Geographies of EU dissatisfaction: Does spatial segregation between natives and migrants erode the EU project?

ABSTRACT This article examines whether the spatial segregation of immigrant groups conditions the relationship between the share of migrants and attitudes towards EU integration. Do varying degrees of clustering of migrants in space diminish, mitigate or exacerbate the effect of immigration on EU attitudes? We combine data from the European Social Survey with fine-grain spatial segregation measures captured by the D4I project across four European Union countries. We find that those who live in regions with a greater share of migrants from Eastern Europe have more positive attitudes towards the EU but that this positive influence diminishes in highly segregated areas. The analysis also shows that the effect is primarily driven by the working class. Our findings have important implications as they show that the joint consideration of both the levels and the distribution of migrants in space is crucial to understanding the relationship between immigration and attitudes towards the EU.


Introduction
This article examines whether the degree of clustering of migrants in spacethat is, spatial segregation-decreases, mitigates or exacerbates the effect of the level of foreigners on attitudes towards the European Union.In those contexts where migrants live spatially segregated from natives, do people express more negative feelings to the EU project compared to places where migrants are less concentrated?Over the past few years, the relationship between spatial segregation and political and social attitudes has been a matter of scholarly interest, as well as a matter of concern for policy-makers.
Yet, the focus on EU attitudes has been more limited.This has been the case even though spatial segregation lies at the very heart of the different strategies envisaged by the EU institutions and it constitutes a core principal within its Cohesion Policy framework.
The relationship between immigration in the European Union and political discontent towards the EU has been a matter of scrutiny for years.Conventional wisdom has largely stated that immigrant levels and immigrant influxes-both from other EU countries and from outside the EU-are negatively related to EU attitudes.This connection is rooted on the idea that citizens often react negatively to newcomers due to perceived cultural threats and/or competition for scarce resources, such as jobs or welfare state support.As it can be seen in Ejrnaes et al. (2023) in this issue, political discontent, often encouraged by the action of political elites, might lead citizens to blame the EU, demanding limits on the free movement of people or even developing strong anti-EU feelings.
We complement these works by arguing that to fully understand this link one needs to consider the potential conditional effect of the immigrant distribution across space.In other words, the spatial segregation of the migrant population may act as a moderating factor that could affect people's attitudes towards the EU.For instance, people living in a territorial unit that holds a large share of migrants (i.e., 70 per cent nationals-30 per cent migrants) can hold different attitudes towards the EU if the unit is highly segregated-both groups live far apart from each other, with little contact-or lowly segregated-both groups live close to one another or, in other words, the probability of encountering any of the two is the same.Thus, this article builds up on recent literature (Schraff & Sczepanski, 2022) that highlights the need to consider both the levels and the spatial distribution of the migrant population in order to explain political attitudes, such as attitudes towards the EU.
Our study delves into the effect of spatial segregation on attitudes towards the EU across four different EU countries.We examine whether spatial segregation changes the relationship between immigration levels and attitudes towards EU unification.As it stands, taking spatial segregation into consideration might offer a different picture than previous works, especially given that the theoretical expectations regarding the effect of spatial clustering of migrants on attitudes are not necessarily clear-cut.Thus, as we explain in the next section, the spatial clustering of migrants may decrease the acceptance of the out-group and eventually decrease support for the EU project.In addition, moderate levels of spatial segregation may not affect public attitudes, such as attitudes towards EU enlargement, as natives may not perceive the potential negative effects triggered by spatial segregation.One could even argue that low to moderate levels of spatial segregation of migrants even has a positive effect on attitudes, as the clustering is not big enough to render potential negative externalities and instead triggers the dynamics posed by the contact theory.Thus, we still know little about whether (and how) the clustering of immigrant groups exacerbates, decreases or mitigates the effect of migration on people's attitudes towards the European Union and the direction of such relationship.Most crucially, the effects can also be different as a function of the type of migrants that are clustered and, as we know from previous works, an individual's social class.
Our empirical research design has different steps.First, we develop in the theory section the different expectations regarding the relationship between the spatial segregation of migrants and attitudes towards the EU.As briefly stated before, the mechanisms outlined by different theories often offer competing and contradictory hypotheses.Second, we make use of a segregation index that provides detailed information on the spatial distribution of migrants (EU and non-EU) in several EU countries.We match this dataset with round 6 of the European Social Survey (ESS), which ultimately allows us to study the micro-macro relationship between spatial segregation and an individual's attitudes towards EU integration in several regions of four EU countries.Second, we distinguish different types of spatial segregation, that is, segregation mainly based on EU migrants and non-EU migrants and on whether migrants come from Western or Eastern European Union countries.Finally, and following the literature on EU support across social groups (G.Evans & Mellon, 2015;Halikiopoulou et al., 2012), we look at the potential moderating effect of segregation by occupational class.
Our results confirm our expectations and show that spatial segregation moderates the effect of immigration levels on EU attitudes, especially when spatial segregation is made up of groups coming from Eastern Europe.People living in areas with large spatially-segregated populations of Eastern EU migrants have a more negative view of European integration.Our models also show that the effect mainly comes from the lower class, who is especially reactive to the spatial segregation of Eastern European migrants.

Theory
Building on studies examining the effect of immigration and political attitudes (Dancygier & Laitin, 2014;Esses, 2021;Hainmueller & Hopkins, 2014), the study of the relationship between immigration and EU attitudes has a long pedigree in political science (de Vreese & Boomgaarden, 2005;S. B. Hobolt & De Vries, 2016).Previous works have documented a relationship between more critical attitudes toward immigrants and political discontent towards the EU (Ejrnaes et al., 2023;S. B. Hobolt & De Vries, 2016;Stockemer et al., 2020;Vasilopoulou & Talving, 2019).For instance, Toshkov and Kortenska (2015) show that immigration from Central and Eastern Europe had a negative effect on support for European integration in the host societies.Other works have also found that immigration is positively related to Euroscepticism (de Vreese & Boomgaarden, 2005;Erisen et al., 2020;Lahav, 2004;Toshkov & Kortenska, 2015;Yeung, 2021) and that people living in regions with more migrants are more likely to vote for parties with Eurosceptic platforms, such as the far right (S. B. Hobolt, 2016;Matthijs, 2020;Vasilopoulou & Talving, 2019).
The relationship between migrants and political attitudes is rooted in several mechanisms.One of the most common approaches is derived from the Social Identity Theory (SIT) (Tajfel & Turner, 1979;Tajfel et al., 1979;Tajfel, 1978).The idea is that individuals form their personal identity through a comparison and categorisation of the in-group-the group they belong to-versus the out-group-the rest.When people are confronted with a different cultural reality they may react negatively, both against the out-group and against the actors that are perceived as facilitators of such situation.This logic has been further developed by the group threat theory (Alesina & La Ferrara, 2002;Biggs & Knauss, 2011;Enos & Gidron, 2018;Vasilopoulou & Talving, 2019), which additionally argues that the perception of threat comes from the fact that migrants tend to have a lower status (social and economic), which may drive down wages, may represent a symbolic and cultural threat to natives and hence negatively affect the status of some natives (objective or perceived).
Notwithstanding the previous theoretical framework, common in the field of EU studies, there is another theory that stands in sharp contrast: the theory of contact.The basic idea of this theory is that stereotypes and negative attitudes towards the out-group diminish if people are in contact with one another.Inter-group contact can lead people to accept their differences and coexist (Pettigrew, 1957;Pettigrew & Cramer, 1959).For instance, Rodon and Franco-Guillén (2014) show that close contact and family contact with immigrants reduce prejudices and, in a study of 22 European cities, Kent (2022) finds that people living in cities with more structural opportunities for intergroup contact view migrants more positively.A metaanalysis conducted in 2006 (Pettigrew & Tropp, 2006) provides empirical evidence in favour of the contact theory.
The tenets of the group threat theory and the contact theory have been previously studied, but less attention has been devoted to whether and how the theoretical expectations underlined by both theories are different when we consider spatial segregation.Spatial segregation lies at the core of the European Commission agenda (van Leeuwen & Vega, 2021) and it has been a popular topic in academia, at least since Key and Heard (1949) suggested that white Southerners in U.S. counties with large African-American populations were more likely to have stronger feelings of white supremacy.As we know, spatial segregation of migrants is a common phenomenon, as they tend to live in regional 'hot spots' (Lichter et al., 2020), that is, they end up living close to people like them.This trend is arguably dependent on contextual factors, such as the policies (or lack thereof) designed by the public institutions.
We argue segregation can have an important moderating effect on the relationship between immigration and attitudes towards the European Union.Only a few studies have indirectly considered this relationship by studying the effect of spatial segregation on support for Eurosceptic political parties.These studies seem to lend support to the group threat theory.For instance, van Wijk et al. (2020) shows that support for the Dutch PVV is high in areas with low shares of minorities and decreases in areas with higher minority shares, up to a tipping point when minorities make up around 25 per cent of the population.Similarly, J.Evans and Ivaldi (2021) concludes that the French Front National in 2017 received a higher support in areas surrounding communities with significantly higher-than-average immigrant populations.However, this is still a relationship that needs further exploration, as works are still uncommon, and some studies point out that the effect of certain variables can be positive or negative as a function of the different levels of spatial segregation (Maxwell, 2012).Last but not least, as other works in this issue show, the characteristics of the regions are important as they can moderate the relationship between variables.For instance, Vasilopoulou and Talving (2023) show the association between growth and EU trust is more pronounced among poor and middle-income regions compared to rich regions (see also Dellmuth, 2023).
Our contention is that spatial segregation is an important factor to understand the relationship between the presence of migrants and attitudes towards EU integration.As already developed by seminal works (Blumer, 1958), prejudice against foreigners may increase with the size and the proximity of the minority group.When the minority group increases in size, the majority group may develop a feeling of threat, either as a result of competing with them for the same resources or from the expectation that the majority's group dominant position is being questioned (Blalock, 1967).The size of the group interacts with how spatially close or spatially distant the in-group is relative to the out-group, which is likely to lead to stereotyping and discrimination (Enos, 2017).All in all, we highlight the importance of studying not only the levels of migrants, but also their spatial distribution.
Different theoretical mechanisms lie at the heart of the relationship between spatial segregation and attitudes towards the EU.Migrants (especially non-EU migrants) are often clustered in space and this may trigger a contrast or a group-threat effect, whereby natives are more likely to be aware of the differences between their situation and that of the other group.When this occurs, natives might blame the EU-at least to a certain extent.Individuals are likely to make this connection through multiple channels.Although immigration is a policy largely at the hands of the EU member states, many voters often blame the EU (S.Hobolt & Tilley, 2014) for not doing enough.Political parties, the media, or even national governments, usually put the EU on the spotlight and criticise the way it handles immigration (S.B. Hobolt & De Vries, 2016), making salient the presence of migrant groups-with its visibility increasing at high levels of spatial segregation.The domestic contestation of EU issues influence national politics (S.B. Hobolt & Rodon, 2020).We can see this dynamic in some recent works.Thus, and examining the Dutch context, Schraff and Sczepanski (2022) find that in places with larger non-Western immigrant groups, citizens have more exclusive national identities, while less segregated contexts show more inclusive identities.Work from the U.S. also points to such moderation effect of spatial segregation in understanding racial attitudes (Enos, 2017).
Notwithstanding the previous logic, there might be another force at play, as suggested by the contact theory.If spatial enclaves are prevented, natives and migrants are more likely to interact and hence attitudes towards the EU may remain unaltered or even improve.If immigration levels are low and even if migrants are spatially scattered, one could argue that this could set in motion the dynamics predicted by the contact theory, that is, people will not be confronted with a large and different group and their attitudes towards the EU will not change.Theoretically, these scenarios could even improve attitudes towards the EU, as citizens might positively consider the way immigration is handled or the idea that diversity in the EU brings prosperity.One could argue that this holds even for moderate levels of spatial segregation, as natives may not perceive the potential negative effects triggered by spatial segregation (group differentiation and group competition) and therefore the conditions to blame the EU are not yet present.
To sum up, following the group threat theory, large migrant populations may be associated with negative attitudes towards the EU, while contact theory would posit the opposite: large migrant populations, unless they are highly spatially segregated, may be associated with more positive attitudes towards the EU.These contradictory expectations have often been referred as to the paradox of space (Enos, 2017).By departing from the idea that spatial segregation acts as a crucial moderator in order to understand the relationship between the size of the group and political discontent towards the EU, we outline two hypotheses: H1a: Following the group threat theory, there is a negative relationship between the migrant population and attitudes towards the EU which should increase when spatial segregation increases.
H1b: Following the contact theory, there is a positive relationship between the migrant population and attitudes towards the EU which should decrease when spatial segregation increases.
The previous theoretical expectations assume that the moderating effect of spatial segregation is constant across different social groups.This is largely unrealistic as not all group of natives have the same probability to be spatially close to the migrant population.As the literature on the relationship between immigration and EU attitudes has shown, it is the lower class, compared to other social groups, who often reacts negatively to the presence of migrants.Thus, migrants tend to be poorer, less educated and have lower access to formal and informal local networks, which means they have a higher likelihood of clustering on space.Spatial segregation also tends to be high among native groups of lower status.Previous works have shown that the lower class is more likely to be affected by the influx of migrants, due to economic factors, as they often compete against natives with similar characteristics for jobs or other scarce resources; cultural factors, as migrants tend to come from different cultures or speak a different language; or because they are often targeted by populist and far-right politicians (Alesina & La Ferrara, 2002;Biggs & Knauss, 2011;Enos & Gidron, 2018;Schraff & Pontusson, 2023), which often use Eurosceptic arguments to garner political support (S. B. Hobolt & De Vries, 2016;Mayda, 2006;Vasilopoulou & Talving, 2019).Given that the less affluent of the society, natives and migrants, are likely to share very similar spaces and be close to one another (Massey, 1979;Piekut, 2021), our expectation is that this proximity may build up resentment and political discontent in which the EU might be perceived as one of the actors to blame.Therefore, and despite the average effect of segregation has different theoretical expectation-as outline before, we expect the moderating effect of spatial segregation on EU attitudes to be negative among the working class, compared to other social groups.
H2: The moderating effect of segregation increases among the working class, compared to other groups.

Data and modelling
We explore the relationships between migrant populations, segregation and support for further European integration using a dataset built from two different sources.The first dataset takes advantage of the Data for Integration (D4I) project.The European Commission's D4I initiative processed 2011 Census data to produce estimates of native and migrant population (disaggregated by origin) at a 100 × 100 metre resolution across for eight EU member states.This index offers high quality data and improves on previous indicators of segregation, which are often built at higher levels of spatial aggregation.After the intersection between the segregation data and survey data at the NUTS-2 level (see below) we are left with four countries.Hence, for the purposes of this paper, we use data for France, Italy, the Netherlands, and Spain at the L1 level.These four cases differ in their overall levels of spatial segregation (the lowest is in France and the highest in Spain), the number of migrants (the minimum is in Spain and the maximum in the Netherlands) and their levels of Euroscepticism (high in the Netherlands and relatively low in the rest).
Despite some global dynamics-migrants tend to be poorer than natives or they tend to self-select in certain urban or sub-urban areas, the policies and the dynamics regarding segregation also differ substantially across these countries (Alessandrini et al., 2017).As reported in previous studies, northwestern Europe (the Netherlands fits the pattern) has lower levels of residential segregation compared to the Southern European countries (Benassi, Bonifazi, et al., 2020;Marcińczak et al., 2023).In the latter the clustering of migrants particularly affects the most disadvantage cities (Benassi, Iglesias-Pascual, et al., 2020).There are also important differences across France, Italy and Spain regarding segregation patterns.France has received more migrants during a longer period of time and the welfare regime and the housing system, together with cultural factors, have triggered increased clustering in some urban and sub-urban areas (Arbaci, 2019;Huttman et al., 1991;Lelévrier, 2023;Oberti & Savina, 2019).However, and despite the differences in public policies and patterns of spatial segregation, the social and political backlash has also recently occurred in both Spain and Italy (Martin-Legendre et al., 2021;Phillips, 2013), where anti-immigrant sentiments have risen, often fuelled by the far-right or other political actors.Overall, and although political, cultural and social differences exist across our cases of study, these offer substantial variation across regions and a constitute a representative look at Western Europe on the different levels of spatial segregation among migrants.
The D4I data includes population estimates for each grid cell by country of birth.We aggregate this data into four population groups: native-born residents, migrants from western EU countries, migrants from eastern EU countries, and migrants from other countries.The data identifies where migrants reside and hence we do not know whether they frequently interact (or not) with natives.However, as developed in the theoretical part, a more frequent interaction between migrants and natives does not necessarily imply a positive impact on EU attitudes, as different theoretical expectations offer opposite relationships.
The data identifies where migrants reside and hence we do not know whether they frequently interact (or not) with natives.Following Jeannet (2018), we use a political logic, instead of a purely geographic one, to separate EU countries into 'west' and 'east'.The EU West group includes countries that joined the EU in 1995 or earlier, while the EU East group includes countries that joined in the 2004 expansion or later.Our analysis utilises the ratios of these four population groups along with segregation indexes for each of the migrant groups.We calculate Theil indexes-a common measure of segregation-at the NUTS-2 level using the segregation R package (Elbers, 2023) to determine the extent to which the migrant groups are segregated from the rest of the population.The Theil Index ranges from 0 to 1. 0 means that each geographic unit-in our case, grid cell-has the same composition as the larger unit-NUTS-2 region-and 1 means that each geographic unit contains only one group-natives or migrants.Here, the geographic units are 100x100m grid cells and the Theil Index is of the level of segregation within each NUTS-2 region.The spatial distribution of the Theil Index across the different regions is shown in Figure 1.As it can be visually seen, the levels of segregation differ both within and across countries.This indicates that there are not only different countrylevel dynamics regarding spatial segregation of migrants, but also different levels of 'ghettoisation' are taking place in some regions within a country.
We combined the region-level population and segregation variables with individual-level data from round 6 (2012) of the European Social Survey.The following question serves as the dependent variable, which captures the dimension of general or diffuse support towards the EU: 'Now thinking about the European Union, some say European unification should go further.Others say it has already gone too far.Using this card, what number on the scale best describes your position?' with 'Unification has already gone too far' corresponding to 0 and 'Unification should go further' corresponding to 10.Although diffuse support captures only a specific set of attitudes towards the EU, it remains an important dimension to analyse, as it considers the legitimacy towards the political entity (Ares et al., 2017;Easton, 1965).We also include gender, age, years of education, country of birth and occupational class as individual-level controls.The country of birth variable is aggregated in the same manner as the region-level variables, while the class variable is constructed using the Oesch 5-class schema (Oesch, 2006a(Oesch, , 2006b)).After the two levels are merged, the resultant dataset has a total number of 6,110 individuals living in 69 regions in four countries.
Our analysis employs multi-level linear models estimated using the lme4 R package. 1 Each model includes random intercepts for regions nested in countries. 2The number of observations in each region range from just 5 in Friuli-Venezia Giulia, Italy, to 338 in Andalusia, Spain, with a mean of 88.6 and a median of 62, and the observations are weighted using the ESS's design and population weights.The number of observations at both levels is stable across models.In addition to the region-level variables, which are added in stages, each model estimates beta coefficients for individual-level age, age squared, years of education, years of education squared, gender, class and country of birth.

Results
Our first models explore support for EU integration as a function of several covariates, including our key variables of interest, that is, spatial segregation and the relative size of each migration population.The null model finds an intraclass-correlation coefficient (ICC) of 0.034 for regions nested in countries, with an ICC of 0.016 for a model with only country as a level and an ICC of 0.029 for a model with only region.This suggests that region-level factors are important sources of variation in our dependent variable and that the multilevel structure of regions nested in countries is suitable for the models with which we test our hypotheses.
Our primary set of models (Models 1 to 3) adds the region-level variables sequentially, which would allow us to identify the value of including segregation data.We plot the coefficients of interest in Figure 2, while the complete model tables can be found in Appendix B. We find that residents of regions with larger relative populations of migrants born in EU East countries hold more positive attitudes toward European integration.Including the segregation variables has little effect in Model 2. However, by including interactions in Model 3 we see that the positive relationship between EU East population and attitudes is weakened in regions where that population is highly segregated-lending support for Hypothesis 1b, based on contact theory.
For populations of migrants from EU West countries or other countries, there is no evident relationship to attitudes toward European integration.Overall, our models show that when we study the effect of the absolute number of migrants without taking into consideration their spatial distribution, results are likely to be misleading.People in places with high levels of EU Eastern European migrants have, on average, a more positive view of the EU.However, when we consider the fact that these migrants are generally segregated across space, we observe that the positive relationship with EU attitudes diminishes.
Given that the interaction in M3 is based on two variables that range from 0 to 1 and hence is difficult to interpret, Figure 3 plots the conditional effects of segregation of migrants from Eastern EU countries on the relationship between the relative population size of that group and predicted attitudes.Although the confidence intervals generally overlap, the figure suggests that in regions with very small migrant populations, segregation may result in more positive attitudes towards the EU.As migrant population size increases, less segregation is associated with greater improvement in EU attitudes.Yet, in highly segregated regions, attitudes decline as migrant population rises.This first set of findings illustrate the idea that the high presence of Eastern European migrants and high levels spatial segregation are associated with negative EU attitudes.Yet, when the size of the group is smaller, segregation is not connected with such negative effects.
As we discussed before and previous works have shown, segregation does not affect everyone equally (Chetty et al., 2022;Wilson, 1959).Thus, in line with the theoretical discussion, we next explore whether the moderating effect of segregation affects some social groups more than others.Figure 4 estimates additional models and explores such heterogeneous effects.
Here, we subset the data by occupational class.The service class model includes only respondents in classes 1 and 2, while the skilled/unskilled workers model includes only those in classes 4 and 5.The results show that the relationships identified above are strongly driven by the working class, lending support to Hypothesis 2. Additionally, the service class model suggests that this group may have less positive attitudes as the population of Western EU migrants increases, though here the finding falls just short of statistical significance.Yet, this result might suggest that the upper class seems to negatively react to Western European migrants, potentially as a way of protecting their status quo position or as a reaction of the rhetoric of some political parties.
As seen in Appendix C, we conduct several robustness checks.First, we repeat models 1-3 but omit the population and segregation controls for EU West and other migrants.Second, we add region-level GDP per capita and other regional-level controls to confirm the relationship when additional confounders are added (individual-level socioeconomic controls are included in all models). 3Finally, we also include the standard deviation of the segregation index for all of the communes within the NUTS-2 regions and the self-reported urban or ruralness of the respondents domicile to test whether within-region variation in segregation levels influences the results.In all of these robustness checks, we continue to find evidence that confirms Hypothesis 1b.

Conclusions
This article aimed at making a contribution to the literature on spatial segregation and support for the EU project.Although spatial segregation has been at the core of many classical and contemporary studies, the evidence on the relationship between spatial segregation and support for EU integration is still limited.We complement previous studies by examining how the unequal spatial distribution of migrants conditions the relationship between immigration and support for further integration into the European Union.As explained in the article, the relationship between these components is not theoretically apparent.The paradox of space illustrates that, while proximity to the out-group might be beneficial and hence natives should not become more negative towards the EU project (contact theory), the same proximity can lead natives to be aware of the differences and experience a change in attitudes and behaviour (group based theory).
We find a high presence of Eastern European migrants and high levels of spatial segregation to be correlated with negative attitudes toward the EU.In addition, this process is particularly strong among the working class.For migrants from non-EU countries we find null effects, perhaps due to the less direct role of EU policies in promoting their arrival compared to the freedom of movement that EU-born migrants enjoy.Our results are in line with Jeannet (2018), who shows that growing populations of Central and Eastern European migrants have negatively contributed to Western Europeans' perception of immigration as an economic threat, and with Jeannet (2020), who shows that immigrants represent an important factor in the growing distrust amongst a relatively small sub-group of citizens-the working class.
Our analysis, however, also suggests that the two theoretical arguments explained in the article remain plausible, in line with the paradox of space and what Enos (2017) finds.In other words, our results point to the idea that a large presence of migrants in non-segregated regions does not seem to fuel political discontent against the EU-instead, it is positive.Yet, in places where segregation coming from EU migrants is high-and hence places where the dynamics outlined by the group threat theory are present-support for EU integration tends to be lower.Taken together, our results point to the need of considering both the and the spatial distribution of migrants in order to understand people's attitudes towards the European Union.They also raise the need to reconsider the relationship between the group threat theory and the contact theory.Despite suggestions that the theories are in opposition, our results point to the idea that they are both valid, but they operate under different conditions, with segregation being an important moderating component of the story.
Our analyses have some limitations.Besides applying a causal design and replicating them as more data becomes available-and at lower levels of geographic aggregation-, the countries considered here might not be similar to other cases or periods of time.One might also explore potential differential effects of segregation on EU attitudes both across countries and over timecomparing them to more recent periods, and in places where the far-right has gained electoral ground.In addition, immigrants do not decide where to live or interact following a random process and hence more refined causal designs are needed in order to gauge the causal relationship between segregation and attitudes or behaviour.Last but not least, our outcome only captures one dimension of EU attitudes and future studies can also look at other dimensions, such as support for specific EU policies or trust towards the EU.Yet, and moving forward, our findings suggest the need to continue looking at the relationship between other political attitudes and segregation and on finding convenient public policies to mitigate the potential negative effect segregation often carries.

Notes
1.If we use Bayesian multi-level models, the findings are consistent.2. Including fixed effects for countries, in lieu of random intercepts, had no substantive effect on the findings.3.If we add other regional-level control variables, such as population density and unemployment (see Table C.2), or if we run the models with country FEs, results remain robust.

Figure 1 .
Figure 1.Map of NUTS-2 regions of Spain, France, Italy and the Netherlands by Theil Segregation Index for migrants born in western EU, eastern EU and other countries, respectively.

Figure 2 .
Figure 2. Coefficient estimates of NUTS2-level variables for linear multilevel models of attitudes toward European integration.

Figure 3 .
Figure 3.Effect of Eastern-EU-born migrant population on attitudes toward European integration conditional on the level of segregation of that group.

Figure 4 .
Figure 4. Coefficient estimates of NUTS2-level variables for linear multilevel models of attitudes toward European integration with data subset by occupational class.