Attitudes Towards Highly Skilled and Low Skilled Immigration in Europe A Survey Experiment in 15 European Countries

Attitudes Towards Highly Skilled and Low Skilled Immigration in Europe A Survey Experiment in 15 European Countries Elias Naumann 1, Lukas Stoetzer 2, Giuseppe Pietrantuono 3 1 University of Mannheim, Germany 2 Massachusetts Institute of Technology, Department of Political Science 3 University of Mannheim, Germany & University of Zurich, Switzerland Introduction Migration policy is at the top of the political agenda not only since the refugee crisis. One hope among policy makers is that migration might help overcome (mainly skilled) labour shortages due to the demographic change (United Nations, 2000; European Parliament, 2007). At the same time fears prevail that (unskilled) migrant workers will displace natives from the labour market and will put downward pressure on natives wages (e.g. Borjas 2003). These concerns fuel opposition against migration and gave rise to several right-wing parties in Europe (Ivarsflaten, 2008). Public opinion is a crucial factor in understanding (migration) politics (Stimson 2004, Burstein 2006, Brooks and Manza 2006, Erikson et al. 2002) and public attitudes towards high-skilled and low-skilled migration restricts politicians room for manoeuvre. It is thus of high importance to understand how people s attitudes toward migration are shaped. Why do people oppose or favor immigration? The extant literature has proposed two basic motivations that shape public attitudes towards immigration, cultural values and economic self-interest (see Hainmueller and Hiscox 1

2014 for a review). The first approach emphasizes the importance of cultural factors in shaping views on immigration. According to this perspective, individuals oppose immigration because foreigners are perceived as a threat to their, i.e. the national identity (e.g., Citrin et al. 1997; Fetzer 2000; McLaren and Johnson 2007, Card et al. 2011). The latter perspective relies mainly on material self-interest to explain attitudes towards immigration. Two strands have emerged within this political economy literature. Most prominently, the labour market competition model predicts that low- skilled migration will increase competition for low-skill jobs, lowering wages and employment in low-skilled sectors while increasing wages for high-skilled natives (Scheve and Slaughter 2001). Consequently, low-skilled natives are assumed to oppose lowskilled migration and rather support high-skilled migration. The opposite holds for high-skilled natives that should support low-skilled migration but oppose high-skilled migration (see also Mayda 2006). A second strand focuses on the fiscal impacts of migration. Assuming that lowskilled migration creates a net fiscal burden as low skilled migrants will possibly be net beneficiaries of welfare and of a redistributive tax system, high-income natives as the net payers of the system should be more strongly opposed to low-skilled migration compared to their low-income native counterparts. The opposite should apply for high-skilled migration (Hanson et al. 2007, Facchini and Mayda 2009, Hainmueller and Hiscox 2010, Tingley 2013). While scholars have consistently shown cultural concerns to be strong predictors of opposition, findings regarding the self-interest hypotheses are highly contested. Therefore, our article focuses on the self-interest explanation of attitudes towards migration. There is only very slowly emerging a consensus in that irrespective of skill-level or income people prefer high-skilled migration over low-skilled migration (Hainmueller and Hiscox 2010, 2

Helbling and Kriesi 2014, but see also Facchini and Mayda 2009). In how far the welfare state and/or fiscal pressures moderate such a relationship is still an open question as the two major experimental studies in the field (Hainmueller and Hiscox 2010, Helbling and Kriesi 2014) rely on single-country studies i.e. the US and Switzerland and use the intra-country variation to explain attitudes towards high-skilled and low-skilled migrants. The advantage of single country studies (and the use of intra-national variation) is the internal validity of its findings as most of the institutional context is held constant. Nevertheless this comes at a cost of external validity and uncertainty in how far these findings also hold in other institutional settings. The major concerns in this respect refer to American exceptionalism (Lipset 1996) and that findings from the US might not apply to European countries that differ largely in their migration history but also in how their welfare states developed (Larsen, 2008). Moreover, the US and Switzerland are very particular cases in that both have very low shares of public social spending (19,2% of GDP in the US, 19,4% of GDP in Switzerland compared to the OECD average of 21,6%, and 28% of GDP and more in Sweden, Italy and France). Against this background it might not be surprising that the fiscal and welfare context in these studies had only limited impact on the relationship between income, skill-level and attitudes towards migration. To test the implications of the labour competition and the fiscal burden models, we use a survey experiment (used in Hainmueller & Hiscox 2010) that was conducted in 15 European countries. Respondents from a variety of institutional settings thus were randomly assigned to give their attitudes towards either high-skilled or low-skilled migration. As one of the (very) few studies in the field of experimental political science we rely on cross-country comparisons and random probability samples of the respective populations. This comparison, how it varies with 3

respondents characteristics and whether it holds across a variety of institutional settings, allows us to reliably test the theoretical self-interest arguments. In addition to the spatial variation, how people form their attitudes towards migration should also vary over time and is particularly affected by salient events (for example see Legewie (2013) on how terrorist attacks have affected attitudes). Therefore, we replicated our experiment in 4 European countries after the onset of the refugee crisis in Europe (Germany, Sweden, Denmark and the UK). We use the refugee crisis as an external shock to the fiscal exposure to migration and examine how increased fiscal exposure affects the patterns of attitudes towards migration. Such a within country comparison nicely complements our crosssectional analysis providing leverage in terms of internal validity and inferring causality from correlations. For the labour market competition model in general we find that both high-skilled and low skilled natives prefer high-skilled over low-skilled migration, which is in line with previous findings stating that the labour market competition model holds only for less skilled natives. However, our robustness checks offer some support (even if not significant at conventional levels) by adopting more fine-grained competition measures for the theoretical predictions also concerning better skilled natives. As for the fiscal burden model, our results show a much clearer picture: the premium attached to high-skilled migration increases with income. Highincome respondents prefer high-skilled over low-skilled migration more than low income respondents since they are mainly concerned about tax increases due to the increased migration. Our results are thus in line with recent findings from Switzerland (Helbing and Kriesi, 4

2014) but contradicts evidence from the US (Hainmueller and Hiscox, 2010) where the premium attached to high skilled migration decreases with income. Moreover, by exploiting the refugee crisis as an exogenous shock we find a stronger relationship between income and support for high-skilled over low-skilled migration if fiscal exposure is high. Economic concerns and immigration attitudes National economies may be impacted by immigrant worker inflows in several ways. Prominently, the literature in the field has put forward and emphasized that anti-immigrant sentiments among natives may arise from two economic concerns: On the one hand, concerns about labor market competition and, on the other hand, fears about the fiscal burden on public services. Theoretically, both channels operate based on economic self-interest even though they rely on different assumptions on how natives react to immigrant worker inflows in relation to their skill-level, and thus, predict different attitudinal and preference based outcomes. Empirically, tests of these two approaches mainly have relied on survey data and have been mixed at best and plagued by methodological concerns (see for example Hainmueller and Hiscox 2010, 2015). Promisingly, survey experiments in the field (Hainmueller and Hiscox 2010, Helbling and Kriesi 2014) address these shortcomings by explicitly test and control for the natives as swell as the immigrants skill-level. Nevertheless, the argument whether antiimmigrant sentiments are contingent upon contextual factors still remains inconclusive. The labor market competition model 5

According to the labor market competition model, attitudes towards immigration are based on heterophyllic preferences in regard to the immigrants skill-level: Low-skilled natives prefer high-skilled immigrants, and high-skilled natives prefer immigrants with lower skills. Income effects are mainly hypostatized based on the fact that immigrant worker inflows shift the distribution of relative supplies of factors of production (these models are often referred to as factor-proportion analysis, see Hainmueller and Hicox 2010, 2015 for more details and Borjas 1999; Borjas, Freeman, and Katz 1996, 1997). This shift threatens natives earning capacities and employment opportunities. An inflow of low-skilled immigrants raises respectively the supply for low-skilled workers leading to a decrease of wages for native low-skilled worker and eventually to a higher unemployment rate and at the same time to an increase in earnings for higher skilled natives. In the same way, an influx of high skilled immigrant workers leads to lower wages for high skilled natives and real wages for low-skilled natives will rise. Empirical tests failed to unambiguously answer the question whether concerns about labor market competition motivate anti-immigrant sentiments: Whereas there is an (established) consensus that low-skilled natives in general oppose migration to a higher level than better educated natives (Hagendoorn and Nekuee 1999; Hjerm 2001; Scheve and Slaughter 2001; Heyder 2003; Mayda 2006), and in particular prefer high-skilled immigrants to low skilled, there is still uncertainty on the preferences of high-skilled natives. The uncertainty arises from the controversial discussion whether education level is linked to xenophobic sentiments, or whether the individual perception of economic threat drives opposition towards immigrants. Generally, several studies show a correlation between education level and anti-immigrant sentiments showing that better educated people are more willing to accept foreign workers. 6

According to these studies natives regardless of their skill-level always prefer high-skilled immigrants. Thus, the labor market competition model offers no explications for attitudes towards immigrants for high-skilled natives (Sniderman et al. 2004; Hainmueller and Hiscox 2010). In contrast, other scholars have suggested that this fact is not due to the higher education level, but is highly mediated through the individual perception of economic threat, which is lower among higher educated. O Connell (2011) argues that even if economic concerns are also relevant for people with high skills the higher specialization they gained protects them from direct competition with immigrants. Once exposed to a higher competition and thus to a higher level of economic threat high-skilled natives are more likely to oppose immigrants with highskills (Hello et al. 2006, Helperin et al. 2007, Facchini and Mayda 2012). However, evidence from experimental studies (Hainmueller and Hiscox 2010, Helbling and Kriesi 2014) suggests that better educated individuals are more in favor of immigrants in general and that the perceived economic threat is lower among highly educated even if exposed to a higher competition. In sum we test the classical prediction from the labor market model that natives oppose same-skilled immigrants. This is low-skilled natives prefer high skilled immigrants (Hypothesis 1a) and that high skilled natives prefer low-skilled immigrants (Hypothesis 1b). Fiscal burden of public services In general the welfare state model assumes that low-skilled immigrants impose a substantial net burden on public finance and challenges the welfare state, whereas highly 7

skilled immigrants are net contributors in terms of taxes. These may lead to a readjustment of the welfare system by either rising taxes or by reducing economic transfers such as welfare state services and benefits (Hanson 2005; Hanson et al. 2008; Facchini and Mayda 2009). Depending on the means chosen by the government to face the increased burden of public services, predictions about the acceptance of immigrant workers vary as high-income natives fear an increase in tax rates and low-income natives lower welfare benefits. In the first scenario, where tax rates are adjusted to face the increased burden on welfare state services high-income natives prefer high-skilled over low-skilled immigrants (more the low-income natives). Progressivity in taxes leads to a relatively higher increase in taxes for wealthier natives that might be required if the demand for welfare state benefits increases (Facchini and Mayda 2009). The second scenario foresees to keep tax rates fixed and rather adjust transfer rates of welfare state services. An increased demand for such services would lead to lower transfer rates. The fear for a reduction in welfare state benefits leads lower income natives to prefer high skilled to lower skilled immigrants (more than high-income natives). Empirically, research shows that these fears are contingent upon the states tax regimes: Hainmueller and Hiscox (2010), Hanson et al. (2007), and Facchini and Mayda (2009) exploit intra-state variation in the US and Helbling and Kriesi (2014) inter-cantonal variation in Switzerland. The observational studies by Hanson et al. (2007) and Fachchini and Meyda (2009) come to the same conclusion interpreting their results as evidence that fears about higher taxes among rich natives, linked to an increased use of welfare services by low-skilled immigrants, 8

lead to anti-immigrant sentiments. Nevertheless, this conclusion has to be drawn with caution as the surveys these analyses rely upon do not distinguish between immigrants skill level and implicitly assume that respondents have low-skilled migrants in mind when stating their attitudes towards migration in general. Two recent survey experiments address this shortcoming and test the welfare state model explicitly by experimentally varying immigrants skill-level. Hainmueller and Hiscox (2010) contrast the predictions of the standard fiscal burden model in which rich natives oppose low-skilled immigration more than poor natives do, and this difference grows larger in states with greater fiscal exposure in terms of immigrant access to public services. They find that rich and poor natives are consistently opposed to low-skilled immigration in general, and that rich natives are less opposed to low-skilled immigration in high-exposure states than in low-exposure states. Moreover, poorer natives are more opposed to low-skilled immigration in states with high fiscal exposure than in states with low fiscal exposure. This leads the authors to conclude that concerns about access to public services may contribute to anti-immigrant attitudes among poorer respondents. Helbing and Kriesi (2014) in contrast conclude from their results that the welfare state model only holds for wealthy natives in regions with low taxes and that poor natives hardly differentiate between low- and high-skilled immigrants. In a high exposure context, they do not find evidence for an income-dependent effect. They conclude that to the extent that context makes a difference in the Swiss case, it is the rich who are particularly context-sensitive (ibid: 605). The authors argue that this is because wealthier citizen tend to choose their residence with tax and spending considerations in mind. 9

We will test in general whether wealthier natives prefer high-skilled immigrants, as they reduce tax burdens (Hypothesis 2a) or whether poor natives prefer high-skilled immigrants, as they fear lower per capita welfare benefits (Hypothesis 2b). Moreover, we will put the theoretical argument to a comparative test assuming that wealthier natives have a preference for high over low-skilled immigrants in national-states with low fiscal exposure as they fear increasing taxes (Hypothesis 3a). Further, we assume that poorer natives are particularly opposed to low- in comparison with high-skilled immigrants in states with a high fiscal exposure, as they fear an erosion of the welfare state (Hypothesis 3b). In sum, previous evidence from observational studies clearly has some fault back in testing the proposed theoretical approaches; particularly by empirically neglecting the central economic mechanism that opposition to immigrants is linked to migrants skill-level. Survey experiments have addressed these concerns, but still remain inconclusive in regard to the impact of crucial contextual factors. Our approach allows us to test whether previous results are idiosyncratic to the single countries that were under scrutiny or whether the results are generalizable across different welfare state and tax regimes. Moreover, we put the welfare state model to a further test by exploiting an external shock in low-skilled migration inflows. Due to the refugee crisis we can validate our results for the welfare state model. The influx of refugees challenges the welfare state as by law refugees will not be net contributors in terms of taxes (i.e. they are not allowed to work), but rather impose a substantial net burden on public finances. In line with the welfare state model we expect both high as well as low-skilled natives will shift their preferences to high skilled immigrants. 10

Survey Experiment We use data from the seventh round of the European Social Survey (ESS) fielded in 21 European countries between August 2014 and December 2015. The ESS is a large-scale, crossnational project that conducts biennial face-to-face interviews in European countries. Individuals are selected by strict random probability methods at every stage of the sampling process so that each country sample is representative of all persons aged 15 and over resident within private households in each country. Response rates vary between 50% and 65% (with the exception of Germany that has a response rate of 31%). The country samples consisted of at least 1,400 and at most 3,000 respondents each. In the following analysis we use data for the following 15 countries: Austria, Belgium, Czech Republic, Denmark, Finland, France, Germany, Ireland, Netherlands, Norway, Slovenia, Spain, Sweden, Switzerland and the United Kingdom. 12 We restrict our analysis to those respondents that were born in the respective country. Our pooled data-set contains 24,951 cases. To account for different sampling strategies affecting individuals probability to be included in the survey and varying sample sizes we use the design weights provided by the ESS. 1 We exclude Hungary, Israel, Lithuania, Portugal and Poland from our analysis since data collection in these countries took place between April and December 2015 and thus overlaps with the beginning of the European refugee crisis. It is of course difficult to say when exactly the European refugee crisis began. One event that gained high media attention across all countries was when a migrant vessel with around 850 refugees capsized in the Mediterranean on April 19. On the next day (April 20, 2015) the European Commission proposed a 10-point plan to tackle the crisis. Three days later, on 23 April 2015, an emergency summit was held and EU heads of state agreed to triple the budget of Operation Triton to 120 million for 2015 2016. Therefore we restricted our analysis to those interviews that took place before April 2015. The majority of interviews used in our analysis (75%) took place in 2014 well before the onset of the refugee crisis. 2 We also excluded Estonia from our analysis for two reasons: first, the Estonian dataset does not provide the income variable in brackets that allow to code income qunitiles. Second, the administration of the survey experiment seems not to have worked well in Estonia as the experimental groups differ quite substantially in size (see Appendix). 11

The experimental design of our study resembles the one used in previous research (Hainmueller and Hiscox 2010; Helbling and Kriesi 2014). Respondents were randomly assigned to two groups. 3 Respondents in the one group were asked about their attitude towards high skill migration (version 1), the other half of respondents about low skill migration (version 2): Version 1: Please tell me to what extent you think [country] should allow professionals from [other country] 4 to come to live in [country]? Version 2: Please tell me to what extent you think [country] should allow unskilled labourers from [other country] to come to live in [country]? Answer options for both versions include Allow many to come and live here, Allow some, Allow a few and Allow none. We recode answers so that high values mean support for immigration. In line with previous research we use educational attainment as a measure for respondents skill level (Hainmueller and Hiscox 2010, Helbling and Kriesi 2014, Hanson et al. 2007, Facchini and Mayda 2009; but see also Malhotra et al. 2013 and Polavieja 2016 for a critique of this 3 Balance checks between the two groups provide evidence whether randomization has worked. As expected there are no significant differences between treatment and control group for 9 of the 10 covariates we are using. Only married or widowed respondents are slightly underrepresented in the treatment group. We cannot say whether these differences are systematically related to the administration of the treatment or whether they occurred by chance. In any case, treatment and control group are not completely balanced and we therefore include marital status as an additional control in our analyses (in addition to the standard socio-demographic variables age, gender and political ideology). 4 The ESS combined this experiment with a second experiment in which the origin countries of the migrants were also varied. The stock and the inflow of migrants but also the Human Development Index of origin countries were used to identify the most significant migrant origin countries inside Europe and outside of Europe. A list of the specific countries is found in the ESS7 Appendix A10. As our interest is mainly focused on the self-interest explanation and less on the impact of cultural distance, we do not make use of this experimental variation but rather include the information as a control. 12

approach). We use the European Social Survey version of ISCED that provides an indicator for the highest level of education a respondent has successfully completed that is comparable across countries. We distinguish between respondents with less than lower secondary education, lower secondary education, upper secondary and advanced vocational education, and tertiary education. Although education is widely used as a measure for skill level, concerns have been raised that it does not capture respondents specific skill levels that are key on the labour market to determine labour market competition (Polavieja 2016). As a consequence, respondents exposure to labour market competition with migrants might differ substantially within the same educational group and might depend more on the specific occupation or the industry someone is working in. As an indirect measure for labour market competition, education might only very roughly capture actual labor market threat (Malhotra et al. 2013). Moreover, education is correlated with many noneconomic attitudes that also motivate attitude towards migrants such as cultural tolerance, social trust and also social desirability bias in survey responses (Hello et al. 2006, Malhotra et al. 2013). We address these concerns with a series of robustness tests using (1) an alternative measure of skill level, (2) a direct measure of labor market competition (see also Malhotra et al. 2013, Hainmueller et al. 2015), and (3) two indirect measures of labor market competition (Polavieja 2016). First, we determined respondents skill level based on their occupation (10 major occupational groups of ISCO 08) and distinguish 4 different skill levels (see Appendix). These skill levels are not only based on the formal education but also take into account the 13

nature of the work performed in an occupation, the amount of informal on-the-job training and skill specialization (ILO 2012). Second, we rely on a direct measure of labor market competition with migrants. The ideal approach would be to move beyond indirect measures, such as education, and assess labor market competition more directly (Malhotra et al. 2013: 394). Even within the same skill-level group, exposure to labor market competition with migrants possibly varies substantially depending on the composition of migrants and their occupations. Moreover, the skill-set of migrants possibly varies between European countries. Therefore, we re-run our analysis with a direct measure of labour market competition. Based on the EU Labour Force Survey from 2013 (with approximately 2.4 million respondents) we calculate the share of foreign-born in each occupation (at an ISCO 3-digit level) for each of our 15 countries separately (LMC1). We code a respondent as being exposed to labor market competition with migrants if the share of foreignborn in the native s occupation is above the country s average number of foreign-born (LMC1 high). Moreover, we will examine a most likely case to find the LMC hypothesis confirmed by further targeting our analysis to respondents working in occupations where the share of foreign-born is 25 percentage points higher than the national average (LMC1 very high). Thirdly, we rely on two occupational characteristics that restrict or limit outside competition: skill specialization and monitoring costs (Polavieja 2016, Ortega and Polavieja 2012). Based on data from previous rounds of the ESS we calculate a measure of skill specialization and monitoring costs for each occupation (at the ISCO 3-digit level) in each country and match these indicators to the respondents in our dataset (see Polavieja 2016). Higher values on these 14

indicators for skill specialization (LMC2) and monitoring costs (LMC3) are related to more outside competition. For a test of the fiscal burden/welfare state model, previous research mainly relied on indirect measures of the fiscal impact of migrants. Countries with public social spending per capita above the mean and a high share of foreign-born are assumed to have a high fiscal burden (Hainmueller and Hiscox 2010, Helbling and Kriesi 2014). Such broad measures neglect that migrants access to social benefits and also their take-up rates of social services differ from country to country (Hanson et al. 2007). Moreover, migrants contribution to the financing of these services via taxes and/or social insurance contributions also depend on the institutional context and migrants labour market access. The ideal indicator for fiscal burden would directly measure migrants contributions via taxes and social security contributions in comparison to the benefits they actually receive. The International Migration Outlook 2013 (OECD 2013) provides a first-time comparative analysis of the fiscal impact of immigration in OECD countries, using high-quality data from the EU-SILC and the Swiss Household Panel Study. The net contribution (in EUR, PPP adjusted) is defined as the difference between contributions paid and the benefits natives and foreign-born receive respectively. Benefits are all government-funded transfers received by households, including family- and children-related allowances, social assistance payments, housing allowances, unemployment benefits, old-age benefits, survivors benefits, and pensions, sickness benefits, disability benefit, and education-related allowances and scholarships. Contributions are all transfers from households to the government, including taxes, applicable tax credits, and social security contributions. We use three different measures of fiscal exposure. Fiscal Exposure 1 only relies on the net contribution of foreign-born. 15

Countries in which foreign-born on average receive more benefits than they contribute in taxes and social security contributions are affected by high fiscal exposure to migration. These countries are the Czech Republic, Ireland, France, and Germany. Focusing only on the net contribution of foreign-born neglects that also the number of foreign-born affects the degree of fiscal exposure. Therefore we multiply the net contribution and the number of foreign-born to calculate our second measure for fiscal exposure (Fiscal Exposure 2). Countries with Fiscal Exposure 2 above the median are classified as having a high fiscal exposure. These countries are Czech Republic, Finland, France, Germany, Ireland and Sweden. Some authors (e.g., Rho and Tomz 2016) argue that the main reason why many studies fail to find self-interested economic preferences is because of respondents economic ignorance and their misperception of economic threats. To address these concerns we rely on subjective perceptions as a third measure of fiscal exposure (Fiscal Exposure 3). Respondents were asked whether they think that people who come to their country take out more than they put in or put in more than they take out. 5 We use these subjective perceptions of fiscal exposure to migration in order to calculate a subjective indicator of fiscal exposure for each country (i.e. the country specific mean). Again we use the median to split our sample of countries in high and low fiscal exposure countries. According to Fiscal Exposure 3, respondents in Austria, Belgium, the Czech Republic, Ireland, France, Spain and in the UK feel a high fiscal exposure to migration. 5 The exact question wording is: Most people who come to live here work and pay taxes. They also use health and welfare services. On balance, do you think people who come here take out more than they put in or put in more than they take out? The Answer scale ranges from 0 Generally take out more to 10 Generally put in more. 16

In order to get a measure for the position of each respondent in the income distribution, we use household s total net income and assign respondents to the respective country specific income quintiles. 6 In the analysis, we further control for respondents sex, age, employment status, political ideology and whether they are married. The controls are mainly included to make the results comparable to previous studies but also to capture the non-random assignment of the moderators (education and income) in our analysis. Attitudes towards highly skilled and low skilled migration in 15 European countries - results from European Social Survey Figure 1 depicts the distribution of preferences for high-skilled and low-skilled migrants. Respondents are more positively disposed towards highly skilled migrants, compared to lowskilled migrants. While around 70% would allow some or many high-skilled migrants, only 40% percent are in favor of allowing some or many low-skilled migrants. The strong opposition towards low-skilled migrants gets especially apparent when comparing the extreme categories. Only 10 % would allow many low-skilled migrants and 26% none. The numbers are practically reversed for high skilled migrants: 8% would allow none and 23% allow many. The disparity of those two unconditional distributions is much in line with the results from previous studies (Helbling and Kriesi 2014, Hainmueller and Hiscox 2010). Moreover, these patterns are found in every single country in our study (see Figure A1 in the Appendix). In the following we will test if 6 The reported household net income has a large number of non-reported and missing values. About 13 % of the total data is missing. 17

those distributions are conditioned as the labor market competition model or the fiscal burden perspective would suggest. [Figure 1 about here] Labour market competition [Figure 3 about here] 18

The central hypothesis that natives should oppose immigrants with similar skill levels but favour immigrants with different skill levels is not reflected in the empirical patterns. Figure 3 plots the percentage of respondents who would allow many migrants for different educational attainments. Two main patterns emerge and contrast the expectations. First, both low and high skilled respondents prefer high-skilled migration over low skilled migration. 37 % of the respondents with tertiary education would allow many high skilled migrants, but only 18 % would allow low-skilled workers. For respondent with lower secondary education the increase is from 4 % to above 12 %. Second, the support for both types of migration increases with respondents own skill level. Educated respondents are generally more open to migration independent of the skill level of migrants. While this can be explained with a variety of other economic and non-economic reasons, the unmet expectation of the labour market competition here is that highly educated should actually reduce their support for potentially high skilled labour competitors (Hypothesis 1b). The same disapproving patterns are observed in Hainmueller and Hiscox (2010) and Helbling and Kriesi s (2014) analysis. 19

Table 1 Support for highly Skilled and Low-Skilled Immigration - Test of the Labor Market Competition Model (1) (2) (3) (4) (5) (6) (7) In Favor of High Skilled Migration In Favor of Low Skilled Migration In Favor Migration In Favor Migration In Favor Migration labor force: in labor force: out Education 0.266 ** (0.014) 0.253 ** (0.014) 0.251 ** (0.014) HS-Frame 0.766 ** 0.750 ** 0.817 ** (0.017) (0.052) (0.029) HS-Frame X Education 0.012 (0.018) < lower secondary -0.399 ** (0.054) HS-Frame X < lower secondary 0.064 (0.067) Lower Secondary -0.156 ** (0.030) HS-Frame X lower secondary -0.098 * (0.040) Tertiary 0.371 ** 0.264 ** (0.020) 0.567 ** (0.078) 0.069 ** (0.025) 0.238 ** (0.021) 0.863 ** (0.071) -0.022 (0.027) (0.033) HS-Frame X tertiary -0.012 (0.043) Controls Yes Yes Yes Yes Yes Yes Yes Country FE Yes Yes Yes Yes Yes Yes Yes N 10110 9888 20077 19998 19998 11348 8613 Ordered Probit Coefficients shown with standard errors in parentheses. All models include a set of covariates (age, gender, marital status, political ideology and income) and country dummies. The reference category for the education dummies is 'upper secondary or vocational education'. * p < 0.05, ** p < 0.01 20

The significance of the effect can be analyzed more thoroughly relying on ordered probit models. This model specification takes the uneven spacing between answering categories into account and further allows controlling for other covariates, including the different levels between countries. Table 1 reports the estimates from seven model specifications. The full specification includes a direct effect of the frame (HS-frame is coded to be one for attitudes towards high-skilled migrants and zero if the respective question refers to low-skilled migrants), a direct effect of educational attainments and an interaction between the two. All models moreover include country dummies and additional controls. The results echo the patterns from the descriptive analysis. Model 1 and 2 in Table 1 show the effect of education for the subset of respondents who are asked to express their support for high and low skilled migration. Both reveal a similar positive effect size, implying that respondents with higher education are more positive disposed towards migrants of either type. Model 3 estimates the effect of the HS-frame on the attitude. The result underscores the more positive evaluation of high skilled migrant inflow. Model 4 includes the interaction effect between education and the HS-frame. Our results confirm previous research and show that higher educated people in general hold more positive attitudes towards migration and that support for high skilled migration is stronger than support for low-skilled migration. Most importantly, the expectation that the effect of the HS-Frame should change its sign the higher educated respondents are, is not confirmed. Instead, the effect is overall positive and appears not to be moderated by education: i.e. independent of the respondents skill level, all respondents prefer high-skilled over low skilled migration. When relaxing the linearity assumption of education and using the education dummies in Model 5, one of the interactions 21

is significant (a negative sign of the lower secondary education HS-Frame). That is, the support enhancing effect of the high skill frame is a bit smaller for those respondents with lower secondary education compared to the baseline, those with upper secondary or vocational training. One reason for why we do not find supportive evidence for the labour market competition model might be that some of the respondents in our analysis are not (directly) affected by labour market competition any more. Restricting the sample to those respondents in the labour force should thus provide a more likely case to find evidence supporting the labor competition model (see also Hainmueller and Hiscox 2007 or Mayda 2006 for a similar approach). In a final step of our analysis we thus split our sample and compare those in the laborforce (the employed, self-employed and unemployed respondents looking for a job) with respondents out of the laborforce (i.e. retired, homemaker, disabled and unemployed not looking for work). We would assume that the reaction to the HS-Frame depends on respondents skill-level for those in the labor force (Model 6) but not for those out of the labor force (Model 7). And indeed we find differences between the two models: the interaction effect of HS-frame and education is significant for those in the labour force (and not for those out of labour force). Nevertheless, against the prediction of the labour competition model, high-skilled natives in the work-force increase their support for inflows of high skilled labour more compared to low skilled natives. Natives do not oppose immigration of workers with similar skill levels. Instead, highly-educated workers seem to attach a stronger value to likewise skilled labour. [Figure 4 about here] 22

The results from the models can further be investigated using the predicted probabilities to allow many migrants from model specification 5. Figure 4 highlights again that independent of a respondent s skill-level, high skilled migrants are preferred over low-skilled migrants. For both the groups of migrants, the probability increases with a respondent s education. The increase is slightly stronger for highly educated respondents. The highest support for immigration is thus observed for highly educated respondents towards highly-skilled immigrants. A finding that clearly contradicts expectations derived from the labour market competition model. Robustness checks: Alternative measures of skill-level and increasing salience of labour market competition As already outlined, the educational level may not fully capture respondents specific skill level and thus, fails to unambiguously test the labour market competition model. Concerns have been raised that first, education as an indirect measure for labour market competition may only very roughly capture actual labor market threat (Malhotra et al. 2013); and that the specific occupation or the industry someone is working in are more determinant to 23

conceptionalize exposure to labour market competition with immigrants. Second, the concerns point to the fact that education is highly correlated with other (noneconomic) attitudes also driving sentiments toward immigration and immigrants (Hello et al. 2006, Malhotra et al. 2013). Taking these critiques into account, we propose a series of robustness checks. The results are reported in Table 2. Model 1 is included as the reference model from the analysis including education as measure for skill level and is identical with model 6 in table 1. Model 2 conceptionalizes respondents skill level based on the ISCO scheme into 4 groups (see table A2 in the Appendix). The results in model two are consistent with the findings we showed in model 1. If anything the coefficient for the interaction term loses leverage and is not significant at conventional levels. Again, against the theoretical assumptions of the labour competition model, high-skilled natives increase their support for inflows of high skilled labour more compared to low skilled natives. Thus, also with this alternative measure of skill-level we do not find (clear) evidence for the labour competition model. In the following we restrict our analysis to (theoretically) more likely cases to find support for the labour market competition model; or in other words, we test the model for settings where competition might be more salient as the occupation specific share of immigrants coworkers are higher (see Hainmueller Hiscox Margalit 2015, Malhotra et al. 2013). For model 3 and model 4 we include a direct measure of labour market competition based on the EU Labour Force Survey from 2013. We restricted our analysis to occupations that are characterized by high competition. We relied hereby on the share of foreign-born in each occupation to construct our LCM1 indicator: LCM1 high indicates whether the share of foreign-born exceeds 24

the country average in the respondents occupation and LCM1 very high if the share of foreignborn is 25 percentage points higher than the national average. Examples of these highcompetition occupations include, for instance, science and engineering professionals, teaching professionals, nursing and midwifery professionals, but also shop salespersons, domestic, hotel and office cleaners, or farmers. By restricting our analysis to these high-competition occupations substantially we find again a more positive evaluation of high skilled migrant inflow. Interestingly, even if not significant at conventional levels, we find as theoretically expected that the effect of the HS-Frame changes its sign the higher educated respondents are. A very similar pattern is revealed by model 5 and 6: Following Polavieja (2016) and Ortega and Polavieja (2012) we include two indirect measures of labor market competition taking two occupational characteristics into account that limit outside competition: skill specialization and monitoring costs. We calculated occupation specific levels of skill specialization and monitoring costs. Higher values for skill specialization (LMC2) and monitoring costs (LMC3) are related to more outside competition. To sum up, we can confirm previous research in stating that the predictions of the labour market competition model clearly are not homogenously applicable to the entire society. We find supportive evidence only for the most likely cases and only when including more sophisticated measures for respondents skill levels. [Table 2 about here] 25

Table 2 Support for Highly Skilled and Low-Skilled Immigration Test of the Labor Market Competition Model with different skill measures in in high risk segments of the labor market (1) (2) (3) (4) (5) (6) Full sample (active) HS-Frame 0.567 ** (0.078) Education 0.264 ** (0.020) HS-Frame X Education 0.069 ** Skill Level 0.728 ** (0.067) (0.025) HS-Frame X Skill-Level 0.014 (0.023) LMC1: high 0.748 ** (0.096) LMC1: very high 1.114 ** (0.220) LMC2: high 0.795 ** (0.228) LMC3: high 0.783 ** (0.147) -0.005 (0.036) -0.126 (0.077) -0.045 (0.131) -0.013 (0.052) Skill-Level 0.226 ** (0.018) 0.268 ** (0.028) 0.361 ** (0.058) 0.360 ** (0.101) 0.206 ** (0.039) Controls Yes Yes Yes Yes Yes Yes Country FE Yes Yes Yes Yes Yes Yes N 11348 11225 4273 760 1408 2212 Ordered Probit Coefficients shown with standard errors in parentheses. All models include a set of covariates (age, gender, marital status, political ideology and income) and country dummies. The reference catgeory for the education dummies is 'upper secondary or vocational education' * p < 0.05, ** p < 0.01 The Fiscal Burden of Public Services The fiscal burden model argument leads to two competing expectations how people with different incomes react to the HS-Frame. On the one hand, the standard model (with fixed welfare benefits) suggests that high income respondents should have a stronger preference for high-skilled over low-skilled migration than low income respondents because they fear the increased fiscal burden (see Helbing & Kriesi 2014 for empirical support). On the other hand, an alternative prediction (assuming taxes to be fixed) suggests that low income respondents should be more supportive of high-skilled migration as low income respondents fear cuts in their welfare benefits (see Hainmueller and Hiscox 2010 for empirical support). Moreover, 26

whereas Hainmueller & Hiscox (2010) show that the latter relationship mainly holds in a context of high fiscal exposure, Helbing & Kriesi (2014) find strongest support for a positive interaction effect in a context of low fiscal exposure. [Figure 5 about here] Our data rather supports the standard argument of the fiscal exposure perspective that tax concerns are the dominant driver of the income gradient in attitudes towards migrants. Consequently, high income respondents preference of highly over low skilled migration is stronger compared to the premium that low income respondents attach to high skilled migration. When plotting the share of respondents that support allowing many migrants, over the different frames, different income quintiles and for high and low fiscal burden context seperately, two major patterns emerge (Figure 5). With increasing income respondents take more positive stances towards immigration. Moreover, the distributions of attitudes towards highly skilled migration and low-skilled migration are similar in the low and high fiscal exposure 27

condition. Except for the outlier of high fiscal exposure and the fifth income quintile, the support across income groups and the reaction to the treatment seem to be rather unchanged for the two different sets of countries. It is exactly this observation; however, that offers support for the perspective that the tax concern of high income earners is even stronger in a context of high fiscal exposure: In a context of high fiscal exposure, 14% of respondents among the highest income quintile would allow many low skilled migrants to come and live in the country. The HS-Frame leads to a drastic increase of support to 43% among the highest income quintile (+ 29 percentage points). In a context of low fiscal exposure, the increase in support among high-income earners due to the high-skill frame is 18 percentage points (from 15% in the low skill frame to 33% in the high skill frame). Based on this observation, it is cumbersome to infer in how far the fiscal burden perspective applies here. The following therefore turns to the analysis of ordered probit models to validate the expected patterns. [Table 3 about here] Table 3 Support for Highly Skilled and Low-Skilled Immigration - test of the Fiscal Burden Model (1) (2) (3) (4) (5) (6) Both Fiscal Exposure: High Fiscal Exposure: Low Both Fiscal Exposure: High Fiscal Exposure: Low HS-Frame 0.672 ** (0.041) 0.629 ** (0.077) 0.691 ** (0.049) 0.752 ** (0.034) 0.772 ** (0.062) 0.745 ** (0.041) income 0.033 ** (0.010) 0.027 (0.018) 0.036 ** (0.011) HS-Frame X Income 0.036 ** (0.012) 0.057 * (0.023) 0.027 (0.014) Q1-0.070 (0.040) -0.125 (0.073) -0.042 (0.047) HS-Frame X Q1-0.034 (0.053) -0.022 (0.098) -0.044 (0.063) Q2-0.028-0.048-0.019 28

(0.037) (0.067) (0.044) HS-Frame X Q2 0.010 (0.050) 0.007 (0.091) 0.011 (0.059) Q4 0.021 (0.036) 0.092 (0.064) -0.012 (0.043) HS-Frame X Q4 0.047 (0.048) -0.124 (0.087) 0.128 * (0.057) Q5 0.069 (0.038) -0.060 (0.071) 0.130 ** (0.045) HS-Frame X Q5 0.131 * (0.051) 0.364 ** (0.098) 0.034 (0.060) Controls Yes Yes Yes Yes Yes Yes Country FE Yes Yes Yes Yes Yes Yes N 19998 6486 13512 19998 6486 13512 Standard errors in parentheses Ordered Probit Coefficients shown with standard errors in parentheses. All models include a set of covariates (age, gender, marital status, political ideology and education) and country dummies. The reference catgeory for the income dummies is 'Q3' * p < 0.05, ** p < 0.01 Table 3 reports the results of six different model specifications that include interaction effects between the HS-frame and the income of a respondent. All models again include additional controls, country dummies and respondents educational attainments. The first model, estimated on the full sample, shows a positive interaction effect between the HS-frame and the income. This implies that high income people prefer highly skilled migration over low skilled migration more than low income respondents. This finding stands in sharp contrast to the findings in Hainmueller and Hiscox (2010) but rather supports the finding of Helbing & Kriesi (2014). As the subsample analysis in model 2 and 3 reveals, this significant interaction is mostly driven by countries with high fiscal exposure where foreign-born net contribution is negative (i.e. where they actually receive on average more benefits than they pay in taxes). Relaxing the linearity assumption of income and including the income quintiles as dummies, shows further that this is mainly because of the highest income quintile (Model 4). Whereas the premium attached to high skilled migration does not vary over the lowest four income quintiles, the 29