Citizens like No Other. Contextual and Individual Explanations of Attitudes towards Roma Population in the EU

Transcription

1 Citizens like No Other. Contextual and Individual Explanations of Attitudes towards Roma Population in the EU Roxana Barbulescu, European University Institute Laurie Beaudonnet 1, University of Montreal Abstract: This paper examines the sources for anti-roma attitudes in the enlarged EU 27. Historically, the Roma population has been one Europe s most marginalized and persecuted minorities. We study the anti-roma attitudes by comparing them to attitudes towards other groups, non-eu migrants and other EU migrants. Recent research has shown that Europeans migrants usually benefit from an "insider" status that softens attitudes towards them, in comparison to non-eu countries (third-country nationals). We argue that it is not the case for Roma. We investigate the individual and contextual explanations for this hierarchy in attitudes, relying on the 2008 European Value Survey. Using multilevel analysis, we study first the impact of salience of the Roma issue in public discourse as well as the size of the Roma population living in the country. 2 We pay a particular attention to potential cleavages between Eastern and Western European Member-States. At the individual level, we test the effects of perceived economic versus cultural threats on negative towards Roma and the impact of attitudes towards Europe. Finally, we study the socio-economic profile of those who are more likely to express positive and negative attitudes towards Roma. To our knowledge, this paper is the first study to systematically explore the sources of Anti-Roma attitudes across European societies. FIRST DRAFT 1 Corresponding author: laurie.beaudonnet@umontral.ca 2 The authors are very grateful to Pavlos Vasilopoulos and Herve Boudou for their help in collecting the issue salience data. 1

2 Introduction Roma people 3 are one of the oldest but also most deprived ethnic minorities in Europe who, despite their relative small number have captured the attention of the European public. The accession to the European Union (EU) of the Central and Eastern European countries which are home to two-thirds of the Roma people in Europe (Council of Europe 2012) gave them the right to move freely and settle in other European countries. After the enlargement, European states could no longer control or restrict the immigration of Roma. Yet, the fact that citizens of the receiving countries largely support more restrictive policies in this area and oppose immigration as a whole (Sides and Citrin 2007, Mclaren 2003, Quillian 1995, Lahav 2004, Ivarsflaten 2005, Hainmueller and Hiscox 2007, Freeman et al 2013) brought more hostility towards the Roma people. How much of a threat can a population as small and socially and economically deprived as the Roma community pose for these societies in objective terms is not so relevant as extremist parties in the East and in the West have managed to achieve electoral success (Norris 2005) by giving voice to an Anti- Roma rhetoric. The success of such strategy in the ballot box spilled over mainstream political parties turning the Roma question into a European-wide matter (Vermeersch 2012 Mark and Braham 2000) and reopening the debate on the integration of the Roma people in all over Europe. Recent conflicts in Italy and France further stirred anti-roma rhetoric and led to exclusionary policies in the two countries specifically targeting the Roma. Without precedent in Europe after World War II, these policies seek the removal of the Roma communities and their so-called voluntary return to their countries of origin (Owen 2012 and Sigora 2011). In both cases these policies were passed by mainstream parties on the left (in Italy) and on the right (in France) cutting through traditional ideological alignments. Violent incidents have also intensified leading to a wave of anti-roma attacks including setting on fire Roma camps in Italy (FRA 2008) and organised anti-roma riots in Hungary, Bulgaria, and the Check Republic (Open Society Institute 2010, BBC 2011, Der Spiegel 2011). This paper seeks to examine the empirical causes of Anti-Roma attitudes in EU27 countries. At the individual level, our empirical findings shows that, while not being migrants in most cases, Roma are 3 We use the term Roma in line with the recommendations of international organization and activist groups for Roma rights. Many Roma people and some activists prefer and identify themselves as Gypsies or Travelers. International organizations such as Council of Europe, which has arguably engaged the most on the Roma rights in Europe, have successively changed its terminology. Since 1969, in it first text relating to the Roma communities, Council of Europe has used Gypsies and other travelers, Nomads (in 1975 and 1983), populations of nomadic origin (1981), Gypsies (1993), Roma (Gypsies) (1995), Roma (1997, 2002), Roma/Gypsies (1995, 1998, 2000), Roma/Gypsies and Travelers (2001), Roma and Travelers (between 2004 and 2010), and Roma since See Jean-Pierre Liégeois The Council of Europe and Roma 40 years of action, 2012, Council of Europe Publishing. 2

3 considered as such by large segments of European populations, and that usual moderators of antiimmigrants attitudes (liberal values and support for European integration) do not apply in their cases. At the country-level, we show that proximity with Roma increases the level of prejudice. In addition to this dimension, a large part of the variation in prejudices against Roma across the EU is explained by the postcommunist specific context. To our knowledge this is the first attempt to identity the factors of Anti- Roma prejudice in all EU27 countries. The next section highlight the theoretical arguments and our hypothesis. We then explain the research design of this study, before presenting and discussing the empirical results. Theorizing anti-roma attitudes across the EU Prejudice may take many forms which can have different social consequences. In this paper we focus on hostility towards having Roma people as their neighbors. Expressing resistance to being neighbors in extremely relevant because it is a direct indicator of resistance to residential integration - one of the most effective tools for de facto inclusion. In addition, mixed neighborhood made up of Roma and non-roma population can have a positive effect on overall integration because they increase contact between groups which in Europe has been shown to have positive effects on tolerance towards immigrants (McLaren 2003). Furthermore, being part of the same neighborhood also gives access to Roma and non-roma children to the same schools. In most European countries, schools serve the nearby neighborhoods which means that in mixed neighborhoods, children from the two groups get to attend the same school and have access to the same educational opportunities at early childhood and adolescence. By contrast, opposition to have a neighbor Roma can have far-reaching social consequences by postponing or even hindering integration. Before we move to review the literature we would like to stress the importance of studying contemporary forms of Anti-Roma prejudice in Europe. Hostility towards the Roma people is not a new phenomenon but the expression of a long history of discrimination. Despite being part of the European nations for nearly a millennium scholars pinpoint their arrival in Europe around 11 th century (Fraser 2000) - Roma people have been subject of prolonged social exclusion and victims of violent acts including pogroms. During WWII, just like Jews, the Roma were victims of Nazi persecution placed in lagers - sometimes in special lagers (Zigeunerlager) - and were murdered in the Holocaust (Muller-Hill 1988, Lutz ad Lutz 1995). It is estimated that one fifth of European Roma lost their life in the Holocaust (US Holocaust Museum Memorial 2012). Nevertheless, the fact that Roma people where victims of Holocaust and the Nazi persecution was not an issue in the Nuremberg trials (Council of Europe 2012: 7) and most European states recognized the Roma as victims of racially motivated acts extremely late. For instance, West Germany recognized them as victims only in 1979, decades after the atrocity. Because of strong 3

4 anti-roma prejudice in the past, monitoring the contemporary levels and understanding modern empirical factors for prejudice are important helping to prevent new escalations in anti-roma sentiment and to inform policymakers about its root causes. Individual factors for Anti-Roma attitudes The first factor we are going to examine is minority membership. Intergroup conflict is lower among members of the non-dominant group because competition is also lower as most resources belong or are perceived to belong to the dominant group. Furthermore, stronger and less prejudiced ideas of attachment tend to predominate towards the own group (Eric and Wong. 2003, Brewer 1999,) while negative bias tends to be associated with members of the out-group leading. For this reason, we expect that members of the non-dominant group to be more tolerant towards other members of the non-dominant group than is the dominant group. Secondly, support for European integration is diffuse in the sense it is the part of support for a political regime that deals with allegiance such as to the regime or feeling of community. Studies (Hooghe and Marks, McLaren 2007; Quintelier and Dejaeghere 2008) have shown that support for European integration is likely to be correlated with higher level of tolerance and inclusiveness along identity lines. Authoritarian views on the other hand tend to be associated with a stronger sense of identity, lower levels of education and lower tolerance towards members of the outgroups. By contrast, more liberal people tend to be more inclusive and more tolerant towards people from other cultures and with different lifestyles. As the stereotyped non-european is often the non-white immigrant Muslim, the indigenous Roma or the Jew (Fligstein 2008). As Mudde (2007: 78) puts it, the non-european gathers the three special enemies against whom Europeans defined their borders and identity for centuries. Indeed, Sides and Citrin (2007) also show that many Europeans regard immigrants from outside of Europe, as well as indigenous minorities in Central and Eastern Europe (CEE), as non-european. For this reason, we expected that anti-roma attitudes are strongly linked with anti-immigrant attitudes. Finally, we consider the effect of following politics in media on prejudice. Media coverage on Roma concentrates nearly exclusively on negative and out of the ordinary events but lacks more comprehensive investigations outside these events which portray the Roma in a negative and stereotypical light (Council of Europe 2012:50-4). Furthermore, the mass media does not give the same level of coverage to cases of discrimination, violence and even hate crimes against the Roma people nor are they contacted when reporting on matters that directly affect them (OSCE 2005). Therefore we expect that the more people follow politics in media, the more exposed they will be to such coverage of the Roma issue, and the more likely they are to have prejudices against this group. 4

5 To sum up, for the individual factors of prejudice, we put forward the following hypotheses: H1 [Minority hypothesis]: Members of minorities are less likely to be prejudiced against Roma than members of the majority. H2 [European attitudes hypothesis]: The more one supports European integration, the less prejudiced against Roma she is. H3 [Liberal attitudes hypothesis]: The more liberal one is, the less prejudiced against Roma she is. H4 [Anti-immigrants attitudes hypothesis]: The more anti-immigrants attitudes one holds, the more prejudiced against Roma she is. H5 [Media hypothesis]: The more one follows the news in the media, the more prejudiced against Roma. Contextual factors for Anti-Roma attitudes Firstly, some studies have found that the size of the non-dominant groups matters and has a positive effect on prejudiced. Larger population, which in a few countries reaches between 7 and 10 % of the total population (Romania, Bulgaria, Slovakia and Hungary) means that they pose a greater material threat to the dominant group. As the largest Roma communities live in Central and Eastern Europe, we expect to have more prejudice in these countries than in the rest of Europe. In addition, we consider the fact that, in countries without large Roma population, the share of EU foreign population might be used as a proxy to evaluate or anticipate the potential threat caused by Roma to the community. As Roma migration is strongly associated in media and political discourse with migrations from Eastern and Central countries, we postulate that the share of EU foreign population has a positive effect on the level of anti-roma attitudes. Secondly, the salience of the Roma issue in the mass media, which typically consists of negative reports about the Roma, can also prime the audience towards hostility. The more salient the Roma issue receives in the mass media the more it exposes the audience to negative and stereotypical news. For instance, in our sample, Italy is the country where the Roma issue has the highest salience: over 2000 news items when even larger countries like Germany only has around 400 news over the same period of time. However in Italy, the Roma issue entered the top political agenda following a murder committed by a Romanian Roma. The news told the story of the murder and reported on the efforts of the Italian police to take special measures which would prevent similar situation, One of the measures was fingerprinting of the Roma people which was regarded as a necessity in Italy and internationally received many criticisms. Lastly, studies (Matras 2000, Barany 1998, 2002) have shown that the Roma people were more likely to suffer discrimination and be subjects of prejudice in Central and Eastern Europe. In the communist era, Roma people were made sedentary and were targeted by social inclusion programs, which helped them 5

6 secure a job and a house in equality of conditions with the non Roma population. However, with the fall of communist regimes across Eastern Europe, and in the transition from a controlled economy to a market economy, the Roma people were among the first to lose their jobs, which significantly impoverished and disenfranchised the community. In addition, ethnic tensions which were controlled under the communist regime came out with renewed force after its fall. Ethnic conflicts leading to violence targeted not only Roma but Muslims in the Balkans and Hungarians in Central Europe. Extremist parties like Jobbik ( Movement for a Better Hungary ), Greater Romania Party, Bulgarian National Guard or Workers Social Justice Party (in Czech Republic) who took from the start an anti-roma agenda also contributed to stirring the conflict between Roma and non-roma people. In what follows, we use the following hypotheses: H6 [Proximity hypothesis]: The larger the Roma population is, the more citizens are likely to be prejudiced against Roma. H7 [Proxy hypothesis]: The larger the EU foreign population is, the more citizens are likely to be prejudiced against Roma. H8 [Issue salience hypothesis]: The more salient the Roma issue in media is, the more citizens are likely to be prejudiced against Roma. H9 [Legacy hypothesis]: Citizens in post-communist countries are more likely to be prejudiced against Roma. Research design: modeling anti-roma attitudes Data To test our hypotheses, survey data must cover attitudes towards Roma, attitudes towards immigrants and attitudes towards European integration. For this reason, we rely on the fourth wave of the European Value Study (EVS) 4, for which data was collected in 2008 and The integrated dataset includes all 27 EU member-states. To test our contextual hypotheses, we used data from three sources. First, the World Bank database for economic indicators 5, second, the Eurostat database for demographic data 6, and data on Roma population, provided by the Secretary General of the Council of Europe for Roma issues (there is no estimates provided for Malta). Finally, we collected data on the salience of the Roma issue in each countries. This collection was done using Google news, using the relevant national edition for each country. Due to coverage limitations of 4 EVS (2010): European Values Study 2008, 4th wave, Integrated Dataset. GESIS Data Archive, Cologne, Germany, ZA4800 Dataset Version ( ), doi: / World Bank, World development Indicators, 6 Eurostat, European Union statistics on international immigration and population stock, online database, 6

7 Google news, data could be collected only for 16 countries: Austria, Belgium (data was collected separately for the two linguistic communities), Czech Republic, France, Germany, Greece, Hungary, Ireland, Italy, Netherlands, Poland, Portugal, Romania, Spain, Sweden, United Kingdom. For each of these countries, the issue salience data was collected for a one-year time period directly preceding the EVS fieldwork. News headlines reported on the country's edition of Google news was used as a measure of issue salience. Headlines were selected based on content, using keywords related to Roma (in average, using two different denominations of Roma) 7, and with search restricted to web pages located in the country itself. Collected data provided us with a count for each country, to be used as a proxy of the Roma issue's salience in the year before attitudes were measured with EVS. Unfortunately, Google news does not exist for all EU countries, more specifically the majority of Northern and Eastern countries are not covered. 8 Dependent variable To measure attitudes towards Roma, we rely on undesirable proximity as a measure of prejudice against Roma: in a list of 15 groups, respondents were invited to mention any group that they would not like to live next to. This binary variable indicates whether respondents would like to have Roma as neighbors (coded 0) or not (coded 1). The exact wording is: "On this list are various groups of people. Could you please sort out any that you would not like to have as neighbours? - Gypsies". We use multilevel logistic regression to model these attitudes while accounting for the non-independence of individual observations within a same country (models are specified with random effects). Independent variables To test the minority hypothesis, we include a measure of membership to another minority, against whom the level of prejudice is high in Europe: Muslims (this binary variable is coded using an original question on religious denomination) 9. To complement it, we use a binary variable coded 1 for respondents born in the country. To measure attitudes towards European integration, we use a set of 5 questions on fears associated with integration, covering social security, national identity and culture, cost, power and jobs, 7 For instance, in the case of the UK, headlines with "travelers" and "Gypsies" were collected; in the case of France and Belgium (French-speaking part), those using the words "les Roms", "tsiganes", "gens du voyages" were collected, for German-speaking countries (Germany and Austria), headlines with "Zigeuner", "Sinti", and "Roma". 8 In order to reduce this bias, we collected data for Romania following a different procedure, as this country has the largest Roma population in Europe and not including it in the analysis seemed to limit it to a large extent. Since there is no Romanian edition of Google news, we counted news headlines of Mediafax, which is the largest and most important news agency in the country. Since in that case data comes from one source only, there is a high probability that the overall salience of the Roma issue is underestimated. 9 Wording is: "Do you belong to a religious denomination?" 7

8 and ranging from 1 to After a factor analysis, a factor score was computed, ranging from negative attitudes to positive attitudes towards integration. 11 In addition, we use a binary variable to capture European identity (coded 1 for 'feeling European", 0 otherwise). Respondents were asked about which geographical group they belong first, and second. The European identity dummy is coded 1 for answering "Europe" as first or second choice. 12 Liberal attitudes are measured with an index, ranging from least to most liberal (it was compiled after a factor analysis including 17 variables on moral values) 13. Attitudes towards immigrants are operationalized with three variables. First a 10-point scale of economic threat (to what extent respondent believes that immigrants take away jobs) and a 10-point scale of cultural threat (to what extent respondent believes that immigrants take over customs) 14. Both range from 1 (most tolerant attitudes towards immigrants) to 10 (least tolerant). Second, a binary variable is included to signal respondents with exclusionary attitudes towards migrants based on ethnic definition of national identity. The variable is coded 1 for respondents for whom having national ancestry is very important for being truly national. 15 To measure exposure to politics in media, we use a five-point scale variable, ranging from following politics every day, to never Wording is: "Some people may have fears about the building of the European Union. I am going to read a number of things which people say they are afraid of. For each tell me if you - personally - are currently afraid of:" 11 Only one dimension was found, with a eigenvalue of 3.2 and all five variables loaded above.8 on this dimension. 12 Wording is: "Which of these geographical groups would you say you belong to first of all / second": locality or town where you live / region of country where you live / country / Europe / the world as a whole. 13 Wording is: "Do you justify: claiming state benefits without entitlement, cheating on taxes, joyriding, taking soft drugs, lying, adultery, bribe money, homosexuality, abortion, divorce, euthanasia, suicide, corruption, paying cash, casual sex, avoiding fare on public transport, prostitution, and insemination or in-vitro fertilization. Each variable ranges from 1"never justifies" to 10"always". Factor analysis resulted in two dimensions (liberalism and morality), out of which the first factor was used to compile the liberalism index (eigenvalue of 4.1, all variables load above 0.55). 14 Wording is: "Please look at the following statements and indicate where you would place your views on this scale? 1) immigrants take jobs away from natives in a country (coded 1)... immigrants do not take jobs away from natives in a country (coded 10). 2) for the greater good of society it is better if immigrants maintain their distinct customs and traditions (coded 1) for the greater good of society it is better if immigrants do not maintain their distinct customs and traditions but adopt the customs of the country (coded 10). 15 Wording is: Some people say the following things are important for being truly [NATIONALITY]. Others say they are not important. How important do you think each of the following is? Having national ancestry. The original variable ranges from 1 "very important" to 4 "not important at all". The binary variable is coded 1 for those who answered "very important", 0 otherwise. 16 Wording is: "how often do you follow politics in media", with intermediate categories being: several times a week, once or twice a week, less often. 8

9 We control for socio-economic status and demographic characteristics. We include gender (reference category is male), age 17, a binary variable for living in an urban area and education level 18. Socioeconomic status is accounted for with a continuous measure of household income 19, and a binary variable for household dependency from social protection 20. Finally, we use four country-level variables to explain intra-country variance in anti-roma attitudes. Two variables measure specific share of population, the first one is EU-foreign population, the second is Roma population, both variables are continuous and measured as share of the total country's population. The salience of the Roma issue is measured with a continuous variable, based on news headline counts. Finally, a binary variable is included to account for the specificity of post-communist countries, in order to test the legacy hypothesis. 21 To be able to compare effects, all continuous individual- and country-level variables are standardized. Results Before presenting our results and explaining the logics underlying anti-roma attitudes in the EU, we start by describing how these attitudes vary within the EU and by comparing them to attitudes towards other minority groups. We then review the impact of individual explanatory factors, before bringing in contextual predictors of anti-roma attitudes. Anti-Roma attitudes in the EU The dependent variable used to model anti-roma attitudes, as presented in the previous section, is part of a question battery that ask respondents to mention any group, in a list of 15, that they would not like to have as neighbors. This allows us to see to what extent people are more prejudiced against Roma than against other groups. Figure 1 below displays the share of respondents who mentioned each group. The three most frequently mentioned groups are drug addicts, heavy drinkers and people with a criminal record, that is, people with morally and to some extent legally reprehensible behaviors. Roma are the fourth most frequently mentioned group, with 40% across the EU saying that they would not like to have Gypsies as neighbors. Other groups with cultural denominations appears far behind, with 22% for Muslims, 17% for immigrants (or foreign workers), 12% for Jews, and 5% for Christians. Despite the fact 17 Age is recoded in 3 categories: years-old (reference category), years-old, and 50 and above. 18 Education is a continuous measure of the highest education level obtained, in 6 categories, with "primary or no education" used as the reference category. 19 Income is the monthly household income, corrected for purchasing power parities in Euros. 20 This binary variable is coded 1 for people who answered yes (for themselves and/or their partner): "During the last five years, have you been dependent on social security at any time?". (social protection here excludes unemployment, disability benefits and pension). 21 This variable is coded 1 for Romania, Bulgaria, Hungary, Czech Republic, Poland, Lithuania, Latvia, Estonia, Slovak Republic, Slovenia, coded 0 for other countries. 9

10 that Roma are European citizens, and in many cases national citizens, negative attitudes towards them are much higher than attitudes towards other minorities, being national, EU or third-country citizens. Figure 2 displays two pieces of information: first, national averages for anti-roma attitudes, and second the hierarchy in prejudices, between Roma, Muslim and immigrant communities. On this map of the EU, the darker the filling, the higher anti-roma attitudes at the national level. Green triangles indicate that, in a particular country, people are most prejudiced against first, Roma, second Muslims, and third immigrants. Orange triangles indicate an inversion in this hierarchy: first Roma, second immigrants, and third Muslims. First, more than 50% of the population in Lithuania, Italy, Czech Republic and Slovak Republic are prejudiced against Roma. Second, among the countries with anti-roma attitudes above the EU average, seven out of 12 are post-communist countries (only Poland, Slovenia, and Hungary are below 40%). Amongst the lowest amount of prejudices, we find Sweden, Luxembourg, and France, with percentages just below 25%. Regarding hierarchy among prejudices, Roma are first in all countries, and Muslims are second, except in Hungary, UK, and Malta, where people are more prejudiced against immigrants than Muslims. In average as well as in each EU country, people are more prejudiced against Roma than against other communities. We now turn to multivariate analysis to understand what explains this high level of prejudices. Figure 1. Prejudices against specific groups in the EU - Figure 2. Anti-Roma attitudes in the EU 80% 70% 60% 50% 40% 30% 20% 10% 0% EU average - European Values Survey 2008 "Don't want as neighbors..." Share of respondents who mentioned each group as undesirable neighbors (questions v49 to v60 in the 2008 European Value Survey), national average. Anti-Roma attitudes are measured as the share of respondents who mentioned Gypsies as undesirable neighbors), national average. Hierarchy in prejudices is based on the respective share of respondents who mentioned 1) Gypsies, 2) Muslims and 3) immigrants / foreign workers as undesirable neighbors (questions v59, v53, v54 in the 2008 European Value Survey), national average. 10

11 Individual predictors of anti-roma attitudes As first step for our analysis, we include the relevant independent variables to test the individual-level hypotheses. Results for the multilevel logistic regression (model 1a) are displayed in table 1 below and reported as odds ratios. In a sample of 15,168 respondents, 5,888 express a prejudice against Roma by mentioning this group as undesirable neighbors, this represent 40% of the sample. Our first hypothesis states that being part of a minority decreases the probability to be prejudiced against another minority (in that case Roma). To test it, we included a binary variable for being Muslim, and a binary measure for being born in the country, as a proxy for people who are most likely to be part of the majority. To be Muslim decreases by 42% the likelihood to be prejudiced against Roma 22, while having been born in the country increases it by 22%. Our second hypothesis deals with attitudes towards the EU, we postulate that positive attitudes towards the EU and European identity decrease the chances to have anti-roma attitudes by creating a feeling of Europeaness, from which Roma would benefit to the same extent than other communities from EU countries. Attitudes towards European integration are measured with an index, ranking from negative to positive, and identity is measured with a binary variable indicating European identity. In both cases, Europeaness decreases the likelihood of anti-roma attitudes: an increase of one point in positive attitudes towards integration yields an 8% decrease in anti-roma attitudes, while feeling European yields a decrease of 11% (respective odds ratios of.915 and.891). Taking into account that attitudes towards integration range from to +2.17, the variation between the least positive attitudes towards integration and the most positive is slightly below 4 standard deviations, which means that the maximum effect of attitudes towards integration is a decrease of 28% in the likelihood to be prejudiced against Roma. According to our third hypothesis, people who hold liberal views will be less prejudiced against Roma. Empirical results do not confirm this, as the coefficient for liberal attitudes is indeed below 1 (so yielding a decreases in the chances to be prejudiced) but is not statistically significant. Hypothesis 3 is thus not confirmed by our results. In the other hand, attitudes towards immigrants (hypothesis five) have the expected significant effect: the three measures ("immigrants take away jobs", "immigrants take over customs" and "it is important to have national ancestry") yield odds ratios higher than 1. The largest effect is caused by economic threat ("immigrants take away jobs"), when a one-point increase increases by 32% the chances to be prejudiced. The variable ranges from -1.7 (immigrants do not take away jobs) to 1.4 (immigrants take jobs away), with a standard deviation of 1, which means a maximum increase of 79% in the likelihood of having anti-roma attitudes. This effect is more than twice larger than the effect of attitudes towards European integration. The effect of cultural threat is less important, with odds ratio of 22 With relative changes in odds calculated as: (1-odds ratio)*100). 11

12 1.087 for the variable "immigrants take over customs", which means a 8.7% increase in the likelihood of anti-roma attitudes for a one point increase in cultural threat (and a maximum increase of 26%). In line, to consider national ancestry as an important feature (to claim citizenship) increases prejudice against Roma by 27% (as opposed to not considering this an important feature). Finally, the last individual-level hypothesis states that news exposure influences the likelihood to be prejudiced. The variable is a five-point scale, ranging from following politics on the news every day, to never. Preliminary analysis (available on request) showed a potential curvilinear relationship with anti- Roma attitudes, inducing a similar effect for the two extremes of the scale. The fact that the squared term has a significant negative coefficient (here translated with an odds ratio below 1) supports this hypothesis. Exposure to political news increases the probability to be prejudiced against Roma when this exposure is frequent or relatively rare. On the contrary, not watching the news at all or watching it every day decreases the probability of such prejudice. Although we do not test for this specifically, it is likely that two logics are at work here. First, people who follow politics on the news every day are likely to have higher level of cognitive mobilization and more sophisticated political analysis, which is likely to decrease prejudices (Inglehart 1971, 1977). Second, if the Roma issue is salient in media, those who never watch the news do not get as much impacted as those who have intermediate level of cognitive mobilization and watch the news on a regular basis. 23 The last set of variables in model 1a consists in socio-demographic controls. Unsurprisingly and in line with the literature on attitudes towards migrants and minorities, we find that the higher the level of education, the less likely to be prejudiced against Roma (for a one-point change in education level, the likelihood to be prejudiced decreases by 6%). In line, people with higher income are less likely to be prejudiced than people from the middle-class. This is showed by the negative coefficient of the quadratic term for income (odds ratio of.824). Income and anti-roma attitudes have a concave curvilinear relationship. This actually shows that people at both extreme of the income scale are less likely to be prejudiced against Roma than people in medium categories. This is coherent with the fact that people who are dependent from social protection are 17% less likely to be prejudiced (odds ratio of.834). Two logics are likely to be at work: on the one hand, people at the higher end of the income scale do not enter in direct competition over material resources such as income jobs, good schools or housing (Bobo 1983). Furthermore, those with more income also tend to be more educated and likely to be more liberal than other categories of the population (Coenders and Scheepers 2003, Hainmueller and Hiscox 2007). On the other hand, the fact that people at the lowest end are also less prejudiced than medium categories do not 23 In a second version of this paper, interaction model to test such logics will be considered, using the data on the salience of the Roma issue. 12

13 fit this logic. We could have here a similar logic to the minority hypothesis described above. People who are dependent from social protection and with very low income might be more tolerant towards Roma because of their own insecure status, just like belonging to a minority reduces the likelihood to have anti- Roma attitudes. Table 1. Individual predictors of prejudices against Roma Model 1a: Prejudices against Roma Odds ratios Odds ratios *Minority hypothesis *Socio-demographic controls Religious denomination: Muslim 0.581*** Gender: female (0.102) (0.035) Born in the country: yes 1.217*** Age: years (0.090) (0.056) *EU attitudes hypothesis Age: 50 and more years 1.109* Attitudes towards European integration 0.915*** (0.061) (0.018) Education level 0.943*** European identity: feels European 0.891** (0.020) (0.047) Monthly household income 1.315*** *Liberal attitudes hypothesis (0.065) Index of liberal attitudes Monthly household income - squared 0.824*** (0.022) (0.037) *Anti-immigrants hypothesis Dependent from welfare provision 0.834*** Immigrants take away jobs 1.320*** (0.045) (0.027) Urban: yes Immigrants take over customs 1.087*** (0.039) (0.020) Constant 0.465*** Important to have national ancestry 1.273*** (0.059) (0.051) Observations 15,168 *Media hypothesis Observations (prejudiced) 5,888 Exposure to politics in media 1.392*** Number of country 27 (0.131) Log-Likelihood Exposure to politics in media - squared 0.721*** AIC (0.069) Estimated residual intra-class correlation Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Measures of goodness-of-fit for the empty model are: Log-likelihood ( ), AIC ( ), Rho (0.06) The specificity of anti-roma attitudes as compared to anti-muslim and anti-immigrants attitudes To put these results in perspective, we investigate to what extent these individual logics are specific to anti-roma attitudes or equally explain attitudes against the other two cultural groups that concentrate the higher level of prejudice in Europe: Muslims and immigrants. Model 1a was replicated twice, respectively using anti-muslim attitudes (model 1b) and anti-immigrants attitudes (model 1c), these models are displayed in Table A1 in Appendix. To summarize, attitudes towards the three groups are similarly impacted (although to various extent) by education level, perception of migrants as cultural and economic threats, belonging to a minority, and exposure to media. Anti-Roma attitudes differ in three ways. To follow up on the effect of socio-economic characteristics, we see a first difference between attitudes towards Roma on the one hand, and attitudes towards the other two groups on the other one. Education has a similar negative effect on prejudices against the three groups. However, neither dependency from 13

14 social protection nor income has a significant effect on anti-muslim or anti-immigrant attitudes. Lower income groups and higher income groups do not express more tolerance towards these two minorities than the average population. Second, liberalism has a clear negative effect on anti-muslim and antiimmigrant attitudes, where it had no effect on anti-roma attitudes. For a one-point increase in liberal attitudes, the chances to be prejudiced against Muslims decreases by 16%, while the chances to be prejudiced against immigrants decreases by 13%. Roma do not seem to benefit from the usual increase in tolerance towards minority groups that liberalism induces. Third, to support European integration yields to less prejudices against all three groups, but the effects vary in size: for a one-point increase in positive attitudes towards Europe, chances to be prejudiced decrease by 8% in the case of Roma (odds ratio =.915, model 1a), by 13% in the case of Muslims (odds ratio =.869, model 1b), and by 15% in the case of immigrants (odds ratio =.850, model 1c). Feeling European decreases prejudices against Muslims and prejudices against Roma to a similar extent: by 11% in both cases, but has no effect on prejudices against immigrants (respective odds ratios =.891 and.885). The fact that positive attitudes towards Europe produces more tolerance is not new, and goes along the argument that people who support integration and are most likely to develop a European identity are those without an exclusive national identity, and with more libertarian attitudes (Hooghe and Marks 2007, McLaren 2002, 2006). However, the novelty lies elsewhere. If we take into account the fact that 56% of immigrants in the EU have a European citizenship 24, it is not surprising that support for Europe has the larger effect on prejudices against immigrants. However, we clearly see that Roma, who are European citizens and often national citizens, do not benefit more from this logic, and to some extent even benefit less, than other groups. Roma do not have the same perceived status than other European citizens. If we combine that with the curvilinear effect of income and the minority hypothesis, we touch upon the dual status of Roma in the EU. In most cases, Roma have the citizenship of the country where they reside, and are thus not migrants. The fact that people in precarious financial situation are less prejudiced against them tends to go in line with this: they are less likely to consider Roma like outsiders, and a similar precarious status induces more tolerance. In such situations, the positive effect of anti-immigrants attitudes on prejudices against Roma can be seen as a side effect of holding less tolerant views about people considered like outsiders to the community, regardless of the actual citizenship of Roma. In other cases, Roma are actual European immigrants (with a citizenship from another EU member-state, in most cases from Romania, Bulgaria and Hungary, the three Eastern countries with the largest Roma population), and in that case they are not as positively perceived as other European immigrants, even by Euro-enthusiasts. 24 Source: Eurostat online database on migrations: 14

15 Bringing context in: how much specific national contexts explain anti-roma attitudes? Until now, we have focused on individual predictors of anti-roma attitudes, while exclusively modeling the contextual dependence of attitudes through multi-level parameters. We now turn to the share of variance that is explained by country-level specificity. Higher-level variables are included one at a time in model 1a, which included individual independent variables and controls. Results for the three models (models 2, 3, and 4) are displayed in Table 2 below. Only higher-level variables are displayed, for full models, see Table A2 in Appendix. To investigate further the distinction between Roma as nationals and Roma as immigrants, we first test the effect of EU foreign population in the respondent's country. Hypothesis 6 (proxy hypothesis) states that the higher this population, as a share of the total population, the higher the chances to be prejudiced against Roma. According to this hypothesis, citizens from host countries are likely to amalgamate Roma with other groups from Eastern and Central European countries. Since Roma concentrate strong negative attitudes, such attitudes are only likely to increase with higher share of EU migrants. In this case, the visibility of EU-foreign population is used as proxy for people to account for the potential share of Roma people in the population. The EU foreign population is calculated as a share of total population and included in model 2. As a control, we also included the share of non-eu foreign population to rule out an effect of foreign population per se, and to establish any specific effect of intra-eu migrations. First of all, we see that there is no effect per se of non-foreign population, but that if something is going on, it is linked to intra-eu immigration, as the share of non-eu foreign population is not statistically significant at all. Second, there is an effect of EU foreign population, but it is the reverse of what we expected. A onepoint increase in the share of EU foreign population, the likelihood to be prejudiced against Roma actually decreases by 15.7%, while we expected a positive relationship. It could be that a large EU immigration, by generating more interactions, increases tolerance vis-à-vis all European citizens, including Roma, similarly to what has been seen for attitudes towards enlargement (Jones & Bijl, 2004). Based on transactionalist argument, they show that trade relations and actual geographical proximity, by increasing the opportunity for interactions, induces a sense of we-feeling, that in turns translate into higher levels of support for enlargement. However, before concluding that this argument applies to attitudes towards Roma, it is worth noting that, among the 12 countries with the lowest share of EU foreign population, ten are post-communist countries, for which we know, after the descriptive analysis presented earlier, that anti-roma attitudes are among the highest in the EU. To rule out any hidden effect, we conduct an additional check, by comparing results from model 2 to those of a supplemental model, run for control purpose. We tested the impact of economic context, operationalized in three different ways: GDP, growth and change in unemployment rate, with the idea that 15

16 the last two variables were capturing the economic crisis. These three models (models 5, 6, 7) are displayed in table A2 in Appendix. GDP has a negative effect on anti-roma attitudes: a one-point increase in GDP decreases the probability to be prejudiced against Roma by 22% (odds ratio of.783 in model 5). Conversely, a one-point increase in growth increases this probability by 16% (odds ratio of in model 6). This apparent contradiction fades away if we consider whose countries have the lower GDP on the one hand, and are experiencing the higher growth rates on the other: out of the twelve countries with the lowest GDP, 10 are post-communist countries, out of the twelve countries experiencing the highest growth, 8 are post-communist countries. The significant effects found for EU foreign population, GDP and growth simply reflect the fact that a) post-communist countries have higher levels of anti-roma attitudes; and b) that they share a set of economic features by being largely sending countries, with GDP lower than the EU average and that, in 2008, most of them continued experiencing economic growth while other EU countries already entered into recession. Our empirical analysis thus shows no effect of economic context and intra-eu immigrations on anti-roma attitudes, but points towards a strong effect of Eastern and Central countries. Table 2. Contextual predictors of prejudices against Roma Prejudices against Roma Model 2 Model 3 Model 4 Proxy Proximity + salience Legacy Odds ratios Odds ratios Odds ratios (Individual-level predictors not displayed) *Higher-level predictors EU foreign population (share of total population) 0.843** (0.073) Non EU foreign population (share of total population) (0.077) Roma population (share of total population) 1.199** (0.095) Salience of the Roma issue (0.173) Post-communist country 1.738*** (0.276) Observations 14,728 9,136 15,168 Observations (prejudiced) 5,743 3,262 5,888 Number of country Log-Likelihood AIC Estimated residual intra-class correlation Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1- Full models are displayed in Appendix. Measures of goodness-of-fit for the empty model are: Model 2: Log-likelihood ( ), AIC ( ), Rho (0.06). Model 3: Log-likelihood ( ), AIC ( ), Rho (0.04). Model 4: Loglikelihood ( ), AIC ( ), Rho (0.06). 16

17 As introduced in the section about theory, Roma are not homogeneously present across the EU, but large variations exists first between Central and Eastern European member-states on the one hand, and Western European member-states on the other, and second within the post-communist countries themselves. Figure 3 below displays Roma population by countries and demonstrates this heterogeneity. For the vast majority of EU countries, the Roma population is marginal, below.05% of the total population. In the former EU 15, the only two countries with Roma population above 1% of the total population are Portugal (8.6%) and Greece (1.5%). Among new member states, the story is different: if Baltic countries and Slovakia s shares of Roma population are very similar to the ones in the EU 15, below 1%, Czech Republic and Slovenia have around 1.5% of Roma population, while Bulgaria, Romania and Hungary have more than 7% of Roma population. Figure 3. Share of Roma population in the EU Estimated Roma population, measured as percentage of total population, national average, 2010 data. Source: The Secretary General of the Council of Europe for Roma issues (no data available for Malta) In line with this diversity, our second contextual hypothesis, the proximity hypothesis, states that the higher the Roma population in a country, the higher the chance to be prejudiced against Roma. An alternative explanation could be that it is not the actual size of Roma population that matters, but the extent to which media and political discourse focus on the Roma issue. In line, our third hypothesis states that the higher the salience of the Roma issue, the higher the chances to be prejudiced. Both logics are tested in model 3, partially displayed in table 2 below (full results available in table A2 in Appendix). Roma population, measured as a share of total population has a significant and positive effect on anti- Roma attitudes. A one-point increase in Roma population yields a 20% increase in the likelihood to be prejudiced against Roma. A respondent in the Netherlands, where the share of Roma population is the smallest, has 56% less chances to be prejudiced against Roma than a respondent in Romania, where the 17

18 Roma population is the largest. Accounting for Roma population reduces by two the remaining variance at the country level (the estimated residual intra-class correlation equals 0.02 in model 3, versus 0.04 in the empty model). Regarding the salience hypothesis, results do not support it, as the measure of issue salience has no significant effect. It is worth noting however, as mentioned in the data section, that data collection for this measure is far from being satisfactory, as data was available for only 16 countries, among which very few Eastern and Central European ones (hence the reduced sample for model 3 and the lower intra-class correlation in the empty model). Despite these results, the salience hypothesis should not be disregarded too fast in future research. Italy is in that respect very illustrative. More than 60% of Italians are prejudiced against Roma (the second more prejudiced population after Lithuania), while actually Roma represent only 0.25% of the total Italian population. However, in the period under scope, the Roma issue is extremely high, with more than 2000 headlines focused on the Roma issue, from October 2008 to October Media salience could be a possible explanation for the high level of anti- Roma attitudes. Obviously, the fact that most Eastern and Central countries are not included in the measure of issue salience due to data availability highlight the need for an alternative operationalization of this hypothesis in order to provide a more thorough empirical test. Figure 4. Salience of the Roma issue in the EU No data Roma issue salience, measured as count of headlines on the Roma issue, as reported in Google news for each country (time period covers one year before the survey data fieldwork). Data gathered by authors. Finally, we test the legacy hypothesis by including in the model a binary variable for post-communist countries. The large difference that we just documented, between Eastern and Central countries on the one hand, and Western countries on the other, regarding Roma population and issue salience, invite to caution regarding the interpretation of such binary variable. Including a binary variable for postcommunist countries is a way to test if the high level of anti-roma attitudes are linked to the specificity 18