Multiple Discrimination against Female Immigrants Wearing Headscarves

Abstract

Western countries have experienced a large influx of Muslim immigrants, and concomitantly the Muslim headscarf has become the subject of major controversy. Drawing on theories of stigma, social identity, and multiple discrimination/intersectionality, this study examines the effect of wearing this headscarf in the German labor market. The author applies the method of correspondence testing that allows measuring discrimination in a controlled field setting. Findings show that when applying for a job in Germany, women with a Turkish migration background are less likely to be invited for an interview, and the level of discrimination increases substantially if the applicant wears a headscarf. The results suggest that immigrant women who wear a headscarf suffer discrimination based on multiple stigmas related to ethnicity and religion.

In Western countries, anti-Muslim sentiments and Islamophobia have grown substantially since the beginning of the 21st century. In the aftermath of September 11, 2001, an increase in anti-Muslim hate crimes has been noted particularly in the United States (ADC-RI 2008). Further, the US Equal Employment Opportunity Commission has reported a rise in religion-related complaints by Muslims since 2001, which suggests that Muslims feel increasingly discriminated against in employment matters. Negative stereotypes depict Muslims as “religious fanatics, terrorists, hostile, evil, barbaric, wild, backward, disorganized people who mistreat and oppress women” (Ghumman and Jackson 2010: 6). Such negative stereotypes can, of course, harm Muslims in their work and social lives.

Islamophobia, however, is not confined to the United States; it has also become more common in European countries, such as Germany, which this study examined more closely. Most Muslims in Germany are members of immigrant communities, and the majority have a Turkish background. During the shortage of labor in the 1960s and early 1970s, “guest workers” from Turkey were actively recruited. Contrary to Germany’s initial intentions, these guest workers ended up staying. Today they and their descendants form an integral part of the German population.

Estimates of the proportion of Muslims in the German population range from 1.9% to 5% (Spielhaus 2013). According to a representative survey (Bertelsmann Stiftung 2015) that was published just before the recent influx of Syrian Muslim immigrants into Germany, 57% of Germany’s non-Muslims perceived Islam as a threat. Of the survey respondents, 40% said they did not feel at home in their own country because of the country’s purported “Islamization,” and 24% believed that Germany should no longer permit Muslims to immigrate. Prejudice against Muslims has also been documented by the World Values Survey, which routinely asks respondents who, out of a list of people, they would not like to have as neighbors. In 2006, 24.8% of Germans were against the idea of Muslims as neighbors (World Values Survey 2006). Given that other groups do not necessarily trigger such a negative response by Germans (in comparison, “people of a different race,”“foreign workers,” and “people of a different religion” were rejected by 7.2%, 13.3%, and 4.3% of the respondents, respectively), it appears that Muslims currently constitute a particularly disliked “outgroup” in Germany.

The Muslim headscarf (or the hijab, which is a particular type of veil that Muslim women wear to cover their head and chest; see Rosenberger and Sauer 2012) is particularly controversial. As Helbling (2014) showed for six European countries (one of them Germany), attitudes of non-Muslims are significantly more negative toward the headscarf than toward Muslims in general.¹ Approximately one-quarter of survey respondents opposed Muslims, whereas nearly 60% disagreed with the practice of women wearing headscarves. According to another survey, 60% of Germans support a ban of Muslim headscarves in public places (Van der Noll 2010). Indeed, several European countries have implemented such laws. In some German federal states, teachers were banned from wearing headscarves in 2003, but in March 2015 the Constitutional Court in Karlsruhe ruled that an absolute ban is incompatible with religious freedom and is unconstitutional. With the recent influx of Muslim immigrants into Europe, however, the discussion on whether women should be allowed to veil their bodies in Muslim style reached a new peak. For example, in the summer of 2016, localities in coastal France stoked the debate by banning the burkini (full-body swimwear worn by Muslim women). France’s highest administrative court soon overturned the ban.

Given the prominence of the topic in the public eye, it is perhaps surprising that no economic study has hitherto empirically examined the effect of wearing the Muslim headscarf in the labor market.² The reason for this gap in the literature is, of course, that empirical data on women wearing headscarves is difficult to come by. Hence, I conducted a field experiment that compared the employment chances of females with backgrounds of migration, some of whom were wearing Muslim headscarves, with those of autochthonous German women.³ The goals of the study were to examine whether wearing the headscarf increased discrimination against immigrant women and to test whether female immigrants with a headscarf were confronted with “multiple discrimination” (see e.g., Ruwanpura 2008) based on ethnicity and religion. The method of sending out fictitious job applications (often called “correspondence testing”), which was applied in this study, is well established and was most famously used by Bertrand and Mullainathan (2004) to examine race discrimination. Recent reviews by Bertrand and Duflo (2016) and Neumark (2018) illustrate the significance of correspondence studies in the experimental literature on labor market discrimination.

The current study makes use of the specifics of the German application process, in which job seekers need to submit a large amount of material, including photographs, to be considered serious applicants. This use of photographs allows employers to see if a woman wears a headscarf. Reference letters from previous employers are also often included in German applications. They describe an employee’s tasks and performance in a firm as well as his or her behavior and personality. This study made use of this custom and included such letters (apparently from previous employers) in a fraction of all applications (see also Kaas and Manger 2012), characterizing the personality of the candidate (e.g., her friendliness and diligence) in a positive manner. The design therefore allowed measurement of the extent to which women wearing headscarves were able to reduce discrimination if they revealed positive information about their personalities.

Theoretical Background and Hypothesis

Social Identity Theory

Social Identity Theory is a theoretical framework that analyzes dynamics between social groups (e.g., Tajfel 1978) and may thus help to explain why women wearing the headscarf are especially discriminated against in the labor market. The theory argues that people gain self-esteem by attributing a higher value to their own social group (“ingroup”) than to others (“outgroup”). Along this same line, Akerlof and Kranton (2000) pointed out that social identity is crucial to our beliefs about who should do what job in the economy. For example, men may fear for their male identity if women do the same kind of work as they do. A “distaste” for women (Becker 1971) in male professions and a gender-segregated labor market may be the result. Similarly, non-immigrants may feel threatened and fear for their presumed superiority if immigrants, particularly Muslims, enter their ranks. To prevent this from happening, employers may exclude immigrants from their immediate workforce. Women wearing headscarves may experience this rejection particularly because they signal greater identification with the Muslim outgroup. Social Identity Theory suggests that individuals who identify more strongly with an outgroup are perceived as more distant by members of the ingroup. As a result, they suffer from increased levels of discrimination. Also, people whose stigmas are visible typically experience more discrimination than do others.

Stigmas

According to Goffman (1963), ethnicity and the headscarf can be interpreted as stigmas of group identity. Stigmas are individual attributes that are seen as personal flaws by mainstream society. As such, they often lead to devaluation, prejudice, and discrimination. Since stigmas are socially constructed, they are affected by normative cultural values and may change over time. For example, public perception of homosexuality has changed over recent decades, resulting in greater acceptance of gays and lesbians, whereas stigmatization of Muslims has increased significantly (Ragins 2008).

The social construction of stigmas also implies that social groups may judge specific attributes dissimilarly. Social identity has been defined as “that part of an individual’s self-concept which derives from his[/her] knowledge of his[/her] membership in a social group or groups together with the value and emotional significance attached to that membership” (Tajfel 1978: 63). Membership of a particular group may thus be based on the very characteristic (e.g., Muslim religion) that is devalued by the majority population. Hence, ingroup members may interpret specific group characteristics in ways that differ from how outgroup members interpret them. For example, while non-Muslims tend to see the Muslim veil as a stigma, Muslim women wearing it interpret it as a positive sign of humility which signals that the wearer is not overtly focused on the body and on materialistic values (Chapman 2016). Also, while many non-Muslims consider the headscarf to be an expression of women’s oppression and the rejection of gender equality (Shadid and van Koningsveld 2005), many Muslim feminists argue that they wear the headscarf by choice and see it as an empowering practice.

The value of and emotional attachment to one’s own social group (that is constitutive of one’s social identity) explain why individuals do not necessarily hide the stigmatized aspects of their identities even if they are concealable, like religion or sexual orientation (Ragins 2008; Tilcsik, Anteby, and Knight 2015). As Ragins (2008) emphasized by drawing on self-verification theory, “individuals have a basic need to affirm their identity, and this need prevails even if their identity is negative or devalued” (p. 198). To conceal a stigma takes a substantial toll on “individuals through psychological strain, emotional stress, and stress-related illnesses” (p. 194), which may be even harder to bear than the stigma itself.

Numerous studies have recently investigated the Muslim veil as a stigma (e.g., Ryan 2011; Chapman 2016; Eijberts and Roggeband 2016). They illustrate how veiled Muslim women challenge negative evaluations of their identity not only by showing that they are “good people” who “have something to offer” (Ryan 2011: 1052) but also by refashioning the veil in a modern way that challenges the conventional stigma while continuing to affirm their Muslim identity (Chapman 2016). Other responses of Muslim women to their stigma are compensation and circumvention (Eijberts and Roggeband 2016), that is, working particularly hard to try to “make up” for their stigma and avoiding adverse situations (e.g., working for a prejudiced employer). Job applicants who want to screen for unbiased companies can do so by attaching a photograph showing them wearing a headscarf (a procedure adopted in this study).

Multiple Discrimination/Intersectionality

The concept of “multiple discrimination” or “intersectionality” has barely been incorporated into economics (exceptions are Brewer, Conrad, and King 2002 and Ruwanpura 2008), even though Akerlof and Kranton (2000: 731) mentioned that identities are “associated with multiple social categories.” Anti-discrimination law is traditionally aimed at specific groups, for instance, at women or at ethnic, religious, or sexual minorities. This focus on a single category of inequality has been called the “unitary” approach. For example, in Germany the General Equal Treatment Act (Allgemeines Gleichbehandlungsgesetz [AGG]) prohibits discrimination based on race or ethnic origin, gender, religion or belief, disability, age, or sexual orientation. Recent research in law, political science, sociology, and gender studies, however, has emphasized that people belong to many groups and can be “discriminated against on the basis of more than one identity category” (Kantola and Nousiainen 2009: 461). For example, immigrant Muslim women may be treated unfavorably because of their sex, their ethnicity, and their religion. Another term used to emphasize how group identities interact and lead to unique forms of discrimination is “intersectionality” (Crenshaw 1989). Numerous recent papers have focused on the intersectional situation of Muslim women with headscarves in society (e.g., Bilge 2010).

Drawing on the theoretical insights from the literature on social identity, stigma, and multiple discrimination, my main research hypothesis is the following: Immigrant women wearing headscarves are discriminated vis-à-vis autochthonous German women, and the level of discrimination they experience is higher than that of immigrant women who do not wear the headscarf. The reason is that immigrant women who wear headscarves are perceived as more distant and are stigmatized not only because of their immigrant background or ethnicity but also because they wear a symbol of the Muslim religion. Multiple discrimination is the result. A sub-hypothesis is also tested as to whether reference letters that describe the personality of the candidate in a positive manner reduce the level of discrimination.

Previous Studies of Unfavorable Treatment of Muslims

Table 1 gives an overview of studies that have previously examined the effect of Muslim religion or Muslim attire. The upper section of Table 1 covers correspondence tests that focused on the Muslim religion of immigrants. In correspondence testing experiments, fake application letters that signal identical qualifications but different demographic characteristics are sent to companies. If, despite identical qualifications, one type of applicant is invited for an interview more often than another type, this is considered discrimination. Most of the studies listed used work and/or volunteer experience with religious organizations in their applicants’ résumés as signals for religion. All of them found lower callback rates for the Muslim candidates.

Table 1.

Studies Examining the Effect of Muslim Religion or Muslim Attire on Job Opportunities

	Authors	Methoda	Country	Signals for religion/headscarf used in study	Results
Muslim religion	Adida et al. (2010)	CT	France	Volunteer experience with Muslim/Catholic organizations in résumés	–	Muslim applicants have lower callback rates than Catholics (72% less likely to receive callback).
	Pierné (2013)	CT	France	Volunteer experience with Muslim/Catholic organizations in résumés	–	Muslim applicants have lower callback rates than Catholics (36% less likely to receive callback).
	Wright et al. (2013)	CT	US	Involvement in religious (e.g., Muslim, Catholic) student organizations in résumés	–	Muslim applicants have lower callback rates than applicants without signal for religion (24% less likely to receive callback); Catholics have no disadvantage.
	Acquisti and Fong (2015)	CT	US	Religion indicated in social network profile	–	Muslim applicants have lower callback rates than Christians, but only in Republican states (in the latter: 87% less likely to receive callback).
Muslim attire	Unkelbach et al. (2010)	LS	Germany	Photos of women with or without a hijab	–	With hijab: more likely rejected; rejection faster, positive responses slower.
	Ghumman and Jackson (2008)	LS	US	Photos of individuals with or without religious identifiers (e.g., turban or hijab) in résumés	+ / –	Highest employability ratings for Muslim women and lowest for Muslim men
	Ghumman and Ryan (2013)	AS	US	Female real-life testers with or without a hijab	–	With hijab: fewer offers to complete a job application, fewer callbacks, higher perceived negativity.
	King and Ahmad (2010)	AS	US	Female real-life testers with or without hijab and abbaya	~ / –	No difference in job callbacks; with hijab: interviews shorter and higher levels of negativity.

^aMethod of study: AS, audit study; CT, correspondence testing; LS, laboratory study.

The effect of wearing a headscarf or a hijab on employment chances has so far been barely researched. In particular, no correspondence study has tested its effect. The studies that do exist are summarized in the lower section of Table 1. In two laboratory studies, students were shown profiles and photos of job applicants, some of whom wore a hijab. In two “audit studies,” female testers, who wore a hijab in half of the trials, applied for jobs in person. The studies yield inconclusive results ranging from preferential treatment of women with a hijab to discriminatory treatment of the women. This ambiguity of results is most likely attributable to methodological issues. For example, laboratory studies may not sufficiently represent real-life scenarios, and the audit studies reviewed have very small sample sizes. Correspondence tests are typically considered to circumvent most conventional methodological problems (Bertrand and Mullainathan 2004). Nonetheless, they measure only the level of discrimination at the first stage of hiring. As Rich (2014) showed on the basis of previous studies though, the vast majority of discrimination in hiring occurs at this initial stage (between 86% and 94%).

Research Context and Experimental Design

Advantages of the German Setting

The study presented here involved a correspondence testing experiment conducted to investigate whether wearing a headscarf significantly affects the employment chances of women with a Turkish background. To isolate the effect of the headscarf, the effect of migration background was also explicitly examined. The study design made use of the large amount of application material that job seekers typically submit in German-speaking countries. In other countries, such as the United States and the United Kingdom, it is not usual to include photographs with résumés precisely because they can enable discrimination. As a result, attaching photographs would be considered awkward and would raise suspicion. In Germany, however, attaching one’s photograph is the norm. In a correspondence study, a photograph provides an ideal opportunity for indicating that the applicant wears a headscarf. In addition, diplomas and certificates are included in German applications. For example, German high school diplomas detail students’ school grades for ten subjects over four semesters in addition to final examination results for four subjects (every subject is evaluated on a 0–15 scale). They also give five different measures of the overall performance of a student in the final years; for instance, the measure “overall points” takes a value between 280 and 840. As this example indicates, German diplomas and certificates give very rich and detailed information regarding students’ educational achievements. Other information typically provided in German applications includes date of birth, marital status, and citizenship.

The necessity of attaching matched diplomas makes correspondence testing experiments very time-consuming in the German setting. The advantage of the large amount of information needed is that it leaves little ambiguity about an applicant’s productivity, which makes statistical discrimination less likely. Phelps (1972) and Arrow (1973) stated that statistical discrimination occurs if information about an individual’s productivity is missing and employers therefore resort to group averages to infer an individual’s productivity. Altonji and Pierret (2001) and Sterling and Fernandez (2018) showed how statistical discrimination relates to the amount of information employers hold on workers. Even though German applications are unusually rich compared to those of other countries (see also Zschirnt and Ruedin 2016), it is, of course, impossible to cover all aspects of an individual’s productivity in an experiment.

Identities Tested

In this experiment, job applications for three fictitious female characters with identical qualifications were sent out in response to job advertisements: one applicant had a German name, one a Turkish name, and one had a Turkish name and was wearing a headscarf in the photograph included in the application material. To indicate a Turkish background, or a lack thereof, I chose first and last names that are common in Turkey (Meryem Öztürk) and Germany (Sandra Bauer). These names were tested with 104 business and economics students, 99% of whom identified the name Meryem Öztürk as Turkish, and 100% of whom identified the name Sandra Bauer as German.

All applications included photographs of the same person, who was wearing a headscarf in some of the cases. The three identities tested in this experiment are shown in Figure 1: Sandra Bauer (bareheaded), Meryem Öztürk (bareheaded), and Meryem Öztürk with a headscarf.⁴ For the fabrication of suitable photographs, a model was hired who could pass equally well as German and as Turkish. That the model was perceived as equally likely to have German or Turkish roots was verified empirically (see Online Appendix). I consulted representatives of the Turkish community in Berlin (Türkische Gemeinde e.V.) to ensure that the portrayals of the Turkish women included in the study were realistic. They also helped with the final selection of photographs. We chose a very modern binding of the headscarf (throat not covered) to signal that the applicant was a young, modern woman who could easily fit into a secular environment and to ensure that she would not be perceived to hold particularly fundamentalist ideals (see Online Appendix for a description of a pretest). By choosing this representation of a Muslim woman, discrimination against her should be minimized.

Figure 1.

Names and Photographs, Indicators of Identity

Apart from the name and the photograph, the applications were strictly identical (e.g., with respect to schooling and job training in Germany and German citizenship). Nothing but her name connected Meryem Öztürk to Turkey. The résumé gave no indication if Meryem Öztürk spoke Turkish (only English was mentioned in addition to German). Thus, Meryem Öztürk would be perceived as a second (or later) generation immigrant, even though she is hereafter referred to as “immigrant” for simplicity. Similarly, the job application contained no signal of religion aside from the headscarf, which Meryem Öztürk may wear only out of cultural tradition.

Randomized Application Procedure

Most correspondence testing experiments send multiple applications, and some even mail a whole battery of comparable résumés to one company. This approach may be problematic in certain circumstances though. As Weichselbaumer (2015) illustrated, sending multiple applications to one firm can lead to biased results, at least in Germany, because the results of correspondence tests have been widely discussed in Germany’s popular media. The method is therefore increasingly well known among personnel managers. Consequently, employers may notice the similarities between fictitious applications, suspect that they are being tested, and present themselves as more minority-friendly than they actually are. The problem is aggravated particularly by the large set of application documents required in Germany. To maintain strict comparability of applicants within the experiment, all documents such as résumés and school reports must be matched. Paired fictitious applications therefore stand out even more from the heterogeneous pool of non-experimental applications.

To avoid detection, this experiment thus followed Weichselbaumer (2015) in that only one profile was sent to each firm. This protocol has the advantage that all applications sent can be virtually identical except for the variables of interest. It also allows photographs of the same person to be used for all identities. By showing the same model in the same pose for all identities, physical appearance was automatically held constant across ethnic backgrounds. The identity of the applicant (German name, Turkish name/no headscarf, Turkish name/headscarf) was randomly assigned to each application by variation of name and photograph.

Occupations

In this study I focused on the employment chances of office workers, that is, secretaries, accountants, and chief accountants. In Germany, chief accountants are responsible for the monthly, quarterly, and annual financial statements and the annual tax declaration. That way, chief accountants do the most sophisticated accountancy work, but do not—as the name might suggest—manage the accountancy department or supervise staff.

The three occupations chosen cover a large fraction of female employees. According to the 2011 German census, company management and organization (which covers various office jobs, especially secretarial work) is the most common occupational field for females and covers a significant share of women with migration background (Zensus 2011). Although all three occupations tested are office jobs, they vary in terms of education required and thus in status: The job of chief accountant is most highly regarded, as it requires advanced training in accounting (as illustrated by a corresponding certificate), followed by accountant and secretary in this order.

What results may be expected from the comparison of occupations? First, the level of education required of the fictitious candidates was highest for chief accountants and lowest for secretaries. It is often assumed that discrimination decreases with increasing level of education because high educational achievements require characteristics (e.g., intelligence, diligence, commitment) that are incompatible with conventional negative stereotypes. Second and relatedly, positions that require higher levels of human capital are typically more difficult to fill. According to Becker (1971), one would assume that discrimination is more likely to occur if employers have a wide range of potential workers to choose from (see also Baert, Cockx, Gheyle, and Vandamme 2015). Third, in contrast to secretaries, accountants have little if any customer contact (AMS Berufslexikon 2019), so one would expect customer discrimination (Becker 1971) to drive up the level of discrimination for secretaries compared to that for accountants. Further, coworker contact, the source of coworker discrimination (Becker 1971), is similar or higher for secretaries than for accountants. Consequently, my hypothesis with respect to occupations was that discrimination would be highest for secretaries and lowest for chief accountants.

Application Material

Following German norms, each application used in this experiment consisted of the following: a letter of application, a résumé, a photograph, a four-page high school diploma (certifying university entry qualification), and a certificate of successful completion of an apprenticeship as an office clerk. The applications for chief accountants also included a certificate of advanced training in accounting. The documentation provided in German applications allows researchers to control for many characteristics of an applicant. For example, examination results for various fields of study, which in Germany are provided in diplomas and certificates, are indicative of the cognitive abilities of a candidate. Thus, in this study, educational achievements were held constant across job applicants. However, statistical discrimination may also be driven by employers’ beliefs about the personality traits of an applicant. In particular, it may be that stereotypes ascribe less favorable personal traits to migrants, especially to those wearing headscarves.

Reference letters that are frequently attached to German applications give information about performance, behavior, and personality of a worker. I therefore employed reference letters to test whether statistical discrimination with respect to these characteristics occurs (Kaas and Manger 2012).⁵ Specifically, the following procedure was implemented: Only a fraction of the applications had a reference letter attached from the company at which the applicant was trained. This letter described the tasks the applicant had performed and the areas in which she was trained, assessed her performance and behavior, and characterized her personality in a positive manner. I prepared the letter on the basis of reference letters of real-life job applicants to signal above-average performance and behavior. German reference letters work with standardized phrases (e.g., provided by human resource guidebooks) that signal a particular level of quality. For the experiment, I implemented phrases that signaled a “good” level of quality, which is the second-best of six available quality levels and is considered “above average” (Huber and Großblotekamp 2006: 65). In addition, the letter referred to the applicant’s fast comprehension, dedication to learning, diligence, reliability, and promptness. Thus, this experimental design enabled testing whether immigrants can reduce the level of discrimination they encounter if they can signal positive personal traits.

I created the documents in strict adherence to rules defined by the German Federal Anti-Discrimination Agency that legally covers testing procedures (Klose and Kühn 2010). Given that only one application was sent to each firm, the applications were identical in every respect except name (and thus also phone number and e-mail address) and photograph. The identities were assigned randomly to the job openings. At the time of the experiment, Sandra Bauer and Meryem Öztürk were 27 years old. On average, Turkish immigrants marry at a significantly younger age than do German non-immigrants. Without information to the contrary, employers would therefore expect the immigrant to most likely be married and the non-immigrant to be single. As marital status may affect employment chances, it was useful that such information is typically provided in German applications. Within the experiment, I identified all candidates as single. Meryem Öztürk thus signaled that she did not resemble the “average female with a Turkish background” and was committed to the labor market.⁶ Thus, the measure for discrimination did not become confounded with effects of marital status.

After primary and secondary schooling, the applicants received their high school diplomas (Abitur) and undertook apprenticeships as office clerks. Afterward, the job candidates had four years of job experience as office clerks or accountants. All applications included information on IT skills and foreign language abilities (English). With more recent immigrants, employers are concerned about language proficiency, transferability of schooling and training acquired abroad, as well as labor market access. In this experiment the fictitious résumé did not leave any uncertainties in these dimensions. It was beyond question that all applicants would be fully proficient in the German language, have full labor market access, and hold all the commonly accepted qualifications.

Procedure

For approximately one year, various popular online job portals were searched weekly for relevant job ads in the six largest cities of Germany: Berlin, Frankfurt, Hamburg, Cologne, Munich, and Stuttgart. In addition, Dresden was included in this study to cover one city located in the former GDR. The application documents were combined in one electronic file that was sent to the companies by way of e-mail. The applications provided full contact information: an address, an e-mail address, and a cell phone number that would forward to voice mail. Including these details allowed companies to contact the fictitious applicants. I restricted the applications to openings for full-time jobs whose basic requirements matched the profile of the fictitious candidates. Because only one application was sent in response to each job posting, comparability of the vacancies included in the study had to be ensured. Each advertisement was therefore carefully checked for its suitability and coded for how well it matched the standardized profile of the applicant. This information was used later as a control variable. If a personnel recruitment agency was involved in the selection process, the corresponding posting was omitted. To avoid detection, each company was contacted only once, even if it advertised multiple suitable jobs in the course of the experiment. All remaining companies that welcomed applications by e-mail were contacted. Invitations for interview and inquiries stating interest in the candidate were coded as positive responses (callback). If the applicant was invited for interview, the appointment was canceled within a day to avoid any inconvenience to the company.

Results

In the course of the experiment, I sent a total of 1,474 applications in response to job advertisements for companies located in Germany (for summary statistics and correlation matrix see Table A.1 in the Online Appendix). The applicant with the German name, Sandra Bauer, was the most successful (see Table 2, column “All occupations”). She received positive feedback from 18.8% of all companies she applied to, followed by the applicant with the Turkish name (Meryem Öztürk) without headscarf, who was contacted by 13.5% of the companies. This difference is statistically significant (t = 2.2633, p = 0.012, one-sided t-test). As hypothesized, the applicant with the Turkish name who wore the headscarf was the least successful—she received positive feedback from only 4.2% of all companies contacted. The difference to both other identities is strongly significant (Ms. Öztürk with headscarf versus Ms. Bauer: t = 7.3975, p = 0.000, one-sided; Ms. Öztürk with headscarf versus Ms. Öztürk without headscarf: t = 5.2395, p = 0.000, one-sided). This result implies that the candidate with the headscarf has to send 4.5 times as many applications as an identical applicant with a German name and no headscarf to receive the same number of callbacks for interview. Compared with Ms. Öztürk without headscarf, she has to send 3.2 times as many applications. These findings confirm the main hypothesis of the article: Discrimination occurs against female immigrants, and the level of discrimination increases if they wear a headscarf, which points to multiple discrimination.

Table 2.

Callback Rates and Relative Callback Rates by Occupation

Applicant identity	All occupations		Chief accountant		Accountant		Secretary
	Callback rate (%)	Ratio†	Callback rate (%)	Ratio†	Callback rate (%)	Ratio†	Callback rate (%)	Ratio†
Bauer (German name, bareheaded)	18.8 (N = 474)		27.3 (N = 77)		29 (N = 107)		12.8 (N = 290)
Öztürk (Turkish name, bareheaded)	13.5 (N = 498)	1.39**	19 (N = 84)	1.44	19 (N = 121)	1.53*	9.6 (N = 293)	1.33
Öztürk with headscarf (Turkish name with headscarf)	4.2 (N = 502)	4.48***	3.6 (N = 83)	7.58***	5.8 (N = 121)	5.00***	3.7 (N = 298)	3.46***

^†The ratio (or “relative callback rate”) is calculated by: callback rate German name/callback rate Turkish name (bareheaded or with headscarf).

^***p < 0.01; **p < 0.05; *p < 0.1.

Table 2 also illustrates that while callback rates differed between jobs (as predicted, accountants were in higher demand than secretaries), discrimination was always highest against Meryem Öztürk when she was wearing a headscarf.⁷ As the relative callback rates (“ratio”) illustrate, differences in the extent of her disadvantage emerged across occupations, but not as anticipated. In the profession of chief accountant, Meryem Öztürk with a headscarf had to send almost 7.6 times as many applications as Sandra Bauer to receive an invitation for interview. The corresponding number for secretaries was “only” 3.5. Thus, the hypothesis that discrimination against the immigrant with the headscarf is highest for secretaries is rejected: Indeed, the opposite is the case, even though the low callback rates for the applicant with the German name confirm that positions for secretaries were easiest to fill. This result is at odds with conventional considerations regarding, for instance, customer and coworker discrimination or the relation between competition and discrimination. One reason may be that wearing a headscarf and thus revealing a stigma is considered even more inappropriate for positions with higher occupational status—an assessment that employers and coworkers may share. It may also be more costly to replace a chief accountant than a secretary if an employment relationship does not work out, so employers may tend to make more conservative decisions when filling a chief accountant position. Another possible explanation is that trust may play a great role in who employers want to assign to the position of a (chief) accountant. As studies have shown (Brewer 2016), it may be the case that outgroup members are trusted less than ingroup members, which may explain the high level of discrimination in the profession of (chief) accountants.

Probabilities of a Callback

A linear probability model was estimated to examine the probability of a positive callback more thoroughly. Since applications were sent to different firms, differences in invitation rates may have been caused by differences in firm or job characteristics of the positions rather than discrimination, even though job postings had been carefully checked for comparability and randomly assigned. To account for such potential differences, a number of characteristics of job vacancies and the corresponding firms were controlled for. For example, applications in particular cities, sectors, or months may be systematically more successful. Further, only a proportion of applications included a reference letter. Employers may consider application packages that include a reference letter more complete and treat them more favorably. The results of the linear probability model are presented in Table 3. The specifications incrementally include the following control variables: occupational dummies (chief accountant, accountant, and secretary), city dummies (Berlin, Dresden, Frankfurt, Hamburg, Cologne, Munich, and Stuttgart), sector dummies (trade, public services, manufacturing, and services), time dummies (in quarters), attachment of a reference letter (yes/no), variables concerning size and internationality of a company (“firm characteristics”), and a number of job characteristics that were derived from the requirements stated in the job advertisement—more specifically, whether the job required a special qualification that the standardized profile did not have (“special qualification requirement” [SQR]) or whether the job involved team and/or customer contact.

Table 3.

Probability of a Callback (linear probability model)

	(1)	(2)	(3)	(4)	(5)	(6)
Variable	Callback	Callback	Callback	Callback	Callback	Callback
Öztürk	−0.053**	−0.055**	−0.054**	−0.054**	−0.055**	−0.057**
	(–2.26)	(–2.37)	(–2.30)	(–2.31)	(–2.35)	(–2.44)
Öztürk with headscarf	−0.146***	−0.148***	−0.147***	−0.148***	−0.148***	−0.149***
	(–7.27)	(–7.40)	(–7.37)	(–7.37)	(–7.39)	(–7.38)
Chief accountant		0.079***	0.074***	0.075***	0.076***	0.077***
		(3.16)	(2.94)	(2.95)	(3.01)	(3.03)
Accountant		0.091***	0.090***	0.090***	0.088***	0.088***
		(4.12)	(4.07)	(4.08)	(3.96)	(3.98)
Reference letter				−0.012	−0.011	−0.012
				(–0.53)	(–0.49)	(–0.53)
Job characteristics
Special qualification requirements (SQR) (Yes = 1)					−0.030	−0.031
					(–1.49)	(–1.49)
Team and/or customer contact (TCC) (Yes = 1)						0.004
						(0.21)
Firm characteristics
Small firm (Yes = 1)						0.004
						(0.15)
International firm (Yes = 1)						0.001
						(0.07)
Time dummies	No	No	Yes	Yes	Yes	Yes
City dummies	No	No	Yes	Yes	Yes	Yes
Sector dummies	No	No	Yes	Yes	Yes	Yes
Constant	0.188***	0.155***	0.129***	0.136***	0.139***	0.139***
	(10.46)	(8.74)	(4.09)	(4.03)	(4.14)	(3.87)
Number of observations	1,474	1,474	1,474	1,474	1,474	1,467
R 2	0.034	0.051	0.061	0.062	0.063	0.063
Adjusted R2	0.0330	0.0486	0.0511	0.0506	0.0512	0.0493

Notes: Robust t-statistics in parentheses. Dependent variable is a dummy for receiving a positive callback. The reference category for identity: Bauer; for occupation: secretary.

^***p < 0.01; **p < 0.05; *p < 0.1.

The results in Table 3 demonstrate that the effects found for the different identities are robust to adding these control variables. Hence, the unfavorable treatment of both job applicants with Turkish names cannot be explained by particular characteristics of either the specific firms or the jobs these women applied to in the randomized application design. In comparison to the applicant with the German name, the bareheaded applicant with the Turkish name (Öztürk) was consistently less successful by 5 to 6 percentage points, whereas the immigrant applicant who also wore a headscarf was almost 15 percentage points less likely than the applicant with the German name to receive a callback for interview.⁸ A test for the equality of coefficients shows that the differences between coefficients are highly significant (at the 1% level). These results again confirm the main hypothesis of the paper: Immigrant women wearing the headscarf are multiply discriminated against. Note that even though the inclusion of a reference letter is often considered standard practice in Germany, the results show that its attachment does not generally lead to a positive effect. In the current setting, this outcome may be attributable to these letters being constructed to give a positive (i.e., “good”) but not excellent impression. Further, if a job advertisement required an unusual qualification (SQR) that the applicant did not have, this did not reduce her callback rate at a significant level. The reason is probably that the fake applicants held typical qualifications for a particular position. If they did not have a special qualification, only a few others would. Similarly, the requirement of team and/or customer contact did not affect the overall probability of a callback.

Reasons for Unequal Treatment

In this study I also tested a set of hypotheses concerning the reasons for unequal treatment. For example, Muslim immigrants may suffer from statistical discrimination if they are ascribed personality traits that are deemed less productive in the job. If this is the case, attaching an identical reference letter should help immigrants more than non-immigrants, as negative beliefs about immigrants are counteracted by positive characterizations. Also, as mentioned above, not only employers but also customers and coworkers may hold discriminatory preferences. For this reason, I created the variable “team or customer contact” (TCC), which captures whether a job advertisement mentioned that such interaction was required at the job. It is also possible that larger firms discriminate less because they follow a more standardized, and therefore more objective, evaluation process than small companies do (Kaas and Manger 2012). For this reason, in the present study the dummy variable “small firm” measured whether a firm had 1 to 20 employees. Companies that operate at an international level may also behave differently toward immigrant applicants because they may be more familiar with diverse groups of people. This possibility was examined with the dummy “international firm.” Further, discrimination may be triggered by employers’ beliefs that immigrants are less proficient in the German language. The variable “German” measured whether a job advertisement emphasized that proficiency in German was required for the job. In this experiment, fluency in German was signaled for all applicants by schooling and training received in Germany. Their callback rates should therefore have been unaffected by German language requirements. Nonetheless, employers may use statements regarding German proficiency in job advertisements as subtle cues against immigrant employees in general, irrespective of their language skills. Whether this is actually the case was tested empirically.

Another interesting question would be how the level of discrimination is affected by the attitudes of recruiters, but it is difficult to obtain proxies for these. In Germany, clerical assistants usually handle the administration of the application process, and thus the identities (e.g., sex and migration background) of recruiters with decision-making power remain unknown. For this reason, I constructed a variable that examined whether a company actively seeks immigrant workers. Some firms explicitly asked for cross-cultural competence in their job advertisements or signaled that they are an intercultural team, which was captured by the dummy variable “interculturalism.” It is likely that “intercultural firms” discriminate less. Finally, some job advertisements mentioned that a neat appearance was required for the job (“appearance”). If conventional looks are of great importance in a certain job, firms may be particularly unwilling to hire an immigrant woman with a headscarf.

To test these hypotheses, in Table 4 I extended the linear probability model to cover not only the full set of control variables (now also including “German,”“interculturalism,” and “appearance”—see column (1)) but also the interaction of each control variable with the variable “Öztürk” and with “Öztürk with headscarf” (from column (2) on). The interaction effects illustrate which variables affect immigrants differently from how they affect autochthonous Germans, and therefore help to identify reasons for differential treatment. Column (2) in Table 4 includes the key interaction effects that test whether the inclusion of reference letters reduces discrimination and whether large firms and firms that indicate an affinity for interculturalism discriminate less. Column (3) adds interaction effects for occupation, column (4) for the other remaining variables. Column (5) presents the fully interacted model. As can be seen, few interaction effects are significant at conventional levels. As mentioned above, Meryem Öztürk with a headscarf was discriminated against to a greater degree in the accountant and chief accountant occupations. Otherwise, small firms may discriminate more, but these effects are, at best, significant at the 10% level. No effects were found concerning special qualification requirements (SQR), team or customer contact (TCC), internationality of a company, and German language requirements. The hypothesis that firms expecting a “neat appearance” from their applicants discriminate more against the headscarf than do others was not confirmed. A marked effect, however, was identified for firms that asked for cross-cultural competence or signaled that they were an intercultural team. This finding indicates that, everything else being equal, Meryem Öztürk with a headscarf increases her relative callback chances by 37 to 39 percentage points if she applies to such a firm (and if she does not wear a headscarf: by 27 to 32 percentage points). Note, however, that only a few firms fall into this category (approximately 2%).

Table 4.

Reasons for Unequal Treatment (linear probability model)

	(1)	(2)	(3)	(4)	(5)
Variable	Callback	Callback	Callback	Callback	Callback
Öztürk	−0.060**	−0.083***	−0.060*	−0.088	−0.053
	(–2.54)	(–2.76)	(–1.84)	(–1.59)	(–0.97)
Öztürk with headscarf	−0.153***	−0.171***	−0.113***	−0.183***	−0.114**
	(–7.60)	(–6.44)	(–3.84)	(–3.79)	(–2.27)
Chief accountant	0.078***	0.079***	0.144***	0.081***	0.161***
	(3.09)	(3.13)	(2.68)	(3.24)	(2.90)
Chief accountant*Öztürk			−0.050		−0.075
			(–0.71)		(–1.04)
Chief accountant*Öztürk with headscarf			−0.143**		−0.157***
			(–2.43)		(–2.59)
Accountant	0.087***	0.087***	0.157***	0.088***	0.164***
	(3.81)	(3.82)	(3.24)	(3.89)	(3.31)
Accountant*Öztürk			−0.061		−0.077
			(–0.99)		(–1.21)
Accountant*Öztürk with headscarf			−0.142***		−0.145***
			(–2.65)		(–2.61)
Reference letter	−0.010	−0.074*	−0.070*	−0.076*	−0.074*
	(–0.44)	(–1.93)	(–1.82)	(–1.90)	(–1.85)
Reference letter*Öztürk		0.105**	0.101**	0.110**	0.109**
		(2.16)	(2.07)	(2.16)	(2.13)
Reference letter*Öztürk with headscarf		0.080**	0.077*	0.085**	0.082*
		(1.99)	(1.90)	(2.00)	(1.94)
Special qualification requirements (SQR) (Yes = 1)	−0.033	−0.031	−0.035*	−0.065	−0.072*
	(–1.60)	(–1.54)	(–1.74)	(–1.56)	(–1.69)
SQR*Öztürk				0.047	0.054
				(0.81)	(0.92)
SQR*Öztürk with headscarf				0.057	0.054
				(1.21)	(1.14)
Team and/or customer contact (TCC) (Yes = 1)	0.004	0.002	0.002	−0.008	−0.010
	(0.21)	(0.12)	(0.12)	(–0.22)	(–0.27)
TCC*Öztürk				0.031	0.033
				(0.63)	(0.67)
TCC*Öztürk with headscarf				−0.005	0.000
				(–0.11)	(0.01)
Small firm	0.007	0.086	0.096	0.086	0.093
	(0.25)	(1.36)	(1.53)	(1.30)	(1.44)
Small firm*Öztürk		−0.118	−0.126	−0.114	−0.120
		(–1.42)	(–1.53)	(–1.32)	(–1.40)
Small firm*Öztürk with headscarf		−0.111*	−0.119*	−0.112	−0.118*
		(–1.67)	(–1.84)	(–1.60)	(–1.72)
International firm	−0.000	0.001	0.001	0.010	0.010
	(–0.01)	(0.03)	(0.05)	(0.25)	(0.26)
International firm*Öztürk				−0.020	−0.021
				(–0.40)	(–0.42)
International firm*Öztürk with headscarf				−0.006	−0.006
				(–0.14)	(–0.15)
German	−0.008	−0.005	−0.003	0.013	0.037
	(–0.42)	(–0.24)	(–0.14)	(0.29)	(0.84)
German*Öztürk				−0.066	−0.090*
				(–1.25)	(–1.66)
German*Öztürk with headscarf				0.020	−0.020
				(0.41)	(–0.40)
Interculturalism	0.134*	−0.171***	−0.170**	−0.181***	−0.194**
	(1.91)	(–2.70)	(–2.23)	(–2.74)	(–2.50)
Interculturalism*Öztürk		0.274**	0.277**	0.306**	0.321**
		(2.28)	(2.17)	(2.51)	(2.49)
Interculturalism*Öztürk with headscarf		0.373***	0.377***	0.367***	0.390***
		(3.08)	(2.95)	(3.01)	(3.03)
Appearance	0.017	0.018	0.014	0.003	0.029
	(0.40)	(0.40)	(0.32)	(0.03)	(0.30)
Appearance*Öztürk				0.007	−0.019
				(0.05)	(–0.14)
Appearance*Öztürk with headscarf				0.036	−0.019
				0.036	(–0.18)
Time dummies	Yes	Yes	Yes	Yes	Yes
City dummies	Yes	Yes	Yes	Yes	Yes
Sector dummies	Yes	Yes	Yes	Yes	Yes
Constant	0.136***	0.146***	0.115***	0.153***	0.113**
	(3.70)	(3.78)	(2.91)	(3.13)	(2.31)
Number of observations	1,467	1,467	1,467	1,467	1,467
R 2	0.067	0.076	0.083	0.079	0.087
Adjusted R2	0.051	0.056	0.062	0.054	0.059

Notes: Robust t-statistics in parentheses. Dependent variable is a dummy for receiving a positive callback. The reference category for identity: Bauer; for occupation: secretary.

^***p < 0.01; **p < 0.05; *p < 0.1.

The attachment of a positive reference letter increased the immigrants’ relative invitation rates. This outcome suggests that some prejudice is indeed at work when no references are provided. Note that this effect occurred in this experiment not because the immigrant improved her absolute chances by attaching a reference letter, but because the non-immigrant applicant reduced hers. This result is probably because the letter described the applicant as good but not excellent, which may be less than what is expected from a non-immigrant (but fits the expectations for an immigrant). Without reference letters, the non-immigrant therefore benefits from positive stereotypes (positive statistical discrimination), while attaching them makes no difference to the immigrant. It should be noted, however, that Neumark (2018) warned of jumping to premature conclusions about statistical discrimination. First, Neumark calls into question whether the manipulated “productivity dimension” for which different levels of information are provided in an experimental study is actually relevant to employers and thus potentially affected by group averages. I contend that, in this study, this is the case, since the information provided in reference letters turned out to affect the callback rates of non-immigrants significantly. Second, Neumark argued that to learn about real-world statistical discrimination, a study would have to provide the amount of information that employers usually have. German personnel managers state that applications with and without reference letters do occur, however, the exact proportion is unavailable. Thus, the conclusion that can be drawn from this study is that stereotyping of immigrant women happens; however, the exact magnitude of the problem in real-life applications remains unclear.

Addressing the Heckman-Siegelman Critique

So far, I have provided evidence for the existence of discrimination against women with a Turkish name, which significantly increased if the women wore headscarves. However, one issue remains that could cast doubt on the empirical findings.

Heckman and Siegelman (1993) raised concerns regarding experimental methods measuring discrimination, one of which is relevant also for correspondence tests: If group differences occur in the variance of unobservable productive characteristics, evidence of discrimination may be spurious. The argument is the following: Even if there is no taste-based discrimination and different groups of job applicants have the same unobservable characteristics on average (thus avoiding statistical discrimination on that front), there may be differences in the variances of unobservables that cause statistical discrimination. What is particularly problematic is that the direction of this effect depends on the correspondence study design, specifically on the quality level the experimenter uses to standardize the observed characteristics of job applicants in the study. If an experimenter chooses a low standard, the group of applicants with a higher variance of unobservable characteristics is more likely to pass the quality threshold of the firms. The opposite is true if the experiment works with high-quality applicants. The experimenter cannot know against what quality of applicants the experimental candidates are competing in order to address this bias. Neumark (2012) developed a method that allows isolating the effect that comes from different variances in unobservables. Applying this method allows an unbiased estimate of discrimination to be obtained.

Table 5 presents the results when Öztürk and Öztürk with a headscarf are compared to Bauer (reference group) in one model (controls include occupational dummies, reference letter, and special qualification requirements). The first two rows (panel A) report the estimated discrimination coefficients for Öztürk (bareheaded) and Öztürk with a headscarf from the probit model. The marginal effects from the probit estimation are very similar to those previously reported from the linear probability model. The next rows (panel B) present the estimated marginal effects from the heteroskedastic probit model. These estimates allow for correcting biases from differences in the variance of unobservables. In panel C, the effects from the heteroskedastic probit model are decomposed into two parts: The “level effects” are considered to be the unbiased estimates, and the “variance effects” are considered to result from the specific experimental design. As shown, the (unbiased) level effects are negative—the effect for Öztürk (without a headscarf) becomes insignificant and the effect for Öztürk with a headscarf almost doubles compared to the results from the probit analysis.⁹ Thus, discrimination, at least against the candidate with an immigration background and Muslim religion, is confirmed irrespective of the specific level of qualification chosen in the experimental design.

Table 5.

Decomposition Results Addressing the Heckman-Siegelman Critique

A. Basic probit model (marginal effects)
Öztürk	−0.042**
	(–2.40)
Öztürk with headscarf	−0.152***
	(–7.59)
B. Heteroskedastic probit model (marginal effects)
Öztürk	0.037
	(0.77)
Öztürk with headscarf	−0.078
	(–1.08)
C. Decomposition
Marginal effect through level (unbiased)
Öztürk	−0.065
	(–0.73)
Öztürk with headscarf	−0.290***
	(–15.08)
Marginal effect through variance
Öztürk	0.101
	(0.99)
Öztürk with headscarf	0.212***
	(3.16)
Standard deviation of unobservables: Öztürk/Bauer	1.787
Wald test statistics: null hypothesis that ratio of standard deviations = 1 (p value)	0.364
Standard deviation of unobservables: Öztürk with headscarf/Bauer	3.361
Wald test statistics: null hypothesis that ratio of standard deviations = 1 (p value)	0.464
Wald test statistic: null hypothesis that ratio of coefficients are equal	0.999
Number of observations	1,474

Note: Robust t-statistics in parentheses.

The Neumark method suggests that in this particular experiment, the applications were of relatively low quality compared to the real-life competitors. I can conclude this from the higher variance of the unobservables for the minority groups (Öztürk w/o or with headscarf)¹⁰ in combination with a larger estimate of discrimination (level effect) in panel C compared to panel B. In the given setting, the Neumark method corrects for firms favoring the higher-variance groups.

Discussion and Conclusion

Heated debate occurs in the West about the apparently inferior position of women in Muslim (immigrant) culture. And yet, little discussion takes place about how Muslim women are actually treated by the Western majority population. To test discrimination against Muslim immigrant women by Western employers, I investigate employment opportunities of female office workers in Germany by sending identical applications that differ only in the applicants’ names and photographs to firms. The results show that applicants with a Turkish name were treated unfavorably at a significant level, in particular when their photograph showed them wearing a Muslim headscarf, which indicates multiple discrimination. Everything else being equal, a female with a Turkish name who wears a headscarf has to send 4.5 times as many applications as does an applicant with a German name and no headscarf to receive the same number of callbacks for interview. The evidence for discrimination against the immigrant woman who wears a headscarf is also robust to addressing the Heckman-Siegelman critique by applying the method developed by Neumark (2012) that deals with potential differences in variances of unobservable characteristics. The levels of unequal treatment found in the present study (especially for chief accountants and accountants) are among the highest that have been measured in the experimental discrimination literature. This substantial disadvantage incurred by wearing the headscarf is all the more remarkable given the modern and progressive headscarf binding used in the current experimental setting. That the headscarf shown in the application photograph did not cover the applicant’s throat could be perceived that she is not particularly strict with respect to her religion. This presentation should have minimized the level of discrimination measured. Discrimination is likely to be even higher with a more traditional binding of the headscarf.

In the public debate, it is often assumed that the unfavorable labor market position of Muslim immigrant women is the result of a lack of education, job experience, or dedication to the labor market. But in this experiment, the women with Turkish names had the exact same qualifications as the women without migration history. Since in German applications an unusual amount of information is revealed, it was possible to control for a large set of variables that may affect hiring decisions (e.g., schooling outcomes and personal characteristics, such as marital status, age, and citizenship). Of course, the fictitious applicants did not only attend the same schools, they also received exactly the same grades. As all applicants grew up in Germany, where they received all their schooling and training, it was evident to employers that they were fluent in German. The many years of job experience signaled that they were eager to work. Given that in Germany it is common to indicate one’s family status, all applicants in this experiment identified as single, signaling to potential employers that they had no family obligations. Given these characteristics and credentials, the scale of discrimination is striking. Attention discrimination (Bartos, Bauer, Chytilova, and Matejka 2016) may be one explanation for the low regard for characteristics and achievements of the immigrant applicants. It is also noteworthy that in the experimental résumé, applicants were not given any spells of unemployment. Given the high level of discrimination against the woman with a headscarf, for her to hold such a résumé is in fact unlikely unless she is substantially more able than the autochthonous German. This finding suggests that discrimination may still be underestimated in this experiment.

When applications included reference letters that described the candidate in positive but not excellent terms, this reduced the level of discrimination. Of interest in this case, it was not that immigrants improved their absolute chances by attaching a reference letter, but rather that non-immigrants no longer benefited from positive stereotypes (positive statistical discrimination). As a result, immigrants did relatively better.

Critics may argue that in this study the woman wearing the headscarf was not discriminated against because of the headscarf per se but because headscarves are considered inappropriate for application photographs. As Unkelbach, Schneider, Gode, and Senft (2010: 382) stated, however, it makes little difference whether discrimination is due to wearing headscarves in application photos not being considered business savvy or due to a distaste for headscarves.¹¹ In any case, attaching a photograph with a headscarf is not an uncommon practice. Syed and Pio (2010) reported that, even in countries in which photos are not commonly included in job applications, some headscarf-wearing women deliberately attach their photographs to screen for non-discriminatory companies. Thus, they avoid wasting time meeting employers who, on principle, do not hire women who wear headscarves.

To examine why and when discrimination occurs, I studied the effects of various firm and job characteristics. For example, it may be that small firms discriminate more or that international firms discriminate less. However, firm and job characteristics helped little to explain unequal treatment. The dislike for immigrant applicants identified by this study may be explained by Social Identity Theory and by the theory on social stigma. To German employers, immigrants represent a stigmatized outgroup, which may lead to their rejection. According to the theory, women wearing the headscarf should be particularly negatively affected, as they carry multiple stigmas: their migration background and a symbol of Muslim religion. This pattern is exactly what I found in the present study. Callback rates for women wearing the headscarf were meager, even in comparison to bareheaded immigrant women. Thus immigrant women wearing headscarves are a social group that experiences particular discrimination, in fact multiple discrimination, in the German labor market.

In practical terms, this study illustrates that, due to discrimination, valuable human capital remains underutilized. Discrimination may discourage immigrant women from gaining higher levels of education and thereby reinforce negative stereotypes. Wooten and James (2004) analyzed various strategies for a company to successfully learn how to reduce discrimination. They argued that the implementation of a division for diversity management can be a successful force for increasing managerial diversity. This claim is in line with Kalev and Dobbin (2006), who also found that such an entity acts as an intensifier for other diversity programs, such as mentoring and networking programs, or for performance evaluation. The implementation of such an entity is obviously more realistic for larger firms. For firms that cannot afford a diversity unit, an encouraging finding of this study is that firms that seek to form “intercultural” teams manage to overcome conventional biases.

In this study I looked at the labor market situation of immigrant women from Turkey. Immigrants from Turkey used to represent the largest fraction of Muslim immigrants to Germany in the late 20th century and, thus, represent a large and long-established immigrant community today. Nonetheless, as this study showed, their wearing of the headscarf constitutes a substantial source of discrimination. Recently, Germany faced a large influx of Muslim immigrants predominantly from Syria, including many women who are and will be entering the labor market. Many of the Syrian women are wearing headscarves. Addressing the discrimination highlighted in this article has therefore become an even more pressing issue.