‘Immigrationization’ of welfare politics? Anti-immigration and welfare attitudes in context

Abstract

Several studies have shown that a person's attitude towards immigration affects his or her support for welfare redistribution. According to one view, negative attitudes towards immigration undermine support for welfare redistribution, as those who hold anti-immigration attitudes are thought to view immigrants as undeserving yet disproportionately drawing upon the welfare state. According to a competing view, however, anti-immigration attitudes awaken a person’s own economic insecurities that in turn spur support for welfare protection and redistribution. This article argues and finds substantial evidence in European public opinion that both of these mechanisms can be at play and have implications that depend strongly on a country’s national-level context. In particular, it is found that anti-immigration attitudes yield lower support for redistribution mainly when a respondent’s country faces more immigration, when welfare-state protections are generous, and when migrants actually rely more than natives on the welfare state.

Keywords:

• Immigration
• redistribution
• welfare state
• public opinion

The European refugee crisis has sparked heated debates about immigration. One of these debates involves whether European welfare states can cope with the increasing numbers of immigrants.¹ Many politicians – and not only those on the radical-right side of the political spectrum – argue that their country’s welfare state cannot cope with too much immigration. Sweden’s social democratic foreign minister, Margot Wallström, typified this view when she said ‘we cannot maintain a system where perhaps 190,000 people will arrive every year – in the long run, our system will collapse’.²

An important result of widespread public debate about immigration and the welfare state is that attitudes about welfare may become increasingly tied-up in the politics of immigration (see Sides and Citrin 2007). Plenty of research suggests that the politics of the welfare state, including attitudes towards welfare redistribution, might be influenced by the level and pattern of immigration flows and stocks into host countries (Soroka et al. 2006; Brady and Finnigan 2014; Burgoon et al. 2012; Crepaz 2008). But the consistent tying-together of immigration and social policy may also mean that attitudes towards welfare redistribution have become ‘immigrationized’ – i.e., support for welfare redistribution becomes dependent upon one’s attitudes about immigration.

Indeed, scholars focussed on the United States have found that welfare attitudes are influenced by conditions of race and ethnicity (Gilens 1995, 1996; Alesina and Glaeser 2004). And diversity in a region’s ethnic composition and negative attitudes about racial minorities shape native support for welfare redistribution (see also Fox 2004; Luttmer 2001). In a recent study, Garand et al. (2017) focus explicitly on what they call ‘immigrationization’ of welfare, where Americans who are negative about immigration are less likely to support the welfare state. This and other studies suggest that this link might reside in how a citizen’s felt disdain or fear towards migrants and immigration makes that citizen less supportive of policies dispensing benefits to beneficiaries seen as less deserving (Van Oorschot 2006). Citizens might also presume that immigrants are disproportionate beneficiaries of welfare programmes, straining the fiscal sustainability of welfare redistribution (Hanson et al. 2007).

Recent studies have demonstrated that such tying-together of immigration and welfare attitudes may also take place in Europe (see Larsen 2011; Boeri 2010; Senik et al. 2009). But in doing so, this literature has also shown that anti-immigration attitudes in that part of the world might lead to more rather than less support for redistribution (Finseraas 2009) – perhaps because anti-immigration attitudes awaken worries that migration undermines one’s own economic security, requiring social protection through welfare redistribution.

Taken together, these contrasting findings open up a genuine, unresolved controversy about how anti-immigration attitudes spills-over into the politics of the welfare state. On the one hand we have an ‘anti-solidarity effect’, where anti-immigration feelings fuel worries that welfare redistribution will go to less deserving residents. This may pose fiscal pressures undermining native access to state benefits, decreasing support for welfare redistribution. On the other hand we have a ‘compensation effect’, where anti-immigration feelings awaken worries about one’s own economic security. This increases support for welfare redistribution to redress insecurities. The empirical evidence for these rival perspectives differs in terms of country, coverage, and specification, rendering the disagreement empirically unresolved. But the more significant empirical shortcoming is that the competing arguments about how anti-immigration sentiment spills-over into welfare attitudes are premised on judgments of socioeconomic conditions that should be understood to vary than instead of just assumed.

In this article, we argue that the degree and direction of immigrationization of welfare-state politics depends on the socioeconomic context of a country with respect to actual immigration, the welfare state and migrant-dependence on that welfare state. We argue that anti-immigration or anti-immigrant sentiment among citizens, whatever its origins, should have off-setting implications for attitudes towards welfare redistribution: on the one hand awakening concerns that such redistribution might go disproportionately to the less deserving migrants one fears or disdains, translating into lower support for redistribution; but on the other hand awakening concerns about one’s own economic insecurities that foster higher support for such redistribution. More importantly, we argue that anti-immigrant attitudes will tend to have less positive or more negative implications for support for welfare redistribution under conditions likely to enhance the first and/or dampen the second of these dynamics of anti-immigration. Anti-immigration will more strongly dampen (less strongly spur) support for redistribution to the extent that there are large numbers of immigrants in one’s country; to the extent that social policy is already generous and hence a vulnerable target to overburdening by migrants; and to the extent that migrants actually disproportionately draw on the welfare state compared with their native counterparts.

Our empirical analyses, based on matching individual-level to aggregate data for 23 countries, finds substantial support for these claims. Patterns in these data suggest that anti-immigration attitudes tend to spur or soften support for welfare redistribution depending on measured foreign-born shares, the size of the welfare state, and the relative social-benefit dependency of migrants. This contextual moderation appears to be strong substantively. The anti-solidarity thesis tends to hold if a society faces high foreign-born shares, high social welfare spending, and high social-benefit dependency of migrants relative to natives. But the compensation thesis holds where a respondent’s society is ethnically homogeneous, social welfare expenditure is low, and when migrants tend to depend no more on social transfers and services than do natives.

‘Immigrationization’ and compensation and anti-solidarity effects

Most debate on how migration relates to welfare state politics has focussed on actual material patterns of migration (e.g. Brady and Finnigan 2014; Eger and Breznau 2017; Kulin et al. 2016; Schmidt-Catran and Spies 2016). The findings informing this debate are mixed. Some studies have found a positive or non-relationship between immigration and support for welfare (Brady and Finnigan 2014; Crepaz 2008; Hjerm and Schnabel 2012; Mau and Burkhardt 2009). Other studies, however, have found stronger, more consistent negative relationships between immigration and welfare support (Dahlberg et al. 2012; Eger 2010; Eger and Breznau 2017; Schmidt-Catran and Spies 2016). Most such studies assume that the mechanism linking immigration to welfare attitudes involves how individual citizens understand these issues: ‘whether immigration depresses support for social welfare should depend on what individuals think about immigration in general’ (Kulin et al. 2016: 3, italics in original). Important extensions of scholarship focussed on objective immigration, therefore, are empirical and theoretical studies that explore how attitudes towards migration and migrants might influence welfare state politics.

While less developed, such studies do exist and yield important findings for contemporary social policy development. This began with work on the relationship between attitudes about race and ideas about welfare redistribution, especially in the scholarship of Martin Gilens (1995, 1996). Gilens focussed on attitudes towards blacks and demonstrated that Americans’ opposition to welfare is, to a significant extent, rooted in ideas about African Americans. Many Americans believe that blacks are, more often than whites, recipients of welfare redistribution (Gilens 1999) and that African Americans are lazy (Gilens 1995, 1996), yielding attitudes of welfare politics that are ‘racialized’. Others have corroborated this link between racial attitudes and attitudes towards welfare states (see Alesina and Glaeser 2004; Fox 2004; Luttmer 2001).

Garand et al. (2017: 149) extend this reasoning to migration: ‘Americans might have shifted their attention from African Americans to immigrants when they think about welfare protections, precisely because of a similar fear that immigrants might absorb welfare resources’. Where Gilens (1999) considers welfare politics to be ‘racialized’, Garand et al. (2017) consider them to be ‘immigrationized’. That many citizens’ attitudes about immigration spill over into ideas about welfare might be due to (unskilled) immigrants indeed being more likely to be welfare recipients in the US (Camarota 2012) and other countries (Boeri 2010). But it could also reflect negative stereotypes about immigrants where immigrants are considered less deserving than natives (Lee and Fiske 2006; Van Oorschot 2006).

This negative correlation between attitudes towards immigration and welfare may well travel to European countries. Larsen (2011) compares the situation in the US with attitudes in Britain, Sweden, and Denmark, and finds that in all three European countries negative perceptions of the out-group reduce support for welfare redistribution. Similarly, Roemer and Van der Straeten (2005, 2006) find that anti-immigrant attitudes reduce the preferred level of redistribution in Denmark and France. And Senik et al. (2009) look at 22 European countries and find that perceived levels of migration tend to dampen support for welfare redistribution among respondents with negative ideas about immigrants. Finally, comparing 21 European countries, Finseraas (2008) reports that opposition to equal rights for immigrants has a negative effect on preference for redistribution in 21 European countries.³

However, Finseraas (2008) also argues and demonstrates that, under some circumstances, ideas about immigration could be related to attitudes about welfare in a different way: the belief that immigrants take jobs away may increase support for redistribution. Those with negative sentiments towards immigration might not be less, but more, likely to support welfare redistribution out of a desire to protect themselves and/or their ‘own’ native group (Finseraas 2008; Burgoon et al. 2012). Indeed, those who are exposed to the risk of income loss (Cusack et al. 2006), and those concerned about fellow (native) citizens with financial problems (Blekesaune and Quadagno 2003) are more likely to support redistribution.

These distinct lines of research generate two contrasting views: (1) one that emphasises an ‘anti-solidarity effect’ where negative feelings towards immigration lead to less support for welfare redistribution (Garand et al. 2017); and (2) a view that emphasises a ‘compensation effect’ where negative feelings towards immigration lead to more support for welfare redistribution (Finseraas 2008; see also Schmidt-Catran and Spies 2016).⁴ These competing positions and underlying empirics constitute genuine controversy about the politics of immigration and welfare states.

Immigrationization and the moderating role of socioeconomic context

We argue that both positions can hold or potentially cancel one-another out in a population. More importantly, we contend that either position can hold depending on the socioeconomic context of countries. Particularly, features of existing social policy and immigration should moderate whether anti-solidarity or compensation effects dominate in public political consciousness.

A citizen’s thinking about immigration and its implications can focus on the in-group (fellow natives) or out-group (immigrants) (Blumer 1958; Semyonov et al. 2006). For a given citizen both of these logics can apply at once, offsetting one another or with one or the other predominating. This translates into three competing hypotheses on the implications of anti-immigration attitudes that we treat as preliminary possibilities (H1a–H1c). The anti-solidarity effect dominates if natives with anti-immigration attitudes evaluate their own socioeconomic protection and the protection of the in-group as less urgent, and therefore shift their attention to the threat they believe is posed by the out-group. Here, disdain towards or fear of immigrants or immigration can yield concerns about or opposition to any government policy interventions that disproportionately help such ill-thought-of immigrants. Since such policy interventions include welfare redistribution involving tax-and-spend provisions of the welfare state, a first hypothesis involves an anti-solidarity effect, captured by arrow H1a in Figure 1:

Figure 1. Hypothesised relationship between Anti-immigration attitudes and Support for welfare redistribution.

Hypothesis 1a. Individuals expressing more anti-immigration attitudes will tend to also express lower support for welfare redistribution than do individuals expressing more pro-immigration attitudes

The opposite, compensation effect may dominate if natives with anti-immigration attitudes believe that the protection of themselves or their in-group is more important than exclusion of the out-group. Here, disdain towards or fear of immigrants or immigration can awaken concerns about one’s own economic security and position, and can spur support for government measures to address such concerns. Among these measures are government tax-and-spend provisions redistributing income, most obviously through redistributive welfare policies and services. Hence, a second hypothesis involves a compensation effect, captured by arrow H1b in Figure 1:

Hypothesis 1b. Individuals expressing more anti-immigration attitudes will tend to also express higher support for welfare redistribution than do individuals expressing more pro-immigration attitudes

Finally, these off-setting anti-solidarity and compensation effects may cancel-out one another, yielding no net effect on support for redistribution. This constitutes the null hypothesis, captured by H1c in Figure 1:

Hypothesis 1c. Individuals expressing more anti-immigration attitudes will tend to express neither higher nor lower support for welfare redistribution than individuals expressing more pro-immigration attitudes.

Our three remaining hypotheses build on the premise that such contrasting implications of anti-immigration attitudes are likely moderated in predictable ways by aggregate socioeconomic conditions of migration and the welfare state. A few studies have devoted attention, albeit limited, to how contextual variables condition the way immigration attitudes affect welfare state support. Finseraas (2008) argued and found some support for type of welfare state as a moderating variable: that there might be a negative moderating effect in social democratic welfare states because their tax-financed and rights-based nature makes the issue of collective solidarity more salient; and that this moderation effect might be more modest, and a compensation effect more salient, in continental welfare states more based on earnings-based contributions (Finseraas 2008: 413–4). Schmidt and Spies (2014), using a different research design, do not replicate this finding, but they do argue for and find evidence that another context-level variable moderates the effect of immigration attitudes on welfare support: the more political parties emphasise policies related to immigration, the less natives with anti-immigrant attitudes support welfare.

Following these studies, we also expect that the relationship between anti-immigration attitudes and welfare support is likely moderated by macro-level socioeconomic variables. Our contribution, however, involves arguments as to which socioeconomic conditions can be expected to do the moderating. As summarised graphically in H2, H3, and H4 in Figure 1, our focus is on exploring the possible moderating role of the three political economic conditions most obviously at play in ‘immigrationization’ of welfare-state politics: levels of immigration, welfare state effort, and migrant welfare dependency.

First, there are the possible moderating effects of a country’s actual immigration or ethnic heterogeneity. If the out-group is large, natives with anti-immigration attitudes can be expected to fear that welfare redistribution will benefit the out-group disproportionally, and, hence, will not support redistribution. However, if the out-group is only small, those with anti-immigration attitudes will not experience a threat coming from this group and will therefore be more likely to focus on their own (group’s) position and support for redistribution. In homogeneous societies, hence, natives with anti-immigration attitudes conceive of the welfare state primarily as a system in which one’s ‘own people’ are helped, whereas in more heterogeneous countries natives see immigrants as ‘others’ who may not deserve the same level of welfare support as one’s ‘own’ people (Garand et al. 2017).⁵ As argued above, various studies have shown that actual levels of immigration, or ethnic heterogeneity, have significant implications for individuals’ attitudes towards welfare distribution (Eger 2010; Burgoon et al. 2012; Larsen 2011; Schmidt-Catran and Spies 2016).⁶ Here, however, we are less interested in the direct effect of sum immigration or ethnic heterogeneity on welfare support than in how actual immigration might moderate the relationship between immigration attitudes and welfare support. This leads to the following hypothesis, summarised in Figure 1 as H2:

Hypothesis 2. An individual’s anti-immigration attitude will tend to decrease (increase) that individual’s support for welfare redistribution to the degree that existing foreign-born stocks are high (low).

Second, the relationship between attitudes about immigration and support for welfare may depend also on the size or generosity of the welfare state. Plenty of scholarship models welfare attitudes as a function of existing levels (or kinds) of welfare protections (see Arts and Gelissen 2001; Dallinger 2010; Jæger 2013; Larsen 2008; Mau 2004; Schmidt-Catran 2016). And Finseraas (2008) has considered the possibility that various attitudes towards migrants might have implications for support for redistribution that might be different in social-democratic versus conservative welfare states.

We build on such insight, but focus on the more basic and broader issue of how the existing size of the welfare state might be an important moderating variable in the process of immigrationization. We expect that anti-immigrant attitudes will foster rejection of income redistribution to the extent that natives have something in welfare-state redistribution that they want to defend: that generous assistance allows citizens to feel secure about their own and their fellow natives’ existing socioeconomic protections. We expect this to be particularly the case in states characterised by generous welfare states with high social expenditure. Conversely, if the generosity of a welfare state and social expenditure is low, and citizens feel less secure about their economic positions, anti-immigration attitudes will lead to more demand for redistribution. We focus in particular on actual government spending, and expect that higher ex ante welfare spending effort, or social policy generosity, may increase the tendency of anti-immigrant sentiment to awaken aversion to such generous social benefits going to help those deemed less worthy of assistance or redistribution.⁷ As an empirical matter, Schmidt and Spies (2014), focussing on 14 European countries, found few such effects on the way to their focus on party framing and signalling. Our theoretical priors, however, suggest that welfare effort should be a strong moderator should one consider more substantial variation in anti-immigration attitudes and social policy support across more countries and time. We therefore hypothesise the following:

Hypothesis 3. An individual’s anti-immigration attitude will tend to decrease (increase) that individual’s support for welfare redistribution to the degree that existing welfare state spending or generosity is already substantial (low and modest).

Finally, the position of migrants relying on social-policy protection can be expected to alter how anti-immigrant attitudes affect welfare sentiments. In particular, there is substantial variation in how much migrants are net beneficiaries of non-pension welfare redistribution and social services relative to their native counterparts (Boeri 2010). And given the common worries about particularly non-contributory social benefits being welfare magnets and problems for host societies, such differences in dependency can substantially moderate the implications that actual levels of immigration have for support for welfare redistribution (Burgoon 2014).⁸ Importantly, such differences may also moderate whether we see anti-solidarity or compensation effects in welfare politics. Anti-immigration attitudes more likely awaken individual concerns than concerns about immigrant welfare-reliance to the extent that migrants rely on non-pension social benefits less than or no more than natives. But where migrants rely disproportionately more on such social benefits, anti-immigration attitudes can more readily awaken opposition to welfare redistribution – precisely because welfare in such settings is going to those who are feared or disdained. Hence, our final hypothesis, graphically summarised as H4 in Figure 1:

Hypothesis 4. An individual’s anti-immigration attitude will tend to decrease (increase) that individual’s support for welfare redistribution to the degree that migrants are more (less) dependent on non-contributory social benefits than are natives.

Other macro- and micro-economic and -political factors might also moderate the effects of anti-immigration sentiments for welfare states. And one can imagine more complicated relationships, such as triple interactions among the hypothesised moderating factors. This study, however, focuses first and most fully on the moderating role of those basic features of immigration and welfare states. In order to better understand if and how attitudes about welfare state are immigrationized, it is of essential importance to assess if, and if yes, how, these fundamental socioeconomic circumstances (beyond welfare state types) condition such immigrationization.

Data and methods

To assess the above Hypotheses, we explore how measures of attitudes towards immigration are associated with measures of attitudes towards welfare redistribution, and we investigate the extent to which this association is conditional upon macro-level characteristics. This approach builds on previous research designs exploring ‘immigrationization’, but it requires large and high-quality samples, ideally including substantial variation across both time and countries, and with well-worded questions on issues of immigration and redistribution. The European Social Survey (ESS) provides such quality coverage and repeated questions on immigration and welfare, that can be linked to country-level data on relevant socio-economic context. We include 7 waves of the ESS (2002, 2004, 2006, 2008, 2010, 2012, and 2014), focussing on 23 countries, including seven Central and Eastern European countries (CEEC) known to be more disproportionately anti-immigrant than their West European counterparts: Austria, Belgium, Switzerland, the Czech Republic, Germany, Denmark, Estonia, Spain, Finland, France, Great-Britain, Greece, Hungary, Ireland, Italy, Luxembourg, the Netherlands, Norway, Poland, Portugal, Sweden, Slovenia, and Slovakia.⁹ Our baseline estimates focus on the full sample of residents, though we also consider and discuss specifications limited to natives or native-citizens in robustness and sensitivity tests.

Our (individual-level) dependent variable is Support for redistribution, measured by how much respondents agree with the proposition: ‘The government should take measures to reduce differences in income levels’. Answer-categories range from ‘1’ (strongly disagree) to ‘5’ (strongly agree).¹⁰ Our baseline specifications focus on binary coding of this parameter: Support for redistribution (binary), where 1 = ‘strongly or somewhat agree’ and 0 = ‘neither disagree nor agree, or somewhat or strongly disagree’. Panel (a) in Figure 2 summarises the sample-country means, smoothing-over the seven waves. Support for welfare redistribution tends to be stronger where such welfare protection and redistribution is less developed, despite outliers (e.g. UK and France). This suggests that ex ante social protection efforts should be taken into account, such as through transformations of the dependent variable (e.g. a respondent’s deviation from the country-year mean). Our robustness tests therefore consider this and other alternative specifications.

Figure 2. Country means of Support for redistribution (binary) and Anti-immigration (scale).

Our main individual-level independent variable is a person’s Anti-immigration sentiment. In our baseline specifications, we measure this with a scale consisting of three 11-point items on immigrants and immigration repeated in all ESS waves. Respondents could answer on a 0–10 scale what they thought of the following three questions: (1) ‘Would you say it is generally bad or good for [respondent’s country]’s economy that people come to live here from other countries?’ (0 = bad; 10 = good); (2) ‘Would you say that [country]’s cultural life is generally undermined or enriched by people coming to live here from other countries?’ (0 = undermined; 10 = enriched); and (3) ‘Is [country] made a worse or a better place to live by people coming to live here from other countries?’ (0 = worse; 10 = better). We have reversed the scale of these items so that ‘0’ refers to an attitude that is strongly pro-immigration and ‘10’ to a strongly anti-immigration attitude (Cronbach’s Alpha = 0.84). Panel (b) in Figure 2 shows the country-means for this Anti-immigration (scale). Icelandic respondents tend to be the most pro-immigration, Greek respondents the most anti-immigration. When assessing relationships between immigration attitudes and redistribution attitudes, Schmidt and Spies (2014) distinguish between the first item (concerning the economy) and the second item (concerning culture). In robustness checks, we consider possible differences across these different aspects of anti-immigration sentiment.

In order to measure the aggregate-level moderating conditions highlighted in Hypotheses 2–4, we rely on both the ESS and aggregate data sources – with country summaries of our baseline measures in Figure 3 (panels a through c). Foreign born is our aggregate-level measure of immigration stocks, as a percent of a country’s population. It might also measure a society’s ethnic homogeneity/heterogeneity (see Brady and Finnigan 2014; Eger 2010; Mau and Burkhardt 2009). Coming from the OECD (2018), the variable is lagged one year to address likely delays in the working of material conditions on attitudes. The parameter is important to explore Hypothesis 2, but also as an aggregate control in shaping the relationship between Anti-immigration sentiments and Support for redistribution. Panel (a) in Figure 3 shows that the country means capture the considerable spread in the European sample, from Poland’s low of less than 2 percent to Luxembourg’s whopping 34 percent.¹¹

Figure 3. Country means of Foreign-born percent, Social welfare expenditure, and Migrant social-benefit dependency.

In order to measure size of welfare-state effort, we used in our baseline estimates Social welfare spending (percentage of GDP, lagged one year), drawing on OECD and Eurostat databases to maximise the country-year spread of our ESS sample. There are, of course, alternative measures, most obviously in terms of generosity of programmatic benefits rather than spending-based metrics (Green-Pedersen 2004). This is important particularly for those seeking to explain welfare state effort or generosity. Our focus on the consequences of social policy effort as an explanatory factor, however, make the spending-based baseline useful to gauge the economic weight of welfare effort and take-up, reasonably captured by total social-policy provisions as share of GDP (Brady 2005).¹² Such social policy effort, in any event, is important not only to explore the moderating role of such effort, testing Hypothesis 3, but also as a control relevant to how individual positions on Support for redistribution likely reflect ex ante social policy provisions at the time of the survey. Hence, the parameter enters as a control in all estimations. As panel (b) in Figure 3 shows, the sample ranges from Estonia’s social spending at roughly 15 percent of GDP to France’s spending at 30 percent of GDP.

Most difficult to measure of our moderating variables is social-benefit dependency of the foreign-born versus the native-born. The difficulty is the lack of comparable measures across European countries in take-up rates or use of various features of social policy, for a significant number of countries and years. To approximate such dependency we use the ESS dataset itself and its high-quality sampling properties, estimating the likelihood that foreign-born respondents rely on non-pension social benefits for their income. Such estimation is based on respondents’ answers to a question in the ESS panel on sources of income, including the ‘unemployment or redundancy benefits’ and other ‘government social benefits’. We construct individual-level incidence of (non-pension) Migrant social benefit dependency (1 = income mainly from unemployment/redundancy benefit or other social benefits; 0 = other sources of income). Using probit models, we then estimate the marginal likelihood that a respondent’s being foreign-born (1 = born abroad; 0 = born in country of residence) predicts that respondents rely on social benefits for their income (i.e. Migrant social benefit dependency = 1), controlling for age, family-composition, and gender of respondents so as to control for possible omitted variable bias and hence isolate the effect of foreign-born basis of dependency. The results provide country-year-specific integration measures of Migrant social benefit dependency, which is the z-statistic of the marginal effect (∂F/∂x) that being foreign-born predicts Migrant social-benefit dependency. The resulting estimates correlate highly with studies of social benefits based on larger samples but for smaller cross-sections of countries (and years).¹³ As measured, Figure 3 panel (c) shows country means for the seven ESS waves: Migrant social benefit dependency is lowest in Portugal, Italy, and Hungary – where migrants are less likely to depend on social benefits than are natives – and highest in Austria, Belgium, and Luxembourg.

Finally, we consider individual-level control variables known to correlate with attitudes towards welfare redistribution, often in interaction with other conditions relevant to economic insecurities (Häusermann et al. 2015). Education is an ordinal variable with 5 categories: (1) ‘less than lower secondary education’; (2) ‘lower secondary education completed’; (3) ‘Upper secondary education completed’; (4) ‘Post-secondary non-tertiary education completed’; and (5) ‘tertiary education completed’. For income, we rely on subjective income, how respondents feel about their household income, ranging from ‘finding it very difficult on present income’ (1) to ‘living comfortably on present income’ (4).¹⁴ We assessed individuals’ unemployed status with a binary variable: 0 = employed, 1 = unemployed. We further assessed a respondent’s age, gender (1 = female), and religiosity (based on how often respondents say they attend religious services, from 1 [never] to 7 [every day]). Finally, we assessed if someone lives with a husband/wife or partner (1 = yes); if one has children living in the household (1 = yes); and if one is foreign-born or not (1 = born abroad; 0 = born in country of residence). Table 1 provides an overview of all variables.

Table 1. Summary statistics.

Variable	Obs.	Mean	Std. dev.	Min	Max
Support for redistribution (binary)	260,856	0.71	0.45	0	1
Support for redistribution (categorical)	260,856	3.83	1.04	1	5
Anti-immigration (scale)	245,029	4.85	2.08	0	10
Foreign born	266,199	10.68	5.73	2	44
Social welfare spending	266,205	23.06	4.23	13	32
Migrant soc. benef. dependency	250,318	1.75	1.50	−2	7
Education: no second	264,293	0.14	0.34	0	1
Education: lower second	264,293	0.19	0.40	0	1
Education: upper second	264,293	0.38	0.49	0	1
Education: post-second	264,293	0.04	0.19	0	1
Education: tertiary	264,293	0.25	0.43	0	1
Income (subjective)	260,238	3.02	0.85	1	4
Unemployed	265,199	0.05	0.21	0	1
Age	256,326	48.84	17.88	18	123
Female	265,888	0.53	0.50	0	1
Foreign-born respondent	265,860	0.09	0.28	0	1
Religiosity	264,997	2.56	1.54	1	7
Partner	264,722	0.60	0.49	0	1
Children	265,426	0.38	0.48	0	1

Our analysis of Hypotheses 1–4 is based on models focussed on how Anti-immigration sentiments shape Support for redistribution in ways moderated by Foreign born, Social welfare spending, and Migrant social benefit dependency. Our baseline models consider all countries and all aforementioned individual and aggregated variables. Since the inclusion of macro-level variables violates assumptions of the standard OLS, logit or probit models (Gelman and Hill 2006), we estimate two-level multilevel random-intercept logit models, wherein individuals (level 1, n = 212,572) are nested in country-years (level 2, n = 128). To minimise inefficiency posed by cross-level interactions, our baseline models consider each cross-level interaction in separate models (Stegmueller 2013). The resulting specifications take the following general form: $${\mathrm{Support\; Redistribution}}_{ij}=y_{00}+y_{01}{\mathrm{AntiImmigration}}_{ij}+y_{10}{\mathrm{ContextualVariables}}_j+y_{11}{\mathrm{AntiImmigration}}_{ij}\times {\mathrm{ContextualVariables}}_j+y_{02}{\mathrm{IndividualControls}}_{ij}+u_{0j}+e_{ij}$$

We explore four models within a given specification: one without the above interaction term, to assess Hypotheses 1a–1c; and then one for each of the three interactions between Anti-Immigration sentiment and a given contextual variable to test Hypotheses 2–4. In our robustness and sensitivity analysis, we consider alternative measures of the dependent, independent, and moderating variables, alternative estimators (such as fixed-effects multi-level logits, multi-level ordered logits of categorical dependent-variables), and an alternative embedding of the multi-level models.

Before proceeding to the results, it is worth flagging limitations of our empirics. The generic challenge we face involves looking at how attitudes are associated with other attitudes. Associations almost certainly reflect not only the direction of relationships relevant to our hypotheses 1–4, but also reverse causality – as redistributive attitudes can be expected to shape, and not just be shaped by, attitudes towards migration. And this dynamic might be further complicated by the deeper historical underpinnings and relations between politics of migration and of welfare redistribution. Our analysis takes some steps to address these issues, but we face serious limits in what can be done to isolate the direction of causality. Multi-country survey instruments, like the ESS, lack a true panel dimension or experimental treatments that would allow a stronger causal identification. Hence, our findings are suggestive, significant steps to testing the four hypotheses developed above.

Findings

Table 2 shows our baseline results. The intra-class correlations (ICCs) are consistently significant – evidence of country-level clustering that justifies multi-level estimation. The controls perform in line with previous studies of support for redistribution. Respondents with lower (subjective) incomes, the unemployed, and the less-educated are significantly more likely to support welfare redistribution, as are older, female, more secular, and attached respondents. None of the macro-level variables exerts a statistically significant direct effect on Support for redistribution.

Table 2. Anti-immigration attitudes and support for redistribution.

	M1	M2	M3	M4
Anti-immigration (scale)	−0.009*** (0.003)	0.018** (0.006)	0.160*** (0.014)	0.009* (0.004)
Anti-immig X foreign born		−0.002*** (0.000)
Anti-immig X social welfare spending			−0.007*** (0.001)
Anti-immig X migrant soc. ben. dep.				−0.010*** (0.002)
Foreign born	−0.004 (0.009)	0.007 (0.009)	−0.003 (0.009)	−0.004 (0.009)
Social welfare spending	−0.005 (0.013)	−0.005 (0.013)	0.029* (0.013)	−0.005 (0.013)
Migrant soc. benef. dependency	−0.060 (0.037)	−0.059 (0.037)	−0.059 (0.037)	−0.010 (0.038)
Education (base = no second)
Lower second	0.035 (0.022)	0.036 (0.022)	0.032 (0.022)	0.038+ (0.022)
Upper second	−0.058** (0.021)	−0.057** (0.021)	−0.060** (0.021)	−0.055** (0.021)
Post-second	−0.193*** (0.031)	−0.192*** (0.031)	−0.199*** (0.032)	−0.192*** (0.031)
Tertiary	−0.473*** (0.021)	−0.473*** (0.021)	−0.481*** (0.021)	−0.472*** (0.021)
Income (subjective)	−0.354*** (0.007)	−0.354*** (0.007)	−0.354*** (0.007)	−0.354*** (0.007)
Unemployed	0.155*** (0.026)	0.156*** (0.026)	0.156*** (0.026)	0.156*** (0.026)
Age	0.009*** (0.000)	0.009*** (0.000)	0.009*** (0.000)	0.009*** (0.000)
Female	0.270*** (0.010)	0.271*** (0.010)	0.269*** (0.010)	0.270*** (0.010)
Foreign-born respondent	−0.028 (0.018)	−0.032+ (0.018)	−0.025 (0.018)	−0.028 (0.018)
Religiosity	−0.011** (0.004)	−0.011** (0.004)	−0.012** (0.004)	−0.011** (0.004)
Partner	0.018 (0.012)	0.018 (0.012)	0.019 (0.012)	0.018 (0.012)
Children	−0.054*** (0.012)	−0.054*** (0.012)	−0.055*** (0.012)	−0.054*** (0.012)
Constant	2.031*** (0.308)	1.897*** (0.309)	1.220*** (0.316)	1.931*** (0.308)
Observations	212,572	212,572	212,572	212,572
Number of country-years	128	128	128	128
Log-likelihood	−117556	−117543	−117481	−117538
Chi-2	6788	6812	6921	6820

DV:1 = Support for redistribution; 0 = Do not support redistribution. All models are random-intercept logit models (individuals within country-years), standard errors (in parentheses).

***p < 0.001, ** p < 0.01, * p < 0.05, + p < 0.10.

With respect to the possible direct effects of Anti-immigration (scale) relevant to Hypothesis 1a-1c, the baseline results in Model M1 suggest statistically significant negative effects of Anti-immigration (scale), meaning that respondents harbouring anti-immigration sentiments tend to be less likely to support redistribution than those with more pro-immigration attitudes. In substantive terms, the predicted effect is quite modest, captured graphically in Figure 4. That figure shows the predicted change in the probability of supporting government redistribution across the full sample distribution in Anti-immigration (scale), holding all other parameters at their means or medians. The full range of anti-immigration sentiment predicts or explains, as it were, a mere one-percent change in Support for redistribution. We have, hence, statistically significant but substantively modest support for Hypothesis 1a, the anti-solidarity effect emerging from Garand et al.’s (2017) study.

Figure 4. Predicted Support for redistribution as a function of Anti-immigration (scale).

Models M2–M4 test Hypotheses 2–4, respectively. All show significant negative coefficients for the interaction terms involving Anti-immigration (scale) on the one hand and Foreign born (M2), Social welfare spending (M3), and Migrant social benefit dependency (M4) on the other. In other words, the interactions are consistently in line with Hypotheses 2, 3, and 4. Anti-immigration (scale) tends to more negatively (less positively) affect the probability of supporting welfare redistribution where and when respondents live in settings with higher Foreign born percent, higher Social welfare spending, and higher Migrant social benefit dependency (relative to natives).¹⁵

What the actual predicted conditional effects are of Anti-immigrant (scale) for Support for redistribution cannot simply be read-off the interaction terms and components. Figure 5, however, summarises the predicted marginal effect of Anti-immigrant (scale) across the full sample variation for each moderating condition. The schedules represent the coefficient values of Anti-immigration (scale) in predicting Support for redistribution across the sample variation in the moderating variables (Foreign-born percent in panel (a), Social welfare expenditure in panel (b), and Migrant social-benefit dependency in panel (c)). Where both the upper and lower confidence intervals are above the value ‘0’ we have a statistically-significant compensation effect, and where both are below this value we have a statistically-significant anti-solidarity effect. The left-hand and right-hand vertical lines show the percentile of the moderating parameter at which such effects of positive and negative significance are reached, respectively. As the panels show, the moderating effects are strong: at low levels of moderators, Anti-immigration sentiments predict statistically-significant increases in Support for redistribution, and at high levels of moderators, Anti-immigration sentiments predict significant decreases in Support for redistribution. As can be seen, the anti-solidarity effect holds significantly for the larger share of total variation in both Foreign born and Migrant social benefit dependency – seen by the respective vertical lines capturing cut-off points. Anti-immigrant sentiment is pretty much as likely to have a compensation effect as an anti-solidarity effect, depending on the level of social policy effort.

Figure 5. Predicted marginal effects of Anti-immigration (scale) on Support for redistribution, across levels of Foreign-born percent, Social-welfare expenditure and Migrant social-benefit dependency.

The substantive size of anti-solidarity and compensation effects can be appreciated using another counterfactual analysis, summarised in a final Figure 6. This shows the predicted level of Support for redistribution across the range of Anti-immigration (scale), conditional upon low levels (10^th percentile) versus high levels (90^th percentile) of Foreign born, Social welfare spending, and Migrant social benefit dependency. This shows again that at low levels of moderating contextual conditions, Anti-immigration sentiments tend to significantly increase the probability of supporting redistribution, whereas at high levels of such moderating conditions Anti-immigration sentiments have the opposite, negative implications. For instance, at high levels of Foreign born (panel (a)), Social welfare spending (panel (b)) and Migrant social benefit dependency (panel (c)), Anti-immigration sentiments yield appreciably stronger decreases in predicted support for redistribution than the average predicted drops seen in the un-moderated results of Model (1). The anti-solidarity effect is strongest amidst high levels of welfare expenditures – roughly fifty percent greater than the un-moderated results.

Figure 6. Predicted Support for redistribution, across levels of Anti-immigration (scale), at low versus high levels of Foreign-born percent, Social-welfare expenditure and Migrant social-benefit dependency.

Robustness checks

The results from our baseline broadly hold-up to a range of robustness and sensitivity tests (see online appendix for fuller discussion). First, working with our baseline measures of explanatory, moderating, and outcome variables of interest, we explored alternatives to our baseline multi-level model specification: (1) a categorical instead of dichotomous DV, and, hence, ordered logits; (2) addition of country dummies to (further) mitigate remaining heteroskedasticity and omitted variable bias; (3) year dummies in addition to country dummies; (4) country- and year-fixed effects (with or without random intercepts of these or other embedding); (5) different multi-level nesting structures (e.g. two-way individuals within countries, three-way individuals within countries within years, etc.) (See online appendix Table A1, supplementary material). Virtually all of these alternative specifications corroborate our baseline results. The exception is a three-level model with individuals nested in country-years, nested in countries. Here the results deviate from the baseline in that only the interaction between Anti-immigration and Social welfare spending remains statistically significant. This recommends extra caution in interpreting the other moderating conditions, but it bears emphasising that such a specification is suboptimal: A three-level model is unnecessarily complex for our purposes, where most of our variation exists between country-waves (the basis of our baseline model structure).¹⁶

Second, we considered alternative measures of our key variables of interest. The results hold up and are corroborated by further exploration of attitudes towards welfare redistribution. For instance, the baseline support for Hypotheses 2– 4 holds should one focus on how much a respondent’s support for redistribution deviates from his or her country mean; and it also holds in the ESS wave gauging respondent support for unemployment assistance (as opposed to the baseline measure of support for government redistribution) (see online appendix Table A2, supplementary material). Interestingly, the results also find support in how Anti-immigration and the macro-level moderating conditions interact in shaping related attitudes about social policy that are upstream mechanisms linking immigrationization on the one hand and support for redistribution on the other. The negative moderation of anti-immigration applies, for instance, to respondent attitudes that migrants are deserving of immediate and full access to social policy, and that social benefits need not be too great a strain on the economy (see online appendix Table A2, supplementary material).

Relatedly, the results hold-up to alternative specifications of the key explanatory and moderating conditions of interest. For instance, they hold up should one consider alternative measures of Anti-immigration sentiment, by looking individually at each of the components (one focussed on economic effects of migrants, one on cultural effects, and one on general effects for one’s country) constituting our baseline scale (see online appendix Table A3, supplementary material). More importantly, the results also hold up to quite diverse specifications of our macro-level moderating factors of interest. Instead of the baseline focus on social welfare spending, we get similar support for Hypotheses 2–4 by focussing on a measure of welfare generosity – net replacement rates of unemployment assistance programmes – or on a measure of explicit redistribution (the difference between pre- and post-tax-and-transfer GINI index, presumably capturing actual tax and social policy interventions) (see online appendix Table A4, supplementary material). And instead of the baseline focus on foreign born stocks, we also get similar results should we focus on net migration rates or on non-EU foreign born stocks (see online appendix Table A4).

Third, we considered an array of alternative and additional controls for factors that might alter the ways our core macro-level variables (migration levels, social policy generosity, and migrant welfare dependency) moderate immigrationization. For instance, the social policy or redistributive effort on which our arguments focus (Hypothesis 3 in particular) might be importantly epiphenomenal of welfare state regime types, the institutional clusters in the design of social policies that have been found by some actors to distinguish worlds of welfare (Esping-Andersen 1990; Hemerijck 2013). Welfare regime types might be important to immigrationization. But controlling for welfare regime types turns out not to substantively or statistically alter the baseline story in support of Hypotheses 2–4 (see online appendix Table A5, supplementary material). Nor does controlling for various other macro-economic conditions like inequality, growth, unemployment, and/or exposure to the global economic and debt crisis (see online appendix Table A6, supplementary material). Nor does controlling for or taking account of interactions with other individual-level conditions, including other socio-economic positions (e.g. occupation) or political sentiment (e.g. attitudes on parties, government capacities, race or altruism). Altogether, these and other specifications suggest that the jury may be out on the net direct effects of Anti-immigration sentiments, with support for Hypotheses 1a–1c depending on how one measures redistribution or anti-immigration sentiments. But we have strong support for our Hypothesis 3 about the role of welfare state effort, and moderately strong support for Hypotheses 2 and 4 on foreign-born stocks and migrant welfare dependency, respectively.

Conclusion

This article has explored the claims that not only do actual material developments in migration flow, but also the public’s attitude towards migration might spill-over into the politics of welfare states. We argue and find evidence that anti-immigration sentiments have off-setting implications for the positioning of voters on welfare redistribution – in some ways awakening concerns that migrant outsiders will unfairly benefit from and perhaps overburden welfare state provisions, but in other ways awakening worries about personal economic insecurities that necessitate social protection. Our principal arguments and survey evidence suggest that macro-level developments in actual migration and the welfare state significantly moderate which of these off-setting dynamics dominates. A robust pattern in the survey empirics is that anti-immigration sentiments tend to undermine support for welfare redistribution (anti-solidarity effect) when and where ex ante foreign-born stocks, actual social-welfare spending, and migrant dependency on social benefits (relative to natives) are all high. Yet such anti-immigration sentiments can actually undergird support for redistribution (compensation effect) where foreign-born stocks, social spending and migrant welfare dependency are low. These patterns go beyond an econometric whim and reflect an important pattern in how welfare politics can be immigrationized in negative and positive directions, depending on a country’s actual immigration and welfare conditions.

These conclusions provide no definitive word on how attitudes towards immigration and the welfare state are linked. We cannot rule out that the interactions capture a different moderating story of anti-immigrant sentiments altering the way macro-level developments like foreign-born stocks play out for attitudes towards redistribution. And we cannot rule out the possibility that the associations we identify may partly reflect attitudes towards redistribution shaping attitudes towards immigration and not the other way around. These possibilities are endemic to the observational data which we (and all other scholars exploring ‘immigrationization’) have used. Whether the Hypotheses 2–4 survive other data and estimation procedures must await further research. We welcome such research, in particular theoretical and empirical exploration into how immigration attitudes spill over into social policy attitudes, and exploration of other ways in which macro- and micro- political and economic conditions of individuals and the polities within which they work and live might moderate such spilling-over. Particularly, such exploration should develop and study more fine-grained measures of attitudes towards immigration and its risks and benefits, and attitudes towards welfare states – including more information on migrants and welfare. And we support investment in combining such data with experimental treatments tiered to exploring the hypotheses on anti-solidarity and compensation effects.

In the meantime, the substantial over-time and cross-country coverage and high quality sampling of the ESS does provide important hints with respect to immigrationization of welfare redistribution in Europe. The results are important and not altogether welcome news for the future of social protection. The moderating macro-level conditions revealed to be ‘switch-men’ in the politics linking migration to redistribution attitudes – foreign-born stocks, welfare spending as a share of GDP, and migrant welfare-dependence – have developed in ways that foster more anti-solidarity than compensation effects of citizens’ attitudes towards immigration.

Acknowledgements

Earlier versions of this article were presented at the University of Amsterdam, University of Mannheim and the European Consortium of Political Research (ECPR). For their constructive criticisms and suggestions, the authors thank participants at these venues and also particular thanks to the editors of West European Politics and the anonymous reviewers.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Notes on contributors

Brian Burgoon

Brian Burgoon is a Professor in the Department of Political Science at the University of Amsterdam (UvA), and the director of the UvA’s Amsterdam Institute for Social Science Research (AISSR). His research focuses on the political conflicts surrounding economic globalisation.

Matthijs Rooduijn

Matthijs Rooduijn is Assistant Professor in the Department of Political Science, University of Amsterdam (UvA). His research focuses on the rise of populist and far left/right parties and its causes and consequences – in particular regarding public opinion and voting behaviour.

Notes

1 See: http://www.bbc.com/news/world-europe-34272111, http://www.theatlantic.com/international/archive/2016/02/welfare-state-refugees-europe/463272/, http://www.nationalreview.com/article/426410/europes-welfare-states-attract-refugees-migrants-looming-crisis (last accessed March 3rd 2016).

2 http://www.thelocal.se/20151030/in-the-long-run-our-system-will-collapse-in-sweden (last accessed March 3rd 2016).

3 See for a discussion on the effect of anti-immigrant sentiment on different types of welfare, Goldschmidt (2015).

4 Note that Finseraas (2008) and Schmidt and Spies (2014) have argued that a distinction should be made between economic and cultural attitudes toward immigration. We assess the difference between these two types of attitudes below, in the section on robustness checks.

5 Note that the argument presented here concerns immigrants in general, and not non-EU or non-western immigrants specifically. Immigrants from outside the EU or from non-western countries are feared more strongly in economic or cultural terms. But the moderation effect presented here should also occur vis-à-vis immigrants from any and all countries, including and perhaps especially EU/western countries. The unskilled Somalian-born might be feared, but so is the sometimes-unemployed Polish construction worker. The article focuses on the full foreign born population, and we leave the focus on non-EU migrants to robustness discussion (see below).

6 Senik et al. (2009) investigated whether perceived presence of immigrants moderates how immigration attitudes affect welfare support. Negative ideas about immigrants were associated with less welfare support independently of the perceived presence of immigrants. However, their study assesses perceived instead of the actual presence of immigrants.

7 We also consider, as supplements to this baseline analysis, the role of government redistribution (taking account, hence, not just redistributive social policies but also of tax provisions by the state), and also the roles of welfare state regime type. We have also assessed the direct and moderating effects of welfare state systems. See the Online appendix and robustness discussion below.

8 Pension benefits are contributory programmes that should matter substantially less for such worries and questions of deservingness, since they are in fact and in popular discussion seen as so clearly tied to years worked in country (Boeri 2010; Burgoon 2014).

9 Not all countries are in every wave. For instance, at the time of this writing, the available countries with sufficient information for our study, the survey rounds between 2002 and 2012 range from 18 to 20, while our sample for the 2014 round comprises 15 countries: Austria, Belgium, Switzerland, Czech Republic, Denmark, Germany, Estonia, Finland, France, Ireland, Netherlands, Norway, Poland, Sweden, and Slovenia. All reported results are robust to exclusion of this wave.

10 Although employing a single-item measure might be criticized, various studies rely on this ESS measure (see Burgoon 2014; Finseraas 2008, 2009; Jæger 2013; Schmidt-Catran 2016).

11 In robustness tests below, we also consider Net migration rates as an alternative measure for exposure to immigrants in a given country-year.

12 Robustness checks below consider net replacement rates, a broad and generosity-based alternative to our spending measure with similar empirical coverage in time and space to our spending-based baseline.

13 See Boeri 2010; Burgoon 2014. As robustness checks below address, the resulting measures are insensitive to no or more controls in specifications to estimate country-year specific social-benefit dependency of foreign-born respondents.

14 We also estimated our models based on the actual household income. See robustness checks.

15 Such negative interactions might be artifacts of macro-level foreign-born percent, social welfare spending and migrant social benefit dependency having implications for Support for redistribution that are negatively moderated by Anti-immigration sentiments. Nonetheless, the interactions (also) fit our Hypotheses 2–4.

16 Another, more logistical drawback is that this model with four-way embedding is a linear model, since the more appropriate logit and ordered logit models did not converge.