Introduction
Mainstream-left parties across Western Europe are not uniform in their immigration and welfare policy positions – particularly at the intersection of the two. The electoral discourse of the mainstream left ranges between proposing the exclusion of immigrants from the welfare state – so-called ‘welfare chauvinism’ – to portraying themselves as universal defenders of an inclusive welfare state for all, with some parties even shifting between the two extremes within short time periods. When it comes to policy reforms, a similar dialectic plays out. A number of parties have pushed for restriction of welfare rights – such as the reform to asylum seeker welfare programmes in 1999, which was legislated by Blair’s New Labour (Consterdine and Hampshire, Reference Consterdine and Hampshire2014) – others have been applauded for their inclusive stance on immigrants’ rights, such as the Swedish social democratic governments in the post-war era (Sainsbury, Reference Sainsbury2012).
A common assumption of many welfare chauvinism scholars is that left-wing parties employ exclusivist welfare policies as a vote-seeking tactic to appease their working-class constituency (Schmitt and Teney, Reference Schmitt and Teney2018; Harris and Römer, Reference Harris2022), who are the most likely to support welfare chauvinism (Van der Waal et al., Reference van der Waal2010; Mewes and Mau, Reference Mewes and Mau2012). Yet, such an assertion may puzzle students of mass public opinion and electoral realignment, whose empirical work shows that, although working-class voters have not completely abandoned the mainstream left (Rennwald, Reference Rennwald2020), their share among social democratic parties has drastically declined in recent years. Instead, a growing literature shows that middle classes have come to dominate constituencies of mainstream-left parties in recent decades and that these voters demand different welfare policies than their working-class co-electorates (Kitschelt, Reference Kitschelt1994; Gingrich and Häusermann, Reference Gingrich and Häusermann2015; Häusermann, Reference Häusermann2018; O’Grady, Reference O’Grady2019). Middle-class voters, due to their higher education levels and more secure income, are less likely to support welfare chauvinism (Careja and Harris, Reference Careja and Harris2022) and more likely to be open to globalisation and immigration.
The implications of electoral realignment for the policy decisions of left-wing governments have been explored vis-a-vis differentiated welfare policies (Gingrich and Häusermann, Reference Gingrich and Häusermann2015), labour market policies (Bürgisser and Kurer, Reference Bürgisser and Kurer2021) and economic policies (Engler and Zohlnhöfer, Reference Engler and Zohlnhöfer2019). However, no paper has systematically investigated whether social democratic parties’ welfare chauvinism is a function of navigating the class realignment dilemma. Thus, in this paper we ask to what extent does electoral demand, namely the divisions between working and middle-class voters, and electoral realignment, namely the share of working- and middle-class voters in social democratic electorates, explain welfare chauvinist policy outcomes when social democratic parties enter government. We empirically test the assumption that when working classes dominate the social democratic electorate, social democratic parties promote welfare chauvinism, whereas when middle classes dominate, reforms to expand welfare programmes to immigrants are more likely. As social democratic parties become less associated with workers’ representation, they also forgo stricter immigration policies to prevent liberally inclined, higher-educated voters to shift their vote choice elsewhere, for example, to the Green parties.
We test our hypotheses using both original and established datasets in a step-wise fashion. At the micro-level, we use European Social Survey (ESSFootnote 1 ) data to assess whether the proposed electoral dilemma between middle-class and working-class voters on welfare chauvinism exists both in the general population and within social democratic electorates. Secondly, we combine ESS and Eurobarometer (EB)Footnote 2 data together to construct country-year class shares of the electorate of social democratic parties. We then use country-year level data from the Immigrant Social Rights Dataset (ImmigSR) (Römer et al., Reference Römer, Leonce and Zobel2023) to analyse whether and how class composition of the social democratic electorate has an impact on how these parties reform social protection rights of immigrants.
At the micro-level, we find that working-class voters are significantly more welfare chauvinist than the middle class in nearly all Western and Southern European countries and that this applies within the mainstream-left electorate, too. On the macro-level, using two-way fixed effects regressions covering 14 countries between 1980 and 2018, we find social democratic parties in government tend to implement reforms expanding immigrants’ welfare rights (welfare inclusion) when their electorate is strongly middle class. However, contrary to assumptions in the literature, the link between welfare chauvinism and working-class electorates cannot be supported. On balance, we consequently argue that greater analytical focus should be paid to the rising influence of middle-class voters within the story of power resources and welfare inclusion on the left.
Social democrats, the progressive’s dilemma and welfare chauvinism
Welfare chauvinism has classically been seen as one response or solution to the so-called “progressives’ dilemma” between open borders and generous welfare states, which is particularly relevant for ‘the Left’ in Western Europe (Kymlicka, Reference Kymlicka2015; Spies, Reference Spies2018). The progressives' dilemma can be summarised as follows: Social democratic parties were important political forces for the development and sustenance of the welfare state in many European countries. As many welfare state benefits are (partially) tax-financed, a certain solidarity towards the poor or those in need undergirds many theories of for support for the welfare state. In this context, immigration may, at least in theory, pose a problem (Freeman, Reference Freeman1986). A popular argument in both scientific and popular debates posits that ethnic heterogeneity depletes solidarity for a strong welfare state (Alesina and Glaeser, Reference Alesina and Glaeser2004). Especially in the USA, this hypothesis has also received widespread empirical support (Soss et al., Reference Soss, Schram, Vartanian and O’Brien2001; Hero and Preuhs, Reference Hero and Preuhs2007), and some studies point to the existence of a similar relationship also in Europe (Eger and Breznau, Reference Eger and Breznau2017; Muñoz and Pardos-Prado, Reference Muñoz and Pardos-Prado2019) although context seems to matter in determining how and in which direction immigration affects welfare support (Brady and Finnigan, Reference Brady and Finnigan2014; Burgoon, Reference Burgoon2014; Soroka et al., Reference Soroka2016, Burgoon and Rooduijn, Reference Burgoon and Rooduijn2021). Regardless of the objective value of the so-called ‚progressives‘ dilemma’, party actors may perceive the need to respond to it. Scholars have suggested that social democratic parties could use welfare chauvinism, and thus prevent immigrants from accessing welfare benefits and services, to maintain their ideological commitment to a strong, national welfare state (Koning and Anthony, Reference Koning and Anthony2017). Yet, the exclusion of the most vulnerable, namely immigrants, from access to the welfare state counterposes fundamental ideas core to social democracy, such as minimum living standards for all and equality. Thus, the argument for welfare chauvinism, particularly at the policy level, is not without trade-offs and is far from trivial for left-wing governments.
The decision to employ welfare chauvinism is not only ideologically conflicting for social democratic parties, but, we argue, poses an even bigger electoral dilemma. Firstly, social democratic parties are classically seen as working-class parties (Häusermann and Gingrich, Reference Häusermann and Kriesi2015). The extensive literature on attitudes towards welfare chauvinism shows that it is especially popular among people with lower education, with higher (perceptions of) economic insecurity and with deteriorating perceptions of social status (Heizmann et al., Reference Heizmann2018; Kros and Coenders, Reference Kros and Coenders2019; Hooijer, Reference Hooijer2021). Therefore, the blue-collar working class has been shown to be the most fervent supporters of welfare chauvinism (Mewes and Mau, Reference Mewes and Mau2012). Following such class-based demand, social democrats could likely use welfare chauvinism in order to appeal to their working-class constituency.
Yet, there is substantial evidence that social democratic parties are no longer primarily parties of the working classes. Rather middle-class individuals, in particular socio-cultural professionals, have become a dominant constituency of Social Democracy (Häusermann and Kriesi, Reference Häusermann and Kriesi2015; Oesch and Rennwald, Reference Oesch and Rennwald2018). This is, on the one hand, due to electoral realignment, through which younger generations of working-class voters have become attached to the radical right rather than the left (Oesch, Reference Oesch2008; Rydgren, Reference Rydgren2012; Rennwald et al., Reference Rennwald, Evans and Jonas Pontusson2021) or have become politically demobilised (Bremer and Rennwald, Reference Bremer and Rennwald2023). Due to processes of tertiarisation, educational expansion and occupational upgrading, the working class has also simply reduced in relative size to the middle-class population more broadly (Oesch, Reference Oesch2008; Evans and Tilley Reference Evans and Tilley2017). Middle-class voters are less likely to support welfare chauvinism given their higher level of education (Mewes and Mau, Reference Mewes and Mau2012; Iversen and Soskice, Reference Iversen and Soskice2020), which has also been shown to increase perceptions of deservingness of welfare recipients (Attewell, Reference Attewell2021), and other authoritarian values along the socio-cultural dimension of political conflict (Bornschier, Reference Bornschier2010; Kriesi et al., Reference Kriesi2012). Thus, should social democratic parties be parties of the middle classes, they should be less likely to promote welfare chauvinism when in power.
A number of studies have shown that the class composition of the electorate impacts social democratic governments in a number of policymaking fields.Footnote 3 Gingrich and Häusermann (Reference Gingrich and Häusermann2015) show that leftist governments with more middle-class voters tend to be less active in promoting social consumption policies when in power, benefits that substantially benefit working-class voters (Häusermann et al., Reference Häusermann, Pinggera, Ares and Enggist2022) (see also Engler and Zohlnhöfer (Reference Engler and Zohlnhöfer2019) for economic policies more generally). Schwander (Reference Schwander2018) and Bürgisser and Kurer (Reference Bürgisser and Kurer2021) also show that the composition of the electorate matters for the positions social democratic parties adopt at elections, vis-à-vis family policy and insider-outsider policies. In the immigration literature, Han (Reference Han2015) demonstrates that the left takes stronger pro-immigration stances in contexts where class voting is weak. On second-dimension (socio-cultural) issues more broadly, Abou-Chadi and Wagner (2019, Reference Abou-Chadi and Wagner2020) show that nowadays leftist parties predominantly, and more successfully, promote culturally liberal positions, presumably reflecting the more middle-class segment of their electorate.
In this paper, we aim to test the supposition that electoral demand leads to different policy outcomes in the field of welfare chauvinism. Schmitt and Teney (Reference Schmitt and Teney2018) in their study of welfare chauvinist policies argued that the working-class base of social democratic parties explains their finding that left-wing governments are less likely to extend social rights to immigrants. Yet, the underlying micro-foundations of this argument have not been tested. Rather, other studies of social democratic policies around immigration and welfare chauvinism have emphasised the importance of coalition governments (Harris and Römer, Reference Harris2022), the intersection of immigration and welfare regimes (Sainsbury, Reference Sainsbury2012) and centralised or decentralised party organisation (Rathgeb and Wolkenstein, Reference Rathgeb and Wolkenstein2022) as important contextual variables. In this paper, we track the argument that electoral class demand matters for welfare chauvinism and welfare inclusion, respectively, assessing both the individual-level behaviour of voters and the response of social democratic parties in power.
Whilst the focus of our paper is to assess the story behind electoral demand and its consequences for partisan effects on welfare chauvinism policy, other variables may moderate the significance of class politics for social democratic governments in their policymaking processes. Most importantly, the presence of challenger parties such as PRRPs may threaten electoral losses, in particular of working-class voters (de Koster et al., Reference de Koster2013; Kriesi et al., Reference Kriesi2012; Afonso and Rennwald, Reference Afonso, Rennwald and Manow2018). In recent years, PRRPs in many countries across Europe have utilised welfare chauvinist rhetoric at elections (Ennser-Jedenastik, Reference Ennser-Jedenastik2018; Schreurs, Reference Schreurs2021) and have promoted excluding immigrants from welfare benefits in coalition governments (Chueri, Reference Chueri2021). Since PRRPs are known to have mobilised working-class voters primarily on anti-immigration platforms generally (Ivarsflaten, Reference Ivarsflaten2005; Oesch, Reference Oesch2008), yet working classes continue to support redistribution for ‘deserving’ natives, welfare chauvinism has been proposed as a strategic option to deter an alleged working class exodus to the populist radical right (Biard et al., Reference Biard, Bernhard and Betz2019; Kopyciok and Silver, Reference Kopyciok and Silver2021). Similarly, the increasing relevance of middle-class voters may also be impacted by electoral competition. Just as PRRPs pose an electoral threat to the working-class constituency of social democratic parties, parties on the ‘New Left’, such as the Greens, may also threaten the middle-class electorate of social democrats. Whilst these parties mostly support cosmopolitan policies (Dolezal, Reference Dolezal2010) towards immigration and also middle-class welfare policies like social investment, they are increasingly popular for middle-class voters (Röth and Schwander, Reference Röth and Schwander2020). It is beyond the scope of this paper to look into these dynamics of party competition, but it can be assumed that, since the 1980s and 1990s, party fragmentation and the salience of immigration have further accentuated the electoral threat of PRRPs and Green parties for different class groups of the social democratic electorate (see Abou-Chadi and Immergut, Reference Abou-Chadi and Immergut2018). We therefore check whether adding the electoral successes of PRRP and Green party challengers in our analysis influences our findings, whilst remaining focused on the significance of class demand in the story of welfare chauvinism. We will reflect on the relevance of party competition in the conclusion and discussion.
Dynamics of electoral demand: hypotheses
To test whether class demand explains policy outcomes, our hypotheses follow two levels of analysis – at the levels of public opinion (micro) and policy outcomes (macro).
Micro-level
Firstly, should class demand be important for social democratic party actors, we follow others and begin with exploring the preferences of their voters. Namely, as discussed above, socio-economic status, such as education level and occupational class, has been shown to explain immigration and welfare chauvinism preferences. Because social democratic parties enjoy support from both working and middle-class, higher- and lower-educated, voters, literature on immigration has argued that their electorates are rather torn on such issues. This has already been explored in regard to immigration issues more broadly. Harris (Reference Harris2022) shows that social democratic party voters are more divided along education lines over economic immigration concerns compared to cultural ones, yet this remains to be shown for welfare chauvinism. At the micro-level, we postulate therefore two main hypotheses:
HYPOTHESIS 1a: Working-class voters support welfare chauvinism more than middle-class voters.
HYPOTHESIS 1b: Within social democratic electorates, working-class voters support welfare chauvinism more than middle-class voters.
Macro-level
Following the class demand hypothesis, we want to test whether the class composition of the electorate for social democratic parties, as it changes over time, is significantly related to their policy reforms. We adopt an actor-centric approach to the policymaking process and consider parties as, first and foremost, rational actors seeking to win elections. As the micro-level hypotheses expect a class divide in welfare chauvinism preferences, we propose the following actor-centred hypotheses regarding electoral demand on policymaking:
HYPOTHESIS 2a: The higher the share of working-class voters in social democratic electorates, the more likely social democratic parties are to enact welfare chauvinist policy reforms.
HYPOTHESIS 2b: The higher the share of middle-class voters in social democratic electorates, the less likely social democratic parties are to enact welfare chauvinist policy reforms.
These hypotheses deal with different units of analysis, and therefore, we proceed in two empirical steps. First, we use survey data to analyse the attitudes towards welfare chauvinism at the individual level and whether class matters. Second, we move to the level of governments and policy reforms and assess if social democratic cabinets are influenced by changes in class share of the electorate.
Empirics: micro-level: does a class divide on welfare chauvinism exist?
Data, measurement and research design
Are working-class voters more welfare chauvinist than their middle-class counterparts (Hypothesis 1a), does such a class divide also exist within social democratic electorates (Hypothesis 1b) across western Europe?
To evaluate the existence of a working-class and middle-class divide, we use individual-level data from the European Social Survey (ESS), specifically from its fourth (2008) and eighth (2016) waves, both of which contain a variable that has been widely used in the literature to capture welfare chauvinist preferences. The ESS allows us to check whether this divide has remained constant between 2008 and 2016 and, more importantly, provides us with data to check the existence of an electoral dilemma for social democratic parties in all the countries on which we also base our macro-level analysis. These are 14 Western and Southern European countries Footnote 4 which differ in terms of their welfare state regime as well as their electoral systems and party system fragmentation.
The dependent variable we use in the micro-level analysis asks respondents when ‘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?’ Becoming increasingly welfare chauvinist, answer categories include the following: 1) Immediately on arrival; 2) After a year, whether or not have worked; 3) After worked and paid taxes at least a year; 4) Once they have become a citizen; 5) They should never get the same rights. Previous studies using this variable have adopted various operationalisation strategies to capture welfare chauvinist preferences. Besides treating the variable as nominal (Reeskens and van Oorschot Reference Reeskens and van Oorschot2012), others dichotomise the variable by defining categories 4 and 5 as welfare chauvinist (Mewes and Mau, Reference Mewes and Mau2012; Heizmann et al., Reference Heizmann2018), treat the variable as ordinal (Heizmann et al., Reference Heizmann2018) or treat it as a continuous variable (e.g., Eger and Breznau, Reference Eger and Breznau2017). For ease of interpretation and illustration, we show findings following the last option. We conduct other variations in the robustness checks.
Our main independent variable measures social class. Instead of using information on subjective class identity or income, we adopt an occupation-based definition of class, following the approach used by Gingrich and Häusermann (Reference Gingrich and Häusermann2015) as well as Engler and Zohlnhöfer (Reference Engler and Zohlnhöfer2019). This approach is primarily based on Oesch’s (Reference Oesch2006) 16 category occupational class schee which relies on the International Standard Classification of Occupations (ISCO) Footnote 5 and secondly based on respondents’ education level. It divides people into four rough classes: Manual working, routine working (i.e., low-skilled non-manual), middle and employer classes. The manual working class includes production workers and lower-educated technicians. The routine working class consists primarily of clerks and service workers who have no upper secondary or higher education. The middle class includes socio-cultural professionals, managerial professionals, and highly educated clerks or service workers. The employer class consists of large employers and small business owners.Footnote 6 Across all our countries and both waves, this leaves us with 9,686 manual working-class respondents, 4,952 routine working-class respondents, 20,269 middle-class respondents and 6,223 employer respondents.
Empirical results
To check whether members of the middle and the working classes indeed diverge regarding their welfare chauvinist preferences, we simply regress welfare chauvinist preferences on the class variable introduced above and show predicted values for each class. Presented here are the results from OLS regression models with country-fixed effects. We control for survey year (2008 or 2016), age and gender but not for any other socio-structural variables. On the one hand, we refrain from doing so because other typical control variables, such as education or income, are too closely conceptually related to our independent variable and, on the other hand, because we are not interested in a causal effect of class but in descriptively showing whether working-class and middle-class voters indeed differ regarding their level of welfare chauvinism. To test Hypothesis 1a, we restrict the sample to respondents who have the right to vote. To test Hypothesis 1b, we restrict the sample to voters who indicate that they have voted for a social democratic party in the last election. We use post-stratification weights.
Figure 1 shows the predicted values of welfare chauvinist preferences on the continuous welfare chauvinism variable (ranging from 1 to 5) for each of our four class categories, firstly for the whole population, and then for the social democratic electorate only. It aggregates data from all our 14 countries and both survey waves. We find strong evidence for Hypothesis 1a, which posits working-class voters to support welfare chauvinism more than middle-class voters. This difference is not only statistically significant but also rather sizeable, amounting to a difference of about 0.18 on a five-point scale. If we compare the predictive power of being part of the middle or a working class for welfare chauvinist preferences to the predictive power of other socio-structural characteristics, class emerges as one of the most important factors (see online Appendix A1). The class divide regarding welfare chauvinism exceeds the gender divide, the divide between union members and non-members and is about equal to a divide between people aged 20 and 80. Hence, class does matter in predicting welfare chauvinist attitudes.
Figure 1. Level of welfare chauvinism by class in the entire electorate, N = 39'726 and the social democratic electorate only (N = 9006).
Notes: The figure for the whole population is based on the regression model ‘Entire Electorate’ in the online Appendix A3, and the second is based on the regression model ‘Social Democratic Voters Only’ in Appendix A3.
Whilst a clear divide exists between the middle and the working classes, we do not find a divide within the working class. Both the old, manual working class as well as the new, routine working class exhibit similar levels of welfare chauvinism. This suggests that when considering the effect of social democratic parties’ class composition on welfare chauvinism, it is not necessary to distinguish between manual and routine workers, but that these can be treated as a single, relatively homogenous working class. Lastly, the employer class is positioned between the poles of the working and the middle classes. These findings are also robust to different methodological treatments of the welfare chauvinism variable. Differences between the classes remain significant if we calculate ordered logistic or binary logistic regressions. Predicted probabilities based on the latter show that whilst members of the working class have a 38% chance to support granting welfare rights to immigrants only when they acquire citizenship or later, middle-class respondents have a likelihood of only 33% (see online Appendix A2).
To test Hypothesis 1b, we create a sub-sample of survey respondents – those who voted for a social democratic party – to illustrate that the attitudinal divide between middle and working-class voters remains relevant even excluding other voters. The right panel in Figure 1 shows that social democratic electorates from all class groups support welfare chauvinism less than the respective class groups in the general population. Most importantly for our purposes, the difference between the social democratic voters from the middle and the two types of working classes even exceeds that of the entire population. Moreover, we find social democratic middle-class voters to have an average value below the population average of 3.14 and the modal option 3, corresponding to giving immigrants welfare rights after they have worked and paid taxes for at least a year – which makes them more generous than the broader middle-class population. Social democratic working-class voters, in contrast, are about as willing to give immigrants early access to welfare than average country voters.
These findings, at the general and social democratic populations, are robust to treating the dependent variable as ordinal or binary rather than continuous (shown in Appendix A2 and A6), including clustered standard errors (see Appendix A3 and A7), across most countries (see appendix A5 and A8)Footnote 7 and across time points, where both for the general population and for social democratic voters, working-class individuals have become slightly less welfare chauvinist over time (see appendix A9) (Dennison and Geddes, Reference Dennison and Geddes2019).
Finally, besides the existence of a class divide, our reasoning hinges on the assumption that both working-class and middle-class voters care enough about the social rights of immigrants to influence party position and behaviour. Whilst previous research has pointed to an asymmetry of issue importance regarding immigration (Kustov, Reference Kustov2022), we have empirical reasons to believe that both working and middle classes attribute importance to welfare chauvinism, although this importance is not measured in the ESS survey used here. In recent survey data from the UK, Germany, Sweden and Spain (Häusermann et al., Reference Häusermann, Ares, Enggist and Pinggera2020) respondents were asked how important it is to them when immigrants obtain the same welfare rights as citizens already living here. 64% of middle-class voters and 55% (manual) to 57% (routine) of working-class voters either responded with ‘very important’ or ‘rather important’ whereas only a minority considers the issue of when immigrants should receive social rights as ‘rather unimportant’ or ‘completely unimportant’ (see Appendix A10 for the results). The electoral divide shown in the ESS data, combined with this other evidence, suggests indeed that social democratic parties – which historically have been most successful when they have achieved in forming a coalition of the working and the middle class – might face an electoral dilemma regarding their welfare chauvinist stances. Yet whether their policies reflect the class demand of their electorate, remains to be tested.
Macro-level: do class composition and electoral demand matter?
Data, measurement and research design
We test our hypotheses at the macro-level using the same selection of 14 Western and Southern European countries as in the micro-level analysis, now extending the time period to between 1980 and 2018.
Data used for the dependent variable welfare chauvinism policy are collated from yearly data on immigrant social rights, which we use as a proxy for welfare chauvinism (reduction in social rights) and welfare inclusion (increase in social rights). We use the ImmigSR dataset (Römer et al., Reference Römer, Henninger and Harris2023) to measure these changes in de-jure access for each of our 14 countries from 1980 to 2018.Footnote 8 This dataset uses a survey of legal experts who were asked to report whether changes occurred in the level of rights for different immigrant groups for the year in question. We use nine main items from the ImmigSR dataset to construct our main dependent variable – welfare chauvinism policy. These cover four dimensions of immigrants’ social rights: eligibility conditions (three items), type of benefits (one item), consequences of benefit receipt (three items) and ‘preventive measures’ (two items). These dimensions are collected separately for labour, family reunification, asylum and refugees, co-ethnics and irregular immigration immigrants (for more details on the variables and the construction of the index see Römer (Reference Römer2017), Römer et al. (Reference Römer, Harris, Jakob and Missler2021)).
We operationalise welfare chauvinism policy by taking the unweighted average of the nine items mentioned above, across different immigrant groups, for each country-year unit which can range on a continuous scale from 0 to 10. These are theoretical minima and maxima – with 0 denoting equal rights for immigrants and citizens and 10 reflecting no access to the welfare state for any group of migrants. At the country-year unit, each value reflects the level of rights. In our models, we use the value of welfare chauvinism for the last year of any particular cabinet. This follows recent studies in welfare (Schmitt, Reference Schmitt2016), education (Garritzmann and Seng, Reference Garritzmann and Seng2016) and welfare chauvinism research (Römer et al., Reference Römer, Leonce and Zobel2022), where scholars argue that country-year designs are likely to bias the effect size of government ideology on policy reforms, because the dependent variable changes yearly, but many political variables (cabinet share of parties, electoral competition variables) do not. For more information on cabinet periodisation, see Appendix A2.1B. Importantly, we include a lagged-dependent variable in our main models. This ensures that we empirically explain changes, that is, reforms that happened during the cabinet period, rather than levels of welfare chauvinism. Two visualisations of the levels and changes of our dependent variable at the cabinet level are shown in Appendix 2A.11, the levels over time separated by social democratic government are shown in Appendix 2A.12, and a full table of cabinets and the respective reform strategies are shown in Appendix 2A.13.
The core independent variable in our hypotheses 2a and 2b is social democratic party cabinet share. This is measured as a percentage from 0 to 100 for the respective cabinet, retrieved from the ParlGov dataset (Döring and Manow, Reference Döring and Manow2017). The second independent variable is class share of the electorate. This is necessary to test hypotheses 2a and 2b, which requires an interaction effect between social democratic party cabinet share and the share of middle-class/working-class voters in the social democratic electorate. To measure the class share, we focus on occupational class groups as shown in the micro-level analysis, but combine manual workers and routine workers to be a broader ‘working class’, both for reasons of size but also that their welfare chauvinism attitudes rarely differed significantly from one another. We follow Gingrich and Häusermann (Reference Gingrich and Häusermann2015) as well as Engler and Zohlnhöfer (Reference Engler and Zohlnhöfer2019) and construct class shares by aggregating individual-level data on vote choice for social democratic parties using the Eurobarometer trend file from 1980 to 2001, and the European Social Survey rounds 1–9 for the years 2002–2018.Footnote 9 To present the changes over time, Fig. 2 shows a three-year moving average of the share of middle-class and of the share of working-class voters for social democratic parties for each country. Footnote 10
Figure 2. Share of working-class and middle-class voters in social democratic electorates over time and by country.
Notes: 3-year moving averages are shown. Missings (i.e., gaps in the lines) are due to a lack of survey coverage. Lines do not add up to 100 as the employer category is not shown here.
We do, as prior studies, observe a trend that social democratic parties become increasingly dominated by middle classes over time (Gingrich and Häusermann, Reference Gingrich and Häusermann2015). Yet, this trend is less pronounced in some countries, especially of Southern Europe such as Spain and Portugal where working-class voters remain more strongly represented than middle-class voters. Yet, in the most recent years, we observe a plateau or a decrease in the share of middle-class voters as well, in some social democratic electorates such as in Denmark, France and Germany. Thus, there exists variance with regard to working- and middle-class shares in social democratic electorates both over time and between countries.
Regarding cabinet-level periodisation, the measurement of class share becomes more difficult as many measurement strategies are prone to error. Namely, the ESS survey asks for respondents’ reported vote choice at the previous election, whereas the EB asks for future vote intention. We devised a strategy such that we average the class share indicated in all the possible survey waves that were relevant for the election which led to the creation of the cabinet. For a full description of the coding process, refer to online Appendix 2.1.
We use a number of control variables which have been shown to influence the dependent variable welfare chauvinism policy and social democratic electoral success and policy agendas. A full table of indicators and sources can be found in online Appendix 2.2. We use the harmonised unemployment rate and GDP per capita (Brady et al., Reference Brady, Huber and Stephens2014) given the relationship between economic pressure and immigrant welfare rights reforms from the left (Harris and Römer, Reference Harris and Römer2023). The functional argument that higher migration flows lead to knee-jerk welfare restrictions from national welfare states (see Boräng, Reference Boräng2018 e.g.) is controlled for by using net-migration flows data. Following Römer (Reference Römer2017), more generous welfare states are found to be generous to immigrants and less welfare chauvinist. Therefore, we include a measure of all social expenditures as a proportion of GDP (Brady et al., Reference Brady, Huber and Stephens2014). Given the relationship between trade unions, labour rights and immigration policies from left-wing governments, we also control for union density. Lastly, we control for cabinet duration as we expect persistent cabinets to have a bigger impact on policy than short-lived cabinets. We also run models that include party competition variables, such as PRRP vote share and Green party share at the last election.
For all independent variables apart from vote share, cabinet length and cabinet party composition, the periodisation process creates cabinet-level variables which are the average of the first half of country-years in a given cabinet period (Schmitt, Reference Schmitt2016).
Empirical results
To account for the cross-section and time-series nature of the data, we run a series of regression models with two-way (country and cabinet period) fixed effects, a lagged-dependent variable and heteroskedasticity-robust standard errors with an autoregressive term of 1, to control for biases that emerge in the structure of the data. We report our findings for our hypotheses using this model specification, but also run a number of robustness checks using alternative specifications to assess the sensitivity of our models, as described below.
We now turn to hypotheses 2a and 2b, which expect that when middle classes dominate the social democratic electorate, social democratic parties as less likely to resort to immigrant-exclusion, whereas when working classes dominate, they are more likely to do so. To assess these hypotheses, we turn to Models 1–3 in Table 1. Model 1 shows the full model, without interactions, Model 2 shows the full model with interaction effect for working-class share, and Model 3 shows the full model with interaction effect for middle-class share. The dependent variable is continuous and ranges from 0 (no welfare chauvinism, same rights as citizens) to 10 (all immigrant groups excluded from welfare, no access).
Table 1. Main table, OLS regression, two-way FES, Huber-White standard errors AR(1)
Note:
* P < 0.1;
** P < 0.05;
*** P < 0.01.
Focusing on our main variable of interest in Model 1, social democratic cabinet share, the results suggest that a one percentage increase in their cabinet seat share leads to an increase in welfare chauvinism of 0.003 (on the 0–10 scale). In practice, all held equal, a cabinet with 100% social democratic seat share, such as in the UK, Spain and France in our sample, increases welfare chauvinism by 0.3. This may seem minimal, but given that often changes in welfare chauvinism are between 0 and 0.25 (see Figure 2A,11 in the appendix), this is no small change. Yet, to test our hypotheses, we move on to Models 2 and 3, where interaction effects are introduced.
Models 2 and 3, which interact social democratic cabinet share with the working-class and middle-class share, respectively, suggest that how social democratic government participation over the last 40 years has mattered for welfare chauvinist reforms seems to depend on the class composition of their electorate. Contrary to hypothesis 2a, however, it is not the working-class share which makes a difference but rather the size of the middle class. The negative interaction effect of −0.003 in Model 3 is significant at the 95% confidence level and indicates that social democratic parties refrain from contributing to welfare chauvinist reforms when their electorates were characterised by a high share of middle-class voters. Indeed, the coefficient of social democratic cabinet share is 0.013 when middle-class shares are at 0 and becomes negative when the middle-class share reaches over 47% of the electorate (see the interaction plots in Fig. 3). Such middle-class social democratic parties are, according to the regression analysis here, less likely to conduct welfare chauvinist reforms. This offers support for hypothesis 2b. Between around 20% and 65%, the confidence intervals become too large for conventional significance, therefore the results should be interpreted with caution. However, in Figure 2 we can observe real-world levels of middle-class share ranging from below 15% (Finland and UK) to above 70% (Norway), suggesting that the values where we find significant marginal effects are not merely hypothethical. The right plot of Figure 3 shows that, in contrast, the working-class share does not impact the coefficient of social democratic cabinet share to any conventional levels of significance. Plots with the distributions of the class share variable are shown in online Appendix 2A.12.
Figure 3. Effect of social democratic cabinet share on welfare chauvinism conditional on the share of middle-class voters (panel A) and the share of working-class voters (panel B) among their electorate.
Robustness checks
In the online Appendix, we show alternative specifications to account for potential biases within our model (Tables 2A.4–11). At the cabinet level, we check to see if the dependent variable as a level, not a change, thus without lagged-dependent variable, remains significant. Second, we remove the year-fixed effects and thus increase the variance within countries that can be explained. Third, we run the same model specification with adjusted data for the cases of Denmark in 1983 and Spain in 2000 as recommended by Römer et al. (Reference Römer, Leonce and Zobel2023, p. 1546). Fourth, we show the results given the inclusion of more party competition variables, such as vote share of PRRP and Green parties. Fifth, we calculate new working- and middle-class shares, using a sample of working-population-only voters, that is excluding non-active labour market participants to whom a number of occupational-relevant class characteristics may apply less. Our main finding regarding the middle-class share interaction with social democratic cabinet share remains significant and negative in all of these alternative model specifications. Finally, we present the models at the country-year unit of analysis, using two-way fixed effects, both with and without the lagged-dependent variable (2A.9–10). Despite the different unit of analysis, the results remain similar, with slightly weaker levels of statistical significance depending on the inclusion of a lagged-dependent variable. Overall, evidence suggests that our main findings are robust.
Conclusion and discussion
In this paper, we addressed expectations in the literature, positing that social democratic parties enact welfare chauvinist reforms to appease their working-class voters and to prevent them from shifting to vote for the radical right. Empirically, we tested these expectations by first assessing whether a class divide in preferences for welfare chauvinism/welfare inclusion of migrants exists generally and within the social democratic electorate. Then, we investigated whether reforms to immigrants’ welfare rights of social democratic governments reflect the class composition of their electorate.
Our findings suggest that social democratic middle-class constituents are much more willing to give immigrants access to welfare benefits than average citizens. Social democratic working-class citizens were, at least in 2008, even slightly more welfare chauvinist than average citizens. Furthermore, our regression analysis of policy reforms suggests that social democratic policies appear to be affected by the share of middle-class voters in social democratic electorates between 1980 and 2018 in the 14 Western European countries we have in our sample. In contrast, the share of working-class voters does not influence social democratic policy regarding welfare chauvinism.
These findings have wide-reaching ramifications for the study of policies at the intersection of immigration and the welfare state. Firstly, our findings contradict classical power resources theory, as we show that mainstream-left parties do not use welfare chauvinism to attract the working classes (Schmitt and Teney, Reference Schmitt and Teney2018), at least not in their policy agendas in government. Working-class power does not ‘beget working-class power’ (Gingrich and Häusermann, Reference Gingrich and Häusermann2015, 51), rather, the preferences of working-class voters for stricter welfare chauvinism are not played out in policymaking agendas of social democratic parties, even those with high shares of working-class voters. This implies that working-class voters may stick with social democrats regardless of their welfare chauvinism – and whilst research has begun to tackle the notion of priorities next to preferences (Häusermann et al., Reference Häusermann, Pinggera, Ares and Enggist2022), the impact of salience on vote choice requires much further investigation. Secondly, we complement other studies which argue that social democratic parties embrace their new, middle-class constituencies, who are supportive of immigration more generally (Abou-Chadi and Wagner, Reference Abou-Chadi and Wagner2020), by refraining from excluding immigrants from welfare access.
This paper also contained several limitations, which can be built upon in future work. Firstly, between voters and policy reforms, electoral rhetoric could be an important second step to assess whether the parties that enter government pledge welfare chauvinism at all (Döring and Schwander, Reference Döring and Schwander2015). This was not integrated here due to lacking data availability for the time span involved and because our research question called for the link between voters and governments to be assessed, but future studies could seek better measurements for party positions on welfare chauvinism over time to assess whether the demand of voters is strongly connected to electoral jousting between parties even more so than at the policymaking level. Secondly, we explore the micro-foundations of the left, but ignore the mainstream right, who are also central actors as government leaders and other work suggests that their electorates are also divided on immigration issues (Harris, Reference Harris2022). Thirdly, we do not study and therefore cannot conclusively determine the reasons for the disparity in welfare chauvinist preferences between the middle and the working classes, that is, whether they differ because of differing education levels, material circumstances or experiences specific to their occupations. Finally, although this paper was focused on the effects of variables on an outcome, more in-depth case knowledge, particularly for pathway cases or outliers, could strengthen the argument presented here and could help to further gauge the direction of causality. Our design does not allow us to determine with certainty whether class composition causally affects social democratic policy or whether, for example, social democratic parties running on pro-immigration platforms become more middle class and are then also more likely to implement less welfare chauvinist reforms. Additionally, the role of threat and ‘contagion’ effects from PRRPs and Green parties’ rhetoric at election time could not be fully modelled in this paper, which could be another variable to explain why class shares are important at some elections and not at others.
On balance, we find that social democratic parties in government over the last four decades have contributed more to welfare chauvinism than other parties. However, our findings also suggest that this may be changing with an increasing share of middle-class voters. Where the middle class has become the predominant constituency of the social democrats, they actively contribute to improving immigrants’ welfare rights. Whilst ongoing processes of educational expansion and occupational upgrading should expand the middle classes further, green and other new-left parties increasingly contest these middle-class votes, proving predictions of further trends of class shares in social democratic electorates difficult. Both structural changes as well as programmatic choices (concerning welfare chauvinism and other issues) will, hence, affect the future class composition of social democratic electorates which – taking our findings seriously – in turn will influence whether social democratic parties are likely to act as defenders or opponents of immigrants’ welfare rights in the future.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S1755773923000346.
Data Availability
See online Appendix A2.2 for a full list of the datasets used. ImmigSR data are currently available on request following the recent end of an embargo period and should be available by contacting [email protected]
Acknowledgements
This paper was benefited from the close reading and comments of Juliana Chueri, Sijla Häusermann, Samir Negash and Leonce Röth, as well as the critical and helpful attendees of CES 2022, ESPANET 2022 and the VelNET conference in Odense in April 2023. Particular thanks also go to the ImmigSR team in Bremen, Friederike Römer and Jakob Henninger and to research assistants Carolina Reiners and Kevin Wang for their support. Finally, we thank the two reviewers and the EPSR editors for their comments and contributions to the final paper. All omissions and mistakes are our own.
Funding
No extra funding was provided for this article.
Competing interests
The authors declare none.
Open access
