The in-group/out-group theory

It has been claimed that people have a natural tendency to develop group identities and to differentiate between “us” and “them” (Allport Reference Allport1954). People are inclined to develop positive feelings towards people who share similar traits – e.g., ethnicity, religion, language (the in-group) – and negative feelings towards people who do not (the out-group) (e.g. Sniderman et al. Reference Sniderman, Hagendoorn and Prior2004). Consequently, the native-born population – the “in-group” – can be less inclined to include the “out-group” in their social benefit schemes. In other words, people exhibit so-called in-group bias, favouring their own kind. Out-group derogation is a related phenomenon, in which an out-group is perceived as being threatening to the members of an in-group, which often accompanies in-group favouritism (Hewstone et al. Reference Hewstone, Rubin and Willis2002).Footnote ²

As argued by Shayo (Reference Shayo2009), the relative size of the out-group can matter for social preferences. Preferences for redistribution in a country can decrease with the size of the out-groups; the reason being that increased distance to other agents (e.g. migrants) in the original group of identity (e.g. low-income class of natives) makes identification with a less redistributive group (e.g. high-income class/the nation as a whole) more attractive. Furthermore, some studies have found that the size of the migrant population – the “out-group” – directly affects discriminatory attitudes and feeling of perceived threat (e.g. Gijsberts et al. Reference Gijsberts, Scheepers and Coenders2004; Kunovich Reference Kunovich2004; Semyonov et al. Reference Semyonov, Raijman and Gorodzeisky2006). Related to this finding, it has been argued that in-group members believe that the welfare system disproportionately benefits ethnic minorities (Gilens Reference Gilens1995; Alesina and Glaeser Reference Alesina and Glaeser2004), which again can trigger welfare chauvinistic preferences. In fact, Mau and Burkhardt (Reference Mau and Burkhardt2009) find support for a negative relationship between the share of non-western immigrants and the willingness to grant equal social rights to foreigners.

Previous research on the relationship between how immigration affects social preferences has mostly focussed on general immigration. However, it is not unlikely that intra-EU immigrants are perceived differently by the native population than non-western (non-EU) immigrants. Some features of the former can make them less threatening to natives than the latter. Most importantly, traditional asylum and family reunification immigrants represent more of an out-group than intra-EU immigrants, the latter creating (comparatively) less heterogeneity. Furthermore, labour immigrants are in the host country to work; this is their raison d'être as immigrants. Studies show that they contribute (much) more to the financing of the welfare state than they take out (e.g. Dustmann and Frattini Reference Dustmann and Frattini2014). Refugee and asylum immigrants, on the other hand, do not migrate primarily for economic reasons. They have often fled armed conflict or persecution and are recognised as needing of international protection because it is too dangerous for them to return home. Because of this, natives can come to believe that labour immigrants contribute more to the welfare system and that they exploit it less, and therefore that they represent less of a threat.

However, other factors can make natives view non-western immigrants as less of an out-group than intra-EU immigrants. It has been argued that citizenship constitutes the basis of eligibility for welfare (Marshall Reference Marshall1950). The state’s welfare programmes are for its members. And it can be argued for a majority of the natives, asylum/refugee immigrants are seen as being more “members” of the host society than intra-EU labour immigrants, primarily because they aspire to become citizens – the primary purpose of their migration is precisely to acquire citizenship (in order to escape persecution/danger). However, this is not the case for intra-EU labour immigrants; they are primarily in the host country to maximise income for a period of time before returning to their home country. Thus, in terms of citizenship, asylum/refugee immigrants are more of an “in-group” than intra-EU immigrants.

Intergroup contact theory

Contrary to the in-group/out-group theory, intergroup contact theory holds that immigration can help to weaken welfare chauvinistic attitudes. According to intergroup contact theory, first proposed by Allport (Reference Allport1954), increasing contact between members of different groups can work to reduce prejudice and intergroup conflict. Related to our research question, this would mean that welfare chauvinism decreases with the amount of immigration (Mewes and Mau Reference Mewes and Mau2013) – under the assumption that more immigration also implies more (positive) contact, which is likely. Indeed, one of the two prevailing measures of intergroup contact is precisely the size of a minority group, the other one being personal contact between members of the majority and minority groups (Stein et al. Reference Stein, Post and Rinden2000, 285). Stein et al. (Reference Stein, Post and Rinden2000) find that the two measures are positively correlated: a higher proportion of immigrants increases the probability of intergroup contact, i.e. that people are more likely to have contact with immigrants when there are relative more of them. This finding is supported by Wagner et al. (Reference Wagner, Christ, Pettigrew, Stellmacher and Wolf2006).

Allport (Reference Allport1954) suggested that positive effects of intergroup contact occur in contact situations characterised by four key conditions: (1) Equal status: While difficult to define precisely (Pettigrew Reference Pettigrew1998), it implies that both groups must engage equally in the relationship – members of the group should have similar backgrounds, qualities and characteristics (Riordan Reference Riordan1978); (2) Intergroup cooperation: Groups need to work together in the pursuit of common goals (Aronson and Patnoe Reference Aronson and Patnoe1997); (3) Support of authorities, law or customs: Both groups must acknowledge some authority that supports the contact and interactions between the groups (Pettigrew Reference Pettigrew1998); and (4) Common goals: Both groups must work on a problem/task and share this as a common goal – exemplified by interracial sport teams (Pettigrew Reference Pettigrew1998). These key conditions seem overall applicable to intra-EU migrants, as they immediately take part in regular life within the new country since they integrate in the labour force.

However, empirical studies have established that contact can reduce prejudice even when these four conditions are not present, i.e. contact between in-group and out-group members reduces prejudice and intergroup conflict even in situations that are not characterised by equal status, common goals, etc. (see Pettigrew Reference Pettigrew1998 for a review of this literature).

At the cross-national level, some empirical studies explicitly support the argument that (a) intergroup contact indeed increases with the amount of immigration, and (b) this, in turn, softens anti-immigrant stances and leads to less prejudice (Wagner et al. Reference Wagner, Christ, Pettigrew, Stellmacher and Wolf2006; Schlueter and Wagner Reference Schlueter and Wagner2008). Another study by Mewes and Mau (Reference Mewes and Mau2013) explicitly examines how intergroup contact directly affects welfare chauvinism, but find little evidence. However, their study operationalises intergroup contact rather differently than ours. Most importantly, they measured intergroup contact through the ‘Konjunkturforschungsstelle’ (KOF) Personal Contacts index, which contains components such as the country’s percentage of international telephone traffic, international tourism, etc.Footnote ³ The KOF index was developed to measure an extent of globalisation, and it taps into how connected a country is to other countries. Furthermore, the KOF index cannot separate the possible effects of specific types of immigration (e.g. intra-EU immigration), something that we are particularly interested in. The share of immigrant population is a different operationalisation of intergroup contact than the KOF index, and one that fits our research aim better.

We want to emphasise that intra-EU immigrants embody a diverse group of immigrants, e.g., they represent 28 different countries. It is therefore plausible that the effect of intra-EU immigration on welfare chauvinistic sentiments depends on from which geographical area in the EU that the immigrants come from. More concretely, it has been argued that in particular Eastern European immigration has been politicised; it is specifically this group of intra-EU immigrants that has been the target of media attacks (Johns Reference Johns2014). In particular, Eastern European immigrants have been accused of putting downward pressure on wages and of abusing the welfare system, receiving benefits they are not entitled to. Right-wing parties, but also mainstream politicians and government agencies, have often reinforced these stereotypes (Johns Reference Johns2014). Furthermore, the Eastern European immigration has been abrupt and escalated dramatically since the enlargement in 2004.Footnote ⁴ Both these features of the Eastern European immigration, the sudden escalation and its politicisation, are the two conditions put forth by Hopkins (Reference Hopkins2010) for an out-group (such as immigrants) to be seen as threatening, and to become the target of political hostility (Reference Hopkins2010, 40). This can have theoretical implications. With respect to the in-group/out-group theory, one can expect Eastern European immigrants to be seen as more threatening than western immigrants, i.e. as more of an out-group. Regarding the intergroup contact theory, one can expect the softening impact of contact (on welfare chauvinism) to be less pronounced when the contact is between natives and Eastern European immigrants, than when it is between natives and western European immigrants. Recall that the positive effect of intergroup contact is particularly pronounced in situations where the natives and the immigrants (out-group) have equal status, and the negative stereotyping of Eastern European immigrants can cause the natives to perceive of them as being of lower status.

The United Kingdom’s (UK) withdrawal from the European Union (known as “Brexit”) is one illustration of the perceived tension between natives and in particular Eastern European immigrants. In the (UK) public perception, the rapid growth in immigration to the UK has been typically associated with the two rounds of EU eastern enlargement that took place in 2004 and 2007; and this immigration is one of the major explanations for why the UK chose to leave the EU (Alfano et al. Reference Alfano, Dustmann and Frattini2016).

Expectations

According to the in-group/out-group theory, welfare chauvinistic attitudes should increase with immigration, and thus,

It can, however, be expected that this negative association is particularly pronounced for nationals of East European countries compared to other EU migrants. As emphasised, politically charged debates related to welfare chauvinistic issues have often focussed on labour immigrants, especially from Eastern Europe.

According to the intergroup contact theory, however, it is expected that welfare chauvinism is decreasing with the amount of immigration, because intergroup contact reduces prejudice and softens anti-immigrant stances; and thus,

We expect that this negative association might be less pronounced when the focus is strictly on Eastern European immigrants, simply because Eastern European immigrants are profiled as more threatening (due to, as argued above, its sudden escalation and degree of politicisation), and therefore that the positive effect of contact is less evident for this immigrant group.

Furthermore, we expect that in welfare states where benefits tend to be proportional to previous income and social security contribution, welfare chauvinism will be lower than in welfare systems providing flat-rate benefits and financed through general taxation (for ease of terminology, we refer to the former system as Bismarckian and the latter as Beveridgean).Footnote ⁵ The argument for this expectation is as follows: we know from previous literature that people differ in their support for various groups of “needy” people (e.g. Van Oorschot Reference Van Oorschot2006). However, we also know that people’s attitudes towards group support seem to be strongly affected by the extent to which the different groups are seen as deserving or underserving, and one of the most important deservingness criteria is reciprocity, i.e. building up a personal entitlement to benefits through working/paying contribution (Van Oorschot Reference Van Oorschot2000, Reference Van Oorschot2006). Therefore, in Bismarckian welfare states it is likely that immigrants will be seen as more “worthy” recipients than in Beveridgean welfare states. We thus control the extent to which a country is Bismarckian, measured by how much of the total amount of social spending is financed by social insurance rather than general taxation (see the Method section for details).

This expectation has implications for our hypotheses. Our first hypothesis – that EU-migration strengthens welfare chauvinistic attitudes – should be particularly pronounced under Beveridgean systems, and less so under Bismarckian systems. We thus include a second, conditional expectation to Hypothesis 1:

Furthermore, the expectation that intra-EU immigration leads to less welfare chauvinistic attitudes, as formulated under our Hypothesis 2, would be specifically pronounced in systems that more resemble the Bismarckian system. Hypothesis 2 can thus be specified to include:

Welfare chauvinism

Welfare chauvinism is tapped in the ESS 2008 with the following question: “Thinking of people coming to live in [country] from other countries, when do you think they should obtain the same rights to social benefits and services as citizens already living here? Please choose the option on this card that comes closest to your view”. People can then choose one of the following answers: (1) Immediately on arrival; (2) After living in [country] for a year; (3) Only after they have worked and paid taxes for at least a year; (4) Once they have become a [country] citizen; and (5) They should never get the same rights.

Main independent variables

Our main independent variable is intra-EU immigration. Eurostat provides data on the number of people living in an EU country who originate from another EU country. This number is then translated into a measure reflecting the percentage of the population. As a further robustness check, we include a second measure of intra-EU immigration. Here, we calculated the percentage of respondents from the ESS 2008 that reported having been born in another EU country, or in Norway or Switzerland. The two indicators of intra-EU immigration are with 0.945 very highly correlated.

Furthermore, the analyses include some other indicators for immigration. The reason for this is twofold. First, when looking at the effects of intra-EU migration, it is important to control for non-EU migration. This allows us to check whether it is intra-EU migration that drives the results, and not a possible correlation between different forms of immigration, i.e. that some countries simply attract more immigrants in general. This is particularly important since the indicator for welfare chauvinism does not specify a specific type of immigration. Second, it is pertinent to examine whether general or non-intra-EU migration has differential effects on welfare chauvinism.

We thus include a variable that indicates the number of all immigrants relative to the population (% migrants), as well as one that indicates the number of immigrants who do not originate from the EU (% non-EU migrants). The World Development Indicators of the World Bank provide data on these measures. Furthermore, as with the measurement of intra-EU migration, we use the ESS 2008 as a basis for a second measure of general and non-EU immigration as a robustness check. The correlation between the two indicators for general immigration is 0.807, and that of the indicators for non-EU immigration 0.587. The Appendix further details the measurement of immigration.

Conditioning variable

We want to test whether the relationship we find holds for all countries equally, or whether the relationship is conditional on specific country characteristics. Following our expectations, we focus on the extent to which a country has a welfare system where benefits are proportional to previous income/contribution. There exist no readymade index for this; however, the percentage of social protection that is financed through social insurance contributions (as opposed to general taxation) is argued to be a good proxy (Bonoli Reference Bonoli1997). The higher this percentage, the more Bismarckian is the welfare system (and, thus, the more likely the system is to offer earnings-related benefits). Conversely, the lower the percentage, the more Beveridgean the system is (and the more likely the system is to offer flat-rate benefits financed through general taxation). Data are provided by Eurostat (see CESifo 2008).

Control variables

Besides the level of intra-EU migration, other factors are likely to affect welfare chauvinism. Therefore, we included a number of alternative variables in order to control for their effects – the inclusion of these variables also helps to isolate the effect of interest here, that of intra-EU migration. However, they do not constitute the main interest in this study. For this reason, we only shortly discuss the control variables here, and include information about their measurement in the Appendix. We will not report or discuss the results regarding these explanations. The results are fully reported in the Online Supplement, in Tables S1 and S2.

The control variables include both individual- and country-level indicators. We emphasise different types of individual-level variables that previous research identified as affecting welfare chauvinistic attitudes (see, e.g. Coenders and Scheepers Reference Coenders and Scheepers2003; Reeskens and Van Oorschot Reference Reeskens and Van Oorschot2012; Mewes and Mau Reference Mewes and Mau2013): political orientation (on a left-right scale); individual characteristics (gender, age, education, income, employment status and whether a person thinks s/he is likely to need some form of welfare assistance); interpersonal trust (as well as how socially active a person is); and lastly, attitudes towards the EU, and whether the government should be in charge of providing welfare benefits.

In addition to these individual-level controls, we include different country-level controls.Footnote ⁷ Most notably, we include an indicator for the size of the welfare state in terms of the percentage of the gross domestic product (GDP) that is spent on social benefits. It has been argued that generous and expansive welfare states seem to foster tolerance towards immigrants and also produce less welfare chauvinism (e.g. Crepaz and Damron Reference Crepaz and Damron2008; Reeskens and Van Oorschot Reference Reeskens and Van Oorschot2012). This is our main and key control variable. Data for this measure are again provided by Eurostat.

There are other variables that may affect welfare chauvinism. However, with just 21 countries, not all of these can be included simultaneously. For each model we include a maximum of three country-level variables, meaning that we need to substitute some controls for others in order to assess their impact. We therefore reanalyse the main model, also including economic inequality, wealth and unemployment rate. Furthermore, it has been observed that unemployment rates are linked to welfare chauvinism (e.g. Mewes and Mau Reference Mewes and Mau2012), and so we control for unemployment. The Appendix details the measurement of these indicators, and the Online Supplement provides the results of the models that include these alternative controls (Table S3).

Estimation process

In order to test our expectations, we use a multilevel ordered logit regression model. The dependent variable that we use is a variable with categories that can be ordered. The distances between the categories, however, are not necessarily of equal distance, and the values attached to the categories do not mean anything except for the order in which they are placed. An ordered logit model does not assume that the distances are meaningful as such, and therefore offers a better option to analyse the data than an ordinary least squares regression, for example.

Moreover, because we are interested in testing a model that includes both country and individual-level indicators, we use a multilevel approach to analyse our data. We explicitly model the multilevel structure of the data in order to avoid too small standard errors, but also as to allow for variation in the level of welfare chauvinism at the country-level. We therefore use a random-intercept model. All models of interest are compared to a simple null-model where no independent variables are included. Both the country-level variance of the null-model and the models of interest are reported, as to allow for a comparison in terms of how well a model performs in terms of explaining this country-level variance.

We thus employ random-effects ordered logit regression in order to analyse the data. This procedure estimates an underlying score as a linear function of the independent variables as well as a number of cut-points. The latent linear equation to be estimated is as follows:

(1) $$Y{\asterisk}_{{iu}} \,{\equals}\,x_{{1iu}} \beta _{1} {\plus}x_{{2iu}} \beta _{2} {\plus}\,\ldots\,{\plus}x_{\rho {iu}} \beta _{\rho } {\plus}z_{{1iu}} u_{{1u}} {\plus}z_{{2iu}} u_{{2u}} {\plus}\,\ldots\,{\plus}z_{{\varphi iu}} u_{{\varphi u}} {\plus}{\epsilon}_{{iu}} $$

where x _iu β are the fixed effect for i individuals and j countries based on the independent variable, z _iu u _u are the random effects based on the variance in the dependent variable on the country-level, and ɛ _iu are the errors which are distributed as logistic. The outcome of the estimated model described in (1) returns values that can be interpreted as changes in the odds ratios, i.e. for a unit change in x, the odds from going to a higher outcome change with β. While the βs can be interpreted in terms of their sign and significance, and the size of the value can be compared within models, the value cannot be interpreted meaningfully without the cut-points. Since we are mainly interested in whether intra-EU migration has a positive or negative significant effect, and whether the effect is different from that of general or non-EU migration, we focus on the interpretation of this latent linear equation.

In addition, we discuss the predicted probability for the different categories of welfare chauvinism to further assess the substantive meaning of the results. On the basis of Equation 1, and using the cut-points that are estimated by the model, the predicted probabilities can be calculated as follows:

(2) $${\rm Pr}\left( {y_{{iu}} \,{\equals}\,v{\rm \,\mid\,}v,\,u_{u} } \right)\,{\equals}\,{\rm Pr}\left( {v_{{v{\minus}1}} \,\lt\,x_{{\rho iu}} \beta _{\rho } {\plus}z_{{\varphi iu}} u_{{\varphi u}} {\plus}{\epsilon}_{{iu}} \leq v_{v} } \right)$$

where v _v is the vth cut-point. In our case, we have five outcomes for the dependent variable, implying that just four cut-points are estimated. In addition, it is assumed that v ₀=–∞ and v ₅=+∞. The predicted probabilities are illustrated for some of the models. The models do not estimate a constant since the cut-points already absorb the constant.

As an additional check for the robustness of the models that we discuss here, we included a replication of all models using a different statistical technique. While a multilevel ordered logit model is the most appropriate for the data that we use, it is also possible to estimate the models with a linear multilevel model, with random intercepts. The results of the replications with this different technique are reported in Table S4 in the Online Supplement, and further support the findings discussed below. Furthermore, we checked whether the results are driven by a particular country by rerunning the analyses excluding one case each time. The results remain largely the same, suggesting that not one case is an outlier with too high leverage.