Identifying barriers to Muslim integration in France

Abstract

Is there a Muslim disadvantage in economic integration for second-generation immigrants to Europe? Previous research has failed to isolate the effect that religion may have on an immigrant family's labor market opportunities because other factors, such as country of origin or race, confound the result. This paper uses a correspondence test in the French labor market to identify and measure this religious effect. The results confirm that in the French labor market, anti-Muslim discrimination exists: a Muslim candidate is 2.5 times less likely to receive a job interview callback than is his or her Christian counterpart. A high-n survey reveals, consistent with expectations from the correspondence test, that second-generation Muslim households in France have lower income compared with matched Christian households. The paper thereby contributes to both substantive debates on the Muslim experience in Europe and methodological debates on how to measure discrimination. Following the National Academy of Sciences’ 2001 recommendations on combining a variety of methodologies and applying them to real-world situations, this research identifies, measures, and infers consequences of discrimination based on religious affiliation, controlling for potentially confounding factors, such as race and country of origin.

Social and political relations between Europe and the Muslim world are fractious (1, 2). Attacks in Madrid (March 2004) and London (July 2005) were perpetrated by Muslim radicals. Political parties in Europe have mobilized opinion against a Muslim threat.^a Relations between the countries and societies of the European Union and the Muslim world have become politically consequential on a number of dimensions: foreign policy in regard to the Middle East, membership into the European Union, and the vast migration of Muslim populations into European Union states.

Surprisingly, several recent studies have found that there are no special problems for Muslims in Europe. The Pew Global Attitudes Project poll of 2006 established that “while there are some signs of tension between Europe's majority populations and its Muslim minorities, Muslims there do not generally believe that most Europeans are hostile toward people of their faith” (3).^b Furthermore, “[s]ubstantial majorities of Muslims living in the European countries surveyed say that [in the 2 years after bombings in Spain and London, and the Cartoon Crisis in Denmark (4)],^c they have not had any personally bad experience attributable to their race, ethnicity, or religion” (3, 5).^d More recently, the 2009 Open Society Institute multicountry study of Muslim and non-Muslim attitudes toward immigration and social cohesion reveals that Muslim respondents are as likely as non-Muslim respondents to report that people in their neighborhood are willing to help each other; only 10% of Muslim respondents reported discrimination by the police, and a similar proportion of Muslims (29.2%) and non-Muslims (31.1%) reported trust in the government (6).^e

At the same time, European states are defined by their historic nationalities, all of them in the Christian tradition, and are seen as having a special problem with Islam going back to the fall of Constantinople to the Ottomans and the reconquest of Spain in the 15th century. Thus, throughout the continent, there is a myth of a “Christian Europe” that is maintained despite its virtually complete secularization in the past century. It manifests itself in recent political events as well as in individual attitudes and perceptions of discrimination. Suspicions run high in the debate over the application of Turkey into the European Union (as opposed to Bulgaria) (7).^f Local issues, such as permits to build minarets in European cities or prohibitions on women wearing the burqa, transmogrify into continental causes célèbres. The 2009 Open Society Institute study paints a deteriorating picture of religious and racial discrimination: 55.8% of Muslim respondents and 43% of non-Muslim respondents, representing a plurality, claim that there is more racial prejudice today than there was 5 y ago^g; 68.7% of Muslim respondents and 55.9% of non-Muslim respondents make that claim with regard to religious prejudice, and more than 90% of both Muslim and non-Muslim respondents agree that Muslims are the ones experiencing this religious prejudice (6).

Thus, conventional wisdom points both to a cosmopolitan Europe open to Muslim immigrants and a closed Europe suspicious of these immigrants. Despite the sound and fury, the question of whether there is a special Muslim problem for Europe in general remains unclear. What is absent is a data-driven answer that can adequately identify and measure Muslims’ failure/success in economic integration into Europe.

The National Academies’ National Research Council's Committee on National Statistics identified this problem on a broader scale when it convened a panel of scholars in 2001 to provide a comprehensive review of the major methods used to measure racial discrimination, including statistical analysis of observational data, laboratory experiments, and field experiments (8). Although the panel argued that no one method can solve all the troubling inferential problems for this notoriously complex issue, the difficulties of relying on observational data, including omitted variable and sample selection bias,^h received special attention, perhaps because at the time of writing, reliance on observational data was thought to be used most often. Laboratory studies received a number of important critiques as well, ranging from the limits of external validity (the inability to generalize findings beyond the laboratory context) to Hawthorne effects (whereby subjects tend to behave differently under a microscope).

The mirror problems of observational and laboratory studies led the panel to give considerable attention to field experiments, such as audit studies, which were praised as “an important and useful means of measuring discrimination in various domains” (8).ⁱ These studies have identified racial discrimination in the housing market, in car sale negotiations, and in preapplication behavior by lenders, among other economic transactions; yet they can draw only limited inferences about racial discrimination. Because it is not feasible to randomize a tester as to whether he or she presents himself or herself as white or black, it is not possible to claim confidently that if tester “A,” who was black, were white, he or she would have done better.

The panel therefore finds great value in combining features of laboratory and audit studies. It reports on Bertrand and Mullainathan's classic curriculum vitae (CV) experiment, today called a “correspondence test,” in which the researchers created two identical resumés stratified by low- and high-skill levels, and assigned a white-sounding name to a random half of the resumés and a black-sounding name to the other random half. The experiment generated striking results. Applicants with white-sounding names received 50% more callbacks than applicants with black-sounding names. Moreover, the return on quality of CV (i.e., reward for high-skill level) was high for whites but nonexistent for blacks. The randomization of race assignment across otherwise identical resumés made it possible to overcome the missing counterfactual and to draw inferences on the effect of race. Because of the relative methodological success of this mixture of experiment and audit, the report calls for greater research investment in the replication of laboratory experiments “in real-world settings with real-world data” (8).

This paper pushes the agenda set by the National Academy of Science panel report by addressing the issue of religious discrimination in Europe. It presents data on the relative success of a matched set of second-generation Muslim and non-Muslim immigrants in France to see if, conditioned on human capital, they face equal opportunities in the labor market, our criterion for economic integration. By doing so, it accomplishes two important goals. First, it contributes to a salient topical debate in a Europe struggling to cope with an era of high religious identification and massive Muslim immigration.^j Second, it addresses and resolves the special methodological challenges involved in measuring religious discrimination. Indeed, identifying religious discrimination has inferential problems quite different from race, inasmuch as the signal of one's religion can be manipulated in front of an employer or a real-estate agent in ways that are far more difficult, on average, for race. A bigger problem in identifying a religious impact on employment opportunity is that of the confound that exists between religion and country of origin. If people from Turkey are most likely to be of Muslim heritage, how can we tell if labor market discrimination in Germany is attributable to suspicion of Turks or religious prejudice? Similarly, how would we know if labor market discrimination in the United Kingdom is attributable to bias against South Asians (from India, Pakistan, and Bangladesh) or Muslims? In France, how might we differentiate between prejudice against North Africans (Maghrebis) and prejudice against Muslims?

This paper introduces a solution to such previously intractable measurement problems. The resulting measure, if valid, would tell us if for Muslim immigrants and their descendants in country X, economic integration faces higher barriers than if everything about these migrants were the same except for their religion. Using a correspondence test in the French labor market and a matching strategy that compares the relative success of two identical candidates who differ only in their religious affiliation, this paper identifies significant anti-Muslim discrimination, controlling for a factor (i.e., country of origin) that has confounded results in previous work. Its principal finding is that, all other things being equal, a Muslim candidate is 2.5 times less likely to obtain a job interview than is his or her Christian counterpart. A high-n survey of immigrants selected via the same matching strategy (e.g., immigrants from the same ethnic group but divided by religion) provides evidence for a substantial income effect consistent with the expectations derived from the correspondence test.

Previous Research on Employment Discrimination in France Among Workers from Immigrant Families

Correspondence testing allows researchers to measure labor market discrimination based on specific characteristics, such as gender, age, race, or religion. A wide range of correspondence testing has been conducted beyond the United States, starting with the experiment of Jowell and Prescott-Clarke (9), which tested whether applicants from Asian backgrounds were discriminated against in the UK labor market. To our knowledge, four of these studies focus on discrimination against immigrants in the French labor market (10–13).^k These studies compare the response rate received by a CV with a Maghrebi-sounding first and last name with the response rate received by a CV with a French-sounding first and last name, with all other characteristics being equal across these CVs. Applicants from a Maghrebi background were found to be strongly discriminated against in the French labor market compared with those from an “authentic” French background. For instance, Duguet et al. (11) compute that for every 100 positive responses for the authentic French candidate, the Moroccan candidate received only 35, with the difference statistically significant at the 99% confidence level.

Although these studies reveal substantial discrimination against applicants of Maghrebi background, they do not allow us to isolate the source of this discrimination. Two confounding factors are at stake: do employers discriminate against Maghrebis or against Muslims?

The difficulties in identifying a religious effect as demonstrated in the CV experiments performed so far are not easily resolved, and this is all the more so in surveys. Data on Muslims in France are hard to get. A 1978 law set prohibitions on the collection of data on the racial, religious, or ethnic identity of its citizens, creating challenges for demographic research. For instance, in a leading sociological study of the economic success of different immigrant groups, researchers could not distinguish the children of Algerian migrants into France from the children of the pieds noirs, those of European ancestry who left after Algerian independence in 1962 (14). Although the law was partially relaxed in 2007, this type of data collection has remained stringently limited (15–17).^l

Equally important for the problem of statistical analysis, those available mass surveys exempt from state oversight rarely (with the Pew Global Attitudes Project poll being the partial exception) include enough Muslims to allow for good data analysis. For example, the World Values Survey included in its latest wave for France only 47 reported Muslims out of a sample of 1,001. Given that there are an estimated 6.3% Muslims in France, this survey is hardly representative of the Muslim population (18).^m

The real killer for identifying a Muslim effect is that most Muslim immigrants to each of the major European states come from a single country or world region. In Germany, nearly all Muslims are from Anatolia; in the United Kingdom, the same is the case with South Asians; and in France, nearly all Muslims are from the Maghreb. In each of these cases, it is nearly impossible to determine conclusively whether any additional problems that these immigrant populations and their descendants have had in economic mobility beyond the problems faced by non-Muslim immigrant families are attributable to the fact that they are Muslim. For example, nearly all immigrants to France from Portugal are Catholic, and nearly all immigrants to France from Algeria are Muslim: once they control for homeland, statistical models cannot distinguish a religion from a country effect on outcomes. To look at the inferential problem another way, any special problems in economic advance faced by children of North African immigrants to France may be attributable to their Muslim religion, the fact that they were a colonized people from a geographic area where the major state fought an insurgency against French imperialism, or some aspects of North African life distinct from Islam (e.g., language, educational systems, history of authoritarian political rule).

To illustrate one aspect of the inferential problem, we draw from a large-n survey of 12,010 randomly selected households with an immigrant then (2002–2003) living in France, one of the few large-n surveys in France in which a self-reported question on religion was posed (19).ⁿ The survey also contains key information on the age, gender, education, region, country of origin, and income of immigrant respondents. These data were collected to study the retirement decisions of immigrants, but they allow us to illustrate the problem of trying to infer the effects of religion on income when there is a high correlation between homeland and religion.

Consider Fig. 1, which analyzes these data, where models 1, 2, and 3 illustrate the problem (the three statistical models are described in the legend for Fig. 1). Model 1 estimates the respondent's income as a function of his or her religion, controlling for gender, age, education, and length of stay in France. With dummies for all religions (and Christianity the omitted category, thereby serving as the point of comparison), we find that Muslim immigrants are significantly poorer than Christian immigrants. In model 2, we add home country fixed effects, a statistical technique that allows us to control for the immigrant's homeland. In model 3, we do the same for region of origin. The Muslim effect previously identified in model 1 loses statistical significance in models 2 and 3. Once we control for country or region of origin, we are unable to identify a Muslim effect, suggesting that the religious effect we measured in model 1 may actually be driven by some other cause related to a respondent's home country or region of origin rather than his or her religion. If the goal is to identify an independent Muslim effect on an immigrant's economic integration, this study (as with the previously reviewed CV studies) yields inconclusive results. A different strategy is needed to address our questions.

Fig. 1.

Impact of religion and geographic origin on a household's yearly income. This figure is based on results in Table S1. The dependent variable is an ordinal variable ranging from the value “0” if the yearly household income is null to “14” if the yearly household income is greater than 68,000 Euros. The independent variable is whether the respondent is “Muslim,” a binary variable, which takes the value “1” if the head of household is Muslim and “0” otherwise. In model 1, Muslim is significant at the P < 0.01 significance level on a two-tailed test. It is not significant by conventional standards in model 2 and in model 3. As for controls, “Female” is a binary variable, which takes the value “1” if the head of household is female and “0” if the head of household is male. “Age” is a continuous variable equal to the actual age of the head of household. “Education” is an ordinal variable ranging from the value “1” for no schooling to “6” for postsecondary education. “Jewish” is a binary variable, which takes the value “1” if the head of household is Jewish and “0” otherwise. “Asian” is a binary variable, which takes the value “1” if the head of household is Buddhist, Hindu, Shintoist, or Confucianist and “0” otherwise. “Atheist” is a binary variable, which takes the value “1” if the head of household is an atheist and “0” otherwise. The reference group is “Christian,” a binary variable, which takes the value “1” if the head of household is Christian and “0” otherwise. “Yrs France” is a continuous variable equal to the number of years that the head of household has lived in France. SEs in the original probit model are robust. In the figure, the dot represents the regression coefficient, the horizontal line marks the 95% confidence level, and the two vertical lines mark the 90% confidence level. F.E. refers to fixed effects.

Our Matching Strategy

Our approach is to use a matching strategy, targeting immigrant groups that are divided religiously, with one portion of them being Muslim and another portion, quite similar to the first culturally, economically, and in education, being Christian. [A literature review on matching strategies can be found in the article by Sekhon (20)]. Comparing the Muslims and Christians in each group allows one to measure the “Muslim effect” without other confounding factors, such as region or country of origin.

Proper matching requires that the two subsets of religiously divided immigrant groups arrive at the host country with relatively equal resources, or else the comparison would be biased. Suppose the target population was the Lebanese population in North America and the comparison were between Maronite Christians and Sunni Muslims. Because the Maronites start off earlier and with rich international networks of banking families, a finding that Maronites achieve higher rates of economic success would tell us little about comparative social and economic barriers in the West attributable to religion (21).^o Furthermore, insufficient overlap on initial income on arrival and on length of residence in the United States would lead us to estimate, in a statistical analysis, coefficients that are model-dependent, and would therefore limit our ability to make confident inferences about the religious effect on immigrant integration.

Careful examination of selected small immigrant groups in France, however, invites opportunities to get a reasonable approximation of an unbiased comparison. In our case, we identified an estimated 10,000 immigrants in France with family backgrounds as Joolas and Serers, two distinct ethnolinguistic communities from Senegal (22).^p These two groups, unlike all other communities in Senegal, have a sufficiently large Christian population to allow for intragroup comparisons (23, 24).^q Moreover, contrary to what we observe for Lebanese Maronite Christians and Sunni Muslims, Senegalese Christians did not benefit from the earlier settlement of a Senegalese Christian diaspora in France. To confirm this, we rely on a survey administered by David Laitin with the French firm Conseils-Sondages-Analyses (CSA), in which 511 Serers and Joolas, both Christians and Muslims, were polled on a variety of issues concerning their integration in France. Data from that survey indicate that the time elapsed since the settlement of the first migrant is 39 y for Senegalese Christian families and 39.3 y for Senegalese Muslim families. This difference is not statistically significant.^r Furthermore, there is no statistically significant difference in the distribution of years of arrival in France between the two groups.

A rather common impression among the French population (and especially among the French population of Maghrebi origin) is that African Muslims are not “real” Muslims because they know little to no Arabic and interact indiscriminately with African Muslims and African non-Muslims (25). Our choice of a target population less spontaneously associated with Islam in the French collective imagination, however, would lead to an underestimation of anti-Muslim discrimination. Our results should thus be interpreted as a lower bound on the magnitude of anti-Muslim discrimination: levels of discrimination among Maghrebis, the real Muslims who are at the center of public debates about the role of Islam in France, would almost certainly be even higher than those we find for Senegalese Muslims.

Keeping with the matching strategy outlined above to separate out region of origin and religion, our experimental design demanded three comparable CVs, two of them from women with an obvious Senegalese surname (Diouf) but one with a well-known Muslim first name (Khadija) and the other with a well-known Catholic first name (Marie). The third CV was from a woman with a typical French republican name, with no religious connotation (Aurélie Ménard). In addition to differences in the first names, we introduced two signals of religious identity related to the work and volunteer experiences of our fictitious candidates. One of Khadija's past positions was with Secours Islamique and one of Marie's was with Secours Catholique, whereas Aurélie worked solely in secular firms. Also, Khadija did voluntary work for the Scouts Musulmans de France, whereas Marie did the same for the comparable Catholic organization, Scouts et Guides de France.^s

The remaining qualifications and backgrounds were identical for all three applicants: all were single French citizens, 24 y of age, and had 2 y of postsecondary education and 3 y of experience on the job market in either secretarial or accounting sectors. We deliberately chose occupations that entailed interaction with clients or company partners, such that recruiters would be paying greater attention to the expected reactions of these potential clients to their employees.

The non-governmental organization, Inter Service Migrants - Centre d’Observation et de Recherche sur l’Urbain et ses Mutations (ISM-CORUM), collected job announcements nation-wide for both types of occupations published on the Pôle Emploi Web site (the French national employment agency) during the spring of 2009. For each pair of job announcements matched by region, sector, company size, and position, ISM-CORUM administrators sent the Aurélie/Khadija candidate pair to one and the Aurélie/Marie candidate pair to the other. Aurélie Ménard was used principally as the “reference” candidate: her CV allowed us to avoid sending Marie and Khadija's applications for the same position. Because these applications were identical in both form and content (except for the religious identity signals), sending both CVs would have inevitably awakened suspicion among recruiters.^t

Fig. 2 summarizes our results. We first observe that the reference candidate, Aurélie Ménard, received the same positive response rate from employers who received Marie's CV and from employers who received Khadija's CV (27% and 25% respectively, with no statistically significant difference between the two). This indicates that the companies receiving Marie's CV were, on average, similar to those receiving Khadija's CV, thus lending greater confidence to our comparison of Marie and Khadija's positive response rates.

Fig. 2.

(A) Substantive effect of Khadija Diouf vs. Marie Diouf. (B) Statistical significance of the Khadija Diouf effect. This figure is based on results in Table S3 from Dataset S1. In B, the dot represents the difference in response rates, the horizontal line marks the 95% confidence level, and the two vertical lines mark the 90% confidence level.

The difference between Marie and Khadija's positive response rates is striking. Although Marie Diouf received a positive response rate of 21%, Khadija Diouf received a positive response rate of only 8% (Fig. 2A). This 13-percentage point difference is statistically significant at the 99% confidence level and indicates that for every 100 positive responses received by Marie Diouf, Khadija Diouf received only 38 positive responses, or 2.5 times less.^u Furthermore, these results hold in a multivariate regression controlling for regional differences, employment sector, company size, occupation, contract type, and whether or not the CV included a photograph.^v Notably, the first 214 applications we sent had no pictures of the candidates. The next 61 included the same picture for the two Dioufs as a signal that they were not North Africans. Furthermore, there were no religious symbols in the photographs, indicating to recruiters that Khadija Diouf is not an orthodox Muslim. The results with and without the photographs were not statistically different from each other. This experiment thus provides a clear indication that in at least one sector of the French labor market and controlling for the candidate's ethnicity among other characteristics, there is significant religious discrimination.

Consequences of Religious Discrimination

Does the discrimination experienced by Muslim candidates in the French labor market correspond to an economic disadvantage on the part of Muslim immigrants relative to their Christian counterparts? To answer this question, we rely on the Laitin/CSA survey of 511 Senegalese Christians and Senegalese Muslims living in France in 2009. We previously established that these two groups immigrated into France in a single wave during the 1970s. To ensure a fair comparison, we must also ask whether the first immigrants of respondent families to France started out on equal footing. The only critical difference on the arrival in France of Senegalese Muslims and Senegalese Christians relates to education. Senegalese Christians were slightly more educated than Muslims: whereas the probability of having a secondary or a postsecondary education is 36% among Senegalese Christians, it is 27% among Senegalese Muslims (a difference that is significant at the 95% confidence level).^w These results are consistent with ethnographic accounts of Senegalese Christians’ access to better quality education through their religious network (i.e., Catholic schools). A proper matching strategy between Senegalese Christians and Senegalese Muslims thus requires that we control for the first migrant's level of education, which, given their high overlap in educational achievement, is statistically feasible.

We can now ask, controlling notably for the educational level of the first migrant, whether Senegalese Christians have been more successful in breaking through social and economic glass ceilings in France than have Senegalese Muslims. With this procedure, we are confident that any differences found between the two groups are the result of some aspect of their religious upbringing or practice, because geographic origin does not vary and initial human capital is controlled for.^x

We estimate the determinants of immigrant income in France today via an ordered probit regression with robust SEs, with the current monthly household income as the dependent variable. The explanatory variables are the religious tradition of the household, the head of household's gender and educational level, and the educational level of the head of household's ancestor who was the first to come to France. (Recall that there is no need for a second model with country fixed effects because all respondents are from Senegal. This is the key to our matching strategy.) As illustrated in Table S4, we find that households with a Christian religious tradition are significantly richer than households with a Muslim religious tradition (significant at the 99% level).^y More precisely, when we estimate marginal effects and hold all other explanatory variables in the model at their predicted sample mean, the results indicate that Muslim households make, on average, 400 Euros less than Christian households each month, the equivalent of 15% of the average monthly income or 17% of the median monthly income for France in 2007 (26).^z Fig. 3 further illustrates the effect of household religion for each income category: it indicates that Muslim households are more likely to fall into lower income categories, whereas Christian households are more likely to fall into higher income categories. In sum, even controlling for the educational level of the first migrant to France, there is a significant negative Muslim effect on present day household income. We can therefore infer that the job discrimination revealed in the section on our matching strategy has broad implications for differences in today's household income for Muslims.

Fig. 3.

Total income effect of Christian vs. Muslim household (Hh). This figure is based on results in Table S4 from Dataset S2 and its codebook (Dataset S3). Predicted probabilities of reaching each income category are calculated for each household. They are then averaged out for Christian households and for Muslim households for cases in which all other independent variables in the model are at their Christian/Muslim household mode. This captures the total effect on household income of being a Christian household vs. a Muslim household. Similar graphs were produced (i) holding all other variables at the sample mode, (ii) holding all other variables at their Christian/Muslim household mean ± 1 SD, and (iii) holding all other variables at the sample mean ± 1 SD. All present the same pattern as that illustrated here.

Summary and Extensions

In this paper, we have shown that both public debate and previous research on the implications of religious difference for integration into Europe, and France in particular, have yielded ambiguous results. Relying on a matching strategy and replicating well-understood CV experiments, we have been able to identify a statistically and substantively strong level of religious discrimination in at least one sector of the French labor market. Relying on that same matching strategy but now through a large-n survey of the descendants of Senegalese migrants into France, we are able to show one potential implication of job discrimination, namely, that over two generations Muslims have done less well economically than comparable Christians.

Some big questions remain unanswered, however. We still have not identified the mechanisms by which Muslims face relative economic failure. Consider the results on job discrimination. They may reflect a “taste” that français de souche (rooted French), those with four grandparents born within the French hexagon, have for fellow Christians. They may reflect instead an ability of Senegalese Christians to communicate trust and desire to succeed better than Senegalese Muslims, who lack comparable civic connections (e.g., through the Church) with French people. Alternatively, they may reflect cultural practices of Muslims that signal to French employers a lower commitment to the job.

To address these core questions on mechanisms and relying on the same matching strategy, we have conducted a range of ethnographic interviews and experimental game-theory interventions focusing on the behavior of Senegalese Christians and Senegalese Muslims toward the French and vice versa. Our subsequent papers for this project will assess the explanatory power of these different mechanisms. Future research ought also to compare matched populations in other European states to see if state policies (e.g., multiculturalism vs. republicanism) differ in their success in fostering the integration of Muslim populations.

In this paper, relying on models that combine the controlled conditions of an experiment with a large-N survey, we have established a clear, albeit uncomfortable, finding. All other things being equal, Muslims have faced barriers to economic integration in France that are higher than they would have been if everything about them were the same save for their religion.