Social class inequalities in educational attainment: measuring social class using capitals, assets and resources

Key takeaways

content: "@connellySocialClassInequalities2021a" -file:@connellySocialClassInequalities2021a

Reading notes

Imported on 2024-06-26 11:21

⭐ Important

• & n this paper, we evaluate the benefits of a social class measure with a Bourdieusian theoretical foundation compared with an orthodox neo-Weberian occupation-based social class measure, the National Statistics Socio-economic Classification (NSSEC). First, we assess how closely we can replicate the Bourdieusian social class measure reported in Savage et al. ([2013]. (p. 280)
• & Second, we aim to compare and contrast the capitals, assets and resources based social class measure with the occupation-based National Statistics Socio-economic Classification, in an analysis of inequalities in school GCSE outcomes. (p. 280)
• & The study of social class has been central to sociology since its foundations (Giddens, 1971). (p. 280)
• & Research on social class has been one of the hallmarks of UK sociology since it burgeoned as an academic discipline after the Second World War (Pevalin & Rose, 2002). (p. 280)
• & Despite the centrality of social class within the sociological enterprise, the definitions and measurements are often muddled and frequently imprecise (see Breen & Rottman, 1995). (p. 280)
• & Historically, occupations and the occupational structure were considered central to the sociological conception of social class (Blau and Duncan, 1967, pp. 6–7). (p. 280)
• & At the core of these schemes is the theoretical conception that employment relations in the labour market are central to the allocation of individuals into social class categories (Erikson & Goldthorpe, 1992). (p. 281)
• & Individuals within a social class are theoretically understood as sharing a similar ‘market situation’ (e.g. levels of income, economic security, and chances for economic advancement), and ‘work situation’ (e.g. authority and control) (Goldthorpe, 1980). Following from this colocation, individuals within a social class are theorised as having similar lifestyles and comparable life chances (Rose & Pevalin, 2003). (p. 281)
• & More recently, Savage et al. (2013) have proposed an alternative theoretical basis for the measurement of social class, which is influenced by Bourdieusian rather than neoWeberian theory. We refer to this as the Capitals, Assets and Resources (CAR) approach. Under this theoretical umbrella, occupations are not theorised as the central indicator of social class. By contrast, and following Bourdieu (1984), the concepts of economic capital (e.g. income and wealth), cultural capital (e.g. engagement with cultural goods and activities) and social capital (e.g. social contacts and networks) are theorised as playing a fundamental role in ascribing the individual’s class position. (p. 281)
• & Our analysis represents a genuine attempt to undertake an assay of Savage et al. (2013) with a source of high quality existing social survey data. The linked administrative data used in this analysis is treated securely and has to conform to standard principles of statistical disclosure control (Department for Education et al., 2015). This means that some numbers are not shown, where this has occurred it is indicated in the tables. (p. 282)
• & Savage et al. (2013) create scales of cultural, economic and social capital and use a latent class2 analysis to identify their seven new social classes. Savage et al. (2013)do not provide detail on how these scales were produced and standardised. Therefore, it is not possible for us to follow this strategy and we do not create scores from the categorical variables before entering these variables into a latent class analysis. Scaling manifest indicators, especially categorical indicators, prior to a latent class analysis is an unorthodox practice and potentially leads to the model reflecting the scaling of the manifest variables rather than reflecting the structure of differences between responses (Bauer et al., 2003). (p. 282)
• & Bourdieu (1977) argues that cultural capital plays a major role in the reproduction of social inequalities through education. Bourdieu and Passeron (1990) argue that those young people who possess cultural capital will gain most from the education system due to their familiarity and competency in dealing with the dominant culture. Savage et al. (2013) use correspondence analysis to aide in the identification of elements of cultural capital, which are then used to create summated ratings scales of ‘highbrow’ and ‘emerging’ cultural capital (see Bennett et al., 2009). (p. 282)
• & Economic capital represents the command of economic resources (Bourdieu, 1986). In Savage et al. (2013) three indicators of economic capital are used, household income, household savings and house price. We use a continuous variable of household equivalised net monthly income, as well as a binary variable indicating whether the household own their home (including with a mortgage) or not. The Understanding Society data also include a measure of house price, however we needed to simplify our models due to convergence issues and this variable was removed. (p. 283)
• & We do not include an explicit measure of household savings in our analysis, however the income variable we use also includes income from investments as well as labour income (see Fisher et al., 2019). (p. 283)
• & The Understanding Society data does not offer social capital measures that closely mirror those used in the original analysis. We have chosen to use the number of friends of the parent with the greatest number of friends, although this has the limitation that we do not know the level of advantage of this social network. We also include measures that indicate whether either parent is active in a trade union or a professional organisation. (p. 283)
• & Latent class models relate a set of observed variables to a set of latent or unmeasured classes (McCutcheon, 1987). (p. 284)
• & As our model contains both continuous and categorical variables it would be best described as a latent class model of mixed observed variables. (p. 285)
• & Our latent class analysis is therefore appropriately weighted, but does not fully adjust for the design of the sample. (p. 285)
• & Table 2 reports the model fit statistics for a series of latent class models. The six class model is an improvement on the five class model, with a lower AIC and higher entropy. The six class model is preferred over the seven class model as it has a lower BIC and a higher entropy. However, we note that there is very little difference between the model fit statistics of the models and there is certainly no definitive class solution. Savage et al. (2013) state that the seven class model they discuss was selected as it minimised BIC, however technical details in later sources note that the eight class solution minimised BIC (Savage et al., 2015). This uncertainty is important to note as latent class analysis rarely identifies a definitive solution (McCutcheon, 1987). (p. 285)
• & We now apply the CAR measure described above in an analysis of GCSE attainment. GCSE subjects are assessed separately, and a subject-specific GCSE is awarded. Each GCSE subject is awarded a grade, historically the highest being grade A and the lowest grade G. From 1994 a higher grade of A* was introduced (Yang & Woodhouse, 2001). (p. 287)
• & The analysis using NS-SEC shows the familiar social class gradient (Demack, Drew, & Grimsley, 2000). The model that contains the CAR-based measure offers only a marginal increase in explanatory power. The substantive pattern shown in the model containing the CAR-based measure does not provide any further insights regarding the association between parental social class and filial educational attainment. The main pattern highlighted by the CAR-based measure is a divide between the ‘precariat’ and the ‘traditional working class’, and the other class categories. It is also important to note that three of the CAR-based classes contain a very small number of participants and therefore have very wide comparison intervals. (p. 289)
• & The first aim of the work was to assess how closely we could replicate the CAR-based social class measure reported in Savage et al. (2013). The overall finding is that it was not possible to recover the seven social classes reported in Savage et al. (2013). The latent class analysis of Understanding Society data recovered six classes, three of which contained only a very small number of sample members. (p. 289)
• & The characteristics of these classes were very different in nature to the classes described in Savage et al. (2013). This may be because we have examined the parents of young people completing their GCSEs and not the wider UK population, but a requirement of a social class measure should be some degree of stability to allow for the examination of class-based inequalities in different scenarios. (p. 289)
• & Both the CAR-based social class measure and NS-SEC illustrate a social class gradient. Pupils with parents in more advantaged social classes, on average, have better school GCSE outcomes. (p. 289)
• & We now reflect on the issues that have emerged from this comparative work. First, the study of population level social inequalities requires access to large-scale data resources. In practice few existing social surveys collect appropriate (manifest) variables suitable for measuring social, economic and cultural capital. (p. 290)
• & The empirical work presented above unambiguously demonstrates that even in a household panel study, which is expressly designed to support a broad range of analyses, the full range of manifest indicators that Savage et al. (2013) identify as being required to construct the CAR social class measure are not available. (p. 290)
• & When analysing datasets that do not contain all of the measures proposed by Savage et al. (2013) researchers will have to make choices about alternative operationalisations (e.g. using substitute measures). This ultimately hangs a question mark over the comparability of social class effects in different studies where alternative operationalisations of a CAR-based social class measure have been used. This issue may be especially stark when comparing social class effects over time. (p. 290)
• & A further related technical issue is the assignment of individuals to latent categories. Here we have used ‘modal’ assignment which is a common method (see Bartholomew et al., 2008). Recently, technical discussions have emerged regarding alternative methods of assignment (see Asparouhov & Muthén, 2014; Bakk, Tekle, & Vermunt, 2013; Heron et al., 2015; Vermunt, 2010). (p. 290)