Mapping the Social Class Structure: From Occupational Mobility to Social Class Categories Using Network Analysis

Abstract

This article develops a new explorative method for deriving social class categories from patterns of occupational mobility. In line with Max Weber, our research is based on the notion that, if class boundaries do not inhibit social mobility then the class categories are of little value. Thus, unlike dominant, theoretically defined class schemes, this article derives social class categories from observed patterns in a mobility network covering intra-generational mobility. The network is based on a mobility table of 109 occupational categories tied together by 1,590,834 job shifts on the Danish labour market 2001–2007. The number of categories are reduced from 109 to 34 by applying a new clustering algorithm specifically designed for the study of mobility tables (MONECA). These intra-generational social class categories are related to the central discussions of gender, income, education and political action by providing empirical evidence of strong patterns of intra-generational class divisions along these lines.

Keywords

class analysis class theory cluster analysis intra-generational social mobility methodology MONECA occupational mobility social class social network analysis

Introduction

This article proposes a new methodology for identifying class boundaries on the basis of a network of mobility between occupations. By deriving the class structure from mobility patterns we are able to stay closer to the theoretical ambition in Weberian class analysis, in which mobility between classes should be rare. This new method therefore contributes to the question of mapping the class structure which has preoccupied sociologists for more than a century (Marx and Engels, 2008 [1848]; Weber, 1978 [1922]). The questions of class boundaries and the correct empirical delimitation of classes are central to this debate. The problem of class boundaries is posed either as a question of the empirical operational outcome of competing class theories (Goldthorpe, 2007; Wright, 2005), as a question of the categories designed to tests various hypotheses about the structure of social mobility (e.g. Blau and Duncan, 1967; Clogg and Goodman, 1984) or as a question of the relative delimitation of clusters of individuals sharing tastes and dispositions (Bourdieu, 1987). Lately, this debate has changed profoundly, as the relevance of class theory, and indeed the very existence of classes has been brought into question (Pakulski, 2005).

Grusky, Weeden and associates have set out to reconstruct class analysis by focusing on the realistic occupational categories expressing the division of labour as the site of production of class-based behaviour and mobility patterns (Weeden and Grusky, 2005, 2012). What is at stake in this long running debate (Erikson et al., 2012; Grusky and Weeden, 2002) is whether class is best explained on the basis of 88 micro-classes (Jonsson et al., 2009), 11 meso-classes (Erikson and Goldthorpe, 1993) or 12 relational classes (Wright, 1989). Such theoretically derived class schemes are the most prominent, but they have been challenged by more data sensitive and descriptive approaches (Savage et al., 2013, 2015).

In the descriptive tradition, this study will demonstrate the viability of a data sensitive approach that derives intra-generational class structure from the mobility patterns between occupations. This is achieved by applying a new clustering algorithm, the Mobility Network Clustering Algorithm (MONECA), which we have developed¹ to register-data drawn from official records covering the entire Danish working population in the period 2001–2007. The result is 34 social classes reflecting the intra-generational social mobility structure. This result can be characterized as a mix of meso- and micro-classes, suggesting that the Danish intra-generational social class structure is not captured well by any of the dominant universal class schemes. Rather, analyses of the distribution of income, education, political action and gender indicates that no single factor alone can account for the intra-generational social class structure revealed. This suggests that future research should investigate how different factors in different ways are involved in the formation of the various social classes of the intra-generational mobility structure.

Irrespective of how different theories have derived class boundaries, a common assumption is that these barriers also represent barriers to social mobility. Indeed, social mobility is at the very heart of class analysis. The power of class as an analytical concept consists in its ability to reflect that members of society cannot easily leave their class position. This entails that those sharing a class position will tend to perceive, and act upon the world in similar ways because their lives are conditioned by the same inescapable structures (Giddens, 1973). In truth, unless they are accompanied by barriers to mobility, class boundaries are only academic abstractions of little practical importance. The profound relation between social mobility and class finds an early and clear formulation in Max Weber’s famous descriptive definition of social class: ‘[a] “social class” makes up the totality of those class situations within which individual and generational mobility is easy and typical’ (Weber, 1978 [1922]: 302, emphasis in original). While both Giddens and Weber place barriers to mobility at the core of their class theory, neither of them had methods that allowed mobility to define class boundaries. This article hopes to provide such a methodology.

None of today’s dominant class schemes take their starting point in the observation of social mobility. The class schemes mentioned above all start from theoretical considerations about class formation, while mobility only plays a role as a, however privileged, variable to test the class schemes. Bourdieu starts from a theory of distinction, but his approach is more empirically sensitive and identifies classes by empirical analysis of latent dimensions or fields (Bourdieu, 1984). However, no observation of patterns of mobility are included in determining the class categories (Andersen and Hansen, 2012; Wacquant, 2013). The same is the case with the CAMSIS-scale, based on observed social interaction patterns of friendship and marriage (Prandy and Lambert, 2003; Stewart et al., 1980). If we acknowledge the fundamental weakness of any class scheme which does not reflect actual barriers to social mobility, the contemporary lack of such schemes challenges us to develop a method by which we can derive a class scheme from the observation of social mobility patterns between occupations. Such a description of the mobility structure will allow us to aggregate occupations into the proper classes within which mobility is easy and typical, and thereby to ground our class scheme solidly in the problem of mobility.

In this article we conceptualize the mobility table as a network of occupations tied together by the social mobility between the occupational categories. We develop an exploratory method that clusters together the occupations in order to derive the intra-generational social class structure. In the following section we elaborate on the intrinsic relationship between class and the structure of mobility in the tradition of class analysis. We argue that, in order to advance class analysis, we need to approach the empirical analysis of classes and their boundaries in a descriptive way, as opposed to a purely explanatory strategy based on statistical model-testing. We then move on to the main contribution, namely, the new analytical method we propose. Subsequently, we apply the method to Danish register-data and analyse 1,590,834 job shifts which took place between 2001 and 2007. After presenting the results, we discuss certain methodological issues and the resulting social class categories in relation to the factors of gender, income, education and political action. Finally, we present our conclusions.

Class and Mobility Structure

For students of social mobility, Weber’s above-quoted definition of a social class constitutes an important starting point. Weber makes two points: first, intra- as well as inter-generational mobility is constitutive of social class. Inter-generational mobility has received much attention, but in this article we are going to analyse intra-generational mobility. This should suffice, because the purpose of this article is to present and demonstrate our method’s ability to overcome the problem of identifying class boundaries. Furthermore, intra-generational mobility has received relatively little attention in the literature on class, and this study contributes to countering this bias. It does imply, however, that the results should not be directly compared to analyses of inter-generational mobility. Different dynamics are at play in the two forms of mobility, which warrants separating the analyses of the two forms of mobility and their class structure. We therefore also hesitate to make direct comparisons to class schemes like those mentioned above, which, even though they may claim to also encompass intra-generational mobility, are often constructed in relation to inter-generational mobility.

Weber’s second, and more crucial point is that social class is made up of patterns of ‘easy and typical’ mobility. Theoretically speaking, a mobility pattern refers to the transition from one position in the economy to another. When operationalized empirically, mobility pattern typically means movement from one occupation to another. The logic of Weber’s definition further suggests that analysis of mobility patterns should be the starting point for the empirical investigation of the social class structure.

Giddens (1973: 48) rephrases and specifies Weber’s definition of social classes as ‘a cluster of class situations which are linked together by virtue of the fact that they involve common mobility chances, either within the career of individuals or across the generations’. What we develop in this article is precisely a method which can identify such ‘clusters of class situations’ based on the observed patterns of mobility at a highly disaggregated level. Also, following Giddens, we separate the question of intra- and inter-generational mobility, and focus on the former. This represents a deviation from Weber who tended to conflate the two forms of mobility. An even more operationalized definition, alluding to social network analysis (SNA), can be found in Breiger’s (1981: 584, emphases in original) summary of Blau, Duncan and Giddens’ Weberian perspective: ‘classes are essentially sociometric “cliques” defined so that mobility chances […] are higher within the cliques than between them’.

The logic of Weber’s definition of social class suggests an approach in which description is prior to explanation. In order to investigate the mechanisms generating boundaries to mobility, that is, explain social classes, we must first identify the exact location of the social class boundaries. To use Giddens’ concepts, we must carefully describe the class-based structure of social mobility in order to investigate the structuration of social mobility. Following this logic, Giddens (1973: 110) underscores that ‘the existence of distinct class “boundaries”’ cannot be ‘settled in abstracto’, because the class structuration of various societies ‘differs significantly according to variations in economic and political development’.

Hence, what is needed is a detailed description of the social class structure, understood as a number of occupational clusters within which mobility is relatively high and between which mobility is relatively low. Such a description will provide the structure that is to be explained. The explanatory effort should take its point of departure in investigation of the identified boundaries. This would enable examining the mechanisms and processes generating the barriers to mobility.

An argument supporting this claim has been provided by Savage et al. (2013) who, echoing Giddens, argue that the need for descriptive and exploratory methods has been accentuated by the growing realization that no single class scheme fits all societies. Rather, if uncritically applied such universal class schemes may be misleading because they make us blind to significant differences between societies, and explanations related to specific societies may be overlooked. This is due to both ‘real cross-nation differences with respect to qualification levels, job autonomy, career prospects (i.e. social mobility), organization of production, etc.’ (Savage et al., 2013: 223) and the multi-dimensionality of social stratification. In our perspective, multi-dimensionality means that the mobility barriers generating the social mobility structure may consist of a variety of processes and mechanisms of selection. This implies that, rather than excluding the impact of factors such as gender, age, institutions, cultural norms, ethnicity and so on on mobility, they are perceived as potential drivers of class formation.

For instance, if a social class turns out to be the result of mobility patterns created by young people enrolled in education who work part-time jobs in the service sector and often change jobs, this is not a problem, but a result in the sense that the existence of this social class is explained by a specific position in the economy tied to age. An example of such a finding is provided in the case of the Emergent service workers identified by Savage et al. (2013: 240ff.). Following this logic, also such factors as, for example, gender-based division of labour may give rise to barriers to job-mobility constitutive of social classes. Such a descriptive and exploratory approach implies that, in the subsequent mapping of the Danish intra-generational social class structure, we do not, a priori, exclude or seek to control for any number of factors that according to a given theory of class are considered extra-class.

The descriptive concept of social class does, of course, relate to class theory, but the distinction is crucial as the concept of social class does not imply a theory of its formation as opposed to the class-concept of a class theory. Thus, several social classes may be combined into one class due to theoretical explanation. For instance, if unskilled workers are separated into different social classes the separate social classes may together constitute a class of unskilled workers. Thus, social classes should be viewed as the building blocks or segments that constitute the starting point for the operationalization of a class theory into a class scheme. However, the social classes impose limitations on the construction of classes. The salience of mobility barriers to class boundaries implies that if a class theory suggests splitting a social class (within which mobility is easy and typical) into different classes, the validity of the class scheme will be undermined. This is due to the permeability of the class boundaries resulting from the split.

The Mobility Table as Network

This study relates to the inductive tradition of research on class and stratification with the occupational mobility table at the centre (Breiger, 1981; Goodman, 1981; Klatzky and Hodge, 1971; Levine, 1972; MacDonald, 1972). However, this study diverts in two ways. First, we focus on intra- and not inter-generational mobility. Second, most scholars in this tradition have aimed at identifying the dimensions driving mobility and have often based their models on scales of the occupational hierarchies (e.g. Duncan, 1961; Treiman, 1977). This study instead aims at describing the intra-generational social class structure in an exploratory manner.

In order to correctly delimit the mobility categories in a descriptive way, we conceptualize the occupational mobility table as a network of occupations. This has been done before by, for example, Griffiths and Lambert (2012), although they analysed marriage-relations and not social mobility. Occupations take the form of nodes in the network and the individuals’ mobility between occupations generates the ties. From such a mapping we can observe between which occupations the labour force flows freely and between which occupations barriers appear to disrupt the flow of labour. The next step is quite literally to identify the ‘essentially sociometric “cliques”’ (Breiger, 1981: 584) that constitute the social class categories of the mobility structure.

The ties connecting the nodes are measured as the relative risk (RR). RR expresses the relative likeliness of the occurrence of job shifts from one occupation to another. RR = 1 represents the ideal situation of ‘perfect’ mobility. The straightforward translation of the RR into a measure of the intensity (or weight) of a network tie would be that if RR ⩾ 1 there is a tie between the nodes. If RR < 1 the nodes are not connected. We can now depict the intra-generational mobility on the Danish labour market in the period 2001–2007 as a directed weighted network which is done in Figure 1. Later we present the data used to construct the network. The size of the nodes expresses a logarithmic function of the size of the occupational category. The colour of the nodes expresses degree of internal mobility. The darker the node, the closer to 100 per cent internal mobility. The alpha of the colour of the edges expresses the intensity of the mobility flow.

Figure 1.

Intra-generational occupational mobility as network.

The Mobility Network Clustering Algorithm (MONECA)²

In a dense network like the one depicted in Figure 1, in which almost all nodes are connected, it is difficult to make sense of the structure. In this case we look for groups of especially tightly connected nodes. The aim is to identify the cohesive and non-overlapping sub-groups of the network in order to derive the social class structure. Conventional SNA concepts such as clique and core are problematic as they produce cluster solutions that overlap. In contrast, clusters generally refer to non-overlapping cohesive sub-groups (Scott, 2000: 126ff.). Cluster analysis has the additional advantage over the sociometric concepts that it is better suited to handle weighted networks (Knoke and Yang, 2008: 80ff.). Others have used cluster analysis on distance-matrices of social mobility (e.g. Hope, 1972). The novelty of this study is the development of a cluster algorithm that clusters on the basis of weighted network ties rather than abstract distances. The principal difference is that traditional cluster analysis is based in the abstract space of the distance matrix where all categories, no matter how remote, have a distance and therefore, in principle, can be clustered. The binary logic of network ties, on the other hand, implies that unconnected categories cannot be clustered together (see Toubøl et al., 2013 for a detailed discussion of this issue).

MONECA is designed to identify discrete clusters of interconnected nodes in a dense network. The logic of the algorithm is closely associated with the concept of the clique. The task of the algorithm is to decide to which clique to allocate the nodes that are in the overlapping area. To answer this question, the algorithm works in an agglomerative manner, considering the connections of the single pairs of nodes. The first step is to pair together the two most intensely connected nodes, which then form a cluster with the properties of a dyadic clique. In step two it proceeds to pair together the second most intensely connected pair of nodes in the same manner as in step one. And so it continues until all connections have been considered. At any subsequent step, if the two nodes under consideration are already members of two different clusters, they can only be joined together if all the nodes in their respective clusters constitute a clique and, thus, can be joined together forming a new cluster. This clique-criterion provides the stop rule for when no more single or sets of nodes should be paired together forming new clusters (for an expanded explanation of the methodology, see the online appendix A).

The Danish Intra-Generational Mobility Structure

The application of the above outlined approach to Danish labour market mobility data covering the period 2001–2007 demonstrates how a network analytical approach may provide a solution to the longstanding problem of identifying class boundaries.

The data for our study were collected by Statistics Denmark and comprise the entire Danish labour market in the period 2001–2007.³ This provides us with yearly information concerning the individuals’ occupations (ISCO) as well as information on job shifts. With this information we can construct career sequences from 2001–2007 consisting of up to seven states in the cases of individuals employed during the entire period. We can determine whether the individuals changed their job or not, as well as from and to which occupation they moved during the transition from one state to another. In total, 11,274,435 transitions are recorded throughout the period. Of these, 1,590,834 (14.1%) represent job shift transitions. We disaggregate the occupational coding to the three-digit level of ISCO leaving us with 109 occupational categories⁴ and develop a 109 x 109 occupational mobility table with 11,881 cells.

For the calculation of the RR, we weight the expected frequencies by the proportion of the total number of employees in the occupational category. To be precise, we calculate the expected frequencies of job shifts as the mean of the column and row proportions of the total number of transitions to the grand total of job-shift-transitions. Thus, the expected frequency of job shifts in a given cell depends on the total size of the row and column cell of the occupation. The reason for doing so is to avoid underestimating occupations with a large proportion of mobile employees, as well as overestimating occupations with a small proportion of mobile employees. Furthermore, despite the large amount of data, some cells in the mobility table are very sparsely populated. In order to avoid false connections due to measurement errors, cells with a frequency below the threshold of five have been erased.

The clusters are identified as described in section 4, meaning they are cliques in which all nodes are mutually connected (i.e. density = 1). The criterion that the mobility pattern has to be mutual in order to constitute a tie is important, as we would otherwise risk joining together occupations which are linked through promotion, such as junior and senior positions in a hierarchical organization. Thus, one-way mobility patterns do not constitute ties considered by MONECA. Forming new categories from the identified clusters, the number of categories is reduced from 109 at the original level one to 56 at level two. MONECA can be used to reduce the number of categories further. We can form a new 56 x 56 directed matrix using the categories identified by MONECA at level two and perform the procedure once more. The result of what constitutes a third level in this agglomerative clustering procedure is shown in the level three graph of Figure 2. The new clusters are marked by lines encircling existing clusters of nodes and single nodes. The new clusters are, by definition, not cliques but are, nonetheless, very dense.

Figure 2.

The five levels of the cluster analysis.

We can repeat the procedure a number of times, reducing the number of categories further (see Figure 2). The amount of possible repetitions is restricted because the matrix is not hollow. As such, for each repetition which results in more aggregated categories, a larger share of the total mobility will be within-mobility located in the diagonal. This leaves less between-mobility to constitute mutual mobility patterns. Eventually, no mutual connections can be detected; hence, no more categories can be merged. Thus, MONECA has a built-in stop point that is conditioned by the chosen cut point (in this case the cut point is RR = 1). Thus, in contrast to most agglomerative clustering algorithms, MONECA does not continue until all cases are merged into one big cluster. In this case, MONECA continues until level five, giving us 34 categories, as can be seen from the level five graph of Figure 2.

Results

The result is a cluster solution of 34 occupational categories, as depicted in Figure 3. Two categories are level five clusters, 12 are level three clusters, seven are level two clusters and 13 are level one clusters (for details regarding the levels of the cluster solution see the online appendix B).

Figure 3.

Result of cluster analysis.

A cluster solution should always be critically inspected and assessed by the researcher. In our case this is especially important with regard to clusters formed at level three and higher. By virtue of the clustering procedure these are not cliques where movement between all occupations is easy. Hence, a critical inspection and evaluation of whether the higher level clusters are too incoherent or their boundaries are too permeable is paramount. Summary statistics of the clusters are provided in Table 1. Numbers correspond to the numbers in Figure 3. Titles are the product of our interpretation of what characterizes the occupations of the clusters. Size refers to the share of the total number of employees at the labour market. The 10 largest categories cover 78 per cent of the labour market. The concentration in size suggests a relatively simple class structure with a few meso-classes and several micro-classes.

Table 1.

Summary of final cluster solution produced by MONECA by size.

#	Title	Size	Density	Within mob.	Nodes
3.7	Sales, services	0.205	0.811	0.750	10
1.54	Care workers	0.111	N/A	0.740	1
5.1	Drivers, construction	0.079	0.476	0.812	7
5.2	Manufacture	0.066	0.603	0.673	17
3.8	Metalworkers, craftsmen	0.066	0.850	0.802	5
3.14	Management, administration	0.058	0.650	0.723	5
3.9	Clerks	0.055	0.667	0.701	3
3.1	Finance	0.054	0.667	0.713	4
3.6	Pre-primary education	0.043	0.833	0.818	4
3.11	Engineers, technicians	0.042	0.833	0.790	4
2.21	Primary teaching	0.027	1.000	0.750	2
2.20	Nurses, midwives	0.024	1.000	0.860	2
3.5	Law, social work/-science	0.023	0.600	0.786	6
2.9	Medicine, science	0.021	1.000	0.848	4
2.7	Computing	0.020	1.000	0.770	2
1.31	Health associate professionals	0.014	N/A	0.820	1
3.4	Secondary teaching	0.013	0.667	0.705	3
2.22	Managers	0.013	1.000	0.617	2
2.6	Sport, culture	0.012	1.000	0.727	3
3.13	Agriculture, forestry	0.011	0.600	0.730	5
2.25	Building caretakers	0.009	1.000	0.652	2
1.67	Painters	0.009	N/A	0.829	1
3.3	Ship, aircraft, fishery	0.006	0.667	0.901	3
1.55	Other personal services	0.006	N/A	0.771	1
3.12	Printing	0.005	0.667	0.680	3
1.56	Protective services workers	0.004	N/A	0.632	1
1.20	Archivists	0.001	N/A	0.709	1
1.23	Religious professionals	0.001	N/A	0.935	1
1.72	Precision workers	0.001	N/A	0.625	1
1.96	Train	0.001	N/A	0.694	1
1.41	Police	0.000	N/A	0.893	1
1.73	Potters	0.000	N/A	0.657	1
1.79	Felt, pelt and leather workers	0.000	N/A	0.653	1
1.101	Street services	0.000	N/A	0.500	1

Density is the share of the total number of possible edges in the cluster; the higher the density, the more coherent the cluster is, because the barriers to mobility within the cluster are low. The ideal is a clique where density equals one. All level one and two clusters are by virtue of MONECA cliques. The opposite is the case for higher level clusters, as they never have a density of one. Still, with the exception of the level five cluster 5.1 Motor vehicle drivers & construction the density for all clusters is high (⩾ 0.6). Within-mobility is the share of the mobility from a given cluster going to a destination within the same cluster. This indicates the discreteness of the clusters: high within-mobility means that a cluster is relatively isolated in the social mobility structure in the sense that the chance of entering or leaving the cluster, that is, crossing the social class boundary, is small.

Inspecting Table 1, the relatively low density of cluster 5.1 draws attention. Close inspection of the map shows that the latest addition, 712 Building frame & related trades workers, is only connected to two of the other six occupations in the cluster, meaning that mobility from this occupation is not easy. Depending on subject matter, it might be advisable to break up the cluster and return to the level four solution, but for the sake of simplicity, and because the purpose of this article is to demonstrate the workings of the method, we stick with MONECA’s original cluster solution.

Before we turn to the discussion, we will provide the main conclusion from three reliability tests. The tests are presented in detail in the online appendix C. The tests concern the RR cut point, the minimum cell-frequency threshold and sensitivity to changes in the network. Figure 4 shows the correlation between the final cluster solution and the manipulated solutions of the three tests.

Figure 4.

Reliability test results: Correlations with final solution.

The cut point of RR = 1 is qualified on theoretical grounds because it is equal to perfect mobility. Nonetheless, it would be critical to the reliability if small variations cause massive change to the cluster solution. This is not the case. When comparing the cluster solutions of different RR cut points to the solution of an RR = 1, the correlation never goes below 0.6 within the entire test-interval of 0.5 ⩽ RR ⩽ 2.0, and within the 0.825 ⩽ RR ⩽ 1.3 it is consistently higher than 0.8. The minimum cell-frequency threshold is more critical as its determination to an extent relies on the judgement of the researcher. However, the test results are reassuring. The correlation with the solution of a threshold of five remains around 0.8 in the entire test-interval from 0 to 20. Finally, we test MONECA’s ability to reproduce the results when erasing up to 20 randomly chosen nodes in the network, which, in turn, affects the relations between the remaining nodes. We compared the results of the reduced networks with the solution of the complete network. This was done 100 times at each level from 1–20. On average, the correlation was never lower than 0.9 and the single lowest correlation among the 2000 was 0.65. From this we conclude that MONECA, in a fairly consistent manner, reproduces the results even when imposing rather comprehensive changes to the network.

Discussion

We begin by considering the sociological status of the categories suggested by MONECA. We then discuss the main limitations of the proposed methodology. Finally, we go beyond the dimension of mobility and consider the correspondence between the categories provided by MONECA and the factors of income, education, political behaviour and gender which represent key analytical aspects of class and stratification. In what follows, we only consider class in relation to intra-generational mobility. We do not include the class structure of inter-generational mobility, as we consider this to be a separate question.

In most studies, 34 class categories are neither analytically relevant, nor practical. As argued earlier, the 34 intra-generational social classes can be aggregated into fewer classes according to, for instance, a certain hierarchical principle, such as education or income, or other theoretical considerations. Mobility within these aggregated classes would naturally not be as likely as within the intra-generational social class categories. However, the barriers between the aggregated classes would remain in place and we are not in danger of creating a class scheme with permeable boundaries.

For theoretically derived class schemes this is not the case. As they are not derived from data on social mobility patterns they will quite likely place occupational categories with strong mutual mobility flows in separate classes, that is, split social classes. This problem is compounded if you impose a class scheme, designed to fit a specific country in a specific historical context on another country in another historical context.

Thus, we argue that the 34 occupational clusters should be seen as intra-generational social classes which, by themselves, are of analytical interest as well as possible building blocks, from which a simpler class-map can be formed. For instance, through an empirical analysis of the categories’ relation to factors such as property-relations, life-chances or social interaction we may find that two or more of the categories share the same position in the hierarchy and, thus, are segments in the same class. The result would then not violate the mobility criterion.

The method we have developed and presented in this article has two important interrelated limitations. First, the occupational classification may weaken the validity of the results if the categories do not fit the actual job-positions in the economy, that is, are invalid. Due to such misfits, the occupational classifications may group together dissimilar jobs or divide similar jobs into different categories. The latter scenario ought not to be a problem to the approach of this article, as two categories with similar jobs implying a high level of inter-categorical mobility would be clustered together. The former scenario, in which dissimilar jobs are grouped together in the same category, is harder to solve. In fact, we cannot be sure that this is not the case. However, the more disaggregated and detailed the baseline occupational classification, the lower the chance that dissimilar jobs are grouped in the same category.

Second, the number of observations in the data imposes limits to how disaggregated a level we can start from. A too sparsely populated mobility table lowers the validity of the subsequent analysis. The methodological approach of this article, designed to start from a very disaggregated level in order to partially avoid the mentioned hazards inherent in any occupational classification, depends on access to large datasets on mobility. Even though few countries collect register-data covering the entire population, as Denmark does, digitalization will most likely result in more such datasets becoming available to researchers in the future. Furthermore, large national survey programmes provide promising sources of such large datasets on mobility by pooling the data from several rounds.

While describing the patterns of intra-generational mobility and -social class segregation in Danish society is interesting in itself, it is crucial to show that these social class categories are of importance to more than mobility. Figure 5 shows the occupations measured by gender (proportion of women), income (the crude mean of the employees’ total disposable income per year after taxes, benefits and payment of interests and alimony from tax-records in DKK (2007 DKK/£ exchange rate ≈ 11)), union density (proportion of employees who were a member of a trade union) and education (mean number of prescribed years of education of the highest level of education received). In Table 2, the same variables are presented by social class in descending order by income (for the exact numbers by occupation see the online appendix D and for statistics summarizing the distributions on income and education see the online appendix E).

Figure 5.

Occupations by income, education, union density and gender distribution in 2007.

Table 2.

Clusters by income, education, union density and gender distribution in 2007.

#	Disp. income (DKK)	Education, years	Union density (%)	Share of women (%)
2.22	573,365	14.1	77.6	33.3
3.3	324,543	12.8^a	57.3	9.4
2.9	319,998	17.2	82.6	47.9
2.7	283,378	14.0	64.5	20.7
1.41	274,551	13.6	98.6	2.9
3.11	263,466	14.5	78.5	31.3
3.14	260,838	13.6	72.5	62.3
3.5	259,903	15.7	89.2	64.6
3.4	257,314	15.3	86.0	48.9
3.1	256,265	13.2	70.2	48.3
1.96	231,657	12.7	97.9	4.5
3.13	231,378	11.9	59.5	23.1
2.6	229,191	14.1	66.9	45.6
1.23	224,456	16.3	82.2	44.7
1.56	222,799	12.7	95.2	13.6
3.12	219,583	12.0	85.9	21.5
2.20	218,122	15.1	93.2	95.4
2.21	216,611	15.3	85.6	66.5
1.20	215,744	14.9	83.6	75.1
1.79	206,694	11.8	70.5	38.8
3.8	203,340	12.5	83.8	2.6
5.1	195,615	11.3	78.7	3.7
3.9	192,325	12.6	72.6	82.7
1.72	192,225	12.5	65.1	55.4
3.6	191,953	14.2	89.6	81.5
1.31	191,830	13.8	88.7	91.3
2.25	189,132	11.6	75.5	10.2
1.67	181,745	11.8	77.7	26.9
5.2	181,414	11.1	82.3	35.3
1.101	178,708	11.6	78.7	4.3
1.73	170,028	11.3	77.7	63.6
1.54	164,446	11.6	77.0	87.8
1.55	145,392	12.0	63.3	86.3
3.7	134,150	11.1	50.7	55.9
Total	201,561	12.7	74.1	51.9

Note: ^aThe very low mean years of education is due to aircraft pilots and air traffic controllers being registered with unrealistically short educations for reasons unknown (Albæk and Thomsen, 2011: 28).

Starting with income, we find that, in general, the occupations within the same social class are in the same income-range. When variation occurs, as in the case of classes 3.13 Agriculture, forestry, 3.14 Management, administration and 3.5 Law, social work/-science, this variation is reflected by the sub-groups of the social class.

Turning to the between variation we find large differences when we inspect the mean disposable incomes in Table 2 which is ordered by income (descending). At the very top of the income scale 2.22 Managers distinguishes itself from the rest with a disposable income of DKK573,365, more than four times that of 3.7 Sales, services (DKK134,150) at the bottom of the table and the income scale. This exclusive management group only makes up 1.3 per cent of the labour market.⁵ Next follows a group of nine high income social classes, in descending order from 3.3 Ship, aircraft, fishery (DKK324,543) to 3.1 Finance (DKK256,265). They make up almost a quarter (23.7%) of the labour market. Then follows a large group going from 1.96 Train (DKK231,657) to 1.101 Street services (DKK178,708). This group constitutes 42.8 per cent of the labour market. Finally, we have a group of four relatively low income social classes going from 1.73 Potters (DKK170,028) to 3.7 Sales, services (DKK134,150). Together they make up almost a third of the labour market (32.2%).

Following these observations, a classification based on a hierarchy of income suggests four aggregate classes: (1) a tiny elite class consisting of top level management; (2) an upper middle class consisting of lower level management, academics, administrators, specialists and the financial sector; (3) a large and heterogeneous middle class consisting of both blue and white collar workers; and (4) a large lower class consisting of service workers.

Such a simple four-group hierarchy becomes problematic if we take education into consideration. The correspondence between level of income and education is far from perfect. For instance, the members of group 2.22 Managers with a mean income of DKK573,365 have a mean of 14.2 years of education. Members of 3.6 Pre-primary education who also have a mean of 14.2 years of education, on the other hand, have an income of less than half that amount (DKK191,953). At the bottom of Table 2, people in low income category of 3.7 Sales, services unsurprisingly also have relatively short educations (DKK134,150/11.1). However, surprisingly, people in the groups 5.1 Drivers, construction and 5.2 Manufacture, who have equivalent levels of education, have a much higher income (respectively DKK195,615/11.3 and DKK181,414/11.1). These discrepancies between income and education indicate that the individuals’ income is not only a product of human capital possession but also depends on their (social) class membership.

The relationship between class and political action is central to the discussion of the relevance of class. Usually, this has been discussed in terms of political partisanship. We take a slightly different approach and look at the degree of unionization. Even though unions are not political parties, they are political organizations and in Denmark, unions have historically been strongly associated with the political left (Toubøl and Jensen, 2014).

In international comparison, the overall union density in Denmark is, at 74.1 per cent, very high. Despite the general high level of unionization, there are considerable variations in union density among occupations. This may indicate that unionization is one of the factors influencing income distribution. An example of this is the contrast between 3.7 Sales, services, which has the lowest union density of all social classes (50.7%) and 5.1 Drivers, construction and 5.2 Manufacture with union densities well above average (78.7% and 82.3% respectively). A possible explanation of why the income level relative to educational level is so much higher in 5.1 and 5.2 compared to 3.7 is that unions are much stronger in 5.1 and 5.2. However, unions do not explain the low income level relative to the level of education of 3.6 Pre-primary education, 2.20 Nurses, midwifes and 2.21 Primary teaching. These categories have a high union density, well above average, but they still have low incomes relative to the level of education. We suggest that the explanation may be related to the extra-class factor of gender.

Gender is related to stratification theory but it is usually treated as a dimension distinct from class. However, the growing awareness of the intersectional nature of inequality (e.g. Walby et al., 2012) calls for the combination of class and gender, among other factors. In Figure 5 and Table 2 we register great variation in the proportion of women among occupations. The intra-generational social class categories reflect these variations. The low-income category 3.7 Sales, services has a relatively high proportion of women (55.9%) compared to 5.2 Manufacture (35.3%) and especially 5.1 Drivers, construction (3.7%). However, we find the highest proportion of women among all social classes in 2.20 Nurses, midwifes (95.4%), while also 3.6 Pre-primary education (81.5%) and 2.21 Primary teaching (66.5%) have proportions of women well above the overall average of 51.9%. These correlations indicate a negative relationship between income and the proportion of women in a social class.

Conclusion

The preceding discussion argues that employing descriptive methods like MONECA makes it possible to investigate the multi-dimensional nature of class structuration. Hence, the above outlined method and approach call for a more nuanced theoretical approach to class. The results also underline the need to focus on society and context-specific explorations of class structure, rather than universal class schemes. The mapping of the Danish intra-generational social class structure 2001–2007 in this article suggests that things are much more complex than what any single universal class scheme and theory is able to grasp. Furthermore, given this complexity, it is only reasonable to expect significant variation between different societies.

The above outlined methodological approach provides two innovations directed to the task of society and context-specific analysis of class structure. First, the conceptualization of the mobility table as a network enables us to depict the social mobility structure in a transparent and intuitively meaningful way, which corresponds with the basic imagery of social class and social mobility theory. Based on a simple probability model of randomness, mobility barriers become visible. Furthermore, we can distinguish between the direction and the relative intensity of the flow and take the differences in proportion of mobile workforce and the proportion of internal mobility into account.

Second, MONECA enables us to identify discrete clusters of occupations between which mobility is easy and typical, and it provides a methodological means to give substance to Weber’s rudimentary definition of social class. The clusters identified provide the basic intra-generational social class categories. These should be perceived as the result of an exploratory and descriptive analysis, mapping the social class categories of the intra-generational social mobility structure. However, subsequent explanatory analysis should have these categories as a basic starting point. The social classes may be aggregated into fewer and larger classes if an extended definition of class is employed. However, a social class category should never be split into different categories within a class scheme as the boundaries of the new class categories would be permeable. The validity of such a class scheme would not meet the mobility criterion.

As indicated by the preliminary and strictly descriptive analysis of the prominent variables income, education, gender and political action, the intra-generational social class map identified, reflects important factors of differentiation. Through the precise identification of boundaries, the social class map derived in the manner outlined above may very well provide a better starting point for the investigation of the factors driving class formation than the hitherto dominant theory-driven approaches.

Going beyond the question of class and social mobility, MONECA is useful for the analysis of any dense, weighted network from which it is meaningful to extract discrete categories based on a principle of connectivity. Thus, the outlined exploratory and data driven approach to categorization may prove very useful in providing the fundamental categories for research into other questions than social mobility.

Footnotes

Acknowledgements

We are grateful to Thomas P Boje, Peter Gundelach, Bue Rübner Hansen, Christian Lyhne Ibsen, Carsten Strøby Jensen and Jens-Peter Thomsen as well as the anonymous reviewers for indispensable comments and suggestions.

Funding

This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.

Notes

Jonas Toubøl is a PhD Fellow at the Department of Sociology, University of Copenhagen. His research interests are in political sociology and social stratification, especially social movements, class and labour market segmentation.

Anton Grau Larsen is Assistant Professor at Copenhagen Business School, Department of Business and Politics. His research interest are in elites, power, field analysis, class and the methodologies of social network analysis and correspondence analysis.

References

Albæk

Thomsen

(2011) Er Kvindefag lavtlønsfag? En analyse af sammenhængen mellem løn og andelen af kvinder i enkelte arbejdsfunktioner. SFI-rapporter. København: SFI–Det nationale forskningscenter for velfærd.

Andersen

Hansen

(2012) Class and cultural capital: The case of class inequality in educational performance. European Sociological Review 28(5): 607–621.

Blau

Duncan

(1967) The American Occupational Structure. New York: John Wiley & Sons.

Bourdieu

(1984) Distinction: A Social Critique of the Judgement of Taste. London: Routledge & Kegan Paul.

Bourdieu

(1987) What makes a social class? On the theoretical and practical existence of groups. Berkeley Journal of Sociology 32: 1–17.

Breiger

(1981) The social class structure of occupational mobility. American Journal of Sociology 87(3): 578–611.

Clogg

Goodman

(1984) Latent structure analysis of a set of multidimensional contingency tables. Journal of the American Statistical Association 79(388): 762–771.

Duncan

(1961) A socioeconomic index for all occupations. In: Reiss

(ed.) Occupations and Social Status. New York: Free Press of Glencoe, 109–138.

Erikson

Goldthorpe

(1993) The Constant Flux: A Study of Class Mobility in Industrial Societies. Oxford: Clarendon Press.

10.

Erikson

Goldthorpe

Hällsten

(2012) No way back up from ratcheting down? A critique of the ‘microclass’ approach to the analysis of social mobility. Acta Sociologica 55(3): 211–229.

11.

Giddens

(1973) The Class Structure of the Advanced Societies. London: Hutchinson.

12.

Goldthorpe

(2007) On Sociology. Stanford, CA: Stanford University Press.

13.

Goodman

(1981) Criteria for determining whether certain categories in a cross-classification table should be combined, with special reference to occupational categories in an occupational mobility table. American Journal of Sociology 87(3): 612–650.

14.

Griffiths

Lambert

(2012) Dimensions and boundaries: Comparative analysis of occupational structures using social network and social interaction distance analysis. Sociological Research Online 17(2): 5.

15.

Grusky

Weeden

(2002) Class analysis and the heavy weight of convention. Acta Sociologica 45(3): 229–236.

16.

Hope

(1972) Quantifying constraints on social mobility: The latent hierarchies of a contingency table. In: Hope

(ed.) The Analysis of Social Mobility: Methods and Approaches. Oxford Studies in Social Mobility: Working Papers. Oxford: Clarendon Press, 121–190.

17.

Jonsson

Grusky

Carlo

et al . (2009) Microclass mobility: Social reproduction in four countries. American Journal of Sociology 114(4): 977–1036.

18.

Klatzky

Hodge

(1971) A canonical correlation analysis of occupational mobility. Journal of the American Statistical Association 66(333): 16–22.

19.

Knoke

Yang

(2008) Social Network Analysis. Thousand Oaks, CA: SAGE.

20.

Levine

(1972) A two-parameter model of interaction in father–son status mobility. Behavioral Science 17(5): 455–465.

21.

MacDonald

(1972) MDSCAL and distances between socio-economic groups. In: Hope

(ed.) The Analysis of Social Mobility: Methods and Approaches. Oxford Studies in Social Mobility: Working Papers. Oxford: Clarendon Press, 211–234.

22.

Marx

Engels

(2008 [1848]) The Communist Manifesto. London: Pluto Press.

23.

Pakulski

(2005) Foundations of a post-class analysis. In: Wright

(ed.) Approaches to Class Analysis. Cambridge: Cambridge University Press, 152–179.

24.

Prandy

Lambert

(2003) Marriage, social distance and the social space: An alternative derivation and validation of the Cambridge scale. Sociology 37(3): 397–411.

25.

Savage

Devine

Cunningham

et al . (2013) A new model of social class? Findings from the BBC’s Great British Class Survey Experiment. Sociology 47(2): 219–250.

26.

Savage

Devine

Cunningham

et al . (2015) On social class, Anno 2014. Sociology 49(6): 1011–1030.

27.

Scott

(2000) Social Network Analysis: A Handbook. London: SAGE.

28.

Stewart

Prandy

Blackburn

(1980) Social Stratification and Occupations. New York: Holmes & Meier.

29.

Treiman

(1977) Occupational Prestige in Comparative Perspective. Quantitative Studies in Social Relations. New York: Academic Press.

30.

Toubøl

Jensen

(2014) Why do people join trade unions? The impact of workplace union density on union recruitment. Transfer: European Review of Labour and Research 20(1): 135–154.

31.

Toubøl

Larsen

Jensen

(2013) A network analytical approach to the study of labour market mobility. Paper presented at the 33rd Sunbelt Social Networks Conference of the International Network for Social Network Analysis (INSNA), Hamburg, Germany, May.

32.

Wacquant

(2013) Symbolic power and group-making: On Pierre Bourdieu’s reframing of class. Journal of Classical Sociology 13(2): 274–291.

33.

Walby

Armstrong

Strid

(2012) Intersectionality: Multiple inequalities in social theory. Sociology 46(2): 224–240.

34.

Weber

(1978 [1922]) Economy and Society: An Outline of Interpretive Sociology. Berkeley, CA: University of California Press.

35.

Weeden

Grusky

(2005) The case for a new class map. American Journal of Sociology 111(1): 141–212.

36.

Weeden

Grusky

(2012) The three worlds of inequality. American Journal of Sociology 117(6): 1723–1785.

37.

Wright

(1989) A general framework for the analysis of class structure. In: Wright

(ed.) The Debate on Classes. London: Verso.

38.

Wright

(2005) Foundations of a neo-Marxist class analysis. In: Wright

(ed.) Approaches to Class Analysis. Cambridge: Cambridge University Press, 4–30.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.90 MB

Mapping the Social Class Structure: From Occupational Mobility to Social Class Categories Using Network Analysis

Abstract

Keywords

Introduction

Class and Mobility Structure

The Mobility Table as Network

The Mobility Network Clustering Algorithm (MONECA) 2

The Danish Intra-Generational Mobility Structure

Results

Discussion

Conclusion

Footnotes

Acknowledgements

Funding

Notes

References

Supplementary Material

The Mobility Network Clustering Algorithm (MONECA)²