in the literature. 
Figure 1.- Four basic phenotypes (illustrative pictures of plausible phenotypes for female job applicants with high physical attractiveness scores).
The importance of photos
Surprising as it may seem, the use of photographs as a treatment is, in itself, a very important contribution of this experiment. Most of what we knew so far about racial discrimination in recruitment came from Anglo-Saxon countries (especially the USA), where, paradoxically, the use of photographs in job applications is either prohibited by law (USA, Canada, Great Britain, New Zealand) or discouraged in practice (Australia). This limitation forces interested researchers in these countries to use "racial" sounding names (e.g. Lakisha or Jamal for African Americans in the USA) as the only way to measure racial discrimination, which is highly problematic because it prevents identifying the specific causal effect of the phenotype and separating it from the effect of other (cultural and socio-economic) cues that are also - and inevitably - provided by the more typical names. A crucial advantage of our study is that it allows us to investigate the role of phenotype and ethnic ancestry as potentially distinct triggers of discrimination in access to employment. 

Key findings: The net impact of the phenotype

Figure 2 shows the estimators of racial discrimination based on physical appearance obtained from the analysis of the responses of the nearly thirteen thousand firms surveyed. These estimators have two key features: 1) they are net, that is, they measure the effect of phenotypic variation isolated from the effect of ancestry region (and also isolated from the gender of the applicant and the characteristics of the vacancy) and 2) they are fully comparable across the three countries in the study (thanks to our standardised design). Figure 2 thus shows us how the net probabilities of positive response (a strong signal of interest in the job applicant) by employers vary for candidates of different phenotypes net of the effect of their ethnic ancestry. We use the white phenotype as the reference category in the statistical significance analyses - which, in turn, inform us about the extent to which the differences observed in our experiment are extrapolable to the national labour market as a whole (defined by the six selected occupations). To simplify the presentation of our main findings, the results for Germany and the Netherlands are shown together (with averaged estimators) under the label "Northern countries", as the average rates and response patterns of firms in these two countries do not differ significantly from each other. 
Figure 2.- Net racial discrimination against non-white phenotypes in Northern Countries and Spain (net probability of positive response and statistical significance)
The first finding to note is that phenotype (or racial appearance) matters in all three countries of the experiment, with its net impact, however, being significantly higher in the northern countries than in Spain. In the Netherlands and Germany, all "visible" phenotypes are penalised, especially the Asian-American and black phenotypes, with average positive response probabilities of 44 per cent compared to 55 per cent for identical candidates (in gender, type of vacancy applied for and region of ancestry) with a white phenotype. 

Due to the very state of the Spanish labour market (in particular, its high unemployment rate), we find a much lower average response rate in Spain than in Northern countries for all job applicants, who on average have to send five applications to get a positive response (compared to two applications in Northern countries). It has been argued in the economic literature that in oversupplied labour markets (i.e. with many applicants per vacancy), as is the case in Spain, employers may be more likely to discriminate, because the opportunity cost of discrimination is lower than in highly stressed markets where vacancies are much more difficult to fill. The rapid and massive increase in migration flows to Spain between the late 1990s and the outbreak of the Great Recession, as well as the particular intensity of the Great Recession in Spain, are two other factors that, in theory, should also favour discrimination against visible minorities. However, net discrimination rates for non-white phenotypes are comparatively lower in Spain than in the two northern countries. In fact, in the Spanish experiment we only found significant evidence of net discrimination against the black phenotype - and only at 90 per cent confidence, not for the other two non-white phenotypes. According to our estimates, applicants with a black phenotype in Spain have an average positive response probability of 18 per cent, which is only 4 percentage points lower than the one we found for curricularly identical applicants with a white phenotype (22 per cent). 

We have tested whether these differences in net racial discrimination between Spain and Northern Europe are statistically significant using a somewhat more complex statistical model (the details of which I will spare you) and indeed they are: in Spain there seems to be less net discrimination by phenotype than in Germany or the Netherlands. Does this mean that phenotype is of little importance in Spain? Not really. What it does mean is that phenotype matters differently in Spain, as its impact on employers' hiring decisions seems to depend on the applicant's ethnic ancestry. This means that discriminatory responses in Spain are triggered only for some specific combinations of phenotype and ancestry, most notably those that result in more stereotypical-looking applicants, as we explain in detail in the study (I will come back to this point later). 

The Combined Impact of Phenotype and Ancestry: The case of Maghreb and Middle Eastern descendants
The estimates presented in Figure 2 capture the reaction of European entrepreneurs to non-white phenotypes, after isolating the effect of applicant ancestry. However, our experimental design also allows us to estimate the cumulative impact of ancestry and phenotype. This cumulative effect can lead to severe levels of discrimination in Europe. To illustrate this claim, we now focus on the case of offspring of parents born in the Maghreb-Middle East (MME) region, a group of particular interest in the European context.
Figure 3.- Discrimination against Maghreb and Middle Eastern (MME) descendants by phenotype in Northern Countries and Spain (probability of positive response and statistical significance)
Figure 3 presents the mean probabilities of positive response for MME offspring with different phenotypes. The reference category, in this case, is job applicants with European parents and white phenotype. Note that the difference between this reference group and the second column, composed of applicants with a white phenotype and MME ancestry, informs us of discrimination that would be attributable solely to ethnic ancestry. In Germany and the Netherlands, having MME ancestry and white phenotype reduces the probability of a positive response by an average of 7 percentage points (from 56 to 49 per cent), a difference that, although statistically significant, is quite small. In Spain, the average difference in positive responses between MME offspring of white phenotype and equally white European offspring is 4 percentage points (from 24 to 20 percent), a difference that does not reach statistical significance in the Spanish experiment. Note that if the main factor triggering discrimination in Europe were the cultural distance associated with Muslim ancestry, as has been argued in much of the literature, then descendants of Maghreb and Middle Eastern countries with white phenotypes should be much more discriminated against. In other words, what our experiment shows is that a fundamental part of discrimination against descendants of majority Muslim countries in Europe has to do with their phenotype. This is true in all three countries in our study, albeit with an interesting difference:

In Germany and the Netherlands a kind of 'colour hierarchy' is observed (which we also find for descendants of other ancestral regions in these two northern countries). (which we also find for descendants of other ancestral regions in these two northern countries). We call it this because the probability of positive response decreases as we move away from the white phenotype in this order: (White)-Brown-skinned Caucasian-Asian/Amerindian-Black. The gap in the likelihood of positive response between the two extremes of this hierarchy is severe in the two northern countries, where, according to our estimates, MME descendants with black phenotype would have to submit up to ((1-(56/37))x100≈) 50 per cent more applications than white applicants of European ancestry (and 20 per cent more than white applicants of MME ancestry) to receive a positive response from firms. 

In Spain, however, we do not find such a clear phenotypic hierarchy (neither in this case, nor in the case of job applicants from other ancestral regions). What we find is that discriminatory responses against descendants from MME countries seem to be limited to applicants with Caucasian Brown-skinned (CPM) and Black phenotypes, which are precisely the two phenotypes most prototypical of this ancestral region. Specifically, in Spain, MME descendants with CPM phenotype would have to send a ((1-(24/14.5))x100≈) 65 percent more job applications than applicants with white phenotype and European ancestry (and 40 percent more than white applicants of MME ancestry) to obtain a positive response; while MME descendants with black phenotype would have to send a ((1-(24/15.5))x100≈) 54 percent more applications than applicants with white phenotype and European ancestry (and 31 percent more than those with white phenotype and MME ancestry). The estimates of discrimination against descendants of the Maghreb-Middle Eastern region and black and brown-skinned Caucasian phenotypes in the three countries in our experiment are comparable in magnitude to the estimates of racial discrimination we typically find in the US against African-Americans. This is severe discrimination.
The Mexican Hat Hypothesis: An Interpretation of the Results of the Spanish Experiment

The structure of the responses of the more than five thousand companies tested in Spain suggests that the phenotype in our country could function more as a reinforcing signal of ethnic ancestry than as an independent trigger of discrimination. This is precisely why its net effect is smaller in Spain than in the two northern countries analysed. However, when the combination of phenotype and name gives rise to more 'stereotypical' candidates, discrimination in Spain can become severe, as we have seen in the case of Maghreb and Middle Eastern descendants with a brown-skinned Caucasian phenotype (CPM). This is a very revealing case because the CPM phenotype is a common (and therefore perfectly invisible) phenotype among Spaniards of indigenous ancestry (see Figure 4), for whom being more or less brown does not carry any penalty. However, associated with an Arab, Turkish or Persian name, the same phenotype (the same photo) seems to trigger rejection. This means that sharing a very wide phenotypic range with our Maghreb and Middle Eastern neighbours (for obvious historical reasons) does not make this phenotype harmless for the descendants of those regions in Spain, although it may be for everyone else. The phenotype in Spain thus seems to act in a way analogous to a Mexican hat on the head of a Mexican-descendant candidate, i.e., reinforcing the ethnic ancestry signal - and triggering its associated stereotypes. This type of mechanism, for which evidence has been found in laboratory experiments, is known in the field of cognitive psychology as a mechanism of phenotypic prototypicality
Figure 4.- The brown-skinned Caucasian phenotype is common in the indigenous Spanish population. 
In Conclusion
To put it bluntly, our results suggest that German and Dutch entrepreneurs are more clearly pheno-racist, i.e. more sensitive to phenotype per se, than Spanish entrepreneurs, whom we might rather describe as ethno-racist. This, as we have seen, does not mean that the phenotype is of little importance in Spain, it just means that its effect on employers' responses is more difficult to dissociate from the effect of the applicants' ethnic ancestry. However, we will need much more empirical research to be able to confirm with greater certainty the existence of these two distinct patterns of racial discrimination that seem to be operating in Europe today, based on the results of this first comparative study.

What we can say, in the light of these same results, is that the widely held belief that discrimination in Europe is primarily culturally based is wrong. Thanks to our study, we now know that phenotype is a factor that affects the employment (and therefore life) opportunities of people of non-European ancestry to an extent comparable to, if not greater than, religious and cultural factors. By ignoring the role of phenotype, previous research has certainly overestimated the role of these cultural factors, since, as we have shown in this study, if all new Europeans of Maghreb and Middle Eastern descent were "invisible" (i.e. white in phenotype), employment discrimination against this minority would be drastically reduced in Europe. 

With these results in hand, to continue to ignore the role of racial discrimination as a driver of inequality in continental Europe is to adopt the ostrich strategy. The danger is that, by the time we want to poke our heads out, the lion may already be too close.

Funding: This study has been carried out in the framework of Project-D (New Frontiers in Research on Ethno-Racial Discrimination in Employment) (PID2020-119558GB-I00), funded by the Ministry of Science and Innovation (MCIN/AEI/10.13039/501100011033). The experimental design, fieldwork and preliminary analyses were carried out in the framework of the GEMM (Growth, Equal opportunities, Migration & Markets) project, funded by the European Union's Horizon 2020 Research and Innovation Programme (GA 649255).  
Se puede leer el artículo original en español 
¿Qué te ha parecido el artículo?