Methodology of Correspondence Testing for Employment Discrimination Involving Ethnic Minority Applications: Dutch and English Case Studies of Muslim Applicants for Employment

: We comment on methodological issues in the use of correspondence testing for discrimination in access to employment—that of submitting identical CVs to employers, but differing by the name (implying their ethnicity) of the candidate. After contrasting changing social structures in Britain and The Netherlands regarding ethnicity and Muslim integration, we report two case studies using correspondence testing for discrimination in employment involving a Muslim woman (in Manchester, England) and a Muslim man (in Rotterdam, Netherlands), outlining the recent socio-political situation concerning ethnic relations in The Netherlands. The methods used indicated apparent discrimination in employment involving both applicants. However, the novel methods we have employed require further veriﬁcation using both traditional and novel methodologies. Findings from the two case studies are discussed and compared, with further research proposed.


Introduction
According to a recent review of research (Bagley and Abubaker 2017), discrimination in access to employment on grounds of "race", ethnicity or religion occurs to a significant degree in England and The Netherlands; that, is despite legal statutes forbidding employment discrimination on such grounds, these laws are rarely enforced, and a number of systematic studies have shown, using "correspondence testing" (submission of personal details of a CV to a potential employer), that black and ethnic minority applications (particularly Muslim women) are much more likely than applications from white European candidates to have their applications rejected. The discrimination considered in this earlier review was at the level of recruitment, although other kinds of discrimination may pertain even after hiring, in terms of levels of appointment, pay, and promotion (Deitch et al. 2003).
Discrimination at the hiring level-the denial of employment to a person from outside the company responding to a public advertisement-has been measured in Europe and America in several ways (Bovenkerk 1992;Riach and Rich 2002;Pager and Shepherd 2008;Pager and Western 2012). Perhaps the most effective method is "situation testing", in which several individuals (usually actors) belonging to different ethnic groups, apply for advertised positions. Discrimination is assumed to occur when, say, ethnic minority candidates are significantly less likely, when interviewed, to be offered employment. However, such studies are expensive and logistically difficult (Bagley 1973).
A more frequently used method is that of "correspondence testing", in which CVs (curriculum vitae, or details of a person's life in terms of education and employment history) are submitted for advertised posts. In this methodology, CVs are usually paired, being identical in type of education and experience, differing only by the name of the applicant, which clearly identifies the individual as belonging to a particular ethnic or religious group. "Net discrimination" is conventionally measured (Wood et al. 2009) by the difference between the proportion of minority candidate CVs that prompts action (e.g., being called for interview) and the proportion of the "white" or "non-Muslim" candidates whose application prompts action.
A further measure of racial, ethnic and religious discrimination both in hiring, and in internal promotion practice can be obtained by analysing large data sets of adults (such as the UK national household surveys) to see if, when relevant factors (e.g., educational level, skills acquired, professional qualifications obtained) are controlled for, minority groups are significantly disadvantaged in terms of wage and salary levels, unemployment history, and status of employment (Wrench and Modood 2001;Carmichael and Woods 2000;Heath and Cheung 2006). These various forms of discrimination amount to what Modood and Khattab (2016) call "an ethnic penalty" in employment.
The rise of Islamophobia in Britain and The Netherlands (Fekete 2008;Fekete and Sivanandan 2009;Van der Valk 2015) has paralleled hostility to Muslims as employees, especially women who dress modestly, and wear the hijab, or head covering (this evidence is reviewed in detail by Bagley and Abubaker (2017)). Muslim women, it appears, bear the heaviest "ethnic penalty" in seeking employment in the British labour market (Modood and Khattab 2016).

Methodological Issues in Correspondence Testing
Constructing CVs for correspondence testing can be challenging, since the researchers must ensure that "fake CVs" appear genuine to a potential employer, and the CVs submitted must be more or less identical, apart from the crucial variable of the applicant's ethnicity, which is implied by their name. The following two quotations from the major study by Wood andcolleagues in 2009 (Wood et al. 2009) of employment discrimination in several English cities, making inferences from employers' response to submitted CVs, illustrate some of these methodological challenges: In principle, for a given number of vacancies, the more applications sent per vacancy the greater the statistical power of the study, but there are some practical limits. Certainly, if too many forms were to be sent after each vacancy then some employers might suspect they were part of an experiment, with the risk that all the forms are rejected. There is also the question of whether sending more than three or four applications is too great a burden on smaller organisations in particular." (Wood et al. 2009, p. 16) Qualifications had to be closely matched between the templates, but also sufficiently different to avoid suspicion. For instance, efforts were made to understand the structure of professional accounting qualifications and to ensure that different awarding bodies were similarly regarded. Constructing work histories was the most complex element. Real organisations that could be verified by employers were listed, although these were generally based outside the local area in case the recruiter knew about them. Years of experience were kept similar between templates, as were the types of organisation and roles carried out". (Wood et al. 2009, p. 22) The description of methodologies employed Wood et al.'s 2009study (Wood et al. 2009) of 2961 CVs submitted to 987 job advertisements (3 for each advertisement, for the different ethnicities described by the CV) did not discuss what seems to us to be a problem: would not an employer detect, in some cases, that three of the vitas submitted were suspiciously similar? This could result in all three vitas being rejected. Wood and colleagues (Wood et al. 2009) attributed the relatively low response of employers to the applications to the economic conditions prevailing following the economic recession of 2008, with strong competition for a declining number of jobs; but there could have been other reasons, such as the confounding effect of vitas being too similar.
In exploring this methodological issue, we submitted only a single vita (either an 'English Christian-name' candidate or 'Muslim-name candidate') to advertisements, until we had sufficient numbers of applications to ensure that the two candidates were applying to organisations whose 'modal' characteristics were similar (Bagley and Abubaker 2017).
Similar methodological problems may have affected the Dutch work of Andriessen and colleagues, and by Nievers and Andriessen between 2008and 2015(Andriessen et al. 2010, 2012Nievers and Andriesson 2010), discussed later.

Comparing Acceptance of Minorities in Britain and The Netherlands
The research focus of this paper began in 1970, when one of us compared Dutch and British social structures and the degree to which they accepted immigrants from their former colonies, and the amount of racism in terms of both attitudes and behaviour, including discriminating against ethnic minority candidates for housing and employment (Bagley 1973).
Analysis of the Dutch 'plural society' system identified separate cultural and religious blocs engaged in legal and social interactions that ensured mutual respect and tolerance between the blocs. Immigrants from the former colonies were usually absorbed into these blocs, with minimal tensions. The exception was the case of Muslim immigrants from North Africa and Turkey, who, not fitting into any of the existing blocs, were often subject to hostility and discrimination.
The crumbling of the blocs of the Dutch plural society (Kremer 2013) coincided with an increase in immigration of Muslim minorities, also coinciding with an increase in anti-Islamic prejudice (Bracke 2013;Fetzer and Soper 2003). This has continued, and has been paralleled by the emergence of a far-right party, the PVV, which prior to the Dutch general election of March 2017 was supported by more than 15 percent of potential voters (Corde 2015). This party advocated the banning of all future immigration of Muslims, the closing of mosques, the imprisonment without trial of "radical" Muslims, and forbidding sales of the Qu'ran, seen as a seditious text (Wilders 2012). However, the supporters of this party tended to live outside of Randstad Holland, the commercial, manufacturing and financial core of The Netherlands, which includes the major Dutch cities, The Hague, Rotterdam, Utrecht and Amsterdam. Voting for both centre and radical parties in the Randstad area was strong, however, pushing the PVV vote into third place in the March 2017 election.
Despite the surge in anti-minority attitudes (Van der Valk 2015), research on those living in Randstad Holland indicated that they had, on average, more favourable attitudes to Muslim minorities than those living in other parts of The Netherlands (Zick et al. 2008;Velasco González et al. 2008). The apparent reason was that, since most Muslims lived in the Randstad area, familiarity with and interaction with Muslims had influenced favourable attitudes-a well-known thesis developed in the work of Thomas Pettigrew (Pettigrew and Tropp 2008).
Some Dutch sociologists, such as Essed and Hoving (2014), have argued that, despite an official claim of "tolerance", there is an undertow of racism in Dutch culture. This we had observed in fieldwork in The Hague; in an overtly tolerant society, we described "race riots" in which Moroccan migrant workers were attacked and their homes burned, by mobs of the white, working class (Bagley 1973). Our argument then was that this kind of "discipline" imposed on a strange minority, outside of the blocs of the plural society, was part of the strict social control by which bloc pillarization was maintained. Now, the blocs have crumbled, and Islamic minorities have to face continued forms of Dutch racism in seeking a stable identity within a rapidly changing society (Essed and Hoving 2014).
There is, despite a façade of liberal legislation, a persistent thread of racism, xenophobia and Islamophobia in both British and Dutch societies. Earlier studies of attitudes showed that about a quarter of Britons were firmly racist in their attitudes to various groups (Bagley and Verma 1979). More recently, a major survey of public attitudes (Lowles and Painter 2011) has shown that around a quarter of Britons still hold very negative attitudes towards ethnic and religious minorities, including Muslims.

Comparing Racial Discrimination in English and Dutch Cultures
The only attempt at systematic comparison of racial discrimination between England and The Netherlands has been Bagley's attempt (Bagley 1973), in fieldwork in 1970, to replicate an English study (Daniel 1968) using young, professional men both as "actors" in applying for vacancies, and accepting a vacancy when it was offered. The three Dutch professionals, with qualifications in industrial management, were "indigenous white", "black colonial immigrant" (Surinamer), and "foreign" (Yugoslavian). Their experience in seeking a response to advertised vacancies, and submitting to interviews when offered, was contrasted with three English actors (Black Jamaican, Hungarian immigrant, and indigenous white). In the 60 "situation tests" of advertised employment, the Dutch West Indian was discriminated against in 30 percent of his applications, while the English West Indian was rejected in 75 percent of the cases in which the white tester was offered an interview. Similar levels of discrimination obtained for applications for accommodation. In The Netherlands the "foreigner" was rejected as an applicant for employment more frequently than the black applicant, quite the opposite of the English situation.
Changes in Dutch society, including the collapse of the "bloc" system, which absorbed certain kinds of immigrants, such as black Surinamers, have been outlined above. By the 1990s, evidence of discrimination against minorities was beginning to emerge, and this has clearly been mapped in the work of Andriessen and colleagues (Andriessen et al. 2010(Andriessen et al. , 2012(Andriessen et al. , 2015Nievers and Andriesson 2010).
Changes in Dutch social structure have resulted, paradoxically, in an increase in hostility towards ethnic minorities such as settlers from the former Dutch colony of Suriname. Nevertheless, today in The Netherlands, as in other parts of Europe, the most stigmatised minorities are Muslims, who, as migrant workers and their descendants, have origins in Turkey and North Africa (Fekete and Sivanandan 2009;Van der Valk 2015). In Britain, Muslim minorities are mostly the descendants of settlers from the Indian Sub-Continent, with substantial numbers also from former British colonies in East and West Africa, the Middle East, and Malaysia.
According to the 2011 National Census, some five percent of the British population were Muslims, most being in the younger age groups, either still undergoing education or in the early stages of working careers. In The Netherlands, Muslim minorities (some four percent of the population) are, like their co-religionists in Britain, victims of Islamophobia. Dutch Muslims have always stood outside of the verzuiling system (Kremer 2013). Indeed, Frank Bovenkerk and colleagues (Bovenkerk et al. 1995) in comparing methodologies for studying racial discrimination in Dutch, French, Swiss and Austrian cultures, found that Dutch Moroccans were subject to considerable degrees of rejection (in 56% of employment applications), compared with other ethnic minority groups.
Andriessen and colleagues (Andriessen et al. 2010(Andriessen et al. , 2012(Andriessen et al. , 2015 Nievers and Andriesson 2010) have examined more recent levels of discrimination surrounding perceived Muslim minorities in Randstad Holland, by submitting resumes for advertised job vacancies, indicating that the applicant was either of "purely" Dutch origin, or had Islamic antecedents. This recent, extensive Dutch research has shown that racial discrimination in The Netherlands is actually increasing in the age of Islamophobia.
In the work of Andriessen and her colleagues, various methodologies were employed over several years, including the use of actors applying for advertised jobs, and the submission of CVs identical except for the name (and implied ethnicity) of the candidate. Either the "twin" CVs were virtually similar, or the minority candidate had enhanced qualifications and experience. Several thousand "situation tests" were made, using varying methodologies. These tests showed that the well-qualified Dutch, "white", candidate was successful in gaining access to an interview in up to a half of their applications, while the Moroccan or Turkish "Muslim" candidate was successful in a little under a fifth of applications; these results give a net figure for discrimination of around 20 percent, which is approximately the same as that obtained in the English work by Wood and colleagues (Wood et al. 2009).
These results showed, then, that there remained significant potential job discrimination against ethnic and religious minorities in The Netherlands, which seemed greater than that in earlier work in Amsterdam and The Hague (Bagley 1973), although differences in methodology make accurate comparisons difficult. We have hypothesised, as have others, that the levels of employment discrimination for minorities may have been worsening with the collapse of the Dutch plural society system (Bracke 2013).

Testing for Discrimination Using Correspondence Testing: Exploring a New Methodology in Manchester and Rotterdam
Since we were puzzled by a problem in the methodology of correspondence testing for ethnic discrimination in employment in which virtually identical CVs are submitted to the same employer (thus inviting the rejection of both "suspiciously similar" vitas), we have explored a new methodology: that of submitting vitas similar in every detail except the ethnicity implied by a religious (Muslim) name to two different sets of employers. We submitted the vitas individually until the contrasted groups of companies were similar, on average, on a variety of factors such as the company's size, trading purpose, geographical location, and role requirements for the advertised vacancy (Bagley and Abubaker 2017).
This correspondence testing in Manchester, England found that "Aminah", a woman in her 20s, qualified at an intermediate level of accountancy, received a positive response to her online submission of her real CV in 151 of her 516 submissions. In parallel the English "Emily" (whose CV was similar to Aminah's in every respect, except her name) was successful in 306 of her 527 applications, for job descriptions and company characteristics which were, overall, similar for both applicants; this yielded an index of net discrimination against the Muslim candidate of 29 percent.
"Muhammad", the real-life correspondence tester in Rotterdam, was in his early 30s, and had worked for six years as a management accountant with the same wholesale supply chain company, and was now earning 55,000 Euros a year. He had BSc and MSc degrees in economics, accounting, auditing and cost control, and was a fully qualified RA (Registered Accountant) in the Dutch system. He was of Turkish ethnic heritage, but had been born in The Netherlands. The company that employed him was small in size, and he had reached the peak of his income level in that organisation. He submitted an online vita for positions paying 70,000 Euros a year or more in 2016-2017. A dummy tester ("Hank Van der Waal") submitted an identical CV for advertised positions, but to different companies, very similar to those to which Muhammad applied. The companies approached (207 for Muhammad, and 214 for Hank) were, overall, similar in profile (location, function, employee numbers, financial turnover). Muhammad was called for further information or interview in 39 per cent of applications; Hank was contacted in 53 percent of submissions, indicating a net discrimination index against the Muslim candidate of 14 percent.

Discussion and Conclusions
These findings, using regionally limited case studies, with relatively small samples of employers, tentatively suggest that a methodologically revised method of correspondence testing for ethnic discrimination in employment practices showed the persistence of discrimination, possibly avoiding the confounding effects of employers rejecting both the mainstream and minority candidate CVs because they were suspiciously alike. However, we must concede a number of criticisms that have been made concerning this approach: that sending CVs singly to different employers may lose an important amount of information on employers who are accepting or discriminating; that, in reality, large employers will always receive many CVs which are almost identical, given the nature of the position advertised, so a revised design is hardly necessary; and in fact the most rigorous tests of discrimination would involve submitting different "ethnic" CVs to the same employer-thus also avoiding the problem that submitting a single CV to each employer could not control for the qualifications and experience levels of competing candidates.
A further critique of our "opportunistic" methodology (using real-life applicants) is that we limit the sociological scope of explanations for discrimination. For example, responses to an applicant with a Turkish-surname may be more favourable than reactions to an applicant with a Moroccan-surname in The Netherlands, in a manner similar to that observed in other cultural settings (Booth et al. 2012;Blommaert et al. 2013;Carlsson 2010). These criticisms, made by reviewers of our paper, we accept as valuable pointers for the next phase of our research.
In exploring the sociological richness of data on discrimination, we must also take into account the "intersection" of statuses (e.g., gender, ethnicity, religion, educational background) in explaining the phenomenon researched more fully (Veenstra 2013). There may be, for example, less discrimination against well-qualified ethnic minority applicants, who have skills that are in short supply (Baert et al. 2015); this may have been the case with regard to the well-qualified, Muslim, Dutch accountant in our study applying for a senior management position. In contrast, a young man from an ethnic minority without qualifications who applies for a basic-wage job may experience considerable discrimination; this was shown in recent Dutch work in which a young, unskilled "Dutch" male with a declared history of child sex offending was significantly more likely to be offered unskilled employment than a young "Moroccan" male with no history of offending (Van den Berg et al. 2017).
The English and Dutch results presented in Table 1 are not directly comparable, since, although both minority group applicants were overtly Muslim, in one case the applicant was female applying for a junior position; in the Dutch case, the applicant was male, older, and applying for a relatively senior position. Differences in acceptance could have been related to gender and qualifications of the applicants. Our method of testing using real applicants who actually were looking for employment, each accepting one of the posts offered, also seems to be unique in the literature on measuring ethnic employment discrimination (Bagley and Abubaker 2017); but this method also limits the sociological scope of the enquiry, and of the search for explanatory models. There was a significantly higher "net discrimination" against the Muslim woman in Manchester than for the Muslim man in Rotterdam, who accepted a senior accounting management post at 70,000 Euros per annum in his home city of Rotterdam. We plan to extend this research in both countries to see if comparable female and male minority applicants from various ethnic groups are treated differently in the two countries. In the further research, we will extend the range of applicants (using dummy CVs), and will compare similarly qualified candidates in both England and The Netherlands.
The tentative Dutch results from our current study do, however, appear to support the findings of Andriessen and her research group, although the discrimination against the Muslim Dutchman was somewhat less than we had expected; nevertheless, it was of sufficient magnitude to suggest that racial discrimination is a persistent problem in The Netherlands, and may well have increased following the crumbling of the pillars of the Dutch plural society. The rise of a political demagogue making overtly anti-Muslim statements does not appear, however, to be reflected in a profound degree of hostility to Muslims in Rotterdam (the major city of Muslim settlement in which our correspondence tester lived) or in Randstad Holland, where, hypothetically, "familiarity breeds affection" (Zick et al. 2008;Velasco González et al. 2008;Pettigrew and Tropp 2008).
These are significant sociological pointers to societal change, and the need for further research in this important field (Blommaert et al. 2013). A critical methodological issue, which we will address in future research, is that the "single CV" method must be compared with the "multiple CV method" (i.e., several contrasted CVs submitted to the same employer) if we are to draw firm conclusions about the reliability and validity of the new method of enquiry that we offer in the present research. Until this is done, the results of the present study remain tentative. We plan to extend our work in England and The Netherlands using the traditional methods of correspondence testing used by previous researchers (Blommaert et al. 2013;Van den Berg et al. 2017), comparing this with a replication of the "sole CV" method described in this paper. We will, in future work, use only hypothetical CVs (i.e., not the CVs of real people), so that characteristics of age, gender, ethnicity and religion, and education and qualifications can be varied in a multi-factorial design.
We urge the importance of such research, both in England and The Netherlands, by declaring the hypothesis that if employment discrimination against young Muslims is widespread (and, implicitly, is known to occur by those who are denied jobs for which they are in fact qualified), degrees of alienation will occur that may lay the seeds of profound personal disaffection, with the potential genesis of violent disorder (Bagley and Al-Refai 2017) 1 . The overt and profound hostility to hijab-wearing women in England in street-level interactions has been described by ethnographers (Zempi and Awan 2017); public hostility, combined with "private hostility" on the part of employers, may have profoundly negative consequences for "quiet civility", the manner that Muslim immigrants and their children traditionally bring to British society (Al-Refai and Bagley 2012).
Author Contributions: M.A. and C.A.B. conceived and designed the experiments; both researchers carried out the UK-based research; C.A.B. carried out The Netherlands-based research. Both M.A. and C.A.B. jointly analyzed the data, and wrote the paper.