Response to Bradley, Mills, Dorling and Rollock.
Authors Note: This paper has now been accepted for publication by Sociology. We have taken the opportunity of making minor corrections, and clarifying, the latent class analysis.
On social class, anno 2014
Mike Savage, Fiona Devine, Niall Cunningham, Sam Friedman, Daniel Laurison, Andrew Miles, Helene Snee, Mark Taylor.
This paper responds to the critical reception of the arguments made about social class in Savage et al (2013). It emphasises the need to disentangle different strands of debate so as not to conflate four separate issues, (a) the value of the seven class model proposed; (b) the potential of the large web survey – the Great British Class Survey (GBCS) for future research; (c) the value of Bourdieusian perspectives for re-energising class analysis, and (d) the academic and public reception to the GBCS itself. We argue that in order to do justice to its full potential, we need a concept of class which does not reduce it to a technical measure of a single variable and which recognises how multiple axes of inequality can crystallise as social classes. Whilst recognising the limitations of what we are able to claim on the basis of the GfK/GBCS, we argue that the seven classes defined in Savage et al (2013) have sociological resonance in pointing to the need to move away from a focus on class boundaries at the middle reaches of the class structure towards an analysis of the power of elite formation.
Keywords: social class, elites, age, Great British Class Survey
On social class, anno 2014.
At the end of our paper reporting our analysis of the BBC’s Great British Class Survey (Savage et al 2013: 244), we concluded
‘We hope that our new model of class will prove a valuable resource for future social researchers in exploring the complex and multi-dimensional nature of social class inequality in the UK in a way which permits us to recognise the ongoing salience of social class divisions in the stratification of British society’.
Have our hopes been realised? Writing less than a year after our paper was published, it is premature to answer, but the interim answer can only be highly ambivalent. On the one hand, the breath-taking scale of interest in the paper, mainly a product of the BBC’s press campaign, revealed a public as well as academic fascination for the topic which far exceeded our expectations and demonstrated an astonishing hunger for talking about class. Within academic debate, as marked by the four critical papers published in Sociology, a signal achievement of which we are proud is the way we have taken the study of class out of different intellectual silos and to have facilitated engagement from different traditions of research.
On the other hand, much of this academic reception has been highly critical. The scale of this criticism varies from the outright hostility of Mills to the more nuanced scepticism of Bradley and the ruminations of Dorling and Rollock. These commentators (as well as others) have raised serious questions about the GBCS as a data source, our mode of analysis, and our conceptualisation of class. To date, few academics (though see Payne 2013 as a partial exception) have seen our seven fold class model as a useful tool for sociological analysis, and in general most academic critics have defended their prior conceptualisations.
How do we make sense of this ambivalent reception? In order to make a considered reply it important to distinguish at least four different issues which are present in our paper but which need analytical separation, as we think that our commentators don’t always realise that a criticism of one does not entail a criticism of others.
Firstly and most obviously are the claims we made about the value of distinguishing ‘our’ seven classes, derived from the latent class analysis of the national GfK data and fleshed out using the GBCS. These were the main focus of our paper, and certainly the centrepiece of the BBC’s press campaign. Whether these seven classes are analytically useful forms the focus of Bradley’s reflections, whilst Mills expresses doubts about the point of developing this classification in the first place (since the NS-SEC is a well-tested and validated measure of class already).
These seven classes, however, are also the lightning conductor for broader issues. Secondly, we need to distinguish the seven classes from the potential of an unorthodox data set – the BBC’s Great British Class Survey (GBCS). As Mills emphasises (and as we made amply clear in Savage et al 2013), our seven classes are very largely derived from the small national GfK survey. Even if our seven classes are found unconvincing, the GBCS data set, which should shortly be in the public realm, might still be a remarkable resource which any researcher will be able to use. We should also note here that our paper was only the first intervention in what we expect to be a much more extensive string of publications based on it. Thus, we have just submitted a batch of articles to Sociological Review conducting more detailed studies of the GBCS, focusing especially on the elite, where the sample skew aids ‘granular’ analysis. As part of this exercise we are also examining the micro-skews in the GBCS (towards particular occupational groups, university graduates, etc). In short, even if our original paper is found problematic, it would be premature to write off the potential of the data itself. We need to play a longer game….
Thirdly, we need to further distinguish what might be termed Bourdieusian, (or ‘capitals, assets, resources’ – see Savage et al 2005) approaches to class analysis from the specific elaboration of the seven classes in Savage et al (2013). The past decade has seen a striking elaboration of these perspectives, sometimes called ‘cultural class analysis’ which seek to make cultural issues central to the analysis of class. These studies include investigations of the cultural aspects of class reproduction (e.g. Devine 2004; Scherger and Savage 2010), of the relationship between class and identity (Savage et al 2005; Savage 2010), as well as conceptualisations of class itself (Savage et al 2005; Devine et al 2004; Bennett et al 2009; Le Roux et al 2008; Atkinson 2010; Atkinson et al 2013; Flemmen 2013). One of the leading aspects of this work is the elaboration of a ‘social space’ approach to class, where geometric methods are used to derive social class groups using clustering methods. Whilst sympathetic to this approach, we actually used latent class analysis in Savage et al (2013). Our point is therefore that the power of Bourdieusian perspectives in general should not be conflated with the specific arguments which we made in Savage et al (2013), and we note that Will Atkinson, is critical of our arguments from a Bourdieusian perspective. This having been said, it is striking that none of the response papers make any reference to the value of the ‘capitals, assets resources’ approach which informed our work. In particular, many of Mills’ questions about how we see the nature and scope of class analysis have been amply outlined in this literature which he does not address (even though much of it is cited in our reference list).
Fourthly, and finally, we also need to observe that the passions sparked off by our paper can only be understood in the context of the huge media interest in the story itself. After all, our previous writings which also include critical reflections on the problems of NS-SEC class categorisations (notably Le Roux et al 2008) have not generated such critical interest. What the media interest in the GBCS has bought to the fore is a contestation over the ‘politics of authorisation’. What is now at stake is the monopoly of the NS-SEC, institutionalised through the Office of National Statistics, to be the only public measure of class. We take this to be one of the reasons for the aggressive tone which Mills adopts in his response, which does not appear to be explicable as a reaction to the tone of our paper which is respectful to different parties, including the proponents of NS-SEC, throughout. In short, this is not simply an ‘academic’ debate, but is itself testimony to the charged nature of class analysis itself in 2014. We have moved well away from the idea that ‘class is dead’. In fact, the sociological analysis of class is now central to public debate. This is a fact which we should celebrate. It is our view too, that having different ways of conceptualising and measuring class – and a recognition of what they can and can’t do – can only aid this welcome development.
We hope we have emphasised that there is more to our paper than simply the ‘seven classes’ which has been the focus of the critical response. In order to recognise the importance of this widening of the debate, we firstly return to the general issue of what classes actually are. This leads us on to our second section on how classes need to be linked to an analysis of inequality. Having conducted this ground work, our third section reflects on how we operationalised the concept of capitals in our paper in order to explicate how our measures of class were derived. The fourth section responds to the challenges about our data and our methods of analysis, addressing both the second issues above. It is only in our fifth and last section that we turn to the specific question of the possible sociological significance of our seven classes.
1: What is class?
It is hardly original to note that class is a contested concept (Calvert 1973; Crompton 2008). Indeed we see the difficulty of domesticating the concept of class to any one paradigm as one of its defining and attractive features. Class spans public and academic discourse, raises issues of politics and science, and is brutally contested within academic paradigms (see e.g. Crompton 2008; Wright 2005). Good! This unruliness is in our view highly productive, being indicative of the telling power of the class concept itself to challenge hegemonic modes of academic expertise and to facilitate a genuinely public sociology in which the expertise of sociologists is itself a matter of public concern. In unravelling this contested approach to class we can identify three different ways of delineating class, all of which are evident in the reception to our study.
Firstly, and amply marked in Mills’ paper, is the emphasis on (i) class as a ‘discrete’ variable, which needs to be delineated and differentiated from any other property with which it might be contingently affiliated. This endeavour to define class as a unitary variable can then lead to a broader project of empirically assessing its significance for other outcomes through using various kinds of multivariate model. It follows that for this perspective, class needs to be differentiated from anything else with which it might be associated (for instance, status, gender, age, ethnicity, residential location, or whatever) so that it is stripped bare as a unitary phenomenon and its net significance registered. Much of Mills’ hostility to our paper appears to be – no doubt deeply and genuinely held – bafflement that class could be anything other than a discrete, validated variable of this kind.
However, as we thought we made clear, our preferred definition of class is different to this. We are seeking a measure of class as (ii) class formation. Here, the crystallisation of different properties renders a class as having a social existence over and above the different factors which make it up. It is in this sense that historians have been interested in classes, not as ‘pure’ variables stripped of contaminants, but as distinctive social formations. In Edward Thompson’s (1963) famous formulation, the English working class was a crystallisation of the cultural traditions of the ‘freeborn Englishmen’, handicraft skills, experiences of state repression and so forth. Subsequent historians have also pointed to other features which Thompson largely ignored such as the masculinist gender cultures (Clarke 1997). In more recent sociology, this perspective has been instantiated by feminist research showing how class identities are simultaneously deeply gendered (Skeggs 1997; 2004), and by arguments by scholars of race and ethnicity that race is ‘“the modality in which class is ‘lived,’ the medium through which class relations are experienced, the form in which it is appropriated and ‘fought through’” (Hall 342). Rollock’s discussion is an exemplary reflection on how ethnic and class identities are intimately interwoven in this spirit. It follows from our preferred definition of class that our model is a heuristic one which is designed to shed light on the nature and significance of class boundaries in Britain today. The seven classes are not ontological or structural entities, but are the product of the interplay of the three different capitals which comprise them. They can thus be seen as the outcome or the effects of these capitals.
Finally, it is also clear that there is a third significant meaning of class, which has come to the fore in the recent debate. Here class is (iii) an ideological contradiction of democratic capitalism. Intense public interest in class resonates with a deep tension between the supposed inclusive and egaliatarian ethos of democratic society on the one hand, set against the stark – and accentuating – inequalities of capitalism on the other. Dorling’s paper valuably explores some of these contradictory dimensions of class in this vein. Here, admitting that classes exist as discrete social groups is offensive to the public’s deeply held egaliatarianism, to believing that ‘everyone is as good as anyone else’ and to resisting snobbish and elitist motifs (on which see Savage et al 2001; 2008; Warde 2012). As Warde and others have shown, it is now deeply unpopular for people to openly deport themselves as if they are superior to others. Yet just as this populist and democratic motif symbolically predominates, contemporary capitalism simultaneously generates massive divisions and economic differentiation, and fundamental inequalities of life chances. Class, therefore, taps a nerve, as an inclusive democratic sensibility confronts the mundane existence of inequality. Therefore, the recognition of class acts as a lightening conductor of profound, though often unacknowledged, cultural dynamics which simultaneously resist the idea of there being ‘classes’, whilst being highly attuned to processes of ‘classing’ and ‘classification’.
It is, we would argue, this third sense which explains why people resist identifying as member of a specific class, at the same time that they are aware of, and often fascinated by, the project of classification. And it is this which explains the kinds of ambivalences around classification which are illuminatingly discussed by Rollock and Dorling. The public response to the GBCS itself is testimony to the way that many people are intrigued by understanding how class operates, at the very same time that they resist the idea that there are in fact distinctive social classes in which people can be categorised. Therefore, whatever model of class is produced, it would be resisted and cause offence. We don’t have the possibility of defining a class schema which would meet public approbation and assent. This is, paradoxically, because of the fundamental centrality of inequalities in the production of class.
2: Class and inequality
This discussion directs us, willy nilly, to the crucial relationship between class and inequality. For, in all three senses above, classes can only exist in relation to other classes, and it is the structural asymmetry between them which is central to understanding them as classes, rather than simply as groups or categories. It is precisely because of this intersection with inequality that the concept of class is different from that of market research typologies, of forms of segmentation, which Mills briefly mentions. As he correctly points out, one can always produce typologies out of a complex data set, but for us to see such typologies as classes, an additional step is needed. But the way that classes are seen to relate to each other is a matter of contestation. Bradley accuses us of having a ‘gradational’ approach to class and argues for a relational perspective. Writers, such as Mills who support a measurement of class associated with the NS-SEC generally see class as a categorical variable, in which different classes have different qualities or properties, here defined by the nature of people’s employment relationships, notably by differentiating employers and employees, and those on wage contract or a more diffuse service relationship, see Goldthorpe 2007).
The fundamental question here how class is related to inequality? None of our critics address this issue head on, though Dorling helpfully recognises its centrality. In this section we want to spell out further our own attempt to provide a more adequate resolution to this question, though we are aware that there is plenty more work to do. Let us pose the issues as starkly and directly as we can, recapping on earlier arguments (Savage 2000; Savage et al 2005).
Historically, the most direct approach to linking class to inequality was through the Marxist theory of exploitation, in which one class is held to derive its privileges by exploiting another. This is presumably how Bradley is invoking the idea of relationality in her comment. The concept of exploitation which underpinned Marx’s theory of class depended on a labour theory of value in which one class was held to derive its resources by systematically appropriating the surplus value from the working class, who did not receive in wages the equivalent value to that which they had spent in producing the commodities for their employers. The labour theory of value was therefore a neat way of linking the theory of capitalism to a theory of class. However, the problem is that the labour theory of value has largely been superseded by economists and even if it appears attractive at an abstract (as well as a rhetorical) level, it is hard to operationalise concretely (see generally, Sorensen 2000; Wright 2000).
The problem with using the labour theory of value as an underpinning for theories of exploitation (and therefore of class) have caused class researchers to move in two different directions. On the one hand, Erik Olin Wright (1985) sought to place the concept of exploitation on a different footing through defining it in game theoretical terms (see the discussion in Sorensen 2000; Savage et al 2005). However, this approach has stalled because game theoretical terms allow numerous axes of exploitation to be defined and followed to their logical conclusion provide the concept of class with no specificity as the concept becomes tautologous. Any difference in pay or conditions could be seen as due to the exploitation of the more deprived by the more advantaged.
The alternative path was pursued by Goldthorpe who sought to pull the concept of class clearly apart from any reference to exploitation at all (see e.g. Goldthorpe and Marshall 1992; Goldthorpe 2000). Increasingly indebted to economists’ analysis of the nature of labour and employment contracts as means of monitoring their workers, he specifically insists that there are no ‘zero-sum’ conflicts between classes. He sees class as associated with employment relations without being committed to the view that different classes are necessarily structurally in conflict. A considerable programme of research shows that his measures of class do indeed map on to significant differences in the nature of employment, such as job security, the nature of remuneration, the regulation of work (e.g. McGovern et al 2007).
This approach deals with the theoretical uncertainty over the concept of exploitation but at the disabling cost of disarming the concept of class itself. In this formulation, there is no analytical added value in defining these categories as ‘classes’ rather than (for instance) as ‘groups with similar kinds of employment contracts’. Using the label of ‘class’ to define such groups is entirely arbitrary. And indeed, when institutionalised into the National Statistics Socio-Economic Classification the term ‘class’ was dropped altogether, apparently with no analytical loss of any kind. Thus, the cost of sundering the concept of class from some kind of theory of inequality is to remove the fundamental point of the concept altogether. Here, we agree with Bradley’s emphasis on the need for the concept to be relational.
Approaches to class therefore need to be anchored in a theory of dis/advantage, our version of which was articulated in Savage et al (2005) but can be re-stated here. Given the problem of operationalising the labour theory of value, we prefer to focus on mechanisms of accumulation as lying at the heart of a relational approach to class. It is the potential of some assets to augment, store, transmit, and convert advantages which is central to the operation of class. Those without such assets are thereby limited relative to those with them. This formulation hence avoids a zero sum conception of class exploitation (where one class gains directly at the expense of another) whilst also endorsing a relational contest in which some groups have unusually marked opportunities to accumulate and hence gain increasing advantages over those who do not.
Let us build on this point to draw out some implications. Firstly, we dispute the view of Flemmen (2013) that our perspective removes the study of class from that of employment or the labour market. Processes of accumulation are hugely dependent on the organisation of the capitalist economy, albeit in ways which vary according to specific organisational contexts. Following the lead of Sorensen (2000) and Grusky and Weeden (2001; 2008), we regard the most promising way of recognising this point that accumulation takes the form of the sequestration of rents, so that certain occupations or employment situations have the potential to allow their incumbents to have future rewards and expectations inbuilt.
However, secondly, we also think that there are other mechanisms of accumulation other than those arising from the labour market alone. Bourdieu’s concept of economic capital usefully broadens out our understanding to incorporate other forms of wealth accumulation, for instance income from savings, investments, housing and the like. In our view, it is quite conceivable that someone who has never been in paid employment, but who has the capacity to draw upon sources of capital such as this can be seen as highly privileged in class terms.
Thirdly, we further follow Bourdieu in seeking sources of accumulating advantage other than those of economic capital. Here, his concepts of cultural and social capital allow those people with certain cultural dispositions and capacities, and with certain kinds of social networks, the potential to accumulate and acquire – for instance educational qualifications, information, skills, etc. Indeed, it is possible that in an increasingly symbolic economy, such kinds of cultural and social mechanisms of accumulation become even more significant.
3: Defining and measuring capitals
According to Mills, and to other critics such as Goldthorpe (2013), our findings are simply a ‘data dredging exercise’. Now, as we emphasised in our paper, it is definitely the case that our analysis is as good as the construction of the variables, and hence ‘rubbish in, rubbish out’ definitely applies. Because our latent class analysis has established seven classes out of the mix of measures which were used to construct them, it does not follow that we have defined seven ‘formed’ classes. For this to be the case, we need to reflect on whether they appear to make sociological sense, whether they might be identifying a group which potentially has some coherence. Mills is of the view that the classes are partly an artefact of our variable construction. In our defence, let us firstly explain why we think our measures of capital are sociologically robust, and then go onto consider the plausibility of the kind of class groupings that the latent class analysis produces.
We think we are quite clear about how we derived our measures of capital in our original paper. Whilst acknowledging that they are not the only way that capitals might be operationalised, we re-iterate that there is extensive prior thinking and research which went into their construction. Our questions of social capital need to be placed in the context of the evaluation and critique of Putnam’s (2000) work over the past decade. Putnam emphasised the significance of membership of voluntary associations for the generation of social capital, but also recognised the role of informal ties, and this has become increasingly recognised in later research (see for instance, Li et al 2007; 2008). This recognition has led to interests in finding a mechanism to measure the nature of people’s social networks using questions on a sample survey. Questions on specific ‘best friends’ or the like, which are used in Understanding Society do not tap the ‘weak ties’ that Granovetter (1973) famously emphasised. The use of the position generator question in the GBCS follows the model which was used in Culture, Class Distinction and has been proven to be effective by Li et al (2008). This approach is drawn from Nan Lin (Lin et al 2001). The idea is that by asking how many people someone knows of a particular occupation it is possible to get a summary measure of the range of connections they possess. By measuring each of these occupations on the Cambridge score, we also found a way measuring whether respondents knew from predominantly high or low status occupations.
Similarly, our thinking about cultural capital is informed by extensive previous research, notably that reported in Culture, Class, Distinction (Bennett et al 2008) and we are surprised that few of our critics, notably Mills, seems to have sought out this book to inspect more fully the underpinning of our thinking. We briefly repeat some especially salient points.
Firstly, whilst obviously recognising that questions on cultural tastes and practices entail measurement error, we still think that they can be used to identify ‘objectified cultural capital’ in a useful way. We recognise Bradley’s point that it is difficult to recognise some informal practices through structured survey instruments, but because the MCA operates relationally, this does not affect the finding that certain practices are relatively associated with specific social groups. Thus, we are not claiming that those without cultural capital in our senses are in some ways un-cultured, we are only claiming that there are some practices which are systematically associated with the advantaged which might allow them to be construed as cultural capital. Mills’ argument is that the our questions on cultural practices (for instance, a taste for certain kinds of music) conflates age and class, but he seems to assert an almost ‘naturalist’ view that being young or old necessarily imparts a pre-disposition to certain cultural practices.
Secondly, Mills claims that the cultural tastes and practices revealed in our Figures 1 and 2 are actually the product of the NS-SEC class divisions which we are supposed to ‘disdain’. We would never deny – and indeed have ourselves argued that – NS-SEC classes are associated with these patterns (as are income, educational qualifications, and other indicators of social hierarchy, see Bennett et al 2008). However, for Mills to make the claim that NS-SEC is ‘predictive’ of these cultural patterns, he needs to engage with the extensive research which we cited in our paper (such as by Chan, Goldthorpe, as well as by ourselves) which argues otherwise. Our own comment which he cites was a discussion of these other sources.
Thirdly, we see the link between age differences and class as a positive merit in our approach, rather than a conflation, as Mills argues. As a side note, in many nations recognition of fundamental cleavages of generation is widely recognised, nowhere better than in France where Louis Chauvel (2006) has emphasised the difficulties of the younger generation compared to their parents. In Britain, class analysis has typically abstracted age from class, seeing these as independent and separate issues. This is despite the fact that class motifs – from yuppie to chav – typically depend on idioms of age. The result of this analytical separation of class from age is that generational divisions have been subsumed in Britain into an anxiety about ‘declining social mobility’, which acts as a proxy for worries about the prospects of the younger generation. In our terms, bringing age more directly into the understanding of class is to be welcomed.
This having been said, it is clearly important to distinguish age, cohort, or generational effects at work. It is unclear whether the younger will become more ‘highbrow’ as they get older. Previous qualitative research suggests that in fact canonical musical forms may be ageing out, and we are struck by the way that these canonical cultural forms, although they are referred to by young professionals, do not convey the intensity and passion of contemporary culture (see Savage and Gayo 2012, and more generally Prieur and Savage 2011). It was from previous research and reflection on these issues across a range of European sites that the concept of ‘emerging cultural capital’ was elaborated (Prieur and Savage 2012). We can also note that we do not regard cultural
It follows that we do not view the association between age and class as a defect of our analysis, indeed the reverse. We do not find it surprising that, or unimportant, that older people might have more expensive houses (and hence more economic capital) than do younger (and Dorling discusses this with respect to house prices, for instance. But it is quite erroneous to see our classes as simply the product of age divisions, as in Mills claim that ‘(L)ife cycle plays a role in distinguishing what Savage et al. term the ‘elite’ and the ‘established middle class’. Because the elite are eleven years older than the established middle classes, he thinks this will explain the superior economic capital which the older ‘naturally’ accrue. But in fact, at least as far as household income is concerned, Goldthorpe and McKnight (2004) show that 57 year olds in class 1 and 2 are actually likely to have marginally lower income than 46 year olds.
4: Data and method
Let us now turn to the vital question of our data and modes of analysis, which is the basis of Mills’ important critique. Mills protests that we actually used the small national GfK sample to derive our classes, and hence that the large GBCS plays no analytical role and is largely therefore, a red herring. He doubts that the GBCS can be used for any substantial sociological purpose. In fact, what we did in Savage et al (2013) was to use the GBCS to explore some of the sociological contours of the seven classes which were indeed derived from the national GfK sample. However, our linking of the two data sets was central to our purpose and ten pages of our paper (pp 233-243) examines each of the classes in turn, explicating their characteristics through drawing on material drawn from the GBCS. This allowed us to reflect on whether the latent classes derived from the GfK are actually sociologically meaningful, and hence is a crucial part of our analysis.
There is no doubt that the GBCS is an unorthodox data set which is not nationally representative. The issue is what follows from this? Do we refuse to have anything to do with data which departs from the ‘gold standard’ of standard large scale nationally representative data sets, or do we try to make the best of what we have and explore using innovative methods to deploy it to its best advantage (notably through its potential granularity because of its large sample size). Our view on this is clear. We do not think it is wise for social scientists to limit their research repertoires at the moment which digitalisation is throwing up a proliferation of possibilities and challenges (see Savage and Burrows 2007; Law et al 2013). We specifically identified our work as part of an ‘experiment’ and we hope that the results will be instructive, even if the GBCS is seen, ten or twenty years down the road, as a white elephant.
Mills asks for more information about the national GfK sample in order for us to judge the extent to which it is nationally representative. This is provided in Appendix 1. In brief, GfK’s quota samples seek to ensure representativeness by controlling for age and sex, and by using geographical location of the interviews as a proxy for social class. Their data included a weight variable which they included to ensure that the sample was representative in their terms (and which we used in our analysis). Following the increased scrutiny of the sample in April we did additional tests (which we report in Appendix 1) which show that the GfK appears representative using NS-SEC measures, as well as those controlling for demographic characteristics and educational qualifications. Given its limits as a relatively small quota sample, we are therefore confident that the GfK is a reasonable nationally representative source.
5: The sociological resonance of our seven classes
It is a striking point that none of the four responses, with the partial exception of Bradley, reflect on whether the seven classes we delineated might have be sociologically resonant or not. We need to make it clear that the latent class analysis does not provide ontological guarantees that the seven classes exist as social formations. They give us a set of patterns, and our challenge was to consider, through sociological reflection, whether the different classes were simply a ‘dogs dinner’ spat out by the LCA. We see this interpretative aspect as crucial to our work, and we remain hopeful that the seven classes offer revealing insights into the fault lines of contemporary British society. Let us recap on what we think the main sociological implications might be.
Probably the most important implication of our work is that the fundamental centrality of the divide between the ‘middle’ and ‘working’ class which has underpinned class analysis since its foundation in the thinking of Marx and Weber needs to be reviewed. Class analysis has developed through a pre-occupation with this ‘collar line’. This obsession was historically marked in myriad ways: the difference between ‘staff’ and ‘line’ between salary and wage, between manual and mental labour and between blue and white collar, allied to the role of gender and ethnicity in articulating these divisions. These debates about the boundary between middle and working class were underscored by the debate between socialists and reformists about the political potential of the industrial working class in contemporary capitalism, which traces a lineage through EP Thompson (1963), TH Marshall (1951), the industrial sociology of John Goldthorpe and David Lockwood (1968/69), and then into cultural studies through Richard Hoggart (1956), Raymond Williams (1961), Paul Willis (1977), Bev Skeggs (1997) and beyond.
Alongside this classic tradition focusing on ‘the problematic of the proletariat’, and juxtaposed to it, lies a second kind of mobilisation of class. Here the middle classes were held to be the backbone of society. Dror Wahrman (2002) has traced this motif back to the 18th century, and as Ross McKibbin and Raphael Samuel have shown how it was then adapted during the 20th century, notably by Conservative politicians seeking to define the middle classes as bastions of national virtue against the dangerous working class. In the post war years it gained further impetus through being injected with a technocratic emphasis on the need for skilled and qualified ‘human capital (see further, Savage 2010).
From the middle years of the 20th century further anxieties were forged around gender, ethnicity and immigration. The male preserves of middle class work were increasingly penetrated by women. But perhaps these women were a ‘white collar proletariat’, leaving the male middle classes in privileged managerial positions (Crompton and Jones 1984; Savage and Witz 1992)? Increasing amounts of immigration caused further anxieties regarding the racialisation of fractions of both middle and working classes, marked in discussions of the putative ‘underclass’ as well as the possible emergence of a ‘black middle class’.
The effect of these anxieties was to focus on the middle reaches of society as the main arena of social concern and boundary drawing. This is nowhere marked so much as in the extensive use of the peculiarly British interest in the ‘lower middle class’ – on the one hand, differentiated from the world of manual labour, and on the other hand, not fully ‘middle class’.
Our analysis has interesting ramifications for this traditional problematic. To be sure, we have ample evidence that very significant divisions within the middle reaches of society can still be found, for instance in the contrast between our ‘established middle class’ and the ‘traditional working class’. However, our analysis suggests these are no longer the fundamental cleavages in British society. Previous models of class, with their concern over the boundaries between middle and working class are supplanted by three other dynamics which the latent class analysis indicate These are (a) the role of the outliers and especially those at the ‘top end’ of the class structure, (b) boundaries of age and generation, and (c) the redefinition of expertise and technique. Let us address these in turn.
On the first point, one of our most striking findings is the delineation of an ‘elite’. Here we are pleased that Bradley also wishes to re-introduce this ‘upper class’ back into class analysis. It is bemusing that Marxist critics of our work abound, given that our account has more affinities to a Marxist focus on the bourgeoisie than other sociological models of class. If one has to detect the most important cleavage in Britain today, it is not between ‘middle’ and ‘working’ class, but between a relatively small corporate (or ‘professional-executive’) elite and everybody else. Over the past thirty years British social science has hived off the study of social class (done mainly by sociology) from the study of elites (done mainly by political scientists). One struggles to read any sustained studies of the social composition of small elites within sociology even though it is clear that their relative income and wealth has increased dramatically. This is why we think that our elaboration of a very wealthy elite at the apex of the class structure is so important. We preferred the label elite to that of the upper class (e.g. Scott 1982) for two main reasons. Firstly, because reference to an upper class conjures up images of the traditional landed gentlemen and senior professionals in their country estates and Mayfair clubs. But this is not the elite which we reveal, which is fundamentally a senior corporate managerial group. Nowhere is the impact of neo-liberal restructuring so apparent as in the power and extreme relative wealth of this small group. The second reason for choosing this term is precisely to strategically recognise the intersection of politics and economic position, to align the terrain of political science and management with that of sociology.
At the extreme is the ‘precariat’. We borrowed this term from the academic Guy Standing (2008) who has done so much to promote the role of the ‘precarious proletariat’ at the bottom of contemporary societies. This term was deliberately used in place of the more conventional ‘underclass’ label which has been used to stigmatise the poor and deprived for decades. And our ‘precariat’ does not fit many of the stereotypes. Our GBCS mapping of its distribution indicates that it is not particularly associated with urban locations, and indeed its location indicates relatively high amounts of rural and suburban poverty. Our model therefore disrupts conventional moralisation of urban poverty and points towards a more complex and systemic picture of social exclusion at the lower reaches.
Next, we can see some significant age differentiated classes, especially the ‘new affluent workers’ and the ‘emergent service workers’. These are both relatively ‘young’ classes, who stand in contrast to other classes with similar economic resources (the established middle classes, and the traditional working class), respectively. What this recognition points toward is the inter-twining of age into class which suggest a very different problematic is now shaping our debates about British society than the older concerns about the collar line. This is the question of historical reference. A wealth of research, including some done by ourselves, indicates that the historical cannon is no longer constitutive of cultural excellence, moral certainty, or tradition. The avant garde, which used to define itself vis a vis the historical canon on which it depends, has been replaced by the themed and fashioned trends with no historical reference points, where the new and contemporary are held to be automatically the marker of excellence. This tension is amply demonstrated in our differentiation between ‘highbrow’ and ‘emergent’ cultural capital, but this only draws on extensive studies within cultural sociology which underscore the power of this divide. Our classes are fractured around this generational politics in a way which is thoroughly appreciated by the Emergent Service Workers party, or by Vice for instance.
Our final tension is that of expertise itself. We have already alluded to this in our earlier comments about classification. One of our classes, ‘the technical middle class’ appears different from the more traditional model of middle class life, oriented towards cultural activity and with extensive social ties. It is, instead a group with restricted social range and limited cultural interests, with tendencies to work in technical occupations and have scientific interests. This ‘technical middle class’ has attracted increasing interest from historians in recent years. Mobilised around a range of technical interventions from new weaponry, research methods, and of course, information technology, this is a group who hardly fit old gentlemanly paradigms (Savage 2010).
This discussion has centred on unravelling the fault lines and points of anxiety and dispute which our model reveals – and suggests how in 2014, there are several sources of ‘classificatory anxiety’. These move us away from classes as ‘variables’ to the fundamentally more sociological issue of ‘class formation’. Here, our arguments can be crisply brought out. The ‘problematic of the proletariat’ has run its course. Although anxieties about crossing from working into middle class positions continue to abound, the fundamental structural division which emerges from our analysis separates out a powerful corporate class against all the other classes. Within the middle reaches of the class structure, age and expertise are major modes of differentiation and contestation. The seven classes which we delineate were elaborated as a means of analytically drawing attention to new fractures and ambivalences which we face today. For, in early 21st century Britain, the politics of class reaches a new moment with the clear partitioning of a powerful and wealthy elite from other classes, and the compounding and fracturing effects of age and expertise, alongside gender and ethnicity, indicate the problems of any straightforward project of ‘working class’, or even ‘popular’ politics.
Let us conclude by noting initially that our critics have valuably pointed to important issues in our analysis reported in Savage et al (2014). We fully subscribe to the principles of social scientific analysis which involves re-testing (using similar or indeed alternative methods) on the same and different data sets. Social science advances through careful inspection, critique and re-specification. If it turns out that further analysis qualifies or indeed completely repudiates our depiction of the seven classes we have distinguished here, then so be it.
We remain confident that it is useful to reflect anew on the nature of class formation today in ways which may challenge more conventional approaches to class. In our paper we are respectful of other ways of conceptualising class, such as enshrined in the NS-SEC, though we also point to critical concerns regarding the limits of these approaches. These arguments about the value of developing new multi-dimensional approaches to class have been made in numerous other publications and we feel raise important issues about how we might conceptualise class formation today. It is interesting – and indeed sociologically revealing – that our paper has been subject to extensive criticism which rarely engages with the reasons we put forward as to why there might be value in developing new a model of class along the lines we sketch out. Anyway, we hope this paper has clarified how our paper (Savage et al 2013) needs to be situated as part of a much wider engagement with theoretical, methodological and substantive issues for class analysis in the current period. We have argued that our paper needs to be put in a wider context of previous research, rather than treated as ‘a bolt from the blue’ Notwithstanding the criticisms which can be made of it, we believe we have the potential to develop a rich understanding of class in fast moving contemporary societies and it is important not to remain in existing intellectual silos.
We are grateful to Magne Flemmen for his comments on this paper.
Atkinson, W., (2011), Class, Individualisation and late modernity, Basingstoke, MacMillan
Atkinson, W., Roberts, S., Savage, M., (eds), (2013), Class Inequality in Austerity Britain; Power, Difference and Suffering, Basingstoke, Palgrave
Bennett, T., Savage, M., Silva, E.B., Warde, A., Gayo-Cal, M., Wright, D., (2009), Culture, Class Distinction, Routledge
Calvert, P., (1982), The Concept of Class: an historical introduction, London, Hutchinson
Chauvel, L., ‘Social Generations, Life Chances and Welfare Regime Sustainability », in Culpepper, P.D., Hall, P.A., and Palier, B., Changing France, The Politics that Markets Make, Palgrave Macmillan, Houndmills, Basingstoke, Hampshire, pp. 150-175.
Clarke, A., (1997), The Struggle for the Breeches: Gender and the Making of the British Working Class, Berkeley, University of California Press.
Crompton, R., (2008), Class and Stratification, Cambridge, Polity
Devine, F., Class Practices, Cambridge, Cambridge UP.
Devine, F., Savage, M., Crompton R., and Scott, J., Rethinking Class, Identities, Cultures and Lifestyles, Basingstoke, Palgrave, 2004,
Flemmen, M., (2013), ´Putting Bourdieu to work for class analysis: Reflections on some recent contributions´, British Journal of Sociology, Vol 64, no 2, pp. 325–343.
Goldthorpe, J.H., (2000a), On Sociology, Oxford, Oxford University Press.
Goldthorpe, J.H., (2000b), ‘Rent, Class Conflict, and Class Structure: A Commentary on Sorenson’, American Journal of Sociology, Vol. 105, No. 6, pp. 1572-1582
Goldthorpe, J.H., (2013), ‘Five minutes with John Goldthorpe’, http://blogs.lse.ac.uk/politicsandpolicy/archives/35681
Goldthorpe, J.H., and McKnight, A., (2004), ‘The economic basis of social class’, LSE: CASE Working Paper, 80.
Grusky DB and Weeden KA (2001) Decomposition without death: A research agenda for the new class analysis. Acta Sociologica 44(3): 203–18.
Grusky DB and Weeden KA (2008) Are there social classes? An empirical test of the sociologist’s favorite concept. In: Conley D and Lareau A (eds) Social Class: How Does it Work? New York: Russell Sage Foundation
Hall, S., (1980) “Race, Articulation, and Societies Structured in Dominance.” Sociological Theories: Race and Colonialism. Paris: UNESCO. 305-345.
Hanquinet, L., Savage, M., and Callier, L, (2012), Elaborating Bourdieu’s field analysis in urban studies: cultural dynamics in Brussels ‘, Urban Geography,
Law, J., Ruppert, E., Savage, M., (2013), ‘Reassembling Social Science methods: the challenge of digital devices’, Theory, Culture and Society, 30 (4), 22-46.
LeRoux, B., Rouanet, H., Savage, M., Warde, A., ‘Class and Cultural Division in the UK’, Sociology , 42, 6, November 2008, pp 1049-1071.
Li, Y., Savage, M., and Warde, A., (2008), ‘Social mobility and social capital in contemporary Britain’, British Journal of Sociology,
Lin, N., Fu, Y.-c. & Hsung, R.-m. (2001) Position Generator: A Measurement for Social Capital, in: N. Lin, K. Cook & RS Burt (Eds) Social Capital: Theory and Research, pp. 51-87 (New York, Cambridge University Press
McGovern, P., Hills, S., Mills, C., (2007), Market, class and employment, Oxford, Oxford UP
Prieur, A., and Savage, M., (2013), ‘Emerging forms of cultural capital’, European Societies, 15, (2), 246-267.
Prieur, A, and Savage, M, ‘Updating Cultural Capital Theory: a discussion based on studies in Denmark and in Britain, Poetics, 2011, (39, 6).
Savage, M., (2000), Class Analysis and Social Transformation, Milton Keynes: Open University Press.
Savage, M., (2010), Identities and Social Change in Britain since 1940: the politics of method, Oxford, Oxford University Press
Savage, M., (2012), ‘Class, Culture and Politics’, in Amenta, E., Nash, K., and Scott, A., The Wiley-Blackwell Companion to Political Sociology, p 294-304.
Savage, M., (2013), ‘The Social Life of Methods: a critical agenda’, Theory, Culture and Society, 30 (4), 3-21.
Savage, M., Bagnall, G., and Longhurst, B.J., (2001), ‘Ordinary, Ambivalent and Defensive: class identities in the North-West of England’, Sociology, 35, 4, 875-892.
Savage, M., Bagnall, G., Longhurst, B.J., (2005), Globalisation and Belonging, London, Sage,
Savage, M., Devine, F., Cunningham, N., Taylor, M., li, Y., Hjelbrekke, J., Le Roux, B., Miles, A., Friedman, S., (2013), ‘A New Model of Social Class? Findings from the BBC’s Great British Class Survey Experiment’, Sociology, 47, (2), 219-250.
Savage, M., Warde, A., Devine, F., (2005), ‘Capital, Assets and Resources; some critical issues’, British Journal of Sociology, 56, 1, 31-48
Savage, M and Gayo, M, (2011) ‘Unravelling the omnivore: a field analysis of contemporary musical taste in the United Kingdom’, Poetics, 2011, 39, 5, 337-357
Scherger, S., and Savage, M., ‘Cultural transmission, educational attainment and social mobility’, Sociological Review, August 2010, 56, 4, 406-428
Skeggs, B., (1997), Formations of Class and Gender, London, Sage
Skeggs, B., (2004), Class, Culture, Self, London, Routledge. ,
Sorensen, A. B., (2000), ‘Toward a Sounder Basis for Class Analysis’, American Journal of Sociology, Vol. 105, No. 6 (May 2000), pp. 1523-1558
Warde, A., (2012), ‘Cultural Hostility Reconsidered’, Cultural Sociology, 5, 3, 341-366
Wright, E.O., (1985), Classes, London, Verso
Wright, E.O., (2005), Approaches to Class analysis, Cambridge, Cambridge University Press.
APPENDIX 1: ADDITIONAL DETAILS ON GfK SAMPLE AND LATENT CLASS ANALYSIS
Mills (2014) understandably requests more information on the GfK sample, which we did not have the space to elaborate in Savage et al (2013). The following tables gives details of the representativeness of the GfK sample, comparing it to the nationally representative British Social Attitudes Survey, 2011 (which was conducted at the same time).
Table 1: demographic representativeness
We should note that the slight age differential is due to the fact that the GfK sampled down to age 16, whereas the BSA samples to age 18. Table 1 indicates that the quotas on age and gender have produced a representative sample in these terms.
With respect to the comparisons with the NS-SEC, we should note that the GfK survey (like the GBCS) asked for job information in free text entry. This does not make it possible for there to be exact comparison with other surveys where structured occupational categories are asked for. In some cases, the free text job is so imprecise that is difficult to pin down where it would fit in the NS-SEC formatting. We have however spent considerable time reflecting on how best to code to NS-SEC categories for every individual case within the GfK. The GfK N here is 935 because in some cases the information is not sufficient to come to a judgement. Overall, the NS-SEC distribution is reassuringly close.
Table 2: GfK and BSA: % in different NS-SEC categories
|1.1 Large employers and senior managers||1.7%||2.2%|
|1.2 Higher professionals||8.1||8.7|
|2 Lower managers||24.1||24.0|
|4 Small employers||8.1||8.4|
|5 Lower supervisors||8.6||8.8|
|6 Semi routine workers||18.0||17.0|
|8 No class||3.0||5.4|
The distribution of the NS-SEC between GFK and BSA shows great comparability, with the higher proportion of no class in GfK probably also explicable in terms of its wider age range.
Finally, we provide more details about the latent class analysis which we used in Savage et al (2014). After various tests, we used the weighted data where the BIC minimised the number of classes at 7. We also experimented with 8 classes where the BIC number was minimised at 8, but these proved difficult to explicate sociologically so we preferred the seven class model. The full details of the BiC minimisation at 7 classes are provided below.
|# classes||AIC||BIC||Classification error||Entropy R-squared|
“Classification error” is defined as E=(Σ_(i=1)^I w_i [1-maxP ̂(x│y_i )])/N, or a measure of the fraction of cases misclassified by modal classification (Vermunt and Magidson, 2005:62).
At the end of Mills’ (2014) paper, we are asked several critical questions. These are all addressed in our paper, though not in the order that Mills lays them out. For ease of reference, here are brief replies, including where they are more fully elaborated in our paper
What is your typology meant to explain?
Our goal in the paper isn’t to identify a latent variable with which to explain the variance in a cluster of other variables. As we explain in section 1, our concern is with understanding class formation. Notably, using a Bourdieusian understanding of capitals, our aim is to identify where the fault lines are in British society, and identify an alternative way of thinking about class boundaries.
This having been said, there are numerous ways in which the seven classes revealed by the latent class analysis might offer strong patterns of association with a range of variables which are not used to construct the LCA itself. In papers which we have submitted to Sociological Review, we have, for instance examined the association between the seven latent classes and attitudinal questions. We have also used the seven classes as a ‘destination class’ in which we can examine processes which might be associated with mobility into them, including comparing them with NS-SEC classes.
Why should we have confidence in a typology built on the basis of such a small amount of data?
We have made the typology in good faith on the basis of the data made available to us and it is for readers to judge how much confidence to place in it. The GfK sample was commissioned by the BBC (to whom we remain grateful for their support) and we believe it offers better prospects of deriving general patterns than the GBCS. The quota sample was selected by the BBC presumably for reasons of cost, and whilst we recognise its limitations, we still believe analyses derived from it have value. Appendix 1 provides more information on the comparability between GfK and larger national representative data sets which we think are relatively reassuring. We should note that there are a number of highly influential studies of class which use only slightly larger data sets, e.g. Marshall et al (1988) and Bennett et al (2008).
We’re aware of the fact that our sample size means the confidence intervals surrounding the estimates of the classes are larger than they would be for larger samples, and we recognise that this will necessarily lead to less precision about the specific size of the classes. However, the number of variables used to construct our measures from which the latent class analysis was conducted is actually very large. , and we couldn’t use any other existing survey to attempt such an analysis.
What will you do when your method is applied to a larger amount of data and you discover, as you undoubtedly will, a larger number of classes?
We will certainly be interested to see what results are produced if similar methods are applied to other, larger, data sets, and as is normal in social science, we will be happy to review the arguments here on the basis of this further analysis. We would emphasise that our approach was not to mechanically define the number of classes simply according to the model selection criteria as where BIC is minimised. As we explain in the paper, the identification of a plausible set of distinct classes isbased on sociological interpretation of the patterns it produces, and using the GBCS as a tool to aid our interpretation. Therefore, if analysis of a larger sample reveals say 10 (or however many) classes where BIC is minimised, we will also seek to explore whether these might be sociologically meaningful, whether some of these classes could be seen as fractions of ‘our’ classes and so on. We see this process of iterative review as central to best practice in social science.
What use can the GBCS (as opposed to the GfK) data actually be put to?
As we explained in Section 4, above, the GBCS is used to interpret the meaning of the seven classes in Savage et al (2014). As we discuss in the introduction to this paper, we have just submitted a series of papers using the GBCS to study the elite, which is a group over-represented in the GBCS. Given the problems of finding suitable data sources on this group, we think the GBCS offers unusual prospects, even whilst fully recognising the need to be careful in making any interpretations because of the possibility of skews based on unobserved factors.
We note that the unusually large sample size of GBCS does allow the possibility of unusually granular analysis. Whilst care will inevitably be required to make inferences from the GBCS given its sample selection bias it contains, in the absence of other sources, we continue to think that it offers potential for innovative analysis. No doubt this can be judged further when our additional analyses of the GBCS are submitted and subject to critical review.
Do you accept that your data show that cultural consumption is related to conventional measures of social class?
Yes. We have never denied this. As well as being something we drew attention to in our paper, there is also a range of other research which some of us have been involved with on this very issue – see for instance Bennett et al (2009) and Le Roux et al (2008) for a full discussion of class (using NS-SEC categories) and the relationship with cultural consumption. We note that our comment on this issue (Savage 2014; 222) was directed towards the findings of other research.
What is gained by relabeling age-cohort and life-cycle stage as ‘social class’?
This depends on our conceptualisation of class, which we have tried to further explicate in section 1 of this paper. As we discuss in Section 4 above, people with different stocks of certain capitals can be meaningfully understood as being in different classes. That these capitals are associated with age or cohort effects doesn’t change that.
Do you really believe that changing one’s social class can be a matter of getting out of bed and making a serious effort to like Brahms or to attract a few more Facebook friends?
Boundary issues are pertinent for any kind of social classification, including the NS-SEC. The point of using MCA to assess the role of cultural consumption in cultural capital is to recognise that entire groupings of cultural practices are involved and it is unlikely that a change of one practice alone would cause a significant enough shift to change class in the terms we use.
More generally, some classes are easier to move between than others – nobody’s going to move from the traditional working class to the elite simply on the basis of the number of Facebook friends they have – but if an individual is able to radically change their social and cultural capital in such a way that both can be exploited, then this is possible. It will be harder than adding a handful of Facebook friends and buying a few CDs if this is to be meaningfully put to work.
 Those especially interested in Mill’s critique, may wish to turn immediately to Appendix 1 which provides details on the GfK survey and the latent class analysis, and Appendix 2 which provides headline responses to the questions he poses in the conclusion to his paper, and are grounds for which are unpacked in the body of this reply.
 See, for instance, ‘we should emphasise that this is a different kind of model to that developed by Goldthorpe and embedded in the NS-SEC, since it is an inductive, rather than deductive class schema’..
 In his more recent work, Wright (1997) invokes three principles of exploitation, firstly (inverse inter-dependence) that party A depends on the material deprivation of party B; secondly that party B are excluded from the productive resources which party A possesses, and finally that it is the appropriation of party B’s labour which allows the first two principles to work. However, this formulation could be applied to any economic differentiation within a market system and ultimately lacks clarity
 There is also the difficulty of deriving a seven fold model of class into a theoretical framework which only readily distinguishes between employees and employers, and between employees on a labour contract and ‘service relationship’.
 Or, to put this another way, the only possible analytic used by Goldthorpe to justify the NS-SEC as classes is to separate out the service class from those on labour contracts, yet this binary divide seems a crude instrument to register the complexity of class inequality.
 Indeed, we have made the same point ourselves, see Bennett et al 2008) Chapter 3 and especially 4, and in Savage (2010), chapter 3.
 Indeed, this view is explicitly contested in Bennett et al 2009 where we show that those who are not engaged with the measures of cultural taste and practice that we asked about on our survey were in fact more likely to be involved with informal ties with kin, friends and neighbours. See further, Savage 2010.
 Mills implies at various points that it is easy to change cultural practices (for instance, in his comment about how changing Facebook friends might entail a change of class) and therefore that these are not sociologically salient. Given the extent of sociological research which emphasises the powerful social structuring of cultural practices, it would be helpful for Mills to have produced sociological evidence for his alternative view.
 Ethnic categories not quite comparable between GfK and BSA. The latter includes options for ‘don’t know’ and ‘refusals’.