Empirical Work in Phase 2

3.
Empirical Work in Phase 2

3.1
Project 3 - The 371 Database

3.1.1 In this section we discuss progress to date on projects 3, 4 and 5. We begin with project 3. Under a separate contract from ESRC, Elias and McKnight have produced a database on the 371 OUGs of the SOC. This database should be useful for SOC revision as well as for the next phase of the Review. In addition, information from the LFS employment relations and conditions variables will be added to the database as part of Phase 3. A report on this project has already been provided (see Rose, 1996a: Paras 4.1 - 4.9). Subsequently, some minor alterations and additions (including the earnings data reported in Para 3.4.5 below) have been made to the database (see McKnight and Elias, 1996).

3.2
Project 4 - The Census Matrix

3.2.1 Obviously, before we could construct the interim SEC from the Omnibus data, we had to produce the necessary Census Matrix relating OUGs to employment statuses. We have, therefore, made the maximum progress possible at this stage on project 4. However, this exercise revealed some anomalies in the Census Matrix design and the rules for invalid combinations of OUG and employment status.

3.2.2 The Census Matrix is designed so that combinations of OUG and employment status are each allocated to a category of SC and SEG (OPCS, 1991:38-53). However, combinations which are not expected or not allowed to occur are left blank and ONS has provided guidelines for dealing with these. In our experience, in some data sets almost 10% of the population need sorting in this way because the combination of their occupational group and their employment status is invalid as far as the Census Matrix is concerned. This raises two specific issues. First, changes in the nature of employment have meant that some of the invalid combinations are now valid and need to be recognised as such in the Census Matrix. The other issue relates to sorting those who fall into genuinely invalid cells. Currently, the logic behind the rules for sorting is that the occupational code is correct and the employment status is wrong. In such cases the employment status of the respondent is changed, according to priority rules, until the combination is valid (Bushnell, 1994). We would suggest that, rather than determining which combinations can exist and then changing the rest to an alternative employment status, a more reasonable approach would be to determine which few combinations are clearly not valid (for good reason) and to force only these to a different employment status.

3.3
Project 5 - Measuring employment relations and conditions: The Omnibus Surveys

3.3.1 The Steering Committee’s conclusion that survey data were required on employment relations and conditions at OUG level raised a crucial issue. Validation studies conducted on data collected in the 1980s have all concluded that the conceptual basis we propose is fundamentally sound, i.e. that it is possible to distinguish between classes in these terms (see Evans, 1992; Evans, 1996; Evans and Mills, 1996 and 1997; Birkelund, Goodman and Rose, 1996). However, the development of the so-called ‘flexible’ labour market may be thought to have complicated the picture so that previously reliable measures of employment relations and conditions may no longer be so discriminating between SEC categories as they once were (see O'Reilly and Rose, 1997b, for more details). This was a matter which clearly concerned government department representatives when the Interim Report was presented to them in April 1996.

3.3.2 In order to test the proposition about the effects on employment relations of a flexible labour market, we chose a variety of questions taken in the main from previous national surveys, such as Social Class in Modern Britain (Marshall et al, 1988) and Employment in Britain (Gallie and White, 1993; cf Gallie, 1996). These questions were then carried on the ONS Omnibus Surveys for April, May and July 1996. The Omnibus Surveys also allowed us to collect data necessary to construct the interim version of the revised SEC. Tables 1 - 5 give the distribution of the Omnibus data, for men and women, for the long versions of the revised SEC and SEG (Tables 1 and 2); for collapsed versions of SEC and SEG (Note 3) (Tables 3 and 4); and for SC (Table 5).

3.3.3 Combining the data from the three month’s Omnibus Surveys yielded the following results. Of 5,775 respondents, 57 per cent were currently in paid work, and 86 per cent of these were employees. Only the 2,822 respondents in paid work as employees were asked our questions on employment relations and conditions (Note 4). Responses were very good: the questions were answered fully by all respondents to whom they were directed with less than 1 per cent falling into the ‘don’t know’ or ‘other’ categories. The only exception to this was with the question which asked ‘are you on an incremental pay scale?’, to which 2 per cent of respondents replied that they did not know.

3.3.4 There was good reason to believe that there were three conceptually separable, though empirically correlated, respects in which employment relations remain differentiated according to whether, in Goldthorpe’s terms, a service relationship, labour contract or intermediate employment relationship exists. These are: (1) forms of remuneration; (2) promotion opportunities; and (3) autonomy, especially as regards time (see Goldthorpe, 1996). However, we also used other employment relations measures in the surveys. The Omnibus data seem to confirm our hypotheses. That is, as shown in Table 6, employment relations and conditions measures do vary systematically across the categories of the interim SEC in ways consistent with the underlying conceptual schema. Further details of our Omnibus analyses and the criterion validation work carried out under project 5 have been reported in separate papers (O’Reilly and Rose, 1997a and b).

3.3.5 The data from the Omnibus Surveys have also been used to test the criterion validity of the interim SEC (see O’Reilly and Rose, 1997b). The interim SEC is devised as a proxy measure for the employment conditions associated with different positions in the labour market. Criterion validation has involved assessing the relationship between the measure itself and some of the criterion variables for which it is a proxy. The Omnibus data have demonstrated that the interim SEC successfully discriminates between positions using a service relationship/intermediate/labour contract typology, and that it does so at least as well as the latest version of the Goldthorpe schema, as well as other categorical classifications.

3.3.6 In line with the recommendations of the Interim Report on Phase 2, the results of these analyses led us to propose (successfully) to ONS that the most discriminating of the employment relations and conditions variables should be carried on the LFS (Note 5). As discussed in Section 4, this will provide sufficient cases to analyse these variables at the OUG level and thus to produce a more definitive version of the revised SEC for use in the further validation studies which are the main element of Phase 3.

3.4
Other empirical work for Project 5

3.4.1 We have also examined the construct validity of the interim SEC using a variety of relevant data sets (GHS, LFS, and the 1% Longitudinal Census Study for examples).

3.4.2 The interim SEC and health outcomes: we have already referred, in Para 2.1.6, to the work of Arber and its relevance both to Phase 2 and the work of the Review as a whole. Arber has also produced other analyses of health data from the GHS for the Validation Group. In particular, a comparison has been made between the interim SEC, SC and (collapsed) SEG which shows that the interim SEC produces comparable health gradients to those of the existing SECs in terms of both limiting long-standing illness (LLI) and self-assessed health (see Arber, 1996b). Dale (1997) has undertaken similar analyses using the Census Sample of Anonymised Records. For men working full-time the interim SEC provides greater definition than SC. For example LLI ranges from 2.6 per cent for higher professionals etc., to 5.1 per cent for routine workers, with routine office workers having relatively high levels of LLI at 5 per cent. These figures compare with a range between 2.7 per cent for SC Class I to 4.9 per cent for SC Class V, with the figure for Class IIIN being 3.8 per cent. Dale’s analyses suggest that Class IIIN is grouping together men who vary considerably on their health status.

3.4.3 The interim SEC and mortality: an analysis has been carried out by Ray Fitzpatrick and Mel Bartley and her colleagues at City University, relating to the construct validity of the interim SEC in relation to mortality in England and Wales. It was postulated that similar patterns of relationships should be identified between categories of the interim SEC and mortality rates in adult males as had been observed from an earlier study using the Erikson-Goldthorpe schema (Bartley et al., 1996). The 1981 cohort of the OPCS Longitudinal Study consists of a one per cent random sample of the population of England and Wales linking the 1981 census to deaths. Occupational and employment status data were recoded according to the interim schema. An analysis of male deaths was performed with age adjusted relative risks (RR) of mortality over the following ten year period. Higher professionals and senior managers, lower professionals and junior managers were found to experience significantly lower than expected death rates (RR 0.70 and 0.94 respectively) and intermediate and other workers experienced significantly higher than expected death rates (RR 1.13 and 1.26). This analysis provides positive evidence for the construct validity of the interim SEC by distinguishing groups with different mortality risks and with patterns of differences similar to those observed by the Erikson-Goldthorpe schema. The latter schema is now extensively used in international analyses of social factors underlying health status differences in European societies. A more formal report of these analyses is shortly to be submitted to the Review committee.

3.4.4 The interim SEC and earnings: Elias and McKnight (1996) have analysed earnings data with reference to SOC OUGs. While earnings information is not relevant to the derivation of the interim SEC (since earnings do not form part of the revised schema’s conceptual basis), nevertheless Elias’ and McKnight’s analyses do serve to highlight certain anomalies which could be useful to examine once we have the LFS data for Phase 3. That is, we must examine whether OUGs which are anomalous in earnings terms are equally problematic when viewed from the perspective of employment relations and conditions. For their analyses, Elias and McKnight took data on average gross weekly earnings for full-time employees from the Quarterly Labour Force surveys (1993 quarter 2 to 1994 quarter 4). After adjusting for the effects of differences in the age structure and geographical location, these data were analysed by gender at the level of SOC unit groups to show occupational differentials within the categories of the interim social classification. This analysis highlighted a number of occupation unit groups which had markedly different earnings from the average within each category of the interim classification.

3.4.5 All of this validation work will be repeated in Phase 3 once we have derived the revised SEC from the LFS data. However, the work on Project 5 reported here is very encouraging in terms of both the validity and utility of our proposed schema.

3.	Empirical Work in Phase 2
3.1	Project 3 - The 371 Database
3.1.1	In this section we discuss progress to date on projects 3, 4 and 5. We begin with project 3. Under a separate contract from ESRC, Elias and McKnight have produced a database on the 371 OUGs of the SOC. This database should be useful for SOC revision as well as for the next phase of the Review. In addition, information from the LFS employment relations and conditions variables will be added to the database as part of Phase 3. A report on this project has already been provided (see Rose, 1996a: Paras 4.1 - 4.9). Subsequently, some minor alterations and additions (including the earnings data reported in Para 3.4.5 below) have been made to the database (see McKnight and Elias, 1996).
3.2	Project 4 - The Census Matrix
3.2.1	Obviously, before we could construct the interim SEC from the Omnibus data, we had to produce the necessary Census Matrix relating OUGs to employment statuses. We have, therefore, made the maximum progress possible at this stage on project 4. However, this exercise revealed some anomalies in the Census Matrix design and the rules for invalid combinations of OUG and employment status.
3.2.2	The Census Matrix is designed so that combinations of OUG and employment status are each allocated to a category of SC and SEG (OPCS, 1991:38-53). However, combinations which are not expected or not allowed to occur are left blank and ONS has provided guidelines for dealing with these. In our experience, in some data sets almost 10% of the population need sorting in this way because the combination of their occupational group and their employment status is invalid as far as the Census Matrix is concerned. This raises two specific issues. First, changes in the nature of employment have meant that some of the invalid combinations are now valid and need to be recognised as such in the Census Matrix. The other issue relates to sorting those who fall into genuinely invalid cells. Currently, the logic behind the rules for sorting is that the occupational code is correct and the employment status is wrong. In such cases the employment status of the respondent is changed, according to priority rules, until the combination is valid (Bushnell, 1994). We would suggest that, rather than determining which combinations can exist and then changing the rest to an alternative employment status, a more reasonable approach would be to determine which few combinations are clearly not valid (for good reason) and to force only these to a different employment status.
3.3	Project 5 - Measuring employment relations and conditions: The Omnibus Surveys
3.3.1	The Steering Committee’s conclusion that survey data were required on employment relations and conditions at OUG level raised a crucial issue. Validation studies conducted on data collected in the 1980s have all concluded that the conceptual basis we propose is fundamentally sound, i.e. that it is possible to distinguish between classes in these terms (see Evans, 1992; Evans, 1996; Evans and Mills, 1996 and 1997; Birkelund, Goodman and Rose, 1996). However, the development of the so-called ‘flexible’ labour market may be thought to have complicated the picture so that previously reliable measures of employment relations and conditions may no longer be so discriminating between SEC categories as they once were (see O'Reilly and Rose, 1997b, for more details). This was a matter which clearly concerned government department representatives when the Interim Report was presented to them in April 1996.
3.3.2	In order to test the proposition about the effects on employment relations of a flexible labour market, we chose a variety of questions taken in the main from previous national surveys, such as Social Class in Modern Britain (Marshall et al, 1988) and Employment in Britain (Gallie and White, 1993; cf Gallie, 1996). These questions were then carried on the ONS Omnibus Surveys for April, May and July 1996. The Omnibus Surveys also allowed us to collect data necessary to construct the interim version of the revised SEC. Tables 1 - 5 give the distribution of the Omnibus data, for men and women, for the long versions of the revised SEC and SEG (Tables 1 and 2); for collapsed versions of SEC and SEG (Note 3) (Tables 3 and 4); and for SC (Table 5).
3.3.3	Combining the data from the three month’s Omnibus Surveys yielded the following results. Of 5,775 respondents, 57 per cent were currently in paid work, and 86 per cent of these were employees. Only the 2,822 respondents in paid work as employees were asked our questions on employment relations and conditions (Note 4). Responses were very good: the questions were answered fully by all respondents to whom they were directed with less than 1 per cent falling into the ‘don’t know’ or ‘other’ categories. The only exception to this was with the question which asked ‘are you on an incremental pay scale?’, to which 2 per cent of respondents replied that they did not know.
3.3.4	There was good reason to believe that there were three conceptually separable, though empirically correlated, respects in which employment relations remain differentiated according to whether, in Goldthorpe’s terms, a service relationship, labour contract or intermediate employment relationship exists. These are: (1) forms of remuneration; (2) promotion opportunities; and (3) autonomy, especially as regards time (see Goldthorpe, 1996). However, we also used other employment relations measures in the surveys. The Omnibus data seem to confirm our hypotheses. That is, as shown in Table 6, employment relations and conditions measures do vary systematically across the categories of the interim SEC in ways consistent with the underlying conceptual schema. Further details of our Omnibus analyses and the criterion validation work carried out under project 5 have been reported in separate papers (O’Reilly and Rose, 1997a and b).
3.3.5	The data from the Omnibus Surveys have also been used to test the criterion validity of the interim SEC (see O’Reilly and Rose, 1997b). The interim SEC is devised as a proxy measure for the employment conditions associated with different positions in the labour market. Criterion validation has involved assessing the relationship between the measure itself and some of the criterion variables for which it is a proxy. The Omnibus data have demonstrated that the interim SEC successfully discriminates between positions using a service relationship/intermediate/labour contract typology, and that it does so at least as well as the latest version of the Goldthorpe schema, as well as other categorical classifications.
3.3.6	In line with the recommendations of the Interim Report on Phase 2, the results of these analyses led us to propose (successfully) to ONS that the most discriminating of the employment relations and conditions variables should be carried on the LFS (Note 5). As discussed in Section 4, this will provide sufficient cases to analyse these variables at the OUG level and thus to produce a more definitive version of the revised SEC for use in the further validation studies which are the main element of Phase 3.
3.4	Other empirical work for Project 5
3.4.1	We have also examined the construct validity of the interim SEC using a variety of relevant data sets (GHS, LFS, and the 1% Longitudinal Census Study for examples).
3.4.2	The interim SEC and health outcomes: we have already referred, in Para 2.1.6, to the work of Arber and its relevance both to Phase 2 and the work of the Review as a whole. Arber has also produced other analyses of health data from the GHS for the Validation Group. In particular, a comparison has been made between the interim SEC, SC and (collapsed) SEG which shows that the interim SEC produces comparable health gradients to those of the existing SECs in terms of both limiting long-standing illness (LLI) and self-assessed health (see Arber, 1996b). Dale (1997) has undertaken similar analyses using the Census Sample of Anonymised Records. For men working full-time the interim SEC provides greater definition than SC. For example LLI ranges from 2.6 per cent for higher professionals etc., to 5.1 per cent for routine workers, with routine office workers having relatively high levels of LLI at 5 per cent. These figures compare with a range between 2.7 per cent for SC Class I to 4.9 per cent for SC Class V, with the figure for Class IIIN being 3.8 per cent. Dale’s analyses suggest that Class IIIN is grouping together men who vary considerably on their health status.
3.4.3	The interim SEC and mortality: an analysis has been carried out by Ray Fitzpatrick and Mel Bartley and her colleagues at City University, relating to the construct validity of the interim SEC in relation to mortality in England and Wales. It was postulated that similar patterns of relationships should be identified between categories of the interim SEC and mortality rates in adult males as had been observed from an earlier study using the Erikson-Goldthorpe schema (Bartley et al., 1996). The 1981 cohort of the OPCS Longitudinal Study consists of a one per cent random sample of the population of England and Wales linking the 1981 census to deaths. Occupational and employment status data were recoded according to the interim schema. An analysis of male deaths was performed with age adjusted relative risks (RR) of mortality over the following ten year period. Higher professionals and senior managers, lower professionals and junior managers were found to experience significantly lower than expected death rates (RR 0.70 and 0.94 respectively) and intermediate and other workers experienced significantly higher than expected death rates (RR 1.13 and 1.26). This analysis provides positive evidence for the construct validity of the interim SEC by distinguishing groups with different mortality risks and with patterns of differences similar to those observed by the Erikson-Goldthorpe schema. The latter schema is now extensively used in international analyses of social factors underlying health status differences in European societies. A more formal report of these analyses is shortly to be submitted to the Review committee.
3.4.4	The interim SEC and earnings: Elias and McKnight (1996) have analysed earnings data with reference to SOC OUGs. While earnings information is not relevant to the derivation of the interim SEC (since earnings do not form part of the revised schema’s conceptual basis), nevertheless Elias’ and McKnight’s analyses do serve to highlight certain anomalies which could be useful to examine once we have the LFS data for Phase 3. That is, we must examine whether OUGs which are anomalous in earnings terms are equally problematic when viewed from the perspective of employment relations and conditions. For their analyses, Elias and McKnight took data on average gross weekly earnings for full-time employees from the Quarterly Labour Force surveys (1993 quarter 2 to 1994 quarter 4). After adjusting for the effects of differences in the age structure and geographical location, these data were analysed by gender at the level of SOC unit groups to show occupational differentials within the categories of the interim social classification. This analysis highlighted a number of occupation unit groups which had markedly different earnings from the average within each category of the interim classification.
3.4.5	All of this validation work will be repeated in Phase 3 once we have derived the revised SEC from the LFS data. However, the work on Project 5 reported here is very encouraging in terms of both the validity and utility of our proposed schema.