Convergences and divergences of a Bourdieusian and occupation-based approach to social class: evidence from Croatia

A disclaimer that social class is a ‘contested’ concept has become common to the point of cliché across much academic writing on social class analysis. On one hand, such a disclaimer encapsulates an acknowledgement of anti-class rhetoric, particularly in reference to Pakulski and Waters (1996) dramatically titled and widely cited book, The Death of Class, in which the authors proclaimed the intellectual bankruptcy of the class paradigm. Contributing to such rhetoric has been evidence that class identities, understood as a sense of class belonging, are weak, and that there is a ‘middle class’ identity bias (Andersen and Curtis 2012). On the other hand, the disclaimer also captures tensions between different approaches to class analysis, ranging from neo-Marxist, neo-Weberian and Bourdieusian theoretical traditions to more interpretative and pragmatist approaches. In particular, the recent theoretical debate within social class analysis has been marked by a positivistic-sounding rivalry over which quantitative approach to class is ‘best’: neo-Weberian, and more specifically Goldthorpean, or Bourdieusian. To illustrate this, Goldthorpe (2013) harshly dismissed Savage et al.'s (2013) Bourdieusian analysis of the social class structure of the UK as ‘a data-dredging exercise – resulting in much conceptual confusion’, and has advocated for the explanatory power of his own influential occupation-based class schema. Conversely, Albert et al. (2018) have claimed that Goldthorpe's approach is insufficient when it comes to, for example, explicating wider cultural and social activities and identities. Underpinning their critique is an advancement of Bourdieusian class analysis as a maximalist, multidimensional and inductive approach to inequalities. Interestingly, this debate does not seem to reflect general trends in social class research, since according to a review of articles published in prominent sociology journals over the last decade, the majority of articles examining class do not draw on any of the main theoretical traditions in class analysis (Cepić and Doolan 2018).

This article juxtaposes a Bourdieusian and an occupation-based approach to social class in order to explore how these approaches converge or diverge when it comes to empirically identifying the size of high-level class groups and exploring the relationship between ‘objective’ social class position and class self-identification in Croatia. The Croatian context adds further layers to social class as a ‘contested’ concept. Specifically, whereas social class analysts in socialist former Yugoslavia debated how to frame inequality dynamics within a political and economic system ideologically committed to classlessness, the class paradigm became taboo in the post-socialist context due to its Marxist connotations. Indeed, as Ost (2015) has observed, in the broader context of post-communist European countries, ‘class was the key concept of the toppled nemesis’ (2015, p. 546). Croatia today illustrates this observation. Before the 1990s, the concept of social class was widely used, albeit for opposing purposes: to sanctify the political and economic status quo as well as to critique it from both a Marxist and liberal standpoint (Dolenec et al.2015). On the other hand, the first two decades of the twenty-first century saw an eclipse of the concept in public and academic discourse (Grdešić 2015).¹ Among other justifications, including the concept's political baggage as a remnant of rejected socialism, one common argument is that the country's transformation into a democracy, a market economy and a nation-state warrants, at best, a deferral of class analysis until the end of the ‘transition period’ (though what the cut-off point for this would be remains unclear), since any conclusion drawn in such a state of flux would be gravely misleading. Such caution indicates an approach to social class which overlooks social class dynamics as a structural feature of any capitalist society, a commitment that, it is worth noting, both Bourdieusians and neo-Weberians share.

The article begins by sketching Bourdieusian and occupation-based approaches to class analysis with a focus on the different ways in which they measure social class. We also discuss how this literature has informed our approach. This is followed by an outline of our survey, which was carried out as part of the project Social stratification in Croatia: structural and subjective aspects, co-funded by the Croatian Science Foundation and the University of Zadar. In the findings section, we present a Bourdieusian construction of social space in Croatia, based on multiple capitals and obtained by multiple correspondence analysis and cluster analysis and contrast it to an occupation-based approach, with a particular emphasis on how both approaches align with class self-identification on the part of our respondents.

In his influential work Pedagogy of the Oppressed, first published in 1968 in Brazil, Freire makes a distinction between a minority dominant class who can eat, dress, wear shoes, be educated, travel and listen to Beethoven, and the dominated, those without food, clothes, education and opportunities for travel, who ‘much less listened to Beethoven’ (1970, 34). Freire's observation, published a decade before Distinction, highlights differences in volume and composition of capitals that constitute the foundation of Bourdieusian social class analysis. For Bourdieu (1984) social class is defined ‘by the structure of relations between all the pertinent properties which gives its specific value to each of them and to the effects they exert on practices’ (1984, p. 106). These properties include volume and composition of capital, but also gender, age, ethnic origin and social origin. More operationally, and using multiple correspondence analysis, Bourdieu (1984) constructs a two-dimensional social space defined by overall volume of capital and the relative weight of the different forms of capital (dominant and dominated). For Bourdieu (1984), capital exists in three forms: cultural, social and economic. Cultural capital can be further divided into three categories: embodied (long-lasting dispositions of the mind and body reflected in, for example, manners, taste in art and music and linguistic competences), objectified (in the form of cultural goods such as artworks, books and dictionaries) and institutionalized (educational qualifications) (1985). Bourdieu (1985) characterizes social capital, on the other hand, as the aggregate of resources, both actual and potential, which are linked to membership in a group. The group provides each of its members ‘with the backing of the collectively-owned capital’ (Bourdieu 1985, p. 248). Bourdieu's measurement of class is based on a range of indicators of economic capital (wealth, income, property ownership) and cultural capital (educational qualifications, relations to art and culture) possessed by individuals located in positions throughout the occupational system (Weininger 2005). According to Bourdieu (1984), overall volume of capital is key to distinguishing between different classes and runs from those best provided with to those most deprived of capitals (1984, p. 114). In this respect, Bourdieu (1984) makes a distinction between the dominant class or the ‘bourgeoisie’, the intermediate class or ‘petty bourgeoisie’ and the dominated ‘working class’. Bourdieu (1984) refines these distinctions by highlighting the distribution of different kinds of capital within classes. For example, in the dominant class there are occupations with higher cultural capital and lower economic capital (e.g. university professors) and those that have lower cultural and higher economic capital (e.g. employers, industrialists).

Contemporary research that has ‘put Bourdieu to work for social class analysis’ (Flemmen 2013) covers different areas of study, from migration (e.g. Oliver and O’Reilly 2010) to ethical eating (e.g. Johnston et al.2012). Most prominently, we find Bourdieu-inspired class research in the areas of cultural consumption and education.² However, as Albert et al. (2018) have noted, whereas social inequalities are regularly studied across different themes, ‘the ‘big’ picture of the class structure at country level is presented less frequently’ (2018, p. 545). Bourdieusian attempts to capture this ‘big’ picture include Flemmen et al.'s (2017) work on cultural manifestations of class divisions in lifestyle differences in Norway using multiple correspondence analysis (MCA), as well as Savage et al.'s (2013) influential development of a model of class for the UK based on indicators of cultural, economic and social capital and using latent class analysis. Similarly, using MCA, Flemmen (2012) and Hjellbrekke and Korsnes (2009) have researched the internal differentiation of the Norwegian upper class, whereas Albert et al. (2018), using latent class analysis, and Cvetičanin and Popescu (2011), using MCA, constructed the social space of Hungary and Serbia respectively.

The work of these authors highlights several important aspects of contemporary Bourdieusian research. First, Bourdieusian class analysis is a European-wide endeavour spanning countries such as the UK, Norway, Hungary and Serbia, to name a few. According to Atkinson (2020), Bourdieu's ‘coherent and recognisable set of concepts can be and has been drawn on to make sense of life in London or Lima’ (2020, p. 1). Second, although there are overlaps in how capitals have been operationalized in research (educational level frequently used as an indicator of cultural capital), there is no universal model. Indeed, the form and number of capital indicators varies: Cvetičanin and Popescu (2011) use respondents musical taste, their educational level and their parents educational level as indicators of cultural capital, whereas Flemmen et al. (2017), while taking the highest level of education of respondents and their parents as measures of cultural capital, also include respondents’ discipline/field of education, education in the military/armed forces and the extent to which a respondent grew up in a home with lots of books, music, art, and other cultural interests. Finally, quantitative Bourdieusian class analysis is associated predominantly with multiple correspondence analysis, though authors also use other data analysis techniques such as latent class analysis to propose Bourdieusian inspired social class schemas. However, as Weeden and Grusky (2012, 157) have observed, it is neo-Weberian class analysis, especially based on the Eriksson-Goldthorpe-Portocarero (EGP) class scheme, that has ‘arguably become the de facto standard within the big-class tradition’. The EGP scheme is most prominently used in studies of social mobility, but also in the areas of work, political attitudes, health outcomes and education.

Whereas Bourdieu's approach to social class focuses on the amount and interrelationship of different capitals, for Weber (1978), we can speak of class when a certain number of people share a specific causal component of their life chances, and when this component is exclusively represented by economic interests in the possession of goods and opportunities for income and is determined by the conditions of commodity and labour markets (1978, p. 927). Chan and Goldthorpe, referring to Weber, define classes as existing insofar as ‘a number of people have in common a specific causal component of their life-chances’ (Chan and Goldthorpe 2007, p. 514). For authors in this tradition, class positions are defined by employment relations; occupation and employment status are used as indicators of these relations (Bukodi et al.2015). Indeed, and with reference to Parkin (1971), Connelly et al. (2016, 1) have highlighted that within sociology occupations have been taken as ‘the most powerful single indicator of levels of material reward, social standing and life chances’. Furthermore, according to these authors, despite the ‘end of class’ argument and the advocacy of a Bourdieusian class scheme by authors such as Savage et al. (2013), ‘there is no strong empirical evidence that dissuades us of the extremely high value of using existing occupation-based measures in the secondary analysis of large-scale social surveys’ (Connelly et al. 2016). Authors using the EGP scheme or one of its heirs tend to examine the effects of class on specific outcomes, such as education or mobility, using regression analyses or structural equation modelling (e.g. Bihagen 2008, Barg 2015, Kulin and Svalfors 2013; Nordlander 2016).

Unlike Bourdieusian class analysis, a modified version of the class scheme developed by Goldthorpe and his colleagues has shaped national and international social class schemes. In its most extensive form, the Goldthorpe schema identifies 11 classes, but there is also a 7 and 4 class version.³ In its 7 class version (as noted by Chan and Goldthorpe (2007), for example) the classes are: professionals and managers higher grade (I), professionals and managers lower grade (II), routine non-manual employees (III), small employers and proprietors (including farmers) and self-employed workers (IV), technicians and supervisors of manual workers (V), skilled manual workers (VI) and nonskilled manual workers (VII). According to Goldthorpe (2000), members of the service class are typically advantaged over members of the working class in terms of income, career prospects, more favourable chances of maintaining continuity of employment and ‘the greater security that they can expect in sickness or old age’ (2000, 166). Importantly, Chan and Goldthorpe (2007) note that in terms of hierarchy, while classes I and II are generally advantaged relative to members of other classes and classes VI and VII are relatively disadvantaged, ranking of the intermediate classes is problematic.

Within a national framework, Goldthorpe's class scheme, slightly modified, is now the official measure of class in the UK government's National Statistics Socio-Economic Classification (NS-SEC). Internationally, classifications aimed at comparative research have been developed. Like the NS-SEC, the European Socio-economic Classification (ESeC) takes the occupational structure as ‘the backbone of the stratification system’ (Rose and Harrison 2007, p. 460). Its link to the Eriksson-Goldthorpe-Portocarero (EGP) schema is explicit: ‘The decision to adopt the EGP classification as the basis for ESeC was made because it is widely used and accepted internationally, is conceptually clear, and has been reasonably validated both in criterion terms as a measure and in construct terms as a good predictor of health and educational outcomes’ (Harrison and Rose 2006, p. 4). ESeC, grounded in the claim that employment relations constitute class, is operationally based on occupation, employment status (employers, self-employed, managers, supervisors and employees) and size of organisation (large and small employers) and each combination of these is assigned a class position (Rose and Harrison 2007, 465). There are ten ESeC classes⁴ and Rose and Harrison (2007) suggest that Classes 1 and 2 are advantaged over the others ‘in terms of greater long-term security of income; being less likely to be made redundant; less short-term fluctuation of income since they are not dependent on overtime pay, etc.; and a better prospect of a rising income over the life course’ (2007, 465).

Finally, and most recently, the European Socio-economic Groups (ESeG) classification has been developed with the aim of enabling the grouping of individuals with similar economic, social and cultural characteristics throughout the EU (Meron et al.2014, p. 7). Similarly to ESeC, though abandoning the concept of employment relations (Christoph et al.2020), in order to assign people to different class positions, information is required on their occupation and employment status (Bohr 2018; Meron et al.2014, p. 7). We use the ESeG classification in this study⁵, as an occupation-based approach validated by Eurostat, in order to examine how it fares against a Bourdieusian approach focused on capitals and their distribution. More specifically, we have used the ESeG-2014 classification which has two core variables: ISCO08 occupation and employment status (employee/self-employed), as well as two additional variables for people not in paid employment: age and status (retired/student/disabled). At its most aggregated level there are 9 ESeG groups which we outline in the methodology section below. ESeG has recently been productively used by Hugree et al. (2017) in their analysis of social class structures in Europe. So far, in the Croatian context, the EseG classification has been used to assess the relationship between food self-provisioning and social class background (Ančić et al.2019b) as well as social class and political behavior (Ančić et al.2019a).

The data for this article comes from a survey conducted on a nationally representative sample of adult citizens living in Croatia (over the age of 18) (N = 1000) that was conducted in November and December 2017. Stratified multi-stage sampling was conducted using two-stage stratification (six regions defined as groups of Croatian counties, four settlement sizes defined by the number or residents). In order to minimize selection bias, a three-step sampling approach was used. First, settlements were selected using the PPS method (‘probability proportionate to size’). Second, households were randomly selected (‘random starting points method‘, ‘random walk method’). Third, the choice of respondents was random (‘next birthday method’). The overall response rate was 45%. As a method of non-respondent adjustment, weighting factors were used which adjusted the structure of data on socio-demographic characteristics with the structure of the population (with respect to gender, age, education, size of settlement and region).

In order to explore how economic, cultural and social capital structure Croatian social space, we first analysed the data using multiple correspondence analysis (Le Roux and Rouanet 2010). As a descriptive and inductive method, MCA is suitable for examining relationships among categorical variables, without making any hypothesis about distributions in the data (Le Roux and Rouanet 2010). In this way, MCA differs from usual multivariate techniques such as regression analysis which examine relationships between independent and dependent variables. What makes MCA especially suited for constructing a social space is its relational character: positions of categories or individuals within ‘the cloud’ are meaningful only in relation to other positions. As a consequence, similarities and differences between individual responses are reflected in the distances between individual points in the cloud (Le Roux and Rouanet 2010). Secondly, in order to identify specific subgroups in the social space, we performed ascending hierarchical cluster analysis on the results of the MCA.

For the MCA analysis, we used indicators of economic, cultural and social capital as active variables. Economic capital was operationalised with four indicators: (1) average monthly net income (salary or pension) of the respondent; (2) estimated value of the real estate of the respondent or his/her household; (3) amount of savings and (4) subjective evaluation of ability to satisfy household needs (ability to ‘make ends meet’).⁶ We used three indicators of social capital: (1) overall network diversity; (2) diversity of friendship network and (3) membership in different types of organisations (sports/recreational, educational/cultural, professional, humanitarian, religious). Both measures of network diversity were derived from a position generator which included 12 occupational positions.⁷ Overall network diversity, which indicates the total number of accessed positions, is one of the standardized measures of social capital (Lin 2001; Erickson 1996). As a continuous variable (range 0-12; M = 6.32; SD = 3.16), it was recoded into three categories (‘low’, ‘average’, ‘high’). Similarly, friendship network diversity was calculated as the number of occupations in which the respondent had a friend (range 0-12; M = 2.84; SD = 2.71). For the purpose of MCA, this was also recoded into three categories (‘low’, ‘average’, ‘high’).

Following Bourdieu, we included indicators of all three forms of cultural capital: institutionalised, objectified and embodied. Five active variables were used: (1) respondent's educational level; (2) respondent's parents’ educational level; (3) estimated number of books in the household; (4) taste in theatre-going and (5) number of foreign languages spoken by the respondent. Institutionalised cultural capital was measured as the highest level of completed education of respondents and their parents, using an ordered scale consisting of 10 different education levels (from incomplete primary school to completed postgraduate university study programme). To measure objectified cultural capital, which indicates possession of cultural goods, an estimated number of printed books in the household was used as an indicator. Respondents were given a scale ranging from 1 (‘0-10 books’) to 6 (‘more than 500 books’). Respondent's embodied cultural capital was operationalized using two indicators: (1) theatre-going and (2) number of foreign languages spoken by the respondent. We work with the assumption that high values of these indicators suggest, as Atkinson (2020) has put it, a symbolic mastery of systems of symbols and signs which are valued in Croatian society. The distribution of active variables used in the analysis is presented in the Appendix Table A1.

Supplementary variables, i.e. variables that are projected into the constructed space and which do not define distances between individual responses, were selected in order to answer three research questions: (1) how do sociodemographic indicators correspond to the constructed social space (2) how does an occupation-based model of social classes in Croatia map onto the distribution of economic, social and cultural capital and (3) how do these pair with class self-identification. As a result, supplementary variables that were projected into the social space, included (1) sociodemographic indicators: age, gender and size of settlement, (2) ESeG occupational groups and (3) respondents’ class self-identification. Following the European Socio-economic Groups Classification (Bohr 2018), the occupational status of the current or last job of respondents together with their employment status was classified into eight occupational groups: (1) managers, (2) professionals, (3) technicians and associated professional employees, (4) small entrepreneurs, (5) clerks and skilled service employees, (6) skilled industrial employees, (7) lower status employees and (8) other non-employed persons. Retired respondents were classified according to their last job.

Class self-identification was measured with one item which asked respondents to choose which class they belong to: lower class, working class, lower middle class, middle class, upper middle class and upper class. Due to the small number of respondents who chose the answer ‘upper class’, their answers were added to the group of ‘upper middle class’. We acknowledge that, on the one hand, ‘It has never been fashionable to study class subjectivities (or indeed identity more generally) using quantitative data’ (Surridge 2007, p. 209), and on the other, that people's indication of which class they belong to does not tell us whether class is a meaningful or ambivalent identity for them. However, responses to this question do suggest some complementarity between ‘objective’ class locations and subjective class positioning. The distribution of supplementary variables is presented in the Appendix Table A2.

To construct the Croatian social space, specific MCA was used (Le Roux and Rouanet 2010), with the 10 ‘missing’ categories of the active variables treated as passive categories. The analysis included 56 categories in total, 25 related to cultural capital, 15 to economic capital and 16 to social capital. Data were analysed using SPAD 9.1. Since the first two dimensions contribute the most to the variance (76.7%) and the contribution of the third dimension is significantly lower (Table 1), we decided to interpret only the plane defined by the first two dimensions. For the interpretation of an axis, only categories with above average contributions are considered (Le Roux and Rouanet 2010). The explaining points for both axes and their contributions are presented in the Appendix Table A3.

This first dimension of the social space (Axis 1, horizontal) explains 66.9% of the total variance and is constituted by the total volume of capital, distinguishing between individuals with a low volume of capital (on the left) or the dominated class, and those with a high volume of capital or the dominant class (on the right). To the left of Axis 1, we find individuals with low economic capital (Income –, Needs -), who are less educated (Edu Prim) and whose parents are less educated (EduF Inc.Prim, EduM Inc.Prim). Here on this left side there are also indicators of low social capital: low values of both indices of network diversity (NetDiv -, NetDiv_F -), as well as indicators of low objectified and embodied cultural capital (Books < 10, Theatre -, Flang -). In contrast, on the right side of the map there is a concentration of indicators of high volumes of capital. First of all, and unlike respondents who have low volumes of capital, on the top right corner of the map, there are categories which indicate high institutionalized cultural capital on the part of both respondents (EduUni) and their parents (EduF_High, EduM_High). Indeed, we can notice educational reproduction on both the far left and far right side of the map. In this group of indicators there is also an indicator of social capital: membership in professional organisations (Org_Prof +). Moving toward the central group of categories on this right side, we find indicators of high volumes of economic capital: belonging to the group of respondents with the highest income level (Income +), having savings (Savings +) and ability to satisfy needs (Needs +). There are also indicators of objectified and embodied cultural capital (Books 100+, Theatre +, FLang 2).

With only 9.8% of the explained variance, Axis 2 is of secondary importance. According to the contributions of categories, this dimension of social space is mostly related to the distribution of cultural capital, especially with regard to educational level. What is also important to notice is the contribution of categories that are related to middle and/or average values of cultural capital: having completed four year secondary education (Edu 4y), having parents with completed three year vocational education (EduF 3y, EduM 3y), having an ambivalent attitude towards going to the theatre (Theatre +-) and having between 10 and 25 books at home (Books 10-25). Briefly, whereas the main opposition is between those with high and those with low volumes of capital, the analysis has also identified an intermediate position. Interestingly, Table A3 in the Appendix shows that indicators of cultural capital, and especially educational level of respondents’ parents, have above average contributions when compared to indicators of economic and social capital. This could be interpreted as a legacy of socialism where, arguably, differences in economic capital were not significant between classes (see Kraaykamp and Nieuwbeerta 2000). As Eyal et al. (1998) have argued, in the social ‘space’ of post-communist capitalism cultural and social capital were more important than economic capital for attaining privilege.

Unlike gender, age has a very strong effect along Axis 1, with the largest deviation (d = 1.047) between the youngest and the oldest group (18-30 and 61+). The difference is also large between the oldest group of respondents and those between 31 and 45 years (d = 0.910). Age differences also appear on the second axis. The largest difference has been found between the two extreme age groups (d = 0.999). The oldest age group is located in the upper left quadrant, associated with a lower total volume of capital, the age group between 46 and 60 is near the center, while the mean points for the two youngest age groups (between 18 and 30 and between 31 and 45) are on the opposite side (Figure 1).

Finally, as Figure 1 indicates, the differences related to the size of settlement are not so large, as the mean points are close to the barycenter (Hjellbrekke 2019). In particular, there is only one large deviation on Axis 1 (d = 0.752), between the two most remote categories (less than 2,000 and more than 100,000 inhabitants). What is important to notice is that the mean point of the smallest settlement size is located on the left, which indicates its association with a lower volume of capital. By contrast, the mean point of the large cities (more than 100,000 inhabitants) is on the right, meaning that they are associated with higher total volume of capital. These findings reveal that there is a rural-urban divide in Croatian society related to the distribution of capital.

With regard to class self-identification, our findings corroborate Andersen and Curtis (2012) observation that there tends to be a middle-class identity bias in surveys examining class subjectivity: more than one third of our sample self-identified as ‘middle class’. However, our analysis also shows that there is an association between ‘objective’ class location, measured both in terms of volume of capital and occupation and employment status, and subjective class identification. From the mean points presented in Figure 2, we can conclude that respondents who perceive themselves as members of the lower class have the lowest volume of capital. Furthermore, skilled industrial and lower status employees tend to self-identify as working class and also have lower volumes of economic, cultural and social capital, as the mean points of these categories are located on the left side. By contrast, professionals with high volumes of capital are more likely than any other occupational group to self-identify as upper-middle class. More specifically, whereas 47% of professionals self-identify as middle class and a following 23% self-identify as upper-middle and upper class, only 2% of skilled industrial and lower status employees self-identify as upper-middle and upper class. Obviously, the largest difference is between the lower class and upper middle class on Axis 1 (d = 1.859). We explore class self-identification in more detail below (Table 2).

Cluster 1 is the smallest cluster, with 12.1% of the sample, best described as the dominant class of urban professionals and managers. This cluster consists of privileged respondents with high volumes of capital.⁹ In terms of cultural capital, there is an overrepresentation of respondents with university education, those whose parents also have university degrees, those with more than one hundred printed books at home and those who can speak two or more foreign languages. Individuals in this cluster also have higher volumes of economic and social capital, when compared to the average values of the sample. In particular, more than half of the people in this cluster have savings, a quarter report the highest income levels and a third the highest value of real estate. Furthermore, respondents in this cluster also tend to have higher values of network variability. Considering supplementary variables, by occupational groups, managers and professionals are overrepresented, as well as those who perceive themselves as members of the upper and upper-middle classes. This cluster is also characterized by an above average percentage of respondents from large cities. Interestingly, whereas the MCA analysis did not point to any significant gender differences in the distribution of capitals, this analysis shows that men form the majority of this class.

Cluster 2 consists of older members of the middle class (13.7%), whereas Cluster 3 (15.4%) of younger members of the middle class. Overall, members of these classes have less capital than the dominant class but more than the working class. Noteworthy are distinctions between the older and younger generation. The younger middle class is, on average, better educated than older members of the middle class. They are also more likely to have more books at home and speak more foreign languages, have savings, a higher income and live in cities. On the other hand, members of the older middle class own more expensive real estate in comparison to the younger generation. In relation to occupational groups, two groups are overrepresented in the older and younger middle class clusters: technicians and associated professional employees, and clerks and skilled service employees. These classes are also dominated by people who tend to perceive themselves as middle class, however it is interesting to note that the younger middle class is less likely than the older to self-identify as working class. There are no significant gender differences between the two groups.

Cluster 4, the largest cluster, includes 24.9% of the respondents and we have labelled it the younger working class. It mostly consists of respondents aged between 18 and 30 and between 31 and 45. The younger working class is dominated by respondents who have four year schooling (none of the respondents have university education) and whose parents have completed three year vocational education. When we consider other indicators of cultural capital, those who possess between 10 and 25 books and who speak one foreign language are overrepresented. Compared to members of the middle classes, this group of respondents tends to have lower volumes of economic capital, in terms of savings, income and real estate. Cluster 5 (19%), on the other hand, encapsulates older members of the working class. They tend to be between 46 and 60 years old and live in smaller settlements. Compared to the younger working class, they are less likely to speak foreign languages, they have lower savings, lower income, lower values of network diversity. However, in both these classes the overrepresented occupational group are lower status employees. In addition, there are no significant gender differences between these groups. Self-perception as members of the working class is also somewhat overrepresented in both these groups, though slightly more for the older working class.

Finally, cluster 6 (14.9%) represents the underclass in the Croatian social space, i.e. those most deprived of capitals. It is most strongly characterized by the oldest age group, with only primary education and the lowest income. Overrepresented categories in this cluster are all related to low volumes of capital, for instance, incomplete primary education of parents, less than ten books at home, low values of network diversity, which is in line with previous research which has shown that network variability diminishes with age (Cepić and Tonković 2020; McDonald and Mair 2010). In comparison to the other classes, a higher proportion of individuals perceive themselves as members of the lower classes (26.2%), while overrepresented occupational groups are lower status employees and skilled industrial employees. When we consider other socio-demographic indicators, there is an overrepresentation of women and inhabitants of smallest rural settlements.

Table 5 presents selected indicators of cultural, economic and social capital, as well as gender, age and size of settlement, across the seven ESeG occupational groups. Expectedly, managers and professionals have the highest volume of capital, whereas skilled industrial employees and lower status employees the lowest. On average, managers, and professionals in particular, are better educated, have more books at home, have savings and are more likely to live in larger cities in comparison to the other occupational groups.¹⁰ On the other hand, lower status employees report the lowest income and lower values of overall network diversity compared to members of other occupational groups. It is interesting to note that among middle-class occupations, small entrepreneurs stand out as, on average, having more expensive real estate and higher network diversity than any other occupational group.

A cautious comparison of the capitals approach and the occupation-based approach in terms of the big picture of the class structure is presented in Table 6 ¹¹ and shows a similar distribution.

Whereas certain authors have advocated for a Bourdieusian approach to social class and have decried the Goldthorpean occupation-based approach as ‘reductionist’, ‘minimalist’ and ‘economistic’ (Flemmen et al.2017), other authors have concluded that a capitals-based approach has not provided them with strong empirical evidence that would dissuade them from using existing occupation-based measures (Connely et al.2016). Contrary to authors who have focussed on the tensions between approaches to class analysis, our results show that class analysis focusing on occupation and a Bourdieusian approach focusing on capitals align with each other when it comes to identifying the ‘big picture’ of class in Croatian society: thirty years after Croatia started its transition from a socialist state, which was part of the Socialist Federal Republic of Yugoslavia, into a democracy, a market economy and a nation-state, Croatia can be described as an unequal society with a small dominant class at the top and the majority at the bottom of the social class ladder. Those at the top, frequently managers and professionals, have high volumes of capital, those at the bottom, skilled industrial employees and lower status employees, have low volumes of capital. The educational level of the parents of those in these dominant and dominated classes indicates the inter-generational reproduction of class inequalities at these two extreme ends.

Our findings on the size of high-level class groups in Croatia corroborate Hugree, Penissat and Spire's (2020) findings that the working classes are overrepresented in peripheral eastern Europe, which they explain with the presence of industrial and agricultural sectors in this region. In the Croatian case, however, the working class is predominantly made up of workers in the service industry and more specifically tourism. Our analysis also shows managers and professionals have similar capital profiles and that small entrepreneurs stand out in terms of their high volume of economic capital. We have also found that there is a distinction in terms of volume of capital between the older and younger middle class, as well as the older and younger working class. Their capital profiles suggest that the new generations have somewhat higher volumes of capital than the older generations. Contributing to this are educational trends (the younger generations tend to stay in education longer) as well as technological advances and in particular the use of social media.

In his criticism of Savage et al.'s (2013) Bourdieusian class schema for Britain, Mills (2014) points out that inventing a social class typology for its own sake, without showing what it can explain, is pointless. In a way the same critique can be applied to the analyses presented in this article since we do not use the two class schemas to explain any particular outcome. Comparing the two approaches in such terms is our next task. However, we have examined how a Bourdieusian and occupation-based approach align when it comes to how they map onto class self-identification and we have found that ‘objective’ class location and class self-identification overlap, irrespective of the class approach one takes. Although we identified a middle-class self-identification bias in our sample, we also found that professionals, i.e. those with the highest volume of capital, are also most likely to identify as upper middle class when compared to the other classes. On the other hand, those with low volumes of capital, lower status employees and skilled industrial employees or the working classes, tend to identify as working class. Although Savage et al. (2001) have suggested that ‘sociologists should not assume that there is any necessary significance in how respondents defined their class identity in surveys’ (2001, p. 875), it is nevertheless noteworthy that in a post-socialist context such as Croatia, where class is not a politicised identity, this association exists.

So far we have focussed on the convergence of a capitals and occupation-based approach to social class, however the two approaches also differ: although they appear to align when it comes to empirically identifying Croatia's ‘big classes’, their theoretical underpinnings, namely Bourdieu's theory of practice and a more pragmatist approach, are incompatible. A strength of an occupation-based approach compared to our Bourdieusian-inspired analysis is that it is relatively straightforward to operationalize for empirical purposes and can be and has been productively used for comparative purposes. On the other hand, when it comes to obtaining a more context-specific and nuanced portrayal of social class distinctions, and in particular identifying those most dispossessed in society, a key merit of a Bourdieusian approach stands out. Whereas an occupation-based approach such as ESeG identifies lower status employees as those at the bottom of the class hierarchy, a Bourdieusian perspective is more nuanced and informative. The underclass largely consists of the elderly members of Croatian society, with only primary education and insufficient state pensions, living in rural areas, who struggle to make ends meet. In addition, the majority of this most vulnerable group are women.

Finally, over the years class analysis has had to weather much dispute and critique both from those unsympathetic to this research agenda generally as well as those conducting social class research from different theoretical angles. With reference to the latter, we have found that, although developed at different points in time and in different contexts, both a Bourdieusian approach and an occupation-based approach such as ESeG yield similar results when it comes to the size of ‘big’ classes in contemporary Croatian society. The triangulation approach we have taken in this article and its resulting findings suggest, therefore, that we should not make sharp distinctions between capitals and occupations in the first instance as we attempt a panoramic view of contemporary inequalities.

We would like to thank Dražen Cepić, Dragan Bagić, Teo Matković, Jeremy Francis Walton, Vjeran Katunarić and the whole STRAT team for their insightful comments and suggestions. Any shortcomings are of course solely our own.

No potential conflict of interest was reported by the author(s).

Željka Tonković is Assistant Professor at the Department of Sociology, University of Zadar. Her research interests include sociology of culture, urban sociology, social inequalities and network analysis. She has researched widely in the field of cultural consumption, with particular reference to the intergenerational transmission of cultural capital.

Karin Doolan is Associate Professor at the Department of Sociology, University of Zadar. She received her PhD in the sociology of higher education from the University of Cambridge. Her research interests include class inequalities with a particular focus on how they are reinforced in educational settings and communities affected by disasters.