Class and social policy in the knowledge economy SILJA HÄUSERMANN*, MICHAEL PINGGERA*, MACARENA ARES°, AND MATTHIAS ENGGIST* *Department of Political Science, University of Zurich, Switzerland; ° Department of Political Science, Universitat Autònoma de Barcelona, Spain Corresponding author, Michael Pinggera, pinggera@ipz.uzh.ch Abstract. Recent studies of welfare state attitudes in the knowledge economy find very high generalized support for generous welfare state policies, both among the working and the middle classes. Has class become irrelevant as a predictor of social policy preferences? Or do we simply misconceptualise today’s class conflict over social policy? To what extent has it changed from a divide over the level of social policy generosity to a divide over the kind of social policy and – more specifically – over the relative importance that should be given to different social policies? Answering these questions is not only relevant to understand welfare politics in the 21st century, but electoral politics as well: only when we understand what working- and middle-class voters care about, can we evaluate the role distributive policies play in electoral processes. We use original survey data from eight West European countries to show that middle- and working-class respondents indeed differ in the relative importance they attribute to social investment and social consumption policies. Middle-class respondents consistently attribute higher absolute and relative importance to social investment. We also show that this emphasis on investive policies relates to the middle class expecting better future economic and social opportunities than the working class. This divide in anticipated opportunities underlies a new kind of working- vs. middle-class divide, which contributes to transforming the class divide from a conflict over the level of social policy to a conflict over the priorities of social policy. Keywords: class, social policy, knowledge economy 1 INTRODUCTION Is there still a class divide over social policy in the knowledge economy of the 21st century, and if yes, what is it fought over and what drives different class preferences? While diverging class interests over social policy seemed obvious, and indeed were at the very centre of early theories of welfare state emergence and development (in particular the power resource theory, e.g. Korpi, 1983), this has become a major theoretical and empirical question over the past two decades. Indeed, as predicted by theories of the new politics of the welfare state (Pierson 2001), generalised support for existing social policy provision today is so massive that it transcends by far both the working class and the electorate of the Left. These high levels of state provision of social security reflect strong generalised welfare state support also among the middle and upper classes, while the working class has always supported generous social benefits. Why are the middle classes so supportive of the welfare state? Pierson himself attributed middle-class support for generous social policy to the endogenous constituency formation of maturing welfare states, with all citizens becoming stakeholders of generous provision, especially in those policies (such as pensions, education, or health) where risks are widely spread (Jensen, 2012; Rehm, 2016). Other theories highlighted more specific drivers of middle- class support for the welfare state, e.g. universalistic values (Korpi & Palme, 1998), skill specificity (Iversen & Soskice, 2001), or the spread of egalitarian values among the new middle class (Beramendi et al., 2015; Kitschelt, 1994). And with existing welfare states strongly bolstering support via feedback mechanisms (e.g. Svallfors, 1997), we today indeed observe generalised levels of support that approach unanimity: Garritzmann et al. (2018a) showed that across Western Europe, about 75-85% of respondents in representative surveys support a further expansion of higher education, childcare services, and labour market training. Data from the ISSP 2016 show similarly high levels of support for old age pension provision: over 90% of respondents across Europe, e.g., disagree with cutting back pension benefits. These findings would imply that all political parties have an incentive to promise expansive social policies and – furthermore – that they should manage to appeal to voters based on these promises. However, that does not seem to be the case. For one, parties still do differ strongly in the extent to which they emphasise social policy in their electoral platforms (Green-Pedersen & Jensen, 2019), with the Left clearly leading the rank order. However, despite these preference patterns, the Left seems to have a hard time mobilising growing electoral shares on 2 the basis of their expansive social policy programs, especially among the working class. One explanation for this puzzle could be that voters simply care more about other ‘second dimension’-issues (such as cultural liberalism, migration flows or international integration), rather than about social policy, a claim that has been evidenced prominently for many years now (e.g. Beramendi et al., 2015; Kitschelt, 1994; Kriesi et al., 2008, Oesch and Rennwald 2018). However, measures of issue saliency among voters cast doubt on the interpretation that this is the entire story: while it is true that new ‘second dimension’-issues have massively gained in saliency, social security keeps ranking among the very top priorities of citizens (Rovny & Polk, 2019; Traber et al., 2018). In this paper, we explore a different explanation: the focus on generalised welfare support and saliency may obfuscate the fact that working- and middle-class voters prioritise different kinds of welfare provision. More specifically, we test whether middle-class voters care more strongly for social investment (SI) policies than working-class voters. SI policies are social policies that ‘create, preserve or maintain human capital’ (Garritzmann et al., forthcoming in 2021), i.e. which invest in the capacity of individuals to support themselves, particularly in the context of the knowledge economy where skills are a crucial factor. We hypothesise that the class divide over social policy in the 21st century may lie in the kind of welfare provision that is prioritised, more so than in the overall level of welfare provision citizens support. Priorities become more relevant, because both elites and citizens perceive the context of distributive politics as being constrained by budgetary limits. In our survey of 12,129 citizens in eight West European countries, no less than eighty percent of the respondents agree with the statement that ‘the limits of taxation have been reached’, and over two-thirds of respondents agree that ‘social policy improvements for one social group sooner or later come at the expense of other social groups’. Hence, budget constraints are real in the perception of a large majority of voters, and this is likely to transform the ways in which social policy matters to voters: while almost all voters may overall support generous policies, they are likely to differ in the importance they attribute to different benefits and services. This is of obvious relevance to our understanding of how and why welfare politics matters politically and electorally: if voters diverge in their welfare priorities, programmatic social policy appeals may only resonate with voters when they match their priorities. Hardly any research to date has studied either these priorities, their determinants or their implications, and this is no surprise, as we largely lacked both theory and adequate data until 3 very recently (as contributions starting to address this idea, see e.g. Bremer & Bürgisser, 2020; Enggist & Pinggera, 2021; Gallego & Marx, 2017; Häusermann et al., 2019; Pinggera, 2020). In this paper, we introduce a newly collected dataset on social policy attitudes in eight Western European countries (Häusermann et al., 2020), which includes extensive and varied measures of priorities for different policy fields and logics of welfare provision, in particular social investment vs. social consumption policies (Beramendi et al., 2015). This data allows us to study the prevalence and determinants of a post-industrial class divide regarding SI reforms. Our findings generate two key contributions. First, we show that working- and middle-class citizens differ consistently and systematically in the importance they attribute to SI. This class divide holds both across different occupational sub-classes, and cross-sectionally. Second, we show that how individuals evaluate their future economic opportunities is an important mechanism linking class and priority. Together, these two contributions provide evidence for a new class divide over social policy in Western Europe. By implication, the mass politics of social policy is neither converged nor depoliticised but plays out along new attitudinal divides. THEORY In the theory and analyses of welfare state politics in the 20th century, different preferences for social rights and redistribution were seen as key to the class divide between left- and right- wing parties and their electorates. Demand for redistribution and social protection among working-class voters fuelled their disproportionate vote for left-wing parties, who defended these claims against the more fiscally restrictive and less redistributive policy claims of right- wing parties and their constituencies. Today, however, a sizeable share of the literature argues that class divides over social policy preferences do not explain electoral and policy outcomes anymore. One strand of this literature sees the reason for declining class politics in the behaviour of elites: the literature on cartelisation, producer group politics and winner-takes-all politics (e.g. Baccaro & Pontusson, 2016; Hacker & Pierson, 2010) has argued that mass politics at the level of voters has become hollowed out, due to hegemonic business interests, as well as policy convergence of the major parties. The second strand of this literature goes even further: it doubts whether voters from different classes still differ markedly in their social policy preferences, at all. In particular, 4 studies on the electoral realignment of mass politics have shown that the key divides in preferences between working- and middle-class voters are not to be found anymore regarding redistribution or welfare spending, but regarding their preferences on policy issues that form a second dimension of political conflict, opposing liberal-universalistic to traditional- particularistic attitudes (Häusermann & Kriesi, 2015; Kitschelt, 1994; Kriesi et al., 2008). This would imply that class politics is very much alive, but it plays out mostly on issues which are not neatly aligned with the traditional distributive left-right conflict (such as minority rights, gender equality, supranational integration, or migration). The decline of ‘class voting’, i.e. the declining capacity of working-class membership to predict vote choice for the left (e.g. Knutsen 2006; Evans and Tilley 2017) is consistent with such a perspective that emphasises the declining relevance of class divides over social policy. Even the welfare state literature itself contributes to this argument about fading class divides: as citizens throughout social strata have become stakeholders of the existing (mature) welfare state, their preferences have converged (Pierson, 2001). Moreover, several studies have shown that class predicts social policy attitudes only when it comes to very specific, concentrated risks (Rehm 2016; Jensen 2012), but much less so when it comes to the pillars of the mature welfare state, such as pensions, education and health. However, all these contributions tend to overlook important substantive developments in welfare politics, which underlie a transformation, rather than a fading out of class divide over the welfare state. In particular, SI policies have become a very influential social policy paradigm across Western Europe since the 2000s (Esping-Andersen et al., 2002; Hemerijck, 2013; Morel et al., 2012; Hall 2020). This emergence of SI politics, we contend, has re- politicised class divides in the area of social policy (Beramendi et al., 2015): in the 21st century, working- and middle-class voters differ in the kind of welfare provision they prioritise, more so than in the overall level of welfare state support. In this study, we thus focus on SI as a key area of welfare state reform on the current agendas of West European countries. SI has spread both as a policy paradigm and ‘toolbox’ from the mid-1990s onwards in the context of ongoing structural dynamics such as globalisation and automation, which result in altered employment patterns and emerging new social risks (Bonoli, 2005). SI policies differ in their logic of welfare provision from traditional forms of social security. Their primary goal is not to de-commodify citizens, but to support citizens in the labour market through ‘creating, mobilizing and preserving human capital and capabilities’ 5 (Garritzmann et al., forthcoming in 2021) 1. They differ from traditional, ‘consumption’- oriented social policies 2 in two ways: first, social consumption policies provide material security, but do not aim at enhancing or improving human capital and citizens’ own chances in the labour market. Second, social consumption policies entail financial transfers that have an immediate distributive effect, and that make the beneficiaries of transfers clearly identifiable. By contrast, when it comes to SI, costs occur in the immediate, but returns are both temporally distant and more uncertain in distributive terms. Hence, since investment and consumption reflect different logics of how social policy sustains security and inclusion, conflict over SI is structured – both at the level of voters and at the level of political parties – in ways that differ from conflict over social consumption (Abou‐Chadi & Immergut, 2019; Bremer & Bürgisser, 2020; Garritzmann, et al., 2018a; Häusermann & Kriesi, 2015). Several studies have hypothesised that the preferences of working- and middle-class voters regarding SI and consumption are likely to differ. Three main mechanisms have been advanced for such a possible difference in demand. First, several studies have shown ‘Matthew effects’ of SI policies, i.e. regressive distributive effects that are due to policy design or to the fact that middle- and upper-class citizens seem to know better how to utilise SI policies (such as universal childcare provision, education and training) than lower-class citizens (Bonoli & Liechti, 2018; Pavolini & Van Lancker, 2018). Alternatively, Beramendi et al. (2015) have argued that universalistic values - i.e. favourable attitudes towards social liberalism, migration and other second dimension issues, which politically motivate many new middle-class voters - resonate with the future-oriented logic of SI, as the benefits of investive social policies are less clearly targeted towards specific beneficiaries. Finally, recent studies have introduced political trust as a determinant of SI support, since allocating funds to temporally distant outcomes requires trust that governments and institutions will be able and willing to deliver (Jacobs and Matthews 2017, Garritzmann et al. 2018b). All three mechanisms – self-interest, values and trust – might explain different levels of SI support among working- and middle-class voters, i.e. differences in actual policy positions. 1 SI, however, is not the same as activation, certainly not negative activation (Bonoli, 2013). While sanctioning activation policies (‘workfare’) pursue the goal of commodification, SI pursues the dual goal of social inclusion and economic prosperity. 2 Terminologically, SI policies have been contrasted to traditional income-replacement policies. Those latter have been termed ‘consumption’, ‘compensation’ or ‘transfer-based’. We use here the term ‘consumption’ because it refers to the temporal distinction between immediate vs. deferred benefit yield. 6 However, while some empirical studies indeed find slightly higher levels of SI demand among middle-class voters, the overall finding is again one of very high generalised levels of policy support across different social groups (Bremer & Bürgisser, 2020; Garritzmann, et al., 2018a). In economic terms, this is highly plausible, since there is little reason not to support expansive SI policies even for working-class voters, as long as there is no budget constraint attached to it. By contrast, we examine in this study whether working- and middle-class voters differ in the importance they attribute to SI. Theoretically, it makes sense to expect that people do hold actual preferences over policy importance: not only are perceptions of fiscal constraints on policy choices extremely widespread in the population (Barnes & Hicks, 2018; Bremer & Bürgisser, 2020), distributive outcomes of policies are also different across respondents, in particular when it comes to SI policies (Bonoli & Liechti, 2018; Pavolini & Van Lancker, 2018). In a context of (perceived) fiscal constraint, these differential benefits become rivals, and thus we would expect priorities to diverge. Empirically, we thus use budget constraints and trade-off scenarios in our measurement strategy to elicit respondents’ preferences regarding the importance they attribute to particular policies. Hence, while we expect both working- and middle-class voters to generally support both SI and social consumption, we expect a stronger class divide over priorities regarding SI. We see three mechanisms affecting differential priorities by class: both political trust and universalistic values are expected to drive priorities, because of the temporal dimension that distinguishes SI from social consumption (Garritzmann et al., 2018b; Jacobs & Matthews, 2017): SI involves costs in the present in order to reap benefits in the future. The benefits that arise from expenses for schools, childcare or workforce training are both temporally distant and relatively indeterminate in their distributive effects. In other words: it is difficult for the individual to foresee to what extent she will be a direct beneficiary of the investment. Therefore, we suggest that working-class voters prioritise more immediate (and secure) benefits, and middle-class voters are more trusting, can afford to be more ‘patient’, and attribute overall more value to distant egalitarian-universalistic outcomes. In addition to these determinants that the literature has already theorised to some extent, we propose an additional, more economic mechanism: we argue that – as with every investment – the importance voters attribute to these SI policies depends on how they evaluate their future. The more positively people evaluate their future opportunities in work and in society, the more value they should attach to investments, even if 7 they are at the cost of present potential income. However, if people evaluate their opportunities more negatively, we expect them to prioritise current compensation. We thus think of the evaluation of opportunities as a mediating variable, linking class differences to SI priorities. Given this emphasis on prospective economic opportunities as drivers of SI priorities, we expect an actual class divide between middle- and working-class respondents, rather than a more specific divide between subgroups of the middle and working classes. Indeed, the existing literature on SI support has argued that socio-cultural professionals are strong supporters of SI because their occupations tend to be tied most closely to the (semi-)public services, and because they hold more universalistic values than other classes (Beramendi et al., 2015; Häusermann & Palier, 2017; Neimanns et al., 2018). However, if – as we hypothesise in this study – the middle-class prioritisation of SI reflects their economic opportunities, prospects, and expected returns on these investments, we would not necessarily expect within-middle- class differences regarding differences in priorities, as those are likely to be similar for socio- cultural professionals, managers, and technical specialists. Consequently, we do not formulate a hypothesis on within-class differences, but we will disaggregate the findings across middle classes empirically3. Hence, the two key hypotheses of this study are the following: H1: Middle- and working-class voters differ in the importance they attribute to social investment. H2: The relationship between class and priorities for social investment is mediated by the subjective evaluation of future economic opportunities. Since we expect the divide between middle- and working-class preferences to be the result of structurally diverging (prospective) economic opportunities, as well as political trust and universalistic values, we hypothesise to find the same divide across all countries under study. However, the case selection, which deliberately includes countries from different welfare regimes, i.e. countries from Nordic (Denmark, Sweden), Anglo-Saxon (Ireland, United Kingdom), Continental (Germany, Netherlands) and Southern European (Italy, Spain) institutional welfare regimes (Esping-Andersen, 1990; Ferrera, 1996), allows us, via 3 Expectations regarding heterogeneity are less substantiated theoretically when it comes to working class heterogeneity, as production and services workers tend to have highly similar preference profiles when it comes to distributive policies (Ares, 2017). 8 observable implications, to gauge the plausibility of an alternative mechanism driving class preferences, i.e. institutional context (on positive and negative institutional feedback effects, see e.g. Fernandez & Jaime-Castillo, 2013). These countries differ systematically on the existing availability of SI policies to different social groups. These differences can be leveraged analytically: if the class differences were mainly a product of institutional availability of SI policies, we would expect to find lower class differences in both Nordic and Southern European countries, where these policies tend to be either universally available or universally scarce (Ferragina & Seeleib-Kaiser, 2015; Horn & Van Kersbergen, forthcoming in 2021; Morel et al., 2012). If strong demand reflected the cost middle-class families – as the main users of the services – pay for childcare and education, we would find particularly high priorities among middle-class families in Anglo-Saxon countries, where costs are highest. However, if the divide is indeed rooted in structural differences between middle- and working-class voters, as we hypothesize, then we should find the same divide across the different institutional contexts. Beyond this, we obviously do expect cross-country variance regarding the levels of prioritisation of the respective policies, as they are likely to depend on national reform agendas and country-specific temporalities of these agendas at the time of our survey. DATA AND OPERATIONALISATION To assess our hypotheses, we use original data from a survey conducted in the project ‘WELFAREPRIORITIES’ 4. Data were collected in eight Western European countries with 1500 respondents each in Denmark, Sweden, Germany, the Netherlands, Ireland, the United Kingdom, Italy and Spain. The questionnaire and sample design were in our hands, while the actual fieldwork was done in cooperation with a professional survey institute (Bilendi) using their online panels. The target population was a country’s adult population (>18 years). The total sample counts 12,129 completed interviews that were conducted between October 2018 and January 2019. Details on how we ensured high data quality and representativeness are presented in the Appendix (A1). The full dataset is also presented extensively in a specific working paper (Häusermann et al., 2020). The survey includes a wide range of items capturing social policy positions as well as priorities. In the first part of the analysis, we show that support for social policies is overall high and that 4 http://welfarepriorities.eu 9 class differences in these positions are relatively small. We establish this by looking at five variables that capture support for pensions, childcare, and higher education in different ways. Specifically, we asked: ‘To what extent do you agree with the following policy reform proposals? The government should increase old age pension benefits’ and ‘The government should reduce old age pension benefits’ to assess support for pensions as the most relevant social consumption policy on the reform agendas of Western Europe and ‘The government should expand access to good-quality childcare services’ as well as ‘The government should invest more in education’ to capture support for SI policies. Furthermore, respondents were asked to speak their minds on work-family reconciliation: ‘It is the government’s responsibility to support working parents’. All items have been answered on a 1 (Disagree strongly) to 4 (Agree strongly) scale. Most importantly, the survey includes several items that enable us to measure social policy priorities. Our analyses focus on childcare and tertiary education as SI policies, and on old age pensions as the typical (and financially most relevant) consumption policy. For this paper, we use variations of two types of questions implemented in the survey: point distributions and a specific trade-off question. In the point distribution questions, respondents were asked to allocate 100 points to six items, reflecting the relative importance they attribute to different strategies of welfare state expansion. The six items were presented in randomised order, so as not to prime the importance given to them. Through this type of question, we can account for the multidimensionality of welfare preferences, while at the same time we pay respect to the constraint that is inherent in the concept of priorities. We implemented two types of these questions: first, we asked interviewees to distribute 100 points to six social policy reforms. The reforms include specific expansive measures in the fields of old age pensions, childcare and tertiary education and vary (per field) in whether they affect all citizens (universal) or only specific groups (targeted) 5. In a second task, respondents were asked to allocate points to social policy fields 6. In the presentation of findings, we focus on policy fields rather than specific reforms, but we include reforms to probe robustness. Using these items, we operationalise our dependent variables as the number of points given to childcare, the number of points given to 5 For old age pensions: increase minimum level vs. benefits for everyone; for access to good quality childcare services and higher education: increase access for low-income families or for everyone. 6 Old age pensions, childcare, higher education, unemployment benefits, labour market reintegration services, services for the social and labour market integration of immigrants. 10 tertiary education, and the number of points given to old age pensions. The variables can take a value between 0 and 100. The second measure of priority consists of a trade-off question: respondents were asked for their evaluation of a specific reform as follows: ‘The government increases the availability of good-quality childcare, at a cost of lowering child benefits’. Answers range from 1 (‘Completely inacceptable’) to 4 (‘Completely acceptable’). Through this wording, the costs of a reform were made very explicit and concrete. Thereby, we gain information about the degree to which aversion to cutbacks or preference for expansion shapes responses to the trade- off. 7 We use both point-distribution and trade-off questions to probe robustness across question types: while the trade-off question requires a choice that the respondent may have a hard time to make (and may not have to make in real life), the point-distribution question does indeed allow for respondents attributing an equal number of points across all fields or reforms. Hence, with this type of question, we do not force respondents to prioritise (consider that a mere rating question of the different fields/reforms would also force the respondent to attribute unequal importance). Nevertheless, the findings are consistent across the two question types. Previous studies have attempted to measure preferences with some sort of budget constraint, of course. Most importantly, several surveys have extended the ‘position’-question by pointing out to the respondents that increases of spending or benefits in the proposed field would come at the cost of higher taxes (e.g. Hansen 1998). Our ‘position’-questions do not include this information, and one may hence wonder whether the small differences we find between middle- and working-class respondents are due to this omission only8. However, previous research has shown that mentioning a tax constraint lowers the average support level only very weakly, and – more importantly for our study – it does not change the distribution of preferences. In the findings by Busemeyer and Garritzmann (2017) e.g., mentioning a rise in taxes or in debt as a consequence of higher education spending does reduce support for this type of spending; however, income (as a proxy for class) does not relate significantly to 7 While we are not aware of another survey using the same point distribution question as this study, both Busemeyer and Garritzmann (2017) and Bremer and Bürgisser (2020) have used similar trade-off questions before. 8 We are grateful to an anonymous reviewer for pointing this out. 11 supporting education, even when the constraint is mentioned. Hence, simply mentioning the constraint does not reveal a class-divide, as it does not seem to validly measure importance. Using such newly developed measures requires careful evaluations of validity, which we have conducted. First, respondent behaviour indicates that they were able and willing to engage with the task at hand: even though they could have avoided the difficult task of indicating the relative importance for particular policies, fewer than 2% of respondents attributed equal point numbers across policy fields. Also, fewer than 6% of all respondents attributed 100 points to one field and 0 to all others. We have also conducted extensive analyses to test the internal validity of our items (Ares et al., 2020). Regarding the point distribution question, we used data on people’s priorities regarding welfare retrenchment (as opposed to expansion) to test if reported preferences are consistent. Our data show that 85-90% of respondents gave consistent answers, i.e. they did not simultaneously prioritise retrenchment and expansion in the same policy field. Regarding the trade-off questions, we were able to confirm that strong support for one side of the trade-off (measured through a direct question e.g., regarding support for childcare services) indeed relates consistently to also prioritising this specific item in the trade-off question. We were also able to replicate this same analysis with the INVEDUC data (Busemeyer et al. 2018), which contains similar trade-off questions to ours, and we found the same positive correlation, confirming internal validity. Regarding external validity, we find roughly the same ‘order of priorities’ in the point distribution question across countries, with pensions being prioritised on average, followed by education and childcare, then labour market policies, and finally, policies for immigrants (see our working paper on the dataset, (Häusermann et al., 2020) 9. Bremer and Bürgisser (2020), who also study the relative importance of social policies find the same ‘priority order’, which is largely consistent with the findings of the literature on welfare deservingness (van Oorschot 2006). These studies are not concerned with the divide between investment and consumption, but they use similar types of data. The fact that the preference patterns are similar across the different datasets increases our confidence in the validity of our measures. More specifically, Busemeyer and Garritzmann (2017) have also investigated the relative importance respondents attribute to education policies relative to pension benefits (which can be roughly interpreted as 9 In the southern European labour market, policies are somewhat more prioritized than in the other countries, which adds further plausibility to the preference distribution in our data. 12 investment vs. consumption). The focus in their study is on ideology as the driving factor of education preferences, but their models - consistently with ours - show that income and education relate positively to supporting education at the expense of pension cuts. The main independent variable is class. More specifically, we study differences in SI preferences between the working and the middle classes. A voters’ class position is measured using the well-established Oesch-class scheme based on marketable skills (high, low) as a hierarchical dimension of class differentiation, and work logic (independent, technical, organizational, interpersonal) as a horizontal dimension (Oesch, 2006). The resulting eight- class scheme allows us to identify the working class that includes production workers, clerks, and service workers who are all characterized by low(er) marketable skills. On the other hand, the middle class includes technical specialists, managers, and socio-cultural specialists with high(er) marketable skills. Occupational classes with an independent work logic, regardless of their level of marketability (large employers and small business owners), have been discarded from the analyses, as our proposed mechanism about future labour market opportunities does not necessarily apply to the self-employed. Using this classification, we end up with 4944 (39.5%) of observations coded as working class while 4436 (35.5%) of respondents belong to the middle class. 695 (5.6%) self-employed have been excluded from the main analyses, as well as 2426 (19.4%) observations for which we do not have information on their occupational class 10. In terms of mechanisms, our focus is on the subjective evaluation of economic opportunities 11. Respondents were asked: ‘The world is changing fast. If you think of your future, how do you rate your personal chances of being in good, stable employment until you will retire?’. The answers on an eleven-point scale (very bad – very good) were normalised (to a 0-1 scale) to simplify comparison with other mechanisms. The literature points to alternative explanations for class differences in priorities, most prominently political trust and universalistic values. These variables were included in the analysis as well. Political trust was captured on a 10- point-scale asking: ‘In general, do you trust politicians in your country?’. A battery of items 10 All models have been re-calculated with an alternative class variable that is based on income and education rather than occupation. Results (see A2-3 in the Appendix) with regard to SI priorities remain stable, or in some cases even become somewhat stronger (opportunity appears as an even better mediator of the class effects). 11 We have also tested the effect of social opportunities with the following question: “Now think of your overall quality of life. How do you rate your personal chances of leading a safe, fulfilled life over your life course?”. The findings are entirely consistent. 13 (similar to the one used to measure universalistic vs. particularistic values in the literature, see Häusermann & Kriesi, 2015) was used to compile an index for universalistic values. Respondents were asked about the following five statements: ‘Immigration is a threat to our national culture’, ‘Gay and lesbian couples should have the same rights to adopt children as straight couples’, ‘All in all, family life suffers when the woman has a full-time job’, ‘Immigration is a threat to the national labour market’, and ‘European integration has gone too far’. An exploratory factor analysis reveals that these five items load on one factor (eigenvalue=2.58). The hypotheses have been tested estimating multivariate regression models. We dichotomised the dependent variables in the position models and run logistic regression models. For the priority models, we run OLS regressions (except for the trade-off model, where we estimate an ordered logit model). To test the mechanism proposed in hypothesis 2, which expects that the class effect on priorities is driven by perceived opportunities (or alternatively political trust or universal values), we use different alternative methods, most importantly structural equation models (using lavaan and lavaan.survey in R). These models are based on a somewhat smaller sample of 9312 observations since it excludes pensioners who were not asked to assess their opportunities, and we applied listwise deletion of missing data. The structural equation model includes a specification of the direct effect (priority regressed on opportunities, trust, universalistic values, and class, as well as controls and country dummies), the mediator models (opportunities, trust, and universalistic values each regressed on class), as well as specifications of indirect effects (from class to priorities through each of the three mediators) and the total effect (sum of the direct effect of class and the three mentioned indirect effects). Regarding the points attributed to childcare and tertiary education the dependent variable was rescaled to range from 0 to 1, while the trade-off variable was kept on the original 1-4 scale. Appendix (A11) presents the exact specification. All models presented in the results section include controls for age, sex, and parental status 12. Further, all regressions include country-fixed effects and are computed using a weighting 12 We intentionally keep the models sparse since we are not primarily interested in the marginal effect of class, but rather in the conflict over welfare policies structured along the lines of class. Hence, typical control variables such as income or education (closely interrelated with class) would rather mask than clarify the structure of conflict. However, in the Appendix (A4), we provide the same models including a range of additional control variables (living with a spouse, single income household, main income earner in household, citizenship, and atypical employment) that do not substantively affect the findings. Also, we replicate the estimations using an alternative operationalisation of class based on income and education, cf. FN 10 (A2-3). 14 variable ensuring a close-to-population distribution of age and sex and educational attainment. Excluding the weights does not change the results substantively. FINDINGS Position vs. priorities We start by establishing that class differences are very small when it comes to positional preferences, both regarding social consumption and SI. Table 1 shows the findings of relating working- and middle-class membership to preferences over the expansion and the retrenchment of old age pensions, the expansion of childcare services, spending on education, as well as to attitudes on whether or not it is the governments’ responsibility to provide services to help working parents reconcile work and care duties. [Table 1 about here] We see that when it comes to pension policy – taken here as an example of social consumption policies – middle-class voters are somewhat less likely to support further pension expansion than working-class voters, which at first glance goes against the assumption our study builds on. Hence, there is a significant but weak class divide regarding social consumption expansion. However, class differences (even including pensioners, who are over-proportionally working class and who have stronger preferences for pension expansion than working-age working- class voters) are non-significant in three countries (Spain, Italy, Netherlands). In addition, old age pension reform nowadays is concerned more strongly with retrenchment than with expansion in most countries. Therefore, we also include position preferences on pension retrenchment in model 2. As expected, we find no class difference there, neither in the full sample, nor in the one limited to active respondents, nor in any of the countries included. Regarding attitudes towards the expansion of services in support of working parents we find that support is slightly smaller among middle-class voters on this typical SI reform, but significantly so only in Ireland. With all other countries individually, whether including or excluding pensioners, we find no significant class differences. Finally, the higher support among the middle class for investment in education stems exclusively from the UK, whereas we find no class differences in the other seven countries. 15 Turning to priorities instead of positions, the findings, as expected, look different. We focus on the pooled findings first and discuss disaggregated findings for occupational class and countries subsequently. Columns 1-3 in Table 2 present findings on class predicting the number of points respondents allocated to the expansion of childcare and higher education, as well as to the expansion of old age pensions for reasons of comparison. On both items measuring SI importance, middle-class respondents differ decidedly from working-class respondents. This finding is robust in several ways: it holds when we exclude pensioners from the sample, and middle-class respondents also attribute higher importance to SI expansion than working-class respondents when we present them with a specific reform proposal, i.e., an increase in the availability of good-quality childcare for all families, instead of policy fields 13. Finally, middle- class respondents attribute overall less importance to the expansion of old age pensions (social consumption). This finding also holds when pensioners are excluded. [Table 2 about here] Moving on to priorities for SI relative to social consumption, column 4 in Table 2 shows that the general finding of middle-class voters prioritising SI more strongly than working-class voters also holds ‘within’ family policy, when asking respondents, whether they would find it acceptable to expand childcare services at the expense of child allowances. On average, middle-class respondents find it more acceptable than working-class respondents to reallocate parts of government expenditures that go into child allowances to childcare services. This finding, however, is weaker, and significant only in half of the countries (we discuss country- specific findings more extensively below). [Figure 1 about here] Figure 1 illustrates the substantive effects for positions (left hand side) and priorities (right hand side). First, the numbers on positions show just how high and generalised support for all these policy reforms are. The probability that respondents support the expansion of pensions, childcare services or education spending lies between 80 and 95% for both classes. Second, as 13 The robustness with the alternative ‘priority-question’ (reform instead of field) is not only relevant because the wording of the answer item changes, but also because the set of possible alternative choices changes. While the reform-question compares only pensions, childcare and education reform, the field-question provides respondents additionally with the option of allocating funds to policies supporting the unemployed or to policies supporting the integration of immigrants. 16 Table 1 above has shown already, differences in position on childcare and education between working- and middle-class respondents are tiny. However, differences between classes turn highly significant if we look at priorities, clearly indicating a class divide over priorities that we cannot observe when looking at positions. As the panels on the right show, middle-class respondents allocate considerably more points to SI policies, here captured by childcare and education, and less points to consumptive measures such as pensions. The point difference on average is between 1-4 points. This may seem little, but it must be judged against the in-sample range, and in light of the two very broad and heterogeneous class categories (std. deviations of the point allocations are between 12-20 points). Appendices A5-6 present an in-depth discussion of the effect sizes and – in particular – demonstrates that the substantive effects presented here are no smaller than that of well-established class differences on other politically salient attitudes (such as cultural liberalism, EU integration or immigration). We also disaggregated the findings of Figure 1 by the more specific occupational classes (distinguishing within the middle class between socio-cultural professionals, managers, and technical specialists, and within the working class between service workers, production workers, and clerks). We report the findings in the Appendix (A7). What they confirm is clearly a divide by overall class, rather than by more specific occupational groups. In particular, managers, technical specialists, and socio-cultural professionals do not substantively diverge in the importance they attribute to childcare or tertiary education 14, and neither do production workers, service workers, and clerks 15. We interpret these findings as suggesting that the class divide we observe is not driven by particular sub-classes and their specific values or sector of employment, but by structural differences between the middle and the working classes regarding their labour market conditions, human capital, and the expected effects of these policies on their opportunities, a hypothesis we will test more systematically in the final part of the analysis. Overall, this section has shown that – contrary to positions concerning SI and social consumption – there is indeed a consistent class divide over the importance attributed to SI. 14 The only noticeable difference is that socio-cultural specialists attribute the least importance to old age pension expansion, but even there the differences are weak and they do not concern our main findings on SI. 15 We also disaggregated these analyses by class and country, and while the samples become too small for meaningful interpretation, they confirm the general finding that the middle class, across its sub-groups, consistently attributes more importance to SI than the working class. 17 Hence, the apparent unanimity in terms of welfare support appears to actually hide a potential new form of class conflict over social policy priorities. To interpret both the substantive importance of this finding and to get closer to the sources of different priorities, we have replicated all the above analyses at the level of the individual countries. If middle- vs. working-class divides were mainly the result of institutional feedback processes, we would expect systematic differences depending on the existing availability of and access to SI policies. However, what we do find (see Appendix A8-A10) is that – despite obviously smaller estimation samples - across all these different institutional contexts, middle- class voters attribute higher importance to the expansion of SI than working-class voters, and the reverse holds for old age pensions. We interpret this as suggestive evidence (being aware that we do not have a strong research design for these analyses, see discussion above) that the divide we find is indeed rooted in structural differences between middle- and working-class voters, which sustain the mechanisms (values, trust, economic opportunities) that we will study explicitly in the next section. The pervasiveness of the middle- vs. working-class divide does not mean that country differences are generally irrelevant for the politics of SI vs. social consumption, of course. In particular, the class divides occur at different absolute levels of prioritisation of the respective policies, as shown in the Appendix. Childcare, for instance, has been an overall more salient policy field in Ireland and Germany at the time of our survey than in the Netherlands and Spain, and tertiary education has been more salient in Italy or Ireland than in Sweden. These differences relate to country-specific temporalities of reform agendas, rather than to systematic institutional (regime) differences. Hence, they do not distract from our main finding that a middle- vs. working-class divide on the prioritisation of SI is observable in Western Europe. Determinants of class divide over priorities In this part of the analyses, we argue that the middle class attributes more importance to SI than the working class, because of their more positive evaluation of future opportunities. Hence, we provide evidence for an economic rationale underlying SI priorities. 18 Table 3 reports the results of structural equation models 16. For each of the dependent variables (childcare, tertiary education, childcare services vs child allowances), they show the total effect of class (if not controlled for mediators), the direct effect of class (controlling for mediators), and the indirect effect of class that runs through the mediators ‘opportunities’, ‘trust’, and ‘universalistic values’. For each of these mediators, the last column reports the share of the total class effect that runs through the mediator. In addition, for each dependent variable, the tables include a test of the Null hypothesis that the indirect effects are identical. We first discuss the results of the pooled sample and then address findings differentiated by occupational sub- class. The results for childcare priorities are shown in the upper part of Table 3. The total effect of class is 0.0078 and significant. We have seen this from model 1 in Table 2 already 17. However, when the mediators are introduced, the direct effect of class decreases and is no longer significant. Hence, the class effect is significantly and substantially mediated by opportunities. 43% of the class effect runs through opportunities. Additionally, a small part of the class effect runs through universalistic values (12%). The significant contrast between opportunities and trust and between opportunities and universalistic values show that the effect of opportunities is distinct from the effect of the other mediators. [Table 3 about here] The middle part of Table 3 shows the results for education priorities. The total effect of class is partially mediated by opportunities (13%) and universalistic values (11%). Beyond that, class still shows a significant direct effect of roughly 1.8 points. The difference between the two main mediators is insignificant. Hence, the difference between the middle class and working class is to some extent mediated in equal part by perceptions of opportunities and universal values. In contrast to the finding for childcare, opportunities do not very strongly mediate the 16 Given the sensitivity of different techniques of mediation analyses, and to increase the robustness of the main findings, we have also run a causal mediation analysis (A12-14) and a simple sequential test where we compare the class effect in a baseline model to the class effect in models including mediating variables (A15-17). The results are in line with the findings reported in the main text. Most importantly, the relative importance of the different mediators is robust across the approaches. 17 Note that the slight deviances in the coefficients are due to the smaller sample size in this part of the analysis: For comparability, all observations with missing values on opportunities, trust, or universalistic values have been excluded (mainly pensioners, since they were not asked the opportunity question). 19 class effect on education, as the coefficient for class remains significant. Also, universalistic values contribute to the explanation of the class effect to about the same degree. Finally, the trade-off between investing in childcare and child allowances is studied in the bottom part of Table 3. Note that the 4-point scale of the trade-off item here is treated as a continuous variable 18. Middle-class respondents are more likely to accept cutting back child benefits in order to expand childcare services. This is partially mediated by the respondents’ assessment of future opportunities (15%) and by the respondents’ level of trust (11%). The difference in these two indirect effects is indistinguishable from zero. Further, to test whether these differences between middle- and working-class respondents are driven by specific parts of the middle class, we decomposed the middle class into its three occupational classes and re-ran the analyses. Specifically, we tested whether the results are similar if we compare, in a single model each, the technical professionals, the managers, and the socio-cultural professionals to the working class. The respective tables can again be found in the Appendix (A18-A20). While the reduced group sizes in some models erase statistical significance, the direction of the effects remains the same in all models. Hence, the conflict over priorities between the working class and middle class is indeed a true class divide and does not seem to be driven by occupational specificities and work logics. Moreover, one may be concerned that the effect of opportunities simply reflects an education effect since the two concepts are correlated (Pearson r=0.24). Put differently, is it really perceived opportunity or rather education that drives SI priorities? In the Appendix (A21-22) we show that even for citizens with lower education levels, a more positive perception of opportunities increases attributed importance to SI. Summarising, the middle class clearly prioritises SI more strongly than the working class. In line with the literature, we find that both political trust as well as universalistic values positively correlate with SI priorities in most of our models. However, analysing SI priorities has clearly shown that respondents’ assessment of their economic opportunities is better suited to explain class differences. Put differently, middle-class citizens are more likely to positively evaluate 18 The lavaan.survey package in R used to compute the structural equation models cannot process ordered endogenous variables. 20 their future, and precisely this more positive evaluation increases the importance they attribute to policies that are oriented towards the future, namely childcare and tertiary education. CONCLUSION Class conflict over social policy is not dead in the 21st century. While much of the literature on democracy in current times has either written off mass politics or assigned it exclusively to the realm of socio-cultural conflict over societal values, we have shown in this paper that working- and middle-class citizens differ clearly and consistently in their social policy preferences. However, the class divide of our times is not so much articulated around the level of benefit provision favoured, or the resistance to cutbacks in existing welfare rights. In line with the key assertions of the ‘new politics of the welfare state’-literature, we find very high generalised levels of support for social policy expansion and equally high levels of opposition against retrenchment. However, working- and middle-class voters do indeed differ in their social policy preferences when it comes to priorities. Which are the problems, beneficiaries, and policies that welfare states should address primarily in times of constrained fiscal resources? When conceptualising the class conflict in such terms of priorities rather than position, we indeed find consistent and robust evidence for diverging preferences. Middle-class voters clearly and consistently attribute more importance to SI than working-class voters. This finding is robust to the field of SI (childcare and higher education), to the formulation of the question, and to including or excluding pensioners. Most importantly, it is robust across the eight highly different institutional contexts of the countries we have studied, which suggests that class differences in preferences relate more closely to structural factors, rather than being endogenously generated institutional feedback effects. As mechanisms reflecting these structural factors, we have studied universalistic values, political trust, and differences in the perception of economic opportunities. In the literature, political trust and universalistic values have so far been identified as key variables driving support for SI at the individual level. We also find at least partial evidence for these factors driving SI priorities. However, we find that a large part of the direct effect of class on SI priorities is mediated by the subjective evaluation of future opportunities of respondents in labour markets. The more positively respondents evaluate their chances of being in good employment, the more importance they attribute to social policies that aim at furthering welfare through sustaining the creation, preservation, and formation of human capital. 21 Why is this finding important? First, it shows that SI is not the welfare reform strategy that the disadvantaged classes prioritise. Against rationalist or technocratic assumptions, people who evaluate their own prospects in the labour market more negatively do not seem to prioritise human capital formation and activation policies to remedy their situation. Rather, citizens prefer the welfare state to support and sustain their individual situation as it is. Hence, a SI emphasis of welfare state reform seems rather unlikely to garner the enthusiasm of working- class voters. Second, class being mediated by the evaluation of economic opportunities is important, because it complements the existing explanations – focused on political trust and universalistic values – with a more economically based mechanism at the source of SI support. Indeed, the existing literature has mainly insisted on the collinearity of political preference patterns when it comes to SI on the one hand, and socio-cultural issues such as cultural liberalism, universalism, or attitudes on migration on the other hand (Beramendi et al., 2015; Garritzmann et al., 2018a). Our findings demonstrate that the determinants of SI support are both socio- cultural as well as economic in nature, both types of considerations dividing respondents in ways that align with other salient second dimension issues, such as cultural liberalism. Indeed, while being clearly an economic-distributive policy, SI very much resonates with the economic and socio-cultural second dimension politics that prevail in the mass politics of our times. What does that imply for class conflict and for electoral politics in the 21st century? It adds further evidence to a scenario of electoral trade-offs (Abou-Chadi & Wagner, 2020): just as with other second dimension issues, when political parties promote SI, this is likely to resonate with middle-class voters, but it may also be unable to reach working-class voters or may even antagonise them, and vice versa when it comes to social consumption policies. Hence, class politics do matter in the 21st century, but in quite different ways and on quite different matters than in the 20th century of welfare state development. These political class dynamics are likely to play out differently across countries, not only because political parties will deal with these trade-offs differently, but also because the relative size of the middle vs. working classes differs strongly across countries (Garritzmann et al., forthcoming 2021). In the knowledge economies of Northern Europe, for instance, medium- and high-skilled middle- class jobs constitute more than half of the workforce, while this share is roughly at a third in the Southern European economies. Hence, welfare politics are likely to diverge when it comes 22 to processes of mobilisation and coalition-formation. Nevertheless, underlying these differences, we observe strikingly similar configurations of welfare priorities across social classes. ACKNOWLEDGEMENTS Previous versions of this paper were presented at the SPSA Annual Congress in February 2019 in Zurich, at the Center for Civil Society Research at the Berlin Social Science Center WZB in May 2019, and at the CES International Conference of Europeanists in June 2019 in Madrid. We wish to thank the participants at these occasions for helpful comments. We are especially grateful for valuable input from Julian Garritzmann, Swen Hutter, Wolfgang Merkel, Line Rennwald, and the anonymous reviewers. REFERENCES Abou‐Chadi, T. & Immergut, E.M. (2019). Recalibrating social protection: Electoral competition and the new partisan politics of the welfare state. European Journal of Political Research 58(2): 697–719. Abou-Chadi, T. & Wagner, M. (2020). Electoral fortunes of social democratic parties: do second dimension positions matter? Journal of European Public Policy 27(2): 246– 272. Ares, M. (2017). A new working class? A cross-national and a longitudinal approach to classvoting in post-industrial societies (PhD thesis). European University Institute, Florence. Ares, M., Enggist, M., Häusermann, S. & Pinggera, M. (2020). Asking about welfare priorities: attitudinal constraint in social policy preferences. Manuscript. Baccaro, L. & Pontusson, J. (2016). Rethinking Comparative Political Economy: The Growth Model Perspective. Politics and Society 44(2): 175–207. Barnes, L. & Hicks, T. (2018). Making Austerity Popular: The Media and Mass Attitudes toward Fiscal Policy. American Journal of Political Science 62(2): 340–354. Beramendi, P., Häusermann, S., Kitschelt, H. & Kriesi, H. (2015). The Politics of Advanced Capitalism. New York: Cambridge University Press. Bonoli, G. (2005). The politics of the new social policies: providing coverage against new social risks in mature welfare states. Policy & politics 33(3): 431–449. Bonoli, G. (2013). The Origins of Active Social Policy. Oxford: Oxford University Press. 23 Bonoli, G. & Liechti, F. (2018). Good intentions and Matthew effects: access biases in participation in active labour market policies. Journal of European Public Policy 25(6): 894–911. Bremer, B. (2021). Public Preferences Towards Social Investment: Comparing Patterns of Support Across Three Continents. In J. L. Garritzmann, S. Häusermann & B. Palier (eds.), The World Politics of Social Investment (Volume I): Welfare States in the Knowledge Economy. Oxford: Oxford University Press. Bremer, B. & Bürgisser, R. (2020). Public Opinion on Welfare State recalibration in Times of Austerity: Evidence from Survey Experiments. SocArXiv. DOI: 10.31235/osf.io/uj6eq. Busemeyer, M.R. & Garritzmann, J.L. (2017). Public opinion on policy and budgetary trade- offs in European welfare states: evidence from a new comparative survey. Journal of European Public Policy 24(6): 871–889. Busemeyer, M.R., Garritzmann, J.L., Neimanns, E. & Nezi, R. (2018). Investing in education in Europe: Evidence from a new survey of public opinion. Journal of European Social Policy 28(1): 34–54. Enggist, M. & Pinggera, M. (2021). Radical right parties and their welfare state stances – not so blurry after all? West European Politics: 1-27. Esping-Andersen, Gosta. (1985). Power and distributional regimes. Politics & Society 14(2): 223–256. Esping-Andersen, Gösta. (1990). The Three Worlds of Welfare Capitalism. Cambridge: Polity Press. Esping-Andersen, Gøsta, Gallie, D., Hemerijck, A. & Myles, J. (2002). Why we need a new welfare state. Oxford University Press, USA. Evans, G. (1999). The End of Class Politics? Class Voting in Comparative Context. Oxford: Oxford University Press. Evans, G. & Tilley, J. (2017). The New Politics of Class. Oxford: Oxford University Press. Fernandez, J.J. & Jaime-Castillo, A.M. (2013). Positive or negative policy feedbacks? Explaining popular attitudes towards pragmatic pension policy reforms. European Sociological Review 29(4): 803–815. Ferragina, E. & Seeleib-Kaiser, M. (2015). Determinants of a silent (r)evolution: Understanding the expansion of family policy in rich OECD countries. Social Politics: International Studies in Gender, State & Society 22(1): 1–37. Ferrera, M. (1996). The ‘Southern model’ of welfare in social Europe. Journal of European social policy 6(1): 17–37. Gallego, A. & Marx, P. (2017). Multi-dimensional preferences for labour market reforms: a conjoint experiment. Journal of European Public Policy 24(7): 1027–1047. 24 Garritzmann, J.L., Busemeyer, M.R. & Neimanns, E. (2018a). Public demand for social investment: new supporting coalitions for welfare state reform in Western Europe? Journal of European Public Policy 25: 844–861. Garritzmann, J.L., Häusermann, S. & Palier, B. (2021). The World Politics of Social Investment (Volume I): Welfare States in the Knowledge Economy. Oxford: Oxford University Press. Garritzmann, J.L., Neimanns, E. & Busemeyer, M.R. (2018b). Trust, Public Opinion, and Welfare State Reform. Manuscript. Green-Pedersen, C. & Jensen, C. (2019). Electoral competition and the welfare state. West European Politics 42(4): 803–823. Hacker, J.S. & Pierson, P. (2010). Winner-Take All Politics: Public Policy, Political Organization, and the Precipitous Rise of Top Incomes in the United States. Politics & Society 38(2): 152–204. Hall, Peter A. (2020). The Electoral Politics of Growth Regimes. Perspectives on Politics 18(1): 185–199. Hansen, J.M. (1998). Individuals, Institutions, and Public Preferences over Public Finance. American Political Science Review 92(3): 513–531. Häusermann, S., Ares, M., Enggist, M. & Pinggera, M. (2020). Mass public attitudes on social policy priorities and reforms in Western Europe. WELFAREPRIORITIES dataset 2020. Welfarepriorities Working Paper Series n°1/20. Häusermann, S. & Kriesi, H. (2015). What do Voters Want? Dimensions and Configurations in Individual-Level Preferences and Party Choice. In P. Beramendi, S. Häusermann, H. Kitschelt & H. Kriesi (eds.), The Politics of Advanced Capitalism. New York: Cambridge University Press. Häusermann, S., Kurer, T. & Traber, D. (2019). The Politics of Trade-offs: Studying the Dynamics of Welfare State Reforms with Conjoint Experiments. Comparative Political Studies 52(7): 1059–1095. Häusermann, S. & Palier, B. (2017). The Politics of Social Investment: Policy Legacies and Class Coalitions. In A. Hemerijck (ed.), The Uses of Social Investment. Oxford: Oxford University Press. Hemerijck, A. (2013). Changing Welfare States. Oxford: Oxford University Press. Hopkin, J. & Blyth, M. (2018). The Global Economics of European Populism: Growth Regimes and Party System Change in Europe. Government and Opposition 1–33. Horn, A. & Van Kersbergen, K. (2021). Legacies of Universalism: Origins and Persistence of the Broad Political Support for Inclusive Social Investment in Scandinavia. In J. L. Garritzmann, S. Häusermann & B. Palier (eds.), The World Politics of Social Investment (Volume II): The Politics of Varying Social Investment Strategies. Oxford: Oxford University Press. 25 ISSP Research Group (2018): International Social Survey Programme: Role of Government V - ISSP 2016. GESIS Data Archive, Cologne. ZA6900 Data file Version 2.0.0, https://doi.org/10.4232/1.13052 Iversen, T. & Soskice, D. (2001). An Asset Theory of Social Policy Preferences. American Political Science Review 95(4): 875–893. Jacobs, A.M. & Matthews, J.S. (2017). Policy Attitudes in Institutional Context: Rules, Uncertainty, and the Mass Politics of Public Investment. American Journal of Political Science 61(1): 194–207. Jacques, O. (2020). The Politics of Public Finance in an Era of Permanent Austerity (PhD thesis). McGill University, Montréal. Jensen, C. (2012). Labor market- versus life course-related social policies. Understanding cross-program differences. Journal of European Public Policy 19(2): 275–291. Kitschelt, H. (1994). The transformation of European social democracy. Cambridge University Press. Knutsen, O. (2006). Class Voting in Western Europe. Lexington Books. Korpi, W. (1983). The democratic class struggle. Routledge. Korpi, W. & Palme, J. (1998). The paradox of redistribution and strategies of equality: Welfare state institutions, inequality, and poverty in the Western countries. American sociological review 661–687. Kriesi, H., Grande, E., Lachat, R., Dolezal, M., Bornschier, S. & Frey, T. (2008). West European politics in the age of globalization. Cambridge University Press. Morel, N., Palier, B. & Palme, J. (2012). Towards a Social Investment Welfare State? Bristol: The Policy Press. Neimanns, E., Busemeyer, M.R. & Garritzmann, J.L. (2018). How Popular Are Social Investment Policies Really? Evidence from a Survey Experiment in Eight Western European Countries. European Sociological Review 34(3): 238–253. Oesch, D. (2006). Redrawing the class map: Stratification and institutions in Britain, Germany, Sweden and Switzerland. Springer. Oesch, D. & Rennwald, L. (2018). Electoral Competition in Europe‘s new tripolar political space: Class voting for the left, centre-right and radical right. European Journal of Political Research 57: 783–807. Pavolini, E. & Van Lancker, W. (2018). The Matthew effect in childcare use: a matter of policies or preferences? Journal of European Public Policy 25(6): 878–893. Pierson, P. (2001). The New Politics of the Welfare State. New York: Oxford University Press. 26 Pinggera, M. (2020). Congruent with whom? Parties’ issue emphases and voter preferences in welfare politics. Journal of European Public Policy: 1-20. Rehm, P. (2016). Risk Inequality and Welfare States. New York: Cambridge University Press. Rovny, J. & Polk, J. (2019). New wine in old bottles: Explaining the dimensional structure of European party systems. Party Politics 25(1): 12–24. Svallfors, S. (1997). Worlds of welfare and attitudes to redistribution: A comparison of eight western nations. European Sociological Review 13(3): 283–304. Traber, D., Giger, N. & Häusermann, S. (2018). How economic crises affect political representation: declining party-voter congruence in times of constrained government. West European Politics 41(5): 1100–1124. van Oorschot, W. (2006). Making the difference in social Europe: deservingness perceptions among citizens of European welfare states. Journal of European Social Policy 16(1): 23–42. 27 Figure Legend Figure 1 - caption: Predicted childcare, education, and pension positions and priorities for working- and middle-class voters - notes: Based on regression models printed in Table 1 and 2, control variables kept at their means. Vertical lines indicate 95% confidence intervals. Table Legend Table 1 - caption: Class as a determinant of social policy position preferences regarding social investment and consumption - notes: ***p < 0.001; **p < 0.01; *p < 0.05. Country-fixed-effects included in all models. Table 2 - caption: Class as a determinant of social policy priority preferences regarding social investment and consumption - notes: ***p < 0.001; **p < 0.01; *p < 0.05. Country-fixed-effects included in all models. Table 3 - caption: Direct and indirect effects of class priorities through opportunities, trust, and universalistic values, based on structural equation models - notes: ***p < 0.001; **p < 0.01; *p < 0.05. 28