The boundary within: Are applicants of Southern European descent discriminated against in Northern European job markets?

Abstract

In the aftermath of the Euro debt crisis, negative stereotypes about Southern Europeans were (re)activated across Northern European countries. Because these stereotypes make explicit reference to productivity-relevant traits, they have the potential to influence employers’ hiring decisions. We draw on a sub-sample of the Growth, Equal Opportunities, Migration and Markets discrimination study (GEMM) to investigate the responses of over 3500 firms based in Germany, the Netherlands and Norway to identical (fictitious) young applicants born to Greek, Spanish, Italian and native-born parents. Using French descendants as a placebo treatment and sub-Saharan African descendants as a benchmark treatment, we find severe levels of hiring discrimination against Southern European descendants in both Norway and the Netherlands, but not in Germany. Discrimination in Norway seems largely driven by employers’ preferences for applicants of native descent, while in the Netherlands discrimination seems specifically targeted against Greek and Spanish descendants. Dutch employers’ propensity to penalize these two groups seems driven by information deficits.

Look at Tsipras [Greek Prime Minister], look at Varoufakis [Greek Finance Minister]. Would you buy a used car from them? When your answer is no, then vote no today!

—Klaus-Peter Willsch, member of the German CDU¹

1. Introduction

Greek, Italian, Portuguese and Spanish migration to Northern Europe has traditionally been associated with a high degree of ‘invisibility’ (see e.g. Tesser and Dronkers, 2007; Favell, 2008, 2013; Eremenko et al., 2017).² Though not immune from xenophobic and paternalistic attitudes, Southern European guest workers participating in the post-war reconstruction of countries such as Belgium, France and Germany were considered as culturally similar and more ‘integratable’ in comparison with non-European laborers (Schönwälder, 2004). Many of these guest workers settled permanently and raised their children in their new home countries. Examples of successful second-generation Southern Europeans born to post-war migrants abound in all types of activities, from arts and sports to politics and business, across Northern European countries, where Southern European guest workers and their descendants have often been depicted as a ‘model minority’ (e.g. in Switzerland) and enjoyed a privileged position in the existing ethnic hierarchies (see e.g. Wimmer, 2013, chap. 5; Strijbis and Polavieja, 2018). This privileged position was further reinforced by the new mobility dynamics set in motion by the Treaty of Maastricht in 1992, which established freedom of movement and residence for all EU citizens, including new Southern European movers, most of whom were highly educated—in sharp contrast to their guestworker predecessors (Favell, 2008). Today, both public discourse and scientific inquiry on migrants’ incorporation in Europe are mostly focused on ‘non-European’ nationals and their descendants, particularly groups having African, Asian or Muslim roots, as well as on Eastern Europeans, while paying little attention to the outcomes of people of Southern–European descent (but see Heath, 2007; Kalter and Granato, 2007; Algan et al., 2010). Southern Europeans have for long being a ‘non-issue’.

In the aftermath of the Great Recession, however, growing migration flows from the South to the North became increasingly politicized, while the Euro debt crisis unleashed a plethora of resentful stereotypes about the alleged ‘national character’ of Southern Europeans across Northern European countries (see Golec de Zavala et al., 2017; Sierp and Karner, 2017). Media and political discourse on the Greek and Portuguese bailouts, as well as on the Spanish banking bailout, often displayed moralistic overtones in countries such as Germany, the Netherlands, Finland or Austria, as their governments championed the implementation of severe austerity measures in Southern Europe in exchange for bailout loans. For example, on March 20, 2017, reflecting on the bailout debates, Jeroen Dijsselbloem, the Dutch finance minister and president of the Eurogroup, told the Frankfurter Allgemeine Zeitung:

As a social democrat, I consider solidarity to be extremely important. But whoever demands it also has to hold up their end of the bargain. I cannot spend all my money on liquor and women and then ask for your support. This principle applies on a personal, local, national and also on a European level. (Quoted from El País, 2017).

Growing migration inflows from Southern Europe were also highly politicized in the context of the Brexit campaign, which placed migration at the very center of the debate (see Goodwin and Milazzo, 2017). Even in Norway, a country outside the EU with comparatively small migration inflows from the South, the mainstream media raised concerns about increasing numbers of ‘Euro-refugees’ who were ‘fleeing the South to work in the North’ (Bygnes, 2016). Today, there is growing academic interest in the dynamics of intra-European inequalities and a certain consensus has emerged, at least among researchers in the ethnic boundary tradition, that previously ‘invisible’ or latent intra-European boundaries and hierarchies might be gaining in salience as a result of these multiple European crises (see e.g. Favell, 2013; Antonucci and Varriale, 2019). Still, there is a dearth of empirical research on new intra-European ethnic boundaries and their potential impact on the distribution of socio-economic opportunities in European societies—in other words, on whether and to what extent symbolic boundaries are translated into social boundaries through processes of social closure and discrimination (Lamont and Fournier, 1992; Lamont and Molnár, 2002).

To fill this gap, we draw on a subsample of data from the GEMM study, the largest harmonized field experiment on ethnic discrimination in hiring ever conducted in Europe (Lancee et al., 2019), to test whether young German, Dutch and Norwegian nationals of Italian, Greek and Spanish descent are discriminated against when looking for employment in their respective countries.³ Because access to employment is a crucial determinant of people’s life-chances, discrimination in hiring can be a very important source of socio-economic inequality. We hypothesize existing stereotypes about Northern and Southern Europeans’ respective ‘national characters’ can be particularly influential in shaping employers’ hiring decisions in a context of incomplete information because these stereotypes typically emphasize traits that are relevant for productivity (i.e. contentiousness/idleness, trustworthiness/dishonesty, competence/negligence, improvidence/foresight, etc.).⁴ By investigating whether country nationals of Southern European parentage are discriminated against when looking for jobs in Northern Europe, we test for a potential source of intra-European ethnic inequality previously overlooked in both the ethnic-boundary and the labor-market stratification literatures.

Our framework draws on insights from social and cognitive psychology, sociology and economics. We are particularly interested in disentangling the mechanisms of discrimination against Southern European descendants. To this end, we propose three empirical tests: (a) a placebo tests, (b) a target-consistency test and (c) a diagnostic test. Our placebo test uses French descendants as an ‘inactive’ treatment to discern whether our discrimination estimates are driven by ingroup favoritism—whereby employers simply favor applicants of native descent—or by employers’ specific rejection of applicants of Southern European descent. We stress that only the latter form of discrimination, which we call targeted discrimination, is consistent with processes of negative stereotyping, which, we argue, were more common in Eurozone lending countries, following the increased politicization of North–South European distinctions. Because negative stereotypes seem to have been particularly targeted against bailed-out countries in the context of the Euro debt crisis, we further test whether discrimination estimates are larger for Greek and Spanish descendants than for Italian descendants—this we call a target-consistency test. Our last test on mechanisms is a diagnostic test, which uses randomized productivity signals to check whether targeted discrimination is driven by informational deficits, as sustained by statistical discrimination models (Phelps, 1972; Arrow, 1973), or else responds to irrational forms of prejudice and/or cognitive bias, as in the seminal correspondence test of Bertrand and Mullainathan (2004) (see also Thijssen et al., 2021). Because we are also interested in gauging the intensity of the observed effects, we compare our discrimination estimates for Greek, Spanish and Italian descendants to those found for identical applicants of sub-Saharan African descent. We chose this group as a benchmark treatment because we know people of sub-Saharan African descent are subjected to particularly severe levels of discrimination in Europe, as consistently revealed by both individual (self-perceived) reports (see European Union Agency for Fundamental Rights, 2017), as well as by the existing field experimental evidence (Weichselbaumer, 2017; Di Stasio et al., 2021; Polavieja, 2022).

To further clarify the contribution of this study, it is also important to discuss upfront what we cannot do. First, because we have no comparable research on hiring discrimination against Southern Europeans prior to the present study, we cannot test whether discrimination has been enhanced by recent political events. While research in social psychology shows distinctions between ingroup and outgroups are much more likely to lead to open hostility and discrimination when they gain salience through politicization (Brewer, 1979, 1999) and mediatization (Fiske, 2002), we can only test whether discrimination against Southern European descendants exists today in each of the three countries studied—and whether observed callback patterns are consistent with stereotype-driven discrimination—but we cannot test whether discrimination has increased—as we suspect.

Second, while the GEMM field experiment allows us to observe the outcomes of employers’ hiring decisions in real-life settings, we cannot observe the mental processes that guide such decisions. In other words, the role of stereotypes as triggers of discrimination can only be tested indirectly in field-experimental research. Yet indirect testing can shed crucial light on discrimination processes when the experimental design includes a rich set of theoretically motivated treatments, several minority groups and country variation in the experiment’s settings, as it is the case of the GEMM study.

Finally, it is important to note that, although the fully harmonized nature of the GEMM study allows us to compare discrimination estimates across countries, the three-country scope of our data means we cannot causally identify the effect of any specific national variable. What we can do, however, is formulate several theoretically informed empirical expectations about which country contexts can potentially produce higher levels of discrimination against Southern European descendants and then test whether the results of the GEMM experiment are consistent with these expectations. To formulate our predictions, we take into consideration the following three factors: (a) the political climate affecting the dynamics of negative stereotyping; (b) the specific migration histories of each ‘minority’ group in each country and (c) some well-known country differences in the degree of institutional regulation of the job application process. We hypothesize all these three factors combined should make the Netherlands the country most prone to targeted discrimination against Southern European descendants, particularly those of Greek and Spanish ancestry.

2. Framework

Social psychologists and cognitive scientists note that humans have a deep-seated cognitive disposition to categorize people into ingroups and outgroups (see Fiske, 1998, 2002; Reskin, 2000). Social categorization is a cognitive shortcut people use to cope with the complexity of a demanding social environment (Fiske, 1998, pp. 362–364). Categorization leads perceivers to view members of outgroups in undifferentiated terms, reducing them to representatives of an assumed stereotyped core, potentially resulting in their depersonalization (Tajfel, 1982; Brown, 2000). Research in psychology suggests that categorical thinking is largely an automatic mental process (see Devine, 1989; Fiske, 1998; and discussion in Macrae and Bodenhausen, 2000). While this mental process seems hard-wired in our cognitive functions, the actual content of stereotypes is defined and maintained by principles of classification that are socially constructed. Stereotypes are thus produced through dynamic social processes that are historically contingent and reflect specific conflicts about symbolic and material resources (Bonacich, 1972; Hardin, 1995). Stereotypes can remain dormant or become activated depending on the socio-political context (Brewer, 1999; Fiske, 2002).

The re-emergence of stereotypes leading to North–South symbolic boundaries in Europe must thus be understood in the context of contemporary intra-European economic inequalities and political struggles over the allocation of resources, decision-making power and the reconfiguration of new institutional structures (Lamont and Molnár, 2002). Within this context, contemporary researchers are increasingly studying how public and political discourse, particularly in times of crisis, categorizes some Europeans (and their governments) as disciplined, honest, frugal and hard-working, while labeling others as corrupt, lazy, extravagant or backward (Chalániová, 2013; Adler-Nissen, 2017). The ‘PIGS’ moniker, which gained currency in the aftermath of the Eurozone crisis, captured this discursive moment by explicitly associating the economies and societies of Portugal, Italy, Greece and Spain with failure and backwardness (Capucha et al., 2014; Van Vossole, 2016). Such categorization processes draw on an essentialist logic by invoking cultural traits and even phenotypical distinctions as the basis for difference (Fox et al., 2012; Mylonas and Noutsou, 2017).

A recent illustrative example of intra-European stereotyping can be seen in the polemical cover of the Dutch right-wing weekly magazine Elsevier, published in the midst of discord over the European Covid-19 Recovery Fund under the headline ‘Not another penny for Southern Europe’ (see https://cdn.prod.elseone.nl/uploads/2020/05/20EWM022kl.jpg).^5,6 The cover contrasted blond-haired individuals performing skilled labor and professional activities, depicting the ‘industrious North’, with dark-haired individuals engaging in frivolous pool-side and terrace activities, depicting the ‘leisurely South’. Note that the Elsevier cover operates on a mirror-image logic of ingroup–outgroup comparison. Consistent with social identity theory, the ingroup derives self-worth by drawing favorable comparison with the outgroup (Tajfel and Turner, 2004), constructed as the polar opposite on a salient frame of reference (Brewer, 1979). The cover also illustrates that negative stereotypes, which thrived in Northern European media and political discourse in the aftermath of the Euro debt crisis (see e.g. Heinrich and Stahl, 2015; Van Vossole, 2016; Van Hecke, 2017), seem to be very much alive today.

Yet, by themselves, these examples do not constitute sufficient evidence that such symbolic boundaries are translating into patterns of social closure and discrimination. Negative stereotypes could be part of a cultural ‘game’ Europeans have been playing together for centuries, or may even be counteracted by positive stereotypes, which also exist—for example, valorizing Mediterranean culture/civilization and warmth (see e.g. Mylonas and Noutsou, 2017).⁷ In order for symbolic boundaries to become social boundaries they must have real consequences for people’s patterns of social interaction and—crucially—for their life chances. Hence, we must find evidence that stereotype-fed boundaries act as mechanisms of social closure, limiting individuals’ opportunities to access material resources (Brubaker, 2009; Connor and Koening, 2013). Under this light, employers’ hiring decisions become a crucial site for the study of boundary-making processes because these decisions have the potential to favor/restrict individuals’ access to gainful employment.

2.1 Discrimination: mechanisms and empirical predictions

Employers never have full and accurate information about a candidate’s potential productivity and therefore hiring decisions always involve contractual risks. Employers will seek to reduce these risks by factoring in any signal that they consider relevant for assessing applicants’ potential productivity. This is when stereotypes can have real consequences for people’s employment opportunities: if, rather than considering each job-applicant as a unique constellation of qualities and predispositions, employers construe them as members of ‘social categories’, to which fixed stereotypical attributes are attached, hiring discrimination will likely follow (Reskin, 2000).

Of course, job applicants display many traits that can be readily categorizable. Yet categories that emphasize productivity-relevant traits are likely to have an ‘activation advantage’ over other categories in the hiring processes (for a discussion of category activation, see Bodenhausen and Macrae, 1998). We note that cognitively unsophisticated employers will be particularly prone to engage in what we could call folk inferencing (to paraphrase Hirschfeld, 1996). This is an irrational form of thinking whereby perceivers fully embrace existing stereotypes by assuming all members of a given social category (e.g. ‘Germans’), to which a defining attribute has been associated (e.g. ‘hard-working’), possess the same value of such attribute (i.e. all Germans are hard-working). Note that this form of thinking is purely prejudicial (i.e. not based on reason or actual experience) because it ignores that variation around the mean is a universal feature of all population distributions. Basing hiring decisions on such ‘faulty and inflexible generalizations’ (Allport, 1955, p. 9, cited in Becker, 1977 [1957], p. 13) will inevitably lead to employment discrimination. This will be discrimination by taste, which implies—in competitive markets—employers will have to pay a price, either in the form of wages or forfeit income, in order to act on their prejudicial beliefs (Becker, 1977 [1957], p. 14).

It is important to note, however, that the effects of social categorization on hiring decisions will not be restricted to irrational forms of thinking alone. If employers believe people belonging to a specific social category are less qualified, reliable or committed on average (when compared with the average ‘majority’ applicant), they will still discriminate against members of this group. In this latter case, stereotypes affect hiring decisions by influencing employers’ beliefs about the distribution of productivity-relevant traits across categorized social groups (i.e. by leading employers to think, e.g. that German descendants are more hard-working on average than, e.g. Greek descendants). Note that beliefs about the distribution of unobserved traits in this case are not inflexible (i.e. they are not immune to reason or experience)—and need not even be faulty on average. This idea was explicitly stated in Phelps’ (1972) original theory of statistical discrimination—albeit posterior formulations of statistical discrimination theory in economics have been less attentive to the role of stereotypes. Phelps’ theory stressed that statistical discrimination is fully consistent with rational processes of profit maximization in contexts of incomplete information. This implies that rational employers will continue discriminating people from a given categorized social group until new contradicting information, either about the group’s average qualities or about the specific job applicant being evaluated, becomes available to them—provided acquiring such information is not too costly (Phelps, 1972; Arrow, 1973). This crucially means negative stereotypes can be trumped by (more) accurate information. Prior positive first-hand experience with members of the categorized group can provide counteracting information on group average qualities to employers who will then dismiss existing negative stereotypes. This is why positive intergroup social contact can reduce prejudice and discrimination (see e.g. Allport, 1955; Pettigrew, 1998). Likewise, the greater the number (and the better the quality) of productivity signals appearing in a given individual application, the lower the need for rational employers will be to draw on group stereotypes in order to forecast individual applicants’ potential productivity. This means statistical discrimination should decline with information about applicants’ productivity. To our knowledge, 12 papers have tested this prediction, from Bertrand and Mullainathan’s (2004) first diagnostic study on hiring discrimination against African Americans in the USA to Thijssen et al.’s (2021) latest study on discrimination against migrant descendants in Europe. We note that most of these studies (10 out of 12) found no support for the statistical discrimination mechanism (i.e. adding additional productivity signals did not reduce discrimination propensity) (see reviews in Bertrand and Duflo, 2017 and Thijssen et al., 2021).

2.2 Ingroup favoritism or targeted outgroup rejection?

Discrimination against any outgroup, for example Southern European descendants, could be driven by two distinctive mechanisms: (a) ingroup favoritism and (b) targeted outgroup rejection. Ingroup favoritism, also known as ethnic homophily, can be defined as the well-observed human tendency to identify with and favor proximate and similar others (in-groups) over distant and dissimilar ones (outgroups) (see e.g. Brown, 2000; Hornsey, 2008). Ingroup favoritism has been observed even when the definition of the group is based on arbitrary and trivial distinctions, which suggests ingroup mechanisms does not require cemented positive stereotypes (Tajfel, 1982). Recent research in social psychology additionally suggests that ingroup favoritism is a primary motivation, which operates independently of outgroup rejection (see Brewer, 1999; Greenwald and Pettigrew, 2014; Perry et al., 2018). Ethnic homophily could thus tilt employers’ decisions toward applicants with ‘native’ surnames against all other applicants in the application pool, regardless of the latter’s specific ancestry. Note that in this case applicants of Southern European descent would still be discriminated, though not due to any negative stereotypes about them in particular. As a mechanism, ingroup favoritism is therefore not consistent with the activation of negative stereotypes on Southern Europeans hypothesized in this study. In contrast, targeted outgroup rejection implies that employers discriminate against a specific ethnic group, which is precisely what we should expect if negative stereotypes of Southern Europeans in particular play a role in Northern employers’ hiring decisions.⁸

2.3 A placebo test

To distinguish between these two mechanisms of discrimination, we compare employer responses to applicants of both native and Southern European descent with their responses to identical applicants of French descent. Like Italy, Greece and Spain, France is a Mediterranean country and a member of the Eurozone. Yet due to its larger and more solid economy, France was better able to shoulder the impact of the Great Recession. Because French descendants are culturally close to Southern European descendants (particularly to those speaking Romance languages) but have not been subjected to negative stereotyping in the current European political context, they can be used as a placebo treatment in our analysis.⁹ The logic of our test is simple: if we find similar levels of discrimination against French descendants as we do for Southern European descendants, we will not be able to conclude that discrimination against the latter is driven by negative productivity-relevant stereotypes specifically targeted to Southern Europeans.

2.4 A target-consistency test

Note also that if negative stereotypes were indeed unleashed in the aftermath of the Euro debt crisis, we should find higher levels of discrimination against people of Greek and Spanish descent, compared with those of Italian descent, given that Greece and Spain, as recipients of bailout packages, were at the epicenter of the political debates surrounding the Eurozone crisis, whereas Italy was not. The distinction between bailed-out and not bailed-out countries thus provides a further test for target-consistency.

2.5 A diagnostic test

Our placebo and target-consistency tests can help us identify if discrimination is targeted—and hence consistent with negative stereotyping—but it cannot shed light on whether such targeted discrimination is driven by information deficits or else responds to employers’ taste and/or bias.¹⁰ To address this latter question, we test how Dutch employers, whose callback patterns strongly suggest targeted discrimination against Greek and Spanish descendants, respond to adding productivity signals in applicants’ résumés, thus contributing to the above-mentioned diagnostic literature on hiring discrimination.

2.6 Cross country expectation

Germany, the Netherlands and Norway are among the most advanced economies in the world and share a high degree of similarity in macro-level indicators, including crucially basic labor-market indicators (see summary measures in Table A1). This makes ceteris paribus assumptions (more) plausible, which in turn facilitates our discussion of country-level differences potentially associated with discrimination propensity. We expect discrimination against Southern European descendants to vary across the three countries of our experiment along three main dimensions: (a) the political climate; (b) the specific migration histories of each ‘minority’ group in each country and (c) the degree of institutional regulation of the hiring process. As noted above, although we cannot causally identify the effect of any of these dimensions individually, together they provide the theoretical underpinnings of our cross-country expectations. We discuss these dimensions in turn.

2.7 Political climate

The GEMM experiment was carried out over a period of 18 months, from 2016 to 2018. This period captures the aftermath of the Euro debt crisis. As discussed above, there is evidence that the Eurozone crisis reinforced negative stereotypes about Southern Europeans in Northern European media and politics. The German and Dutch contexts are particularly relevant as the two Eurozone countries (together with Finland and Austria) that most clearly championed severe austerity measures in exchange for partial (Spain) or total bailout funds (Portugal and Greece) at the time. It is in this specific political context that negative stereotypes about Southern Europeans were most clearly activated and used politically to blame bailed-out countries for their misfortunes. Because Norway is not part of the EU, the Eurozone debt crisis did not impose major costs or risks for Norwegian taxpayers. As a result, Norway’s political climate seems to have been less prone to fostering targeted stereotypes against Southern Europeans.¹¹

2.8 Migration histories

Positive stereotypes for long-established ethnic groups could protect them from the effects of a strained intra-European political climate. As discussed above, Southern Europeans have largely enjoyed a good reputation in old migration countries that received a significant influx of guest workers from the South after WWII.¹² The preferred destinations of these migrants were West Germany, France, Switzerland and Belgium, while the largest migration inflows came from Italy and Spain (followed by Greece). Germany is the country with the largest population of Southern European descendants, who make the second largest second-generation ethnic group after Turkish descendants (Algan et al., 2010). Comparatively high rates of intermarriage between post-war Southern European male migrants, particularly Italians, and German women have been interpreted as a clear sign of their successful assimilation in Germany (Klein, 2001). The Netherlands, on the other hand, met most of its post-war labor demands through migrants from its former colonies and, although a non-negligible inflow of Southern European guest workers joined the foreign-born Dutch workforce in the 1960s, most of these workers returned to their home countries (Tesser and Dronkers, 2007). As a result, there are no Southern European countries in the top-10 ancestry countries for people of foreign descent (allochtonen) in the Netherlands. Norway is a new immigration country outside the European Union and has never been a traditional destination for Southern Europeans.¹³ In accordance with social contact theory (Allport, 1955; Pettigrew, 1998), we should expect the existence of sizeable, long-established and well-regarded minorities (as it is the case of Southern Europeans in Germany) to increase the possibilities for positive social-contact, which, in turn, could help employers counteract the negative stereotypes bolstered by a strained intra-European political climate.

2.9 Institutional regulation of the hiring processes

In their comprehensive meta-analysis of correspondence tests on hiring discrimination in OECD countries, Zschirnt and Ruedin (2016) demonstrate that German-speaking countries show consistently lower rates of ethnic discrimination.¹⁴ They attribute this finding to the highly regulated nature of application procedures in these countries. Job applications in German-speaking countries typically require, not only a CV, photograph and cover letter, but crucially also official education and training reports, as well as reference letters from former employers. Such detailed and standardized application packages are expected to reduce the scope for statistical discrimination by reinforcing the reliability of the productivity signals included in applicants’ résumés (Zschirnt and Ruedin, 2016).

Table 1 summarizes cross-country differences across the three macro-level dimensions considered. Based on these differences, we expect hiring discrimination against Southern Europeans to be particularly high in the Netherlands, as this country combines intense negative stereotyping against Southern European countries, a low presence of second-generation Southern European communities (i.e. low chance for positive counteracting social contact), and unregulated application procedures (i.e. less reliable productivity signals). Discrimination in Germany should be significantly less intense given its long-established (and traditionally well-regarded) communities of Southern Europeans and because detailed application packages reduce the scope for statistical discrimination for all minorities. We also expect discrimination against Southern Europeans (targeted rejection) to be lower in Norway than in the Netherlands because, as a non-EU country, negative stereotyping of Southern Europeans was likely less common in the Norwegian media and politics in the first place.¹⁵ Finally, we expect differences in discrimination estimates between Greek and Spanish descendants, on the one hand, and Italian descendants, on the other hand, because, as discussed above, the nature of the Eurozone crisis implied that bailed-out countries (and their governments) were the main targets of negative stereotyping in public and political discourse.

3. Design, data and methods

3.1 Measuring discrimination

Field experiments can detect discrimination by observing the outcomes of employers’ hiring decisions in real-world settings. The most developed field experiments on hiring discrimination are the so-called correspondence (or résumé) tests. In correspondence tests, researchers send fictitious job applications to real job vacancies and record employers’ callbacks as measures of employers’ interest in each candidate. Fictitious applicants are identical in all relevant characteristics but the treatment/s of interest. Randomization of the treatment/s allows us to attribute any significant difference in employers’ callback to treatment effects. By capitalizing on the strengths of experimental and observational methods, field experiments provide the strongest basis for studying hiring discrimination (see reviews in Pager, 2007; Zschirnt and Ruedin, 2016; Bertrand and Duflo, 2017).

We draw on a sub-sample of data from the GEMM study, the largest field experiment on ethnic discrimination in hiring conducted in Europe to date (Lancee et al., 2019). Our analytical sample includes the responses of roughly 3600 firms to the same number of fictitious applications (cover letter and CVs), which were sent to real vacant jobs advertised online in Germany, the Netherlands and Norway. The experiment was conducted over a time span of 2 years, from 2016 to 2018.¹⁶ As in Ahmed et al. (2013) and Weichselbaumer (2017), the GEMM study used an unpaired design and sent one application to each vacancy. This design allows researchers to test the effect of multiple treatments simultaneously, while minimizing detection risks, harm to employers and masking and induced-competition biases.¹⁷ The GEMM study covers national labor markets for the same seven selected occupations, which were carefully chosen to provide variation in skills and customer contact.¹⁸ Together, these occupations cover between 15% and 20% of the workforce within each country. This design allows researchers to estimate average employer behavior within occupations and countries. Fictitious job applications include fixed characteristics, which are identical across applicants for the same occupation, and randomized treatments.

3.2 Fixed characteristics

All fictitious applicants are citizens of—and have obtained all their education and work experience in—the country of study (Germany, the Netherlands or Norway).¹⁹ Education varies as required for each occupation, while work experience is fixed to 4 years in the same sector of the job vacancy for all occupations (all applicants report having worked for two different companies in this period). Because there are obvious differences in the length of schooling required for each occupation, the age of applicants varies from 22 to 26 (being fixed for all applicants within each occupation).

3.3 Randomized treatments

Ancestry

The key treatment of this experiment is country of ancestry, which is defined as the country of origin of job applicants’ parents. The GEMM study included a total of 44 different ancestries, which were randomly assigned to each application within the following strata: 25% applicants of native ancestry, 25% for the two most representative minorities in each country and 50% randomly assigned to 31 different ancestries (see Lancee et al., 2019). All the non-native ancestries used in the present study come from this latter stratum. We use a total of five non-native ancestries, three Southern European ancestries, one ‘benchmark’ ancestry and one ‘placebo’ ancestry. Southern European ancestries are: Greek, Spanish and Italian. The benchmark ancestry is sub-Saharan African, which includes applicants of Nigerian and Ugandan parents. As explained above, we chose this latter ancestry as benchmark because we know sub-Saharan descendants are one of the most (if not the most) strongly discriminated groups across Europe and thus provide an obvious yardstick with which to compare the intensity of the other discrimination estimates (Weichselbaumer, 2017; Di Stasio et al., 2021; Polavieja, 2022). Finally, we use French descendants as a placebo ancestry to distinguish between ingroup favoritism and targeted outgroup rejection as two distinct drivers of ethnic discrimination, as explained above. The analytical sample of this study includes a total of 921 applications of non-native ancestry (roughly 150 for each single national group) plus over 2600 native applications (total N = 3596). This imbalance produces no estimation bias but implies that standard errors around parameter estimates for non-native descendants are large for single national groups (see below).

In the GEMM application packages, country of ancestry was indicated using three simultaneous signals: first, naming applicants using typical family and first names for the majority population of each country of ancestry. The names chosen were popular, recognizable as male or female and free from class (or religious) connotations (see Table 2 for the list of chosen names). Second, in addition to the respective home country language, a second mother tongue, for example, ‘Italian (mother tongue)’ was explicitly signaled in the skills section of the applicant’s CV. Finally, because names (and languages) are often imprecise signals of specific national origin, the cover letter contains a statement that the family of the job candidate migrated from the ancestry country to the region of the advertised job. All these three signals combined should convey clear information on country of ancestry (for further details, see Lancee et al., 2019).

Gender

Applicants are randomly assigned a male or female name and their gender (male or female) is explicitly indicated in the CV. Because gender is orthogonal to ancestry it needs not be controlled for in our statistical models.

Productivity signals

The GEMM study includes three different treatments associated with applicants’ potential productivity: (a) academic performance, (b) job performance and (c) soft skills. High academic performance was signaled by randomly providing information on high grades in 50% of the résumés (vs. no information), whereas the two latter treatments were signaled by randomly adding a self-descriptive paragraph in the cover letter, indicating, respectively, high job motivation leading to high performance (vs. no information) and the possession of soft skills (vs. no information).^20,21 Productivity treatments will allow us to test whether targeted discrimination is driven by information deficits, as explained above.

3.4 Outcome variable

Callback

Employers’ callbacks is a binary variable differentiating positive response (signal of interest) from negative response (no signal of interest) to each application. Firms can indicate their interest in applicants in three different ways. First, they can issue a formal invitation to the job candidate for an interview (invitation); second, they can inform the applicant that they have passed an early selection process (pre-selection) and third, employers can request specific additional information or ask the applicant to be called back, thus revealing interest in the candidate (interest). We take all three responses as positive signals of interest. The category ‘No signal of interest’ includes both explicit rejections of the job application or lack of response 12 weeks after the application is sent (automatic confirmation of receipt is not considered as a signal of interest). To minimize harm to employers, GEMM researchers promptly and politely declined any invitation to a job interview or request to provide additional information.

3.5 Estimation

We estimate country-specific logistic models with controls for occupational skill requirements.²² We regress employer callbacks on applicant ancestry using applicants of native-born parents as the reference category. The ancestry coefficients provide an estimate of average differences in callback probabilities within each country, with associated standard errors. To better gauge the magnitude of our estimates, we also provide callback ratios (CBRs) for significant ancestries. The CBR is the proportion of applicants of native descent that receive a positive response by employers relative to the proportion of applicants of foreign descent that receive a positive response. We calculate CBRs for Southern European ancestries, as well as for French and sub-Saharan African descendants.

We use three different specifications depending on the degree of (dis)aggregation of the Southern European category: in the first models, all Southern European descendants are grouped together and their average callback is compared with that of natives; in the second model, we split Southern European descendants into two groups: those associated with bailed-out countries (Greece and Spanish descendants) and those with Italian ancestry (not bailed-out) and finally, in model 3, the three Southern European ancestries are tested separately against applicants of native descent. We note that this latter model will yield large standard errors due to small-n (as we have roughly 50 observations per country of Southern European ancestry for each country of the experiment) and this increases the chances of Type II error—that is, failing to capture a true significant effect. Caution is thus recommended when interpreting significance tests in the most disaggregated specification. As explained above, all models include additional callback estimates for applicants of French (placebo) and sub-Saharan African (benchmark) descent, which will help us interpret the nature and the magnitude of discrimination against Southern European descendants. Finally, to test for the effect of information deficits on discrimination propensity, we use the Dutch sample (the country that most clearly shows targeted discrimination) to fit an interaction between ancestry and the number of productivity signals randomly appearing in the applicant’s résumé.

4. Results

Table 3 presents the results of fitting the first model (where all Southern European countries of descent are tested together) to each of the countries of the experiment using a logistic specification (odds ratios). We note that all our results are fully replicable using linear probability models (available upon request). Callback rates are shown graphically in Figure 1 (panels a–c). The combined model already reveals four of the five main findings of this study. First, we find no signs of discrimination against Southern European (nor French) descendants in Germany. The extent to which this is due to the effect of positive contact with long-established Southern European minorities, exposure to positive stereotypes, or due to the more detailed and standardized application procedures in Germany, we cannot tell.²³Second, in contrast to Germany, we find significant levels of discrimination against Southern European descendants in both the Netherlands (CBR = 1.2) and Norway (CBR = 1.4). Third, and in line with our expectations, discrimination against Southern Europeans seems targeted in the Netherlands, as we find no discrimination against equivalent French descendants; whereas in Norway we find all applicants with non-native parents receive significantly fewer callbacks than identical applicants of native descent, which suggests ingroup favoritism could be the main driver of discrimination in this country. Forth, we find applicants of sub-Saharan African descent consistently show the lowest callback rate in all three countries of this study, thus confirming the high level of discrimination experienced by this group in the European context.

Targeted discrimination, according to our own argument, should penalize Greek and Spanish descendants in particular because these two countries (together with Portugal) required bailout packages and this placed them at the epicenter of the Euro debt political crisis. Table 4 presents the results of splitting Southern European descendants into those associated with bailed-out and non-bailed-out countries, whereas Table 5 presents results fully disaggregated by country of ancestry. Callback probabilities and CBRs (for significant treatment effects) are presented graphically in Figure 2, which combines the results from Tables 4 and 5. We note that, as expected, discrimination in the Netherlands is specifically targeted to Greek and Spanish descendants, whereas no discrimination is found for applicants of Italian (or French) descent. Discrimination by Dutch employers against these two specific Southern European ancestries seems actually sizeable, as noted specially by the magnitude of our discrimination estimate for applicants of Spanish descent (CBR = 1.4), which is not too far off that found for applicants of sub-Saharan African descent (CBR = 1.5). Findings for the Dutch experiment thus seem fully consistent with targeted stereotype-driven discrimination. Although disaggregation makes the Norwegian picture somewhat more blurred, we still note our placebo (French descendants) and benchmark (sub-Saharan African descendants) tests strongly suggest ethnic homophily is the main driver of discrimination in this country—even if callback estimates for Italian (CBR = 1.2) and Spanish (CBR = 1.4) descendants do not reach standard levels of statistical significance. Again, no signs of discrimination are found in the German experiment for any of the analyzed ancestries but sub-Saharan African descendants. It may be interesting to note that our data show Spanish descendants in Germany receive higher callback rates than native German descendants, while Italian descendants have lower callback rates. That German employers are different in their responses to applicants of Southern European descent—when compared with Dutch or Norwegian employers—is confirmed by testing a country–ancestry interaction in a pooled model (see Table A2).

4.1 Does targeted discrimination decline with productivity signals?

Having established the existence of targeted discrimination against Greek and Spanish (but not Italian or French) descendants in the Netherlands, we now turn to test the extent to which this form of discrimination is primarily driven by information deficits, as proposed by statistical discrimination theory (Phelps, 1972). Figure 3 shows the results of fitting an interaction between discrimination propensity and the number of potential productivity signals randomly appearing in the application. We show the predicted callback probabilities of native descendants versus Greek and Spanish descendants (panel a), Italian descendants (panel b) French descendants (panel c) and sub-Saharan African descendants (panel d) (full models are presented in Table A3). The results of this interaction, which is significant at the 90% level for Greek and Spanish descendants (P > |z|= 0.086 in LPM estimation) but no other ancestry group strongly suggest targeted discrimination against these groups is significantly reduced when additional information on applicants’ potential productivity is introduced in the application. This finding suggests that Dutch applicants of Greek and Spanish descent need to show (up to three) more signs of potential productivity than equivalent native descendants to be called back by employers. This further reinforces our argument that targeted discrimination against these Southern European groups likely stems from negative productivity-relevant stereotypes—which employers might be likely to draw on to reduce contractual hazard in the absence of (contradicting) information. When such information is available, however, stereotypes seem to lose their signaling value for employers, as originally proposed by Phelps (1972). Interestingly, however, we note that discrimination against sub-Saharan African descendants seems largely resistant to the introduction of productivity signals, which suggests discrimination against this group is probably taste-based (or else driven by implicit bias).

5. Discussion and conclusions

Although the recent (re)emergence of national stereotypes in Europe has attracted the attention of researchers in the ethnic boundary tradition, their focus has been mostly placed on the symbolic/discursive aspects of ethnic boundary making, while paying comparatively less attention to the structural dimension of ethnic boundaries—that is, their potential to affect people’s life chances through processes of social closure. We have argued contemporary discourses contrasting industrious Northerners with leisurely Southerners have social-closure potential because they stress traits that are relevant for productivity. Our explanation of how stereotypes affect employers’ hiring decisions has built on insights from social and cognitive psychology and discrimination theories in economics, two literature strands that have too often run in parallel (see Pager, 2007). Using a broader sociological perspective, we have additionally discussed under which macro-level societal conditions, stereotype-driven discrimination against Southern European job applicants is more likely to occur. We have considered three national characteristics: the political climate (as the main activator of negative stereotypes), the specific migration histories of Southern European migrants (as a potentially crucial source of stereotype-neutralizing information), and the degree of standardization of the application process (as a key factor in reducing noise around applicants’ productivity signals). Two very clear empirical expectations have followed from our discussion, namely that targeted discrimination should be largest in the Netherlands and, relatedly, that targeted discrimination should be larger for applicants of Greek and Spanish descent (bailed-out countries) when compared with applicants of Italian descent (not bailed-out).

We have tested these predictions using a sub-sample of data from the GEMM study, a uniquely harmonized source of field-experimental data on ethnic discrimination in Europe. While randomization of treatments ensures our discrimination estimates are unbiased, small within-country sample sizes for individual foreign-ancestry groups inevitably produce large standard errors around discrimination estimates, thus reducing estimation precision. Statistical power issues notwithstanding, our results are consistent with the main empirical predictions that followed from our theoretical model. As expected, we find the largest levels of targeted discrimination in the Netherlands. Dutch employers are significantly less likely to call back applicants of Greek and Spanish (but not Italian) descent when compared with applicants of native descent, whereas they show no specific aversion toward applicants of French descent, which we have used as a placebo treatment. Discrimination estimates for applicants descending from our two bailed-out countries are sizeable, particularly for Spanish descendants, who, according to our estimates, would have to send 45% more CVs to get the same callbacks as applicants of Dutch descent (sub-Saharan African-ancestry applicants would have to send 50% more CVs).²⁴ Consistent with statistical discrimination models, we have further shown discrimination against Greek and Spanish descendants in the Netherlands significantly decreases with the number of productivity signals introduced in the application. This suggests that Dutch employers use targeted group stereotypes to fill informational gaps about the potential productivity of individual applicants of Greek and Spanish descent. In contrast, we find no evidence that the responses of Dutch employers to applicants of sub-Saharan African descent are sensitive to the number of additional productivity signals, which suggests discrimination against this group could be either taste-based or driven by implicit bias. This latter finding is particularly relevant because sub-Saharan descendants are probably the most discriminated minority in Europe (European Union Agency for Fundamental Rights, 2017). To our knowledge, ours is the first paper that suggests different discrimination mechanisms might be at play depending on the targeted group, which could explain why extant research on informational mechanisms finds contradicting results (for reviews, see Bertrand and Duflo, 2017; Thijssen et al., 2021).

Native to Southern European CBRs are even larger in Norway (the CBR for Greek descendants reaches 1.8), but, in this case, the evidence suggests discrimination is not targeted against Southern European ancestries in particular but seems to respond to a general mechanism of ingroup favoritism, whereby Norwegian employers seem to prefer applicants of Norwegian descent over everybody else. Had we not included a placebo treatment in the analysis, it would have been impossible to distinguish between targeted and untargeted discrimination, a distinction that bears great analytical import, but which has seldom been addressed empirically in the field-experimental literature on hiring discrimination. We believe future discrimination research will benefit greatly from including placebo treatments in experiment design.

Most children of Southern European migrants in Europe live in Germany. This is good news because we find no signs of discrimination against Southern European descendants in this country. We have claimed this might be partly due to the stricter job-application procedures that are typical of all German-speaking countries and partly due to the specific migration histories of Southern Europeans in Germany, which might have provided German employers with a reservoir of positive experiences and perhaps also favorable stereotypes with which to neutralize negative stereotypes coming from the media and the political arena. We note, however, all our claims about the specific drivers of cross-national differences in discrimination estimates are ultimately speculative, as we cannot possibly identify any single macro-level effect with only three countries included in this harmonized experiment.

Alternative explanations of our findings are of course possible and should be acknowledged. In line with the recent literature on intra-European boundaries, we have placed great emphasis on the impact of recession, recession-driven South–North migration and the Euro debt crisis as the main triggers of current stereotyping dynamics in Europe. Yet our discrimination estimates could admittedly respond to other sources of negative stereotyping, both proximate and distant. One such proximate source of negative stereotypes could be Northern Europeans’ own knowledge of Southern European countries, typically filtered through experiences of leisure. While all three countries included as Southern European ancestries in this study are tourist magnets for Northern Europeans, Greece and Spain have specialized in ‘sun and beach’ tourist packages, which could reinforce stereotypical images for these groups. A second alternative explanation concerns stereotypical images of Spanish descendants in particular, as we must consider the possibility that images for this group could additionally be tainted with exoticized (mis)conceptions of ‘Hispanic’ cultures more generally, which may lead Northern employers to perceive Spanish-ancestry applicants as a more distant outgroup.²⁵

It is also worthwhile to consider the possibility that the ‘exoticization’ of the Southern European category involves dynamic racialization processes. The example of the Dutch magazine cover discussed above might be particularly revealing of such processes. In this cover, the construction of the Southern European ‘other’ rests on multi-layered visual tropes of phenotypical difference and sexualized (dressed vs. undressed) social distance. Our findings could therefore be contextualized in terms of racialization processes, for example, liminal whiteness (see e.g. Magbouleh, 2017) and not only in terms of intra-European ‘cultural’ stereotyping. Yet, as the Elsevier cover itself illustrates, racialization and cultural stereotyping are likely intertwined processes and hence should be better understood as complementary rather than alternative forms of boundary making (see Wimmer, 2013, chap. 1).

A final complementary explanation worth considering is what we could call the long shadow of history. By this, we specifically mean the potential role of negative stereotypes construed in the context of imperial rivalries and Protestant hostility, particularly against the Spanish empire (see Greer et al., 2007). While we do not wish to deny that the so-called Spanish ‘black legend’ might have played an important role in cementing deeply seated negative stereotypes against Spain in Protestant Europe, especially in the Netherlands, which gained independence from the Spanish crown in 1648 after the Eighty Years’ War, going so far back to explaining current discrimination dynamics seems, in our view, unwarranted. The black legend might have provided a cultural sediment—that is, a certain script—that favors the rooting of contemporary stereotyping dynamics, but we believe it is these contemporary dynamics, and the contemporary political struggles they express, that better explain today’s intra-European ethnic boundary making processes. The severity of the discrimination estimates found for Greek and Spanish descendants in the Netherlands suggests that these boundaries are not just symbolic but social. We call for greater attention to these intra-European boundary-making processes previously overlooked in the socio-economic literature.

Footnotes

Acknowledgements

Previous drafts of this study were presented at the Research Network on Experimental Social Sciences Seminar, Spanish National Research Council; the Department of Applied Sociology, Complutense University of Madrid and the Department of Sociology and Communication, University of Salamanca. We wish to thank all seminar participants for their comments. We also wish to thank Frank Kalter, Edvard Larsen and the SER referees for their helpful input. The usual disclaimer applies.

Funding

This research is part of the project ‘Pushing the Boundaries of Research on Ethno-Racial Discrimination in Hiring’ (grant PID2020-119558GB-I00) funded by the Spanish Ministry of Science and Innovation (MCIN/AEI/10.13039/501100011033). The GEMM discrimination study, which provides the data for this article, was funded by the Horizon 2020 Program of the European Commission (grant H2020 649255).

References

Appendix