Attention to the influence of social groups on citizens’ political behaviour has rebounded in recent years.Footnote 1 In particular, recent research has shown a strong relationship between voters’ social identities and their political behaviour (for example, Achen and Bartels Reference Achen and Bartels2016; Bornschier et al. Reference Bornschier, Häusermann, Zollinger and Colombo2021, Reference Bornschier, Haffert, Häusermann, Steenbergen and Zollinger2024; Fraga et al. Reference Fraga, Velez and West2024; Hopkins et al. Reference Hopkins, Lelkes and Wolken2025; Huber et al. Reference Huber, Meyer and Wagner2024; Kinder and Kam Reference Kinder and Kam2009; Zárate et al. Reference Zarate, Quezada-Llanes and Armenta2024; Zollinger Reference Zollinger2024). Likewise, voters’ expressive perceptions of the social group composition of parties’ electoral alliances have been found to influence political behaviour (for example, Ahler and Sood Reference Ahler and Sood2018; Kane et al. Reference Kane, Mason and Wronski2021; Mason Reference Mason2018; Mason et al. Reference Mason, Wronski and Kane2021). Significant and valuable as these contributions are, however, this literature has not systematically dealt with a set of more wide-ranging connections between, on the one hand, voters’ perceptions of social groups and their connections to the parties and, on the other hand, their political behaviour.
Thus, while voters’ social identities – that is, their knowledge and evaluation of their membership of a given social group – undoubtedly influence political behaviour (for an overview, see Huddy Reference Huddy, Huddy, Sears and Levy2013), voters are also influenced by their evaluations (positive as well as negative) of groups to which they do not necessarily belong – what is commonly referred to as reference groups. Put simply, the theory maintains that voters vote more for parties that are seen as related to positively valued groups (and vice versa). While the approach is pioneered in older work (Miller and Wlezien Reference Miller and Wlezien1993; Miller et al. Reference Miller, Wlezien and Hildreth1991; Wlezien and Miller Reference Wlezien and Miller1997; see also Brady and Sniderman Reference Brady and Sniderman1985; Campbell et al. Reference Campbell, Converse, Miller and Stokes1960; Conover Reference Conover1988; Converse Reference Converse2006 [1964]; Nelson and Kinder Reference Nelson and Kinder1996; Sniderman et al. Reference Sniderman, Brody and Tetlock1991) and lies implicit in some research on specific groups (such as racial or ethnic groups; see below), most contemporary studies of group effects on political behaviour have not systematically addressed this idea as a complementary, overall approach (although see Elder and O’Brian Reference Elder and O’Brian2022; Kane et al. Reference Kane, Mason and Wronski2021; Kinder and Dale-Riddle Reference Kinder and Dale-Riddle2012; Mason et al. Reference Mason, Wronski and Kane2021; Piston Reference Piston2018; Stuckelberger and Tresch Reference Stuckelberger and Tresch2024).Footnote 2
Furthermore, the focus on the important, expressive side of voters’ perceptions of and relations to parties – that is, the view that voters’ party assessments are driven by their images of the social group composition of parties’ electorates, of ‘Who’s at the party’ (Kane et al. Reference Kane, Mason and Wronski2021; see also Claassen et al. Reference Claassen, Djupe, Lewis and Neiheisel2021; Green et al. Reference Green, Palmquist and Schickler2002) – neglects a relevant alternative. Thus, voters may also base their evaluations of the parties on more instrumental concerns about whether or not they see the parties as fighting for the interests of various groups. As is the case for their expressive concerns, we should expect voters’ instrumental evaluations of the parties to moderate the relationship between their group evaluations (or identities) and vote choice (for similar arguments, see, for example, Bornschier et al. Reference Bornschier, Haffert, Häusermann, Steenbergen and Zollinger2024; Evans & Langsæther, Reference Evans and Langsæther2021; Nicholson and Segura Reference Nicholson and Segura2012; Robison et al. Reference Robison, Stubager, Thau and Tilley2021; Stubager and Slothuus Reference Stubager and Slothuus2013). The aim of this article is to examine the veracity of these underexplored perspectives on the (subjective) social group–vote choice relationship. Over and above what we already know about the influence of voters’ objective positions and identities, can we find evidence that choices at the ballot box are also influenced by voters’ evaluations of groups with which they do not identify? And to what extent are such relationships moderated by voters’ instrumental perceptions of the parties’ relations with these groups? These are the main questions pursued here.
Focusing on the Danish case and using both observational data and a preregistered survey experiment about the relationship between voters’ evaluations of social classes and their perceptions of parties’ agency on their behalf, the study provides evidence in support of the reference group perspective. Thus, the analyses show, with high internal as well as external validity, that the influence of social groups on party choice extends beyond the expressive, identity-based processes that dominate current research on the topic. Voters’ choice of party is also influenced by groups of which they are not themselves members. Sympathy for the working, upper-middle, and upper classes is significantly related to voting for parties seen as fighting for these classes, even when controlling for class identity. For the working and upper classes, further, the relationships are mediated by individual-level perceptions of the extent to which parties are perceived to fight for the classes. In contrast to the view from a dominant line of research (for example, Kinder and Kam Reference Kinder and Kam2009), hence, out-groups do not only serve as adversaries. Perceptions of out-groups can be positively related to support for parties seen as fighting for the groups – a result not foreseen by identity perspectives.
Thereby, the article offers several contributions to extant knowledge. First, the analyses document – for the first time with experimental evidence of the core effect – the importance of also considering reference group mechanisms when investigating the role of social groups in voters’ decisions. That the effect of social groups extends beyond expressions of social identities constitutes an important addition to our knowledge about the political role of social groups. Thus, the underlying reference group theory offers an alternative, complementary perspective on the group–politics relationship. This perspective permits an analysis of the political implications of a wider range of groups than the prevalent identity perspective, which struggles to explain, for example, how small minority groups can be viewed positively also by those who do not themselves identify as part of the group. In that way, the reference group perspective opens other avenues for understanding contemporary politics.
Second, by demonstrating these effects on vote choice, the study moves beyond the prevalent focus on attitudes or party identification (see, for example, Elder and O’Brian Reference Elder and O’Brian2022; Kane et al. Reference Kane, Mason and Wronski2021; Robison and Moskowitz Reference Robison and Moskowitz2019); it moves from attitudes to (intended) behaviour, in other words. Third, the study documents the existence of group-based processes also outside the polarized US context on which most existing work is focused. Finally, and despite claims about the ‘death of class’ (Clark and Lipset Reference Clark and Lipset1991, Reference Clark and Lipset2001; Beck Reference Beck2007; Dalton Reference Dalton2019), the study adds a new, reference group-based aspect to the discussion about the relevance of social classes to politics in advanced industrial democracies.
The following section outlines the theoretical basis of reference group process and instrumental perceptions of parties’ group relations as an alternative to expressive, identity-based processes and perceptions – resulting in a set of hypotheses to guide the analyses. The empirical part is divided into two studies, an observational and an experimental. For each, I begin by setting out the design, data, measurement, and modelling before presenting the analyses. The conclusion discusses the implications of the results.
Groups in Voters’ Minds
The question of how groups influence political behaviour has spawned scholarly interest for decades. The seminal contribution by Converse (Reference Converse2006 [1964], 41) suggests that for groups to exert an influence, ‘individuals must be endowed with some cognitions of the group as an entity and with some interstitial “linking” information indicating why a given party or policy is relevant to the group’. The current study focuses on exactly these two steps in the process: voters’ cognitions of the groups and how they link them to the parties. Groups are, in accordance with Kinder and Dale-Riddle (Reference Kinder and Dale-Riddle2012, 11; see also Kinder and Kam Reference Kinder and Kam2009, 32), defined as ‘[a]ny collection of people that constitutes a psychological entity for any individual’.
Voters’ Group Cognitions
As noted, most research on voters’ thinking about social groups has been focused on their identification (or not) with such groups. The core idea (going back to classic studies like Berelson et al. Reference Berelson, Lazarsfeld and McPhee1954; Campbell et al. Reference Campbell, Converse, Miller and Stokes1960; Lipset and Rokkan Reference Lipset, Rokkan, Lipset and Rokkan1967) is the distinction between in- and out-groups, that is, groups to which individuals either see themselves as belonging or not (see Huddy Reference Huddy, Huddy, Sears and Levy2013). According to Social Identity Theory (SIT), social identities involve the individual’s ‘knowledge of his membership in a social group (or groups) together with the value and emotional significance attached to that membership’ (Tajfel Reference Tajfel and Tajfel1978, 63). Social identities are, thus, potentially deeply held. Identifying with a group implies that you have a stake in how the group fares. Furthermore, while an identity tells you who you are (that is, your in-group), it often also tells you who you are not (that is, your out-group). And the expectation is that in-groups receive preferential treatment (see also below). In many cases, furthermore, out-groups are treated distinctly worse because they ‘are assumed to be … unfriendly, uncooperative, unworthy of trust, dangerous and more’ (Kinder and Kam Reference Kinder and Kam2009, 8; although see below). Focus is, in other words, on the individual and her sense of group belonging as the driver of political behaviour.Footnote 3
Despite theoretical roots as old as those of the identity approach, the reference group approach has – as an overall perspective – enjoyed considerably less attention (although see Elder and O’Brian Reference Elder and O’Brian2022; Kane et al. Reference Kane, Mason and Wronski2021; Kinder and Dale-Riddle Reference Kinder and Dale-Riddle2012; Mason et al. Reference Mason, Wronski and Kane2021; Miller and Wlezien Reference Miller and Wlezien1993; Miller et al. Reference Miller, Wlezien and Hildreth1991; Piston Reference Piston2018; Stückelberger Reference Stückelberger2019; Stuckelberger and Tresch Reference Stuckelberger and Tresch2024; Wlezien and Miller Reference Wlezien and Miller1997).Footnote 4 Interest in the political implications of reference group processes is motivated by the ‘fact that men frequently orient themselves to groups other than their own in shaping their behavior and evaluations’ (Merton and Rossi Reference Merton, Rossi and Merton1968, 288; italics in the original). Focus is, in other words, on how groups function as ‘reference points for the formation of attitudes and decisions about behavior’ (Campbell et al. Reference Campbell, Converse, Miller and Stokes1960, 296; see also Converse Reference Converse2006 [1964], 70; Lewis-Beck et al. Reference Lewis-Beck, Jacoby, Norpoth and Weisberg2008, 306). The starting point, thus, is voters’ evaluations of social groups in society – including of groups other than the ones with which they identify or that are opposed to their identity groups (but we would expect similar mechanisms for such groups). This is a crucial difference to identity-based approaches: individuals do not need to have something at stake personally when being influenced by reference groups. Reference group processes may, therefore, be particularly important in relation to small groups such as some ethnic or gender minorities that, despite having very limited membership, are subject to intense political debate. In these debates large numbers of people who do not identify as members of the minority groupFootnote 5 – that is, for whom the minority is an out-group – still harbour positive evaluations of the group. Such positive evaluations cannot be understood from an identity perspective. They make good sense from the perspective of reference group theory, however, which therefore offers a complementary perspective to that of identity theories.
Although not the core issue here (see Piston Reference Piston2018, 37–8 for an extended discussion), group evaluations may have different origins. While for some people and groups, they may be grounded in early socialization (see Elder and O’Brian Reference Elder and O’Brian2022, 1409), for others they may be based on observations of society during which people recognize the existence of various groups about which they form evaluations. Such evaluations can be formed as a relatively effortless process that does not necessarily involve considerations of the individual’s own relationship with the groups in question. According to Brady and Sniderman (Reference Brady and Sniderman1985, 1075), ‘likes and dislikes are easy to form and, even more important, easy to remember’ (for related perspectives, see Cuddy et al. Reference Cuddy, Fiske and Glick2007; Fiske Reference Fiske2011, Reference Fiske2019; Sears Reference Sears and Kuklinski2001). In particular, Conover (Reference Conover1988, 57; see also Piston Reference Piston2018) highlights the dominance of deservingness considerations – of ‘who is getting what and whether they deserve it’ – in forming group evaluations. This point is backed up by the extensive literature on deservingness perceptions (for example, Attewell Reference Attewell2021, Reference Attewell2022; Petersen et al. Reference Petersen, Slothuus, Stubager and Togeby2011). Indeed, the deservingness literature could be seen as an implicit application of reference group theory to the domain of welfare support.Footnote 6
Group evaluations are subsequently used as a basis for forming opinions and guiding vote choice. Basically, if a policy or party is seen as favouring a liked group, support for the policy or party is positively related to evaluations of the group (see Miller et al. Reference Miller, Wlezien and Hildreth1991, 1137). There is, however, also an easily overlooked flip-side to this expectation. Thus, if a party or policy is seen as adversarial to a positively evaluated group we should expect to see a negative relationship between group evaluations and support for the party or policy (see also Elder and O’Brian Reference Elder and O’Brian2022, 1409; Kane et al. Reference Kane, Mason and Wronski2021, 1794–5). These theoretical considerations can be summarized as the reference group hypothesis, claiming that evaluations of social groups exert a positive effect on voting for parties seen as connected to the groups.
Identity versus Reference Group?
There is no inherent conflict between identity and reference group processes. They can very well function alongside each other. As hinted at, however, while the theories would in most cases lead to the same predictions regarding in-groups (making these less interesting when testing reference group theory), they part company over the role of out-groups. For social identity theory in-groups are the focus of attention, so to speak, and behaviour is expected to be driven by the desire to enhance the in-group’s standing (Brewer Reference Brewer1999, Reference Brewer, Sibley and Barlow2016). This does not necessarily evolve into outright antagonistic behaviour vis-à-vis the out-group (see, for example, Herring et al. Reference Herring, Jankowski and Brown1999) but – per Kinder and Kam’s (Reference Kinder and Kam2009) theory of ethnocentrism – it may very well. In this perspective, however, there is no theoretical basis for understanding why or how an out-group can be viewed positively, and that political behaviour can aim to further the interests of out-groups. This is not to say that the existence of such positive out-group reactions should be seen as disconfirming or refuting social identity theory; but they call for a different theoretical apparatus – and that is exactly what reference group theory offers. According to reference group theory, thus, out-groups can also be viewed positively, as objects of striving or of compassion (Piston Reference Piston2018, 34–6; see also Sirin et al. Reference Sirin, Valentino and Villalobos2016 for a related perspective). Therefore, the central question for establishing the relevance of reference group theory is whether we can find evidence of a positive group influence on party choice also for groups of which people are not members themselves, that is, over and above the influence of social identities (and objective group memberships).
‘Linking’ Information: Perceived Group–Party Relations
The second part of the process relating group cognitions (be they group identities or evaluations) to party choice is voters’ perceptions of parties’ relations with groups – Converse’s (Reference Converse2006 [1964], 41) ‘interstitial “linking” information’. Overall, this aspect has received less scholarly attention than the cognitions discussed above. Research on the topic has, however, also been dominated by the identity perspective. Thus, most existing work (Ahler and Sood Reference Ahler and Sood2018; Claassen et al. Reference Claassen, Djupe, Lewis and Neiheisel2021; Green et al. Reference Green, Palmquist and Schickler2002; Kane et al. Reference Kane, Mason and Wronski2021; Rothschild et al. Reference Rothschild, Howat, Shafranek and Busby2019; Titelman and Lauderdale, Reference Titelman and Lauderdale2023) is built on the premise that voters’ perceptions of and reactions towards parties are driven by a desire to express their social identities through their (party identification and/or) vote choice. According to this approach, voters choose parties based on the match between their group identities and their perceptions of the group composition of parties’ electorates (Green et al. Reference Green, Palmquist and Schickler2002, 8, 10). Voters’ perceptions of ‘Who’s at the party’ (Kane et al. Reference Kane, Mason and Wronski2021, 1783) influence which party they want to attend, so to speak.
Just as reference group evaluations provide a non-expressive alternative to identity processes, there is also a non-expressive alternative to evaluations of ‘Who’s at the party’ as ‘“linking” information’. This alternative type of information is not about which voters are attracted by the parties, but about how the parties themselves behave vis-à-vis groups. The idea has roots in cleavage theory (for example, Lipset and Rokkan Reference Lipset, Rokkan, Lipset and Rokkan1967; see also Bornschier et al. Reference Bornschier, Haffert, Häusermann, Steenbergen and Zollinger2024, 9–11; Stuckelberger and Tresch Reference Stuckelberger and Tresch2024), which portrays parties as group mobilizers or even as agents of social groups, acting on their behalf in the political arena. What matters are the instrumental considerations regarding the extent to which parties are seen as fighting for the interests of a given group. As noted by Wlezien and Miller (Reference Wlezien and Miller1997, 629), such perceptions ‘reflect historical alignments between groups and parties and the more recent behaviour of groups and political party leaders and candidates themselves’.
While the importance of the historical aspect is indicated in some analyses (for example, Nicholson and Segura Reference Nicholson and Segura2012; Stubager and Slothuus Reference Stubager and Slothuus2013; see also Petrocik Reference Petrocik1996), recent analyses (Dolinsky Reference Dolinsky2023; Evans & Langsæther, Reference Evans and Langsæther2021; Finseraas et al. Reference Finseraas, Heath, Langsæher and Smets2025; Horn et al. Reference Horn, Kevins, Jensen and van Kersbergen2021; Huber, Reference Huber2022; Piston Reference Piston2018; Robison et al. Reference Robison, Stubager, Thau and Tilley2021; Thau Reference Thau2021) have shown how parties’ direct appeals to social groups (in which they claim to stand for the group) condition the relationship between voters’ group memberships and party choice. On this basis, one should expect the extent to which voters perceive parties as fighting for a given group to also condition the relationship between group evaluations and vote choice. Specifically, the instrumental party perception hypothesis maintains that the more a party is seen as fighting for a group, the more positive should be this relationship.
The phrasing of the hypothesis as regarding a ‘more positive’ relationship is purposefully chosen. Thus, it does not make sense to expect a stronger perceived association between a group and a party to lead to a stronger negative relationship between group evaluations and voting for the party. Seeing a party as fighting for a group should lead to a stronger positive relationship between evaluations of the group and voting for the party – that is, a higher probability of voting for the party if one evaluates the group positively and a lower probability if one evaluates the group negatively, than if the party is not seen as fighting for the group.
Expressive versus Instrumental Considerations?
As is the case for the two types of group cognitions, some element of overlap between the two types of linking information is to be expected, particularly for larger groups and parties. However, as Tables B2 and B3 in the appendix show (see below for an introduction of the measures), the overlap is far from perfect in the data applied here.Footnote 7 Further, the two types of information differ both theoretically and practically. Theoretically, instrumental considerations about parties’ group-related agency constitute a complementary alternative to the expressive considerations that dominate extant research (see, for example, Bornschier et al. Reference Bornschier, Haffert, Häusermann, Steenbergen and Zollinger2024, 58). They offer an account of voter behaviour also on this point that is not driven by expressive, identity concerns but by a concern for ‘who’s getting what’ from the parties.
In particular, and as is the case for reference group evaluations (see above), perceptions of the extent to which parties fight for various groups provide a better basis for grasping the political role of small minority groups (for example, some ethnic or gender groups) that cannot, by virtue of their limited size, make up a larger part of any party’s electorate – but that may still feature prominently (in both positive and negative ways as seen from the group) in parties’ political programmes and, in that way, influence voters’ perceptions of the parties. The instrumental perceptions, thus, avoid the problem that the base rate of some groups is very low.
Practically, instrumental perceptions also avoid problems associated with the base rate of parties, that is, that larger parties are likely to attract more voters from all groups (Bornschier et al. Reference Bornschier, Haffert, Häusermann, Steenbergen and Zollinger2024, 58), which may create perceptual difficulties for voters, probably more so in multiparty systems with parties of varying size. Moreover, and very basically, it is not given that the two types of party perceptions are related to party choice in the same way nor that they moderate the relationship between group evaluations and party choice in the same way and with the same strength. Figure A1 in the appendix provides an overview of the different theoretical perspectives discussed.
Social Class Groups and Party Choice
In the following, the general theoretical considerations laid out above are applied to the case of classes. First, however, it is worth briefly taking stock of existing knowledge about the role of subjective perceptions of class as an influence on voter behaviour. While most extant work has focused on the influence of objective indicators of class (for example, occupation) – generally finding an effect on both attitudes and vote choice (for reviews, see Evans & Langsæther, Reference Evans and Langsæther2021; Lindh and McCall Reference Lindh and McCall2020; Manza and Crowley Reference Manza, Crowley, Albarracin and Johnson2018) – relatively little work exists charting the subjective side of the class–politics relationship that is in focus here. Despite the claims about ‘the death of class’ (Clark and Lipset Reference Clark and Lipset2001; see also Beck Reference Beck2007; Dalton Reference Dalton2019; Franklin Reference Franklin, Franklin, Mackie and Valen1992), recent analyses have, however, documented that class continues to be a meaningful concept and basis of identification for voters (see, for example, D’Hooghe et al. Reference D’Hooge, Achterberg and Reeskens2018; Franko and Witko Reference Franko and Witko2023a, Reference Franko and Witkob; Harrits and Pedersen Reference Harrits and Pedersen2018; Heath and Bennett Reference Heath, Bennett, Frankenburg, Cleary and Curtice2023; Hout Reference Hout, Lareau and Conley2008; Stubager et al. Reference Stubager, Tilley, Evans, Robison and Harrits2018). Class is alive in people’s minds, in other words. Likewise, class identification has been found to influence attitudes and party choice, although to somewhat varying degree across countries (D’Hooghe et al. Reference D’Hooge, Achterberg and Reeskens2018; Evans et al. Reference Evans, Stubager and Langsæther2022; Sosnaud et al. Reference Sosnaud, Brady and Frenk2013).
There is less evidence about classes as reference groups. Although a few socio-economic groups feature in some reference group-based analyses (for example, Miller et al. Reference Miller, Wlezien and Hildreth1991; Kane et al. Reference Kane, Mason and Wronski2021; see also Bartels Reference Bartels2016, 113–16), they are not labelled as classes and in subsequent analyses they are bundled together with other groups usually associated with the same party, thereby preventing an assessment of class-specific relationships. Only Piston (Reference Piston2018) pays extensive, direct attention to class – but under the labels ‘rich’ and ‘poor’. We do not know, thus, how class groups labelled as such may or may not serve as reference groups for voters. While this is, on the one hand, a semantic difference, it is also, on the other hand, a difference that may have real implications in that different labels may have different political connotations. Thus, particularly the label ‘working class’ has a distinct status as the explicit social group mobilized by Social Democratic and Labour parties when they formed in the late nineteenth and early twentieth century. A working-class identity was and is to many a source of strength and pride to an extent that is difficult to associate with the term ‘poor’.
Moreover, labels such as ‘rich’ and ‘poor’ do not refer to groups to which very many people see themselves as belonging. In fact, Piston (Reference Piston2018, 34–5) chose ‘rich’ and ‘poor’ as the focal points exactly for this reason. While this of course implies that these groups stand out as ‘pure’ reference groups it also – by design – reduces the influence of class identification in the analysis. When only very few people identify with a given group, identification should play less of a role. This is unfortunate both because people do in fact identify with other class groups and because class identification has been shown in previous research to influence vote choice. By excluding from focus groups with which the majority of the population identifies, we become unable to assess whether reference group processes add insights over and above the influence of class identification. The theory is not subjected to a particularly strong test, in other words.
With respect to voters’ perceptions of parties’ relationships with classes, there is also very little evidence to stand on. Just a few recent studies (Evans & Tilley, Reference Evans and Tilley2017; Grant and Evans Reference Grant and Evans2024; see also Rothschild et al. Reference Rothschild, Howat, Shafranek and Busby2019, 432) have looked directly at linkages between classes and parties, while others (Bornschier et al. Reference Bornschier, Haffert, Häusermann, Steenbergen and Zollinger2024; Piston Reference Piston2018) do so using different labels for class-related groups. Overall, the studies find that, in Germany, France, Switzerland, and the UK, Social Democratic or Labour parties as well as the US Democrats are associated with the working class (or ‘the poor’ or ‘people who do hard and tiring work’) while the British Conservatives are associated with the middle class and US Republicans with ‘the rich’. The British studies explicitly look at the parties’ perceived attention to class interests, but the measures used in the other studies do not discriminate between different types of perceived group–party linkages.
Class Evaluations and Vote Choice in Denmark
Following standard terminology, the study covers four classes – the working, middle, upper-middle, and upper class. The working, middle, and upper classes span, so to speak, the class hierarchy and are frequently referred to in discussions about class. However, previous research (for example, D’Hooghe et al. Reference D’Hooge, Achterberg and Reeskens2018) has shown (and the data presented below corroborate) that only very few identify with the upper class. Thus, the upper-middle class is included as a more palatable category at the higher end of the spectrum. This implies that identity process can be expected to operate for the working, middle, and upper-middle classes while the upper class serves as a ‘pure’ reference group in that (almost) no one identifies with it. The three former classes are, thus, particularly important for testing the influence of reference group process over and above identity processes.
How should we expect class evaluations to influence vote choice in the Danish multiparty system? To form more detailed expectations, we may rely on voter perceptions of the extent to which parties fight for the interests of the classes presented in Table B2 in the appendix (see below for the exact measure).Footnote 8 As is clear from the table, the traditional left-wing parties such as the Social Democrats, the Red–Green Alliance, and the Socialist People’s Party are clearly seen as fighting for the working class and much less for the upper-middle and upper classes. Thus, we should expect support for these parties to be higher among those who evaluate the working class positively and/or the upper-middle and upper classes negatively (and vice versa). On the opposite side, the Liberals, the Conservatives, the Liberal Alliance, and the New Right are all seen as fighting for the upper-middle and upper classes and not for the working class. For these four parties, we should, therefore, expect the reverse pattern.
The Social Liberal Party and the Danish People’s Party are mostly focused on so-called second-dimension issues such as immigration (Stubager et al. Reference Stubager, Hansen, Lewis-Beck and Nadeau2021) which do not straightforwardly relate to classes partly because the underlying conflicts are not primarily economic. This is reflected in the aggregate pattern of perceptions of the parties’ class relations in Table B2 where the two parties are associated with the classes in rather indistinct ways. Therefore, I expect support for these parties to be unrelated to evaluations of the classes.
The role of the middle class is more ambivalent. On the one hand, the class is the largest in terms of identifiers in Denmark, selected by more than 40 per cent of respondents when asked to identify with a class (see Table B1). This means that we should expect it to occupy a substantial role in people’s thinking about classes and class relations simply because it is so large. On the other hand, the position of the class in the middle of the class hierarchy implies that it has a less distinct position also in political terms. Parties’ relations with the middle class may, further, have become unclear due to the realignment of the segment of socio-cultural professionals towards left-wing parties on the basis of second-dimension issues (see Abou-Chadi and Hix Reference Abou-Chadi and Hix2021; Oesch and Rennwald Reference Oesch and Rennwald2018) – a development that is not captured when focusing on standard class labels (cf. the use of alternative labels by Bornschier et al. Reference Bornschier, Haffert, Häusermann, Steenbergen and Zollinger2024). This lack of clarity is reflected in Table B2 where there is very little variation between the parties’ perceived agency on behalf of the middle class. For this reason, I do not present any specific hypotheses regarding the middle class.
The party perception hypothesis, for its part, is easier to specify in that it maintains that all of the hypothesized relationships will be more positive the more a party is seen as fighting for a given class. Table A2 provides an overview of the detailed expectations.
Study 1
The first study takes an observational approach to test the hypothesized relationships, thereby prioritizing the external validity of the conclusions.
Data, Measurement, and Modelling
Data for the study were collected online by YouGov from 18- to 70-year-old members of the company’s standing panel of respondents during the period 17 to 31 August 2021. A total of 3,615 panellists were asked to participate in the survey, 2,065 of which provided complete responses, thus producing a response rate of 57.1 per cent. The resulting sample is representative of the population on gender, age (in three categories), education, and region of residence.
Measurement
The dependent variable is respondents’ party choice, assessed by asking respondents which party they would vote for if parliamentary elections were held tomorrow. Because several of the parties gather limited support, they were combined into five groups (with the individual parties in parentheses) to obtain reliable estimates: Left Socialists (Red–Green Alliance, Socialist People’s Party), Social Democrats, Social Liberals, Danish People’s Party, and Economic Right (Conservatives, New Right, Liberal Alliance, Liberals).Footnote 9
There are two main independent variables in the study. The first, class evaluations, has been assessed using two different measures included in current analyses in the field. First, and mimicking the thermometer scales used in some US studies (for example, Wlezien and Miller Reference Wlezien and Miller1997; Lewis-Beck et al. Reference Lewis-Beck, Jacoby, Norpoth and Weisberg2008), the survey contained a simple sympathy measure: ‘Then we have some questions about different groups in society. Please place each group on this scale from 0 to 10 where 0 means that you think very badly about the group and 10 means that you think very highly of the group.’Footnote 10 Respondents were asked to evaluate the working class, the middle class, the upper-middle class, and the upper class. The groups were presented in randomized order mixed up with eight other groups that are not analysed here.Footnote 11
Second, the study included an alternative measure which has appeared in some more recent studies (Bornschier et al. Reference Bornschier, Häusermann, Zollinger and Colombo2021, Reference Bornschier, Haffert, Häusermann, Steenbergen and Zollinger2024; Kane et al. Reference Kane, Mason and Wronski2021; Mason and Wronski Reference Mason and Wronski2018). The measure asks respondents ‘How close do you feel to the following groups? By “close” we mean people who are most like you in their ideas, interests, and feelings.’ The response scale ran from 0 ‘Not at all close to the group’ to 10 ‘Very close to the group’.Footnote 12 The measure was asked for the same groups (also in randomized order). In order to compare the relationships of the two evaluation measures with party choice without risking confounding across the two measures, they were each asked of a randomized half of the sample. As shown in Appendix C, however, there are only very minor differences in the relationship between the two measures and party choice. For this reason, the two measures have been (standardized and) combined in the analyses.
The second, main independent variable is used to test the party perceptions hypothesis. A randomly selected half of the sampleFootnote 13 was asked about their perceptions of the extent to which parties fight for the working, middle, upper-middle, and upper class (again in randomized order):Footnote 14 ‘To what extent do you think that the parties below are fighting for [X class]? Please respond on a scale from 0 to 10 where 0 means “not at all” and 10 means “to a very great extent”.’ Then followed a randomized list of the nine largest parties. In the analyses, perception scores for each of the five party groups are constructed as the mean of the scores for each party in a given group. The measure is rescaled to 0–1 in the analyses.
The survey also included an extensive set of control variables. First and foremost, respondents’ class identification was assessed using a version of the standard International Social Survey Programme item asking respondents ‘People sometimes talk about the existence of different social groups or social classes. In which of these social classes would you place yourself?’Footnote 15 with the response categories lower class, working class, lower-middle class, middle class, upper-middle class, and upper class (in addition to don’t know). Thirteen respondents identifying with the upper class have been combined with the upper-middle class except for the analyses where they are excluded (see below).
Respondents’ gender, generation, parents’ occupation (based on father’s occupation; if missing, mother’s), education, household income, and occupation (see Table B1 for coding of the variables) were also included in the models (although with the exception noted below).Footnote 16 Since, as Elder and O’Brian (Reference Elder and O’Brian2022) have shown, group evaluations may influence policy attitudes, the models did not include controls for respondents’ political attitudes in order to avoid post-treatment bias. However, controlling for both cultural and economic attitudes does not alter the main results reported below (even if the relationships are weakened for some parties; see Appendix C). See Appendix B for descriptives of all variables.
Modelling
Multinomial logistic regression is used to test the reference group hypothesis by assessing the relationship between class evaluations and party choice while controlling for the background factors. In a second step, parties’ perceived relations with the classes (as per the party perception hypothesis) are interacted with class evaluations (again controlling for the background factorsFootnote 17 ). To avoid collinearity, I estimated separate models for the influence of party perceptions regarding one class at a time. Because the party perceptions are attributes of the alternatives (that is, the party groups) in the analysis, these models are estimated using conditional logit.
All models include evaluations of all four classes – the working, middle, upper-middle, and upper classes – as well as the controls for class identity and the socio-structural factors. However, to effectively separate reference group processes from identity processes, the results presented are based on models excluding, one class at a time, respondents who identify with each of the four classes. For example, when evaluating the influence of evaluations of the working class or perceptions of the extent to which parties fight for this class, respondents identifying with the working class have been excluded from the analyses to prevent their identification from influencing the strength of the evaluation and party perception variables (for a similar approach, see Kane et al. Reference Kane, Mason and Wronski2021, 1793–4).Footnote 18
Results
The first step of the analysis tests the reference group hypothesis. The core results reported in Figure 1 and the underlying models (see Table D1 in the appendix for the coefficients) show evaluations of the working and upper classes to be overall significant in all models (that is, p < 0.05 in Wald tests of the block of coefficients associated with each variable). Evaluations of the upper-middle class are only significant in the model excluding upper-middle-class identifiers and very nearly so (p = 0.06) in the model excluding upper-class identifiers. Overall, thus, class evaluations are related to party choice over and above both objective class positions and subjective class identification. This is the first important finding supporting the veracity of reference group theory.
Class evaluations and party choice. Predicted probabilities.
Note: N = 789–1,455. McFadden’s R 2 = 0.13–0.16. Left Soc: Left Socialists, SD: Social Democrats, Soc Lib: Social Liberals, DPP: Danish People’s Party, ER: Economic Right. See the text for model description. Control variables are left at their values in the sample. Respondents identifying with the class in focus in each sub-plot have been excluded from the analysis for that sub-plot. Vertical lines indicate the interquartile range on class evaluations. Error bars indicate 95 per cent confidence intervals. The x-axes span the relevant variation for each variable.
Figure 1 adds more nuance about the direction of the relationships. The general pattern conforms to the expectations. Thus, we see the expected positive relationship between working-class sympathy and voting for the Left Socialists and the Social Democrats, while the relationship for the parties on the Economic Right is negative. This is entirely as expected. We can observe the opposite relationships for the upper-middle and upper classes: those with more positive evaluations of these two classes are less likely to vote for the Left Socialists and the Social Democrats and more likely to vote for the Economic Right. For the working and upper classes, all relationships are significant at least at p < 0.05 and this also applies to the differences between the predicted values across the interquartile range.
For the upper-middle class, only the difference between voting for the Left Socialists and the Economic Right comes out significant in the model, while the difference between voting for the Social Democrats and the Economic Right obtains a p-value of 0.09. These weaker results for the upper-middle class are due to the inclusion of evaluations of the middle and upper class in the model. If either of these variables are removed, upper-middle-class evaluations are significant overall and for all parties except the Social Liberals.Footnote 19 This reflects that the upper-middle class occupies a middle position between the middle and upper classes, with which it shares either one or the other part of its name. When evaluations of all classes are included in the model simultaneously, those for the upper-middle class are suppressed.
As expected, furthermore, evaluations of the classes are with one exceptionFootnote 20 not significantly related to voting for the Social Liberals and the Danish People’s Party. Finally, and also in accordance with expectations, evaluations of the middle class turn out insignificant in both the overall Wald tests of their contribution to the model as well as when testing for changes in the predicted probabilities across the interquartile range. These results reflect the ambivalent position of this class in the political competition – evaluations of it do not appear to be related to vote choice.
All in all, therefore, the reference group hypothesis clearly meets with support in the test. With the exception of the ambivalent role of the middle class, voters’ class evaluations have a statistically and, in many cases, also substantially significant relationship with their propensity to vote for the different parties. Thus, the differences in the predicted probability of voting for, for example, the Social Democrats or the Economic Right range around 11 percentage points across the interquartile range on evaluations of the working and upper classes. Given the extensive set of controls employed in the models – including respondents’ class identification – as well as the procedure of excluding those identifying with a given class when examining the influence of evaluations of the class, relationships of this strength are of substantial significance. Thereby, they clearly indicate the existence of reference group processes over and above identity processes and that not only for groups with which no one identifies (cf. the discussion of extant research).
The conditioning role of party perceptions
The second step of Study 1 adds the measure of voters’ perceptions about the degree to which parties fight for the four classes to the model (now estimated by means of conditional logit) one by one in order to test the instrumental party perceptions hypothesis. The results are reported in Table D2. Before discussing the interactions that are central to the hypothesis, it is interesting to note the unconditional relationships between the party perception measures and party choice. Thus, all perception variables are significantly related to party choice. While perceptions of parties’ agency on behalf of the working, middle, and upper-middle classes are positively related to voting for a given party, perceptions of parties’ agency on behalf of the upper class are negatively related to party choice. I return to these relationships below.
That aside, the analysis provides support for the party perception hypothesis for three out of the four classes. Thus, the relationship between, on the one hand, working, middle, and upper-class evaluations and, on the other hand, party choice is significantly more positive if a party is seen as fighting for the class in question than if it is not.Footnote 21 This applies irrespective of whether fighting for a class has a positive (as for the working and middle classes) or negative (as for the upper class) main relationship with party choice. Focusing on the parties and classes that show the most consistent relationships across the two steps of Study 1,Footnote 22 Figure 2 plots the predicted probability of voting for the Left Socialists, the Social Democrats, or the Economic Right across evaluations of the working (left column) and upper (right column) classes separately for those who perceive the party to be at either the 75th or the 25th percentile on the party agency variables.Footnote 23
The conditioning role of perceptions of parties’ agency on behalf of the working and upper classes for the relationship between evaluations of the two classes and party choice. Predicted probabilities.
Note: See the text for model description. Control variables are left at their values in the sample. See Table D2 for the underlying models. The vertical lines indicate the interquartile range on working and upper-class evaluations. Error bars indicate 95 per cent confidence intervals. Respondents identifying with the class in focus in each sub-plot have been excluded for that sub-plot.
The results in the two columns of Figure 2 are parallel in the sense that those for the Left Socialists and the Social Democrats look alike while those for the Economic Right deviate. Thus, Figure 2 reproduces the overall finding from Figure 1 that there is a positive relationship between working-class evaluations and voting for the Left Socialists and the Social Democrats while the relationship is negative for the Economic Right – and vice versa for the upper class. The crucial new information added in Figure 2, however, is that these relationships are conditioned by the extent to which the parties are seen as fighting for the two classes. Among those who see a party as a protagonist of one of the two classes (that is, places the party at the 75th percentile), the relationship is consistently – and as expected – more positive than among those who see the party as rather indifferent to the classes (that is, place it at the 25th percentile).
For the Left Socialists and the Social Democrats, this implies that the positive relationship between working-class evaluations and voting for the party is clearly strongest for those who see the party as a strong protagonist of the working class, whereas there is essentially no relationship between working-class evaluations and support for the party among those who do not see the party as fighting for the class. Likewise, but as a mirror image, the results for the Economic Right also support the expectations. But because the unconditional relationship between working-class evaluations and voting for these parties is negative as shown in Figure 1, the positive conditioning influence of the party perceptions results in a weaker (that is, more positive) relationship between working-class evaluations and voting for these parties among those who see them as strongly related to the working class than among those who see them as weakly related to the class. For upper-class evaluations, we see the opposite pattern.
The figure also reveals the strength of the relationship between the party perception variables and party choice. Thus, we can see that among respondents scoring above −2 on the working-class evaluation measure, the light-grey curve lies significantly above the dark curve in the plots. For these respondents, seeing a party as fighting for the working class has a positive relationship with voting for the party. For the upper class, we see the opposite pattern in that the relationship between party perceptions and vote choice is stronger at the negative end of the evaluation scale with those not seeing a party as fighting for the upper class being more likely to vote for it (although differences are smaller compared to the working class). These results may reflect an underlying sense of deservingness according to which the working class is seen as deserving more political attention than the upper class, but more research is needed to elucidate the matter.
As shown in Appendix F, parallel models estimated using a measure of expressive perceptions of the extent to which parties attract voters from the different classes (designed as a multiparty version of the measure used by Kane et al. Reference Kane, Mason and Wronski2021) produce much weaker results. Thus, none of the interactions with the class evaluation variables turn out significant. On this basis, instrumental party perceptions about the extent to which the parties fight for the classes appear a more important moderator than do expressive concerns about which voters the parties attract, although more research is needed to draw a firmer conclusion.
Study 1 Summary
All in all, the results from Study 1 support both the reference group and instrumental party perceptions hypotheses, although to varying degrees for the four classes. Results are clearly strongest for the working and upper classes where we find the expected patterns for the three parties (/party groups) for which they were expected – and not for the two where none was expected. As also expected, the analyses found insignificant results for the middle class, most likely due to the politically indistinct position of the class. For the upper-middle class, the results were overall weaker and slightly mixed. This is due at least in part to the suppressing influence of evaluations of the middle and upper classes. The results thereby provide observational evidence of a relationship between group evaluations and party choice over and above the previously investigated relationships with objective and subjective (that is, identity) group memberships. Also those who are not members of either the working, upper-middle, or upper classes seem to care about their evaluations of the classes – and the parties perceived agency on behalf of the classes – when they go to the polls. And so the relationship between group perceptions and party choice is not just a question of ‘Us against Them’ (Kinder and Kam Reference Kinder and Kam2009; for parallel results using a different theoretical approach, see Bornschier et al. Reference Bornschier, Haffert, Häusermann, Steenbergen and Zollinger2024). Furthermore, and in contrast to the approach of Piston (Reference Piston2018), the processes also operate for classes with which substantial numbers of people actually identify.
However, while representative, observational data of the kind employed in Study 1 provide some certainty that the relationships found mirror those in the population, the spectre of endogeneity hovers over the results presented. Thus, both perceptions of parties’ agency on behalf of classes and evaluations of the classes may be subject to reverse causality or the relationships may be driven by unmeasured prior factors (see also Clifford et al. Reference Clifford, Simas and Suh2025 for a related argument regarding the relationship between group evaluations and policy attitudes). To address these issues and provide evidence of a causal component to the relationships documented above, Study 2 employs a preregistered survey experiment designed on the basis of the results from Study 1.
Study 2
Data and Design
Data for Study 2 was collected online by YouGov during the period 2 July to 5 August 2024. The sample size is 7,090Footnote 24 and respondents are representative of the adult Danish population on gender, age (in three groups), region of residence, and education (in three groups). Due to the considerable requirements with respect to respondents and the difficulty of administering trustworthy manipulations regarding more than one class or party at a time, the design focuses on reproducing one set of the results from Study 1, viz. the relationship between, on the one hand, evaluations of the working class and the extent to which the Socialist People’s PartyFootnote 25 (SPP, forming part of the Left Socialists in Study 1) is seen as fighting for it and, on the other hand, party choice. Thus, the experiment is designed to show that evaluations of the working class have a causal effect on support for the SPP and that this effect is moderated by the extent to which the party is perceived to fight for the class. This means that the study involved two independent variables that are both subjected to manipulation: evaluations of the working class and the perceived extent to which the SPP fights for the working class. To achieve this, the experiment is designed as a 2 × 2 factorial with randomization taking place in two rounds.Footnote 26
After the introduction to the survey (including informed consent), respondents first had to respond to two attention check measures. Those who failed these are (as preregistered) removed from the analysis. Then followed the class identity measure applied in Study 1. Next was the first treatment in which respondents were randomly assigned to either a positive evaluation or a negative evaluation condition. In both conditions, respondents saw three manufactured news stories each of roughly similar length. The contents of the stories were chosen to associate the working class with either generally valued things or behaviours, or the opposite. In the positive evaluation condition, the stories described the working class as important for Denmark’s economic wellbeing, crucial for the transition to a green economy, and more friendly towards other people – all three arguments ascribed to (loosely defined) expert sources. In the negative evaluation condition, the stories described the working class as being more delinquent, more engaged in tax evasion, and less friendly towards other people – again ascribed to expert sources.Footnote 27 After reading the three stories (respondents were prevented from moving on in the survey for 30 seconds to induce them to read), respondents answered two questions. The first one required recall of the group mentioned in the stories in order to reinforce the treatments.Footnote 28 The second asked respondents to evaluate the working class on the 0–10 sympathy scale used in Study 1.
Then followed the second manipulation where respondents were again randomized into two equally sized groups: party fights for working class or party fights against working class. Respondents saw two new manufactured stories. In the party fights for working class condition, the stories mentioned that a higher than usual number of SPP MPs are from the working class and that the party had voted in favour of many laws to the benefit of the working class. In the party fights against working class condition, the stories were the opposite. Respondents were again prevented from moving on (this time for 20 seconds) before answering two questions about which party and group were mentioned in the stories, again to reinforce the treatment.Footnote 29
Then followed a measure of the perceived extent to which the SPP fights for or against the interests of the working class (assessed on a 0–10 scale, labelled as such: 0 Fights against the interests of the working class, 5 Fights neither for nor against the interests of the working class, 10 Fights for the interests of the working class). Finally, the main dependent variable – propensity to vote for the SPP – was assessed using a 0–10 scale (0 Very unlikely, 10 Very likely).Footnote 30 All dependent variables also included a Don’t know option; these responses were excluded from the analysis. At the end of the survey, respondents were informed that the news excerpts they had read were fictitious and offered the opportunity to withdraw their consent and have their responses deleted from the survey.
The analyses were conducted using OLS with results presented in graphical form (see the underlying coefficients in Appendix G). Following the approach in Study 1 and the preregistration, respondents identifying with the working class were excluded from the analysis in order to remove the influence of identification with the class. This leaves a total of 5,601–5,706 respondents for the analyses (depending on missing data on the dependent variable).
Manipulation Check
Before presenting the main results, Figure 3 shows the effect of the treatments on the two variables that they were intended to manipulate. Each plot is constructed so that respondents who received the same treatment on the second manipulation (intended to influence perceptions of the SPP’s agency on behalf of the working class) are grouped together while different colours are used to indicate the treatment on the first manipulation (intended to influence evaluations of the working class). As can be seen, the treatments worked as intended. In the left panel, thus, we can observe that respondents treated to have a more positive evaluation of the working class are indeed more positive towards the class than those treated to have a negative evaluation.Footnote 31 The difference between the two groups is both statistically and substantially significant.
Treatment effects on sympathy for the working class and perceptions that the SPP fights for the working class. Scale scores.
Note: N = 5,705/5,706. Bars represent mean scores for each of the four experimental groups on sympathy for the working class (left panel) and the perceived extent to which the SPP fights for the working class (right panel). Error bars represent 95 per cent confidence intervals.
The same applies in the right half of the figure. Here, the bars within each block are of similar height and consistent with the treatments. The second group, which was treated to perceive the SPP as fighting for the working class, shows the highest levels of such perceptions, while the first group, which was treated to perceive the party as fighting against the working class, scores below the neutral mid-point, that is, they perceive the party as fighting against the interests of the group. Again, the effect is statistically and substantially significant. We see no evidence that the two treatments affect each other: the effect of the sympathy manipulation is the same irrespective of the party treatment and vice versa.
Main Analysis
The first step in the analysis is, as in Study 1, the test of the unconditional relationship between working-class evaluations and the vote. This is simply the effect of the first treatment on the propensity to vote for the SPP averaged across the two conditions on the second treatment. The effect comes out significant (p < 0.05)Footnote 32 with an estimate of 0.2 on the 0–10 scale. At this basic level, hence, the experimental evidence of Study 2 replicates the finding from Figure 1, thereby supporting the reference group hypothesis. The estimated effect is smaller in absolute sense than the effect of the treatment on working-class evaluations themselves (see above), which may suggest the existence of pre-treatment effects that are not easily overturned with treatments like the ones employed here.
Moving on to testing the instrumental party perception hypothesis, Figure 4 presents the predicted propensity to vote for the SPP across all four combinations of the two treatments. Again, the results conform to the predictions. Thus, the first two bars reveal that evaluations of the working class do not influence the propensity to vote for the SPP when it is seen as fighting against the interests of the group. This replicates the observational result from the upper-left-hand plot in Figure 2 where we see a flat line for those who see the Left Socialists – of which the SPP is part – as fighting less for the working class (that is, places the party at the 25th percentile on the scale).Footnote 33 The last two bars also replicate the result from Figure 2 in that we find the expected positive effect of working-class evaluations on the propensity to vote for the SPP when it is seen as fighting for the group. The differences between the last two bars as well as between them and the first two are statistically significant (at least at p < 0.05), as is the overall interaction between the two treatments (see Table G1). Incidentally, the figure also replicates the positive main effect of being seen as fighting for the interests of the working class found in Figure 2.
Treatment effects on the probability of voting for the SPP. Scale scores.
Note: N = 5,601. Bars represent the mean probability of voting for the SPP for each of the four experimental groups. Error bars represent 95 per cent confidence intervals.
Summary Study 2
The experimental test of the reference group and instrumental party perceptions hypotheses in Study 2 provides support for the expectations. Thus, we can observe the expected positive relationship between working-class evaluations and the propensity to vote for the SPP both unconditionally and, as expected, more strongly when the party is seen as fighting for the group. The analysis not only provides further support for the expectations, it also replicates precisely the observational results from Figure 1 and, particularly, Figure 2. The size of the estimated effects on the vote propensity variable should be evaluated in light of the fact that the variable is not directly manipulated and is subject to numerous other influences generating substantial pre-treatment effects. In this perspective, the effects found are substantial.
Conclusion
Across the two studies, the analyses provide support for both the reference group and instrumental party perception hypotheses. Thus, evaluations of social groups – the working, upper-middle, and upper classes in the current study – influence vote choices; an influence that is moderated by the extent to which parties are seen as fighting for the interests of the classes (at least for the working and upper classes). The results match up, furthermore, across the observational and experimental designs. To the best of my knowledge, this study is the first to provide direct, experimental evidence for a reference group influence on the vote.Footnote 34 Likewise, the (direct as well as) conditional effect of instrumental party perceptions on the vote has not been documented before.
The study contributes to different strands of literature. First, to the general political behaviour discussion about the drivers of vote choice among which evaluations of reference groups, that is, also groups of which voters are not necessarily themselves members, can now be confidently counted. Thus, the effects found go beyond those of objective and subjective (that is, identification) group memberships, just as the study goes beyond the dominant focus on attitudes or partisan identity in demonstrating the effects for vote intention. Thereby, the study establishes reference group theory as an overall approach to understanding the influence of social groups on the vote independently of social identities and/or group memberships that are central in other approaches such as cleavage theory (see, for example, Bornschier et al. Reference Bornschier, Häusermann, Zollinger and Colombo2021, Reference Bornschier, Haffert, Häusermann, Steenbergen and Zollinger2024; Westheuser and Zollinger Reference Westheuser and Zollinger2025). In contrast to such approaches, reference group theory offers a theoretical lens for understanding the political implications also of positive evaluations of out-groups and thereby of a wider selection of groups.
The study also moves beyond the geographic and group contexts of extant research. Finding these effects in the Danish multiparty context shows how group processes exist also in countries characterized by far less social and political polarization (see, for example, Stubager et al. Reference Stubager, Hansen, Lewis-Beck and Nadeau2021) than the United States, where most extant studies originate. Finally, the finding that reference group processes operate for groups defined in class terms, and with which substantial shares of the population identify, provides evidence that class continues to be a relevant consideration to citizens of Western democracies despite the ‘traditional view’ (Heath and Bennett Reference Heath, Bennett, Frankenburg, Cleary and Curtice2023, 5) that class differences have lost their social and political relevance.
The two latter points, however, also serve to highlight the importance of the social and political context. Thus, as the results for the middle class demonstrate, the perceptions and political linkages of classes vary and in ways that are likely to also vary across countries. Specifically, perceptions of the middle class and its relationship with political parties vary considerably between countries (Stubager et al. Reference Stubager, Tilley, Evans, Robison and Harrits2018; Evans et al. Reference Evans, Stubager and Langsæther2022), a problem exacerbated by the realignment of socio-cultural professionals away from the traditional middle class, as discussed above. This means that the exact results for combinations of classes and parties shown above cannot necessarily be replicated for other countries. If expectations are appropriately calibrated to the national context including the relevant group labels, however, similar mechanisms should be expected also in other contexts. That said, more research is needed to determine whether the mechanisms found here are limited to classes placed towards the ends of the class hierarchy (for example, the working and upper classes) or also extend to classes in the middle of the hierarchy (like the middle class).
Likewise, and relatedly, it is also relevant to explore further how the mechanisms identified may work for groups for which the influence of parties’ perceived agency is more balanced than is the case for the working and upper classes in the current study. Specifically, it would be interesting to pursue whether the absence of an interaction between evaluations and party perceptions for the middle class is a general result for such ‘middle groups’ – thereby implying a restriction in the scope conditions of the theory – or merely a feature of the middle class (at least in Denmark).
At the overall theoretical level, the results presented here provide further evidence for a more flexible view of the relationship between social groups and political behaviour as suggested by Clifford et al. (Reference Clifford, Simas and Suh2025). This relationship is not, as one might assume on the basis of the dominant stream of research in the area, exclusively founded on expressive, identity-based considerations of in- and out-groups and which parties they prefer. Rather, potentially less deep-seated evaluations of social groups and instrumental considerations about the extent to which parties work for or against group interests also play a role. This suggests a more malleable component to the group–politics relationship than previously assumed, thereby proposing a way for political entrepreneurs to both intensify and de-intensify political group polarization.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S0007123425000444
Data availability statement
Replication data for this article can be found in Harvard Dataverse at: https://doi.org/10.7910/DVN/ZTIBQC.
Acknowledgements
The paper has benefitted from generous and helpful comments and suggestions from David Attewell, Simon Bornschier, Ruth Dassonneville, Christoffer Hentzer Dausgaard, Lena Huber, Leonie Huddy, Spencer Piston, Rune Slothuus, Julie Wronski, and Lene Aarøe, as well as the three anonymous reviewers. Previous versions of the paper have been presented at seminars at Aarhus University and the University of Gothenburg, as well as the annual conferences of the Danish Political Science Association (2022), ECPR (2023), EPOP (2023), and the MPSA (2022). I am grateful for all comments and suggestions received; the responsibility for any remaining errors etc. remains mine.
Financial support
Work on the paper was supported by grants from Research Fund Denmark (Grant# 1327-00113) and the Carlsberg Foundation (#CF23-0813) to the author. Work on the paper was finalized while the authors was a fellow at the Aarhus Institute of Advanced Studies.
Competing interests
None to disclose.
Ethical standards
As discussed in Appendix H of the Supplementary material, the research complies with APSA’s Principles and Guidance for Human Subjects Research.
Open access
