Implementation of Review Recommendations and Some Further Comments

2.
Implementation of Review Recommendations and Some Further Comments

2.1
Implementation

2.1.1 The most important element of the Phase 2 Interim Report concerned the proposed conceptual basis, properties, relation to existing SECs, operationalisation and other characteristics of a revised SEC. Recommendations as set out below were made. So far as is possible at this stage of the Review, these have been implemented in an interim version of the revised SEC. This interim version has been subject to initial validation tests using data collected in the ONS Omnibus Survey (see O’Reilly and Rose, 1997a for details) and further validation work in Phase 3 will be undertaken using the recently and especially collected data from the Labour Force Survey (see Sections 3 and 4 below).

2.1.2 Conceptual basis for the SEC: in terms of its conceptual basis, the interim revised SEC follows a well-defined sociological position that employment relations and conditions are central to delineating the structure of socio-economic positions in modern societies (see for example, Erikson and Goldthorpe, 1992; and cf. Goldthorpe, 1980; Lockwood, 1958/1989). As we noted in both of our previous reports (Rose, 1995 and 1996a), SEG speaks theory without knowing it and is therefore already amenable to this conception, capturing the essential elements of a truly social scientific SEC quite well. Thus the interim revised SEC makes explicit the latent conception which underlies SEG. Moreover, as we shall see, the conceptual basis for the revised SEC is one for which there is already a good deal of impressive criterion and construct validation. Appendix 1 gives a diagrammatic representation of the way in which the revised schema is derived.

2.1.3 Properties of a revised SEC: the interim revised SEC has the hierarchical or nested properties envisaged in Phase 1, i.e. a long version of the SEC (see Appendix 2) which can be collapsed into a smaller number of categories (see Appendix 3) for analytic purposes. However, unlike SC, the collapsed version need not necessarily be an ordinal measure; both ordinal and nominal collapsed versions will be made available, so that analysts can make choices about which they use, depending upon the requirements of theory and analysis.

2.1.4 Relation of the revised SEC to current SECs: since SEG is closer than SC to a measure of employment relations and conditions, the revised SEC will be, in its long version, as similar as possible to the current SEG. In its collapsed version it will resemble SC (see Appendices 1-3). Because of the close relationship between SEG and SC (Martin, 1996), this strategy will yield high continuity between the revised SEC and both of the current classifications.

2.1.5 Creating the revised SEC: it should be noted that, with the exception of agreed minor changes to the Census Enumeration Form, the data required and the method used for creating the revised SEC from the Census and from social surveys will be the same as is currently required for SC and SEG. It will, of course, be possible to produce a version of the SEC for use with registration and administrative data where only occupation is recorded.

2.1.6 Population coverage: in response to user demand, it is important to include within the revised SEC as many as possible of those adults not in paid employment. To achieve this it may be necessary for some purposes to have a special category for the ‘never worked’. Recent research by Marshall et al (1996) demonstrates that classifying individuals not currently in paid employment by their last main job is a satisfactory procedure, even for those who have been out of the workforce for many years. Moreover, research by Arber using GHS data demonstrates strong class gradients in ill-health for unemployed men and the retired. Arber has concluded that the non-employed should not be separately treated as a group within analyses of health variations because there are stronger class inequalities among non-employed than employed men of working age. For the retired, class (in terms of last main occupation) continues to be associated with health throughout retirement (see Arber, 1996a, 1997). It was for this reason that we recommended that both the Census and government surveys collect information on last main job for all those not in paid employment.

2.2
Some further comments on recommendations

2.2.1 Conceptual issues: the choice of employment relations and conditions as the conceptual basis of the SEC was discussed in the Phase 2 Interim Report (and cf Rose, 1996b). A number of government department representatives subsequently asked for further explanation of certain issues arising from this conceptual model. This is provided in Paras 2.2.4 - 2.2.7. As noted above (Para 2.1.2), SEG is already differentiated by employment relations and for this reason it is the backbone of the interim SEC. Our preliminary revision of SEG - using ONS Omnibus data - has attempted to make explicit what is latent in SEG categories by reference to employment status characteristics that are widely recognised as significant in the literature (such as mode of payment, promotion prospects and autonomy) and are partially defining features of the Goldthorpe class schema.

2.2.2 <The Goldthorpe schema: the Interim Report also explained why we have not recommended the adoption of the Goldthorpe schema as currently operationalised. In brief, apart from the important issues of continuity with the current SECs, we must recall that the Goldthorpe schema, as operationalised for the UK, was never fully validated (in criterion terms) from the beginning because the data required (on employment relations and conditions at SOC OUG level for example) were not available. Therefore there was always an intuitive and inductive element to the allocation of occupations to Goldthorpe’s classes. This was why the Review Committee arrived at the conclusion that proper validation in terms of the conceptual construction and operationalisation of the revised SEC, both now and in the future, required new data on employment relations and conditions at OUG level (see Section 3 and cf. O’Reilly and Rose, 1997a). Finally, the usefulness of the Goldthorpe schema’s concept of the ‘service class’ for identifying class boundaries is something that we shall wish to examine more closely in the light of the LFS data (see para 4.2.4 below).

2.2.3 Operationalisation and maintenance of the revised SEC: although the long version of the revised SEC will be based on SEG, it is not possible to produce the final version of the SEC by simple revision of SEG categories. This is largely because some categories are problematic in terms of homogeneity and/or size (especially SEGs 5, 6, 7 and 13-16; cf Table 2 below). The operationalisation of the revised SEC requires the collection of primary data on employment relations and conditions for the 371 OUGs of the SOC. To achieve this, the Validation Group agreed on a relevant set of indicators of employment relations and conditions and, using the ONS Omnibus Survey, pre-tested these. The ONS data were then subjected to validation analyses as discussed in Section 3. On the basis of these analyses a final set of seven questions was chosen for inclusion on the December 1996-February 1997 Quarter of the Labour Force Survey (LFS). Analysis of these data will be central to Phase 3 in order to produce a final version of the revised SEC. This type of exercise could then be repeated inter-censally for SEC validation and revision purposes.

2.2.4 Causal narratives: why do we need an SEC conceived and constructed in the manner we have recommended? One of the strengths of the approach we have taken in this Review, indeed its underlying principle, is that our revised SEC offers not necessarily improved statistical associations over the current SECs, but that it lends itself to the possibility of explaining the associations we find. Because we know what our proposed SEC is measuring - employment relations and conditions, i.e. aspects of the work situation and the labour contract - we can construct causal narratives which specify how the SEC links to a range of outcomes via a variety of intervening variables (see Breen and Rottman, 1995; Marshall 1997: Introduction; Wilkinson, 1997: 593). This is decidedly not the case for the current government SECs, since it is not clear what they measure. They may show statistical associations with dependent variables of interest, but they do not lend themselves to causal explanations (see Bartley et al, 1996).

2.2.5 The example of health inequalities: one of the major uses of SECs has been in studies of fertility, morbidity and mortality, i.e. as a means of obtaining a macro or societal perspective on these issues. Our revised SEC defines structural positions which can be seen conceptually to exist independently of the individuals who occupy those positions at any particular time. The positions condition and shape the lives of their occupants. Thus, for example, health inequalities are differences between class categories in respect of morbidity and mortality. The study of these inequalities makes the social factors in the production of health outcomes more visible. We are therefore linking health with social organisation. This is vital for a range of public policy and monitoring issues (e.g. WHO targets for reducing social differences in health). Of course one can use other independent variables than class to study health and other outcomes of interest, for examples, income, education, housing, consumption, but none of these alternatives is designed to capture the basic structuring principles of society in the way that social class does. Thus when we pose questions about how the social structure shapes outcomes, social class is of prime importance. Moreover we need to keep the idea of social class analytically distinct from the possible consequences which the occupancy of a position may give rise to, e.g. income or housing. This will allow us to examine the mechanisms which link class to outcomes. The same is true for individual attributes which are necessary for the occupation of positions, e.g. education or skill level. Therefore, SECs should not be based on measures of skill, for example, although they will tend to correlate highly with it (see, for example, Gallie, 1995).

2.2.6 The Whitehall Studies: Breen and Rottman (1995) have recently pointed to the need to ‘hypothesise and test a number of different intervening variables that would represent alternative mechanisms linking class and outcome’, i.e. specifying causal narratives (and cf. Wilkinson, 1997). Phase 3 will pursue this suggestion wherever possible, building on a solid basis of prior work in this direction. For example, there is growing evidence that the amount of control and autonomy a person has at work are important factors in explaining heart disease (Bosma et al, 1997). The ‘prospective perspective' associated with secure, career employment among top managers and professionals is associated with greater control and autonomy at work, more self-esteem, greater self-care with regard to factors such as diet and exercise, more choice over medical treatment and so on. This we have learned, for example, from the Whitehall Studies which show that, contrary to popular belief, it is those at the bottom of employment hierarchies who are most stressed (Marmot et al, 1991; Bosma et al, 1997; and cf. Davey-Smith et al, 1997).

2.2.7 What we require, therefore, are more multivariate analyses which show how class effects are mediated via specific intervening variables. How class has its effect will vary according to what it is we wish to explain. We must thus construct and test different models designed to link what the SEC measures to a range of different outcomes (Note 2). These comments lead naturally to a discussion of the work we have undertaken in the past year - especially the selection of appropriate indicators of employment relations and conditions, the design and pre-testing of relevant questions for use on the ONS Omnibus Survey, the analysis of these data and the final choice of questions to be asked on the LFS to provide the data required for Phase 3.

2.	Implementation of Review Recommendations and Some Further Comments
2.1	Implementation
2.1.1	The most important element of the Phase 2 Interim Report concerned the proposed conceptual basis, properties, relation to existing SECs, operationalisation and other characteristics of a revised SEC. Recommendations as set out below were made. So far as is possible at this stage of the Review, these have been implemented in an interim version of the revised SEC. This interim version has been subject to initial validation tests using data collected in the ONS Omnibus Survey (see O’Reilly and Rose, 1997a for details) and further validation work in Phase 3 will be undertaken using the recently and especially collected data from the Labour Force Survey (see Sections 3 and 4 below).
2.1.2	Conceptual basis for the SEC: in terms of its conceptual basis, the interim revised SEC follows a well-defined sociological position that employment relations and conditions are central to delineating the structure of socio-economic positions in modern societies (see for example, Erikson and Goldthorpe, 1992; and cf. Goldthorpe, 1980; Lockwood, 1958/1989). As we noted in both of our previous reports (Rose, 1995 and 1996a), SEG speaks theory without knowing it and is therefore already amenable to this conception, capturing the essential elements of a truly social scientific SEC quite well. Thus the interim revised SEC makes explicit the latent conception which underlies SEG. Moreover, as we shall see, the conceptual basis for the revised SEC is one for which there is already a good deal of impressive criterion and construct validation. Appendix 1 gives a diagrammatic representation of the way in which the revised schema is derived.
2.1.3	Properties of a revised SEC: the interim revised SEC has the hierarchical or nested properties envisaged in Phase 1, i.e. a long version of the SEC (see Appendix 2) which can be collapsed into a smaller number of categories (see Appendix 3) for analytic purposes. However, unlike SC, the collapsed version need not necessarily be an ordinal measure; both ordinal and nominal collapsed versions will be made available, so that analysts can make choices about which they use, depending upon the requirements of theory and analysis.
2.1.4	Relation of the revised SEC to current SECs: since SEG is closer than SC to a measure of employment relations and conditions, the revised SEC will be, in its long version, as similar as possible to the current SEG. In its collapsed version it will resemble SC (see Appendices 1-3). Because of the close relationship between SEG and SC (Martin, 1996), this strategy will yield high continuity between the revised SEC and both of the current classifications.
2.1.5	Creating the revised SEC: it should be noted that, with the exception of agreed minor changes to the Census Enumeration Form, the data required and the method used for creating the revised SEC from the Census and from social surveys will be the same as is currently required for SC and SEG. It will, of course, be possible to produce a version of the SEC for use with registration and administrative data where only occupation is recorded.
2.1.6	Population coverage: in response to user demand, it is important to include within the revised SEC as many as possible of those adults not in paid employment. To achieve this it may be necessary for some purposes to have a special category for the ‘never worked’. Recent research by Marshall et al (1996) demonstrates that classifying individuals not currently in paid employment by their last main job is a satisfactory procedure, even for those who have been out of the workforce for many years. Moreover, research by Arber using GHS data demonstrates strong class gradients in ill-health for unemployed men and the retired. Arber has concluded that the non-employed should not be separately treated as a group within analyses of health variations because there are stronger class inequalities among non-employed than employed men of working age. For the retired, class (in terms of last main occupation) continues to be associated with health throughout retirement (see Arber, 1996a, 1997). It was for this reason that we recommended that both the Census and government surveys collect information on last main job for all those not in paid employment.
2.2	Some further comments on recommendations
2.2.1	Conceptual issues: the choice of employment relations and conditions as the conceptual basis of the SEC was discussed in the Phase 2 Interim Report (and cf Rose, 1996b). A number of government department representatives subsequently asked for further explanation of certain issues arising from this conceptual model. This is provided in Paras 2.2.4 - 2.2.7. As noted above (Para 2.1.2), SEG is already differentiated by employment relations and for this reason it is the backbone of the interim SEC. Our preliminary revision of SEG - using ONS Omnibus data - has attempted to make explicit what is latent in SEG categories by reference to employment status characteristics that are widely recognised as significant in the literature (such as mode of payment, promotion prospects and autonomy) and are partially defining features of the Goldthorpe class schema.
2.2.2	<The Goldthorpe schema: the Interim Report also explained why we have not recommended the adoption of the Goldthorpe schema as currently operationalised. In brief, apart from the important issues of continuity with the current SECs, we must recall that the Goldthorpe schema, as operationalised for the UK, was never fully validated (in criterion terms) from the beginning because the data required (on employment relations and conditions at SOC OUG level for example) were not available. Therefore there was always an intuitive and inductive element to the allocation of occupations to Goldthorpe’s classes. This was why the Review Committee arrived at the conclusion that proper validation in terms of the conceptual construction and operationalisation of the revised SEC, both now and in the future, required new data on employment relations and conditions at OUG level (see Section 3 and cf. O’Reilly and Rose, 1997a). Finally, the usefulness of the Goldthorpe schema’s concept of the ‘service class’ for identifying class boundaries is something that we shall wish to examine more closely in the light of the LFS data (see para 4.2.4 below).
2.2.3	Operationalisation and maintenance of the revised SEC: although the long version of the revised SEC will be based on SEG, it is not possible to produce the final version of the SEC by simple revision of SEG categories. This is largely because some categories are problematic in terms of homogeneity and/or size (especially SEGs 5, 6, 7 and 13-16; cf Table 2 below). The operationalisation of the revised SEC requires the collection of primary data on employment relations and conditions for the 371 OUGs of the SOC. To achieve this, the Validation Group agreed on a relevant set of indicators of employment relations and conditions and, using the ONS Omnibus Survey, pre-tested these. The ONS data were then subjected to validation analyses as discussed in Section 3. On the basis of these analyses a final set of seven questions was chosen for inclusion on the December 1996-February 1997 Quarter of the Labour Force Survey (LFS). Analysis of these data will be central to Phase 3 in order to produce a final version of the revised SEC. This type of exercise could then be repeated inter-censally for SEC validation and revision purposes.
2.2.4	Causal narratives: why do we need an SEC conceived and constructed in the manner we have recommended? One of the strengths of the approach we have taken in this Review, indeed its underlying principle, is that our revised SEC offers not necessarily improved statistical associations over the current SECs, but that it lends itself to the possibility of explaining the associations we find. Because we know what our proposed SEC is measuring - employment relations and conditions, i.e. aspects of the work situation and the labour contract - we can construct causal narratives which specify how the SEC links to a range of outcomes via a variety of intervening variables (see Breen and Rottman, 1995; Marshall 1997: Introduction; Wilkinson, 1997: 593). This is decidedly not the case for the current government SECs, since it is not clear what they measure. They may show statistical associations with dependent variables of interest, but they do not lend themselves to causal explanations (see Bartley et al, 1996).
2.2.5	The example of health inequalities: one of the major uses of SECs has been in studies of fertility, morbidity and mortality, i.e. as a means of obtaining a macro or societal perspective on these issues. Our revised SEC defines structural positions which can be seen conceptually to exist independently of the individuals who occupy those positions at any particular time. The positions condition and shape the lives of their occupants. Thus, for example, health inequalities are differences between class categories in respect of morbidity and mortality. The study of these inequalities makes the social factors in the production of health outcomes more visible. We are therefore linking health with social organisation. This is vital for a range of public policy and monitoring issues (e.g. WHO targets for reducing social differences in health). Of course one can use other independent variables than class to study health and other outcomes of interest, for examples, income, education, housing, consumption, but none of these alternatives is designed to capture the basic structuring principles of society in the way that social class does. Thus when we pose questions about how the social structure shapes outcomes, social class is of prime importance. Moreover we need to keep the idea of social class analytically distinct from the possible consequences which the occupancy of a position may give rise to, e.g. income or housing. This will allow us to examine the mechanisms which link class to outcomes. The same is true for individual attributes which are necessary for the occupation of positions, e.g. education or skill level. Therefore, SECs should not be based on measures of skill, for example, although they will tend to correlate highly with it (see, for example, Gallie, 1995).
2.2.6	The Whitehall Studies: Breen and Rottman (1995) have recently pointed to the need to ‘hypothesise and test a number of different intervening variables that would represent alternative mechanisms linking class and outcome’, i.e. specifying causal narratives (and cf. Wilkinson, 1997). Phase 3 will pursue this suggestion wherever possible, building on a solid basis of prior work in this direction. For example, there is growing evidence that the amount of control and autonomy a person has at work are important factors in explaining heart disease (Bosma et al, 1997). The ‘prospective perspective' associated with secure, career employment among top managers and professionals is associated with greater control and autonomy at work, more self-esteem, greater self-care with regard to factors such as diet and exercise, more choice over medical treatment and so on. This we have learned, for example, from the Whitehall Studies which show that, contrary to popular belief, it is those at the bottom of employment hierarchies who are most stressed (Marmot et al, 1991; Bosma et al, 1997; and cf. Davey-Smith et al, 1997).
2.2.7	What we require, therefore, are more multivariate analyses which show how class effects are mediated via specific intervening variables. How class has its effect will vary according to what it is we wish to explain. We must thus construct and test different models designed to link what the SEC measures to a range of different outcomes (Note 2). These comments lead naturally to a discussion of the work we have undertaken in the past year - especially the selection of appropriate indicators of employment relations and conditions, the design and pre-testing of relevant questions for use on the ONS Omnibus Survey, the analysis of these data and the final choice of questions to be asked on the LFS to provide the data required for Phase 3.