Phase 3

4.
Phase 3

4.1
Aims and timetable

4.1.1 ONS agreement to allow us to use the LFS to produce a more definitive allocation of OUG and employment status combinations to categories of the revised SEC effectively creates a third and final phase of the Review. Joint funding for this has been agreed between ESRC and ONS.

4.1.2 The aims of Phase 3 are (1) to analyse Labour Force Survey data on employment relations and conditions at the Occupation Unit Group level of the Standard Occupational Classification in order to operationalise the revised government social classification; (2) using a variety of datasets, to undertake validation studies with the revised classification; (3) to produce a new Census Matrix; (4) to bridge the current ONS social classifications to the new one; (5) to write a final report on all of this work for ESRC and the ONS; and (6) to provide users of the revised SEC with full details of its conceptual basis and construction. In undertaking this work we shall therefore be completing projects 4-6.

4.1.3 Phase 3 timetable: we understand that the LFS data will be available for analysis in April 1997. These data will then need to be subjected to the analyses set out below (Section 4.2) before the SEC can be finalised and validated (Section 4.3). Assuming that we do obtain the LFS data by April 1997, we estimate that it will take nine months to undertake the necessary analyses for the establishment of the revised SEC and the associated Census Matrix, as well as to complete further validation analyses using the revised SEC. Beyond this point, the proposed project 6 in Phase 2 - bridging between old and new SECs - and the application of the revised SEC to registration data will still need to be undertaken. This will take a further three months. Even beyond this point, further work might remain, for example in specifying the details of the SEC and its operational rules for inclusion in the next edition of SOC Volume 3. In addition, we believe it is most desirable that the results of our LFS analyses be included in a revised version of the 371 database. ESRC has agreed to fund this as part of its contribution to Phase 3. Three months is estimated for this task, to run concurrently with other work. A summary timetable for Phase 3 is given in Appendix 6.

4.2
The LFS Analyses

4.2.1 The analysis of LFS data is a relatively straightforward exercise, similar to what we have already done with the Omnibus data, but now with OUGs as the main focus. Essentially we wish to create SEC categories which are as homogeneous as possible with respect to employment relations and conditions as we have measured them on the LFS.

4.2.2 Continuity with existing SECs: in order to retain continuity with SEG, we need to attempt to use the LFS data to allocate a few large, problematic OUGs to the revised SEC rather than try to operationalise a whole new measure of social class. Given that the twenty largest OUGs account for 36 per cent of those in paid employment and 33 per cent of all employees; and that among full-time employees, the fifty largest OUGs account for 56 per cent of all cases, it will be readily apparent how important to the whole exercise is the proper allocation of these OUGs to the categories of the revised classification. Our initial efforts must therefore be to this end. We must also examine whether there is any evidence from the LFS that employment relations and conditions (as we define them) vary between full-time and part-time employees or between men and women in the same OUGs.

4.2.3 Refining the interim SEC: assuming (as the Omnibus and other validity exercises have demonstrated to date) that analysis of LFS data show that there are distinct groups or classes in terms of our employment relations indicators, and thus that the basic theory is sound, we can then use the LFS data to refine our interim SEC. That is, we can, where necessary, move individual OUGs from their current SEC categories to others, if the evidence indicates that this is appropriate. This cannot, however, be done wholly on the basis of statistical analysis because some OUGs are likely to be too heterogeneous with respect to employment relations. Equally, even with LFS data we will have too few cases in some OUGs to make statistically-driven decisions. We may also have to make some compromises necessary for continuity between the revised and existing SECs. No schema is perfect, and so we must be careful how this part of the exercise is handled. Consequently we will take expert advice before making final decisions.

4.2.4 Collapsing from long to short versions of the SEC: since presenting our Interim Report on Phase 2, we have made some small changes to the categories of the interim revised SEC as a result of our analyses of the Omnibus data (Note 6. However, the final decision on how best to collapse the long - SEG based - version of the revised SEC to the shorter version can only be made once we have completed our initial LFS analysis. In particular, this will involve a closer examination of the employment conditions of professionals by comparison with managers, and also those of lower managers by comparison with technicians and supervisors of office employees.

4.2.5 Category size: more pragmatically, we have to consider the relative size of the categories of the final collapsed version of the revised SEC. For example, some have already argued that category 7 - other workers - is too small at 6 per cent of employees and should therefore be combined with category 6 - currently intermediate workers. However, the beauty of the schema we are proposing is that it is flexible enough to allow researchers to make their own decisions about how to collapse from long to short SEC versions, provided they stay within the limits set by the schema’s conceptual basis.

4.3
Validation Exercises

4.3.1 The criterion validation aspect of Phase 3 (i.e. the continuation of project 5) involves ensuring that the revised classification is measuring what it is intended to measure. In our validation exercises we shall follow similar procedures to those developed by Evans (1992, 1996), and by Evans and Mills (1996), and especially the latent structure analysis techniques pioneered by Birkelund, Goodman and Rose (1996) and recently followed by Evans and Mills (1997). We shall examine both the criterion and construct validity of the revised SEC.

4.3.2 It has been agreed that the revised SEC should be based conceptually on employment relations and conditions. This sociological approach treats ‘class’ as a clearly-defined concept rather than as a theory tout court. What needs to be tested is that the instrument used, the revised SEC, fits the concept - that it measures what the concept purports to measure (see Evans and Mills, 1997). For Goldthorpe (1996) class analysis ‘explores the interconnections between positions defined by employment relations in labour markets and production units in different sectors of national economies’. It is these positions which are captured by the measurement of ‘class’. When the concept is operationalised it reflects these positions in the labour market. The positions themselves cannot easily be measured, yet the conditions which tend to cluster in distinctive mixes around the positions can. These conditions of employment form criteria related to the concept of class, and indicators of these will therefore be used to test the criterion validity of the revised SEC.

4.3.3 The results of the Omnibus data analyses already give us a high degree of optimism in this regard (see O’Reilly and Rose, 1997a and b). The definitive measure of the revised SEC which we devise with the aid of LFS data (and, following Goldthorpe’s comments (1996), data garnered from other sources on career patterns at OUG level and on pensions entitlements) will necessarily be subjected to further analyses.

4.3.4 Construct validity: it has been argued that criterion validity has limited usefulness in the social sciences (Carmines and Zeller, 1980). Where possible, one should also assess the construct validity of a measure. This involves assessing how the measure relates to other variables in ways predicted by theory, and in substantive areas of importance to government researchers and social scientists. The construct validation exercise will be theoretically driven, will use appropriate methodological procedures, and will ensure that other variables used have themselves been properly validated. Only under such conditions can we confidently conclude that correlation is evidence of construct validity. Here we shall be using the same datasets as are currently being examined - BHPS, GHS, the 1% LS etc.

4.3.5 As we have already noted, and most encouragingly from our perspective, the Goldthorpe schema has recently been positively evaluated by researchers interested in measuring socio-economic inequalities in health. Evidence from both the UK and Europe suggests that the Goldthorpe schema does have validity in these terms and may prove a better measure for these purposes than alternatives, including the current government social classifications (see, for example, Bartley et al, 1996; cf. Kunst and Mackenbach, 1994). Some of this research (e.g. that by Bartley et al) is continuing and will be monitored and supplemented during Phase 3.

4.3.6 Once we have created the revised SEC, simultaneously producing a new Census Matrix (project 4), we must validate it and test it against the current government SECs. ESRC is providing funding to allow us also to test it against the Goldthorpe schema, the Cambridge Scale and, possibly, (using the British Household Panel Study) IPA Social Grade.

4.4
Other Issues to be resolved in Phase 3

4.4.1 There remain a number of other important issues to be addressed and resolved in Phase 3. We indicate and comment on these in this sub-section.

4.4.2 Coverage: It has always been our aim to create a revised SEC which is as inclusive of the adult population as possible. This was one reason why we recommended that the Census asked all respondents to give details of their last main job (and see Para 2.1.6 for more details). However, it has been decided that the 2001 Census will operate a ten-year cut-off rule for occupational information. Obviously this has the effect of reducing the population coverage of the revised SEC. Arber (1995) analysed GHS data to see what might be discovered about those who have never been in paid employment, and thus cannot be straightforwardly captured within the revised SEC. This analysis shows that, among people aged 16-59, only 0.9 per cent of men and 2.6 per cent of women are recorded by the GHS as having ‘never worked’. The never-worked, therefore, are a very small category and of less concern perhaps than the much larger group of those who have been out of paid employment for more than ten years. It may be possible to apply special rules for classifying the retired and others who have been out of the labour market for more than ten years by reference to (a) pensions information in the case of the retired and (b) educational attainment in relation to non-pensioners. Census questions on sources of income and highest educational qualification, for example, might help us to classify the non-employed population for whom there is no occupational information. We shall examine these possibilities. However, the fact remains that the SEC we are proposing is essentially derived from occupational information. There could therefore be some groups which the revised SEC may not be able to capture in an entirely satisfactory manner.

4.4.3 Household Reference Person: this problem connects with that of population coverage. In the 1991 Census many household heads could not be classified to the SECs because they had not had an occupation in the previous twelve months. The ten-year rule will improve the picture but not entirely resolve the problem. We recognise the importance of this issue for local authorities in particular and we are aware that we shall need to give advice to ONS on the issues involved. Of course, the issue of the household reference person is one which is quite distinct from our main task of deriving a social classification for all individuals from whom occupational information can be gathered. However, we are equally aware that the HRP question becomes very important where data are collected from one household member only, as well as in the case of Census outputs which use ‘head of household’. We will therefore consider and advise ONS on the various options for whom the HRP should be. If our analyses justify it, we shall make a recommendation for a standard HRP.

4.5
The Revision of SOC and the Issue of SIC

4.5.1 Here, we reiterate our concerns regarding any revision of the Standard Occupational Classification. We understand that ONS Census Division is conducting a review of the SOC and the extent to which it needs to be revised. Given the operational role of the SOC in relation to the SEC, it will be readily apparent that this is a matter in which we have an acute interest.

4.5.2 In the Phase 2 Interim Report (Para 2.17 (c)), we noted the need for the revisions of SEC and SOC to be integrated. We repeat this request here because of our concern that any revision of the SOC after we have completed Phase 3 will inevitably lead to the necessity for further alterations to the revised SEC, almost certainly involving re-analysis of our LFS data and additional validation work, etc. Moreover, we are convinced on the basis of our work so far that the SOC does need to be revised (see Elias, 1996) and that our LFS data will be of great assistance to SOC revision.

4.5.3 SIC: we are also concerned about the issue of industry coding. We understand that Census Paper QWG23 proposes 100 per cent industry coding but also an automated coding process which reduces the details about industry to be requested or recorded. However, industry information is required at something like 3-digit level of SIC’92 in order to code occupation to OUGs - the building blocks of the SEC. We would thus be against any proposal to reduce the amount of detail available for industry coding.

4.6
Beyond Phase 3

4.6.1 Beyond Phase 3, there are prospects of improved international comparative research arising from the Review which may be of interest to ONS. The Siena Group (an organisation composed of senior members of the statistical offices of most of the EU countries, Eurostat, Canada, the USA and Australia) recently invited Rose to a meeting on social classifications and expressed enthusiasm for the proposed SEC because it could be relatively easily operationalised using the EU version of the International Standard Classification of Occupations (see Rose, 1997). Subsequently Rose has been invited by INSEE to a meeting on the revision of the French social classification system - PCS. INSEE is considering whether to adopt in France the approach we are taking to the revision of UK social classifications.

4.	Phase 3
4.1	Aims and timetable
4.1.1	ONS agreement to allow us to use the LFS to produce a more definitive allocation of OUG and employment status combinations to categories of the revised SEC effectively creates a third and final phase of the Review. Joint funding for this has been agreed between ESRC and ONS.
4.1.2	The aims of Phase 3 are (1) to analyse Labour Force Survey data on employment relations and conditions at the Occupation Unit Group level of the Standard Occupational Classification in order to operationalise the revised government social classification; (2) using a variety of datasets, to undertake validation studies with the revised classification; (3) to produce a new Census Matrix; (4) to bridge the current ONS social classifications to the new one; (5) to write a final report on all of this work for ESRC and the ONS; and (6) to provide users of the revised SEC with full details of its conceptual basis and construction. In undertaking this work we shall therefore be completing projects 4-6.
4.1.3	Phase 3 timetable: we understand that the LFS data will be available for analysis in April 1997. These data will then need to be subjected to the analyses set out below (Section 4.2) before the SEC can be finalised and validated (Section 4.3). Assuming that we do obtain the LFS data by April 1997, we estimate that it will take nine months to undertake the necessary analyses for the establishment of the revised SEC and the associated Census Matrix, as well as to complete further validation analyses using the revised SEC. Beyond this point, the proposed project 6 in Phase 2 - bridging between old and new SECs - and the application of the revised SEC to registration data will still need to be undertaken. This will take a further three months. Even beyond this point, further work might remain, for example in specifying the details of the SEC and its operational rules for inclusion in the next edition of SOC Volume 3. In addition, we believe it is most desirable that the results of our LFS analyses be included in a revised version of the 371 database. ESRC has agreed to fund this as part of its contribution to Phase 3. Three months is estimated for this task, to run concurrently with other work. A summary timetable for Phase 3 is given in Appendix 6.
4.2	The LFS Analyses
4.2.1	The analysis of LFS data is a relatively straightforward exercise, similar to what we have already done with the Omnibus data, but now with OUGs as the main focus. Essentially we wish to create SEC categories which are as homogeneous as possible with respect to employment relations and conditions as we have measured them on the LFS.
4.2.2	Continuity with existing SECs: in order to retain continuity with SEG, we need to attempt to use the LFS data to allocate a few large, problematic OUGs to the revised SEC rather than try to operationalise a whole new measure of social class. Given that the twenty largest OUGs account for 36 per cent of those in paid employment and 33 per cent of all employees; and that among full-time employees, the fifty largest OUGs account for 56 per cent of all cases, it will be readily apparent how important to the whole exercise is the proper allocation of these OUGs to the categories of the revised classification. Our initial efforts must therefore be to this end. We must also examine whether there is any evidence from the LFS that employment relations and conditions (as we define them) vary between full-time and part-time employees or between men and women in the same OUGs.
4.2.3	Refining the interim SEC: assuming (as the Omnibus and other validity exercises have demonstrated to date) that analysis of LFS data show that there are distinct groups or classes in terms of our employment relations indicators, and thus that the basic theory is sound, we can then use the LFS data to refine our interim SEC. That is, we can, where necessary, move individual OUGs from their current SEC categories to others, if the evidence indicates that this is appropriate. This cannot, however, be done wholly on the basis of statistical analysis because some OUGs are likely to be too heterogeneous with respect to employment relations. Equally, even with LFS data we will have too few cases in some OUGs to make statistically-driven decisions. We may also have to make some compromises necessary for continuity between the revised and existing SECs. No schema is perfect, and so we must be careful how this part of the exercise is handled. Consequently we will take expert advice before making final decisions.
4.2.4	Collapsing from long to short versions of the SEC: since presenting our Interim Report on Phase 2, we have made some small changes to the categories of the interim revised SEC as a result of our analyses of the Omnibus data (Note 6. However, the final decision on how best to collapse the long - SEG based - version of the revised SEC to the shorter version can only be made once we have completed our initial LFS analysis. In particular, this will involve a closer examination of the employment conditions of professionals by comparison with managers, and also those of lower managers by comparison with technicians and supervisors of office employees.
4.2.5	Category size: more pragmatically, we have to consider the relative size of the categories of the final collapsed version of the revised SEC. For example, some have already argued that category 7 - other workers - is too small at 6 per cent of employees and should therefore be combined with category 6 - currently intermediate workers. However, the beauty of the schema we are proposing is that it is flexible enough to allow researchers to make their own decisions about how to collapse from long to short SEC versions, provided they stay within the limits set by the schema’s conceptual basis.
4.3	Validation Exercises
4.3.1	The criterion validation aspect of Phase 3 (i.e. the continuation of project 5) involves ensuring that the revised classification is measuring what it is intended to measure. In our validation exercises we shall follow similar procedures to those developed by Evans (1992, 1996), and by Evans and Mills (1996), and especially the latent structure analysis techniques pioneered by Birkelund, Goodman and Rose (1996) and recently followed by Evans and Mills (1997). We shall examine both the criterion and construct validity of the revised SEC.
4.3.2	It has been agreed that the revised SEC should be based conceptually on employment relations and conditions. This sociological approach treats ‘class’ as a clearly-defined concept rather than as a theory tout court. What needs to be tested is that the instrument used, the revised SEC, fits the concept - that it measures what the concept purports to measure (see Evans and Mills, 1997). For Goldthorpe (1996) class analysis ‘explores the interconnections between positions defined by employment relations in labour markets and production units in different sectors of national economies’. It is these positions which are captured by the measurement of ‘class’. When the concept is operationalised it reflects these positions in the labour market. The positions themselves cannot easily be measured, yet the conditions which tend to cluster in distinctive mixes around the positions can. These conditions of employment form criteria related to the concept of class, and indicators of these will therefore be used to test the criterion validity of the revised SEC.
4.3.3	The results of the Omnibus data analyses already give us a high degree of optimism in this regard (see O’Reilly and Rose, 1997a and b). The definitive measure of the revised SEC which we devise with the aid of LFS data (and, following Goldthorpe’s comments (1996), data garnered from other sources on career patterns at OUG level and on pensions entitlements) will necessarily be subjected to further analyses.
4.3.4	Construct validity: it has been argued that criterion validity has limited usefulness in the social sciences (Carmines and Zeller, 1980). Where possible, one should also assess the construct validity of a measure. This involves assessing how the measure relates to other variables in ways predicted by theory, and in substantive areas of importance to government researchers and social scientists. The construct validation exercise will be theoretically driven, will use appropriate methodological procedures, and will ensure that other variables used have themselves been properly validated. Only under such conditions can we confidently conclude that correlation is evidence of construct validity. Here we shall be using the same datasets as are currently being examined - BHPS, GHS, the 1% LS etc.
4.3.5	As we have already noted, and most encouragingly from our perspective, the Goldthorpe schema has recently been positively evaluated by researchers interested in measuring socio-economic inequalities in health. Evidence from both the UK and Europe suggests that the Goldthorpe schema does have validity in these terms and may prove a better measure for these purposes than alternatives, including the current government social classifications (see, for example, Bartley et al, 1996; cf. Kunst and Mackenbach, 1994). Some of this research (e.g. that by Bartley et al) is continuing and will be monitored and supplemented during Phase 3.
4.3.6	Once we have created the revised SEC, simultaneously producing a new Census Matrix (project 4), we must validate it and test it against the current government SECs. ESRC is providing funding to allow us also to test it against the Goldthorpe schema, the Cambridge Scale and, possibly, (using the British Household Panel Study) IPA Social Grade.
4.4	Other Issues to be resolved in Phase 3
4.4.1	There remain a number of other important issues to be addressed and resolved in Phase 3. We indicate and comment on these in this sub-section.
4.4.2	Coverage: It has always been our aim to create a revised SEC which is as inclusive of the adult population as possible. This was one reason why we recommended that the Census asked all respondents to give details of their last main job (and see Para 2.1.6 for more details). However, it has been decided that the 2001 Census will operate a ten-year cut-off rule for occupational information. Obviously this has the effect of reducing the population coverage of the revised SEC. Arber (1995) analysed GHS data to see what might be discovered about those who have never been in paid employment, and thus cannot be straightforwardly captured within the revised SEC. This analysis shows that, among people aged 16-59, only 0.9 per cent of men and 2.6 per cent of women are recorded by the GHS as having ‘never worked’. The never-worked, therefore, are a very small category and of less concern perhaps than the much larger group of those who have been out of paid employment for more than ten years. It may be possible to apply special rules for classifying the retired and others who have been out of the labour market for more than ten years by reference to (a) pensions information in the case of the retired and (b) educational attainment in relation to non-pensioners. Census questions on sources of income and highest educational qualification, for example, might help us to classify the non-employed population for whom there is no occupational information. We shall examine these possibilities. However, the fact remains that the SEC we are proposing is essentially derived from occupational information. There could therefore be some groups which the revised SEC may not be able to capture in an entirely satisfactory manner.
4.4.3	Household Reference Person: this problem connects with that of population coverage. In the 1991 Census many household heads could not be classified to the SECs because they had not had an occupation in the previous twelve months. The ten-year rule will improve the picture but not entirely resolve the problem. We recognise the importance of this issue for local authorities in particular and we are aware that we shall need to give advice to ONS on the issues involved. Of course, the issue of the household reference person is one which is quite distinct from our main task of deriving a social classification for all individuals from whom occupational information can be gathered. However, we are equally aware that the HRP question becomes very important where data are collected from one household member only, as well as in the case of Census outputs which use ‘head of household’. We will therefore consider and advise ONS on the various options for whom the HRP should be. If our analyses justify it, we shall make a recommendation for a standard HRP.
4.5	The Revision of SOC and the Issue of SIC
4.5.1	Here, we reiterate our concerns regarding any revision of the Standard Occupational Classification. We understand that ONS Census Division is conducting a review of the SOC and the extent to which it needs to be revised. Given the operational role of the SOC in relation to the SEC, it will be readily apparent that this is a matter in which we have an acute interest.
4.5.2	In the Phase 2 Interim Report (Para 2.17 (c)), we noted the need for the revisions of SEC and SOC to be integrated. We repeat this request here because of our concern that any revision of the SOC after we have completed Phase 3 will inevitably lead to the necessity for further alterations to the revised SEC, almost certainly involving re-analysis of our LFS data and additional validation work, etc. Moreover, we are convinced on the basis of our work so far that the SOC does need to be revised (see Elias, 1996) and that our LFS data will be of great assistance to SOC revision.
4.5.3	SIC: we are also concerned about the issue of industry coding. We understand that Census Paper QWG23 proposes 100 per cent industry coding but also an automated coding process which reduces the details about industry to be requested or recorded. However, industry information is required at something like 3-digit level of SIC’92 in order to code occupation to OUGs - the building blocks of the SEC. We would thus be against any proposal to reduce the amount of detail available for industry coding.
4.6	Beyond Phase 3
4.6.1	Beyond Phase 3, there are prospects of improved international comparative research arising from the Review which may be of interest to ONS. The Siena Group (an organisation composed of senior members of the statistical offices of most of the EU countries, Eurostat, Canada, the USA and Australia) recently invited Rose to a meeting on social classifications and expressed enthusiasm for the proposed SEC because it could be relatively easily operationalised using the EU version of the International Standard Classification of Occupations (see Rose, 1997). Subsequently Rose has been invited by INSEE to a meeting on the revision of the French social classification system - PCS. INSEE is considering whether to adopt in France the approach we are taking to the revision of UK social classifications.