The Migration-Welfare Nexus

Transcription

1 This paper was prepared for the 11 th European Consortium for Political Research (ECPR) General Conference, 6-9 September 2017, Universitetet i Oslo, Norway. The Migration-Welfare Nexus To what extent does immigration influence national welfare state generosity in Europe? Clare Fenwick Department of Economics, Leiden University Abstract The question of whether immigration undermines native support for welfare provision has received considerable attention in the academic literature. Remarkably, only a few studies have explored if changes in support are echoed by changes in policy, such as the retrenchment of the welfare state. This study explores whether immigration plays an influential role in the determination of welfare state generosity. It investigates the relationship between stocks of migrants, the foreign-born population, on two different indicators for welfare generosity - social welfare spending as a percentage of GDP, as is common convention in welfare state literature, and a welfare generosity index. The results show that the foreign-born population has a positive and statistically significant effect on social welfare spending, but no effect on the welfare generosity index. Crucially, the findings provide no evidence to support the hypothesis that higher levels of immigration lead to reduced levels of social welfare provision. On the contrary, immigration may lead to welfare state expansion rather than retrenchment. Key words: Political economy; welfare state; government spending; immigration; Europe Acknowledgements I would like to thank Olaf van Vliet, Kees Goudswaard, Koen Caminada and Alexandre Afonso for their thoughtful and insightful feedback and suggestions on previous versions of this paper. Additionally, this study is a part of SOLID: Solidarity Under Strain and is funded by Interaction between Legal Systems (ILS 2.0).

2 Introduction Milton Friedman famously once said You cannot simultaneously have free immigration and a welfare state (Friedman, 1999). Indeed, it is to be expected that increasing immigrant inflows can present new difficulties and challenges for the welfare state and solidarity among citizens. Some authors go as far to argue that increasing immigration in Europe will eventually lead to the Americanisation of European welfare states and politics (Alesina et al., 2001, 2004; Freeman, 1986). The welfare state can be understood as a mechanism, or a social arrangement, which enables a population to deal with collective risks and reduce social inequality. In Europe, it is something that has been closely linked with the development of nation-states and has contributed to the forging of bonds between citizens. The welfare state does, however, restrict rights and benefits to insiders and thus exclude outsiders. As a result, increasing immigration becomes a challenge for modern social security institutions built on inclusion, integration and homogeneity (Mau & Burkhardt, 2009). Consequently, immigration can expose the tensions between the inherently closed system of the welfare state and the relatively open economies of developed nations. This led Freeman to conclude that, ultimately, national welfare states cannot coexist with the free movement of labour (1986). This issue is especially salient in light of the fact that immigration is consistently one of the most pressing issues for citizens and currently forms a central pillar in electoral campaigns in Europe (Afonso & Devitt, 2016). In particular, Western European countries tend to have well established welfare states, but have also been dealing with large-scale migration for several decades and migration discourse is often fuelled with controversy. As Burgoon et al. wrote public opinion regarding the economic and cultural impact of immigrants tends to be negative (2012). Crucially, some authors predict that weakening solidarity due to increasing ethnic diversity will undermine the welfare state (for a survey of the literature see Stichnoth & Van der Straeten, 2013). Despite this, Castles & Schierup (2010) wrote that immigration and growing ethnic diversity are important but often neglected factors in the evolution of welfare systems in Europe. In globalisation literature, most authors investigate the impact of trade and capital on the welfare state but ignore its third facet, the movement of people. This paper aims to provide insight into the migration-welfare nexus through investigating whether or not immigration plays an influential role in the determination of welfare state generosity in Europe. Earlier quantitative research is mixed, it is not clear to what extent immigration impacts welfare generosity as most researchers use only social welfare spending as a proxy for generosity (Gaston & Rajaguru, 2013; Lipsmeyer & Zhu, 2011; Soroka et al., 2006; Soroka et al., 2016). Starke (2006) argues that studies researching welfare policy change should complement expenditure data with additional quantitative measures. Consequently, this study extends previous research through complementing 2

3 social welfare spending data with a welfare generosity index developed by Scruggs et al. (2004, 2014). Moreover, the analysis includes the years following EU expansion and so hopes to shed light on the speculations and predictions that EU enlargement would have negative consequences for the European welfare state. My empirical findings suggest that there is no evidence to support the conclusion that increasing immigration is detrimental or incompatible with European welfare states. In the following section I examine the previous literature surrounding immigration and the welfare state. This is followed by a section on the research design, which includes my hypotheses, data and method. Then, I present the results and analysis before finally concluding the paper. Immigration and Welfare in Contemporary Debate National welfare states, whatever their internal principles, exist in a global political economy (Freeman, 1986). The relationship between national welfare states and globalisation is complex, and previous literature has tended to focus on the impact of capital mobility and trade liberalisation rather than on increasing mobility. The research that has been conducted on how immigration impacts the generosity of welfare states is typically split into two competing camps. One side advocates that increasing inflows should lead to the retrenchment of the welfare state (e.g. Alesina & Glaeser, 2004; Alesina et al., 2001; Beine et al., 2015; Burgoon, 2014; Schmidt-Catran & Spies, 2016), while the other proposes that increasing immigration should actually drive its expansion (e.g. Brady & Finnigan, 2014; Finseraas, 2008; Steele, 2016; Walter, 2010). However, political economy theory proposes that in theory empirical evidence should find increasing immigration leads to reduced levels of welfare generosity. Median Voter Theory In 1981, Meltzer and Richard developed a political economy model in order to demonstrate that demand for redistribution by voters is dependent on the level of economic inequality. The model predicts that when the mean income rises relative to the median income, then demand for redistribution will increase and taxes will rise. This is based on the assumption that voters act with economic self-interest, that those with an income lower than the median income choose candidates who favour higher taxes and greater redistribution. Whereas, voters who have an income above the median desire lower taxes and less redistribution (Meltzer & Richard, 1981). Magni-Berton (2014), uses Meltzer and Richard s median voter model to show how immigration can reduce demand for redistribution. First, it is assumed that immigrants to a new country have a lower income than the median voter there. Subsequently, due to an absence of voting rights for immigrants, this means that the income level of the median voter does not change, but the general mean income of 3

4 the entire population does. Consequently, immigrants close the gap between the mean and median income and subsequently reduce support for redistribution. 1 Based on this theory and building on the concept that a higher proportion of immigrants in Europe are considered low-skilled and work in lower-paid jobs (UN-DESA & OECD, 2013), I would expect that immigration has led to reductions in welfare state spending and generosity. This notion is supported by the welfare magnet hypothesis, which also relies on assumptions about the economic self-interest of people. The Welfare Magnet Hypothesis The welfare magnet hypothesis proposes that if location choices made by immigrants are guided by income-maximizing behaviour, then the generosity of a welfare state will act as a considerable pullfactor in those decisions. As a result, states are expected to reduce their welfare effort in order to avoid becoming a magnet for immigrants. The hypothesis finds its roots in George J. Borjas seminal work. Borjas (1999) investigates the location decisions made by immigrants arriving in the United States of America (US) using the 1980 and 1990 Public Use Microdata Sample (PUMS) of the U.S. census. He finds evidence to suggest that immigrant welfare recipients in the US are more heavily clustered in welfare-generous states than natives or immigrants who do not receive welfare benefits. Borjas suggests this is because of migration costs; immigrants have already decided to incur these costs and so choosing the state that offers the highest benefits is costless. Whereas for native welfare recipients, the cost of migration deters them from seeking out welfare-generous states. Borjas concludes then that a relatively generous state becomes a magnet for immigrants and will lead to a very different geographic sorting of welfare recipients in the immigrant and native populations (1999). He does stress, however, that the statistical significance of his results is weak and that there may be alternative explanations for the evidence. To investigate the welfare magnet theory in a European context, De Giorgi and Pellizzari (2009) use data from the European Community Household Panel (ECHP). They find that the relative generosity of a welfare state influences the decisions of migrants. However, the results also show that the number of migrants influenced by the welfare state of a particular country is minor when compared to other reasons for emigrating. Regardless, the authors conclude that their findings may present serious implications for further expansion of the EU. In addition, Razin and Wahba (2015) develop two theoretical models, the first a free-migration regime and the second a restricted-migration regime, and then test them empirically and find support for the welfare-magnet hypothesis. They conclude that in a free-migration regime, a generous welfare state 1 For more political economy models involving immigration, taxes and redistribution see Hansen (2003), Ortega (2004), Felbermayr and Kohler (2007), Nannestad (2007) and Razin and Sadka (2012). 4

5 attracts unskilled migrants while skilled migrants are deterred. Thus, they conclude that in a freemigration regime such as the EU, benefit levels should act as a pull factor within the EU labour market. The welfare magnet hypothesis is crucial in the debate on how migration will impact welfare generosity because in particular, the public and politicians have become increasingly concerned that the welfare systems in Europe will be negatively affected through welfare migration. One such result of this, is the hypothesis that governments may enter into strategic interactions with neighbouring districts, states or countries in a race to provide the lowest levels of welfare to avoid becoming a magnet for poor immigrants. This has been dubbed the Race-to-the-Bottom. In America, Schram and Krueger (1994) and Brueckner (2000) both present evidence to show that strategic interactions between states have taken place, and both papers conclude that it is due to belief in welfare magnets. Likewise in Europe, Kvist (2004) argues that EU 15 member states have been engaging in strategic interactions and, in light of further EU enlargement, concludes that this may intensify in the future. Similarly, Dahlberg and Edmark (2008) investigate whether or not there is a race-to-the-bottom between municipalities in Sweden. They conclude that if a neighbouring municipality reduces welfare spending by 100 SEK, then a municipality reduces their welfare spending by approximately 41 SEK. Although, benefit levels between municipalities or US states are more transparent than benefit levels between countries and may explain the differing evidence on strategic interactions. Interestingly, based on a comparative analysis of the EU-15 and 7 non-eu OECD countries, van Vliet (2010) determines that rather than a social race-to-the-bottom, social expenditures in the EU-15 have converged and increased on average. In addition, using EU-LFS data, Skupnik (2014) find no evidence of a race-to-the-bottom in benefit levels due to the mediating effect of transitional labour market restrictions on new EU members. Since 2014 however, these restrictions have all been lifted, and so the impact of immigration on benefit levels may have changed. However, whether or not a social race-to-the-bottom is taking place, there is a hypothesis that also supports the belief that increasing immigration should lead to reduced levels of welfare generosity, just through a different mechanism. The Anti-Solidarity Hypothesis The individuals who agree to share according to need have to experience a sense of solidarity that comes from common membership in some human community. (Freeman, 1986) Solidarity is considered important for the survival of the welfare state to support this, previous research has found that solidarity is significantly and positively related to support for the welfare state (Burgoon et al., 2012; Crepaz, 2007; Kymlicka & Banting, 2006). Consequently, the anti-solidarity 5

6 hypothesis proposes that increasing racial heterogeneity challenges solidarity and thus undermines support for the welfare state. Alesina at. al. (2001) argue that ethnic diversity is an crucial factor for explaining why the US does not have a welfare state similar to those found in Europe. They find that if the probability of two people drawn at random from a population will belong to different ethnic groups increases by just one percentage point, then social spending as a percentage of GDP is expected to reduce by 7.5 percentage points. Following their research in America, Alesina at. al. (2001) argue that as Europe s heterogeneity increases because of immigration, rising ethnic divisions will be used as a challenge to generous welfare states. Alesina and Glaeser (2004) further their argument through expanding their analysis to 54 countries and directly comparing the US and Europe. They find a negative correlation between racial fractionalisation and social welfare spending. They conclude that European countries, in particular those in Scandinavia, are largely homogenous and have generous levels of welfare state spending. and argue that generous welfare states are contingent on a homogeneous society because solidarity between citizens depends on common linkages, such as culture and language. Alesina and Glaeser (2004) suggest that increasing immigration in Europe, potentially through the expansion of the EU, will challenge Europe s comparatively generous welfare states, as they find it has in the US. Focussing on European OECD Countries, Mau and Burkhardt (2009) believe that the conclusion that migration poses a threat to European welfare states is over-exaggerated. While they find a negative influence of ethnic diversity on support for the welfare state, it is very weak and when controlling for certain factors, such as GDP and unemployment, there is a mediating influence. Mau and Burkhardt use five different measures for ethnic diversity; ethnic fractionalisation, the proportion of foreign population, foreign-born population, non-western foreign-born population and migration inflow (2009). Additionally, Finseraas (2009) finds no evidence for the anti-solidarity hypothesis. Instead, he 2 determines that it is xenophobia which undermines support for the welfare state. He finds evidence to suggest that voters with typically left-wing views on redistribution but right-wing views on immigration tend to follow their immigration preferences at the ballot box. The parties that tend to support tough stances on immigration are usually right-wing, and those same parties typically advocate for reductions in welfare state generosity. 2 There is a large body of research that focuses on prejudice, racism and xenophobia, which investigates people s tendencies to favour an in-group as reasons to why ethnic diversity may reduce support for social welfare and redistribution. This is also related to a strand of literature on trust and the trust of outside groups where there are a variety of empirical, observational and experimental studies (Banting & Kymlicka, 2006; Crepaz, 2007; Nannestad, 2007; van der Meer & Tolsma, 2014). 6

7 Magni-Berton (2014) uses data from the 2008 European Values Survey covering 45 European countries, and finds that immigration reduces support for redistribution primarily through concerns of expected competitiveness on the labour market, which is increased when people believe there is a high number of immigrants, while the impact of native s solidarity with immigrants is comparatively weak. Furthermore, Burgoon (2014) finds that the relationship between stocks of foreign-born and support for redistribution is conditional upon the level of economic integration of immigrants; the less economically integrated immigrants are, the more likely natives are to oppose redistribution. Support for Redistribution: The Compensation Hypothesis The flip-side of the anti-solidarity hypothesis is the compensation hypothesis, which advocates that as immigration is perceived to increase the risk of income loss, then support for redistribution should increase as a result. Proponents argue that governments in open economies expand the welfare state order to insure citizens against the risks posed by globalisation. For example, Finseraas (2008) finds evidence to support the compensation hypothesis. Using the European Social Survey, he shows that individuals who believe immigration lowers average wages are more likely to support higher benefit levels. In Switzerland, Walter (2010) finds differences between globalisation losers and winners. She shows that globalisation losers are more likely to experience feelings of economic insecurity and also support greater expansion of the welfare state. Moreover, Walter s results show that whether a person is a globalisation loser or winner is highly dependent on their skill level, losers typically have lower levels of education than that of winners. Looking at survey data from 17 European countries, Burgoon et al. (2012) find that exposure to increasing immigration at the occupational-level leads to greater support for government redistribution because it can raise individual economic uncertainties. Whereas, at the national level, they find that an increasing foreign-born population has little to no effect on an individual s support for increasing welfare benefits. Migration and Social Policy We should consider how and whether broad public attitudes putatively shaped by immigration actually influence party and policymaking agendas and ultimately revenue and spending policies of states. (Burgoon et al., 2012) Much of the previous literature focuses on the demand-side of welfare - how demand/support for redistribution changes as immigration increases. There is much less research, particularly in Europe, focused on the supply-side of welfare, and whether supplied levels of social welfare social spending and generosity - have increased or decreased as a result of immigration. To test this, Soroka et al. (2006) combine two OECD social spending databases to cover 18 OECD 7

8 countries over the period They investigate the impact of changes in stocks of foreign-born on changes in social welfare spending changes being the current year minus the preceding year. They find that in countries with higher rates of immigration, welfare spending grows less than in countries with limited migration. Notably, they do not find that spending decreases in countries with higher rates of immigration, just that the rate of growth in welfare spending slows. Soroka et al. (2016) build on their previous research through separating social spending into nine different sub-categories. They find that there are different effects in different spending categories and that the areas most affected are those subject to moral hazard or rhetoric about moral hazard. Overall though, they find further support to suggest that increases in immigration lead to smaller increases in social welfare spending. However, the suitability of the model used by Soroka et al. (2016) has been debated in previous literature (Plümper et al., 2005). In a comparative study across 15 European countries from , Lipsmeyer and Zhu (2011) investigate the impact of immigration on unemployment benefits measured as replacement rates - and find that domestic political pressures are more important for explaining variation in unemployment compensation. Gaston and Rajaguru (2013) use data on government social expenditures and migration from the OECD's Social Expenditure database (SOCX) and the Continuous Reporting System on Migration (SOPEMI) database and find no negative relationship between migration and social spending. Instead, they determine that depending on the countries included in the sample, immigration can have a positive effect on social welfare spending. Research Design: Hypothesis The relationship between migration and the welfare state is complex, as evidenced by the mixed results and conclusions in the literature. However, there are two key directions in which this analysis could turn. Based on the theoretical arguments of the median voter, the welfare magnet hypothesis, the antisolidarity hypothesis, and the social race-to-the-bottom, I would hypothesis that increased immigration leads to a reduction in welfare effort. Indeed, Soroka et al. (2006, 1016) find evidence to suggest immigration reduces growth in social welfare spending. However, the compensation hypothesis predicts the opposite result, and following empirical results from Gaston and Rajaguru (2013) and Lipsmeyer and Zhu (2011), which fail to find support for the antisolidarity hypothesis, I could also hypothesise that increased immigration leads to an increase in welfare effort. Consequently, I formulate two hypotheses to examine in my analysis: 8

9 Hypothesis 1: Increased immigration reduces support for the welfare state. This change in demand is reflected in reduced social spending and generosity. Hypothesis 2: Increased immigration increases demand for greater redistribution. This change in demand is reflected in increased social spending and generosity. Research Design: Data To test these hypotheses, I primarily draw on Organisation for Economic Cooperation and Development (OECD) data, with spending from the OECD s Social Expenditure (SOCX) Database (2017b), which covers 26 European countries 3 that fall within the EU and/or the Schengen area, between the years of 1990 and Additionally, I have included the social welfare generosity index from the Comparative Welfare Entitlements Dataset (CWED), developed by Scruggs, et al. (2014), which is available for 16 European countries 4. I also use data on stocks of foreign-born from the OECD (2017a). Data on migration such as stocks of foreign-born or immigration rates are consistently different between various sources. These differences may arise from varying definitions between sources and countries, for example whether or not returning citizens are included in immigration inflow data. Moreover, measurements may be taken at varying times of the year between countries. Data have also been gathered from the Quality of Government Dataset (Teorell et al., 2017), the ICTWSS: Database on Institutional Characteristics of Trade Unions, Wage Setting, State Intervention and Social Pacts in 51 countries between 1960 and 2014 (Visser, 2016), the KOF Globalization Index database (Dreher, 2006), the Comparative Political Data Set (Armingeon et al., 2016) and the World Bank (2017). The Dependent Variable Problem: operationalising social welfare generosity The commonly used proxy indicator for welfare generosity or welfare effort is social spending as a percentage of gross domestic product (GDP)(Allan & Scruggs, 2004). There are clear advantages to this measure; for example, authors believe it provides a good indication of the generosity of a welfare state as a whole, there is no need to correct for inflation and exchanges rates, and it is well recorded so data is readily available for the majority of European countries over an extended period of time. These measures have been criticised and there is a debate in the literature about whether or not this measure is a suitable indicator for depicting welfare generosity (Allan & Scruggs, 2004; Caminada et al., 2010; Clasen & Siegel, 2007; Esping-Anderson, 1990; Green-Pedersen, 2004; Starke, 2006; van Oorschot, 2013; Wang & van Vliet, 2016). 3 Austria, Belgium, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Lithuania, Luxembourg, Netherlands, Norway, Poland, Portugal, Slovakia, Slovenia, Spain, Sweden, Switzerland, United Kingdom. 4 Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Netherlands, Norway, Portugal, Spain, Sweden, Switzerland and United Kingdom 9

10 First, Allan & Scruggs (2004) argue that levels of spending are not directly relevant to the levels of protection provided because changes in the number of beneficiaries, such as higher unemployment and an aging population, can mask changes at the individual level and therefore may not reflect policy changes. An example is early 1980s Britain where the Conservative Government rolled back individual entitlements but aggregate spending actually increased. However, authors often argue that these issues can be mediated by carefully chosen control variables (Allan & Scruggs, 2004; Scruggs, 2008; Stichnoth & Van der Straeten, 2013; Wang & van Vliet, 2016). Second, there are differences in the tax treatment of transfers and social benefits between countries - such as varying tax structures and/or income tax exemptions. This could mean that levels of disposable income of benefit recipients vary despite the same levels of social spending. The tax system is increasingly used as an alternative transfer mechanism - a notable example is the United Kingdom s Working Families Tax Credit and gross spending data do not capture this. Unfortunately, the data for net social spending, which accounts for tax expenditures, is sparse. Consequently, this makes comparing social protection systems using spending more difficult (Allan & Scruggs, 2004; Caminada et al., 2010). Third, welfare effort is not completely restricted to the public domain; there is a large variety of private arrangements that can act as substitutes to public programmes. Finally, it may be that retrenchment in areas such as pensions may not be visible immediately because they have a long phase-in period and do not affect the current beneficiaries but have a large effect on the entitlements of future recipients (Pierson, 1996; Starke, 2006). Taking into account the dependent variable problem, I use two different dependent variables to approach the question of how immigration influences social welfare generosity. Initially, I use social welfare spending as a percentage of GDP as is common convention. I then take the welfare generosity index from the Comparative Welfare Entitlements Dataset (CWED) 5 developed by Scruggs et al. (2004) in order to compare and contrast the results of the two different indicators for welfare generosity. The welfare generosity index contains information on the generosity of social benefits. It covers unemployment insurance, sick pay insurance, and public pensions. Unemployment insurance only covers national insurance provisions that are earned without income testing and so excludes programmes such as the UK s income-based Jobseeker s Allowance or Germany s unemployment assistance. Sick pay insurance covers the benefits that are paid in the instance of short-term nonoccupational illness or injury. Public pensions covers only mandatory public programmes except the nominally private Finnish earnings-related fund (Scruggs et al., 2014). 5 See Scruggs, L., Jahn, D., & Kuitto, K. (2014). Comparative Welfare Entitlements Dataset 2 Codebook. Version University of Connecticut & University of Greifswald. for further information on the dataset, the countries included, the index and its methodology. 10

11 Limitations of this index includes the lack of recent data, it extends from 1990 to 2010 and it is only available for 16 European countries. Additionally, it does not cover aspects such as maternity leave benefits, certain child/family benefits and publicly provided health insurance/universal healthcare. Thus, the index could underestimate the generosity of some welfare states. Table 1 shows social welfare spending as a percentage of GDP as an average over the period for each country. France tops the table with 29% of its GDP spent on social welfare, on average. Unsurprisingly, the Scandinavian countries Sweden, Finland and Denmark - also have high social welfare spending. Notably, Norway is somewhat lower, with 22% percent of its GDP spent on social welfare on average this is the same as Luxembourg, Spain and Hungary, countries that are not traditionally associated with generous welfare states. Overall, it appears that Western European countries, those that are traditionally associated with more generous welfare states, do indeed have higher welfare spending as a percentage of their GDP. Although the welfare state typologies do seem to exist (Esping-Anderson, 1990), with the Social Democratic and Conservative welfare state typologies dominating the top half of the table. In contrast are the post-soviet states and the Liberal welfare state typology (United Kingdom, Iceland and Ireland), they spend a much lower percentage of their GDP on welfare. When we compare welfare spending with Table 2, which shows the Total Welfare Generosity Index 6, we can see that they do not correspond exactly the correlation is 0.4. Of particular note is Norway, which leaps from 9 th place in social welfare spending, to 1 st place in the generosity index. Also interesting, is Austria and Belgium. Both countries spend a similar amount of their GDP on welfare on average, but when looking at the generosity index we can see that Belgium scores almost 8 points higher. However, as discussed previously, this index could be improved through including a larger range of social programmes, Austria may be generous in ways that the index does not capture. Table 1: Social Welfare Spending as a Percentage of GDP, average: , OECD SOCX Country Mean Country Mean Country Mean France 29 Luxembourg 22 Czech Republic 18 Sweden 28 Spain 22 Switzerland 17 Finland 27 Hungary 22 Slovakia 17 Belgium 26 Greece 20 Ireland 17 Austria 26 Netherlands 21 Iceland 15 Denmark 26 Poland 21 Estonia 15 Germany 25 Slovenia 21 Latvia 14 Italy 24 Portugal 20 Norway 22 United Kingdom 19 Average 22 6 There are a different set of countries as the OECD dataset covers more European countries than the CWED. 11

12 Table 2: Total Welfare Generosity, average: , CWED Country Mean Country Mean Country Mean Norway 42 Denmark 36 Portugal 31 Belgium 41 Finland 35 Italy 28 Sweden 41 Germany 35 Greece 28 France 38 Spain 34 United Kingdom 27 Netherlands 37 Austria 33 Switzerland 36 Ireland 31 Average 35 Also of interest, is how social spending and welfare generosity have changed over time. Figure 1 shows how social welfare spending as a percentage of GDP has changed over time. It appears that most European countries have seen steady increases in spending levels or stayed reasonable stable. Welfare generosity on the other hand, as shown in Figure 2, appears to be more heterogeneous. Germany, Denmark and Finland have seen steady declines in their generosity indexes. Others like Greece and Italy have seen steady increases. Sweden has seen a dramatic decrease, while Ireland has seen a dramatic increase. Figure 1: Social Welfare Spending as a Percentage of GDP, Austria Belgium Czech Republic Denmark Estonia Finland France Germany Greece Hungary Iceland Ireland Italy Latvia Luxembourg Netherlands Norway Poland Portugal Slovakia Slovenia Spain Sweden Switzerland United Kingdom Year 12

13 Figure 2: Welfare Generosity Index, Austria Belgium Denmark Finland France Germany Greece Ireland Italy Netherlands Norway Portugal Spain Sweden Switzerland United Kingdom Year Explanatory Graphs by Country Variable As is common convention in the previous literature (e.g. Burgoon et al., 2012; Burgoon, 2014; Gaston & Rajaguru, 2013; Mau & Burkhardt, 2009; Soroka et al., 2006, 2016), I use foreign born as a percentage of the population, which serves as an indicator of the stock of migrants in a country, as the main explanatory variable. The standard definition of foreign-born is all persons who have ever migrated from their country of birth to their current country of residence (OECD, 2017a). People who were born abroad as nationals of their current country of residence are included in the foreign-born data. Mau and Burkhardt (2009) found that the percentage of foreign-born, as a proxy for ethnic diversity, was one of the most useful indicators for explaining variation in attitudes towards the welfare state. An alternative measure is net migration, as used by Lipsmeyer and Zhu (2011). However, Soroka et al. (2006) argue that high emigration in some countries can, and does, mask considerable inflows of migrants. Table 3, displays the percentage of the population that is foreign born. There is considerable variation between countries. To start, Luxembourg has the highest foreign-born population; unlike the majority of other EU countries, most of the foreign-born in Luxembourg are other EU nationals (Eurostat, 2017; Kollwelter, 2007). Switzerland is a similar case where just four EU nationalities - Italian, German, Portuguese and French - make up almost half (49%) of the total foreign-born population (Nguyen, 2016). 13

14 Estonia s high foreign-born population is somewhat of an outlier; it is because of the large number of recognised non-citizens. These are mainly former Soviet-Union citizens and are permanent residents, but have not acquired any other citizenship (Eurostat, 2017). Finally, there are no surprise that Poland, Romania and Bulgaria have, on average, small foreign-born populations; all are traditionally considered countries of emigration. Whereas Sweden, Germany and Austria, traditionally considered countries of immigration, have relatively higher foreign-born populations. Table 3: Foreign-born as a percentage of the total population, average: , OECD Country Mean Country Mean Country Mean Luxembourg 36 France 11 Norway 9 Switzerland 24 Slovenia 11 Portugal 7 Estonia 17 Spain 10 Czech Republic 6 Austria 14 Greece 10 Denmark 6 Ireland 13 Netherlands 10 Finland 4 Belgium 12 Iceland 9 Slovakia 4 Germany 12 Italy 9 Hungary 3 Sweden 12 United Kingdom 9 Poland 2 Average 11 Figure 3 shows that over time, most Western European countries have seen a steady increase in their foreign-born populations. Italy, Ireland and Spain are interesting because they are usually considered countries of emigration, yet all three have seen large increases in their foreign-born populations. Italy, Ireland and Spain all record both high immigration on non-nationals and emigration of nationals, helping to push the foreign-born population higher (Eurostat, 2017). The only country to see their foreign-born population reduce since 1990 is Estonia. The foreign-born populations of other EU/Schengen countries have remained relatively stable. 14

15 Figure 3: Foreign-born as a percentage of the population, Austria Belgium Czech Republic Denmark Estonia Finland France Germany Greece Hungary Iceland Ireland Italy Luxembourg Netherlands Norway Poland Portugal Slovakia Slovenia Spain Sweden Switzerland United Kingdom Control Variables Population Demographics. I control for certain population demographics by including the population under 15 and the population over 64 in the model specification - as is advocated by Soroka et al. (2006, 2016). It is reasonable to assume that spending would increase as a larger proportion of the population become dependent on working age tax payers. Infants are expected to affect health care and child-care costs, school-age children affect education costs and the elderly impact spending on pensions (Soroka et al., 2006). Pensions are typically the largest expenditure in social welfare budgets, this is due to a combination of increased life expectancy and a decline in birth rates that cannot be mitigated by current increases in the retirement age or immigration (Ervasti et al., 2012). Alternatively, Gaston and Rajaguru (2013) use the dependency ratio to account for the impact of the dependent population. In the sensitivity analysis, I substitute the population under 16 and over 64 and find that the results are not altered. Economic Controls. I control for the effect that a country s domestic economic status may have on social welfare generosity using three key variables. First, I use GDP growth as Gaston and Rajaguru (2013) have done. They find that GDP growth has a significant and negative association with social spending. They explain that this reflects the fact that 15

16 the denominator (GDP) grows more slowly than the numerator (SOCX) when the economy slows and social expenditures are politically difficult to decrease in a downswing of the business cycle (Gaston & Rajaguru, 2013). Moreover, a growing economy is expected to act as a pull factor and attract more immigrants than one that is contracting (Massey, 1988). Second, I include the national unemployment rate as a control for the domestic labour market as is done by Gaston and Rajaguru (2013), Lipsmeyer and Zhu (2011) and Soroka et al. (2006, 2016). Higher unemployment would indicate more spending on unemployment benefits. Plus, immigrants may take the prospective job market into account when deciding on a destination country (Davanzo, 1978). Third, I use female labour force participation as in Soroka et al. (2006, 2016). Female labour force participation affects social spending through increased public childcare infrastructure to support working mothers. It is also thought to be negatively related to migration as female labour acts as an imperfect substitute for migrant labour (Afonso & Devitt, 2016; Soroka et al., 2006, 2016). Political Institutions. One of the strongest generalisations that can be made about the origins and growth of the welfare state is that where trade unions and social democratic parties are strong, the welfare state has thrived (Freeman, 1986). It is generally thought that left-leaning governments, who traditionally have electoral ties to the working-class and unions, will support greater redistribution (Armingeon & Giger, 2008; Lipsmeyer, Philips, & Whitten, 2017) although the importance of partisan politics is still debated (Allan & Scruggs, 2004; Pierson, 1996; Starke et al., 2014). Freeman (1986) argues that migrant labour threatens the welfare state through dividing the working class and thus breaking the unity of organised labour movements. Moreover, some authors find that a strong left or strong trade unions can counteract the potential negative effects of diversity on welfare generosity (Lipsmeyer et al., 2017; Lipsmeyer & Zhu, 2011; Taylor-Gooby, 2005). As a result, I control for the ideology of the government in power by including the percentage of cabinet posts held by social democratic and other left-wing parties, weighted by the number of days in office in a given year as done by Soroka et al. (2006, 2016) and Lipsmeyer and Zhu (2011). As an alternative, Gaston & Rajaguru (2013) use a 1 to 5 scale to account for cabinet ideology, with 1 being a hegemony of right-wing parties and 5 being a hegemony of left-wing parties. In the sensitivity analysis, I switch the percentage of cabinet posts held by social democratic and other left-wing parties for the cabinet ideology indictor and find that the results do not change. In addition, I use trade union density net union membership as a share of wage and salary earners in employment - as a control for the bargaining power of domestic labour as done by Lipsmeyer and Zhu (2011) and Soroka et al. (2006, 2016). Economic Globalisation. Globalisation has been argued to both reduce public spending and increase it (Gaston & Rajaguru, 2013; Iversen & Cusack, 2000; Lipsmeyer & Zhu, 2011). Thus, to control for economic globalisation, I use the KOF economic globalisation indicator, as do Gaston and Rajaguru 16

17 (2013). The KOF economic globalisation indicator ranges between 0 and 100, with higher values indicating a higher degree of economic globalisation. Economic globalisation is here defined as the long distance flows of goods, capital and services as well as information and perceptions that accompany market exchanges. It is measured by actual flows of trade and investments, and by restrictions on trade and capital such as tariff rates (Dreher, 2006). An alternative indicator for economic globalisation is trade openness as used by Lipsmeyer and Zhu (2011). Trade openness is measured as the sum of exports and imports of goods and services as a share of GDP. I use this instead of the KOF in the sensitivity analysis and the results remain unchanged. Soroka et al. (2006, 2016) do not account for globalisation in their model specification. Research Design: Method and Model Specification A common practice for this type of study is to use the de facto Beck-Katz standard panel-corrected standard errors with country fixed effects and a lagged dependant variable as done by Gaston and Rajaguru (2013), Lipsmeyer and Zhu (2011) and Soroka et al. (2006, 2016). However, it is argued that the lagged dependent variable is a considerable source of bias known as Nickell bias (Nickell, 1981). It is reasoned that the lagged dependent variable is highly correlated with the dependent variable and thus causes an upward bias in the standard errors. Subsequently, the estimation model does not provide an accurate coefficient for the key explanatory variable. Therefore, for the empirical analysis I use the more appropriate panel-corrected standard errors with country and year fixed effects and Prais-Winsten correction for serial correlation of errors as recommended by Plümper et al. (2005). In addition, I lag the explanatory variable and all the control variables by one year - Gaston and Rajaguru (2013) do the same, while Lipsmeyer and Zhu (2011) and Soroka et al. (2006, 2016) lag a selection of their variables. One key reason is that it can help mitigate endogeneity issues arising from reverse causality. Furthermore, in the case of certain variables it makes theoretical sense; policy decisions can take time to be reflected in spending levels - by lagging the variables, this can be better taken into account. Finally, I use country and year fixed effects. By using country fixed effects, I hope to not only control for welfare states that typically spend more because of how they were constructed/built-up over the years but also to capture cultural influences that may determine a country s tendency towards favouring a more Bismarckian or Beveridgean welfare state. Moreover, by using year fixed effects I aim to account for external shocks that may have taken place in certain years; for example, the expansion of the European Union, the lifting of labour market restrictions on new member states and the financial crisis. 17

18 Results and Discussion The estimate in column 1 of Table 4 indicates that higher immigration leads to higher social spending. Levels of foreign-born are positively and significantly associated with levels of social spending; a 1 percent increase in foreign-born is associated with, on average, a percent increase in social welfare spending, ceteris paribus. These findings are similar to Gaston and Rajaguru (2013) who also use social spending as a percentage of GDP and conclude that immigration has a relatively modest effect on welfare state spending. On the other hand, Gaston and Rajaguru also find that when they reduce their sample to 11 European countries the effect disappears. Whereas, I find a highly significiant effect with 16 European countries, and when that sample is expanded then the magnitude of the effect lessens but the result still remains highly significant. The crucial difference between our two models is the way we correct for serial correlation. Gaston and Rajaguru (2013) use a lagged dependent variable, whereas I use the Prais-Winsten correction - advocated by Plümper et al. (2005) as the more appropriate method. Moreover, Gaston and Rajaguru (2013) do not control of population demographics in their model, nor the unemployment rate. A first explanation is that my results lend support for the compensation hypothesis, Gaston and Rajaguru (2013) also use their results to say they find some support for the compensation r exposure effect. Some previous literature focusing on support for redistribution found that if natives feel economically insecure when exposed to increased movement of labour, particularly when exposed at an occupational level, then they support more compensation and greater redistribution from the government (Burgoon et al., 2012; Finseraas, 2008; Walter, 2010). Thus, the higher spending associated with immigration could be an indication that this demand is being reflected in policy decisions. However, when I replace the dependent variable with the generosity index, the coefficient for foreignborn is not statistically significant. I find no effect for either increased or decreased generosity as a result of increasing immigration. It is interesting that the results for these two indicators of welfare generosity are different it suggests that the two indicators are in fact, measuring different things. Instead of the increase in spending suggesting support for the compensation hypothesis, spending may increase for a different reason. Perhaps the foreign-born population costs the welfare state more in terms of education or child-care costs, especially considering immigrants in Europe tend to have more children than the native population (Boeri, 2010; Freeman, 1986). Moreover, it is possible that immigration is linked to increased spending through increased unemployment, either of natives or immigrants, rather than increased generosity. On the other hand, the foreign-born population on the EU is, as a whole, younger than the native population and is less likely to use the health system or draw a pension (Eurostat, 2017). Just as social spending is a flawed indicator, so is the welfare generosity indicator with certain programmes not 18

19 included in its make-up. Consequently, the lack of a statistically significant relationship between foreignborn and the welfare generosity index does not mean that the increase in social spending is not indicative of increasing generosity. Table 4: Effect of Foreign Born on Social Welfare Spending as a Percentage of GDP and the Welfare Generosity Index, (1) (2) Variable Spending Generosity Foreign-bornt *** (0.070) (0.085) Population under 15t ** (0.275) (0.250) Population over 64t ** (0.122) (0.151) Unemployment ratet ** (0.058) (0.048) GDP growtht *** (0.048) (0.049) Female labour force participationt *** (0.055) (0.050) Left Seatst *** (0.002) (0.003) Union Densityt * *** (0.050) (0.055) KOF - economict (0.038) (0.043) Intercept * (6.660) (6.592) Country Dummies YES YES Year Dummies YES YES N Adj R Standard errors in parentheses * p <.1, ** p <.05, *** p <.01 The difference in the effect of foreign-born on spending and generosity is not unique to the explanatory variable. For example; it is not surprising that the size of the population over 64 years old has a significant and positive impact on social spending considering pensions tend to be the biggest expenditure for welfare states. However, there is no statistically significant effect on the generosity index. Likewise for those under the age of 15 and for the unemployment rate. I find that GDP growth is significant and positively associated with social spending - this is the same result as Gaston and Rajaguru (2013) who explain that in an economic downturn, the denominator (GDP) will grow more slowly than the numerator (social spending) and so is not unexpected. This 19

20 explanation also makes sense for why there is no statistically significant relationship between GDP growth and the generosity index. A particularly interesting result is female labour force participation, which has no significant effect on social spending. Whereas for generosity, female labour force participation has a large, positive effect and is highly statistically significant. Alongside arguments that female labour force participation increases welfare generosity through the need for improved child-care infrastructure, Soroka et al. (2016) suggest that it may also be linked to increased costs associated with a larger work force such as training, employment insurance and leave. Moreover, in the medium to long-term, it is thought that women s participation in the labour force will increase women s demands for redistribution (Huber & Stephens, 2001). Furthermore, the proportion of left cabinet seats is positively and significantly associated with social spending, but appears to be unimportant for explaining welfare generosity. There is debate in the literature about the impact of partisan politics on the expansion and retrenchment of the welfare state. The effect of left seats on social spending would appear to support the Allan and Scruggs (2004) camp that partisanship is important, while the impact of left seats on the generosity index would appear to support Pierson (1996) and his new politics of the welfare state. So as Starke stated: "the debate on the relevance of political parties and ideas [ ] is still far from settled" (2006) Union density is positively and significantly associated with both social spending and the generosity index. Although, the magnitude of the effect of union density is much greater for the generosity index and the significance level is higher. It appears that union membership and support is important for social spending and generosity, this is similar to the results found by Lipsmeyer and Zhu (2011) who argue that domestic political pressures are important for explaining higher unemployment compensation in an age of increasing immigration. In contrast to Gaston and Rajaguru (2013) who find a statistically significant negative relationship between economic globalisation and social spending, I find no significant impact of the KOF economic globalisation indicator on either social spending or the generosity indicator. Sensitivity Analysis In this section, I discuss the results of various tests that were undertaken in order to check the robustness of my results. In order to see if my results are sensitive to the control variables chosen, I substitute a number of the variables with various alternatives that have also been chosen by other authors. The results are shown in Table 5. The result for foreign-born appears to be robust; it is positive and statistically significant for all the estimation models using social welfare spending, and for all the estimation models using generosity is consistently statistically insignificant. Table 5: Alternative control variables,