Social Mobility in Modern Britain: Class versus Merit
Saunders, P. (1995) 'Might Britain be a Meritocracy?', in Sociology, vol 29, no 1: 23 - 41
This paper challenges the Nuffield findings on social mobility, and the subsequent work of Goldthorpe and others who believe that 'recruitment into higher class positions is socially biased in favour of those who themselves originate in these classes, and that those born into the higher classes are to some extent insulated against the possibility of falling into the working class' (23). The aim instead is to see if the data might fit a rival hypothesis -- that Britain is a meritocracy.
The first task is to summarise Goldthorpe and others on social mobility:
(a) The Nuffield studies, and subsequent work by Goldthorpe, Marshall, and Payne all indicate considerable absolute social mobility. '34 per cent of the men and 30 per cent of the women in the service class had started life in the manual working class ... In all these studies, around one half of those born into the top class... failed to stay there... Britain seems a remarkably open society' (24).
(b) However, these researchers still argue that Britain is closed and unfair, by stressing relative mobility rates. The argument is that all classes have benefited from economic growth, and that service class children have tended not to fall back into the working class. Relative mobility rates are seen as more important in judging the 'fairness' of Britain, which has not disproportionately benefited children from working-class backgrounds.
Saunders argues that the stress on relative mobility has problems:
(a) The increase in absolute opportunity is sidelined
(b) More technically, the Nuffield and Essex studies used 'odds ratios' to measure open-ness, and these have problems. Odds ratios results when you 'calculate the chances of the service class child remaining in the service class relative to the chances of... falling into the working class... make a similar calculation regarding the chances of a working-class child rising into the service class as compared with the chance of remaining in the working-class... and then divide the first by the second' (25 - 26). This enables Goldthorpe to suggest that service class ratios are 13-19, while those for Class VII come out at 36. For him, a fair odds ratio would be 1. But odds ratios 'combine both success and failure rates... which... multiplies up any differences in mobility patterns' (26). Thus they would not be reduced by much simply by more upward social mobility, but would require more downward social mobility as well. This would be a radical form of egalitarianism.
Saunders suggests that we used 'disparity ratios' instead, which calculate the relative chances of success or failure of children from different classes ['class chances'?]. Using the Nuffield data, the disparity ratio was between 4 and 3-1:'In other words, a child from the service class has a three times better chance of ending up in the service class than does a child from the working class' (26).
(c) There is an assumption that there should ideally be no association between classes of origin and destination. This assumes 'no differences of aptitude between the members of different social classes in each generation', and that any unevenness in disparity ratios must result from purely social barriers. There is an 'assumption that talent and effort are equally distributed between the classes' (27). The notion of a meritocracy would not make these assumptions and would predict unequal outcomes, but Goldthorpe and his associates simply disowned this as a model for Britain, and gathered no actual data about either the intelligence or the effort displayed by the socially mobile. At worst, this seems to have been a refusal even to discuss differences of IQ, for example. Goldthorpe himself is content to rely on 'a presumption that [meritocracy] cannot be operating' (28). Saunders suggests that this results in a circular approach, where data gathered confirms this presumption and is never intended to be used to test meritocracy as a model .
In the absence of any hard data, it is necessary to explore the possibility that innate ability exists and that middle-class parents may be able to pass on intelligence or motivation. Apparently, the Halsey variant of the Nuffield study came close to admitting such a possibility, but did not explore it. Halsey's study admits that merit plays a role in educational attainment, but did not attempt to measure abilities. Instead, [in order to estimate the effects of social class membership against IQ] they estimated 'average IQ levels for each of the three social class groups (estimated at 109 for the service class, 102 for the intermediate classes and 98 for the working class)' (31). They used these figures to estimate the percentage of working-class boys going to selective schools who should have got there on merit alone (28 per cent), and compared them to the real proportion (24 per cent). Even these figures show that measured intelligence plays a considerable role, of course. Saunders calculates disparity ratios on these figures, and concludes that 'while class background are still operating as a factor in educational selection, most of the apparent "class bias" was a function of differences in average levels of measured intelligence between the classes' (32). Heath's subsequent study on the indirect routes [working your way up the hierarchy] also indicates the importance of personal achievement: in his own words 'social origins are not the only, or even the most important, influence on individual's subsequent career' (Saunders 32). The higher residual factors suggest that differences of ability might be 'playing a key role' for Saunders (33).
If only on policy grounds, however, we should investigate further. Saunders draws on some rather old work by Eysenck here. First, the review of twin studies (page 30) does seem to indicate that 'there are differences in innate ability reflecting the different genetic endowments of different individuals... people's mental abilities are partly a product of their genes, and... IQ scores do partly express genetically determined differences of mental ability' (31). Secondly, Saunders is going to use Eysenck's work to estimate the amount of this ability that gets passed on to children. He does this in trying to construct a model of a pure meritocracy, using the same assumptions about social mobility and the importance of males as does Goldthorpe:
(a) The first step is to assume that all the fathers have achieved their class position only on the basis of their measured intelligence -- 'We know from Goldthorpe's data that just 14 per cent of these fathers were in service class occupations while 55 per cent of them were in manual working-class occupations. Had a purely meritocratic system been operating, the service class fathers would have been in the top 14 per cent of the normal distribution of IQ, while the working-class fathers would all have been in the bottom 55 per cent of the distribution' (34)
(b) Assuming that IQ scores are 'normally distributed with a mean of 100 and a standard deviation of 15', we can calculate the actual IQ scores involved. Thus all the service class fathers should have an IQ of 116 or better, and all the working-class fathers an IQ of 102 or lower.
(c) We can then calculate expected IQ scores for sons, according to the Goldthorpe data: '26.5 per cent of the sons overall were in the service class, while 43.8 per cent were in the working class... [giving expected]… IQ scores of 109 or more for service class and 98 or less for the working class this time.
(d) Using principles established by Eysenck, we would not expect the perfect transmission of IQ scores across the generations, but should anticipate a certain regression to the mean. This in itself would stop self-recruitment to classes ( i.e. produce some social mobility). Views that classes will eventually become fully closed are exaggerated. We can now predict the pattern of IQ scores in children, bearing in mind this characteristic:
(1) 'we would predict that around 59 per cent of the children produced by parents with an IQ of 116 would have an IQ of better than 108... our model of perfect meritocracy therefore predicts that about 59 per cent of the children…. [of service class parents]… would themselves end up in the service class' (36).
(2) 'about 21 per cent all for children born to parents with an IQ of 116 or more would have an IQ of 98 or less... therefore... 21 per cent of service class children should end up in the working-class' (36)
(3) Similarly, for working-class parents: 18 per cent or of those born to parents with an IQ of 102 will meet the threshold entry of 109, where 58 per cent of them will have an IQ of 98 or less.
(e) These predictions can then be compared with Goldthorpe's actual findings, and 'the model fits the data almost exactly!' (37). Thus 'the apparently "gross" class bias all but disappears' (37).
There are still problems, including recruitment into elite positions, and whether the patterns will apply to women as well as men. Actual data would be required. However, the Nuffield studies have not provided data that conclusively rejects the meritocracy model. Critics have alleged right-wing bias, but Saunders merely pleads for more research, and announces an intention to do some of his own, using National Child Development Study data.
Marshall, G and Swift, A (1996) 'Merits and Mobility: a Reply to Peter Saunders', in Sociology, vol 30, no 2: 375 -- 386
Saunders' critique of their work is selective, and he has ignored published evidence relating to his interest.
(1) Saunders says that M and S have underestimated the effects of absolute changes in mobility. However the issue is equality of opportunity, the distribution of opportunities rather than the undisputed fact that there are now more opportunities. There is therefore always a comparative element involved.
(2) Saunders prefers disparity ratios rather than odds ratios, and accuses M&S of a radical definition of mobility that would imply levelling down. However, odds ratios are perfectly acceptable and they are calculated from the same basic data as disparity ratios. M&S provide both . If focusing on disparity ratios alone, it can be said that 'men who were the sons of service class fathers were almost four times more likely to be found in service class positions than were men who were the sons of working-class fathers; and, conversely, that men whose origins are in the working-class were nearly four times more likely to be found there than were the sons of service class fathers' (377). These facts remain, however expressed statistically.
(3) Saunders implies that mobility researchers have simply assumed that abilities and aptitudes are equal. However, M&S have tested the meritocratic model themselves, but by estimating the impact of educational attainment rather than choosing IQ. The assumption is that attainments indicate merit, although not perfectly. Thus their work shows that 'there is still a substantial direct effect of class background on class destinations which is not mediated by educational achievement... even when they have attained some low levels of education, children from service class backgrounds are more likely to arrive at service class destinations than are children whose social origins lie in the working class' (378). More technically, their attempt to fit a model of merit selection to their data proved unsuccessful, 'whereas by including a direct origins/destinations term we could reduce the deviance in the meritocratic model by more than 85 per cent, and achieve a good fit to our data' (379). Nor did examination of any residuals support the meritocratic idea. This analysis was not conclusive and may have missed distinctions in the types of educational achievement. At the same time, assuming that education reflects merit alone over estimates the effects of merit, although there are reasons to suspect that education is not class neutral.
Thus, overall, 'we find that equals are here being treated unequally... People from different class origins have unequal chances, not only of educational but also of occupational success, despite taking actual credentials into consideration' (379). Saunders simply fails to report these findings, possibly because he thinks that education is not a good indicator of merit. However, it is a conventional way to operationalize merit.
Saunders also fails to refer to the 'many published studies' that relate to the issue of meritocracy. For example, Heath et al conclude that 'even among people with similar educational levels, those from service class origins are more likely to be found in the service class themselves than are those from working-class origins. Britain is by no means a pure meritocracy' (381). There are comparative studies to cite as well [Marshall and Swift cite studies of Sweden and Ireland, page 381, and there are American studies by Bowles and Gintis, which are not cited, but which deny meritocracy there].
Saunders might wish to contest the idea that educational attainment indicates merit. There are other kinds of merits, such as the ability to work hard, and this may be distributed differently across social classes. These occupation-specific merits might also be transmitted differently among the classes, which would explain the apparent class effect. Motivation might be important. Equally, though, educational attainment itself might overlook other kinds of merit, especially those transmitted to working-class children [some idea of 'practical merit' here?].
Saunders seems to prefer IQ to educational attainment, although there are problems with this too -- Jencks, for example, says that intelligence tests may measure 'miscellaneous information picked up at home and on television, and the inclination (mainly instilled by school teaching) to try harder at this particular task than at any other' (383). Saunders would have been better to advocate the notion of 'innate intelligence', which could be assumed to be normally distributed and genetically transmitted. However, that would then raise the issue of justice -- should such a quality be used as a basis for reward? This is the old issue of whether one rewards achieved or ascribed characteristics -- Saunders actually advocates using achieved characteristics as the basis of meritocracy. It may also be true that the ability to work hard 'is itself a chance attribute and thus an improper basis for reward' (383). The main point of these complexities is that Saunders needs to demonstrate that differences in social mobility outcomes are based on some 'morally relevant differences between individuals, not just any differences whatsoever' (383).
Even if Saunders continues to prefer IQ rather than educational attainment, there are studies which he has ignored, including those of Douglas and his associates, which show substantial class effects on entry into selective schools, or staying on at school. A study by Wordsworth (page 384) has demonstrated class effects even when test scores and type of schooling are controlled [with a kind of sliding scale effect as in Douglas]. Saunders' simulation is less plausible than studies which have assessed such empirical evidence.
Saunders, P. (1997) 'Social Mobility in Britain: an empirical evaluation of two competing explanations', Sociology Vol 31, No 2: 261 -- 88
I am inverting the actual sequence of argument to begin with the bit which claims to be a reply to Marshal and Swift. One or two points make detailed reference to the text, which I can ignore, but generally:
(1) Saunders says his main interest was to critique Goldthorpe and others, not really to rebuke Marshall and Swift [although he did while he was there]
(2) Marshall and Swift have looked at the effects of educational qualifications, but Saunders is not convinced that this is an adequate measure of ability [he is quite contemptuous at the thought, which contrasts with his acceptance of other widespread assumptions about IQ scores and meritocracy]. His own work below suggests that ability is an independent factor instead, and one which operates alongside and indeed after the effects of educational qualifications as such.
(3) Marshall and Swift probably chose educational qualifications instead of more direct measures of ability such as IQ, simply because data on qualifications are easier to gather in a mass survey. They also seem to share the usual ideological opposition to such measures. Saunders admits that they are far from perfect, but reasserts that IQ scores are a reasonable measure to investigate: 'IQ scores correlate positively and strongly with different dimensions of ability... with other independent measures of mental agility such as mental reaction times... they are relatively stable over time... and even correlate with measures of infant intelligence... they correlate strongly and positively with academic performance and with various measures of job performance in adult life.' (278) [functionalism?]
(4) Marshall and Swift want to defend the use of odds ratios, although Saunders insists that these do underestimate the real advances in opportunity that have arisen. Saunders also relies on an argument from Noble, that pure mobility cannot be separated out from structural mobility. To decode this slightly, Saunders is arguing that particular patterns of mobility will always give high odds ratios -- especially where service class opportunities have increased, and working-class jobs have been diminished. In this case, there simply must be a rise in the chances of upward social mobility (which is acknowledged in Goldthorpe), but equally, a fall in the rate of downward mobility (and this is not acknowledged as having an effect): odds ratios will pick this up and give a misleading impression of the extent of closure in modern Britain. Even though real opportunities will have increased, the odds ratio will not register this improvement in these circumstances (pages 280 -- 81).
(5) Marshall and Swift appeared to be changing their ground by arguing that even if Britain were a meritocracy, it would not be a just social system. Saunders has argued somewhere else that meritocracy involves not just innate ability, but the use to which ability is put -- 'Occupational positions are earned. They are not allocated like sweets as rewards for doing well in IQ tests' (282) [functionalism again – pretty much like Davis and Moore here, certainly innocent of any marxism]. In any event, like it or not, 'the meritocratic principle is the necessary guiding principle of any competitive market society, and in Britain, it is the one which attracts widespread popular endorsement' (283). [However, it is worth noting that Young, who is cited for the basic definition of a meritocracy, was a thorough critic of the meritocratic principle and his book is in fact a satire].
Turning now to the new findings in this piece, Saunders wants to redefine the meritocratic model, probably to give more weight to a possible role for social classes in socialising their offspring [which does move him a bit closer to the usual sociological explanations of the effects of class -- the 'social advantage/disadvantage' or 'SAD' thesis]. However, he still insists on a separate effect of merit, a combination of ability and effort.
He has investigated some data from the National Child Development Study dataset (NCDS), a longitudinal study with quite a large sample, originally based on a panel of children born in 1958. There were several problems with this dataset when Saunders turned to read the 1991 results. The article explains these fully, but particular groups seem to have dropped out of the panel disproportionately, especially unskilled workers and women. Comparing the proportions left to the proportions in the 1991 census permits the effects of these withdrawals to be noted -- for example 'social classes I and II are slightly over-represented relative to the 1991 census figures and the proportion of those entering class V is substantially under-represented' (263). While this might affect any study of the stability of the working-class groups, we are on safer ground when analysing actual mobility both up and down. However, panel members were 33 years old when this study was performed, and so they are probably not finished with social mobility. At the same time, there are advantages in looking at younger people, such as that these people are more likely to reflect the most recent social reforms and economic changes.
Having got that out of the way, the data is used to complement the Goldthorpe study, even though the class categories are not exactly the same. Nevertheless, the NCDS data show that 52 per cent of the panel had been mobile compared with either their fathers or mothers. One third of middle-class (= service class ) children had been downwardly mobile, although they had not fallen very far. 25 per cent of working-class children had been upwardly mobile all the way into the middle /service class. In brief, absolute mobility rates were broadly consistent with the Goldthorpe findings: 'upward mobility into the middle class is common, even among those starting out at the bottom, while downward movement across the whole range is much less in evidence' (266).
The differences arise when looking at relative mobility, as measured by disparity ratios. There still are advantages for all children born to middle-class fathers, who are 'just over twice as likely to achieve middle-class positions by the age of 33 as compared with children born to fathers in semi-skilled or unskilled occupations, and they are about three-and-a-half times less likely to end up in semi-skilled or unskilled jobs' (266). This might be compared with the disparity ratio in Goldthorpe of between 3:1 and 4:1, not 2:1 as here. The unrepresentativeness of the panel might explain the differences, since lower working-class members have dropped out disproportionately: perhaps this will mean that those who remain are unusually upwardly mobile? Against this, the NCDS study defines working-class membership more narrowly, which might underestimate upward mobility. Saunders thinks that overall there has been a real reduction in disparity ratios, despite the need for this sort of caution.
Happily, the NCDS sample was also tested for ability and indeed a number of other family and background matters. Saunders believes that the general ability tests used do provide a useful measure of the sorts of skills and abilities required for social mobility [happily functionalist again?] (267). Scores on this ability test were correlated with the social class of parents [correlation coefficient of 0.24, rather a low one], and there was a class gradient in test scores [which supports SAD?]. This is probably the result of both genetic and environmental factors, Saunders thinks, but leaving that aside, we have some basic data to test out the effects of merit and social class.
Measured ability also correlates with the social class arrived at by the 33 year-olds, but this time much more strongly [correlation coefficient of 0.37]. Although there is another gradient, ability and class of destination is particularly strongly associated 'at the lower end of the class system' (267). [There are some handy tables, eg on page 268]. This result is better news for the meritocracy thesis than for the SAD thesis. In a meritocracy 'any association between class of origin and ability should be weaker than that between ability and class of destination' (268). The strong association between low ability and low class destination also supports the thesis.
Saunders is not content with this general finding, however, and wants to see if the entire disparity ratio, 2:1, can be explained by these differences in measured ability. He begins this by repeating the kind of experiment we saw earlier:
(1) As 43 per cent of the sample were in service class occupations, they should also be the ones in the top 43 per cent of the range of measured ability (that is with a score of 49 or better on the particular ability tests used)
(2) The actual data shows that only 62 per cent of those entering the service class scored as high as that: 'Put crudely, 38 per cent of those arriving in the service class were not bright enough to be there!' (268).
(3) Of those less able entrants, far more came from service class families than working-class families (32 per cent as opposed to 17 per cent). Looked at differently, 41 per cent of relatively low scoring children from service class families still managed to end up in the service class. Such children are 'twice as likely to succeed as low ability children from semi-skilled and unskilled manual worker homes' (269). [So – more support for SAD]. Thus 'low ability is not necessarily a barrier to later occupational success’
(4) Conversely, high ability does appear to offer reasonable guarantee against failure, for only five per cent of children in the top ability court trial ended up in class IV/V, while 65 per cent of them made it to class I/II' (269).
So much for ability, but Saunders also argues that effort is a key part of merit. There are some data in the NCDS survey that can help here too, such as those gathered by a motivation scale, an 'absenteeism' factor and a 'job commitment' factor [these seem to be composite measures based on some factor analysis to identify connections between responses to more specific questions] (269). These factors can be compared with similar ones suggested by the SAD thesis, principally measures of parental class, but also level and type of pre-school education, extensive parental support and interest, and a level of parental ambitions, including different ones for children of different genders. Saunders also draws upon some equally 'traditional' sociology of education, going back at least as far as Douglas, Fraser and Plowden, to measure the educational effectiveness of the home. Thus level of parental education, the social class of grandparents [emerging as a factor in a very early 1962 study by Jackson and Marsden, Saunders tells us in a note], and the basic amenities of the home are also measured [for a critical discussion of these measures, check out the notes on Fraser].
We can then enter these different factors into various statistical analyses, and see which ones best predict upward social mobility. [I have not given full details of the ingenious measures and experiments that Saunders engages in here -- read them for yourselves page 271 on]:
(1) In one experiment 4 meritocracy variables and 12 SAD variables are 'used to predict whether individuals born to class IV/V parents will remain where they are or move all the way up to class I/II. It demonstrates clearly that it is meritocratic variables -- ability, motivation and attitudes to employment -- which are the key factors distinguishing successful lower working-class children from those they leave behind them' (271).
[I don't understand 'logistic regression', but the technique seems to involve testing each variable at a time to estimate its 'predictive accuracy', which seems to be a measure of fit with the actual data. The factor 'ABILITY' increases predictive accuracy the most, followed by measures of motivation, worker attitudes, gender and the social class of grandparents in that order. For a non-numerate person like me, it is at least clear that these factors also display decreasing coefficients when correlated with working-class success: ability has a correlation coefficient of 0.26, motivation 0.16, attitudes 0.15, gender 0.10, and grandparents' class 0.09. Only the last two provide any support for SAD. Many variables assumed to be effective seem not to have any measured effect at all, including attendance at fee-paying schools, material deprivation, parental levels of education and so on.]
(2) The same technique can be used to predict failure, that is downward mobility from service class families. Here, 'none of the variables helps very much in predicting downward mobility. Nevertheless, to the extent that we can predict it, ability is again the key factor [correlation coefficient 0.19] twice so strong as that of any other variable in the model, and motivation is entered second. Private schooling appears as the most important of the SAD [variables]… suggesting that... [it]... may offer middle-class parents some means of insuring their less able offspring against downward mobility' (273).
(3) There is a need to develop an even stronger test, to establish whether individual qualities positively overcome the effects of different class environments -- 'For example, are dull middle-class children still likely to do better than bright children from class IV/V backgrounds?' (273). Saunders thinks he has shown that they are not. Ability is a major factor in determining who will succeed given the same social class background. It is also the case that 'class IV/V children entering the middle class have higher average ability scores than class I/II children leaving it' (273). This can be shown by using a multiple regression model using all the factors to try and explain the social class position achieved at age 33. Again, the table on page 274 provides the data in detail, but:
(a) When looking at entry into differently ranked occupations, '[A]bility... has by far the strongest effect... or motivation at school enters second... absenteeism and work attitudes enter 4th and 6th... The strongest SAD variable is parental class [which is third in the list]' (274). The first four variables explain nearly all of the effect.
(b) Meritocratic variables, especially exams ['based on the numbers and grades at GCSE, O level and A-level passes]; qualifications ['based on NVQs']; first job ['occupational status of the first job taken after completing full-time education']; and ability, are also much better at explaining the status of the first job achieved [this specific definition of social class is not explained very well in the actual study, but things are clearer in a note summarised below]. Although meritocratic factors explain more than class factors, 'over three-quarters of the variance in occupational prestige scores remains unexplained', however (275). Saunders thinks that 'bright people tend to end up in higher status jobs, partly because they accumulate more qualifications, but also because their ability comes to be recognised and rewarded independently of their paper qualifications.' (275).
In conclusion, Saunders thinks that meritocratic factors clearly outweigh the effects of parental class with its relative advantages and disadvantages:
(1) 'ability correlates more strongly with class of destination than with class of origin, and this must mean that the occupational class system is to some extent selecting by ability irrespective of social class origins' (281).
(2) 'ability and motivation are the key predictors of low working-class success and of middle-class failure' (281).
(3) 'class destinations... reflect individual merit... much more than class background' (281). Many variables which have seemed so important in past studies -- including cultural factors in the home -- seem to have very minor effects on class destinies. Factors measuring intellectual capacity have been ignored on ideological grounds, but seem far more important.
(4) Both meritocratic and class theses have something to offer to an overall explanation, but 'the occupational class system in Britain is more meritocratic than has commonly been assumed' (282)
(5) The high residuals left unexplained indicate that other important factors are also at work [Saunders seems to give no indication as to what they might be].
The notes make many fascinating additional comments and enter a number of important reservations. For example:
Breen, R and Goldthorpe, J (1999) 'Class inequality and meritocracy: a critique of Saunders and an alternative analysis', in British Journal of Sociology, Vol 5, No 1: 1 -- 27
With this piece, the technical arguments are dominant, and the level of discussion of these is way beyond my competence. If I have understood the logic of the argument correctly, Breen and Goldthorpe are suggesting that there is a more profitable and less controversial way to test the effects of class inequality and meritocracy than that pursued by Saunders. They intend to use the same NCDS dataset as did Saunders, but with a different view about how to select the relevant variables. They want to criticise Saunders' choice of variables in a number of ways, including a reference to the criticism we made ourselves in the last summary -- that some of the variables, such as parental interest, are likely to reflect the class judgements of teachers. There are much more technical criticisms too.
Breen and Goldthorpe have a different methodology as well: instead of comparing class variables and meritocracy variables directly, and letting them compete or 'race' against each other to explain the variations in question, Breen and Goldthorpe suggest that we first establish from the original data the association between class of origin and class of destination, and then secondly to calculate the effects of adding meritocracy variables. If class remains the major determinant, adding these extra variables should produce very little change in the measures of association. There are, apparently good technical reasons for testing the effects of the variables in this way, but these reasons are largely borne by various asides and footnotes, and these stray way beyond my expertise. Substantively, what they claim here, in the Abstract, is that 'while merit, defined in terms of ability and effort, does play a part in determining individuals' class destinations, the effect of class origins remain strong' (1). However, much of their dispute with Sanders does turn on the methodological arguments -- Breen and Goldthorpe think tangling with the methodological issues is the only way to help us choose between the rical versions
Breen and Goldthorpe also want to deny Saunders' dismissal of educational qualifications as relevant, and, indeed, calculate a third table to allow for the effect of such qualifications. Finally, they want to pull off a neat trick and report some unnoticed findings from Saunders' own calculations. The Abstract again summarises the substance -- 'Children of less advantaged class origins need to show substantially more merit than children from more advantaged origins in order to gain similar class positions' [original emphasis] (1).
The paper begins with a very useful summary of the debate with Saunders so far, noticing in particular the accusation that the focus on odds ratios and relative social mobility seriously underemphasises the actual growth of mobility in Britain. The other main accusation is that the processes through which people become mobile have been relatively neglected: that work which has looked at educational qualifications has been misguided, and researchers should have examined IQ scores and measures of effort directly. The last paper in the series (see above) attempted to rectify this omission by looking at NCDS data, with the results that we know already -- that merit apparently plays a large part in determining class destinations. Breen and Goldthorpe want to agree that researchers should turn to the actual processes of mobility, but think that Saunders has underestimated the difficulties with his approach.
Breen and Goldthorpe agree straight away that the chances of absolute mobility have increased, that there is indeed more 'room at the top'. However they still want to insist that it is important to look at relative mobility. They suggest that 'inequality of opportunity could be increasing, decreasing or unaltered, despite the "upgrading" of the opportunity structure overall' (4).
The first technical issue then arises, about the relative merits of odds ratios and disparity ratios. Breen and Goldthorpe suggest that odds ratios give a much fuller position of mobility chances, since they establish the chances in terms of entering one particular class of destination, but then go on to compare chances of achieving entrance into that one class with the chances of appearing in other destinations. Odds ratios simply combine estimates of these chances together, while Saunders prefers to keep them separate. Saunders thinks that combining separate disparity ratios leads to pessimistic and extreme findings, but the reverse is also true -- that single disparity ratios are used by Saunders mostly to refer to upward rather than to downward mobility chances.
Saunders also repeats the particular criticisms made by Noble, that odds ratios rarely seem to fall if there is a constant increase in chances for upwards social mobility . Breen and Goldthorpe suggest that it is 'entirely realistic' for there to be both such increases in chances and changes in the relative opportunities given to children of different social classes. Thus chances of upward mobility could be combined with genuine egalitarian reforms to alter the chances of children from working-class origins achieving an increased share of these newly provided opportunities. Something like this apparently happened in Sweden, they argue.
There's an even more technical argument about the superiority of odds ratios, pages 5 - 6. I'm not sure I understand it, but I think Breen and Goldthorpe are suggesting they are a better measure of social fluidity, and a more general one -- 'indicating the association that exists within the mobility table between class origins and class destinations, net of the origin and destination main effects: i.e. the larger the ratio, the stronger the association' (6).
We now turn to discussing different statistical models and strategies to test the effects of merit and class. The 'most straightforward way' for Breen and Goldthorpe would be:
(i) establish the extent and pattern of the association between class origins and destinations; and then (ii) observe what happens to this association when variables indicative of individual merits are brought into the analysis. If a substantial association remains, the meritocracy thesis is undermined; if the association largely disappears, the thesis is supported. In a perfect meritocracy, class of origin and class of destination would be statistically independent once merit was taken into account (6)Saunders does not stick to this approach and instead develops 'logistic regression to try to do something clearly different and something that is both less apposite and technically far more difficult: that is, to evaluate the relative importance of variables indicative of merit and of social advantage in predicting in which of two destination classes children of a particular origin class will be found' (6). In the final stages, using multiple regression, Saunders has to change the terms of analysis [as we have seen above], no longer measuring class position but rather the score on a scale of occupational prestige. Saunders tends to think of 'all the important indirect effects', those left unexplained by class of origin directly, as variables expressing 'measures of ability and motivation' (7). Since these are more important, statistically speaking, meritocracy appears to explain more than social class.
However, there are considerable theoretical and statistical problems involved in this less straightforward procedure. I don't understand the arguments carried by the references on page 7, but the gist of it appears to be that measurement problems can seriously affect the results. In particular 'when measuring variables to be set in competition with each other, great care must be taken that this is not done in such a way that the result of the race is more or less "fixed" from the start' (7). This is apparently what Saunders has done.
For example, Saunders uses the Registrar General's classification in measuring class of origin, although this is apparently known to be deficient [with a reference to Marshall et al 1997]. In Saunders' work, the original six categories are collapsed to three and are then formed into a three-point scale -- 'as in effect a continuous variable but one with a very restricted range of values' (8). This is likely to show a stronger relationship with merit [because it is also a continuous variable?], and thus Saunders' results run the risk of being 'simply a statistical artefact' (8). [The only light I can shed on this argument is to refer back to a similar criticism made of Fraser's work. As far as I can see, the point is that if the measurement characteristics of two variables are similar, especially if they are continuous variables with a number of values on a proper interval scale, then this similarity will be shown up and artificially emphasised in any attempt to show a statistical connection between them.]
Breen and Goldthorpe point out other problems in Saunders' use of NCDS data:
(1) There is the problem of variable drop-out which makes the data increasingly unrepresentative -- Saunders notices this. The data itself varies in terms of its quality, and the 'documentation of the data-set... is... still in some respects inadequate' (page 8). Thus any conclusions must be subject to revision.
(2) Saunders includes men and women in the earlier analyses, but not in the multivariate regression analysis. He also simply excludes those with no occupation at age 33, while Breen and Goldthorpe prefer the 'standard practice', which is to allocate them on the basis of their last occupation.
(3) Breen and Goldthorpe think it best to use a classification of social class which is similar to the Goldthorpe class schema, and not the Registrar-General's scheme as Saunders does. Apparently it is straightforward to do this using NCDS variables themselves.
(4) There are a number of variables relating to academic ability. Saunders uses a combination, but Breen and Goldthorpe use only one -- scores on the general ability test taken at age 11. They argue that the other measures refer more to attainment than ability, and claim that attainment is more affected by a class background. These test scores are the closest equivalent to IQ scores.
(5) Effort is not so easily measured from NCDS data, and Saunders again composes his own measures. In particular, measuring attitudes to work at age 33 is likely to be affected by class position at that age and cannot be a causal variable. There are problems with measures such as absence from school, which may be due to constraint not choice or 'motivation'. Breen and Goldthorpe rely on just the one scale of academic motivation.
(6) Saunders measures school exam results at 16 and then reported qualifications at age 33. Breen and Goldthorpe prefer to work with data on the highest level of qualification achieved. There are apparently six categories here, and the intention is to turn these into a continuous variable, just like ability and effort. [Thus all the variables have similar measurement characteristics, eliminating one source of error as described above].
Breen and Goldthorpe then proceed to perform their own analysis, first establishing a relationship between class origins and destination, and then trying to estimate any changes after introducing merit variables. Specifically, they first construct 'inter-generational class mobility tables for men and women', and calculate association between origins and destinations [using a log-linear model for the technically minded, the same as in the original work and in one of Goldthorpe's follow-ups]. Then they introduce 'ability effort and educational qualifications' and examine the effects if any on the associations demonstrated in the first model.
There is a table on page 11 to demonstrate the results. Goldthorpe and Breen also want to discuss more abstract measures of association to set alongside the demonstrated more concretely by the table. A technical discussion ensues about how best to do this (page 12). I don't understand again, but apparently the data in the table can be transformed into a 'topological model', with different levels reflecting the 'strength of association between the rows and columns... [that is]... between class origins and destinations' (12). Each particular combination of parental class and class of destination, each cell in the table, can be relocated on one of the levels of the model [achieving a bit more of an economical model for the data, possibly] . This topological model was then fitted to NCDS data, and a good fit was achieved after some slight modifications, reducing the first eight level topological model to a six level one for men, and a five level one for women. In a way which leads to further confusion and inadequacy for me, the model can then be used to provide the relative mobility chances for any particular combination of parental class and class of destination -- for example 'the chances, under our model, of men born into Class I... being found in Class I rather than in Class VII... relative to the chances of men born in Class VII being found in Class I rather than in Class VII' (13). The actual chances in this case, incidentally, are 'estimated at over twenty to one' (13).
The model can then be used at the second stage, when merit variables are introduced. The technicalities elude me yet again, but are described at the top of page 14. Thus additional variables are entered one at a time, general ability test scores first, then academic motivation, in order to give the best indication of the effects of merit 'as Saunders might conceive it' (14). Finally, or rather thirdly, the effect of educational qualifications can be included, first separately, and then combined with the other merit variables.
Again tables indicate the results, pages 14--16. Apparently, the addition of ability and effort variables does improve the fit, and both are statistically significant. This shows that the 'log-odds [chances] of being found in any other class relative to Class I do, in part, depend on ability and effort' (17). Estimating the strength of these factors involves comparing the changes in social fluidity as they are added to the original data. The results for men show a variable and rather modest effect, so that 'substantial inequalities in class mobility chances are still clearly in evidence' (17). For example, the 'odds ratio for relative mobility chances as between Class I and Class VII was 20.7. When individual ability and effort are introduced into this model... this odds ratio is still 11.1' (17), and there are smaller changes in the odds ratios between other classes.
For women, there appears to be 'less inequality in mobility chances than among men', originally, and more, and more consistent, reductions as merit variables are added. However, merit variables do not completely replace class variables, and 'considerable' differences remain. As an example,'the odds ratio for relative mobility chances as between Class I and Class VII is, under our original model, 16.3, and is still 6.3 after ability and effort have been taken into account' (17) [Is this a substantial and significant reduction, or a sign of the stubborn influence of class? Is the glass half full or half-empty?].
On including educational qualifications, it seems that the effects are stronger, for both men and women, although again 'statistically significant and substantively important' class effects remain. Little is added if educational qualifications are combined with measures of ability and effort overall. The data show that educational qualifications seem to be more significant in reducing class inequality than do ability and effort, the opposite of what Saunders thinks. There is also the key finding that considerable effects of social class remain, although they don't seem to be transmitted through ability, effort or educational attainment, in other words through 'merit'. [Might they be the same ones as identified by Bowles and Gintis -- 'behaviour', 'personality', a willingness to take on the values of the docile worker? NB These are the special province of education as such, for B&G] .
Thus meritocratic variables are a long way from explaining social mobility, although they do seem to have an effect. The degree of the effect can be estimated by constructing an ideal picture or model [much as Saunders does in his original piece with IQ scores]. The techniques are baffling again, but they seem to involve predicting 'ideal' values for the topological model described briefly above. The idea seems to be to establish some 'ideal' picture of 'pure' social mobility again, where the effects of social class have been reduced to negligible values. This prediction can then be compared with actual observations, particularly those measuring 'outflow' from each social class. Apparently, there are substantial differences between the pure predictions and the actual observed cases, 'indicative of persisting class inequalities in mobility chances' (19). However, [and curiously], 'outflows from Classes III and IV (those of routine non-manual employees and of lower grade technicians and supervisors of manual workers) show no notable departures from perfect mobility expectations' [so there does seem to be considerable social mobility or social fluidity for those classes? I wonder why].
Overall 'meritocratic selection by ability and effort, as it operates in modern British society, mitigates the influence of class to only a very limited degree' (20), which simply contradicts Saunders' findings. The maligned SAD tradition seems to have been right after all.
Why are there these differences, though? Is it just a matter of using different models and 'complex techniques, by means of which... [one] can demonstrate more or less anything' (21)? There are methodological differences with Saunders, and Breen and Goldthorpe think their choice of variables is more appropriate and less biased. It is also also a matter of emphasis. Breen and Goldthorpe clearly confirm 'that mobility chances are indeed influenced by ability and effort, and in turn it follows that merit, understood in terms of ability and effort, may in some cases enable individuals to overcome the disadvantages of their class origins. But what is also shown is that merit can operate in the way it does without the effects of class being thereby annulled or even much reduced' (21). This seems possible only because 'children of disadvantaged class origins have to display far more merits than do children have more advantaged origins in order to obtain similar class positions' (21).
A close examination of Saunders' own work would suggest this too. It would be quite possible to calculate the amount of merit needed to compensate for class disadvantage, using Saunders'own data. When this is done, 'it turns out that children of working-class... origins would need to outperform children of middle-class... origins by roughly... 15 points on the 80 point scale [of the general ability test]' (22). Similar results apply to effort. Overall, 'the rewards that accrue to merit themselves vary depending on the class origins of those who possess it. If this makes Britain a meritocracy, it is then one of a very peculiar kind' (22). [However, Saunders also says that service class children who stayed on in the service scored more highly than those who entered it from below].
The flaws in Saunders'analyses are fatal. But there are genuine difficulties, especially in measurement, for any study. Better measures might well indicate more of a role for merit, and better definitions of social class origins might work the other way around. The best form of analysis would be to consider different birth cohorts, where different generations could be compared, even using the same flawed measures. We would at least have some estimates about whether merit is becoming more important over time. There is apparently a British Cohort Study which might be compared with NCDS in the future.
The notes include some interesting asides as well:
Saunders, P. (2002) 'Reflections on the meritocracy debate in Britain: a response to Richard Breen and John Goldthorpe', in British Journal of Sociology, Vol 53, No 4: 559 - 574.
As the Abstract indicates, this paper consists first of all of a critical discussion with Breen and Goldthorpe (a reply to their criticisms, criticisms of their work); a new study of the NCDS data, using measures that Breen and Goldthorpe suggest; an attempt to establish propositions that indicate what middle ground remains after the lengthy debate over the original 1995 paper. The paper starts with an excellent summary of the discussion so far.
Goldthorpe has replied to some of Saunders' original criticisms but not to all. In particular, his original work suggested that a genuine meritocracy would feature complete equality of opportunity (odds ratio of 1:1, to be technical). Saunders thinks he has argued very successfully that a meritocracy would still provide certain inequalities, and that he now assumes the Goldthorpe agrees:
Proposition I: Evidence of inter-generational transmission of class identities does not of itself demonstrate that meritocratic selection is failing to occur, nor that class advantages and disadvantages are necessarily influencing individual outcomes (561).
Saunders has suggested all along that there is one area where class of origin does seem to have an effect, and that is in preventing downward social mobility from the service class to the working class, and this suggestion was borne out by the subsequent research (including a paper with Rod Bond that I have not summarized -- Bond, R. and Saunders, P (1999 'Routes of success: Influences on the occupational attainment of young British males', in British Journal of Sociology, Vol 50, No1:1-27). None of this denies that ability is the main factor explaining upward social mobility, however:
Proposition II: The main factor preventing a true meritocracy from operating in Britain is not the failure of bright working-class children to rise in the occupational class system, but is the ability of middle-class parents to reduce the chances that their less talented children will fail (562).
Saunders says that this finding leads to a rather novel policy implication -- that we 'should force more middle-class children to fail'. Apparently, he and his colleagues agree that this would be impossible, however, especially as the middle-class still seems to be expanding. [However, we seem to have arrived at an issue first discussed in the functionalist work on social mobility, especially that of Hopper (Hopper, E. (1972) 'Notes on stratification, education and mobility in industrial societies', in Hopper, E. (ed) Readings in the Theory of Educational Systems, London: Hutchinson) . Hopper suggested that a main function of the education system should be to 'cool out' untalented middle-class children, persuading them to lower their ambitions. Bright working-class children, on the other hand, should be 'warmed up'. As Hopper notes, a classic study by Clark (Clark, B. (1960) The Open Door College : a case study, New York: McGraw-Hill Book Company Inc.) indicates that the US Community College would be an ideal institution to perform such cooling out. It is worth noting that the UK government has recently flirted with the idea of community colleges too].
Turning now to the methodological controversies:
(1) Saunders still thinks he was right to use combined measures of ability and motivation, and, indeed, in the paper with Bond, even more complex scales were defined. Of all the criticisms made by Breen and Goldthorpe of this approach, Saunders is prepared to accept only that he should not have used measures of work attitudes at age 33. He says he cannot see how using combined measures can bias the results: 'use of multiple indicators is the standard procedure for minimising measurement error' (563). [Even I know that this is not always the case, however. It only applies if the multiple indicators have different sources of measurement errors and that such errors can be assumed to cancel each other out]. However, Saunders is prepared now to use the same simpler measures as chosen by Breen and Goldthorpe, and claims 'We shall see that it makes little difference to the final result' (563).
(2) There is no problem in choosing between the Registrar General's scheme of class classification and the Goldthorpe scheme. The NCDS data reveal a close overlap between those two (a correlation of 0.782) (564). Saunders wants to acknowledge the criticism about using the class schema as an interval variable, and agrees to re-run his analysis using the scheme preferred by Breen and Goldthorpe [treating class positions as dummy variables -- I am not skilled enough to follow the implications here, although one of the notes helps clarify some of the logic].
(3) Saunders still thinks he was right simply to exclude those who did not have a job at age 33. Estimating their social class position from their last jobs, as Breen and Goldthorpe suggest, is unacceptable because those jobs might well refer to a position in late teens or early twenties, right at the start of a career. This in turn would underestimate the amount of mobility that most people would expect.
(4) Saunders reduces the size of the sample by excluding those without jobs, but Breen and Goldthorpe exclude still more, by deleting 'cases where information on any single variable is missing' (565). Since there are many variables involved, a large number of cases end up by being deleted. Saunders'option is to substitute 'sample mean values... in cases where data were missing' (565). This strategy does have an effect, however, in 'compressing the total variance in each variable, thereby reducing all measures of association between them' [I think this relates to the point raised earlier that the actual distribution across variables has an independent effect on measures of correlation between them]. However, this is likely to affect both class and merit variables equally.
(5) Saunders' technique involved comparing the relative 'predictive strength' of both class and meritocracy variables [and the different ways in which this was actually done, using different statistical techniques, are explained on page 566]. Breen and Goldthorpe point out problems with this 'variable race' approach, 'because some variables are more accurately measured than others' [I thought the issue was that some variables are measured on an interval scale and produce wide distributions, which should give them a greater say in prediction].
(6) Breen's and Goldthorpe's own approach also has problems, however. The logic of their approach is to establish the effects of class origins on destinations, and then introduce various meritocratic variables and see what differences remain. The main flaw is to think that any remaining difference must be due to the effects of class. In general, Saunders argues, this involves assuming that all the variables have been measured correctly, and that there are no third variables which could account for the link between class origins and class destinations. [I think that neither of the parties have been very imaginative here in thinking additional variables to explain the residual variance, which both have admitted is high. Both parties need to look at the work of Bowles and Gintis, in my view, which offers a number of important additional ways in which class affects schooling which then affects social mobility]. Still at the general level, Saunders suggests that Breen and Goldthorpe are arguing that 'residual variation in occupational outcomes that is not explained by variables included in [their models] should automatically be attributed to class origins, but much of it should actually be explained as due to measurement error' (566). At least in Saunders's approach, the effect of each variable was calculated separately, and the variance left at the end was seen as unexplained. Breen's and Goldthorpe's approach sees the SAD thesis as the default explanation, and allows it to claim all the unexplained variance.
Saunders then takes most of the objections on board and re-runs his own regression models with modified variables. He also decides to proceed in a different way, by entering the SAD variables first -- 'it is well known that variables entered first in a multiple regression can appear stronger than they really are, for they can claim all the variance that they share with other variables which are not entered until later', and Saunders wants to 'ensure that the SAD hypothesis gets all the help we can possibly give it ' (567). He suggests a second step, adding 'all the other variables which might point to the importance of middle-class advantages or working-class disadvantages in determining occupational outcomes', and leaves the merit variables until the third step, and qualification variables as the final (4th ) one. The results appear as a table on page 568.
There is a further discussion about the precise use of the coefficients involved [which I am not competent to comment on]. However, overall, Saunders's results show that 'at step 3, ability (0.26) and motivation (0.14) far outweigh any of the other variables entered at earlier stages, and that at step 4, ability (0.16) in particular is still having a substantial effect, even after school and post-school qualifications are taken into account' (569). These results hold at a high level of statistical significance [and of the variables used, ability, motivation and qualifications achieve this high significance, as do 'three of the class dummies (Classes I, II and IV) -- (569)].
There are variations within the overall picture, however, for example 'if the father is a member of the service class (particularly the "upper" service class), or if he is a self-employed small businessman (Class IV), then class origins also play a significant part in influencing class destinations. This is not, however, true of any other class background' (569). [The relative lack of fluidity in Class IV is noted in the original Goldthorpe studies, and probably arises because small businessman and farmers have a very direct way of passing advantages on to their offspring, simply by leaving them the business. The relative lack of fluidity in the upper service class is probably more politically significant, however, and lends force to the suspicion that elite occupations are far less fluid].
There are two general points of particular interest:
(1) 'class origins of the main effect through the ability of middle-class parents to prevent their children from sliding downwards' (569 )
(2) 'Knowing whether or not somebody has working-class origins simply does not help us predict where they end up at age 33... class origins, defined and measured as Goldthorpe wants them to be, explain at most just 7 per cent of the variance in occupational destinies at age 33. Adding other measures of socially advantaged or disadvantaged backgrounds raises this to 10 per cent... [adding] the ability tests scores at age 11 and a crude scale measuring motivation at school [doubles the proportion of variance explained] to19 per cent, and when qualifications are added, we can raise this again, to 27 per cent' (570).
In fact, these results are not very far from those found by Breen and Goldthorpe, who noted a substantial drop in odds ratios after adding ability and motivation, and then educational qualifications. They suggested, however, that what remained was 'far from negligible'. Saunders points out that, on the contrary, 'class origins explain only a small proportion of what is left over', especially when we allow that what remains in the Breen and Goldthorpe analysis is amplified by much that 'is undoubtedly due to measurement error and the effect of other variables not included in the model rather than to the influence of class' (571). Thus:
Proposition III: Class origins plays some part in influencing class destinations (mainly as a result of middle-class parents protecting their children against downward mobility). Individual ability and effort, however, are much stronger predictors of class of destination than our class origins, as are the qualifications which individuals achieve at school and thereafter (571).
This does not mean that Britain is a perfect meritocracy, but that Britain is a society that rewards talent and hard work, and places far less emphasis on class origins.
The notes are as usual very interesting: