Look where you’re going: the cultural and economic explanations of class voting decline

Abstract Scholars have shown a consistent decline in class voting over time, arguing that social class no longer structures political competition. Yet, the loss of the explanatory power of social class should be attributed to the shift of left-wing parties towards less friendly issue positions for their class-based electorate and not entirely to demand-side explanations. This article focuses on the explanatory factors underpinning the decline of class voting, in particular economic and cultural ones. It does so by relying on a longitudinal dataset including survey data from the EES and party manifesto data from the CMP over the last 25 years in 12 European countries. Results show that when class-based parties adopt pro-labour positions, class voting is reinforced. By contrast, when class-bloc parties stand out for their pro-multiculturalism positions, class voting weakens significantly. Supplemental data for this article can be accessed online at: https://doi.org/10.1080/01402382.2021.1987108 .

1986). Accordingly, the persistence of 'old' cleavages (or the lack thereof) (Bartolini and Mair 1990;Emanuele 2021;Franklin 1992) was inextricably linked with the vanishing of social classes (Clark et al. 1993;Clark and Lipset 1991). Evidence in this regard made it clear that a) the predictive power of cleavages (including the class cleavage) in determining voting behaviour was declining (Franklin 1992) and b) class voting in general was declining (Nieuwbeerta and De Graaf 1999). What is still a matter of debate are the reasons behind this drop.
The literature on class dealignment puts the economy at centre stage, arguing that the decline in class voting is mostly due to the convergence towards the centre of class-based parties on traditional economic issues, such as redistribution, which in turn produced the alienation of the working-class electorate. However, other dimensions of political contestation (i.e. cultural issues) are less frequently investigated (for a notable exception see Abou-Chadi and Wagner 2020), even though they appear to have an impact on the dynamics of class voting (van der Waal et al. 2007). Our aim is to close this gap further by re-evaluating the relevance of economic issues and exploring the importance of the cultural dimension in explaining the class voting decline.
We do so by focussing on the effects on class voting of (class-based) party positions on specific economic and, most importantly, cultural issues. In line with other studies, we find that the link between the working class and the class bloc is robust when the class bloc adopts pro-labour positions. On the other hand, when the class bloc stands out compared to other parties on multicultural issues the link weakens significantly.

The decline of class voting
The electoral allegiance between the working class and leftist parties remained undisputed for more than a century (Bartolini and Mair 1990). In his precursory work, Alford (1962) found a strong link between the working class and leftist parties. However, starting from the 70's, the strength of this link has been challenged (Franklin 1992). Class voting as well as the association between left-wing parties and the working class has been generally deemed to be in decline in Western societies (De Graaf et al. 1995;Nieuwbeerta 1995;Nieuwbeerta and De Graaf 1999). According to this scholarship, this process occurred as a consequence of socio-structural changes, for example increased levels of affluence and education, and reduction in size of the working class. These processes generated new social alliances based on social interests, which cross-cut traditional class hierarchies (Clark et al. 1993;Dalton 1996;Gingrich and Häusermann 2015;Inglehart 1997).
Yet, another strand of the literature argues that, rather than declining, class voting is 'realigning', and that dealignment is, at best, overrated (Hout et al. 1993). In the words of Kitschelt andRehm (2015: 1780), realignment means that 'voters continue to coalesce around parties on the basis of durable socioeconomic interests and policy preferences, but since political-economic post-industrialisation […] has changed the distribution of preferences, established parties have been compelled to alter their appeals […]' .
Critics of the dealignment theory (Evans 2000;Manza et al. 1995) postulate that the 'traditional approach' to class voting (i.e. à la Alford), which envisions a simple two-class (working class vs middle class) and two-party system (leftist vs rightist parties) oversimplifies reality. Furthermore, the traditional approach to class voting overlooks total class voting, since the only possible association is the one between the working class and leftist parties and the middle classes and right-wing parties (Hout et al. 1993). This implies that each deviation from traditional class voting is interpreted as a sign of class dealignment. Instead, deviations from traditional class voting might be interpreted as forms of realignment (Pisati 2010).
To overcome these shortcomings, new categorizations of social class have been proposed following the changing pattern of industrialisation (Erikson and Goldthorpe 1992;Oesch 2008). Moreover, more articulated models of voting behaviour and alternative estimation techniques have been advanced (see e.g. Evans 1999) to gauge the overall ability of social class to structure political preferences.
Nonetheless, recent contributions seem to converge on the decline thesis (see especially Jansen et al. 2013): the traditional allegiance between the working class and leftist parties has been dropping over time (Gingrich and Häusermann 2015). However, while class voting is declining, the explanations for this decline are a matter for debate.

Why is class voting in decline?
We provide four main explanations for the decline (see Pisati 2010). The first looks at the dilution of class inequalities over time, as a consequence of increased living standards and affluence. According to this argument, the homogenisation of socio-economic conditions across different classes translates into political choices as well (Pakulski 2001;Pakulski and Waters 1996). The second explanation concerns the role of increased levels of education (and mobilization) of citizens due to cognitive mechanisms independent from class identity (Dalton 1984;Franklin 1984b). From these changes stemmed the post-materialist 'silent revolution' that replaced class-based political choices (Inglehart 1997). A third argument points to the growing importance of cross-cutting issues, such as race, gender and identity (Huckfeldt and Kohfeld 1989;Norris and Inglehart 2004), which cannot be simply subsumed under the traditional economic left vs. right dimension, as new cultural allegiances (demarcation vs. integration) have now become as relevant as the left vs. right divide (Hooghe and Marks 2018;Kriesi et al. 2012). Finally, a fourth explanation focuses on the role of political parties (supply side) as a source of dealignment, rather than on structural or individual level factors Tilley 2012, 2017;Jansen et al. 2013;Rennwald and Evans 2014).
This last line of argumentation reconnects the dynamics of class voting to the dimension of party strategies: with the growth of the middle class, the leftist parties' conundrum was to either adopt an electoral platform to appeal to the traditional working class (aware that the working class was declining) or shift towards a catch-all platform (appealing to the society at large to expand the electoral basis). To accommodate the latter, mainstream parties focussed on cultural issues, whose salience grew dramatically in the last decades (Kriesi et al. 2012).
Party strategies are thus crucial in identifying why class voting has changed. In this respect, Evans and colleagues found that the decline in class voting occurs when class-based parties converge towards the centre on economic issues (Evans and Tilley 2012;Rennwald and Evans 2014). Jansen et al. (2013) reached a similar conclusion: they found that left-right party positions alone do not have any effect on class voting; yet high levels of polarisation along the left-right dimension reinforce the association between class and the vote. In other words, class voting declines when class-bloc party positions on the economy become indistinguishable from those adopted by parties outside the bloc.
Whereas the role of party positions on the economy has been extensively analysed, the effect of class-based party positions on the new cultural dimension has been widely overlooked, as the link between classes and parties was mainly framed in economic terms (the working class supports leftist parties because of its economic interests for redistributive policies). One notable exception is a study by Abou-Chadi and Wagner (2020) which found that a party's position shifts on cultural issues do not influence the likelihood of the working class voting for a social democratic party.
However, the focus of the authors is only on the effect of parties' absolute positions on the cultural dimension: thus the relevance of party differentiation on cultural issues is overlooked. In this study we try to overcome this limitation a) by investigating the effect of class-based parties' positions on economic and cultural issues on class voting and b) by evaluating the effect of their positions on both dimensions, compared to all the other parties of the system.

Framework of analysis
Against this backdrop, we propose four different hypotheses (see below). In so doing, we stick to traditional class voting analysis (Knutsen 2007). Although we acknowledge the relevance of recent studies on total class voting, 1 our focus is specifically on the supply-side conditions which allow the traditional allegiance between the working class and leftist parties to either resist or decline, i.e. a top-down approach Tilley 2012, 2017;Rennwald and Evans 2014).
Supply-side changes are believed to occur due to the convergence towards the centre of mainstream parties' policy positions. In particular, the literature noted a shift of mainstream class-based parties, e.g. social-democracy, towards (neo)liberal positions (Pierson 2001).
During the Great Recession for example, social democracy, while becoming sceptical towards economic liberalism, accepted the need for fiscal consolidation and budgetary cuts (Bremer 2018). This ideological drift also occurred among many radical left parties, the other party family of the class bloc, which increasingly became a pro-systemic (yet anti-neoliberal) party family (March and Keith 2016). Thus, we hypothesise that traditional class voting declines if left-wing parties shift to a more moderate position on economy, as working-class voters are supposed to 'rationally' defect from parties that no longer deliver programmatically. This hypothesis was already proposed by Jansen et al. (2013: 72), who 'did not find evidence for the idea that left-right positions of left-wing parties alone influence the association between class and vote' . HP1 The association between class and voting decreases when class-based parties move toward the centre in the economic field. However, the shift towards market-oriented positions from the left cannot be equated to a systematic convergence of the left and the right poles, since, theoretically, (conservative and liberal) parties outside the bloc can adapt their positions as well, shifting towards more pro-orthodox positions. As Evans and De Graaf highlighted (2013: 7), '[p]ositional convergence weakens the strength of the signals from parties to voters' . Therefore, 'voter responses to party polarisation […] depend upon the choices voters are offered (the supply side), as well as the presence of differences in ideological and value preferences within the electorate' .
Theoretically, there may be two trends that parties outside the bloc may undertake: although counterintuitive, mainstream liberal and conservative parties may have converged towards less pro-market positions to accommodate the pressure for maintaining social benefits standards, in the same way as social democratic parties became more sceptical towards economic liberalism (Bremer 2018). If this is the case, then the shift of class-based parties may be less pronounced. If this is not the case, that is if mainstream right parties maintain or further emphasise their classical pro-market position, then political systems may still diverge in the economic dimension despite the shift of social democratic parties to the centre. Besides mainstream right parties, the other relevant party family outside the bloc, i.e. the radical right family, can shift its position in the same way. The radical-right 'winning formula' (Kitschelt and McGann 1995), which comprised a mix of authoritarian and neoliberal proposals in the 90's, has changed due to the blurring position of radical right parties on the economic side (Rovny 2013). Yet, this does not mean that the parties of this family have no position whatsoever on the economy. Afonso and Rennwald (2018) showed that welfare expansion has become more salient among radical right parties, most notably because their electorate changed its composition through a process of 'proletarization' (Afonso and Rennwald 2018;Arzheimer 2013;Oesch 2008). Our second hypothesis, thus, focuses on the economic side, but it differs from the previous one as it implies that class voting resists only when a significant polarisation occurs in a political system. No matter the direction of the polarisation 2 , once it occurs working-class voters should find it easier to detect, from a purely rational perspective, which parties stand for their preferred economic platform.
HP2 The association between class and voting decreases when class-based parties programmatically converge with parties outside the class bloc in the economic field.
A supply-side explanation of class voting would be one-sided, if it only focuses on party shifts on economic issues (van der Waal et al. 2007). Class-bloc parties have progressively adopted catch-all strategies, including social-liberal cultural issues in their programmatic platforms. These changes reflect the new cultural cleavage, orthogonal to the left-right divide (Hutter and Kriesi 2020;Kriesi et al. 2012). When adopting social liberal positions on cultural issues, class-based parties overcome traditional class boundaries and become much more middle-class oriented (Gingrich and Häusermann 2015). Indeed, when it comes to liberalism in non-economic issues, the middle class has been traditionally identified as more progressive than the working class (Lipset 1959;Inglehart 1977).
Consequently, the link should remain stable as long as class-based parties are responsive to working class economic interests. However, this link might weaken when political systems incorporate new non-economic cleavages and class-based parties shift their programmatic platform towards more culturally progressive positions, i.e. those positions which are less appealing for the traditional working class and more in line with middle class orientations. We then hypothesise that the adoption of culturally progressive positions weakens class voting. HP3 The association between class and voting decreases when parties of the class bloc reinforce their social liberal positions on cultural issues.
The shift on cultural issues could be inconsequential if class-bloc positions on cultural issues are indistinguishable from the positions of parties outside the bloc. This line of reasoning mirrors the one proposed above. Thus, we anticipate that the effects of cultural progressivism should emerge especially when it becomes a distinctive feature of class-bloc parties compared to all the others. Following HP2, we propose a further hypothesis based on the polarisation of the political system on cultural issues.
HP4 The association between class and voting decreases when parties of the class bloc diverge from and adopt social liberal positions on cultural issues relatively more than parties outside the bloc.

The data
We test our hypotheses by employing a time-series, cross sectional dataset of the European Election Study-Voter Study (EES), spanning the period from 1989 to 2014. The dataset relies on the pooled dataset (1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004) collected by Marsh and Mikhaylov, 3 to which we add two further waves collected respectively in 2009 and 2014. We excluded the 1999 wave, because of the lack of comparable measures for social class. To ensure data comparability, we restricted our analyses to those countries surveyed in all waves. The final dataset encompasses five time-points (1989,1994,2004,2009,2014), in 12 European countries (Belgium, Denmark, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain and the UK), including a total of 125,215 individual observations. The data has been collected using consistent sampling procedures over time, thus improving the comparability of the data at different time-points. Also, the dataset covers a fairly long period of time. Finally, all waves include a survey question for the vote recall of respondents in the last national general elections (our dependent variable).
We added data concerning the supply side using the Comparative Manifesto Project (Volkens et al. 2019). The methodology adopted by the CMP, widely validated by previous literature (Budge and Pennings 2007;Pennings 2011;Volkens et al. 2009), uses content analysis of party manifestos to measure the proportion of (quasi) sentences that a party dedicates to a given topic in any single election covered by the CMP. The dataset includes a distinction between positive and negative mentions on a set of coded issues, allowing party positions on different issues to be extrapolated.
In order to capture the effect of party positions on class voting, we matched each EES round with party manifestos relating to the closest previous general election. For example, for the EES 2014 in Germany, we considered parties' electoral manifestos for the general elections of 2013, which was the closest election held before the data collection of the 2014 EES in Germany (and for which the EES has a question related to the 2013 election). In this way we impose a lag on the measurement of party positions, based on the very simple idea that party position shifts need time to be able to affect individuals' behaviour.

The dependent and independent variables
Our dependent variable is the vote choice for the class bloc in the last national general election (i.e. vote recall: 'Which party did you vote for in these last parliamentary elections?'). For all countries we select the legislative elections, thus we do not take into consideration either European or presidential elections. We opt not to take into account European elections, for the second-order effect (Reiff and Schmitt 1980) that might significantly affect vote choice.
By 'class bloc' we mean the leftist parties which traditionally appeal to the working class. These are the parties belonging to the communist and social-democratic party families. 4 The inclusion of parties in the class bloc has been preliminarily based on the classification provided by the CMP, which is the same one used by the EES to code party families. A further screening has then been performed based on the information retrieved from the ParlGov dataset. Overall, we coded 85 parties belonging to the class bloc over the entire 1989-2014 period. 5 The identification procedure has been carried out ex-post: once political parties have been identified as part of the class bloc, they are believed to remain within the bloc. Therefore, we do not take into consideration when a specific party entered into political competition in each country or the changes in the composition of the bloc per se, since we intend to analyse the link between the working class and the class bloc as a whole.
Once we identified the class-bloc parties, we constructed our dependent variable as a dichotomy with 1 = voted for a class-bloc party and 0 = all other voters. 6 Respondents who either did not declare any party preference or abstained have been excluded from the analysis and thus coded as missing values. The overall number of valid observations is 94,523.
Our key independent variable is the social class of respondents. Although there is much debate on the operational definition of social class (see e.g. Evans and Tilley 2017: 2), there is a consolidated consensus on privileging an objective measure of social class. We thus measure social class by means of occupational class. The latter is operationalised as a dichotomy, distinguishing manual workers (both skilled and unskilled manual workers, as well as farm labourers) (=1) and non-manual workers (=0). 7 It is worth noticing that other studies have worked on more fine-grained operationalizations of occupational class, which proved to be very useful from both a theoretical and empirical perspective, e.g. the Erikson-Goldthorpe-Portocarero class scheme and more recently the class scheme elaborated by Oesch (2008). However, our dataset does not allow us to take such an approach as it lacks the relevant information for a more refined operationalisation of social class. We acknowledge this limitation in our data, but we are also confident that our measure can capture the dynamics of class voting. First, as shown by Nieuwbeerta and De Graaf (1999), different measures of class voting yield substantially similar results. In particular, the Erikson-Goldthorpe measures of class voting produce the 'same picture as that observed when using a manual/ nonmanual class distinction' (Nieuwbeerta and De Graaf 1999: 35). Secondly, Jansen et al. (2013) reveal a pattern of convergence of class voting for the routine non-manual class, the service class and the self-employed when compared with the manual working class in the period in between the 1960s-1970s and the 2000s. This means that, in general, for non-manual social classes the trend in class voting has followed a similar pattern. Along the same lines, skilled and unskilled workers do not seem to differ to a great extent (Nieuwbeerta and De Graaf 1999). Finally, we are more interested in the specificities of the manual working class (i.e. traditional class voting, Knutsen 2007) than a more nuanced reconstruction of the different voting paths of different social classes.
As control variables, we include gender (1 = Male), age (discrete) and education. As for education, the EES includes a standard variable which indicates the year when the individual stopped full-time education. To avoid inconsistencies across time and ensure the comparability of the data, in each EES round we have recoded the variable as follows: 1 = Low education (including respondents not having formal education or who stopped their full-time education when they were 15 years old or less); 2 = Intermediate (including respondents who stopped full-time education when they were aged between 16 and 19 years); 3) High (including respondents who stopped full-time education when they were aged 20 or more); 4) Respondents who are still studying.
As for the operationalisation of variables related to party position (economic and cultural), we derived them from the CMP dataset. To measure party positions on traditional economic class-issues we use the CMP items measuring party mentions of labour groups (per701 = favourable references to labour groups; per702 = negative references to labour groups). We use the labour item, as it has been widely shown that it is an issue owned by centre-left and labour parties, i.e. traditional parties of the class bloc (Otjes and Green-Pedersen 2021;Seeberg 2017).
For party positions on cultural issues, we use multiculturalism as a proxy. This measure is obtained from the items reporting positive and negative references to cultural plurality and diversity within domestic societies (per607 = favourable mentions of multiculturalism; per608 =negative mentions of multiculturalism).
Following previous studies (Emanuele et al. 2019), for each of the two items we calculated party positions as the difference between the positive and the negative pole for each election. This value has then been weighted for the overall salience of the issue for each single party, obtained as the sum of the proportion of positive and negative mentions on each item: This yields two variables with positive values measuring supportive positions on left-wing economic issues and on multiculturalism, and negative values measuring right-wing positions on both issues. We also weighted each party score for its electoral strength (measured as the percentage of votes gained by the party) and we then averaged the party scores by class bloc belonging on each of the two dimensions. Thus, we obtain a summary measure of pro/anti-labour and pro/ anti-multiculturalism positions for both class-based parties and parties outside the class bloc.
Besides the absolute measure of class-bloc positions, we also analyse the effect of the class-bloc position on economic and cultural dimensions relative to the position of all the other parties. As hypothesised above, if class-based parties hold pro-labour positions, this might not be sufficient to attract voters if other parties hold the same position. A positive effect of labour issues on class voting might occur only when this position is a distinctive feature of class-bloc parties. To measure the position of class-bloc parties on labour and multiculturalism relative to the other parties of the system, for each issue we simply subtract the mean position of all parties outside the bloc from the class-bloc average score. This yields two variables measuring the support of class-bloc parties for labour and multiculturalism relative to parties outside the bloc. Therefore, higher values of the two variables indicate that class-bloc parties are relatively more supportive of labour and multiculturalism compared to parties outside the bloc.

The modelling strategy
In our pooled dataset, respondents are cross classified within countries and years, i.e. respondents are independently nested both within countries and years. Variables at the party level instead vary across time and country, i.e. country*year. Given the nature of the data, a traditional logistic model is not appropriate to account for the hierarchal structure of the dataset. We opted for a modelling strategy in which we control that 1) observations clustered in one country are more similar compared to observations clustered in another country; 2) observations clustered in one year are more similar compared to observations clustered in another year (even when clustered in the same country); 3) observations clustered in one country-year combination are more similar compared to observations clustered in another country-year combination.
Thus, following Schmidt-Catran and Fairbrother (2016: 25), we employ a cross-classified multilevel logistic model which further includes, at a higher level, a grouping variable which distinguishes all possible country*year combinations. In particular, we include random effects for (12) countries, (5) years and (60) country*year combinations. Thus, the models treat country-year combinations as cross-classified within countries and years, and individuals as simply nested in country-year combinations. A maximum likelihood estimation technique is used to estimate different models' specifications.

Class voting decline
Before testing our hypotheses, we first explore the effect of belonging to the working class on voting for a class-bloc party in national elections. Second, Note: *p < 0.05, **p < 0.01, ***p < 0.001. we assess this relationship over time. 8 These models enable us to provide the reader with a broad overview of trends in class voting and to compare our data with the most recent literature on class voting. In this respect, these results also provide some information about the quality of our data.
In Table 1 we present three logistic regressions with random country intercepts to account for unexplained variance at the country level: in Model 1.1 we assess the effect of social class on voting for a class-bloc party, including as control variable only time fixed effects; in Model 1.2 we let time interact with social class (social class*time) to explore the effect of social class over time; in Model 1.3 we assess the robustness of Model 1.2 including standard sociodemographic controls such as gender, age and education. 9 In Model 1.1 we find confirmation for the endurance of the association between the working class and class-bloc parties. The coefficient of social class is positive and highly significant, thus confirming that the working class is still more likely than other social classes to vote for a left-wing party.
In Model 1.2 we test the robustness of class voting over time and find evidence of a declining trend in class voting. The coefficient of the interaction term is indeed negative and highly significant. A visual exploration of the effect of class over time is provided in Figure 1, where we report the average marginal effect of class in different years. The decline is confirmed to be consistent and statistically significant. Moreover, we find a stable and steady decline in class voting, rather than a dynamic of dealignment during the first decade of the XXI century, without any sign of realignment. Thus, while working-class voters significantly prefer class-based parties compared to other parties, this association has declined consistently over time. This result is also robust to the inclusion of sociodemographic controls, as shown in Model 1.3.
In line with previous literature, the analysis thus supports the overall endurance of the association between the working class and the class bloc. However, this association is declining over time. These findings are in line with the most recent studies on class voting.

Position of the bloc on the economy
The second part of the analysis focuses on the parties' supply side, under the broad hypothesis that parties themselves might be responsible for the demise of class voting, to the extent that they abandon their traditional programmatic platforms to adopt catch-all strategies. To test our hypotheses we estimate four models, each including our key independent variable (social class) and standard sociodemographic controls. As we now also include in our analysis variables that are measured at the country*year level (i.e. class-bloc positions on economic and cultural issues), we employ a cross-classified multilevel logistic model (Table 2). 10 In presenting our results, we first discuss whether parties' positions on traditional economic issues have a role in displacing class voting.    Note: *p < 0.05, **p < 0.01, ***p < 0.001. all models are based on 90528 individual observations and data from 12 countries, 5 years, and 60 country*year combinations. Models estimated with stata's xtmelogit command. please note that, to avoid multicollinearity issues, in the models reported in table 2 we include the labour and multiculturalism (absolute and relative) party positions separately. However, in online appendix B we present additional models including all the interaction terms in the same multilevel logistic model. in particular, in Models 1 and 2 of table B1 we report the results of two logistic regressions which include the interactions between class and class-bloc absolute position on economy and multiculturalism together. in models 3 and 4 of table B1 we report the results of two logistic regressions including the interactions between class and class-bloc relative positions on economy and multiculturalism together. results are in line with the empirical findings presented in the article. Finally, in Models 5 and 6 of table B1 we estimate the effects of absolute and relative class-bloc positions on labour (Model 5) and multiculturalism (Model 6) simultaneously. overall, we find that while the moderating effect of absolute class-bloc positions on labour does not reach statistical significance, the moderating effect of relative class-bloc positions on both labour and multiculturalism remains significant and in line with the results presented in the article.
In Model 2.1, aside from social class and control variables, we include a variable measuring the pro-labour position of class-bloc parties in each country*year and we let this variable interact with social class to test whether party positions on labour issues do indeed affect the link between the working class and class-bloc parties. The model provides insightful results. The coefficient of the interaction term is positive and statistically significant, a clue that when class-bloc parties adopt pro-labour positions, class voting is reinforced. The result is even more straightforward if we look at the average marginal effect of social class at the different levels of class-bloc parties' support for labour issues (Figure 2).
In order to empirically test the argument of HP2 on the position of the class bloc relative to parties outside the bloc, we plugged in the relative pro-labour position variable and let this variable interact with social class (Model 2.2). The interaction term between the relative pro-labour position of the class bloc and social class is positive and statistically significant, thus lending support to our second hypothesis. This result is further confirmed if we look at the average marginal effect of social class on voting for a left-wing party at different levels of relative support for pro-labour issues. Indeed, class voting significantly increases as the pro-labour position of class-bloc parties increases compared to that of other parties (Figure 3).
Overall, our results confirm both HP1 and HP2: class voting is not only reinforced when class-bloc parties adopt pro-labour positions, but also when they are relatively more pro-labour compared to parties outside the bloc.

Position of the bloc on cultural issues
In order to test HP3 and HP4, we replicate the previous analysis applied to cultural issues. In Model 2.3 and 2.4, we test whether the position of the class bloc on multiculturalism affects class voting and whether the interaction between class and multiculturalism depresses class voting. Overall -  and in contrast with the results for the economy -the position of the class bloc on multiculturalism alone does not affect significantly class voting (Model 2.3, Figure 4). However, the picture is quite different when we take into account the relative position of class-bloc parties (compared to parties outside the bloc) on multiculturalism. We include the relative position of class-bloc parties on multiculturalism and, again, we interact this variable with social class (Model 2.4). Our results show that class voting is significantly depressed when class-bloc parties adopt more favourable positions on multiculturalism compared to other parties. Figure 5 in this regard shows that multiculturalism downsizes class voting. In a nutshell, when class-based parties adopt a distinctive position on multiculturalism (they are relatively more pro-multiculturalist than parties outside the bloc) class voting decreases. This result is extremely interesting, as it shows that if we want to capture how the new cultural dimension is affecting class voting, it is not sufficient to account for the simple party position on this dimension, but we should rather focus on the distinctiveness of its position, i.e. the extent to which it stands out in the party system for its multiculturalism. Therefore, we find support for HP4 (relative position of class-bloc parties), but not for HP3 (absolute position of class-bloc parties).
To sum up, we have shown that the relative position of parties on both economic and cultural issues has a significant impact on the dynamics of class voting. Yet, the absolute position on the two issues has a different impact: when class-bloc parties hold favourable positions on labour issues, independently from the positions of other parties, class voting is reinforced. This is not the case for the absolute position on cultural issues.

Robustness tests
As robustness tests, rather than focussing on single CMP items (labour and multiculturalism) to measure class-bloc positions, we aggregate economic and cultural items together to generate two indexes of class-bloc  economic and socio-cultural positions. We decided here not to opt for the classical RILE scale because it collapses economic and cultural items together in one indicator (see Online Appendix B for a detailed discussion).
The two dimensions have been operationalised as the weighted (by salience and electoral strength) difference between (economic/ socio-cultural) left-wing and right-wing items:

Party position
Left items Right items The specific items for each dimension are available in Table B2 of Online Appendix B.
We replicate the analyses presented in previous sections by simply plugging the two indexes into our interaction models. The results are robust, yet with a caveat: unlike the labour item, the absolute class-bloc position on the new economic index does not produce any effect on class voting ( Figure 6).
As for the absolute position on the cultural dimension, the results prove to be consistent with our main findings (Figure 7): the absolute position does not affect class voting.
The effects of the relative class-bloc positions on both the economic and cultural dimensions, instead, are fully in line with our previous analyses. Class voting is reinforced when class-bloc parties adopt relatively more leftist positions on the economy compared to other parties (Figure 8). On the contrary, social-liberal positions, relative to other parties' positions, weaken class voting (Figure 9).
Robustness tests are consistent with HP2 and HP4, while yielding different results in relation to HP1, which nonetheless does not affect our main research question.
Further robustness tests include the replication of our models, but this time including party positions not weighted for issue salience. Results,  available in Table B4 of Online Appendix B, are strongly supportive of our findings.

Conclusion
In this article we have investigated the supply-side conditions under which class voting declines. The literature so far has focussed mainly on the economic aspect, neglecting the new cultural cleavage. As the new cleavage is gaining prominence in explaining changes in political systems, we test whether the shifting of class-bloc parties' positions towards more progressive stances depresses (or reinforces) class voting. Focussing on 12 European countries and covering a period of 25 years, we have first shown that class voting, albeit present and significant, is declining over time. In our supply-side analysis, we show that the allegiance between the working class and class-bloc parties is strong and robust when class-bloc parties adopt pro-labour positions, both in absolute and in relative terms. On the contrary, class voting is weakened when the pro-multicultural stances of class-bloc parties become predominant compared to other parties. Yet, in this case the absolute position of the class bloc does not impact class allegiance. These two findings show that whereas class-bloc positions on economic issues matter a priori, cultural issues play a (negative) role only when and where class-based parties are distinguishable from parties outside the bloc. Thus, class-bloc parties are able to mobilise working-class voters by adopting favourable positions on labour issues. Yet, they lose the working-class vote when progressive positions on multiculturalism mark a distinction with other parties outside the bloc. As tautological as it might sound, working-class voters reward class-bloc parties when it comes to the 'old' economic cleavage and they are also less attracted by class-bloc parties which are overly attentive to typical middle-class issues, i.e. cultural ones.
We believe these findings have important implications. As the literature acknowledges, class voting decline can be imputed to structural changes in Western European countries, which accordingly caused the restructuring of party systems along new cleavages. Yet, the appearance of new cleavages does not automatically translate into a change in political positions for all social classes, nor has the importance attributed to materialist issues by working-class voters completely vanished in the post-Cold War environment. To the contrary, working-class voters seem to be able to recognise who 'stands for' their materialist interests and reward them electorally. Even when new issues increasingly shape voters' preferences, the economic ones, i.e. issues related to the distribution of resources and power across classes, are still key to disentangling where the working class, though reduced in number, is going ideologically. Further research is still needed in relation to this topic. First, we acknowledge that while our definition of working class provides a broad conceptualisation, other works (Oesch 2008;Oesch and Renwald 2018) have disentangled the Manichaean distinction between members of the class bloc and workers outside the bloc, showing the intra-class nuances in the vote choice. Secondly, our article has not confronted the saliency of materialist and cultural issues in the party systems as a whole nor has it analysed voting behaviour through the lens of a more fine-grained distinction between social classes, from which it is possible to derive whether all classes behave the same when it comes to economy or, rather, middle-class workers are more inclined to privilege cultural issues.

Notes
1. These studies relax the two-party and two-class system assumption to capture total class voting. In so doing, they retrieve the patterns of vote choice of the different social classes within each system (see for example Jansen et al. 2013). 2. Theoretically, a polarization to the left means that the class bloc significantly moves to the left while parties outside the bloc do not change their positions or move to the right, while a polarization to the right implies that parties outside the bloc move to the right while parties of the bloc do not change their positions or move to the left. 3. The dataset is available at: https://www.tcd.ie/Political_Science/people/mi-chael_marsh/ees_trend_file.php 4. We adopt here a conventional classification which excludes green parties from the class bloc (e.g. Emanuele 2021; Jansen et al. 2013;Rennwald and Evans 2014). 5. The full list of parties is provided in Online Appendix A. 6. We have also used a more restrictive solution, which includes among parties outside the class bloc only right-wing parties (liberal, conservative and radical right parties). This solution implied the exclusion from our analysis of regionalist and green parties. Empirical results (available upon request) are in line with the findings presented in the manuscript. 7. We constructed the dichotomy relying on two variables which record the current and past employment of respondents, and which consistently report information about skilled and unskilled manual workers (i.e. the necessary information to separate the working class from other classes). To avoid loss of cases, information about past employment has been used to assign retired and currently unemployed respondents to either the manual or non-manual categories. Non-responses have been excluded from the analysis. 8. Descriptive statistics are available in Online Appendix A. 9. In addition, as a robustness test, we replicated the analysis presented in Table 1 using a cross-classified logistic model (see Table A1, Online Appendix A). Results are consistent with the ones presented in the manuscript. 10. Fully specified models reported in Table 2 are available in Table A1, Online Appendix A.