The more the merrier? A study on how further enlargement of the European Union and increased internal migration affects EU attitudes Tove Linder and Lovisa von Schéele Supervisor: Yun Xiao Master’s thesis in Economics, 30 hec Spring 2023 Graduate School, School of Business, Economics and Law University of Gothenburg, Sweden Abstract The free movement of workers within the European Union has strong support among its citizens. At the same time, the free movement of labour has been put forward as one of the most politically controversial features of the European common market. This paper examines how a set of EU attitudes are affected by the inclusion of Bulgaria and Romania in the union through changed intra-EU migration. The analysis is conducted in 25 European countries using the responses from the Eurobarometer surveys during the period 2004 to 2019. The analysis is conducted using an IV approach utilizing the distance from Bulgaria and Romania interacted with opening of the border as an instrument. The results indicate that increased intra-EU migration is negatively associated with positive EU attitudes. The results also indicate that increased intra-EU migration is negatively associated with the willingness for further enlargement of the EU. Further, the analysis displays a heterogeneous effect between the founding members of the EU and new member countries where the EU image of founding member countries is positively affected by increased intra-EU migration while the effect is negative for new members of the EU. 1 Acknowledgement We would like to express our gratitude to our supervisor Yun Xiao who has supported us throughout this thesis process and provided excellent advice. We would also like to thank our family and friends for the moral support they have provided during this process. 2 Table of Contents 1. Introduction 4 2. Literature review & Conceptual framework 6 2.1 Literature review 6 2.1.1 Removing mobility barriers 6 2.1.2 Immigration effects on attitudes 8 2.2 Conceptual framework 10 2.2.1 The labour market competition model 10 2.2.2. The fiscal burden approach 11 2.2.3 Compositional concerns of immigration 12 2.2.4 Applying the framework on EU attitudes 13 3. Institutional settings 13 4. Data and Methodology 15 4.1 Research questions 15 4.2 Data 15 4.2.1 Dependent variables 17 4.2.2 Independent variable 19 4.2.3 Instrumental variables 20 3.2.4 Interaction variables 20 4.2.5 Control variables 21 4.3 Empirical models and strategy 22 4.3.1. OLS with fixed effects 22 4.3.2. Instrumental Variable regression 23 4.3.2.1. Choice of instrument 23 4.3.3 2SLS specifications 25 4.3.4 Relevance and validity of instrument 27 5. Results 29 5.1 Simple OLS model 29 5.2 IV regressions 32 5.2.1 First stage 32 5.2.2 Second stage 33 5.3 The effect on EU attitudes across different countries 36 5.4 Robustness- and sensitivity checks 40 6. Discussion & analysis 41 6.1 Internal validity 42 6.2 External validity 44 7. Conclusion 45 References 46 Appendix 50 I. Description of variables and data sources 50 II. Regression tables & figures 52 3 1. Introduction The free movement of workers is one of the four pillars of the European single market. While free movement of workers has strong support among EU citizens (European Commission, 2017) it has also been put forward as one of the most politically contentious elements of the European common market (Dorn & Zweimüller, 2021). Today, more than 460 million citizens in European countries are allowed to work anywhere they want within the union without having to apply for a work permit. While there are economic gains from free movement of labour between countries, it also tends to be a controversial topic (Dustmann & Preston, 2019). The inability of the United Kingdom to control the inflow of labour immigration from Eastern and Central Europe is argued to be a crucial argument for leaving the European common market in the Brexit referendum (Dorn & Zweimüller, 2021). Increased skepticism towards immigration is also closely intertwined with the rise of right-wing European populist parties which is also strongly correlated with a general mistrust in the European Union (Schmuck & Matthes, 2017). In this paper, we study how public opinion towards the European Union is affected by increased internal migration within the union. To do so, we utilize the enlargement in 2007 when Bulgaria and Romania joined the EU to analyze how changes in intra-EU migration composition affects EU public opinion. Using a Two-Stage Least Squares approach, we conduct a cross-country analysis for the period 2004 to 2019. Drawing inspiration from Angrist and Kugler (2003), we exploit the distance from Bulgaria and Romania interacted with a time dummy variable as an instrument to predict migration. The time dummy indicates when an EU country removed its mobility barriers toward the recently accepted countries. We construct a unique dataset using migration data from Eurostat, the Eurobarometer survey to construct a set of EU attitudes and CEPII1 statistics on bilateral distances between countries. The OLS results propose a positive correlation between intra-EU migration and EU attitudes, while the IV results, which are utilized to combat the endogeneity problems in the OLS, display contrary results. The results from the IV regressions show that changes in intra- EU migration due to the inclusion of more countries in the union are negatively associated with a positive image and trust in the EU, as well as negatively associated with the preference for further enlargement of the EU. Further, the analysis displays a heterogeneous effect between the founding members of the EU and new member countries where the EU image of founding 1 French center for research and expertise on the world economy 4 member countries is positively affected by increased intra-EU migration while the effect is negative for new members of the EU. Previous studies on labour market integration have shown that the removal of transnational mobility barriers in many cases causes an influx of migrants. Card (1990) demonstrates how the Mariel boatlift in Cuba in 1980, which did permit Cubans to freely leave Cuba if they wished to, led to a labour supply shock in Miami where the Cuban migrants increased the Miami labor force by seven percent. Dustmann et al. (2017) showed that after the fall of the Berlin wall when a policy allowed Czech workers to seek employment in German border municipalities, this increased the local labour markets by ten percent. In Europe, the enlargements of the European labour market in 2004 and 2007 increased the inflow of migrant workers from the recently accepted countries (EU8 and EU22) to old member states (EU153) considerably (Baas et al., 2009; Barrell et al., 2010; Holland et al., 2011). For example, within two years after Bulgaria and Romania joined the EU, 4.1 percent of the population had migrated to EU15 countries (Holland et al., 2011). Free movement has however been argued to create tensions in high income countries between immigration policy and welfare services (Meltzer et al., 2018). Further, individual economic interests are often attributed to explain attitudes on immigration policy (Scheve & Slaughter, 2001). However, the economic adverse effects from immigration, with higher unemployment, lower wages and an increased fiscal burden in the destination country have been argued to be limited (Card, 1990; Barrell et al., 2010; Hainmueller & Hiscox, 2010). Hence, Card et al. (2012) argues that compositional concerns, that immigration changes the demographic composition, are more critical for an individual's perception of immigration rather than economic concerns. The impact of immigration on public EU support is also argued to be associated with whether immigration is seen as a cultural threat or a threat to national identity (Ringlerova, 2022). On the other hand, for candidate countries to the EU, the support towards the EU is driven by expectations of economic benefits and as a tool for further democratization (Peshkopia, 2020). While many studies have examined how immigration affects political attitudes and immigration policy concerns, the question of how attitudes towards the European Union are affected by changes in intra-EU migration is a relatively unresearched area. Thus, this paper 2 EU8 refers to the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Slovakia and Slovenia, while EU2 refers to Bulgaria and Romania. 3 EU15 refers to Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, the Netherlands, Portugal, Spain, Sweden and the United Kingdom 5 contributes to the literature by examining if increased intra-EU migration in the European common market affects the public opinion about the institution responsible for the policy decisions. This result is interesting from a policy perspective as we have eight candidate countries to join the European Union, which most likely would increase intra-EU migration further. The enlargement of the EU2-countries is especially interesting to examine given their long path to become EU members. When the rest of the eastern European countries joined the EU in 2004, Bulgaria and Romania lagged behind because of issues related to their institutions, socioeconomic conditions and domestic political choices (Noutcheva & Bechev, 2008). Although Bulgaria and Romania eventually succeed to qualify for EU membership, their struggles to meet the conditions for joining the union could be compared to the issues the Western Balkan countries face, where they need to reform both political and judicial institutions before qualifying for membership in the union (Strelkov, 2016). The rest of the paper is organized as follows. Section 2 provides a literature review and a conceptual framework, Section 3 provides a brief explanation of the institutional setting, while the data and methodology are presented in Section 4. Section 5 provides the results, Section 6 presents a discussion and analysis and Section 7 provides a summary of the results. 2. Literature review & Conceptual framework This paper aims at examining whether the inclusion of more countries into the EU affects the public opinion of the EU through increased intra-EU migration. To enhance the understanding of this topic, this section provides a literature review and a conceptual framework. The literature review covers previous studies on the effects of migration and effects of mobility barriers on migration while also reviewing studies on how migration affects attitudes. The conceptual framework dive into the potential links between migration and migration attitudes. 2.1 Literature review 2.1.1 Removing mobility barriers The effects of migration and global labour mobility is an extensively researched topic. In a paper covering the global gains of labour mobility, free movement, and open borders, Dustmann and Preston (2019) claim that the most obvious economic benefit of free movement of labour is that it allows workers to move to a place where their skills are the most productive and valued. However, Dustmann and Preston (2019) also put forward that while labour mobility 6 can increase the world output, it can also cause adverse effects for workers at the receiving location. That removing mobility barriers between countries has effects on the movement of people is a well-known phenomenon. Several papers have used this phenomenon when examining the effect of immigration on various outcomes. As earlier mentioned, Card (1990) illustrates how the local labour market in Miami experienced a labour supply shock by Cuban immigration due to the removal of restrictions for Cubans to leave their country. Examining the effect of this shock, Card (1990) found that the inflow of immigration had virtually no effect on wages or unemployment for both natives and immigrants. In contrast to Card’s results, Dustmann et al. (2017) found that when Czech citizens were free to move across the border to Germany after the fall of the Berlin wall, this had adverse effects on local wages and local employment for Germans. The adverse wage effects were most noted among the younger locals. These results are also supported by Moritz (2011) who found the same effect of free movement of Czech citizens on wages for low-skilled young German men. Considering the common European labour market, Dorn and Zweimüller (2021) highlights that the European integration process has remarkably reduced mobility frictions between the union’s members by giving EU-citizens the same legal access to a country’s labor market as the domestic citizens have. Including more countries in the European Union and granting them access to the European common market have subsequently led to increased migration within the union. When examining the enlargement of the EU in 2004, the stock of migrants from new member states (EU8) to old member states (EU15) increased from almost 893,000 in 2003 to more than 1.91 million by the end of 2007 (Kahanec et al., 2009). Holland et al. (2011) further argue that the differences in when the mobility barriers were removed among the member countries, that is, when the countries opened their borders, might have had a permanent effect on migration patterns in Europe. The argument is based on the network effects of migration, i.e., that people tend to move to places where they already have a network. Furthermore, when analyzing effects of the enlargement in 2004, Baas et al. (2009) found that migration associated with the European labour market integration has had substantial gains. From 2004 to 2007, the aggregate GDP in the European integrated area increased around 0.2 percent, which translates to approximately 24 billion euros (Baas et al., 2009). Their analysis also found small labour market effects, where the wages and unemployment in the receiving EU15 countries decreased by about 0.1 and 0.3 percent respectively in the short run. However, in the long run the adverse effects on wages and unemployment in the old member states proved to be nonexistent (Baas et al., 2009). 7 In regard to the enlargement in 2007 when Bulgaria and Romania joined the EU, which is the enlargement we intend to examine in this paper, the literature is more scarce. Many studies group together the two enlargements in 2004 and 2007 when examining the effects of EU-enlargement, both regarding the effect enlargement has on intra-EU migration and economic effects of increased integration. Roman (2019) raised this issue in a paper comparing the effects of the two different enlargements on the UK labour market, since she found that the effects of the EU2 enlargement had been largely debated but relatively unresearched. In her analysis, Roman (2019) found that EU2 migrants in many cases display different characteristics regarding work, self-employment and overall labour market outcome compared to other European migrants but are not significantly different when it comes to welfare take-up. In this regard, our paper will further contribute to the limited literature on effects of the EU2 enlargement in 2007. 2.1.2 Immigration effects on attitudes However, the economic effect of immigration is far from the only field of research when it comes to aspects of immigration. Both within economics and other research fields, many studies examine how different sets of attitudes are affected by immigration. In this review, we will mostly cover how immigration affects opinions on immigration policy and how immigration relates to public opinion of the EU. Research has suggested that people tend to be more positive toward immigration from ethnically or culturally close groups rather than remote groups (Brader et al., 2008; Hainmueller & Hopkins, 2014). This notion can be linked to social identity theory which suggests that cultural differences between migrants and natives creates a narrative of “us” and “them”, which in extension will affect the formation of attitudes on immigration. However, the origin of the “us” and “them” narrative has been debated. Sniderman et al. (2000) argues that the formation of an in- and out-group is enough for the creation of an opposition towards each other. In the paper, Sniderman et al. (2000) conducts an experiment based in Italy where respondents are asked about social problems caused by either eastern European immigrants or African immigrants. The results found that, to a large extent, the respondents could not distinguish the groups apart. Misperceptions about immigration are also found to be widespread. Blinder (2015) identified that when the British public considers immigration, they tend to imagine asylum seekers and permanent arrivals but often fail to consider international students or spouses. Hence, Blinder (2015) proposes that these misperceptions of immigration should be taken into 8 consideration when examining immigration attitudes and forming policy based on public preferences. Further, Gorodzeisky and Semyonov (2020) affirmed that inflated misperceptions are found to contribute to opposition towards immigration. In the context of the EU2 enlargement, Roman (2019) points out that the potential high inflow of EU2 migrants became a constant theme in pre-Brexit debates and Roman argues that it was the perception, rather than the actual inflow, of EU2 migrants that was influencing the Brexit decision. A general mistrust in the European Union has accordingly been argued to be closely intertwined with increased skepticism towards immigration as well as right-wing populism (Schmuck & Matthes, 2017). Increased euroscepticism has also been linked to negative media framing, since it has been noted that when the media puts forward risks associated with integration of the European Union, individuals become less likely to support further integration of the union (Schuck & Vreese, 2006). When framing the EU as either a cultural threat or as problematic in a democratic point of view, individuals express less support for the EU (Abbarno & Zapryanova, 2013; Schuck & Vreese, 2006; Usherwood & Startin, 2013; Lecheler & Vreese, 2010). Consequently, immigration has been argued to have a negative effect on EU support. To analyze the causal mechanism for this effect, Ringlerova (2022) uses an experimental public online survey on immigration policy in the EU where the different policy options are framed in accordance with different theories about immigration effects on attitudes. For some of the countries in the study, Ringlerova (2022) finds that negative information about immigration policy has a moderately strong negative effect on EU support, while for some countries she found no systematic effect. Ringlerova (2022) experiment does however focus on immigration from outside of the EU. In an empirical analysis of the effect of immigration from Central and Eastern European countries to western European countries after the fall of the Berlin wall, Toshkov and Kortenska (2015) found that immigration had a negative effect on the support of European integration in the host countries. Yeung (2021), on the other hand, found no significant effect of either EU migration and non-EU migration on EU support during the period 2009–2017 in a study using data from Eurostat and the Eurobarometer. When empirically studying the effects of immigration, the main issue is that migration decisions are based on several factors, making it difficult to compare regions that experience more immigration with regions that experience less immigration since migration patterns are not random. Thus, some form of external variation is needed to combat endogeneity problems in empirical studies on immigration effects. Many studies in this research field uses survey designs and experiments to find causal effects (Sniderman et al., 2000; Brader et al., 2008; 9 Blinder, 2015; Ringlerova, 2022), while other papers have tried to find natural experiments to examine causal relationships (Card, 1990; Dustmann et al., 2017; Moritz, 2011; Roman, 2019). However, previous studies examining immigration effects on EU support use different OLS approaches to examine how actual immigration correlates with public opinion about the EU (Toshkov & Kortenska, 2015; Yeung, 2021). Hence, these studies are likely to suffer from endogeneity problems since migration flows rarely are randomly distributed. Hence, this paper will contribute to the research field on how actual immigration affects EU attitudes by trying to solve the problem with endogeneity by using an instrumental variable approach. 2.2 Conceptual framework As stated earlier, this paper aims at examining whether the inclusion of more countries into the EU affects the public opinion of the EU through changed intra-EU migration. In the purpose of understanding the chosen model specification, this section creates a framework for the empirical analysis and dives into the potential links between migration and migration attitudes as well as EU attitudes. The inclusion of additional countries in the EU is assumed to affect the perception of the EU as follows (Figure 1): Figure 1. Flow chart of the potential link between EU enlargement and changed EU attitudes. Inclusion of additional Increased intra-EU Negative perception Negative perception co untries in the EU migration towards immigration towards the EU We have identified three channels through which immigration is assumed to affect immigration attitudes from the literature on migration and attitudes. The different approaches, or channels, can be divided into (1) the labour market competition model, (2) the fiscal burden, and (3) the compositional concerns of immigration. The first two channels cover the economic aspects of migration and attitudes while the third channel covers the psychological aspects. We argue that it is possible to connect migration and EU-attitudes based on previous literature that has shown that perception of the EU and migration attitudes are closely intertwined (Schmuck & Matthes, 2017; Ringlerova, 2022). 2.2.1 The labour market competition model One of the most important contributions to the linkage between immigration and immigration attitudes is provided by Scheve and Slaughter (2001) who links the effects of immigration on attitudes to the impact immigration has on the labour market. The Factor 10 Proportions (FP) model assumes a national labour market and a single aggregate output market, as well as perfect substitutability between natives and immigrants. According to the model, an increased supply of workers in a sector lowers the wages in the same sector. Thus, Scheve and Slaughter (2001) predicts that low-skilled workers would prefer reduced low-skilled immigration4. This prediction was confirmed by Mayda (2006) who found that natives’ skills are correlated with their support for immigration. She also found that the correlation is stronger in countries with bigger differences between natives' skills and immigrants’ skills. Further, Dancygier and Donnelly (2013) shows that natives employed in growing sectors are less opposed to immigration, suggesting that individuals are less concerned about immigration if it is not perceived to increase the competition in their sector of the labour market. Even though empirical findings suggest that wage effects from immigration are small (Card, 1990; Barrell et al., 2010; Hainmueller & Hiscox, 2010; Hainmueller et al., 2015), it is fair to argue that immigration effects on attitudes can be linked to the effect migration has on labour market competition. In the context of our study, an increased supply of workers, either low- or high-skilled, to a national labour market would be assumed to affect the native worker, which in extension could affect the native’s support for immigration, free movement, open borders, and a common labour market. 2.2.2. The fiscal burden approach The fiscal burden theory suggests that immigration attitudes might not only be correlated to an individual's own situation but also the perception of the national economy. This could explain why high skilled immigrants are preferred over low skilled and why workers are preferred over refugees (Aalberg et al., 2012; Blinder, 2015; Hainmueller & Hiscox, 2010; Sides & Citrin, 2007; Valentino et al., 2019). To test this theory, Hanson et al. (2007) elaborates on the FP model and includes public finance into the model, allowing immigration to affect the net salary of workers. While low-skilled immigration increases the fiscal burden, high-skilled immigration lowers the fiscal burden, thus affecting the native’s net salary differently. Using U.S. data, Hanson et al. (2007) find that high-income natives are less likely to be in favor of immigration in states that already have high fiscal pressure. Hainmueller and Hiscox (2010) on the other hand oppose this hypothesis, arguing that U.S. states with a substantial growth of immigrants during the period 1990 to 2004 had lower increases in income taxes than states 4 The FP predicts that the opposite effect would be true about preferences of high-skilled immigration inflows. 11 without such inflow of immigrants. Furthermore, Brader et al. (2008) show that the public's reaction to the fiscal burden of immigrants varies depending upon the characteristics of the immigrants. Schneider (2007) proposes that a fear of different values and culture play a larger role in the creation of immigration attitudes rather than economic competition. The fiscal burden of increased low-skilled immigrants related to including Bulgaria and Romania into the European common market can thus be disputed. First of all, Roman (2019) found that migrants from EU2 countries do not display more welfare take-up than other EU migrants. Secondly, since medium-skill level workers are overrepresented among mobile EU workers (European Union, 2011) the fiscal burden might not be substantially affected from increased migration due to further enlargement. 2.2.3 Compositional concerns of immigration While empirical evidence to some degree supports the theories that immigration effects on attitudes are channeled through the economic impacts of immigration, researchers have pointed out that this might not be the sole explanation of how immigration affects attitudes. When Hainmueller and Hiscox (2007) tested the FP model in a European setting, they found that high-skilled natives, opposed to the FP model, were more positive towards all kinds of immigration than low-skilled natives. Hainmueller and Hiscox (2010) and Hainmueller et al. (2015) also examined this in a U.S. setting, and found no evidence that natives are more skeptical towards immigration of the same skill set as themselves. Instead, they found that natives with higher education were less likely to oppose immigration regardless of the skill set of the immigrants, indicating that the correlation is rooted in cultural differences in attitudes towards immigration rather than a concern about lower wages. Thus, while labour market competition theories and fiscal concerns consider attitudes to be shaped by economic self-interest, compositional concerns put emphasis on the effect immigration has on culture, local society, and the country as a whole (Hainmueller & Hopkins, 2014). Since most studies have found modest economic impacts of immigration, Card et al. (2012) challenged the labour market competition theory in a paper on public opposition to increased immigration. Card et al. (2012) put forward that immigration, rather than affecting wages and taxes, affects the composition of the population regarding nationalities, languages, and cultures. The effect of compositional concerns on immigration attitudes is drawn from economic theories of neighborhood choice, which suggests that externalities arise since people value the composition of their local society (Card et al. 2012). For example, studies have shown that when neighborhoods become more heterogeneous, white natives tend to move to other 12 neighborhoods (Card et al., 2008; Bohlmark & Willen, 2020). Card et al. (2012) thus argues that such preferences play an important role in mediating opinions on immigration. Correspondingly, their study finds that compositional concerns are two to five times more important than economic concerns in shaping immigration opposition among individuals. 2.2.4 Applying the framework on EU attitudes Compositional concerns of immigration can also be related to the impact of immigration on EU attitudes, where one possible explanation of public support of the EU is linked to culture and identity (Ringlerova, 2022). Ringlerova (2022) put forward that an individual’s support of the EU can be associated with whether an individual perceives immigration, and in extension membership in the EU, as a cultural threat or threat to their national identity. Many studies that analyze the effect of immigration on attitudes have put forward differences in cultural aspects (e.g., ethnicity, religion, culture) as explanations for increased skepticism towards immigration (e.g., Brader et al., 2008; Hainmueller & Hopkins, 2014). Immigration from ethnically or culturally close countries are however often considered to not impose compositional or cultural threats (Brader et al., 2008; Card et al., 2012; Hainmueller & Hopkins, 2014; Quillian, 1995). What we intend to capture through this paper is whether including more countries in the EU invoke a “we” sentiment, a European sense of belonging, or if there might be a risk of an “us” and “them” sentiment between member countries due to cultural differences. The inclusion of additional countries in the European Union would affect the composition of the intra-EU migration, which in extension could affect public opinion of the EU. 3. Institutional settings This paper aims to examine the consequences of including additional countries in the European common market on EU attitudes, from the perspective that inclusion into the union gives citizens in newly accepted countries access to free movement within the union. Thus, we will not use the inclusion date in 2007 when Bulgaria and Romania joined the union, but rather focus on the introduction to free movement according to the transitional arrangements5. The transitional arrangements allow countries to restrict the right of free movement of workers temporarily, hence providing a gradual inclusion of new member countries into the EU free movement zone. This temporary period spans over seven years, with three phases. For the EU2 countries these periods were 2007, 2009, 2012 and 2014 (see Table 1 below). 5 This does not refer to the Schengen agreements which focus on the removal of border controls. 13 Table 1. The year countries removed their mobility barriers to Romania and Bulgaria Countries 2007 2009 2012 2014 Austria Yes Belgium Yes Cyprus Yes Czechia Yes Denmark Yes Estonia Yes Finland Yes France Yes Germany Yes Greece Yes Hungary Yes Ireland Yes Italy Yes Latvia Yes Lithuania Yes Luxembourg Yes Malta Yes Netherlands Yes Poland Yes Portugal Yes Slovakia Yes Slovenia Yes Spain Yes Sweden Yes United Kingdom Yes Sources: European Union. Notes: The removal of mobility barriers to Bulgaria and Romania for country i is indicated as “Yes” in the table. In countries that used the transitional arrangements to restrict the right of free movement of workers temporarily, workers had to apply for a work permit through the usual work permit system, although the system allowed some reliefs. For example, in Germany graduate students in some fields were exempted from a labour field test, although they still needed a work permit. In Italy workers did not require work permits for working in some sectors, ranging from seasonal work or work within the tourism sector to high skilled work (European Union, 2011). 14 4. Data and Methodology This section provides a presentation of the data used to conduct the empirical analysis as well as the empirical approach where the econometric models and methods are presented. The section is divided into three subsections, where Section 4.1 provides an overview of the research question, Section 4.2 provides information about the data, data collecting process, construction of variables as well as descriptive statistics and Section 4.3 presents the method and econometric models used in the analysis. 4.1 Research questions The main research question for our analysis which aim answer how internal EU migration affects EU attitudes follows: 1. Does the increase of intra-EU migration, due to the inclusion of additional countries in the European common labour market, affect the public opinion of the European Union among the member countries? Free movement is generally supported among EU citizens while, on the other hand, large inflows from new EU countries have been raised as an issue, especially in the countries that receive the largest inflow of workers. In other words, we want to understand how public opinion of the European Union is affected when more countries are included in the union and in extension increases the potential pool of migrants that are allowed to move without restrictions within the union. We also want to understand if this potential effect differs between countries in the union. This leads us to our second research question which is of a more exploratory characteristic: 2. Does the effect of an increased intra-EU migration, due to the inclusion of additional countries in the European common labour market, on the public opinion of the European Union differ between member countries? The enlargement studied in this analysis is relevant from a policy perspective since countries that are currently candidates for membership in the European Union are similar to Bulgaria and Romania in many senses when it comes to macroeconomic factors and institutional challenges to become eligible for EU membership. Hence, the results could give 15 insight into how inclusion of more countries into the European common market will affect EU attitudes. Today there are eight candidate countries to the EU: Albania, Moldova, the Republic of North Macedonia, Montenegro, Serbia, Turkey, Ukraine, and Bosnia and Herzegovina. To add, there are potential candidate countries such as Kosovo and Georgia (European Union, n.d.1). The result in our analysis could therefore give an indication how these future enlargements might affect EU opinion among the already existing member countries. 4.2 Data To answer our research question of how changes in intra-EU migration affects EU attitudes, we construct a unique dataset with aggregate survey data from the Eurobarometer and country-level data on migration from Eurostat during the period 2004 to 2019. Thus, we have data on migration before and after the transitional agreements were implemented. The Eurobarometer was initiated in 1974 by the European Commission as an instrument to regularly monitor the public opinion about issues and attitudes related to the European Union as well as issues of political and social nature (European Union, n.d.2). The Eurobarometer survey uses random sampling, where the sample from each country consists of at least 1000 randomly selected persons6 over the age of 15, and the total sample is weighted to get a geographical and demographic representative sample of the European population (European Union, n.d.2). The standard Eurobarometer survey is generally conducted through face-to-face interviews in the appropriate national language, and the respondents are not told at the beginning of the interview that the survey is commissioned by the EU. The response rate of the survey differs between countries. As an example, in the 2019 survey the response rates ranged from 17.6 percent in Finland to 78.0 percent in the Netherlands (European Commission, 2019). The mean response rate for the 2019 survey was 44.17 percent (European Commission, 2019). Since this paper focuses on changes in attitudes due to the enlargement of the EU in 2007, the analysis covers the attitudes of the countries that were part of the EU27 (2007–2013). Croatia is thus excluded since they joined the EU during the time span of the analysis. Since the analysis examines the changes in EU attitudes in the pre-existing EU member countries, the attitudes of Bulgaria and Romania are also excluded. This leaves us with 25 EU countries in the sample, covering 16 years, providing 400 observations. 6 If the country has less than 1 million inhabitants, the sample consists of at least 500 randomly selected persons. 16 Table 2. Summary statistics VARIABLES N mean sd min max Dependent variables: Average EU Image (scale 1–5) 400 3.181 0.252 2.370 3.870 Very Negative EU Image (%) 400 4.285 3.725 0 23 EU Trust (%) 400 46.02 11.61 17 71 For EU enlargement (%) 400 46.55 13.72 17 78 Migration variables: Migration from EU27 countries to country i* 355 50 405 82 895 62 456 174 The log of migration from EU27 countries to country i* 355 9.592 1.833 4.127 13.031 Instrument variables: Average distance from Bucharest and Sofia to capital in country i (in 100 km) 400 14.83 5.748 6.311 28.661 Open border to Bulgaria and Romania 400 0.623 0.485 0 1 Interaction variables: Founding members of the EU 400 0.240 0.428 0 1 EU budget contributor 400 0.395 0.489 0 1 Eastern Europe 400 0.160 0.367 0 1 Economic controls: GDP per capita (€) 400 27 894 17 156 5180 100 360 Unemployment rate (%) 399 6.048 3.113 1.583 19.239 Asylum applicants 399 20 003 53 127 5 745 160 Demographic controls: Female (%) 400 51.845 1.325 49 56 Average age 400 46.271 2.035 40.70 51.60 Educational attainment (in years) 400 19.134 2.068 14.50 26.10 Living in rural area or village (%) 400 34.910 11.273 5 67 Notes: *Except for the reporting country. 4.2.1 Dependent variables The questions used for the dependent variables are part of the standard battery questions for the standard Eurobarometer, which allow us to collect the answers from these questions for each year included in the analysis. We have collected data from the autumn Standard Eurobarometer, where the standard battery questions are included, which is carried out in October–November and published in December each year. The set of EU attitudes that we are examining consists of four variables: Average EU Image, Very Negative EU Image, EU Trust and For EU Enlargement. The Average EU Image and Very Negative EU Image are collected from the same question in the Eurobarometer while EU Trust and For EU Enlargement are collected from separate questions. Starting with the Average EU Image and Very Negative EU Image, the following question is used: 17 In general, does the EU conjure up for you a very positive, fairly positive, neutral, fairly negative or very negative image? The respondent answers on an ordered five-level scale, from very negative to very positive. The Average EU Image is then constructed by multiplying the share of each answer by the corresponding numerical value in the ordered scale7. This gives each observation a value between 1 to 5 depending on the answers in the survey. The variable Very Negative EU Image on the other hand is used to capture the extreme value of individuals' EU attitudes. The variable therefore includes the share of the respondents that choose the alternative very negative image. Thus, while the Average EU Image tracks how the average perception of the EU changes, the Very Negative EU Image tracks whether the share of the most negative responses changes. The variable EU Trust is constructed by using a question where the respondents can choose from three different answers. For this variable we are interested in the share of the respondents who answered that they tend to trust the EU. I would like to ask you a question about how much trust you have in certain media and institutions. For each of the following media and institutions, please tell me if you tend to trust it or tend not to trust it: The European Union where the respondent can answer that they Tend to trust, Tend not to trust or that they Don’t know. Lastly, the For EU Enlargement-variable collects the share that agrees that a further enlargement should take place from the following question. What is your opinion on each of the following statements? Please tell me for each statement, whether you are for it or against it: Further enlargement of the EU to include other countries in future years. where the answers are For it, Against it, or Don’t know. 7 EU Image = 1 ∗ very negative image + 2 ∗ fairly negative image + 3 ∗ neutral + 4 ∗ fairly positive image + 5 ∗ very positive image 18 4.2.2 Independent variable The independent variable used in the empirical models consists of the migration from all EU27 countries8 to a given country i and is collected from Eurostat. As seen in Graph 1, the intra-EU migration does on average display an increase over the years included in the analysis. Graph 1. Migration during the period 2004–2019 Notes: Intra-EU Migration refers to EU27-Migration, and the variable is logged. Since Eurostat does not provide a variable for EU27 migration for the whole period examined, our variable is constructed by using three different variables from Eurostat: EU27 countries (2007-2013) except reporting country, EU28 countries (2013-2020) except reporting country and migration from Croatia9. These variables track immigration, which applies to people that have their usual residence for a period that is, or is expected to be, longer than a year in another country than previously. In other words, the variable tracks all kinds of immigration regardless of the reason for moving. Eurostat’s data are collected from national statistical institutes, that in turn uses the appropriate data source at disposal within the framework that adheres to the definition of migration statistics for harmonizing statistics across 8 This variable includes migration from Bulgaria and Romania before and after they officially became EU members, i.e., during the years 2004-2007. 9 Eurostat does not report data on the variable EU27 countries (2007–2013) after 2013. Thus, the EU28 countries (2013–2020) variable is used where the migration from Croatia is deducted from this variable. 19 countries. The data is typically collected from either administrative sources, sample surveys, census data, or a combination of the above (Eurostat, 2022). The independent variable suffers from missing values10, causing an unbalanced panel. The missing values are primarily occurring in the early years of the sample and for some specific countries, which could cause a bias since the missing values are not totally randomly distributed in the sample. The potential problem with an unbalanced panel is further discussed in section 5.4. 4.2.3 Instrumental variables The instrument utilized in the IV-regression is the distance from the EU2-countries interacted with a time dummy variable indicating when EU countries removed the mobility barriers towards the recently accepted countries. The distance variable, Average distance to Bucharest and Sofia, is constructed using data from CEPII (French center for research and expertise on the world economy). The CEPII data provide distances in kilometers between country pairs where the longitude and latitude of the capital cities in each country pair is used to calculate the geodesic distance following the great circle formula, i.e., the equation for calculating distances between two dots on a sphere (Mayer and Zignago, 2011). From this data we select the country pairs that are of interest for our analysis, i.e., Romania + country i and Bulgaria + country i. For each country i, we take the average distance to Romania and Bulgaria11. The distance variable used in the analysis is scaled to 100 km (which roughly translates to a one-hour car ride). To get variation in the distance variable over time, we interact it with the variable Open border. This is a dummy variable that takes the value of 1 when a country has opened their borders towards Bulgaria and Romania. This variable is constructed using the information of when EU-countries opened their borders according to the transitional agreements that can be seen in Table 1 in Section 3. 4.2.4 Interaction variables As seen in the literature review, countries can be affected in different ways by changes in migration flows depending on the characteristics of the country (Mayda, 2006; Meltzer et al., 2018). To analyze whether there are any heterogeneous effects between countries within 10 Belgium: 2004–2006, 2008–2009, Cyprus: 2004–2008, France: 2004–2005, Greece: 2004–2007, Ireland: 2004–2005, Italy: 2004–2007, Latvia: 2004–2010, Malta: 2004–2007, Poland: 2008, Portugal: 2004–2007, Slovakia: 2004–2007, UK: 2004–2006 11 Since Bulgaria and Romania are neighboring countries, we argue that taking the average distance captures the distance effect on migration. 20 the European Union, we have constructed a set of interaction dummies. Using data on EU budget accounts from Eurostat we construct the dummy EU budget contributor which takes the value 1 if a country contributes more to the EU budget than they receive in returns, 0 if a country is a EU budget receiver. The countries contributing more than they receive are also in general the ones with the highest GDP per capita. Thus, this division explores the attitudes of the countries that bear the biggest fiscal burden in the union, and at the same time also receives the most migration of new member states. The interaction variable Founding members include the founding members of the EU (France, Germany, Belgium, Netherlands, Italy, and Luxembourg) and aims to analyze whether the effect of migration on EU attitudes differs for these countries compared to the newer members of the EU. We also construct a dummy for the Eastern European12 countries in the EU since Meltzer et al. (2018) have pointed out that Eastern European countries are the most positive toward free movement within the EU. This interaction variable could also indicate differences in attitudes due to migration between countries responsible for sending the most migrants to other EU countries and EU countries that are on the receiving end. Figure 2. Divisions for the heterogeneity analysis A. Countries contributing to EU budget B. Founding members C. Eastern Europe Notes: Bulgaria and Romania are highlighted in green in the maps. Sources: OpenStreetMap, Eurostat, EU Commission, UNSD 4.2.5 Control variables To decrease the risk of omitted variable bias, two sets of controls are applied to all models. The control variables are divided into a set of economic controls and a set of demographic controls. These sets are added in a stepwise approach in the empirical models. 12 Based on the ISO 3166 division of geographical regions 21 The economic controls are included to capture macroeconomic shocks that are not captured by the country and time fixed effects. To control for the effects of both the financial crisis in 2008 and the European sovereign debt crisis in 2009, we control for GDP per capita and Unemployment rate for each country since these are macroeconomic factors that can be used to indicate how a country has been affected by a crisis (Iacus & Porro, 2015). For example, in Greece the unemployment rate was 28 percent at the highest in the years to come after the euro-crisis (OECD data, 2022). Another event during this period that potentially could affect the outcome is the Syrian refugee crisis in 2015, where more than 1 million Syrians applied for asylum in Europe (UNHCR, 2021). We thus control for Asylum applicants in each country to control for the potential effect the refugee crisis had on EU attitudes. The demographic controls refer to population characteristics which might affect EU attitudes and are all collected from the Eurobarometer surveys. Female capture the share of females in the survey for each country each year. We also control for average age in each survey as well as the average educational attainment, which have been identified as important factors in explaining differences in attitudes overall (Tosun et al. 2023). We also include the share of respondents that have stated that they live in a rural area or village since studies have shown that political views tend to differ between inhabitants in rural and urban areas in the same country (Mitsch et al., 2021; Tosun et al., 2023). 4.3 Empirical models and strategy This section provides an in-depth explanation of the model specifications used in the analysis. In Section 4.3.1 the OLS model is explained and in the following sections the IV model is discussed. The models used in the analysis are based on the literature and conceptual framework in Section 2 as well as the institutional settings presented in Section 3. We have also taken inspiration from previous studies on causes of migration patterns and macroeconomic impacts of EU enlargements (Angrist & Kugler, 2003; Pedersen & Pytlikova, 2008; Llull, 2018; Roman, 2018) 4.3.1. OLS with fixed effects To assess whether increased migration due to the inclusion of additional countries in the EU has affected EU attitudes, we begin by conducting a simple OLS regression with country and year fixed effects over the years 2004 to 2019 for the EU27 countries. The model examines the correlation between migration from the EU27 countries and a set of attitudes, namely the Average EU Image, Very Negative EU Image, EU Trust and For EU Enlargement. 22 𝐴𝑡𝑡𝑖𝑡𝑢𝑑𝑒 = 𝛼′ ′𝑖𝑡 0 𝑋𝑖𝑡 + 𝛼1 𝑍𝑖𝑡 + 𝛼2 𝑂𝑝𝑒𝑛 𝑏𝑜𝑟𝑑𝑒𝑟𝑖𝑡 + 𝛽 𝐼𝑛𝑡𝑟𝑎– 𝐸𝑈 𝑀𝑖𝑔𝑟𝑎𝑡𝑖𝑜𝑛𝑖𝑡 + 𝛾𝑖 + 𝜃𝑡 + 𝜀𝑖𝑡 (1) In model (1) the dependent variable Attitude is any of the four attitudinal variables previously presented and the regressor Intra-EU Migration refers to migration from all EU27 countries to country i13. 𝛽 is the parameter of interest since this parameter indicates the average effect of a one unit increase of the regressor on the outcome. Furthermore, we include the pre/post dummy Open border to control for a potential effect from removing mobility barriers on EU attitudes. This control variable indicates the moment in time each country opened their borders to Bulgaria and Romania. In this model, we control for a set of economic covariates, 𝑋𝑖𝑡, as well as a set of demographic covariates, 𝑍𝑖𝑡 (see Section 4.1.5) which are added in a stepwise approach. 𝛾𝑖 denotes the country fixed effects while 𝜃𝑡 refer to the year fixed effects. 𝜀𝑖𝑡 is the error term and the model is estimated with heteroskedasticity-robust standard errors14. 4.3.2. Instrumental Variable regression The OLS model (1) previously presented can, even though we control for country and year fixed effects and include two sets of control variables, only provide correlational results since both reversed causality as well as omitted variables could affect the results. The simple OLS regression is likely to suffer from omitted variable bias, which arises when the independent variable and the dependent variable also is determined by a variable not included in the model (Verbeek, 2017). Reverse causality could also be an issue, where the relationship between intra-EU migration and EU opinion could be reversed, i.e., when choosing a country to move to, people might consider the public opinion in that country. Hence, to gain exogenous variation and handle issues related to endogeneity, we design a Two-stage Least Square setup (2SLS) with country and year fixed effects using a distance variable interacted with a time variable as an instrument to infer random variation in the migration variable. 4.3.2.1 Choice of instrument When Two-stage Least Squares is used as the empirical method, the choice of instrument is crucial. Distance has in many studies been proven to have a significant negative effect on migration. Ever since Ravenstein in 1885 proved that most migrants only move short 13 Except reporting country 14 Since we use aggregated individual level data - we cluster at country year level, thus we control for correlation for a country within a year. However, we are not able to control for serial correlation over time with clustered standard errors because of the statistical power in our analysis due to the limited dataset available. 23 distances, the notion that greater distance deter migration has gained a large body of evidence (Lucas, 2001). The proposition that migration is negatively affected by distance can be seen both when looking at industrialized countries as well as developing countries and seems to be true for both internal and international movement (Lucas, 2001). Hence, the instrument takes advantage of this widely observed phenomenon between migration and distance. Both neoclassical theory and international trade models, as the gravity model, has provided theoretical explanations on why migration and distance display a negative relationship. Lim (2019) presents a neoclassical human capital model of migration, where migration is considered an investment decision for the individual. Lim (2019) incorporates the distance effect in the model as both a monetary and a nonmonetary fixed cost that increases the risk premium. As a monetary cost of distance, Lim (2019) put forward transportation costs which increase with distance, especially if the soon-to-be migrant plans to travel home frequently. As an example of a non-monetary cost, Lim (2019) suggests information costs that might increase with longer distances. For Lucas (2001), an explanation to the fact that many migrants only move short distances could be that a concentration of fellow countrymen in places nearby would lower the costs of migration through providing a network and decreasing information costs. The negative effect of distance on migration can also be seen through the gravity model, a model most used in trade literature, but which has become more frequently used in the context of migration patterns. The model states that bilateral trade decreases with distance, i.e., countries conduct most trade with countries nearby and less trade with countries far away. In papers exploiting the gravity model in the context of migration patterns, the same effect is found (Karemera et al., 2000; Poprawe, 2015). The further away a country is located, the less migration there is, all else equal. However, since distance is constant over time and we utilize panel data with country and time fixed effects, bilateral distance cannot be used as an instrument by itself since it does not provide any variation over time. To combat this issue, we take advantage of the fact that EU countries opened their borders to Bulgaria and Romania at different points in time. The heterogeneity due to the differences of transitional agreements between EU members have been used in previous studies to examine migration patterns in relation to enlargement of the European Union (Pedersen & Pytlikova, 2008; Roman, 2018). Both Pedersen and Pytlikova (2008) and Roman (2018) argue that, from a methodological point of view, the differences in opening years could be seen as a “natural experiment” since it causes exogenous variation between countries. 24 Using bilateral distance interacted with a time dummy as an instrument for migration has previously been exploited by Angrist and Kugler (2003) in a cross-country analysis of the effects of immigration on native’s labour market participation in Europe. Specifically, they examined the effect of the vast inflow of European immigrants from Yugoslavia during the Balkan war on the immigrant share in a given country. As their instrument for migration, Angrist and Kugler (2003) use distance from Sarajevo/Pristina to the capital in each country15 interacted with a time dummy for the war period16. In a similar setting, Llull (2018) uses bilateral distance interacted with a push factor (war, natural disaster, political regime changes or economic variables) as an instrument for immigration to examine the effect on native male wages. Llull (2018) argues that the relevance of both his and Angrist and Kugler’s instrument comes from the fact that distance in this case mitigates the effect of the push factor on migration. In our case, the time interaction in our instrument can be seen as a pull factor of migration, since the removal of mobility barriers is something that would attract migrants to another country rather than something that pushes them out of their own country. 4.3.3 2SLS specifications Building on Angrist and Kugler’s paper (2003), we utilize a similar strategy in our empirical approach. We use the break in time-series due to the status of transitional agreement towards Bulgaria and Romania to examine changes in public opinion towards the EU for countries further away or closer to these countries. In other words, the distance to Bulgaria and Romania is assumed to mitigate the exogenous shock that comes with removing mobility restrictions towards recently accepted EU countries on the migration inflow for a given country. Thus, when interacting the time dummy variable with the distance, we gain an instrument that varies over both time and countries. The first stage of the IV regression is therefore as follows: 𝐼𝑛𝑡𝑟𝑎– 𝐸𝑈 𝑀𝑖𝑔𝑟𝑎𝑡𝑖𝑜𝑛 = 𝜋′ ′𝑖𝑡 0𝑋𝑖𝑡 + 𝜋1𝑍𝑖𝑡 + 𝜋2𝑂𝑝𝑒𝑛 𝑏𝑜𝑟𝑑𝑒𝑟𝑖𝑡 + 𝛾 𝐼𝑛𝑠𝑡𝑟𝑢𝑚𝑒𝑛𝑡𝑖𝑡 + 𝜓𝑖 + 𝜏𝑡 + 𝜂𝑖𝑡(2) where 𝐼𝑛𝑠𝑡𝑟𝑢𝑚𝑒𝑛𝑡𝑖𝑡 = 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 𝑡𝑜 𝐵𝑢𝑐ℎ𝑎𝑟𝑒𝑠𝑡 & 𝑆𝑜𝑓𝑖𝑎𝑖 ∗ 𝑂𝑝𝑒𝑛 𝑏𝑜𝑟𝑑𝑒𝑟𝑖𝑡 15 Or the nearest city with a population of at least 10 000 16 One dummy for 1991-95 (Bosnia War), one dummy for 1996-97 (inter-war period) and one dummy for 1998- 99 (Kosovo War). 25 is the instrument used to predict migration. The average distance is measured in 100 km from the capital in a given country i. Accordingly, 𝛾 is the parameter for the effect of the instrument on intra-EU migration. In the first stage specification, 𝜓𝑖 is the country fixed effects, 𝜏𝑡 the year fixed effects and 𝜂𝑖𝑡 is the error term. The second stage specification is the same as the OLS specification (1) 𝐴𝑡𝑡𝑖𝑡𝑢𝑑𝑒 = 𝛼′ ′𝑖𝑡 0 𝑋𝑖𝑡 + 𝛼1 𝑍𝑖𝑡 + 𝛼2 𝑂𝑝𝑒𝑛 𝑏𝑜𝑟𝑑𝑒𝑟𝑖𝑡 + 𝛽 𝐼𝑛𝑡𝑟𝑎– 𝐸𝑈̂ 𝑀𝑖𝑔𝑟𝑎𝑡𝑖𝑜𝑛𝑖𝑡 + 𝛾𝑖 + 𝜃𝑡 + 𝜀𝑖𝑡 (3) with the difference that the migration variable is estimated by the instrument. Subsequently, 𝛽 is the parameter which estimates the average effect of intra-EU migration on EU attitudes, given the intra-EU migration has been predicted by the instrument. This specification also includes Open border to control for the effect of removing mobility barriers on EU attitudes in itself. As before, 𝑋𝑖𝑡 and 𝑍𝑖𝑡 are sets of economic and demographic covariates, 𝛾𝑖 and 𝜃𝑡 are country and year fixed effects and 𝜀𝑖𝑡 is the error and the model is estimated with heteroskedasticity-robust standard errors17. Lastly, to examine whether the effect of intra-EU migration on EU attitudes displays any heterogeneity between countries, we utilize a model that estimates interactions between migration and country groups, specified as three interaction variables (see section 4.1.4). 𝐴𝑡𝑡𝑖𝑡𝑢𝑑𝑒𝑖𝑡 = 𝛼 ′ 0 𝑋𝑖𝑡 + 𝛼 ′ ̂ 1 𝑍𝑖𝑡 + (𝛼2 + 𝛼3𝐺𝑖) 𝑂𝑝𝑒𝑛 𝑏𝑜𝑟𝑑𝑒𝑟𝑖𝑡 + (𝛽0 + 𝛽1𝐺𝑖) 𝐼𝑛𝑡𝑟𝑎– 𝐸𝑈 𝑀𝑖𝑔𝑟𝑎𝑡𝑖𝑜𝑛𝑖𝑡 + 𝛾𝑖 + 𝜃𝑡 + 𝜀𝑖𝑡 (4) In this specification, 𝐺𝑖 indicate a specific country group interaction. We look at three different types of country divisions (founding member states of the EU, EU budget contributors and Eastern Europe countries). The parameter 𝛽0 capture the average effect of intra-EU migration on any of the four EU attitudes when the country group interaction equals zero. The parameter 𝛽1 on the other hand, estimate how this effect changes given that the country is included in one of the country group interactions. Hence, to get the effect of increased intra-EU migration on EU attitudes for a country included in one of the interactions, we sum the average effect 𝛽0 with the effect difference 𝛽1. The instrument as well as open border is augmented with the group interaction dummy in each setup of this specification. 17 Since we use aggregated individual level data - we cluster at country year level, thus we control for correlation for a country within a year. However, we are not able to control for serial correlation over time with clustered standard errors because of the statistical power in our analysis due to the limited dataset available. 26 4.3.4 Relevance and validity of instrument The instrument should fulfill two assumptions: the relevance assumption and the exogeneity assumption. The relevance assumption states that for an instrument to be valid, it has to be correlated with the explanatory variable. If the instrument proves to be significant and correlated with the endogenous regressor in the first stage regression, the relevance assumption is considered to be fulfilled. In Table 6 (Section 5.2.1), we present the results from the first stage where the instrument proves to be significant and displays the correct sign, i.e., distance mitigate the effect opening borders have on migration. For an instrument to be considered a good instrument, the first stage correlation should preferably be strong (Verbeek, 2017). The F-statistic from the first stage regressions give an indication of the strength of the instrument. For our instrument, the F-statistics ranges around 7, which is quite weak. When dealing with weak instruments in a 2SLS, the second stage estimations should be interpreted with caution since these estimates are more likely to be biased (Murray, 2006). There might however be a general weakness with instruments that use distance interacted with a time dummy. The F-statistics for Angrist and Kugler’s (2003) instruments do for example ranges between 0.14 and 6.96, and in Llull’s (2018) paper the F-statistics for the instruments ranges from 6.35 and 12.84. When dealing with weak instruments, special attention should be given to establish that the instrument fulfills the exogeneity assumption (Murray, 2006). This assumption states that the instrument should only affect the dependent variable through the endogenous regressor. If more than one instrument is used to infer exogenous variation in the regressor, an overidentifying restrictions test (Sargan J test) can be used. However, since we only use one instrument the exogeneity assumption must be fulfilled through economic reasoning and knowledge of the subject of interest. For the exogeneity assumption to be fulfilled for our instrument, the average distance to Bulgaria and Romania has to be argued to not be more or less correlated with attitudes after opening the border given that there was no increase in the number of migrants. When the average distance to Bulgaria and Romania was regressed on attitudes for each year in the sample (see Appendix table I), almost all estimates showed statistically insignificant results and we could not detect any pattern between distance and attitudes18. Thus, this indicates that distance does not seem to be correlated with EU-attitudes. Further, looking at EU 18 For the outcome For EU enlargement the pattern is however consistently negative and displays some significant estimates. Thus, we need to be more careful when interpreting the results for this outcome. 27 parliament election data from 2019, EU-skeptic parties are spread over the member countries, with EU-skeptic parties having hit prosperity in countries from Poland to Italy, the UK as well as Sweden which further proves that distance and attitudes do not display any significant geographical pattern (European Parliament, 2023). The random distribution of attitudes can also be viewed graphically when examining how trust in the EU is spread across the member countries. In Figure 3 is it apparent that the attitudes are uncorrelated with the distance to the EU2-countries (graphics on remaining attitude sets are found in Appendix). Therefore, there is no reason to believe that the average distance to Sofia and Bucharest should correlate with, or have a direct impact on, the public opinion of the EU for a given country. Figure 3. Trust in the EU spread over the EU member countries A. Trust in EU 2004 B. Trust in EU 2019 Sources: OpenStreetMap, Eurobarometer Furthermore, to make sure the instrument is uncorrelated with attitudes given that distance is interacted with the removal of mobility restrictions for each country, we regressed the time dummy for when a country opened their borders towards Bulgaria and Romania on EU-attitudes (see Appendix table J). The results showed no significant effects of opening borders for the attitudes Average EU Image, Very Negative EU Image and EU Trust. This provides us with further confidence that the instrument is valid. However, for the outcome For EU Enlargement we do get some significant results, thus we need to be more careful when 28 interpreting the results for this outcome in the IV regressions. To correct for the potential effect opening borders have on EU-attitudes in itself, we add this time variable as a control in all of our specifications19. However, since the instrument utilizes two different exogenous measures in combination to predict migration: distance and the status of transitional agreements, we do argue that the instrument does fulfill the exogeneity assumption for at least the first three outcomes. 5. Results This section provides a presentation of the results from the equations (1–4) in the previous section. The section is divided into three subsections, where Section 5.1 provides the results from a simple OLS model, Section 5.2 provides the results for the main IV-regression, while Section 5.3 presents the results from the IV-model with interactions. 5.1 Simple OLS model This section provides a presentation of the results from the OLS specification (1) presented in Section 4.3.1. The results aim to give a first glance on the problem and examine the correlation between increased intra-EU migration, from the inclusion of Bulgaria and Romania into the common labour market, on the perception of the EU. Table 4 presents the results from the OLS panel data model, divided between the set of attitudes examined. The R-squared is rather high across all different specifications, varying between 0.723 and 0.881, indicating that the model specification explains much of the variation in the dependent variable. Although, since the model specification suffers from endogeneity problems, i.e., other factors which are not specified in the model could play a role in the creation of attitudes and where people choose to migrate, the results should be interpreted with caution. Migration increasing by one percent is correlated with a 0.005 point increase in Average EU Image on the scale between 1 and 5 (Column 1), holding all else constant (significant at 1 % level). When controlling for economic and demographic factors, the correlation is no longer significant (Column 2–3). The same pattern appears when examining the correlation with Very Negative EU Image. The estimate is significant at 1% level in the baseline, indicating that a one percent increase in migration is correlated with an 0.005 percentage point average decrease 19 The distance is controlled for by itself by using fixed effects. 29 in the Very Negative EU Image, holding all else constant. When economic and demographic controls are applied the estimates are insignificant (Column 2–3). The estimated correlation between migration and EU Trust is larger. A one percent increase in migration is related to an on average increase by 0.018 percentage points in the EU Trust in the baseline specification. The estimate turns insignificant in Column (2–3). For the outcome For EU Enlargement neither of the columns provide significant coefficients, but the estimates for the OLS regressor are consistently negative. To sum, the results presented in Table 4 indicate that countries gain a slightly better image of the EU and display slightly more trust in the EU when the intra-EU migration increases. 30 Table 4. OLS regressions Dependent variable: 1. Average EU Image 2. Very Negative EU Image 3. Trust EU 4. For Enlargement (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) Intra-EU Migration 0.052*** 0.017 0.012 -0.545*** -0.010 0.027 1.833*** 0.750 0.502 -0.188 -0.483 -0.534 (0.015) (0.014) (0.014) (0.166) (0.173) (0.168) (0.666) (0.699) (0.669) (0.381) (0.445) (0.445) Open border 0.009 0.029 0.012 -0.053 -0.284 -0.115 1.775 2.156* 1.527 4.771*** 4.797*** 4.919*** (0.029) (0.026) (0.029) (0.409) (0.356) (0.418) (1.337) (1.295) (1.427) (1.026) (1.029) (1.109) Economic controls No Yes Yes No Yes Yes No Yes Yes No Yes Yes Demographic controls No No Yes No No Yes No No Yes No No Yes Country & Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Constant 2.514*** 0.008 -0.732 12.06*** 63.51*** 63.11*** 26.67*** -83.66* -92.43 37.31*** -20.02 -137.8** (0.158) (0.945) (1.352) (1.726) (13.78) (15.96) (7.025) (46.81) (71.07) (4.187) (37.31) (55.64) R2 0.739 0.771 0.793 0.765 0.799 0.806 0.723 0.741 0.765 0.870 0.875 0.881 N 355 353 353 355 353 353 355 353 353 355 353 353 Notes: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01, Intra-EU Migration refers to the log of migration from EU27 countries to country i. Open border is a dummy- variable that takes the value of 1 when a country removes their mobility barriers to Bulgaria and Romania. Economic controls and Demographic controls consist of two sets of covariates, for full list see section 4.1. Economic controls are standardized. 31 5.2 IV regressions To combat the endogeneity problems in OLS-regressions we conduct IV-regressions, see specifications 2–4 in Section 4.3.3. The IV models aim to give more robust results and ideally provide a causal interpretation of the effect rather than only correlations. The results from the first stage regressions are presented in Table 6 and the results from the second stage regressions are presented in Table 7. 5.2.1 First stage In the first stage, we estimate the effect of the instrument Average distance to Bucharest and Sofia x Open Border on our endogenous variable Intra-EU Migration. The results from this are found in Table 6, below. Table 6. First stage Dependent variable: Intra-EU Migration (1) (2) (3) Instrument -0.044*** -0.033*** -0.035*** (0.016) (0.013) (0.013) Open border 0.659** 0.571** 0.641** (0.296) (0.259) (0.273) Economic controls No Yes Yes Demographic controls No No Yes Country & Time Fixed Effects Yes Yes Yes Constant 10.174*** -5.904 -8.480 (0.276) (4.591) (6.613) R2 0.906 0.921 0.922 F-statistic 7.685 6.819 7.068 N 355 353 353 Notes: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01, Intra-EU Migration refers to the log of migration from EU27 countries to country i. The instrument consists of an interaction between a time dummy variable for when a country removes their mobility barriers to Bulgaria and Romania and the average distance to the capitals of Bulgaria and Romania. Open border is a dummy-variable that takes the value of 1 when a country removes their mobility barriers to Bulgaria and Romania. Economic controls and Demographic controls consist of two sets of covariates, for full list see section 4.1. Economic controls are standardized. F-statistic for first stage obtained through estat firststage. The instrument has a significant effect on migration, indicating that the instrument is relevant since it is correlated with the regressor. The estimates of the instrument have expected signs where distance mitigates the effect of opening the border on migration which is in line 32 with previous literature and theory. However, the F-statistic of the instrument is below 10, thus the first stage is quite weak. Because of this, all coming results in the second stages should be interpreted with caution. We discuss this issue in detail in Section 6. The reduced form estimates are also significant for the instrument (see Appendix Table A) which is necessary for estimating an effect in the second stage (Verbeek 2017). 5.2.2 Second stage In the second stage, the predictions on intra-EU migration from the first stage are used to estimate the effect of changes in EU attitudes, where EU attitudes refers to the set of four EU attitudes, Average EU Image, Very Negative EU Image, EU Trust and For EU Enlargement. Examining the first outcome variable in the set of EU attitudes, we see that increased intra-EU migration lowers the Average EU Image. When the migration increases with one percent, the Average EU Image decreases by on average 0.002 points on the 1–5 point scale, holding all else constant. While the baseline estimate is not significant, the estimates in Column (2–3) are significant at 5 % respectively 10 % levels. When estimating the effect on the share of the most negative respondents, using the outcome variable Very Negative EU Image, we see that when the migration increases by one percent, the Very Negative EU Image increases on average by 0.027 percentage points (Column 1). Applying controls, the effect of a one percent increase in intra-EU migration causes an on average 0.045 points or 0.040 percentage points increase in the Very Negative EU Image (Column 2 or 3), holding all else constant. The estimate in the baseline is significant at a 10 % level, when applying the economic and demographic controls the estimates are significant at 1 % level. Looking at the EU Trust, when intra-EU migration increases with one percent, the EU Trust decreases on average between 0.120 to 0.146 percentage points (depending on whether one uses the Column 2 or 3 estimates), holding all else constant. The estimates for intra-EU migration are significant (5 % level) when controls are applied. Thus, the trust in the EU is negatively associated with increased migration. The estimates for intra-EU migration on For EU Enlargement are significant in all columns, when migration increases with one percent, For EU Enlargement decreases on average by approximately 0.089 to 0.120 percentage points (Column 1–3) holding all else constant. The estimated effects of intra-EU migration on different EU-attitudes from the main IV- models can appear to be rather small. However, when estimating the effect on EU attitudes, we look at the effect of a one percent increase in migration on the public opinion of the EU. This 33 result is not very telling since changes in migration from including an additional country to the EU are much larger. As an example, the intra-EU migration to Germany increased by 17.5 percent when they removed the mobility barrier towards Bulgaria and Romania in 2014 (Eurostat, 2023). Thus, the size of the effect of opening borders on EU attitudes could be more significant than one might be led to believe by estimating the effect of a one percent increase. As a comparison to the IV estimates presented in this section, in Appendix (table G) we present the results from the same model specification, using the EU27 migration variable without the log-transformation. These results display that an increase of 10,000 EU-migrants causes approximately a two percentage point decrease in EU trust. This implies that an inflow of 100,000 EU-migrants would be associated with an on average 20 percentage point decrease in citizens’ trust in the EU. We will discuss the implications of these results in section 6. 34 Table 7. IV regressions Dependent variable: 1. Average EU Image 2. Very Negative EU Image 3. EU Trust 4. For EU Enlargement (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) Intra-EU migration -0.111 -0.214** -0.156* 2.655* 4.489** 3.963** -9.575 -14.550** -11.956** -8.878** -12.047** -9.971** (0.109) (0.108) (0.090) (1.610) (1.751) (1.545) (6.020) (6.618) (5.510) (3.812) (5.110) (4.325) Open border 0.005 0.041 0.027 0.029 -0.517 -0.485 1.482 2.945 2.698 4.547*** 5.393*** 5.806*** (0.033) (0.031) (0.032) (0.536) (0.533) (0.567) (1.801) (1.866) (1.915) (1.415) (1.466) (1.410) Economic controls No Yes Yes No Yes Yes No Yes Yes No Yes Yes Demographic controls No No Yes No No Yes No No Yes No No Yes Country & Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Constant 4.177*** -1.415 -2.041 -20.733 91.320*** 93.755*** 143.578** -178.210** -189.433* 126.379*** -91.483 -211.254*** (1.108) (1.336) (1.759) (16.669) (24.149) (30.105) (61.850) (78.741) (106.976) (39.438) (69.180) (81.793) R2 0.599 0.536 0.670 0.536 0.422 0.521 0.404 0.264 0.454 0.739 0.685 0.756 N 355 353 353 355 353 353 355 353 353 355 353 353 Notes: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01, Intra-EU Migration refers to the log of migration from EU27 countries to country i. The instrument consists of an interaction between a time dummy variable for when a country removes their mobility barriers to Bulgaria and Romania and the average distance from the capitals of Bulgaria and Romania. Open border is a dummy-variable that takes the value of 1 when a country removes their mobility barriers to Bulgaria and Romania. Economic controls and Demographic controls consist of two sets of covariates, for full list see section 4.1. Economic controls are standardized 35 Comparing the results of the OLS and IV, it can be noted that the OLS estimates of intra-EU migration on Average EU Image are positive, but only significant for the baseline. However, for the IV-regression, the estimates are not significant in the baseline specification but are significant at 5 % and 10 % level respectively when the sets of controls are added. The application of the IV-model also causes the estimate to change signs and turn negative. The opposite is true for the OLS-estimates for migration on Very Negative EU Image where the sign of the baseline estimates is negative, while the IV estimates are positive. Thus, the OLS- and IV-regressions provide contradictory results for increased intra-EU migration on the public image of the EU. This is most likely caused by endogeneity problems in the OLS. Given that the estimates in the OLS become insignificant when we add the set of economic controls in the second step of each model, it could be argued that other events during the period which impacted EU countries differently, e.g., the European sovereign crisis, influenced both the intra-EU migration flows and public opinion of the EU, causing omitted variable bias. The same pattern is found when examining the estimates for migration on EU Trust where the OLS estimates are positive while the IV-estimates are negative. The OLS-estimates are non-significant for For EU Enlargement; thus, the OLS-estimates cannot be used for a comparison with the IV-estimates. 5.3 The effect on EU attitudes across different countries This section provides the results from the model specification (4)20, where the 2SLS is used in combination with an interaction model to provide a picture over how the effect of intra- EU migration on EU attitudes differs depending on different country factors. Table 8 provides the estimates of the effect of intra-EU migration on Average EU Image, given that a country is a founding member of the EU, is a net contributor to the EU budget or is an Eastern European country. Most of the estimates from these specifications are not significant, hence it is not possible to reject the null hypothesis that there is no effect. The fact that the model does not provide significant estimates is however not surprising given the limited dataset used. In Table 8 and 9, the estimate for Intra-EU migration represents the effect of migration on attitudes when the interaction term equals zero (founding members = 0, budget contributors = 0 or Eastern Europe = 0). This implies that in Columns 1–3 for example, the Intra-EU 20 Results with dependent variable EU Trust and Very Negative EU Image are found in Appendix, see table B and C. 36 migration estimates the effect of increased intra-EU migration on countries that are non- founding members to the union. Furthermore, the estimate for the country group interaction (for example Founding members x Intra-EU Migration) represents the difference in effect between the group of countries the interaction compares, e.g., between founding members and non-founding members of the European Union. Consequently, the effect of increased migration on attitudes for an interaction country group is calculated as the sum of the estimates for the Intra-EU migration and the estimated difference between the two groups of countries (e.g., Founding members x Intra-EU Migration). Table 8. IV-regression with interactions. Outcome: Average EU Image Dependent variable: Average EU Image (1) (2) (3) (4) (5) (6) (7) (8) (9) Intra-EU migration -0.072 -0.124* -0.091 -0.006 -0.106 -0.051 -0.263 -0.244* -0.199 (0.085) (0.070) (0.063) (0.093) (0.100) (0.085) (0.194) (0.132) (0.131) Open border 0.089** 0.116*** 0.092*** 0.004 0.018 -0.027 -0.009 0.038 0.025 (0.037) (0.032) (0.034) (0.042) (0.043) (0.038) (0.040) (0.033) (0.034) Founding members x Intra-EU Migration 0.197 0.325** 0.279** (0.132) (0.130) (0.131) Founding members x Open border -0.214*** -0.209*** -0.191*** (0.056) (0.047) (0.048) Budget contributors x Intra-EU Migration 0.044* 0.028 0.020 (0.024) (0.022) (0.017) Budget contributors x Open border -0.027 0.007 0.053 (0.072) (0.075) (0.064) Eastern Europe x Intra-EU Migration 0.163 0.026 -0.004 (0.194) (0.145) (0.146) Eastern Europe x Open border 0.104 0.021 0.060 (0.078) (0.070) (0.065) Economic controls No Yes Yes No Yes Yes No Yes Yes Demographic controls No No Yes No No Yes No No Yes Country & Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Constant 3.806*** -0.259 -0.276 2.640*** -0.604 -1.803 5.796*** -1.485 -2.369 (0.867) (1.052) (1.583) (0.952) (1.305) (1.537) (2.011) (1.425) (2.193) R2 0.689 0.705 0.761 0.702 0.692 0.779 0.389 0.495 0.596 F-statistic excluded instrument 9.221 13.860 15.114 3.973 3.947 3.892 6.306 6.149 5.860 F-statistic excluded instrument x 5.592 6.521 7.547 5.232 4.671 5.847 3.347 3.152 4.326 interaction N 355 353 353 355 353 353 355 353 353 Notes: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01, Intra-EU Migration refers to the log of migration from EU27 countries to country i. The instrument consists of an interaction between a time dummy variable for when a country removes their mobility barriers to Bulgaria and Romania and the average distance from the capitals of Bulgaria and Romania. Open border is a dummy-variable that takes the value of 1 when a country removes their mobility barriers to Bulgaria 37 and Romania. The instrument and open border are augmented with the interaction dummy in each setup. Economic controls and Demographic controls consist of two sets of covariates, for full list see section 4.1. Economic controls are standardized. F- statistic for first stage obtained through estat firststage. The estimates for the interaction between migration and founding members are significant at a 5 % significance level when applying the set of economic and demographic controls (Columns 2–3). When the migration increases with one percent at the same time as the country is a founding member, the Average EU Image increases with on average 0.002 points on the scale between 1 and 5, holding all else constant (Column 2). The result thus suggests that founding members do react differently to increases in intra-EU migration than new members where the results suggest that intra-EU migration could potentially increase the average image of the EU for founding members. This pattern is also seen when examining the heterogeneous effects on the Very Negative EU Image (see Appendix, Table C). Since these countries founded the EU with the aim to preserve peace after WWII through increased collaboration and trade between European countries, this could potentially explain why these countries react differently to increases in intra-EU migration compared to new members of the union. The interaction between intra-EU migration and budget contributing countries is significant (at 10% level) in the baseline specification, but insignificant when including economic and demographic controls. The effect of a one percent increase in migration increases the Average EU Image with on average 0.0004 points on the scale between 1 to 5 if a country is a budget contributor compared to if a country is a budget net beneficiary. Therefore, this contradicts what the fiscal burden theory would suggest. The estimates of migration for the Eastern European countries are insignificant and inconsistent, hence this suggests that the effect of intra-EU migration on EU attitudes for Eastern European countries is not significantly different from other countries in the EU. While Eastern European countries have proved to be the most positive toward free movement among all countries in the EU (Meltzer et al., 2018), Eastern Europe also mainly consists of countries from which people migrate from. Previous studies have in this regard argued that the sending countries could experience adverse effects from the migration, e.g., in terms of brain drain for the sending countries (Kahanec et al., 2009). In Table 8, we see that there are heterogenous effects on the Average EU Image in regard to whether the country was a founding member or not. To further examine the heterogeneous effects, we have analyzed the effects on the attitude towards a further enlargement as well. The estimates for the regressions are displayed in Table 9. 38 Table 9. IV- regression with interactions, EU Enlargement Dependent variable: For EU Enlargement (1) (2) (3) (4) (5) (6) (7) (8) (9) Intra-EU migration -8.134*** -9.534*** -8.113*** -6.377** -9.563* -7.399* -12.493** -12.904** -10.423* (2.910) (3.215) (2.862) (3.036) (4.892) (4.020) (5.937) (5.775) (5.786) Open border 7.307*** 8.015*** 7.606*** 5.123*** 5.353*** 4.632*** 4.312*** 5.340*** 5.795*** (1.593) (1.544) (1.530) (1.741) (2.021) (1.709) (1.562) (1.414) (1.402) Founding members x Intra-EU migration 3.598 8.832* 7.920* (4.490) (4.630) (4.621) Founding members x Open border -6.969*** -7.072*** -5.314*** (2.034) (1.714) (1.772) Budget contributors x Intra-EU migration 1.199* 0.749 0.527 (0.616) (0.680) (0.559) Budget contributors x Open border -1.812 -0.767 0.958 (2.411) (2.795) (2.380) Eastern Europe x Intra-EU migration 5.035 2.922 1.010 (6.937) (6.617) (6.785) Eastern Europe x Open border 1.490 -0.838 -0.173 (2.585) (3.043) (2.824) Economic controls No Yes Yes No Yes Yes No Yes Yes Demographic controls No No Yes No No Yes No No Yes Country & Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Constant 119.241*** -53.646 -162.388** 88.407*** -65.875 -202.469*** 164.836*** -82.652 -203.518 ** (30.438) (55.140) (79.118) (31.185) (67.969) (75.052) (61.954) (66.240) (91.685) R2 0.772 0.769 0.806 0.807 0.759 0.818 0.690 0.700 0.756 F-statistic excluded instrument 9.221 13.860 15.114 3.973 3.947 3.892 6.306 6.149 5.860 F-statistic excluded instrument 5.592 6.521 7.547 5.232 4.671 5.847 3.347 3.152 4.326 x interaction N 355 353 353 355 353 353 355 353 353 Notes: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01, Intra-EU Migration refers to the log of migration from EU27 countries to country i. The instrument consists of an interaction between a time dummy variable for when a country removes their mobility barriers to Bulgaria and Romania and the average distance from the capitals of Bulgaria and Romania. Open border is a dummy-variable that takes the value of 1 when a country removes their mobility barriers to Bulgaria and Romania. The instrument and open border are augmented with the interaction dummy in each setup. Economic controls and Demographic controls consist of two sets of covariates, for full list see section 4.1. Economic controls are standardized. F-statistic for first stage obtained through estat firststage. When intra-EU migration increases by one percent, given that the country also is a founding member, the variable For EU Enlargement decreases by on average 0.007 respectively 0.002 percentage points (depending on whether on estimates from Column 2 or 3), holding all else constant. This could be compared to the effect of a one percent increase in intra-EU migration for new member countries, which decreases For EU Enlargement by on average 0.081 or 0.095 percentage points. The estimates are nonsignificant in the baseline, but 39 significant at 10 % level when applying the set of controls (Column 2–3). As discussed earlier, EU attitudes of countries that are founding members thus prove to be less sensitive and relatively unaffected to increases in intra-EU migration compared to new members of the union, potentially because part of why they founded the EU was to increase European integration. When migration is interacted with countries that are net contributors to the EU budget, the effect is positive and significant at 10 % level at the baseline. Thus, countries that contribute more to the EU budget are less negative towards a further EU enlargement given an increase in intra-EU migration. When the migration increases with one percent, the contributing countries are on average 0.052 percentage points less prone to be in favor of an EU enlargement compared with on average 0.064 percentage points if the country is an EU budget receiver (Column 4), all else equal. Finally, the estimates for the Eastern Europe countries are consistent but insignificant throughout the columns. 5.4 Robustness- and sensitivity checks To examine if the results are sensitive for model misspecifications, the same regression models (Specification 2–3) have been rerun with a set of different corrections, all to be found in the Appendix. Since the independent variable suffers from non-random missing values, we have altered the migration variable slightly. First, we use a migration variable containing the EU27 countries (2007–2013) and the EU28 countries (2013–2020). The only difference between this variable and the main variable is that migration from Croatia is not subtracted for the years 2013–2019. The data covering Croatia has missing values for some countries and years, thus we have not been able to subtract the migration from Croatia for all years and countries. To examine how much this might have affected our results, we reran the regression without subtracting the migration from Croatia. The results were close to similar, probably because the migration from Croatia constitutes such a small share of the total intra-EU migration variable, which indicates we should not have to be concerned about this measurement error affecting our results significantly (see Appendix table D). Secondly, we utilized data for the migration from each country in the EU and created an index for the mean migration2122 and concludes that our results are robust to this change as 21 The index for migration is calculated as the mean of the migration from each country and is calculated manually after approximating missing values with the mean of the closest observations. 22 The reason for not using this variable in the final main analysis is that the country specific migration statistics suffered from more missing values than the EU27-migration variable. 40 well (see Appendix table D). Furthermore, we also reran the regression using a variable covering migration from the whole of Europe. While the migration variable for Europe is not as specific and includes information for migration from countries outside the EU, the variable still provides similar results as the EU27-migration variable (see Appendix table D). Lastly, to identify that the effect of the instrument on the EU27 migration is due to migration from Bulgaria and Romania, we rerun the IV regression using migration from only these two countries.23 This regression showed the same result, i.e., the instrument had a negative effect on migration from Bulgaria and Romania24 to a given country (see appendix table E). Moreover, since attitudes might not be formed simultaneously as the border opens, an analysis where the attitudes were forwarded one year was conducted (see appendix table H). Since the results were robust to this change in the dependent variable, we decided to use the non-forwarded variable to keep as many observations in the sample as possible. Since the data on attitudes is collected from the autumn Eurobarometer surveys, which is conducted in October–November each year, this variable might already capture the fact that attitudes could take some time to develop. 6. Discussion & analysis In this section we discuss the results in relation to our research question and how the results relate to previous literature on this subject. We will discuss the internal and external validity of the study, as well as what policy implications can be drawn from the results. The results from our IV regressions show that increased intra-EU migration, due to the inclusion of more countries in the common labour market, is associated with lower trust in the EU, a less positive view of the EU and less support for further enlargement of the union. Subsequently, the results show that intra-EU migration is positively associated with the share of the population that have a very negative view of the EU. The results from the IV regressions are consistent and robust to changes across the independent variable as well as to different sets of controls. These results are in line with Ringlerova (2022) who used migration from outside the EU to examine the link between migration and EU attitudes and found that immigration has been one of the drivers of EU-skepticism. Ringlerova (2022) also proposes that the EU support 23 This robustness is inspired by Angrist and Kugler (2003) that conduct a similar robustness test in their paper. 24 Since this variable exhibits a large number of missing values, this result should only be used as a comparison/indication. 41 is linked to culture and identity, thus we argue that expanding the EU might affect the citizens identity as Europeans and negatively impact their image of the EU. Lowered EU support from internal migration in the EU could also be based on a feeling of lost control since countries lack instruments to control the immigration flows internally in Europe which in extension threaten their national identity (Dorn & Zweimüller, 2021). Further, when including country specific interactions in model (4), the results display a significant heterogeneous effect between founding members of the EU and new member countries of intra-EU migration on attitudes. For the founding members of the union, increases in intra-EU migration is associated with an increase in EU image, although this effect is quite small. Correspondingly, the share of the population that have a very negative image of the EU is negatively affected by increased intra-EU migration for founding members. Since these countries founded the EU with the aim to preserve peace after WWII through increased collaboration and trade between European countries, this could explain why these countries' EU attitudes are affected differently by increases in intra-EU migration than newer members of the union. However, we did not find that countries that receive more immigrants, i.e., richer countries, in our analysis mentioned as contributors to the EU budget, are more negative towards the EU. Thus, this result does in a sense challenge Toshkov and Kortenska (2015) who found that immigration from Central and Eastern European countries to western European countries had a negative effect on the support of European integration in the host countries. The results from the interaction models might therefore suggest that political factors and history better explain differences in how immigration affects public opinion of the EU between countries rather than economic factors. 6.1 Internal validity The results from the empirical analysis could however be argued to suffer from internal validity problems. One problem which affects the interpretation of the results is the fact that our instrument has F-statistics from the first stage regression which ranges around 7, which is quite weak. The choice to use weak instruments in an IV-regression approach has been debated among economic researchers. It has been argued that using a poor instrument that does not fulfill the assumptions is more likely to provide wrong statistical results and inference than a simple OLS regression (Verbeek, 2017). When dealing with weak instruments in a 2SLS, one must interpret the second stage estimations with caution since these estimations are more likely 42 to be biased (Murray, 2006). Hence, the implication from the IV regression (see Appendix table G) that an increase in 100,000 EU-migrants would be associated with a 20 percentage point decrease in EU trust is probably a bit excessive. Further, the IV-estimates of intra-EU migration on For EU enlargement should be interpreted with extra caution since the exogeneity assumption potentially could be violated for this outcome. The problem with having a weak first stage might be due to a general weakness with instruments which use distance interacted with a time-dummy, since Angrist and Kugler’s (2003) and Llull’s (2018) instruments also displayed weak first stages. However, the weakness of the instrument could also be due to the few observations in our sample and the limited ability to track intra-EU migration patterns. Overall, the statistical power in our analysis is low. Using migration data on smaller geographical units would probably improve the empirical analysis. However, with the IV approach we are able to combat the problems that other events, such as the euro crisis and the refugee crisis, could bias our results. We are also able to combat problems with potential labour reforms or framing of the EU in the media. Since we know that the instrument displays a weak first stage, the sizes of the IV-estimates are most likely overestimated. Hence, the results should be interpreted with caution and rather be used as an indication of the direction of the effect intra-EU migration has on EU attitudes. Other threats to the internal validity that we have identified are related to the data used in the analysis. One such weakness is the low response rates of the Eurobarometer. However, researchers which examined the relationship between response rates and support of the EU did not find any evidence that samples with lower response rates were more supportive of EU than samples with higher response rates (De Vries & Hoffmann, 2020). The European commission states that the Eurobarometer is a reliable instrument to measure public opinion over time, which is how the Eurobarometer has been used in this paper. Also, regarding the implications that can be made from the results, it should be stated that a high trust in the EU or a positive EU image does not necessarily imply that the support for the EU is high. When asked about the trust in the EU, people might respond if they trust the EU institutions e.g., no corruption and democratic processes. Although, we argue that using questions that are formulated differently, covering different aspects of the EU provides a broad picture of the general perception towards the EU. 43 6.2 External validity Regarding the external validity of the study, the results do give some implications for further enlargements of the EU. Given that earlier studies have shown that people respond similarly to different groups of immigrants (Blinder, 2015), it is possible to assume that the results could be applicable in the coming enlargements of the EU. We have eight candidate countries waiting to join the EU, and given our results, one could expect that including them in the union would have a negative effect on the public opinion of the EU. When considering the external validity of our analysis, the period should also be considered. The analysis is based on the period before Brexit, Covid-19, and the war in Ukraine. These events could have changed the public's opinion about increased intra-EU migration and cooperation with additional countries, or to be more concise, the sensitivity in which people react negatively towards the EU from increased migration. While Covid-19 might have put the light on the problems with coordination within the union, which might lower the willingness to include further parties in the union, the war in Ukraine might have affected the public's opinion in the other direction, suggesting that we should tie the candidate countries closer to the EU, rather than risking that authoritarian countries do so. While this paper finds that inclusion of additional countries in the European Union is related to an increased negative public opinion of the EU, it does not examine the cause of the negative relationship between immigration and EU attitudes. The negative effect might be due to increased compositional concerns of intra-EU migration due to e.g., increased cultural differences. However, the negative relationship might also be due to the increase in labour supply and labour market competition due to an increased pool of migrants within the free movement zone, or the perceived increased fiscal burden of immigration on the nation state. Hence, to draw any policy implications from our findings of the negative association between intra-EU migration and EU attitudes, further studies on the cause of the negative relationships should be carried out. Furthermore, since the available data on immigration does not allow for a division between labour migration and students, we cannot differentiate the effect of labour immigration and other immigration on EU attitudes due to EU-enlargement. Consequently, we cannot examine if different types of immigration affect attitudes differently. Given that Blinder (2015) showed that when examining the effect of immigration, the public tend to imagine asylum seekers and permanent arrivals but fail to consider international students or spouses. This would also be an interesting subject for further research. 44 In conclusion, this study has contributed to the research field by examining the effect of the EU2 enlargement on EU attitudes through increased intra-EU migration. Furthermore, our paper also contributes to the research field of immigration effects on attitudes by providing an alternative approach to combat endogeneity problems. The results of this study should be considered as a first indication of the potential effect increased intra-EU migration, due to the inclusion of more countries in the common labour market, has on public opinion of the EU. However, this topic would need further research before any policy implications could be drawn from the results. 7. Conclusion This paper aims to examine the impact of increased intra-EU migration on EU attitudes. To do so, we utilize the inclusion of Bulgaria and Romania into the EU in 2007 to examine if changes in intra-EU migration composition influence the public opinion of the EU. Since previous studies on the effects of actual immigration on EU attitudes are likely to suffer from endogeneity problems, our paper has tried to combat such problems by using an IV approach. Covering the period 2004–2019 and 25 EU countries, we utilize the average distance from the capitals of Bulgaria and Romania combined with the removal of mobility barriers as an instrument to gain external variation in the intra-EU migration variable. The results from the IV regressions show that changes in intra-EU migration after a country has opened their border to Bulgaria and Romania are negatively associated with a positive image and trust in the EU, as well as negatively correlated with further enlargement of the EU. The analysis also shows a heterogeneous effect of intra-EU migration on attitudes between the founding members of the EU and new member countries, where the attitudes of the founding member countries are generally less affected by intra-EU migration than new members of the EU. The results from this paper might suggest that the EU image is negatively associated with further enlargements. Thus, the potential effect of intra-EU migration on EU attitudes could be considered a political cost for the EU. However, there are also potential political gains from including further countries, but these are not considered in this paper. Policy implications would thus need to be further researched. 45 References Aalberg, T., Iyengar, S., & Messing, S. (2012). Who is a 'Deserving' Immigrant? An Experimental Study of Norwegian Attitudes. Scandinavian Political Studies, 35(2), 97–116. Abbarno, A., & Zapryanova, G. (2013). Indirect Effects of Eurosceptic Messages on Citizen Attitudes toward Domestic Politics. Journal of Common Market Studies, 51(4), 581– 597. Angrist, J., & Kugler, A. (2003). Protective or counter-productive? labour market institutions and the effect of immigration on eu natives. The Economic Journal (London),113(488), 302–331. Baas, T., Brücker, H., & Hauptmann, A. (2009). Labor mobility in the enlarged EU: Who wins, who loses?. In EU labor markets after post-enlargement migration (pp. 47–70). Berlin, Heidelberg: Springer Berlin Heidelberg. Barrell, R., FitzGerald, J., & Riley, R. (2010). EU enlargement and migration: Assessing the macroeconomic impacts. JCMS: Journal of Common Market Studies, 48(2), 373-395. Blinder, S. (2015). Imagined Immigration. Political Studies 63(1), 80–100. Bohlmark, A., & Willen, A. (2020). Tipping and the Effects of Segregation. American Economic Journal. Applied Economics, 12(1), 318–347. Brader, T., Valentino, N., & Suhay, E. (2008). What Triggers Public Opposition to Immigration Anxiety, Group Cues, and Immigration Threat. American Journal of Political Science, 52(4), 959–978. Card, D. (1990). The Impact of the Mariel Boatlift on the Miami Labor Market. Industrial & Labor Relations Review, 43(2), 245–257. Card, D., Mas, A., & Rothstein, J. (2008). Tipping and the Dynamics of Segregation. The Quarterly Journal of Economics, 123(1), 177–218. Card, D., Dustmann, C., & Preston, I. (2012). IMMIGRATION, WAGES, AND COMPOSITIONAL AMENITIES. Journal of the European Economic Association, 10(1), 78–119. CEPII, Bilateral distance. http://www.cepii.fr/CEPII/en/bdd_modele/bdd_modele_item.asp?id=6 Accessed: 2023-02- 20. Dancygier, R. M., & Donnelly, M. J. (2013). Sectoral Economies, Economic Contexts, and Attitudes toward Immigration. New York, USA: Cambridge University Press. The Journal of politics,75(1),17–35. De Vries, C. E., & Hoffmann, I. (2020). The Eurobarometer controversy: EU popularity, response rates and one journalist's critique. Eupinions. https://eupinions.eu/de/blog/the- eurobarometer-controversy Accessed: 2023-04-18. Dorn, D., & Zweimüller, J. (2021). Migration and Labor Market Integration in Europe. The Journal of Economic Perspectives, 35(2), 49–76. Dustmann, C., & Preston, I. (2019). Free Movement, Open Borders, and the Global Gains from Labor Mobility. Annual Review of Economics, 11(1), 783–808. Dustmann, C., Schönberg, U., & Stuhler, J. (2017). Labor supply shocks, native wages, and the adjustment of local employment. The Quarterly Journal of Economics, 132(1), 435– 483. European Commission. European Neighbourhood Policy and Enlargement Negotiations (DG NEAR). https://neighbourhood-enlargement.ec.europa.eu/enlargement-policy/overview- instrument-pre-accession-assistance_en Accessed 2023-05-07. European Commission. EU spending and revenue 2004-2020. https://commission.europa.eu/strategy-and-policy/eu-budget/long-term-eu-budget/2014- 2020/spending-and-revenue_en Updated: 2022-09-28. Accessed 2023-03-01 46 European Commission. (2019). Standard Eurobarometer 92 – Autumn 2019. Public opinion in the European Union, First results. https://europa.eu/eurobarometer/surveys/detail/2255 Accessed: 2023-04-05. European Commission. (2017). “Standard Eurobarometer 88.” Autumn 2017. https://europa.eu/eurobarometer/surveys/detail/2143 Accessed: 2023-04-05. European Parliament. (2019). European election results, https://www.europarl.europa.eu/election-results-2019/en/national-results-overview/ Updated:2023-05-09. Accessed: 2023-05-09 European Union, Conditions for membership. https://neighbourhood- enlargement.ec.europa.eu/enlargement-policy/conditions-membership_en Accessed 2023-04- 18. European Union. Eurobarometer. https://europa.eu/eurobarometer/surveys/browse/all/series/4961 Accessed: 2023-02-10. European Union. (n.d.1). Joining the EU. https://european-union.europa.eu/principles- countries-history/joining-eu_en Accessed 2023-04-18. European Union. (n.d.2). Eurobarometer - About Eurobarometer. https://europa.eu/eurobarometer/about/eurobarometer Accessed: 2023-04-05. European Union. (2011). Frequently asked questions: The end of transitional arrangements for the free movement of workers on 30 April 2011. https://ec.europa.eu/commission/presscorner/detail/en/MEMO_11_259 MEMO/11/259 Brussels, 28 April 2011. Accessed 2023-04-18. Eurostat, The Statistical Office of the European Union. Asylum applicants by type of applicant, citizenship, age and sex - annual aggregated datahttps://ec.europa.eu/eurostat/databrowser/view/MIGR_ASYAPPCTZA__custom_497443 2/default/table Updated: 2023-01-16. Accessed:2023-02-10. Eurostat, The Statistical Office of the European Union. Asylum applicants by citizenship till 2007 - annual data https://ec.europa.eu/eurostat/databrowser/view/MIGR_ASYCTZ__custom_5015314/default/t able Updated: 2022-04-07. Accessed: 2023-02-17. Eurostat, The Statistical Office of the European Union. Gross domestic product (GDP) at market prices - annual data https://ec.europa.eu/eurostat/databrowser/view/TIPSAU10/default/table Updated: 2023-02- 02. Accessed: 2023-02-18. Eurostat, The Statistical Office of the European Union. Immigration by age group, sex and citizenship https://ec.europa.eu/eurostat/databrowser/view/MIGR_IMM1CTZ__custom_4904073/default /table?lang=en Updated: 2023-01-28. Accessed: 2023-02-10. Eurostat, The Statistical Office of the European Union. Immigration (migr_immi), Metadata in Euro SDMX Metadata Structure (ESMS) https://ec.europa.eu/eurostat/cache/metadata/en/migr_immi_esms.htm Updated: 2022-06-01. Accessed: 2023-04-20. Eurostat, The Statistical Office of the European Union. Population by sex, age, citizenship and labour status (1 000) https://ec.europa.eu/eurostat/databrowser/view/LFSQ_PGANWS__custom_5030006/default/t able Updated: 2023-01-13. Accessed: 2023-02-15 Eurostat, The Statistical Office of the European Union. Population on 1 January by age and sex. https://ec.europa.eu/eurostat/databrowser/view/DEMO_PJAN/default/table?lang=en&categor y=demo.demo_pop Updated: 2023-01-28. Accessed: 2023-02-10. Gorodzeisky, A., & Semyonov, M. (2020). Perceptions and misperceptions: actual size, 47 perceived size and opposition to immigration in European societies. Journal of Ethnic and Migration studies. 46(3). 612–630. Hanson, G.H., Scheve, K.F., & Slaughter, M.J. (2007). Public finance and individual preferences over globalization strategies. Economics & Politics. 19(1). 1–33. Hainmueller, J., & Hopkins, D. (2014). Public attitudes toward immigration. Annual Review of Political Science, 17(1), 225–249. Hainmueller, J., Hiscox, M., & Margalit, Y. (2015). Do concerns about labor market competition shape attitudes toward immigration? New evidence. Journal of International Economics, 97(1), 193–207. Hainmueller, J., Hiscox, M. J. (2007). Educated Preferences: Explaining Attitudes Toward Immigration in Europe. New York, USA: Cambridge University Press. International organization, 61(2), 399–442. Hainmueller, J., & Hiscox, M. J. (2010). Attitudes toward Highly Skilled and Low-Skilled Immigration. American Political Science Review, 104(1), 61–84. Holland, D., Fic, T., Rincon-Aznar, A., Stokes, L., & Paluchowski, P. (2011). Labour mobility within the EU-The impact of enlargement and the functioning of the transitional arrangements. National Institute of Economic and Social Research, 2978. Iacus, S. M., & Porro, G. (2015). EU regional unemployment as a transnational matter: An analysis via the Gompertz diffusion process. Papers in Regional Science, 94(4), 703–726. Kahanec, M., Zaiceva, A., & Zimmermann, K. F. (2009). Lessons from migration after EU enlargement. In EU labor markets after post-enlargement migration (pp. 3–45). Berlin, Heidelberg: Springer Berlin Heidelberg. Karemera, D., Oguledo, V., & Davis, B. (2000). A gravity model analysis of international migration to North America. Applied Economics, 32(13), 1745–1755. Lecheler, S., & De Vreese, C. (2010). Framing Serbia: The effects of news framing on public support for EU enlargement. European Political Science Review, 2(1), 73–93. Lim, S. S. (2019). Migration decisions under ambiguity: Migration distance effects revisited. Review of Development Economics, 23(4), 1520–1539. Llull, J. (2018). The effect of immigration on wages. The Journal of Human Resources, 53(3), 608–662. Lucas, R. (2001). The effects of proximity and transportation on developing country population migrations. Journal of Economic Geography, 1(3), 323–339. Mayer, T., & Zignago, S. (2011). Notes on CEPII’s Distances Measures: The GeoDist Database. CEPII Working Paper, 2011(25). Mayda, A. (2006). Who Is against Immigration? A Cross-Country Investigation of Individual Attitudes toward Immigrants. The Review of Economics and Statistics, 88(3), 510–530. Meltzer, C. E., Eberl, J. M., Theorin, N., Lind, F., Schemer, C., Boomgaarden, H. G., Strömbäck, J. & Heidenreich, T. (2018). Perceptions of the impact of immigration and attitudes towards free movement within the EU: A cross-national study. Working Paper Compass/Reminder Oxford; 2018 Mitsch, F., Lee, N., & Morrow, E. R. (2021). Faith no more? The divergence of political trust between urban and rural Europe. Political geography, 89, 102426. Moritz, M. (2011). The Impact of Czech Commuters on the German Labour Market. Prague Economic Papers, 2011(1), 40–58. Murray, M. (2006). Avoiding Invalid Instruments and Coping with Weak Instruments. The Journal of Economic Perspectives, 20(4), 111–132. Noutcheva, G., & Bechev, D. (2008). The Successful Laggards: Bulgaria and Romania's Accession to the EU. Los Angeles, CA: SAGE Publications East European politics and societies, 22(1), 114–144. OECD data. Unemployment rate. https://data.oecd.org/unemp/unemployment-rate.htm 48 Published in 2022. Accessed 2023-04-20 Pedersen, P. J., & Pytlikova, M. (2008). EU Enlargement: Migration Flows from Central and Eastern Europe Into the Nordic Countries: Exploiting a Natural Experiment. Department of Economics, Aarhus School of Business. Peshkopia, R. (2020). An ally to the people: EU membership conditionality and Albanians' attitudes toward EU membership. European Societies, 22(2), 266–289. Poprawe, M. (2015). On the relationship between corruption and migration: empirical evidence from a gravity model of migration. Public Choice, 163(3-4), 337–354. Quillian, L. (1995). Prejudice as a Response to Perceived Group Threat. American Sociological Review, 60(4), 586. Ringlerova, Z. (2022). The Impact of Immigration on Attitudes toward the EU: Evidence from a Three‐Country Survey Experiment. Journal of Common Market Studies, 60(2), 391– 407. Roman, M. (2019). Work and welfare take-up of enlargement migrants in the United Kingdom. Applied Economics Letters, 26(5), 341–344. Scheve, K., & Slaughter, M. (2001). Labor Market Competition and Individual Preferences over Immigration Policy. The Review of Economics and Statistics, 83(1), 133–145. Schmuck, D., & Matthes, J. (2017). Effects of Economic and Symbolic Threat Appeals in Right-Wing Populist Advertising on Anti-Immigrant Attitudes: The Impact of Textual and Visual Appeals. Political Communication, 34(4), 607–626. Schneider, S. L. (2007) Anti-Immigrant Attitudes in Europe. European Sociological Review 24(1), 53–67. Schuck, A., & De Vreese, C. (2006). Between Risk and Opportunity. European Journal of Communication (London), 21(1), 5–32. Sides, J., & Citrin, J. (2007). European Opinion About Immigration. British Journal of Political Science 37(3), 477. Sniderman, P. M., Peri, P., de Figueiredo, R. J. P., & Piazza, T. (2000). The Outsider: Prejudice and Politics in Italy. Princeton University Press. Strelkov, A. (2016). The EU and rule of law promotion in Western Balkans - a new role for candidate states' parliaments. Abingdon: Routledge East European politics, 32(4), 505–524. Toshkov, D., & Kortenska, E. (2015). Does Immigration Undermine Public Support for Integration in the European Union? Journal of Common Market Studies, 53(4), 910–925. Tosun, J., Schaub, S., & Marek, C. (2023). Europeans’ Attitudes Toward the Goals of Agricultural Policy: A Case of Rural–Urban Divide?. Political Studies Review, DOI: 14789299221149505. Usherwood, S., & Startin, N. (2013). Euroscepticism as a Persistent Phenomenon. Journal of Common Market Studies, 51(1), 1–16. UNHCR. (2021). Syria Refugee Crisis – Globally, in Europe and in Cyprus. https://www.unhcr.org/cy/2021/03/18/syria-refugee-crisis-globally-in-europe-and-in-cyprus- meet-some-syrian-refugees-in-cyprus/ Updated: 2023-04-17. Accessed: 2023-04-20. UNSD, Countries or areas/geographical regions. https://unstats.un.org/unsd/methodology/m49/ Accessed: 2023-03-01. Valentino, N., Soroka, S., Iyengar, S., Aalberg, T., Duch, R., Fraile, M., . . . Kobayashi, T. (2019). Economic and Cultural Drivers of Immigrant Support Worldwide. British Journal of Political Science, 49(4), 1201–1226. Verbeek, M. (2017). A guide to modern econometrics (5.th ed.). John Wiley & Sons. Yeung, E. (2021). Does immigration boost public Euroscepticism in European Union member states? European Union Politics, 22(4), 631–654. 49 Appendix I. Description of variables and data sources This section includes a table with descriptions of the variables used in the empirical analysis in the paper and where the data for each variable was collected from. The list contains all variables that were used to construct the variables used in the final analysis. Table A1: Variable list Variable Tag Years Description Source Unit EU image - Very Attitudes 2004- Q: In general, does the EU conjure Eurobarometer Percent positive 2019 up for you a very positive, fairly positive, neutral, fairly negative or very negative image? A: Very positive EU image - Fairly Attitudes 2004- Q: In general, does the EU conjure Eurobarometer Percent positive 2019 up for you a very positive, fairly positive, neutral, fairly negative or very negative image? A: Fairly positive EU image - Neutral Attitudes 2004- Q: In general, does the EU conjure Eurobarometer Percent 2019 up for you a very positive, fairly positive, neutral, fairly negative or very negative image? A: Neutral EU image - Fairly Attitudes 2004- Q: In general, does the EU conjure Eurobarometer Percent negative 2019 up for you a very positive, fairly positive, neutral, fairly negative or very negative image? A: Fairly negative EU image - Very Attitudes 2004- Q: In general, does the EU conjure Eurobarometer Percent negative 2019 up for you a very positive, fairly positive, neutral, fairly negative or very negative image? A: Very negative Average EU image Attitudes 2004- Average EU image for each Eurobarometer Ordered 2019 country. Average image = 1 * Very scale negative + 2 * Fairly negative + 3 (1-5) * Neutral + 4 * Fairly positive + 5 * Very positive For EU enlargement Attitudes 2004- Q: What is your opinion on each of Eurobarometer Percent 2019 the following statements? Please tell me for each statement, whether you are for it or against it: Further enlargement of the EU to include other countries in future years. A: For EU27 (2007–2012) Migration 2004- Migration from EU27 countries Eurostat Persons Migration 2012 except reporting country EU28 (2013–2020) Migration 2004- Migration from EU28 countries Eurostat Persons Migration 2019 except reporting country Croatia Migration Migration 2013- Migration from Croatia except Eurostat Persons 50 2019 reporting country Female Demographic 2004- The share of female respondents in Eurobarometer Percent 2019 the survey for each country Age Demographic 2004- The average age for the Eurobarometer Years 2019 respondents in the survey for each country Educational Demographic 2004- The average educational Eurobarometer Years attainment 2019 attainment for the respondents in the survey for each country Living in rural area Demographic 2004- Q: Would you say you live in a...? Eurobarometer Percent 2019 A: Rural area or village Population Economic 2004- Population in the country Eurostat Persons 2019 GDP Economic 2004- Main GDP aggregates per capita: Eurostat Euro 2019 Gross domestic product at market prices GDP per capita Economic 2004- Main GDP aggregates per capita: Eurostat Euro 2019 Gross domestic product at market prices/Population Unemployed persons Economic 2004- Number of unemployed persons in Eurostat Persons 2019 each country Labour population Economic 2004- Number of persons in the labour Eurostat Persons 2019 population in each country Unemployment rate Economic 2004- Number of unemployed persons in Eurostat Percent 2019 each country/number of persons in the labour force in each country Asylum applicants Economic 2004- Number of Asylum applicants in a Eurostat Persons 2007 country Asylum applicants Economic 2008- Number of Asylum applicants in a Eurostat Persons 2019 country Founding member of Interaction Time 1 if country were the founding European Union Binary EU invariant member of the EU, 0 if not EU budget netto Interaction 2004- 1 if country contributed more to the Eurostat Binary contribution 2019 EU budget than they received (netto), 0 if not Eastern Europe Interaction Time 1 if country are in Eastern Europe, European Union Binary invariant 0 if not Distance to Distance Time Geodesic distance between the CEPII Km Bucharest invariant capital i EU-country i and Bucharest Distance to Sofia Distance Time Geodesic distance between the CEPII Km invariant capital i EU-country i and Sofia Average distance to Distance Time Average geodesic distance CEPII Km Bucharest & Sofia invariant between the capital i EU-country i and Bucharest & Sofia Open border Time dummy 2004- 1 when a country opened their European Union Binary 2019 borders towards Bulgaria and Romania, 0 if not open. 51 Average distance Instrument 2004- The instrument captures the Km Bucharest & Sofia x 2019 mitigating effect distance has on open border the opening of borders to Bulgaria and Romania on the inflow of migration from EU countries for country i. II. Regression tables & figures This section contains the regression outputs for all the specifications used in the robustness- and sensitivity checks section of the paper (section 5.4). The section also includes excluded regression outputs for the main analysis; the reduced form results of the 2SLS and the results of the interaction IV-models for the EU-attitudes EU trust and Very negative EU image. Finally, the section also includes the excluded figures of the changes in geographical distribution of the EU-attitudes studies in the paper between 2004 and 2019. Tables: 52 A) 2SLS Reduced form Dependent variable: 1. Average EU Image 2. Very Negative EU Image 3. EU Trust 4. For EU Enlargement (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) Instrument (open border x distance) 0.007** 0.010*** 0.009*** -0.235* -0.306*** -0.294*** 0.424*** 0.493*** 0.477*** 0.436*** 0.438*** 0.418*** (0.004) (0.003) (0.002) (0.121) (0.094) (0.087) (0.163) (0.133) (0.116) (0.104) (0.105) (0.105) Open border -0.128** -0.141*** -0.150*** 4.599** 5.206*** 5.500*** -5.505** -6.112*** -6.542*** -3.051* -2.869 -2.699 (0.062) (0.048) (0.048) (2.201) (1.728) (1.717) (2.773) (2.294) (2.178) (1.846) (1.881) (1.954) Economic controls No Yes Yes No Yes Yes No Yes Yes No Yes Yes Demographic controls No No Yes No No Yes No No Yes No No Yes Country & Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Constant 3.112*** -0.853 -0.838 26.577*** 194.408*** 182.518*** 49.274*** -135.956*** -103.981* 36.821*** -16.964 -75.238 (0.053) (0.698) (1.125) (1.705) (23.432) (35.989) (1.992) (31.984) (57.309) (1.850) (30.778) (47.459) R2 0.681 0.765 0.787 0.729 0.797 0.812 0.686 0.741 0.768 0.868 0.874 0.876 N 400 398 398 400 398 398 400 398 398 400 398 398 Notes: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01, Instrument refers to an interaction between the opening of the border and the average distance to Bulgaria and Romania. Open border is constructed as a dummy-variable that takes the value of 1 when the border for country i is opened. Economic controls and Demographic controls consist of two sets of covariates, for full list see section 4.1. Economic controls are standardized. 53 B) Interactions EU Trust Dependent variable: EU Trust (1) (2) (3) (4) (5) (6) (7) (8) (9) Intra-EU Migration -8.724* -11.637** -9.630** -4.884 -10.154* -7.445 -10.291 -9.796 -8.719 (4.932) (4.565) (3.872) (4.728) (5.991) (4.701) (8.780) (6.184) (6.252) Open border 2.899 3.633* 2.936 -1.189 -1.566 -3.006 1.878 3.398** 2.947* (2.171) (2.062) (2.052) (2.128) (2.331) (2.039) (1.743) (1.488) (1.644) Founding members x Intra-EU Migration 4.473 11.182* 9.528 (6.956) (6.230) (6.037) Founding members x Open border -3.573 -2.470 -0.898 (3.128) (2.442) (2.425) EU budget contributors x Intra-EU Migration 1.253 0.316 0.070 (0.990) (0.936) (0.736) EU budget contributors x Open border 4.207 7.106** 8.766*** (3.099) (3.237) (2.723) Eastern Europe x Intra-EU Migration 7.936 3.710 3.291 (8.673) (6.143) (6.262) Eastern Europe x Open border -5.404 -8.494** -6.783* (3.974) (3.888) (3.752) Economic controls No Yes Yes No Yes Yes No Yes Yes Demographic controls No No Yes No No Yes No No Yes Country & Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Constant 135.299*** -159.956** -193.129* 80.898* -196.690*** -237.765*** 152.043* -128.822** -133.463 (51.036) (66.014) (102.523) (49.174) (71.948) (89.133) (91.270) (59.845) (97.992) R2 0.466 0.451 0.575 0.624 0.519 0.667 0.487 0.571 0.637 F-statistic excluded instrument 9.221 13.860 15.114 3.973 3.947 3.892 6.306 6.149 5.860 F-statistic excluded instrument x interaction 5.592 6.521 7.547 5.232 4.671 5.847 3.347 3.152 4.326 N 355 353 353 355 353 353 355 353 353 Notes: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01, Intra-EU Migration refers to the log of migration from EU27 countries to country i. The instrument used is an interaction between the opening of the border and the average distance to Bulgaria and Romania. Open border is constructed as a dummy-variable that takes the value of 1 when the border for country i is opened. The instrument and open border are augmented with the interaction dummy in each setup. Economic controls and Demographic controls consist of two sets of covariates, for full list see section 4.1. Economic controls are standardized. F-statistic for excluded instruments obtained through estat firststage. 54 C) Interactions Very Negative EU Image Dependent variable: Very Negative EU Image (1) (2) (3) (4) (5) (6) (7) (8) (9) Intra-EU Migration 2.314* 3.317*** 3.006*** 1.475 3.610** 3.034** 3.247 3.197** 3.055* (1.352) (1.226) (1.118) (1.354) (1.531) (1.329) (2.521) (1.601) (1.704) Open border -0.967* -1.352** -1.221** 0.062 -0.034 0.242 -0.054 -0.642 -0.560 (0.572) (0.528) (0.567) (0.736) (0.815) (0.665) (0.535) (0.417) (0.477) Founding members x Intra-EU Migration -1.702 -4.275** -4.041** (1.592) (1.829) (1.810) Founding members x Open border 2.514*** 2.354*** 2.190*** (0.697) (0.641) (0.652) EU budget contributors x Intra-EU Migration -0.485 -0.159 -0.119 (0.313) (0.318) (0.268) EU budget contributors x Open border 0.236 -0.622 -0.936 (1.272) (1.409) (1.217) Eastern Europe x Intra-EU Migration -2.720 -1.269 -1.217 (2.444) (1.608) (1.736) Eastern Europe x Open border 1.312 2.480*** 2.127*** (0.903) (0.825) (0.817) Economic controls No Yes Yes No Yes Yes No Yes Yes Demographic controls No No Yes No No Yes No No Yes Country & Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Constant -17.448 77.897*** 74.747*** -3.516 88.990*** 95.883*** -27.270 76.610*** 74.940*** (14.074) (19.971) (27.110) (14.332) (25.294) (27.985) (26.316) (18.619) (25.994) R2 0.603 0.608 0.654 0.674 0.555 0.644 0.575 0.667 0.690 F-statistic excluded instrument 9.221 13.860 15.114 3.973 3.947 3.892 6.306 6.149 5.860 F-statistic excluded instrument x interaction 5.592 6.521 7.547 5.232 4.671 5.847 3.347 3.152 4.326 N 355 353 353 355 353 353 355 353 353 Notes: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01, Intra-EU Migration refers to the log of migration from EU27 countries to country i. The instrument used is an interaction between the opening of the border and the average distance to Bulgaria and Romania. Open border is constructed as a dummy-variable that takes the value of 1 when the border for country i is opened. The instrument and open border are augmented with the interaction dummy in each setup. Economic controls and Demographic controls consist of two sets of covariates, for full list see section 4.1. Economic controls are standardized. F-statistic for excluded instruments obtained through estat firststage. 55 D) IV with different independent variable Dependent variable: (1) (2) (3) (4) (1) (2) (3) (4) (1) (2) (3) (4) (1) (2) (3) (4) 1.1 2.1 3.1 4.1 1.3 2.3 3.3 4.3 1.4 2.4 3.4 4.4 1.5 2.5 3.5 4.5 Bulgaria & Romania -0.001 0.059 -0.202 -0.027 (0.003) (0.046) (0.213) (0.108) EU27 & EU28 -0.145* 3.658*** -11.036** -9.204** (0.081) (1.375) (4.870) (3.813) Europe -0.045 2.072 -9.015 0.422 (0.091) (1.706) (7.713) (3.927) EU27 proxy -0.387 8.264 -21.886 -32.699* (0.329) (5.307) (14.851) (19.438) Open border -0.036 1.2029* -2.560 0.816 0.023 -0.380 2.380 5.542*** -0.013 0.569 -0.792 2.285 -0.028 0.849 -0.169 1.607 (0.052) (0.677) (2.956) (1.823) (0.031) (0.543) (1.825) (1.360) (0.040) (0.457) (2.254) (1.360) (0.046) (0.727) (2.180) (2.700) Constant 2.9809* 16.28 20.64 -18.62 -1.842 88.689*** -174.151* -198.508** 1.428 50.487* -89.528 -78.803 3.552 -29.737 95.324 204.927 (1.727) (19.74) (122.3) (61.11) (1.697) (28.342) (101.569) (77.217) (1.631) (29.603) (144.876) (72.653) (3.255) (51.533) (160.274) (194.729) R2 0.728 0.616 0.580 0.889 0.689 0.563 0.502 0.777 0.751 0.659 0.528 0.891 0.482 0.119 0.412 0.468 F-statistic 0.760 0.760 0.760 0.760 8.185 8.185 8.185 8.185 1.446 1.446 1.446 1.446 2.940 2.940 2.940 2.940 N 249 249 249 249 353 353 353 353 245 245 245 245 334 334 334 334 Notes: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01, for the dependent variable (1) Refers to Average EU Image, (2) Refers to Very Negative EU Image, (3) Referts to EU Trust, (4) Refers to For EU Enlargement when Economic and Demographic controls, as well as Country and Time fixed effects, are applied. Economic controls are standardized. The Bulgaria & Romania migration variable is measured in 10,000 migrants, while the EU27 & EU28 migration, Europe migration and EU27 with proxy migration refers to the log of each variable. Open border is constructed as a dummy-variable that takes the value of 1 when the border for country i is opened. F-statistic for excluded instruments obtained through estat firststage. 56 E) First stage migration from Bulgaria & Romania Dependent variable: Bulgaria & Romania migration (1) (2) (3) Instrument (open border x distance) -2.020* -1.436 -1.438 (1.132) (1.054) (0.970) Open border 15.298 11.697 8.038 (14.501) (13.343) (12.685) Economic controls No Yes Yes Demographic controls No No Yes Country & Time Fixed Effects Yes Yes Yes Constant 2.557 -185.759 249.648 (6.326) (118.196) (226.410) R2 0.704 0.727 0.760 F-statistic 3.183 1.855 2,200 N 250 249 249 Notes: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01, The migration variable is upscaled: 1 unit = 1000 migrants. Instrument refers to an interaction between the opening of the border and the average distance to Bulgaria and Romania. Open border is constructed as a dummy-variable that takes the value of 1 when the border for country i is opened. Economic controls and Demographic controls consist of two sets of covariates, for full list see section 4.1. Economic controls are standardized. F-statistic for excluded instrument obtained through estat firststage. 57 F) First stage Migration EU27 (not logged) Dependent variable: Migration EU27 (1) (2) (3) Instrument (open border x distance) -0.244** -0.205** -0.210** (0.102) (0.087) (0.083) Open border 4.358*** 4.351*** 3.676*** (1.633) (1.455) (1.391) Economic controls No Yes Yes Demographic controls No No Yes Country & Time Fixed Effects Yes Yes Yes Constant 4.709*** -3.810 45.415* (0.755) (12.882) (24.179) R2 0.892 0.903 0.908 F-statistic 5.659 5.592 6.402 N 355 353 353 Notes: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01, Migration EU27 refers to the migration from EU27 countries to country i. Instrument refers to an interaction between the opening of the border and the average distance to Bulgaria and Romania. Open border is constructed as a dummy-variable that takes the value of 1 when the border for country i is opened. Economic controls and Demographic controls consist of two sets of covariates, for full list see section 4.1. Economic controls are standardized. F-statistic for excluded instrument obtained through estat firststage. 58 G) IV Migration EU27 (not logged) Dependent variable: 1. Average EU Image 2. Very Negative EU Image 3. Trust EU 4. For Enlargement (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) Intra-EU Migration -0.020 -0.035* -0.026 0.476 0.731** 0.664** -1.716 -2.368* -2.004*** -1.591** -1.961** -1.671** (0.021) (0.021) (0.017) (0.314) (0.343) (0.289) (1.222) (1.323) (1.063) (0.774) (0.933) (0.785) Open border 0.018 0.070 0.024 -0.295 -1.131 -0.387 2.649 4.938* 2.403 5.629*** 7.043*** 5.559*** (0.036) (0.043) (0.034) (0.586) (0.754) (0.607) (2.201) (2.799) (2.131) (1.716) (2.102) (1.605) Economic controls No Yes Yes No Yes Yes No Yes Yes No Yes Yes Demographic controls No No Yes No No Yes No No Yes No No Yes Country & Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Constant 3.145*** -0.285 0.467 4.039** 67.599*** 29.978 54.236*** -101.328* 2.971 43.537*** -27.823 -50.782 (0.108) (0.996) (1.705) (1.693) (15.951) (24.870) (6.503) (55.393) (108.549) (4.533) (45.212) (64.199) R2 0.618 0.592 0.697 0.590 0.546 0.618 0.425 0.347 0.497 0.761 0.741 0.787 F-statistic excluded instrument 5.659 5.592 6.402 5.659 5.592 6.402 5.659 5.592 6.402 5.659 5.592 6.402 N 355 353 353 355 353 353 355 353 353 355 353 353 Notes: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01, Intra-EU Migration refers to migration from EU27 countries to country i. The instrument used is an interaction between the opening of the border and the average distance to Bulgaria and Romania. Open border is constructed as a dummy-variable that takes the value of 1 when the border for country i is opened. Economic controls and Demographic controls consist of two sets of covariates, for full list see section 4.1. Economic controls are standardized. F-statistic excluded instrument obtained through estat firststage. 59 H) IV with dependent variable forwarded one year Dependent variable: 1. Average EU Image 2. Very Negative EU Image 3. Trust EU 4. For Enlargement (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) Intra-EU Migration -0.196* -0.313** -0.240** 2.825* 4.614*** 3.785** -9.558* -14.693** -11.795** -6.266* -8.465* -6.576 (0.118) (0.124) (0.102) (1.554) (1.730) (1.478) (5.784) (6.414) (5.314) (3.291) (4.708) (4.099) Open border -0.015 0.017 0.009 0.225 -0.191 -0.136 0.269 1.513 1.435 3.403*** 3.731*** 4.135*** (0.039) (0.041) (0.038) (0.531) (0.572) (0.568) (1.822) (1.989) (1.951) (1.313) (1.401) (1.385) Economic controls No Yes Yes No Yes Yes No Yes Yes No Yes Yes Demographic controls No No Yes No No Yes No No Yes No No Yes Country & Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Constant 4.940*** -0.959 -2.685 -21.625 75.259*** 99.664*** 137.989** -159.061* -194.921* 97.140*** -64.703 -245.588*** (1.217) (1.769) (2.014) (16.038) (25.740) (29.926) (59.525) (84.028) (101.496) (34.281) (58.593) (71.354) R2 0.458 0.280 0.509 0.572 0.422 0.565 0.419 0.241 0.447 0.801 0.775 0.826 N 330 328 328 330 328 328 330 328 328 330 328 328 Notes: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01, Intra-EU migration refers to the log of migration from EU27 countries to country i. The instrument used is an interaction between the opening of the border and the average distance to Bulgaria and Romania. Open border is constructed as a dummy-variable that takes the value of 1 when the border for country i is opened. Economic controls and Demographic controls consist of two sets of covariates, for full list see section 4.1. Economic controls are standardized 60 I) Average distance from Bulgaria and Romania regressed on EU attitudes for each year in the sample Regressor: average distance to Bucharest and Sofia Year: 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Dependent variable: EU Image 0.005 0.001 -0.006 -0.004 -0.004 -0.003 -0.009 -0.009 -0.010 0.000 0.003 0.010 0.012 0.013* 0.016** 0.010 (0.010) (0.009) (0.010) (0.009) (0.007) (0.007) (0.007) (0.007) (0.009) (0.009) (0.008) (0.007) (0.008) (0.007) (0.007) (0.007) Very Negative EU Image -0.030 0.041 0.058 0.050 -0.001 -0.035 0.014 -0.024 -0.070 -0.132 -0.153 -0.247* -0.220 -0.202 -0.150 -0.060 (0.082) (0.100) (0.090) (0.110) (0.083) (0.078) (0.123) (0.151) (0.190) (0.172) (0.137) (0.101) (0.142) (0.138) (0.123) (0.101) EU Trust -0.242 -0.390 -0.619 -0.357 -0.340 -0.238 -0.472 -0.513* -0.336 -0.387 -0.119 0.243 0.467 0.477 0.472 0.361 (0.345) (0.420) (0.393) (0.329) (0.296) (0.362) (0.385) (0.295) (0.344) (0.365) (0.330) (0.313) (0.283) (0.294) (0.315) (0.310) For EU Enlargement -0.671 -0.900* -1.031** -0.609 -0.612 -0.457 -0.939** -0.769* -0.449 -0.461 -0.279 -0.239 -0.129 -0.349 -0.0417 -0.468 (0.406) (0.454) (0.444) (0.401) (0.383) (0.462) (0.444) (0.389) (0.354) (0.370) (0.334) (0.369) (0.378) (0.415) (0.431) (0.501) N 25 25 25 25 25 25 25 25 25 25 25 25 25 25 25 25 Notes: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01 61 J) The effect of open border on EU-attitudes Dependent variable: 1. Average EU Image 2. Very Negative EU Image 3. Trust EU 4. For Enlargement (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) Open border -0.015 0.006 -0.010 0.995 0.526 0.993 1.001 1.441 0.777 3.648*** 3.839*** 3.712*** (0.029) (0.024) (0.026) (0.979) (0.846) (0.919) (1.299) (1.187) (1.251) (0.958) (0.946) (1.003) Economic controls No Yes Yes No Yes Yes No Yes Yes No Yes Yes Demographic controls No No Yes No No Yes No No Yes No No Yes Country & Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Constant 3.096*** -0.716 -0.924 27.069*** 190.066*** 185.305*** 48.385** -128.948*** -108.507* 35.905*** -10.740 -79.202 (0.051) (0.754) (1.163) (1.655) (24.817) (36.674) (1.918) (34.499) (59.083) (1.640) (31.536) (48.383) R2 0.676 0.755 0.778 0.724 0.790 0.805 0.676 0.729 0.757 0.861 0.867 0.870 N 400 398 398 400 398 398 400 398 398 400 398 398 Notes: Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01, Open border is constructed as a dummy-variable that takes the value of 1 when the border for a given country is opened. Economic controls and Demographic controls consist of two sets of covariates, for full list see section 4.1. Economic controls are standardized 62 Figures: A) For Enlargement of the EU 2004 B) For Enlargement of the EU 2019 C) Average EU Image 2004 D) Average EU Image 2019 E) Very Negative EU Image 2004 F) Very Negative EU Image 2019 63