What Determines Religious and Racial Prejudice in Europe? The Effects of Religiosity and Trust

Abstract

This paper analyzes the effects of different dimensions of religiosity and trust on religious and racial prejudice in Europe. The sample is based on 37 European countries that are current or potential members of the European Union (EU). Using multi-level logistic regression modeling and the latest wave from the European Values Study data, we test the effects of both individual and country-level variables. Our results suggest that religious particularism is correlated with more religious and racial prejudice. Doctrinal belief and individual spirituality are both correlated with less religious prejudice only. Nonreligious individuals have the highest religious prejudice compared to members of religious denominations. Individual and country-level generalized trust, as well as trust in the EU, are all negatively correlated with religious and racial prejudice.

1 Introduction

The European Union (EU) has been founded on the principles of both economic union and cultural coexistence. However given the very diverse ethnic and religious cultures hosted by the EU, Europeans’ individual experiences and attitudes could be different than the official government positions. Despite enormous efforts by the governments, prejudice against various groups in the EU is very abundant. Prejudice has been conceptualized as an attitude consisting of negative feelings, beliefs, and behavioral intentions toward other social groups (e.g. Dovidio et al. 2000; Jackman 1977). The two most notable groups that receive such prejudice are Jews and Muslims.

Anti-Semitism is known to be prejudice against, or hostility toward, the Jewish people based on hostility to Judaism, and toward Jews as a religious group. The most recent overview of manifestations of anti-Semitism in the European Union, a report prepared by the European Union Agency for Fundamental Rights (FRA), was published in November (2013). This report presented the results of a 2012 FRA survey on experiences and perceptions of hate crime, discrimination, and anti-Semitism among Jewish persons in the EU. According to the survey findings, two-thirds of the almost 6,000 survey respondents consider anti-Semitism to be a problem in the EU Member State where they live; three quarters believe that anti-Semitism has increased in that country over the past 5 years. Anti-Semitism remains an issue of concern in the EU that must be tackled. Similarly, judging from data collected by the British organization Tell Measuring Anti-Muslim Attacks (MAMA), nearly 1,200 anti-Muslim attacks were registered in England and Wales between January 2012 and June 2013. The majority of victims were females, whereas more than 75 % of the perpetrators were reported as being male and under the age of forty. Almost three quarters of anti-Muslim hate crime incidents occurred online.

On the other hand, prejudice against non-Muslims has also been widely documented in Muslim countries and the broader Arab world (Tausch 2014; Tessler and Robbins 2007; Moaddel and Karabenick 2008; Inglehart et al. 2006). Tausch (2014) reports high acceptability for honor killings, stoning, and suicide bombing among a third or more of the entire population of current and potential future EU-candidate countries. Turkey, which is one of the potential candidate countries to the EU, has seen a particularly significant rise in conservatism and religiosity over the last decade (Yeşilada and Noordijk 2010; Erişen et al. 2013; Negron-Gonzales 2012; Şimşek 2013). These trends are considered to be a threat to the very fundamentals of the European Union, which state that: “The Union is founded on the values of respect for human dignity, freedom, democracy, equality, the rule of law and respect for human rights, including the rights of persons belonging to minorities. These values are common to the Member states in a society in which pluralism, nondiscrimination, tolerance, justice, solidarity and equality between women and men prevail” (Article 1).¹

There are numerous other studies on prejudice around the world. Most of these studies have been conducted in the USA, and have mainly focused on anti-black prejudice. In Europe, the number of studies on prejudice has increased considerably since the 1970s (Strabac and Listhaug 2008), and there are good reasons for this increased interest in prejudice and inter-ethnic hostilities in Europe. Pettigrew (1998) argues that since the economic recession in the early 1970s, immigration from developing countries has been perceived as a major problem, prompting four types of reactions: prejudice, discrimination, political opposition, and violence. By the 1980s, immigrants in Europe started experiencing a hostile reaction from sectors of the native populations that felt especially threatened. These threats were exacerbated in countries which experienced unemployment and economic recession. In addition, there is a large degree of xenophobia in the Arab world, which has shown a high level of prejudice against people of a different race or religion, and against immigrants or foreign workers (Tausch 2014). Specifically, Tausch (2014) evaluates Arab public opinion through the “Arab Opinion Index,” which covers 12 Arab countries with 85 % of the population of the entire Arab world. Overall, although Tausch’s results suggest that there is indeed overwhelming support for democracy and change in the region, at the same time, the data imply real weaknesses in civil society support for the basic structures of democracy. The real civil society support for democracy, measured as the percentage of people saying political freedom and civil liberties are a requirement of democracy, in the Arab world ranges from 17 % in Morocco to 49 % in Sudan. These results are also consistent with some other studies (Moaddel and Karabenick 2008; Inglehart et al. 2006).

With the massive immigration in Europe, exploring religious and racial prejudice has become even more important. Most of the relevant studies conducted using European data have focused on anti-immigrant or anti-Muslim prejudice (Scheepers et al. 2002a, b; Strabac and Listhaug 2008). While these types of prejudice target specific groups, this paper presents a broader focus on prejudice. In this paper, we are interested in analyzing the effects of different dimensions of religiosity and trust on religious and racial prejudice, within 37 European countries. Specifically, we analyze prejudice against neighbors who are from a “different” religion or race. This specification of prejudice makes no distinction between, say, attitudes of Christians toward Muslim neighbors and attitudes of Jews toward Christian neighbors, or between attitudes of different racial and ethnic groups.

We use data from the 2008 European Values Survey (EVS), covering 37 countries, and apply multi-level logistic regression modeling to look at prejudice against anyone who belongs to a religion or race that is different than the respondent’s. This study uses individual and country specific variables. In this paper, we are concerned with three main questions: (1) Do different dimensions of religiosity have an effect on religious and racial prejudice? (2) Do different dimensions of trust have an effect on religious and racial prejudice? (3) Is there any variation among levels of religious and racial prejudice between countries?

2 Literature Review

Given the complex nature of prejudice, it is difficult to define and measure the concept. The best known attempt is probably Allport’s (1954) [1979] definition of ethnic prejudice: “an antipathy based upon a faulty and inflexible generalization” (p. 9). Although different research and theories of prejudice suggest different starting points for the formation of prejudice, this study adopts Allport’s (1954) [1979] eclectic approach. This model stresses that no single theory of prejudice is adequate, and that multiple historical and sociocultural perspectives on the actual group differences are needed to understand prejudice. Following Allport’s approach, this study considers the importance of both micro-level (individual-level) and macro-level (country-level) determinants of prejudice. Theories of prejudice on each level have been tested and supported, and we review them in the next section.

2.1 Theories of Prejudice and Prior Research

2.1.1 Individual (Micro-Level) Theories

According to Davis (1959), relative deprivation theory concludes that when a deprived person compares himself with a non-deprived person who is within the same in-group, the resulting state is called “relative deprivation.” Especially in situations of economic hardship, relative deprivation intensifies prejudices about an outgroup, and is often expressed as aggression. Split labor market theory (Bonacich 1972) is another dominant perspective on the sources of ethnic and racial antagonism, mobilization, and conflict (Kunovich 2004). A split labor market exists when the price of labor differs across ethnic or racial groups (Bonacich 1972). Typically, those in the dominant ethnic or racial group seek to prevent direct competition in the labor market, by excluding ethnic or racial groups who might be willing to perform comparable work for less money. Thus, prejudice is more common among dominant ethnic or racial group members who are competing directly with ethnic or racial minorities. Prior research supports both approaches. Lower social class (specifically lower levels of income and education) is associated with higher levels of prejudice (Carvacho et al. 2013), and unemployed people are also found to have more prejudice (Kunovich 2004). In addition, prior research focusing on individual characteristics consistently finds that men and older adults exhibit more prejudice than their counterparts (Stewart et al. 2009; von Hippel et al. 2000). Married individuals and those whose parents are both citizens are more prejudiced, whereas those who have ever lived abroad are less prejudiced (Kunovich 2004).

In addition to these individual characteristics, prior research on religion and prejudice within the psychology of religion literature has generally been concerned with types of religiosity that predict prejudice or tolerance. Allport and Ross proposed that (religiously) intrinsically motivated individuals internalize religious values related to “humility, compassion, and love of neighbor” (1967, p. 441) and thus are not prejudiced, whereas more extrinsically religious individuals are utilitarian and more dogmatic in their social attitudes as well as their religion, and thus endorse prejudiced ideologies that promote their group’s interests. Research generally demonstrated that an intrinsic orientation was related to positive outcomes, while an extrinsic orientation was inversely related to positive outcomes (Park et al. 1990; Rowatt and Kirkpatrick 2002; Salsman et al. 2005). Other concepts, such as religious fundamentalism and church attendance, have also been linked with prejudice (Hunsberger 1996; Kirkpatrick 1993; Eisinga et al. 1990; Scheepers et al. 2002b). People who have a fundamentalist approach to religion were found to have higher levels of prejudice (Altemeyer 1996; Altemeyer and Hunsberger 1992).

Moreover, frequency of church attendance has been linked with less prejudice (Allport and Ross 1967) and some cross-cultural research has also replicated this pattern (Eisinga et al. 1990). This finding, however, is not consistent in prior literature. In an extensive review of (mostly American) studies of the association between prejudice and indicators of religiosity conducted between 1940 and 1990, Batson et al. (1993) conclude that within this framework of analysis, “the more religious an individual is, the more prejudiced he or she is likely to be’’ (p. 296). Most European studies find no or a very weak relationship between religiosity and prejudice (Hunsberger and Jackson 2005; Scheepers et al. 2002b; Konig et al. 2000; Duriez and Hutsebaut 2000). Prior research has been mostly consistent about the effects of doctrinal beliefs (i.e., subscribing to religious beliefs) and religious particularism (i.e., to what extent people believe there is only one true religion) on prejudice, whereas doctrinal belief significantly reduced prejudice and religious particularism significantly increased prejudice (Eisinga et al. 1995; Konig et al. 2000; Glock and Stark 1966; Scheepers et al. 2002b). Prior research on the effect of individual spirituality (i.e., the spirituality people experience) on prejudice is limited; the results, however, suggest that individual spirituality was found to significantly reduce prejudice (Hood et al. 1996; Scheepers et al. 2002b).

2.1.2 Intergroup (Macro-Level) Theories

According to Group Conflict theory, prejudice is caused by intergroup conflict. When two groups compete for limited resources or the same goal (e.g., a limited number of jobs), it causes frustration and therefore conflict, prejudice, and discrimination. Prejudice and intergroup tensions are exacerbated when groups are, or perceive themselves to be, in conflict with other groups for valued resources such as money or power (e.g., Sherif 1966). Thus, when different groups compete for resources, negative attitudes toward out-groups may be exacerbated. Strabac and Listhaug (2008) find some support for group conflict theory, in that unemployed individuals in their study had significantly higher anti-Muslim prejudice.

Prejudice towards another group can arise from a perceived threat from that group. Group Threat Theory differentiates between two kinds of perceived threats, namely “realistic” and “symbolic” threats. Increased competition in the labor market and for other resources in the economy could be examples of realistic threats. Symbolic threats, on the other hand, involve the perceptions that the prejudice-targeted group might disrupt the cultural and religious values of the dominant group. Several researchers analyze the causes of prejudice by focusing on these two possible explanations (Schlueter et al. 2008; Tolsma et al. 2008; Savelkoul et al. 2011; Hooghe et al. 2013). However, all of these researchers focus on one or two countries or several regions within the same country. Furthermore, they only consider the prejudice of citizens against immigrants. Prejudice by the minorities against citizens cannot be captured by this specification.

According to group threat theory indicated by the relative size of immigrant populations and the state of the economy, the presence of a large minority population is expected to produce prejudice. Strabac and Listhaug (2008), however, find no support for the group threat theory where the percentage of Muslims in a country is not significantly correlated with anti-Muslim prejudice. Where group threat theory suggests that the presence of a large minority population living near whites arouses economic and/or political threats among whites, thus resulting in prejudice (Blalock 1967; Quillian 1995), intergroup contact theory proposes that the presence of a large minority population living near whites may lead to more frequent interactions between minorities and whites (Sigelman et al. 1996; Quillian 1996). Depending on its quality, among other things, this contact is associated with reduced prejudice (Allport (1954) [1979]).

Social identity theory (Tajfel 1981) proposes that we see ourselves as belonging to categories (e.g., racial group, gender), and compare the groups we are in (in-groups) to other groups that we do not belong to (out-groups). Our self-esteem is increased by regarding our own group as superior to others, which leads to a biased view of in- and out-group members, causing prejudice. Tajfel proposes that mere identification with a group is enough to cause hostility toward other groups. Those who have strong social identity are expected to be less socially integrated. Durkheim (1893) argued that integration of a group of people into the mainstream of society is important for social cohesion. Using social integration theory, a society characterized by high levels of social trust could achieve the integration of immigrants more easily than a society with lower levels of social trust. Moreover, higher interpersonal trust at both individual and regional levels were the strongest and negative predictors of anti-immigrant attitudes (Herreros and Criado 2009; Rustenbach 2010), and trust was also found to reduce inter-religious prejudice (Tam et al. 2008).

In addition, certain cultural traits and institutional factors might affect the level of prejudice across countries. In addition to being individual-level characteristics, generalized trust and religiosity are also considered to be two important cultural traits. Sources of prejudice and discrimination are often rooted in particular historical and social contexts which are influenced by these cultural traits, and which then shape institutional structures and practices (Alesina and Guiliano 2013). This would explain some of the differences in prejudice levels among different countries. Prior studies have suggested that a recent history of traumatic experiences, belonging to a group that historically felt discriminated against, being economically unsuccessful in terms of income and education, and living in a racially mixed community and/or one with a high degree of income disparity are all associated with low trust, which then affect levels of prejudice (Guiso et al. 2009; Alesina and La Ferrara 2000, 2002). In addition, religiosity is an important cultural trait and there is a positive correlation between religiosity and trust, but a negative correlation between religiosity and tolerance among different cultures (Guiso et al. 2003).

2.1.3 Extension of Prior Research

Our analysis departs from previous literature in several ways. First, we use the most recent wave from European Values Survey (EVS) data, which includes information on 37 European countries, to analyze the prejudice against people from other religions and races, as measured by respondents’ willingness to accept them as neighbors. To the best of our knowledge this is the most comprehensive cross-country analysis in Europe of the influence of various dimensions of religiosity and trust on both religious and racial prejudice. The indicator we use from EVS data (“On this list are various groups of people. Could you please sort out any that you would not like to have as neighbors?”) has been used in prior research to measure social tolerance or prejudice. Reeskens (2013) used this same indicator to measure social tolerance, but he collapsed all the answer categories into one group. Strabac and Listhaug (2008) also used this indicator to measure anti-Muslim prejudice, but no other types of prejudice.

Our specification departs from both these approaches. Contrary to Reeskens (2013), we do not collapse different groups into one category, but instead focus on “other religion” and “other races” separately. This allows us to identify differences between religious and racial prejudice in Europe. Also, contrary to Strabac and Listhaug (2008) we do not focus only on prejudice against a specific religion. We instead analyze prejudice against anyone from “another” religion or race. Finally, we utilize both individual level and country specific variables using multi-level logistic regression modeling. Although there are various theories and thus different starting points for the formation of prejudice, this study focuses on trust and religion-related variables. Thus, we formulate the following six hypotheses:

Hypotheses

• 1. Subscribing to doctrinal beliefs, individual spirituality, and more frequent religious service attendance are expected to be correlated with less religious and racial prejudice.

• 2. On the other hand, religious particularism is expected to be correlated with more religious and racial prejudice.

• 3. Those who belong to a religious denomination are expected to have less religious and racial prejudice.

• 4. A higher generalized trust is expected to be correlated with less religious and racial prejudice.

• 5. A higher trust in the EU is expected to be correlated with less religious and racial prejudice.

• 6. In addition to the individual level variables, we expect some country-level predictors, such as GDP, unemployment rate, population density, average trust, and ratio of immigrants, to have an effect on religious and racial prejudice.

3 Data and Variables

3.1 Sample

The European Values Study (EVS) is a large-scale cross-national survey that collects information on the values of people living in Europe. The survey has been conducted every 9 years since 1981, with more countries participating each year. The survey’s main goal is to investigate how Europeans think about life, family, work, religion, politics, and society. There are around one thousand respondents from each country in each wave. The last wave of the survey includes 47 countries, some of which are not part of European Union but are located in the region.

In this paper we utilize the last wave of the survey, which was conducted in 2008. We focus only on the countries that are current members of the European Union, candidate countries of the EU and potential candidate countries.² Our final data includes 37 countries, with a total of at least 27,000 observations. Beyond religion and trust related variables, we also use demographic variables and country level aggregate variables in our models. The list of the variables, along with their descriptive statistics, can be found in Table 1.

Table 1 Descriptive statistics and definitions of all variables

In our sample 34 % of the respondents are found to have religious prejudice, whereas only 16 % have racial prejudice. These numbers vary considerably among individual countries (Table 2). Turkey has the highest level of religious prejudice (68.7 %), followed by Estonia (64.7 %). Northern Cyprus takes the lead in racial prejudice with 55 %, followed by Turkey with 42 %. Hungary ranks last in both categories, with 9.6 % of respondents having religious prejudice and only 1.1 % having racial prejudice. Definitions and descriptive characteristics of all variables can be found on Table 1.

Table 2 Average religious and racial prejudice by country (in percentages)

3.2 Dependent Variables

We have two different dependent variables, both of which are binary. The first one measures the level of prejudice against neighbors from a “different” religion, and the second one measures the prejudice against neighbors from “other races.” The question on neighbors is particularly important because individuals could have different attitudes when thinking about people living in their country as opposed to people living next door. In this paper we only focus on religious and racial dimensions of prejudice.

The dependent variables were created in the following manner. There are three questions about accepting neighbors from various religions; the options are “Muslims,” “Christians,” and “Jews.” The responses to these questions are matched with the religious denomination of the respondents. In the master file of EVS 2008, there are 9 different religious affiliations that could be chosen by the respondents. Muslim and Jewish are two of the options. There is no single “Christian” option, but respondents can choose Protestant, Roman Catholic, or Orthodox options. Therefore we coded anyone who chooses these options as a Christian.³ Then, if anyone has chosen an answer of not wanting a neighbor who belongs to a religious group other than his/her own, that respondent is given a value of 1 for religious prejudice. Otherwise the value is 0. For those respondents whose religious affiliation is other than the three main categories mentioned above (including the no-religion category), we look at whether or not they declared neighbors from any one of the three categories as unwanted. Racial prejudice is measured by using the responses to another question. If the respondents answer affirmatively to the question “do you not want people from other races as neighbors?” then racial prejudice variable is coded as 1; otherwise it is coded as 0.

3.3 Independent Variables

There are two main dimensions of our independent variables, religiosity and trust. In this study, to measure religiosity we create the following five variables. Religious particularism is measured by asking the respondents to what extent they think there is only one true religion. The four response categories are: (1) “There is only one true religion”, (2) “There is only one true religion, but other religions do contain some basic truths as well”, (3) “There is not one true religion, but all great world religions contain some basic truths”, (4) “None of the great religions have any truths to offer.” The answer categories are reverse coded so that a higher number indicates more religious particularism. Individual spirituality is measured by asking the following question: “Whether or not you think of yourself as a religious person, how spiritual would you say you are, that is how strongly are you interested in the sacred or the supernatural?” The answer categories are: (1) “very interested,” (2) “somewhat interested,” (3) “not very interested,” and (4) “not at all interested.” The answer categories are reverse coded so that a higher number indicates more spirituality.

Third, doctrinal belief is measured by asking the respondents the following question: “Which of the statements comes closest to your beliefs?” The answer categories are: (1) “There is a personal God,” (2) “There is some sort of spirit or life force,” and (3) “I do not really think there is any sort of spirit, God or life force.”⁴ The answer categories are reverse coded so that a higher number indicates higher doctrinal belief. We also create a variable to measure frequency of attendance at religious services, by using the response to the question “Apart from weddings, funerals and christenings, about how often do you attend religious services these days?” The answer categories range from 1 (“more than once a week”) to 7 (“never, practically never”). We reverse coded this item so that higher numbers for this variable indicate more frequent attendance. Finally, religious denomination is used as a dummy variable to distinguish between non-religious people (the reference category), Catholics, Protestants, Orthodox Christians, Muslims, and those who belong to other religious groups (i.e., Jew, Hindu, Buddhist, Other). All of these variables are intended to capture the different aspects of religion in respondents’ lives.

We measure the reliability and validity of all these measures for religiosity. The Cronbach’s alpha of these five indicators is 0.73, indicating a high reliability. The exploratory factor analysis suggests that these five indicators load on one factor, which indicates that these five indicators measure the same concept. Finally, we run confirmatory factor analysis (CFA) to test whether and how well these five religion indicators measure religiosity. The fit indices show that the model fits the data well (χ ² = 408.13, df = 5; CFI = 0.99, RMSEA = 0.04). Finally, we add the correlation matrix in Table 7 in “Appendix” to show that there is no concern for high multi-collinearity among these variables.

We also include two dimensions of trust in this paper: (1) trust in individuals and (2) trust in institutions (Paxton 1999). Trust in individuals or generalized trust is measured by responses to the question “Generally speaking, would you say that most people can be trusted or you can’t be too careful in dealing with them?” This question has two categories, where 0 refers to “can’t be too careful” and 1 refers to “most people can be trusted.⁵” In order to include trust in institutions, we also include confidence in the EU. This measure is created as a dummy variable by using responses to the question “How much confidence do you have in the European Union?” There are four answer categories ranging from 1 (“a great deal”) to 4 (“none at all”). We code those respondents as 1 (“have confidence in the EU”) who reported either “a great deal” or “quite a lot” of confidence, and 0 (“don’t have confidence in the EU”) who reported either “not very much” or “none at all.”

3.4 Control Variables

Following Reeskens (2013) and Strabac and Listhaug (2008), we include several individual level control variables in our estimation. These variables are age, gender, education, immigrant status, marital status, and employment status. One additional control variable is the size of the city the respondent is living in. It could be argued that one’s neighbor’s religion and race are more important in small cities where there is more interaction among the neighbors. Finally, we also control for household income. Based on the country-specific annual household income, we create a dummy variable which includes those individuals who are in the lowest quartile (25th percentile) within each country, and then group together those individuals across all 37 countries.

Finally, we include some country level variables that could be potential correlates of religious and racial prejudice. These variables are per capita GDP, unemployment rate, population density, mean generalized trust in that country, and mean percentage of immigrants in the respondent’s country. Some of these variables have also been used by Strabac and Listhaug (2008) to explain anti-Muslim prejudice in Europe. The variable capturing average level of generalized trust in the country, however, is unique to our study. We construct this variable by measuring the percentage of respondents in each country in our data set who said that they “trust people in general.” Similarly, the share of immigrants is also calculated by using the EVS 2008 data. Other variables are obtained from outside sources.

The first three of the variables above are compiled from EUROSTAT⁶ and the numbers are for the year 2007. Since EVS was conducted in 2008, we used the country level variables corresponding to the previous year. The mean generalized trust and percentage of immigrants in the surveyed countries are constructed by using EVS 2008 data, and then respondents in each country are assigned the corresponding values.

4 Analytical Strategy and Results

We analyze the individual and contextual correlates of religious and racial prejudice using multi-level logistic regression (MLR) analysis. This is a strategy that is eminently suitable to account for clustering, especially when respondents in cross-national datasets are sampled within countries. The multilevel modeling is performed using the xtmelogit command in Stata 11. For multilevel logistic regression analysis with a small number of level-2 units (N = 37 countries), the estimates of higher-level variance terms and their standard errors are known to be unreliable (Hox 2002). Due to this, and since our main interest is in the effects of individual and country-level predictors and not in changes in variance of level-2 residuals, we only present the fixed parts of the models in the final tables.

Latent response multilevel logistic modeling can be modeled as follows. Let y _ij be the prejudice level of individual i in country j. Since this level is not observed, we have the latent dummy variable that only shows if the respondent has prejudice or not:

$$ y_{ij} = \left\{ \begin{gathered} 1,\quad y_{ij}^{*} > 0 \hfill \\ 0,\quad otherwise \hfill \\ \end{gathered} \right. $$

Then we can model the observed response as a function of individual and country level observable variables, as well as country level unobservable variables.

$$ y_{ij}^{*} = \beta_{0} + \beta_{1} x_{1ij} + \beta_{2} x_{2j} + \eta_{j} + \varepsilon_{ij} $$

with η _j  ∼ N(0, Σ) and ɛ _ij has standard logistic distribution, and is independent of η _j . η _j is a measure of country level unobservable characteristics. x _1ij , x _2j correspond, respectively, to individual and country level covariates. MLR allows us to estimate the effect of covariates by taking into account the unobservable characteristics between countries.⁷ Thus β _i (i = 1,2) shows the effect of increasing the value of the covariates by one unit without changing the country. β ₀ shows the log-odds of prejudice when all the covariates and country-level unobservable characteristics are equal to zero. MLR also allows us to calculate the Variance Partition Coefficient (VPC), which is the proportion of the residual variability in the prejudice level that is due to between-country variation. The formula for VPC is given below:

$$ VPC = \frac{\varSigma }{\varSigma + 3.29} $$

We start testing the nested models to determine which model fits the data the best. The results are shown in the last three rows in Tables 3 and 4, which show the goodness of fit of the different models for explaining religious and racial prejudice. We have six models for each of the two dependent variables. Model 1 is a null model, which is the logistic regression that includes only an intercept and allows only for individual-level variation. Model 2 estimates a model that incorporates country-level variations in the intercept. In this random intercept model, we allow the intercept to vary between countries. Next, we include all control variables in Model 3 (except religiosity and trust measures). Then, we test the effects of religiosity measures by including all five dimensions of religiosity in Model 4. In Model 5, we test the effects of two trust measures by including generalized trust and trust in EU in the model. Finally, in Model 6, we add all five country level characteristics to the model. We replicate the same pattern for both religious and racial prejudice.

Table 3 Results of multilevel logistic regression analyses for religious prejudice (N = 23,560)

Table 4 Results of Multilevel logistic regression analyses for racial prejudice (N = 27,586)

4.1 Multilevel Logistic Regression Results

The results for religious prejudice and racial prejudice are shown in Tables 3 and 4, respectively. The log-likelihood values for each model are represented at the bottom of the tables, and they indicate that the best model is the model where we have all the individual and country level variables (for both religious and racial prejudice). Likelihood ratio tests also show that these models are better than ordinary logistic regression models. Thus we only interpret the estimates from the best model (Model 6) in both tables.⁸

Model 6 in Table 3 reveals that generalized trust, trust in EU, and all of the religious variables except religious service attendance are significant predictors of religious prejudice within Europe. These effects are all in the expected direction. For instance, religious particularism is correlated with higher religious prejudice, whereas doctrinal belief and individual spirituality are correlated with less religious prejudice. Higher generalized trust and trust in the EU are both correlated with less religious prejudice. For example the odds of expressing religious prejudice decrease by 35 % for those who trust people in general, and by 10 % for those who trust the EU, compared to their counterparts.⁹ People who belong to any religious denomination are also less likely to have religious prejudice compared to non-religious people. Effects of country-level predictors presented in Model 6 (Table 3) do not show much support for group-level theories of prejudice, but only the mean level of generalized trust has a significant and negative effect on religious prejudice. Thus, both individual and aggregate levels of generalized trust significantly decrease the odds of religious prejudice in Europe.

The effects of our independent variables on racial prejudice (in the final model) are shown in Model 6 of Table 4. The only significant religious variable is religious particularism: more religious particularism is correlated with more racial prejudice. The odds of expressing racial prejudice increase by 14 % with an increase in the level of religious particularism. Generalized trust and trust in the EU are both significant in predicting racial prejudice, in the expected direction. The odds of expressing racial prejudice in Europe within a country decrease by 25 % for individuals who trust people in general and by 11 % for those who have trust in the EU, compared to their counterparts. Effects of country-level predictors presented in Model 6 (Table 4) do not show much support for group-level theories of prejudice, but only the mean level of generalized trust has a significant and negative effect on racial prejudice.

Among the demographic variables, the results suggest that those who have a higher education level, immigrants, females, and those who live in larger cities are less likely to have religious and racial prejudice than their counterparts. In addition, those with less wealth are more likely to have racial prejudice, but this effect is not significant for religious prejudice. The odds of racial prejudice increase by 12 % for those who are in the lowest quarter of the income distribution. The odds of religious prejudice decrease by 31 % for immigrants compared to natives. All of these effects hold for individuals who are within the same country and who have the same characteristics. In order to find out the between-country variation of these coefficients, we calculate the variance partition coefficients (VPC). The VPC for each model is also reported in each table. We can see that about 10–15 % of the variation in religious prejudice and about 11–18 % of the variance in racial prejudice can be explained by between-country differences that were not controlled for in our current models. This is consistent with the predictions of Alesina and Guiliano (2013), who showed that cultural traits are important for predicting different prejudice levels. Our finding calls for further attention to more country-specific variables in future studies’ estimations.

In additional analyses, we also tested whether religious denomination, immigrant status, and wealth moderate the effects of trust and religiosity on religious and racial prejudice.¹⁰ We show only the significant interactions in Tables 5 and 6. As shown in these tables, generalized trust is correlated with less religious and racial prejudice, whereas religious particularism is correlated with more religious and racial prejudice. Both of these effects are stronger among those in the lowest 25th percentile of the income distribution, Muslims, and Orthodox Christians (for religious prejudice only), and weaker for Protestants (for racial prejudice only). The positive effect of religious particularism on religious prejudice is stronger among those who belong to any religious denomination, whereas the same effect is stronger among Catholics and Orthodox Christians for racial prejudice only. The negative effects of doctrinal belief, individual spirituality, and religious service attendance on religious prejudice are stronger among those who belong to any religious denomination. The negative effects of doctrinal belief and individual spirituality on racial prejudice are both stronger among only Muslims and Orthodox Christians. In addition, the negative effect of individual spirituality on racial prejudice is stronger among Catholics.

Table 5 Results of multilevel logistic regression analyses including interaction variables (religious prejudice) (N = 23,560)

Table 6 Results of Multilevel logistic regression analyses including interaction variables (Racial Prejudice) (N = 27,586)

5 Discussion and Conclusion

The aim of this study was to analyze the effects of religiosity and trust on religious and racial prejudice among respondents from 37 current and potential member countries of the European Union, using 2008 EVS data. There are several important findings. The multilevel logistic regression results suggest that religious particularism is correlated with more religious and racial prejudice. Doctrinal belief and individual spirituality are both correlated with less religious prejudice, but not correlated with racial prejudice. Nonreligious individuals have the highest odds of having religious prejudice compared to people in religious denominations. These results are consistent with the findings of Eisinga et al. (1990), who argue that religious particularism is a strong contributor to prejudice. According to Eisinga et al. (1990), when people consider their religion to be the only true religion, they tend to have an unfavorable attitude toward other religion out-groups, and in fact to all religious groups not following the traditional Christian teachings. The present study’s results also support the conclusion that people with experiences of a spiritual, supernatural, or other-worldly nature may put inner-worldly tensions between in-groups and religious ethnic out-groups into another perspective, i.e. one of mutual tolerance (Scheepers et al. 2002b).

These results support the social integration theory. Based on our results, individual generalized trust, country-level generalized trust, and trust in the EU are all negatively correlated with religious and racial prejudice. The finding on generalized trust is consistent with the findings of Torpe and Lolle (2011), who suggest that the traditional survey question used to measure social capital is not capturing trust in strangers as “people one meets for the first time” but as “people of different nationality and religion.” The negative effect of generalized trust at the country level on both religious and racial prejudice indicates that those who live in countries with higher levels of social capital (as captured by average generalized trust in that country) are less likely to be prejudiced against neighbors from another religion, as well as against those who are from a different race. This result indicates that one effective way of fighting against prejudice would be to promote social capital, both at the individual level and at the aggregate level. Improving trust among residents at the individual level as well as at the country level will have a direct impact on reducing their religious and racial prejudice. This should be an important implication for the policy makers interested in reducing religious and racial prejudice across the EU. Lastly, the results also support the relative deprivation theory. Specifically, immigrants and those with higher education have less religious and racial prejudice, whereas those who are poor have more racial prejudice.

The findings of prior research on the effects of religiosity on prejudice are not consistent (Village 2011; Batson et al. 1993; Hunsberger and Jackson 2005; Scheepers et al. 2002b; Konig et al. 2000; Duriez and Hutsebaut 2000). The differential effects of different dimensions of religiosity on prejudice support similar findings from prior research (Scheepers et al. 2002b; Strabac and Listhaug 2008), emphasizing the importance of broad measurement of this concept. The fact that all religiosity measures we use are significant in predicting religious prejudice, and that only religious particularism significantly and positively affects racial prejudice, gives support to the conclusion that each form of prejudice should be considered separately, rather than combining different forms of prejudice into one scale. Finally, our study suggests that the effects of individual and country-level variables are expected to vary depending on the type of prejudice under discussion.

With regards to religious denomination, we find that individuals who belong to the Islamic religion have the highest likelihood of having religious prejudice compared to members of other major religious denominations in Europe. As mentioned above, other researchers have shown high levels of prejudice among Muslim countries and in the Arab world (Tessler and Robbins 2007; Moaddel and Karabenick 2008; Yeşilada and Noordijk 2010; Tausch 2014; Inglehart et al. 2006). Our findings also indicate that there is significant religious prejudice (the highest is among Muslims) within the EU, and that policy-makers should be careful about these issues at the political level.

Several extensions of our work are possible. This study has a cross-sectional design, so future research could test whether the effects of these individual and country-level predictors of religious and racial prejudice change over time. This argument is consistent with the findings from a prior longitudinal study, which found that some country-level variables had significant effects on anti-immigrant prejudice between 1988 and 1997, but none of the same variables had significant effects in the year 2000. The authors concluded that the effects had been diminishing over time (Semyonov et al. 2006). Considering the longitudinal design of the EVS as well as other datasets such as the World Values Survey (WVS), a time series analysis among countries could be carried out and changes in prejudice levels, as well as the changing effects of individual and country-level predictors, can be analyzed. In addition to the predictors we use in this study, consistent with prior research, other potentially useful country-level predictors could include religious heterogeneity, immigration policy, and national left/right political leaning. On the individual level, political affiliation and religious fundamentalism could be added to the current analyses (Rustenbach 2010; Hjerm 2007; Laythe et al. 2002; Scheepers et al. 2002b).

Future studies could also measure prejudice in different ways. As stated before, in our study, we do not make any distinction between attitudes of different racial groups, or between different religious groups. Using large datasets, future studies could make these distinctions and test the levels of prejudice between various groups. Finally, along the same lines, future studies could also test the levels of prejudice within the same religious and/or racial group. For example, we found that in our dataset that almost none of the Christians had religious prejudice against other Christians, whereas prejudice among Muslims against other Muslims is noteworthy. We hope that future research could explore some of these extensions. Finally a methodological extension of this work could include using random slope multi-level modeling, which assumes that the effects of covariates are different between countries. This could shed more light on questions such as whether a Muslim living in Turkey has different levels of religious prejudice than a comparable Muslim (i.e., someone who has similar socio-demographic and background characteristics) living in Germany.

Overall, this study makes a significant contribution to prior studies on prejudice by using both individual and country-level variables to test the effects of religiosity and trust on religious and racial prejudice among 37 countries in Europe. Results on both types of prejudice show support for social integration theory, with regards to the effect of generalized trust at both the individual and country levels. Trust in institutions (i.e., trust in the EU) also had a significant and negative effect on both religious and racial prejudice. In addition, all five dimensions of religiosity received significant support for predicting religious prejudice, whereas only religious particularism had a significant positive effect on racial prejudice. Moreover, consistently with some prior research (Semyonov et al. 2006; Strabac and Listhaug 2008; Scheepers et al. 2002a), the results suggest that compared to country-level predictors, individual-level predictors (and thus individual-level theories) receive more support.

Appendix

See Table 7.

Table 7 Correlation matrix between religiosity measures