An old model of social class? Job characteristics and the NS-SEC schema

Abstract

This article explores the relationship between the job characteristics underlying the Goldthorpe model of social class (work monitoring difficulty and human asset specificity) and those underlying theories of technological change (routine and analytical tasks) highlighted as key drivers for growing inequality. Analysis of the 2012 British Skills and Employment Survey demonstrates monitoring difficulty and asset specificity predict National Statistics Socio-Economic Classification (NS-SEC) membership and employment relations in ways expected by the Goldthorpe model, but the role of asset specificity is partially confounded by analytical tasks. It concludes that while the Goldthorpe model continues to provide a useful descriptive tool of inequality-producing processes and employment relations in the labour market, examining underlying job characteristics directly is a promising avenue for future research in understanding over time dynamics in the evolution of occupational inequalities.

Keywords

employment relations job characteristics NS-SEC social class technological change

Introduction

One of the most widely used models of social class is that associated with John Goldthorpe and colleagues. This model places two salient job characteristics as the basis of social class: how difficult work is to monitor and the level of human asset specificity required to perform the job (Goldthorpe, 2007). Differences in these two job characteristics explain the broad differentiation in employment relations (service relationships and labour contracts). This differentiation in employment relations in turn gives rise to class-based patterns of stratification. Since data on underlying job characteristics and employment relations are not normally available in social surveys, empirically, social classes are derived by aggregating detailed occupational categories into broader occupational groupings that are supposed to share similar underlying job characteristics, so therefore, employment relations and economic conditions. This model of social class culminated in the National Statistics Socio-Economic Classification (NS-SEC) classification of occupations (Rose and Pevalin, 2003, 2005), the model employed in this article. The Goldthorpe model and its predecessors have been widely used in social mobility studies, and, given its occupational basis, in work sociology as an explanatory variable (e.g. Chatzitheochari and Arber, 2009; Goldthorpe and McKnight, 2006; Warren, 2015).

This article explores how the two salient job characteristics providing the explanatory basis to the Goldthorpe model relate to two further job characteristics that are central to accounts of growing inequality – routine and analytical tasks – which are connected to developments in technology (Autor, 2013; Liu and Grusky, 2013). Technological change may have implications for the explanatory basis of the Goldthorpe model if these job tasks confound the roles of monitoring difficulty and asset specificity in determining class positions and employment relations. Given the growing popularity in technology-based explanations and the predominance of social class explanations in sociology, this article investigates their relationship using individual-level data on job characteristics. Implications for theory and research are discussed.

The theoretical basis of the Goldthorpe model

According to the Goldthorpe model, social classes are analytical constructs delineating differing solutions to the ‘contractual hazard’ inherent in the employment relationship (Goldthorpe, 2007). It makes the distinction between a ‘service relationship’ and a ‘labour contract’, with ‘mixed forms’ in between. Under service relationships (typically managerial and professional occupations), earnings are less connected to productivity, job tenures are longer and internal labour markets offer greater opportunities for advancement. Earnings are less connected to day-to-day productivity because work is harder to define and monitor. Employment is generally more secure because work requires higher asset specificity (specialist knowledge) so employers have much to gain in tying workers to the organization. Conversely, labour contracts are more akin to ‘spot contracts’ (typically semi- and lower-skilled occupations), where earnings are more directly connected to inputs and/or outputs and tenures typically shorter in duration, and offer fewer opportunities for advancement. As work is easier to define and to monitor, employers can control workflows more closely to work demands; for instance, by varying piece rates or paying by the hour. Since work in such jobs requires lower asset specificity, employers have little to gain in tying employees to the organization as employees are more easily expendable. Employment is therefore generally less secure and more exposed to market conditions. In summary, it is differences in these two salient job characteristics – difficulty of monitoring and human asset specificity – that give rise to differentiation in employment relations, which in turn lead to class-based inequality patterns (Goldthorpe, 2007).

Since information on monitoring difficulty and asset specificity are rarely available in social surveys, empirically, differences in these underlying job characteristics and resultant employment relations are proxied by aggregating detailed occupational codes to broader groupings (‘social classes’) that are supposed to share similar job characteristics and employment relations. The practice of aggregating detailed occupations to fewer groups culminated in the widely used NS-SEC schema (Rose and Pevalin, 2003), the model employed in this article, which standardizes the aggregation of occupations to classes.¹ The NS-SEC categories along with the underlying job characteristics emphasized by the model and resulting employment relationships are summarized in Table 1.

Table 1.

The National Statistics Socio-Economic Classification (NS-SEC).

Abbreviation	NS-SEC category	Monitoring difficulty	Asset specificity	Employment relations	Top 3 SOC 2000 occupations (4-digit)
HMP	Higher managerial and administrative occupations	Difficult	High	Service relationship	Marketing and sales managers; production, works and maintenance managers; software professionals
HMP	Higher professional occupations	Difficult	High	Service relationship
LMP	Lower managerial and professional occupations	Difficult	High	Service relationship (modified)	Nurses; medical radiographers; secondary education teaching professionals; primary and nursery education teaching professionals
I	Intermediate occupations	Moderate	Moderate	Labour contract (modified)	Customer care occupations; police officers (sergeant and below); call centre agents/operators
LST	Lower supervisory and technical occupations	Moderate	Moderate	Mixed	Sales and retail assistants (supervisor); cleaners (supervisor); heavy goods vehicle drivers (supervisor)
SR	Semi-routine occupations	Easy	Low	Labour contract	Sales and retail assistants; kitchen and catering assistants; retail cashiers and checkout operators
SR	Routine occupations	Easy	Low	Labour contract	Cleaners; heavy goods vehicle drivers; other goods handling and storage occupations n.e.c.

Sources: McGovern et al. (2007); Rose and Pevalin (2005). Top 3 SOC 2000 4-digit occupations from the 2012 British Skills and Employment Survey (BSES).

n.e.c., not elsewhere classified.

Previous research has demonstrated the NS-SEC schema and predecessor models have strong construct validity. Various job characteristics such as difficulty of monitoring and asset specificity strongly predict membership to the expected schema categories (Evans and Mills, 1998, 2000; McGovern et al., 2007). Additionally, previous research has also demonstrated class categories show considerable criterion validity in terms of being predictive of various aspects of employment relations such as earnings, the provision of fringe benefits, unemployment risks and control over work processes (Evans, 1992; Evans and Mills, 1998; Goldthorpe and McKnight, 2006; McGovern et al., 2007; Rose and Pevalin, 2003). However, one shortcoming of this research stream is that it has generally not examined these salient job characteristics in relation to the alternative explanations, nor has it examined how the relationship between the characteristics emphasized in the model and resulting employment relations may have changed over time.

Technological change, the Goldthorpe model and growing inequality

A burgeoning literature in economics has begun to converge on stratification research in highlighting the importance of job characteristics for inequality patterns. However, this literature, typified by the work of David Autor and colleagues, focuses on job tasks (Acemoglu and Autor, 2011; Autor, 2013). Whereas the Goldthorpe model makes the distinction between a service relationship and a labour contract based on monitoring difficulty and asset specificity, the task-based literature makes the distinction between jobs that can be replaced by technology and those which are complemented by it, based on their tasks. Task-biased technological change, as this body of work is sometimes known, purports that technology can easily replace routine tasks, but complements non-routine analytical ones, making them more productive. As technology advances and becomes more widespread, the demand for routine tasks falls and the demand for non-routine and analytical ones grows, deteriorating the earnings and working conditions for routine jobs, but upgrading them for non-routine analytical jobs (Autor, 2013; Autor and Handel, 2013).

Empirically, since information on job-level tasks are typically not available in social surveys, this stream of research maps task information from specialized task databases (such as O*NET in US research) onto detailed occupation codes in social surveys. Mean task scores at the detailed occupational-level are then used to map employment trends and wage differentials relating to tasks at the occupational-level with a technology-based interpretation (Autor, 2013).² Indeed, similar data limitations with respect to the lack of information on monitoring difficulty and asset specificity in social surveys is one rationale behind the NS-SEC schema (Mills, 2014: 438). Although the underlying measures for tasks and categorization of occupations in the technological change literature are not standardized to the same extent as with NS-SEC, the literature does consistently highlight significant and growing wage differentials according to the extent to which occupations are routine and analytical (Liu and Grusky, 2013; Zhou, 2015).

Although the Goldthorpe model emphasizes differences in monitoring difficulty and asset specificity and resulting employment relations, there are at least two reasons why they should be investigated in relation to the job characteristics emphasized by technology-based explanations. First, empirically, research in both economics and sociology finds that much of the growth in wage inequality over the last few decades in Britain is attributable to growing wage differentials between occupations – whether defined in terms of tasks (Goos and Manning, 2007) – or social classes (Williams, 2013). Investigating the relationship between the job characteristics emphasized by both theories together sheds light on the extent to which inequality trends may be due to shifts in the task structure related to technological change, or whether class-based changes in inequality are largely unrelated.

Second, even though technology-based theories primarily provide an over time explanation, there are theoretical reasons to believe changes in job tasks may have implications for the relative roles of monitoring difficulty and asset specificity in differentiating classes and employment relations. Given developments in technology, some types of knowledge have become more easily codified and stored by technology, meaning asset specificity may be becoming less influential for the type of employment contract offered by employers. The extent to which asset specificity still matters in determining the type of employment relations will therefore increasingly depend on the underlying tasks being performed, since some jobs where asset specificity was previously important can now have this ‘asset’ substituted with technology. In other words, substitutability of tasks with technology may be becoming a more important dividing line between classes than how easily one employee is substituted with another. This then has implications for the extent to which asset specificity predicts class categories and also how it predicts the type of employment relationship offered by employers.

Similarly, the role of monitoring difficulty in determining the type of employment relationship could also increasingly depend on the types of task. As routine tasks can increasingly be substituted by technology, tasks in some jobs which were previously easier to monitor may have become more difficult due to growing non-routine tasks. Conversely, tasks in jobs which were previously difficult to monitor may have become easier to monitor due to developments in technology, save for only those involving the most complex and analytical tasks. As with asset specificity, developments in technology may mean the dividing line between classes and the type of employment relations offered may increasingly be due to the extent to which tasks are routine and analytical, and less to do with how difficult work is to monitor. In summary, technological change may confound the theoretical basis to the Goldthorpe model in not only differentiating class categories, but also in determining employment relations.

It is not yet known the extent to which classes are simply a close relative to the sorts of tasks highlighted by technology-based explanations, or whether they are an analytically distinct explanation because both research streams tend to rely on occupation codes as proxies. A limited number of careful attempts to scrutinize the Goldthorpe model alongside alternative explanations do exist (Oesch, 2013; Tåhlin, 2007), but even these do not explicitly address task-based approaches. As Lambert and Bihagen (2014: 484) note, existing validation exercises of social class schemas generally do not consider other candidate factors. Part of the reason for this is the gap in knowledge due to the paucity of decent data relating closely enough to the relevant underlying job characteristics. Even though existing approaches to the measurement of underlying job characteristics have their own limitations, such exercises are still worthwhile given the predominance of social class and the growing popularity in technology-based explanations in sociology.

The British Skills and Employment Survey

Data come from the 2012 wave of British Skills and Employment Surveys (BSES) (Felstead et al., 2014). The BSES is one of the few surveys available that includes job characteristics central to the Goldthorpe model and technology-based explanations. The BSES uses random probability sampling so provides a nationally representative portrait of employees in Britain aged 20 to 65. After excluding 465 self-employed respondents and 332 cases with missing data, the final sample consists of 2398 employees. Following the standardized NS-SEC procedure, respondents are allocated to one of the ‘analytic’ NS-SEC classes in Table 1 based on their 4-digit SOC 2000 code and employment status (Rose and Pevalin, 2005). Higher managerial workers are merged with higher professionals to provide sufficient numbers of observations, leaving six NS-SEC classes. We use sampling weights in all analyses.³

One challenge of using secondary data on job characteristics is that the data were not collected with the researcher’s exact question in mind, often meaning subjective judgement calls in choosing between the plethora of items (Autor, 2013: 191). In this article, single items were selected which best matched difficulty of monitoring, asset specificity, routineness of work and intensity of analytical task-use.⁴ Difficulty of monitoring is captured by an item asking respondents ‘how closely supervised is your work?’, with the possible responses: ‘very closely’, ‘quite closely’, ‘not very closely’ and ‘not at all closely’. The reasoning behind this item is that work which is not closely supervised must also be work where monitoring is difficult, whereas the opposite is the case for work that is very closely supervised. While the BSES does contain items on the incidence of various monitoring methods such as appraisals, performance-related pay, clients, etc, these items do not indicate the underlying difficulty in monitoring work per se because the list is not exhaustive. Asset specificity is captured by an item asking respondents to rate ‘how important is specialist knowledge to perform your job?’, with the possible responses: ‘not at all important’, ‘not very important’, ‘fairly important’, ‘very important’ and ‘essential’. Asset specificity, as conceptualized in Goldthorpe (2007), refers to specialist knowledge broadly defined, including specialist to an occupation, field, or organization.

As for the job tasks highlighted by technology-based explanations, the extent to which tasks are routine is captured by the item ‘how frequently does your job involve short repetitive tasks?’, with possible responses: ‘never’, ‘rarely’, ‘sometimes’, ‘often’ and ‘always’.⁵ This measure is based on the assumption that jobs involving mostly repetitive tasks are more easily programmable or codified, and is broad enough to cover both manual and non-manual jobs. Analytical tasks are captured by an item asking respondents to rate ‘how important is analysing complex problems in depth?’, with possible responses: ‘not at all important’, ‘not very important’, ‘fairly important’, ‘very important’ and ‘essential’. This is included as an indicator of higher-level analytical tasks highlighted by technology-based explanations which are presumed to be especially complemented by developments in technology, but not easily substituted by it. Since the number of response categories across all items is not uniform, they are standardized to have a mean of 0 and standard deviation of 1.⁶

The relationship between job characteristics and several indicators of employment relations are investigated: hourly pay (logarithm);⁷ whether an employer pays into a pension (binary indicator); a scale on ease of dismissal (derived from averaging responses to two items reporting how long it will take for a respondent to be dismissed for performing poorly and repeatedly turning up late, with identical six-point scales ranging from ‘never’, ‘more than a year’, ‘within a year’, ‘within six months’, ‘within a month’ and ‘within a week’);⁸ whether unemployed in the last five years (binary indicator); and the degree of control over start and finish working times (a four-point ordinal scale from ‘strongly disagree’ to ‘strongly agree’). These specific characteristics are examined as they closely resemble ones used in previous validation exercises of the Goldthorpe model.

The analysis consists of two steps. First, the extent to which job characteristics predict NS-SEC category are assessed to examine the possible overlap in the characteristics highlighted by the Goldthorpe model and technology-based theories in determining NS-SEC membership. Second, the extent to which these job characteristics predict several indicators of employment relations are then considered. With both these exercises, the effect of including job tasks on the roles of monitoring difficulty and asset specificity is what is of focal interest.

The Goldthorpe model, job characteristics and the NS-SEC schema

In the first step, the relationships between monitoring difficulty and asset specificity with NS-SEC categories are explored through a simple multinomial logistic regression predicting membership to NS-SEC categories with the Higher Managerial and Professional class as the reference.⁹ The extent to which routine tasks and analytical tasks confound the predictive strength of these is examined.¹⁰ While in linear models (ordinary least squares, OLS), the magnitude of confounding on the variable of interest is ascertained relatively straightforwardly by comparing the change in the coefficient of the variable of interest following the inclusion of a confounder; in nonlinear models (e.g. logistic regression) it is not so straightforward. In nonlinear models, coefficients and the error variance are not separately identified so that changes in the variable of interest are not comparable across models since they are not measured on the same scale. The solution proposed by Karlson et al. (2012) (KHB) is therefore adopted. This method effectively controls for this problematic rescaling effect giving unbiased estimates of confounding.¹¹ This then allows for a simple decomposition of coefficients into total, direct and indirect effects in a similar way as with linear models.

The results from these two multinomial logistic regressions are displayed graphically in Figure 1,¹² which show respectively the associations between difficulty of monitoring (Panel A) and asset specificity (Panel B) with NS-SEC category relative to the Higher Managerial and Professional class. Examining total effects first (the sum of direct and indirect effects), the figure demonstrates that monitoring difficulty and asset specificity are strongly predictive of NS-SEC categories in ways expected by the Goldthorpe model. A standard deviation above average monitoring difficulty (Panel A) or asset specificity (Panel B) is associated with increasingly negative log odds of being in other classes relative to the Higher Managerial and Professional class (the reference category) the further one moves down the NS-SEC schema, displaying a clear class gradient. The total effects then broadly replicate findings in McGovern et al. (2007) in that the NS-SEC schema demonstrates good construct validity with respect to the two salient job characteristics emphasized by the model with newer data and different measures.

Figure 1.

Monitoring difficulty, asset specificity, job tasks and the NS-SEC schema.

In examining the confounding by tasks (routine and analytical), the predictive strength of monitoring difficulty is reduced between about a fifth and a quarter for most classes, while for human asset specificity it is more considerable across classes. This suggests that technology-based explanations do have some overlap with the Goldthorpe model in predicting NS-SEC membership. As might be expected, analytical tasks are more closely related to human asset specificity in predicting NS-SEC membership, whereas monitoring difficulty is more closely related to the extent to which tasks are routine. In particular, the confounding of asset specificity by analytical tasks is fairly considerable, suggesting that analytical tasks may in fact be an important consideration for the role asset specificity plays in determining class position, notably in terms of the dividing line between the Higher and Lower Managerial and Professional classes. Across classes, analytical tasks confound asset specificity in determining class positions by about two- to three-fifths. However, it must be noted that job tasks do not fully confound the roles of the two job characteristics highlighted by the Goldthorpe model, even in the case of asset specificity and analytical tasks, as statistically significant direct effects remain and overall patterns are left intact (see Table A1 in online-only appendix for the full results). This implies that the NS-SEC schema continues to be directly explained by the theoretically important job characteristics emphasized by the Goldthorpe model, but overall suggests the concept of asset specificity does overlap with analytical tasks, suggesting some theoretical refinement may be needed here.

The Goldthorpe model, job characteristics and employment relations

In the second step of the analysis, the relationships between job characteristics and several employment relations indicators are examined using multivariate regression techniques appropriate to each outcome of interest. Several standard control variables are included in all models (listed in Table 2). This exercise provides evidence on whether the underlying job characteristics highlighted by the Goldthorpe model directly predict employment relations in expected ways (which has never been demonstrated before) and the extent to which these are through the types of job tasks. Again, this is achieved using the KHB method.

Table 2.

The Goldthorpe model, job tasks and employment relations indicators.

	(1)	(2)	(3)	(4)	(5)
	Hourly pay (logarithm)	Employer pension	Dismissal scale	Unemployed (last 5 years)	Control over working time
Monitoring difficulty
Total	0.084^***	0.056	−0.118	−0.198^*	0.451^***
	(6.75)	(0.75)	(–1.71)	(–2.10)	(7.27)
Direct	0.078^***	0.036	−0.115	−0.189^*	0.428^***
	(6.22)	(0.48)	(–1.66)	(–1.99)	(6.88)
Indirect	0.006	0.0203	−0.003	−0.009	0.0238
	(1.30)	(1.50)	(–0.59)	(–0.86)	(1.63)
Via routine tasks	0.005	0.018	−0.001	−0.002	0.016
Via analytical tasks	0.001	0.002	−0.001	−0.007	0.008
% Via routine	6.3	32.1	1.2	1.0	3.5
% Via analytical	0.97	4.0	1.5	3.5	1.7
Asset specificity
Total	0.100^***	0.295^***	−0.190^***	−0.174^*	0.172^***
	(10.82)	(4.92)	(–3.36)	(–2.43)	(3.55)
Direct	0.057^***	0.186^**	−0.165^*	−0.072^*	0.032^**
	(5.45)	(2.75)	(–2.49)	(–1.98)	(2.59)
Indirect	0.042^***	0.109^***	−0.026	−0.102^*	0.140^***
	(6.29)	(3.29)	(–0.86)	(–2.56)	(5.04)
Via routine tasks	0.006	0.028	−0.001	−0.003	0.022
Via analytical tasks	0.036	0.081	−0.015	−0.100	0.117
% Via routine	6.0	9.5	1.1	1.6	12.9
% Via analytical	36.7	27.5	13.4	57.1	68.2
Controls	Yes	Yes	Yes	Yes	Yes
Estimation method	OLS	Logit	OLS	Logit	Ordered logit
R²/pseudo-R²	0.41	0.26	0.14	0.09	0.12
N	2072	2398	2398	2398	2398

Source: Employees aged 20 to 65 in the 2012 British Skills and Employment Survey (BSES).

Notes: Decomposition into direct and indirect effects derived using the Karlson et al. (2012) procedure. Controls: age; female; whether belong to an ethnic group; whether married; whether hold a degree; whether a union member; and whether part-time. Log odds ratios reported in Columns 2, 4 and 5. T-/z-statistics in parentheses.

Statistical significance: *p < 0.05, **p < 0.01, ***p < 0.001 (except for indirect effects for which standard errors cannot be calculated).

OLS, ordinary least square.

As implied in Goldthorpe (2007), we expect monitoring difficulty and asset specificity to relate to indicators of employment relations in different ways, depending on whether an indicator is more of an indicator for a service relationship or a labour contract. Table 2 appears to demonstrate this. Examining total effects first, those indicators relating to longer-term employment relations that aim to harness and develop asset specificity (pension and ease of dismissal) have significant positive and negative associations, respectively, with asset specificity, but not with monitoring difficulty. On the other hand, control over start and end times are predicted by monitoring difficulty, but not asset specificity, suggesting employers impose greater control when labour inputs are easier to define. Hourly pay and whether experienced unemployment – perhaps the most critical indicators of labour market inequality considered here – are predicted by both job characteristics in ways expected by the model. When examining the extent to which significant total effects are confounded by job tasks, if little or no significant confounding is found, then the theoretical basis of the Goldthorpe remains valid in predicting aspects of employment relations. In all cases, significant direct effects (not through job tasks) for monitoring difficulty and asset specificity remain. However, again we find asset specificity is confounded the most by job tasks, and again by analytical tasks ranging from one- to two-thirds. Again, this suggests that the role of asset specificity in the Goldthorpe model may be impinged by developments in technology, and this suggests this aspect may need some refinement as technology continues to advance.

An old model of social class?

Monitoring difficulty and asset specificity are the two salient job characteristics highlighted by the Goldthorpe model as the basis for social classes. This article extends previous research by considering how these relate to developments of technology by examining two further job characteristics – the extent to which jobs are routine and the extent to which their tasks are analytical – central to emerging technology-based explanations for labour market inequality. While job tasks confound monitoring difficulty in predicting NS-SEC membership and employment relations to a small extent, a more substantial role is found for the role of analytical tasks confounding asset specificity. Asset specificity is essentially an indicator of the replaceability of an employee with equivalent human assets in the wider labour market. The findings in this article suggest that the concept may need some refinement in the future to account for replaceability by technology as well as in determining class positions and the types of employment relationships offered by employers.

It must be stressed that job tasks do not completely confound the job characteristics emphasized by the Goldthorpe model, even in the case of analytical tasks with respect to asset specificity. Thus, the NS-SEC schema still represents a useful analytical tool in describing broad occupational groups which share certain work characteristics and employment relations. It also suggests that the theoretical basis is still fairly sound. However, the partial confounding of asset specificity by analytical tasks highlights that the underlying basis to social class may be shifting.

To an extent, the fact that classes may not mean the same thing over time and are in need of refreshing from time to time is something of a trivial point if they are merely descriptive tools. However, if class categories are to explain the dynamics of over time shifts in inequality, it is less than trivial, as we must also assume the theoretical basis is more or less constant. This is a strong assumption given developments in technology and shifts in the occupational structure. One response to this general problem is to abandon models of occupational class altogether in favour of models incorporating non-occupational factors (Savage et al., 2013 c.f. Crompton, 2010). Another, implied by this article, and other research which suggests a strengthening of occupational inequalities (Williams, 2013), is that future research would benefit from analysing job characteristics measured directly, to better understand these trends. By examining job characteristics measured directly, we can better distinguish the inequality-producing processes and, crucially, their over time dynamics. Relying on aggregated schemas of any kind comes at the cost of tapping into the dynamic inequality-producing processes that they represent.¹³ On the other hand, collecting data on job characteristics is costly and demanding. However, there may be a middle ground. Sociologists could take note from the recent task-based literature which aggregates mean job characteristics scores to the detailed occupational-level and maps them into other social surveys as explanatory variables for the sorts of technology-related factors emphasized by these theories. Future sociological research, then, may want to extract monitoring difficulty and asset specificity from surveys such as the BSES to examine their theories more directly in a similar way.¹⁴

Footnotes

Acknowledgements

I would like to thank Ying Zhou for helpful comments on an earlier draft. I would also like to thank the Editor and three anonymous referees for helpful comments.

Funding

The author received no financial support for the research, authorship and/or publication of this article.

Notes

Mark Williams is a Lecturer in HRM and Employment Relations at the Department of People and Organisations, University of Surrey. Prior to that he was a Fellow at the Department of Management, LSE. He holds a DPhil in Sociology from Nuffield College, Oxford. His main areas of interest are on pay and occupational inequality, work quality and employment relations.

References

Acemoglu

Autor

(2011) Skills, tasks and technologies: implications for employment and earnings. In: Ashenfelter

Card

(eds) Handbook of Labor Economics, vol. 4b. London: Elsevier, 1043–171.

Autor

(2013) The ‘task approach’ to labour markets: an overview. Journal of Labour Market Research 46(3): 185–99.

Autor

Handel

(2013) Putting tasks to the test: human capital, job tasks, and wages. Journal of Labor Economics 31(2): S59–96.

Chatzitheochari

Arber

(2009) Lack of sleep, work and the long hours culture: evidence from the UK Time Use Survey. Work, Employment and Society 23(1): 30–48.

Crompton

(2010) Class and employment. Work, Employment and Society 24(1): 9–26.

Evans

(1992) Testing the validity of the Goldthorpe class schema. European Sociological Review 8(3): 211–32.

Evans

Mills

(1998) Identifying class structure: a latent class analysis of the criterion-related and construct validity of the Goldthorpe class schema. European Sociological Review 14(1): 87–106.

Evans

Mills

(2000) In search of the wage-labour/service contract: new evidence on the validity of the Goldthorpe class schema. British Journal of Sociology 51(4): 641–61.

Felstead

Gallie

Inanc

Green

(2014) Skills and Employment Surveys Series Dataset, 1986, 1992, 1997, 2001, 2006, and 2012 (Computer file), 2nd Edition. Colchester: UK Data Archive (distributor).

10.

Goldthorpe

(2007) Social class and the differentiation of employment contracts. In: Goldthorpe

(ed.) On Sociology: Volume Two: Illustration and Retrospect. Stanford, CA: Stanford University Press, 101–24.

11.

Goldthorpe

McKnight

(2006) The economic basis of social class. In: Morgan

Grusky

Fields

(eds) Mobility and Inequality: Frontiers of Research in Sociology and Economics. Stanford, CA: Stanford University Press, 109–36.

12.

Goos

Manning

(2007) Lousy and lovely jobs: the rising polarization of work in Britain. The Review of Economics and Statistics 89(1): 118–33.

13.

Karlson

Holm

Breen

(2012) Comparing regression coefficients between same-sample nested models using logit and probit: a new method. Sociological Methodology 42(1): 286–313.

14.

Koumenta

Williams

(2016) An Anatomy of Zero-Hour Contracts in the United Kingdom. London: CIPD.

15.

Lambert

Bihagen

(2014) Using occupation-based social classifications. Work, Employment and Society 28(3): 481–94.

16.

Liu

Grusky

(2013) The payoff to skill in the third industrial revolution. American Journal of Sociology 118(5): 1130–374.

17.

McGovern

Hills

Mills

White

(2007) Market, Class, and Employment. Oxford: Oxford University Press.

18.

Mills

(2014) The Great British Class fiasco: a comment on Savage et al. Sociology 48(3): 437–44.

19.

Oesch

(2013) Occupational Change in Europe: How Technology and Education Transform the Job Structure. Oxford: Oxford University Press.

20.

Rose

Harrison

(2010) Social Class in Europe: An Introduction to the European Socio-Economic Classification. London: Routledge.

21.

Rose

Pevalin

(2003) A Researcher’s Guide to the National Statistics Socio-Economic Classification. London: SAGE.

22.

Rose

Pevalin

(2005) The National Statistics Socio-Economic Classification: Origins, Development and Use. Basingstoke: Palgrave Macmillan.

23.

Savage

Devine

Cunningham

Taylor

Yaojun

Hjellbrekke

(2013) A new model of social class? Findings from the BBC’s Great British Class Survey Experiment. Sociology 47(2): 219–50.

24.

Tåhlin

(2007) Class clues. European Sociological Review 23(5): 557–72.

25.

Warren

(2015) Work-time underemployment and financial hardship: class inequalities and recession in the UK. Work, Employment and Society 29(2): 191–212.

26.

Williams

(2013) Occupations and British wage inequality, 1970s–2000s. European Sociological Review 29(4): 841–57.

27.

Zhou

(2015) Occupational inequality in pay and task discretion: a multilevel approach. In: National Institute of Economic and Social Research seminar, London, 21 April.