London Guildhall University
Footnote 1: The points made in this paragraph were confirmed and amplified by the Workshop proceedings (cf. Section 10).
17 February 1995
1. Introduction
1.1 The following report summarises presentations and discussions from a one-day Workshop on Social Classifications (SECs) as part of the OPCS-funded ESRC Review of Government SECs. The speakers and other participants included academics (both users of official SECs and experts in classification), market researchers and representatives from local authorities and quangos. The Workshop Programme and participant list are attached at Annex 1.
1.2 The Workshop was intended to provide evidence for the Steering Committee of the ESRC Review of Government Social Classifications concerning: (1) the need for Government SECs; (2) the uses of SECs, both by different kinds of users and across the most relevant disciplinary and substantive areas; (3) problems arising in their use and how these might be resolved; (4) alternative classifications; and (5) suggestions for revised or replacement SECs and their validation. This Report does not attempt a synthesis of the proceedings. Rather, it summarises each presentation and the discussion which followed it. However, a synthesis will be produced at a later stage.
2. The ESRC Review
2.1 Professor David Rose, ESRC Research Centre on Micro-social Change, University of Essex (Convenor of the Steering Committee for the ESRC Review) explained the purposes of both the Review and the Workshop's place within it. He noted that the Report on the Review which had to be produced for OPCS must first demonstrate the continuing need for government social classifications why they were still required and what benefits they brought to users and others. However, none of the evidence received by the Committee had called for the abolition of government SECs. On the contrary, all users (including those in central government) wished to retain them. Similarly, oral and written evidence was clear about the users and uses of SECs and existing alternative SECs. However, there were widely ranging views on the need for revision or replacement of the existing government SECs - Registrar General's Social Class based on Occupation (RGSC) and Socio-economic Groups (SEG).
2.2 First, the `off-the-peg users' of SECs: a few were saying `leave things as they are'. Rather more said `leave things as they are unless you can prove to us that a revision will provide real benefits over the existing SECs. And even then, if you do make changes, make sure we can relate the old scheme to the new one because of the important time series which the RGSC gives us'. Rose emphasised that these were opinions related almost entirely to RGSC: fewer users seemed to employ or understand SEG; it was usually RGSC to which they referred and which they wished to preserve in recognisable form. However, SEG had its champions, too, being used by geo-demographic researchers and, usually in collapsed form, by government researchers (on the GHS and LFS, for example) and academics on surveys such as the British Social Attitudes Survey.
2.3 Most off-the-peg users recognised the deficiencies of the RGSC, especially in terms of its population coverage. Put simply, there were doubts about the extent to which any occupation-based SEC could be wholly adequate, both because it failed to classify individuals who were not economically active and because occupational SECs were regarded as deficient in relation to the classification of women’s employment. Some also recognised its conceptual poverty. While few went so far as to say that occupational SECs should be abandoned, there was an across-the-board view that it would be good to have a more inclusive non-occupational classification (`NOC') based on consumption or lifestyle - and there was also a good deal of demand for an income question in the Census. These off-the-peg user views were quite similar for both government and for academic researchers, and especially those in the health and medical fields. Many of these users were less concerned with SECs as incorporating theories of society than with the ability of SECs to give good population coverage, to display social variations and to provide an ordinal measure yielding an appropriate gradient (or what might be called the Registrar General's Convenient Summariser of Social Variations).
2.4 These views differed from those expressed by the experts in social classification, mostly sociologists. They tended to see little merit in the RGSC (primarily because of its ad hoc nature), although the time-series/continuity argument was appreciated (but not entirely accepted - see the discussions in Sections 3 and 10). However, in so far as revision of the current SECs was concerned, sociologists viewed SEG as a better candidate for revision than RGSC. This was because SEG combines information about employment status, authority and establishment size with occupational (Standard Occupational Classification - SOC) data. Thus SEG, because it is primarily about employment status, comes closer to a sociological conception of class than does RGSC. The sociologists were also more aware of existing alternatives, of course, such as the Goldthorpe class scheme and the Cambridge scale. Some also shared the concerns of the off-the-peg users about the coverage of any occupationally-based classification, but were less fazed by this - rules could be (and have been) developed to deal with the problem. Some experts gave more importance than others to maintaining continuity so that comparisons could be made diachronically; and a number were critical of the higher levels of aggregation within SOC because they grouped occupations with very dissimilar work and market situations and did not distinguish the self-employed from employees. This point was very important, since the sociologists were mainly concerned to have the right building blocks for social classifications which they could then derive for their own research purposes. It was at this point - the building blocks issue - where any apparent impasse between `off-the-peg users' and SEC experts might find some resolution. How could the building blocks of SECs be improved to the benefit of all? Again, a strong argument was put forward by some that this required the development of an occupationally-based scheme which more satisfactorily covered female employment patterns; and SECs which gave rules for the assignment of difficult cases based on the results of empirical research.
2.5 What was also clear from the oral and written evidence was the multiplicity of uses of SECs and of the ways in which they are used. This raised the question of the role which a publicly sanctioned and provided SEC should play, especially at a time when both the building blocks for any classification and automated coding were increasingly available to researchers. The experts wanted more information at the most disaggregated level possible; they could then do their own classifying. The `off-the-peg' users wanted a standard, reliable and valid census-based measure which covered the maximum number of people, conveniently summarised social variations, and related intelligibly to other phenomena of interest (Footnote 1).
2.6 Finally, Rose indicated the purposes of the Workshop. Given the problems with the current government SECs from the perspectives of users in different disciplinary, policy and substantive areas and with different levels and types of `SEC expertise', the following key questions needed to be addressed:
i) How might some of the perceived problems be resolved? For example, to what extent and in what particular ways was it possible to revise the current SECs and provide a better conceptual rationale? Could this be done without loss of continuity? How might the building blocks need to be improved?
ii) What were the alternatives to the current SECs which should be evaluated in the Review? What were the possibilities for creating `NOCs’? Did participants accept the case for NOCs?
iii) What were the implications for the Census Form? i.e. what new or revised questions or codeframes might be needed?
iv) What might all this imply for future research?
In other words, the Workshop should concentrate on revisions, alternatives and their evaluation. These were the fundamental questions for Phase 1 of the Review.
3. Social Classifications and Health Research
3.1 The presentation on SECs from the health perspective was given by Dr Ray Fitzpatrick, Nuffield College, Oxford. He noted that, almost without exception, the SEC which had been used to explain mortality and morbidity was the RGSC. In its basics, RGSC had been in use since the 1911 Census. It thus provided researchers with a long-standing, continuous time-series which could be used for comparative purposes. RGSC also produced a stepwise gradient with respect to mortality. Its importance to health researchers was made even more vital because of its application to vital registration records. Nevertheless, no one had really examined what the health patterns revealed by RGSC meant since it was not clear what exactly the RGSC was measuring, and therefore what it was telling researchers about health outcomes and their explanation. In addition, the population coverage of the RGSC was confined to those in paid employment; and RGSC was also less effective in explaining women's health outcomes, mainly because of the way in which women's occupations were classified and categorised, many ending up in Class IIIN.
3.2 Despite the problems of RGSC, there had been little research on the use of alternative classifications. For example, the other government SEC - SEG - had not been used in health research and so nothing was known of its power to explain health outcomes. However, the OPCS 1% Longitudinal Survey of the Census did have the potential for an exploration of other alternative SECs and some researchers had analyzed health in relation to Census consumption variables such as housing tenure and access to motor vehicles. These sorts of Census variables might form the basis for a non-occupational classification with much greater population coverage, including coverage of some of the most disadvantaged groups in society. In addition, researchers had used area-based classifications in the form of deprivation indices constructed from Census variables. These indices described the social characteristics of quite small geographical areas and yielded powerful relations with health outcomes (e.g. the Townsend Index, the Jarman Index and the Carstairs Index). However, area classifications could not replace SECs based on individual occupations, not least because these indices incorporated census information on SEG and/or RGSC. Also, in epidemiology much research was related to the specific job hazards of individuals and. as such, provided a driving force for the collection of occupational data. Finally, analyses based on area classifications were always prone to ecological fallacies and could thus be misleading; and, in dealing with social deprivation, area schemes addressed only a part of what health researchers were concerned to explain.
3.3 Fitzpatrick acknowledged that consumption-based measures were of importance and could play a role over and above that of occupation-based measures. However, consumption measures also tended to be much more transient and so could only add to, rather than supersede measures such as the current SECs. While health research could benefit from trying alternative occupational SECs to those produced by government, nevertheless so many factors affected health outcomes that any occupational schema would be limited in terms of what could be explained by such measures alone.
3.4 In discussion, Dr Peter Goldblatt, Department of Health, was concerned that, in terms of health research, many important groups of people were moving out of the scope of the RGSC. This was a worrying problem and needed to be addressed. Nor did he see that area classifications had problems any different from the current SECs: just as individuals can be atypical of the areas in which they live, so can individuals be different from others in the same occupational group, with women's occupations being a prime example. Professor Brian Jarman, St Mary's Hospital Medical School, noted that his index should not be interpreted as an area index in the same sense as Townsend's. It was not designed as a deprivation index but as an index of factors which affected GPs' workloads.
3.5 Professor Raymond Illsley, University of Bath, noted that RGSC used to be the only explanatory measure available to health researchers. However, as understanding of health issues improved so had new measures emerged. RGSC was now outgrown and more sophisticated measures were required. These could be obtained from the Census. If RGSC were to be continued, it was necessary to be wary about what its alleged continuity meant. For example, RGSC Class V was not now the same as it was 20 years ago. The RGSC was a constantly changing measuring rod: moreover it changed in ways which were not understood. While its categories might be called the same, they did not mean the same from Census to Census. Illsley had no objection to a changed SEC provided it could be interpreted sensibly. This was a much more important issue than continuity.
3.6 Dr John Goldthorpe, Nuffield College, rejected the implications of Rose's distinction between off-the-peg users who wanted SECs to be convenient data summarisers and experts who were concerned with theories of society. Any SEC, whatever its purpose, must be clear conceptually so that its construct and criterion validity could be investigated. This was an area where sociologists were much more cavalier than psychologists. Researchers must know whether a measure is measuring what it is supposed to be measuring. On this criterion the RGSC failed. Goldthorpe agreed with Illsley that validity checks needed to be maintained over time. Surely it would be beneficial to off-the-peg users to know what processes were generating empirical regularities, so validity was vital. Dr Ken Prandy, University of Cambridge, supported Goldthorpe's views. Predictive validity must be examined. Off-the-peg users should want the best; false oppositions between types of users were not helpful. In concurring with Goldthorpe and Prandy, Professor Rose noted that he had been reporting a distinction between types of users which existed, not one of which he approved.
4. Social Classifications and Political Behaviour
4.1 Dr Anthony Heath, Nuffield College, in discussing SECs in relation to research on political behaviour, pursued the issue of predictive validity raised by Goldthorpe and Prandy. He noted that neither RGSC nor SEG was widely used in political research; preference was for the measure of social grade used in market research (see Section 9) or for the Erikson-Goldthorpe-Portocarrero class scheme (usually known in the UK as the Goldthorpe scheme).
4.2 Social class remained a powerful predictor in political research at the aggregate level (e.g. in predicting which party would win a particular parliamentary constituency). It was somewhat less powerful as a predictor of individual political behaviour (e.g. how a particular individual would vote). However, the predictive power of the ordinal RGSC was not impressive when compared to either Goldthorpe's scheme or SEG, both of which were categorical. Indeed this was one of the reasons why categorical class schemes were to be preferred. When collapsed into fewer categories, SEG showed much better systematic and theoretically intelligible variation than RGSC (e.g. the self-employed were much more likely to vote Conservative, a fact which is comprehensible). In fact, collapsed SEG approximated the Goldthorpe scheme. In commenting on Heath's SEG analyses, Professor David Lockwood, University of Essex and Chair of the ESRC Review Committee, observed that SEG seemed to speak theory without knowing it - something which could not be said for RGSC.
4.3 As in health research, Heath reported that some political scientists and sociologists had attempted to use consumption-based measures, and especially housing tenure, as independent variables. This gave as good a prediction of individual behaviour as Goldthorpe class. However, housing tenure was now a skewed variable (74% owner occupied against 18% council tenants in the British Election Survey) and this made it very undesirable as a discriminator. In analysis, housing tenure should only be used as a supplementary explanatory variable to social class. Moreover, consumption variables must be related to issues of theoretical concern. For what were they supposed to be proxy measures?
4.4 The advantage of social grade was that it was all-inclusive in population terms. However, this resulted in Grades D and E being extremely heterogeneous: for example Grade E included both young people who had never worked and retired people who relied solely on the state pension for an income. Social grade was good for explanations of consumption behaviour, but useless for explaining political behaviour.
4.5 How should researchers tackle the population coverage problems of class measures? Heath pointed to research which he had undertaken in which last main occupation was given to the economically inactive and predictive power had been satisfactory. He noted that very few people had never had an occupation, but those in this position were a very heterogeneous group mainly of young men and older women. They could not, therefore, be regarded as forming a single category. However, the unemployed could be used as a category.
4.6 Prandy, while recognising the importance of self-employment, did not see why this should mean a rejection of an ordered classification scheme. Heath had not shown any analyses using the Cambridge Scale. In fact the CS demonstrated a steady gradation in voting patterns which implied a continuous scale rather than categorical properties. Should measures of inequality be based on scales or categories?
4.7 Diane Mathewson, MRC Biostatistics Unit, Cambridge, agreed that consumption variables such as housing and car ownership were transient because they had been subject to rapid change, but argued that the same could be said of job descriptions used for coding to SOC occupations were becoming more transient, too. Were there any more stable and enduring measures such as education?
4.8 Dr Gordon Marshall, Nuffield College, pointed to the wide-ranging consensus that information on occupation and employment status should continue to be collected in the Census and on major surveys. Surely, since these formed the building blocks for all class schemes, the issue should be how to improve these rather than arguing about the schemes themselves. In response, Jean Martin, OPCS Social Survey Division, disagreed. Building blocks could not be taken for granted. For example, the employment status information derived from the Census, from surveys and from registration data was not all of the same order. While accepting this, Goldthorpe argued for nested classifications using SOC information at various levels, depending on the level of detail in the data available. This type of procedure was common in other countries using ISCO - the International Standard Classification of Occupations.
5. Social Classifications, Education and Training
5.1 Professor John Bynner, Social Statistics Research Unit, City University, discussed SECs in relation to training and education. He immediately related his comments to those of Prandy and Goldthorpe on validity, but argued that validity became a simpler issue if the fitness of a measure for its purpose was first addressed. In education research, social class was used for many purposes.
5.2 First, class was used to explain the reproduction of inequality. Second, as in the work of Paul Willis for example, it had been used to explore subcultures and norms. Third, class had been employed to investigate education as the generator of human capital. Fourth, class had been seen as an indicator of multiple disadvantage. Finally, class could be used as a means for identifying marginal groups, such as an incipient underclass.
5.3 However, in its fundamentals, education must be viewed as longitudinal and inter-generational. Hence, all approaches to educational research using class had to confront the problem of the life-course or life-cycle. In explaining educational performance, what was important about class, and whose class was important, changed. At early ages, father's class was the most salient explanation of a child's performance. By the teenage years, a child's own past performance became more important. From the beginning of adulthood, class had to be defined in terms of the individual's own occupation and education and occupation became closely interlinked.
5.4 Professor Robert Erikson, Swedish Institute for Social Research, emphasised the importance of distinguishing between different aspects of social stratification. Class should not be regarded as an explanatory panacea. It could not, and was not intended to measure everything. Erikson also commented on Swedish experience with official SECs as a warning to British colleagues. In Sweden, there had been a pincer movement by experts on the one hand and politicians on the other against official social class measures. For a time Statistics Sweden was forced to abandon social class and attempted to produce composite measures instead. This failed and a new class scheme based on occupation and employment status was developed in the 1970s which was similar to the Goldthorpe scheme. For people not in the labour market, rules had been devised to allocate them to classes - such as last main job, partner's job, etc. A validation exercise in the 1980s led to the removal of the distinction between semi and unskilled workers. The basic building block was a five-digit occupation code which could be used to produce Goldthorpe's scheme, ISCO and the Swedish government scheme. Moreover, 70% of the coding could be done using automated techniques. Not only did the Statistics Sweden scheme relate well to international measures, but SS makes money from marketing it. Finally, Erikson noted the vital importance of the UK retaining a scheme which could be used for international comparisons.
5.5 Dr Simon Szreter, St John's College, Cambridge, agreed with those who argued for the creation of the best possible social class measure. He also considered that classification schemes were very powerful in their political and communication uses, providing historical and international comparability. He concurred with Marshall that the main task was the creation of better building blocks and commended Elias' work in these terms.
6. Social Classifications and Employment
6.1 Dr Duncan Gallie, Nuffield College, discussed the use of class schemes in relation to research on employment. For economic sociologists the issue was not whether class schemes were important, but which class scheme was best. The Goldthorpe and RGSC schemes were differently conceived, yet little was known about the properties of either. Using analyses of his recent national employment survey, Gallie showed that the two schemes correlate at about 0.8. The RGSC is essentially skill-based, but skill is multifaceted. Nevertheless, whether using a skill measure or Goldthorpe's scheme or the RGSC similar results are obtained from discriminant analyses using skill attributes, although the Goldthorpe scheme is somewhat better at correctly classifying cases in terms of skill. Surprisingly, correct classification is not greatly improved in the case of the Goldthorpe scheme when variables relating to the service relationship/labour contract distinction (fundamental to the scheme) are added to the analysis.
6.2 Gallie's presentation prompted a discussion about data on occupations. Dr Peter Elias, Institute for Employment Research, University of Warwick, pointed to the difficulties in defining what an occupation is. How do people describe their occupations? Where does skill fit in? SOC was the common language developed to tackle these issues and was now widely used across a range of important datasets. It was also agreed that data should be collected at the most detailed level, including the actual text. However, there were often problems about access to these data, especially with the Census but also with the Census Sample of Anonymised Records. Goldthorpe saw no point in improving the data atoms unless these, rather than aggregated data, could be used by researchers.
7. Social Classifications and Housing Research
7.1 Professor Chris Hamnett, Kings College, London, discussed housing research in which he saw occupational class data as crucial. While having information on income and consumption would be desirable, he agreed with other speakers that these could not be regarded as replacements for a class variable. In the case of housing, Hamnett echoed Heath in observing that, as owner-occupation rose to such high levels, housing tenure became a less discriminating variable. Researchers were now attempting to construct measures which discriminated among types of owner-occupier. Similar considerations applied to car ownership where the type of car was an issue. People might be home-owners and car-owners, but could they afford their upkeep, especially when things go wrong with them and repairs are required?
7.2 In reporting results of analyses of British Household Panel Study data conducted for the Rowntree Inquiry into Income and Wealth, Hamnett observed that collapsed-SEG was a good predictor of housing value. The scale of housing gain was also related to occupational measures. Hamnett's research had led him to the conclusion that, while housing tenure was an important aspect of stratification, its role should be viewed in relation to area, class and income. He was therefore very wary of using housing tenure as a substitute for class.
7.3 In response, Goldblatt argued that a measure of housing tenure sorted out the `muddle-in-the-middle' of the RGSC scheme. Others thought that it would therefore be better to sort out the muddle-in-the-middle by revising RGSC. For Illsley, Hamnett's presentation took him back to his earlier point: we must know what the basis of any SEC is. Had housing tenure been chosen as the basis for a SEC in the 70s, it would now have to be abandoned for lack of stability. The advantage of occupation-based SECs was the persistence of occupation. Goldthorpe concurred. He saw no need for composite NOCs when advanced multivariate techniques, software packages and computer power were readily available. Why did people want an omnium gatherum NOC? What was the rationale? Marshall regarded the unique strength of occupational class measures in terms of their significant relationships to wide and varied social outcomes. Thomas warned that no measure could ever be totally persistent since occupations change. Equally, all occupation-based measures were composites because they included reference to other variables such as employment status.
8. Social Classifications and Geo-demographic Classifications
8.1 Keith Dugmore, CACI, explained the origins, derivations and uses of area classification schemes (ACs) such as ACORN. Essentially ACs classify and analyze people in terms of where they live. They incorporate data from a myriad of sources, including the use of Census SECs and SEC data from government surveys. Indeed SECs were vital to classifications such as ACORN, with SEG the preferred measure. In CACI's case, RGSC and SEG were both used for other purposes, too. For Dugmore, it was important that SEG be retained and be coded from the Census at 100% level rather than the current 10% level. From CACI's viewpoint, to lose SEG and RGSC would be the worst option; it would be bad to revise the SECs and leave them incompatible with Social Grade; it would be good to make compatibility between SEG and Social Grade; it would be best to have an income question on the Census.
8.2 Dugmore's pleas on behalf of market analysts prompted a similar set of requests from local authority representatives. Eileen Howes, London Research Centre, also wanted income and 100% occupation coding on the Census. She also agreed that SEG was a much more important measure to retain than RGSC. However, she also argued for LAs to have a reputable, well-researched, Census-based NOC at both area and individual levels, since these were vital to LAs' task of making equitable distributions of resources within their areas.
9. Social Classifications and Market Research
9.1 Corrine Moy, NOP and MRS, gave a presentation on the IPA Social Grade (SG), widely used in market research. SG was used for many purposes - defining target markets, planning and selling advertising, discriminating across markets and products, etc. Six grades (A,B,C1,C2,D and E) covered the range of occupational groups and those on state benefits and pensions. The measurement of SG was based on employment status, current or previous occupation, organisation type, grade within the organisation, number of employees, number of people supervised or managed and qualifications. SG could be approximated to a collapsed version of SEG. One of its principal advantages to the market research world was its stable profile of the population over time. Primarily evaluated on a household basis, SG also had rules for the inclusion and classification of `nonworkers'. However, its application in the field by interviewers and in the office by coders was admitted to involve some subjectivity and to lead to problems of replicability. In terms of a wish list, the MRS needed social grade output from the Census, and a project to derive SG from the Census matrix was in train. The government SECs were seen as poor measures of social grade.
9.2 There was some discussion of the coding of SG. Several people were puzzled by the claim that SG was easy to code given that it required more raw information than government SECs. Nick Moon, NOP, said that in practice full coding was only done on the National Readership Survey. More usually coding was done in the field by interviewers who were trained to ask all the questions necessary for classification to SG. In many cases occupational title was all that was required. However, Corrine Moy admitted that this was where the subjectivity problems of SG entered the picture. Gallie asked about the validity claims in respect of SG to which Moy had referred: what was the evidence? Moon explained that this was purely a reference to SG's validity against other possibilities which would be available in market research and not against academic or government SECs. It was difficult to make comparisons between SG and other SECs because there was no basis of comparison across market research, academic and government datasets. Elias noted that comparisons would be possible if MR raw data were maintained rather than only the aggregated SG data. Moon and Moy thought clients would not be prepared to pay for this; Elias countered that SG could not be produced from Census data because all the information needed was not collected in the Census: comparability depended upon MR companies retaining raw occupational data. Finally, Marshall noted that while SG was obviously good at what it was designed for - strategic market research and the explanation of consumption patterns SG did not work for most of what social scientists were interested in explaining.
10. Concluding Session
10.1 David Rose led the concluding discussion. Despite the earlier strictures of Goldthorpe and Prandy, Rose was left with the impression that there was a divide of both understanding and use of SECs between most `off-the-peg' users and most `experts' in classification. Fitzpatrick had said that it was uncertain what RGSC was measuring; Illsley had stated that whatever it measured was not the same in meaning in 1995 as it was in 1911, 1931, 1951 or 1971, even though RGSC categories and their labels might have remained the same. If RGSC was not defined so that we did not know what was doing the explaining, and if RGSC also lacked continuity because of societal and occupational change, why not say that the best solution was to have a definition of whatever a new or revised SEC was to be called so that we knew what was supposed to be being measured?
10.2 Rose also observed that demand for keeping SEG had emerged more at the Workshop than before and that this was an across-the-board demand academics, market researchers, market analysts and LAs. This could create a dilemma for the Committee. If NOCs were also in demand, either SEG or RGSC might have to go, assuming that OPCS would not want to produce three social classifications. However, some people still did not understand what a NOC meant, as Goldthorpe had observed. The committee needed more information on what a NOC might look like and what its conceptual rationale would be.
10.3 Finally Rose asked how hard it would be to persuade `off-the-peg users' to lose RGSC. The Review had revealed a surprisingly low level of understanding of social classifications issues by `off-the-peg users'. Equally Rose admitted to being a little ashamed that the experts had not done more to try and educate `off-the-peg' users. Discussions across the divide were needed. Lockwood, in agreeing with Rose's summary, further noted that government users seemed to prefer RGSC, but many outside government apparently favoured SEG. Secondly, if revision of SECs was to be undertaken, a sound conceptual rationale was required. This point had come across clearly at the Workshop.
10.4 Dr Rosemary Crompton, University of Kent, queried whether it was true that government would not be prepared to produce more SECs - CASOC and its successors should make classification easy. It was collecting the raw data that cost the money. Martin said this was not entirely true: what cost money was developing and maintaining the Census matrix for different SECs. Also we should recognise that the reasons why people used either SEG or RGSC were not to do with theoretical arguments but convention. Choice of RGSC was also a matter of convenience - RGSC had fewer categories than SEG. Rose said that he thought the real concern with existing SECs was coverage. `Off-the-peg users' were largely ignorant of ways of giving class position to the economically inactive yet, as the Workshop had demonstrated, there were a range of ways in which this could be achieved, whether using SG, RGSC, SEG or Goldthorpe. Illsley remained concerned about the unemployed and disabled for whom it was more difficult to obtain a true estimate of their occupational position by referring to past occupations. While these groups might be a small proportion of the population, their mortality rates were often higher than Class V. Redistributing them would have considerable impact.
10.5 Non-occupational Classifications: Eileen Howes was not impressed by Goldthorpe's argument concerning multivariate analysis as an alternative to a NOC. Not only were non-experts often unable to manipulate data in this way, but vital to LAs was an accepted standard classification which could be used for policy purposes and for national and European comparisons, both of which were vital in the process of obtaining and distributing resources. Martin agreed with Howes on the importance of standard social classifications for comparative purposes. There was increasing demand for greater comparability and standardisation from central government departments so that data from a variety of sources could be compared. She also sympathised with the off-the-peg view regarding analysis: many people used only published sources; only expert researchers wanted and could use raw data. Elias supported this perspective and called attention to the fact that the recent Rowntree Inquiry made no reference to occupation or social classification because the relevant information was not sufficiently available. Multivariate solutions were not the answer because of high inter-collinearity. Psychologists would see a latent variable in that situation and therefore would seek to condense the data. The other way of dealing with the problem would be to classify. On the basis of the evidence from Hamnett's, Fitzpatrick's and Heath's presentations, Lockwood remained unconvinced about the case for a Census-based NOC.
Problems with existing SECs: there was further discussion of the lack of coverage of the current SECs and their inadequate classification of women. Goldblatt noted that the Census Small Area Statistics necessarily excluded large numbers of people who did not have occupations. Dr Karen Hurrell, Equal Opportunities Commission, was concerned that SOC and the SECs did not adequately capture the nature of women's working lives. In response Rose reminded participants of the recent independent work of both Heath and Marshall which showed how economically inactive people could reasonably be classified by reference to their last main occupation and also to Heath's contention that few people had never had contact with the labour market. However, Rose agreed that there were problems with SOC which needed attention in relation to women's occupations. He also observed that Elias' paper began to address this issue and that Martin and Roberts in their 1984 publication on women and employment had produced, faute de mieux, a revised RGSC for women.
10.6 Elias concluded the meeting by reiterating his plea for eclecticism in data collection, data analysis and data presentation so that people could then `do their own thing'. Researchers must be allowed to look at raw information in different ways and so it was vital to retain the information collected and preserve the detail. At the same time the SECs would have to be revised. Elias favoured the promotion of `skill'-based classifications, but recognised that there was no one best SEC. Even so, efficient methods for maintaining and developing SECs were required. Above all, SECs had to take account of and reflect social and economic change: continuity would have to be sacrificed.