Popular Posts

Caveat Emptor

The opinions expressed on this page are mine alone. Any similarities to the views of my employer are completely coincidental.

Wednesday 20 October 2010

Smear tactics and guilt by association

A good rhetorical  tactic if you want to cast doubt on the credibility of someone without directly calling them  a liar is to associate their name with the name of someone else who has a reputation for dishonesty. All the better if both parties are dead and can't defend themselves. I came across  something like this in Peter Saunders' old 1996 pamphlet Unequal but Fair: A Study of Class Barriers in Britain. To make my point I have to invoke the doctrine of fair use and quote Saunders at length:

It is a sad reflection on the rigour, vitality and integrity of much mainstream post-war British sociology that the Glass findings were so readily endorsed by so many leading sociologists for so long. The evidence was hopelessly dated (more than two-thirds of the fathers in the study had first entered the labour  market when Queen Victoria was still on the British throne), yet these findings were accepted as a valid guide to British social structure well into the 1970s, thirty years into the long post-war economic boom and long after the first wave of post-war social and educational reforms should have had some sort of impact. Even worse, the research was seriously flawed. Geoff Payne has meticulously demonstrated that the data are unreliable, for taking account of the twentieth century expansion in white-collar and contraction in blue-collar jobs, and of the higher fertility rates in working-class than in middle-class families, the Glass data could only have been valid if the number of white collar jobs had declined by 18 per cent in the course of a generation. In reality, however, the number of such jobs had increased over this period by 17 percent. The findings, in other words, were quite simply impossible given the occupational changes documented by censuses through the first fifty years of this century.

Payne likens David Glass’s standing within social mobility research to that of Charles Darwin in evolutionary theory. A more appropriate parallel might be with the infamous Cyril Burt and his influence upon psychological thinking about intelligence. Like Burt, Glass’s work went uncriticised for many years despite clear evidence that the data were fallacious. There is no suggestion that Glass manufactured his data, but Payne does note with some frustration the apparent unwillingness within the discipline of sociology to cast doubt upon Glass’s study. This may have had something to do with Glass’s standing within British sociology, for in the early 1950s he was a major figure with considerable influence, but it also probably reflects the reluctance of left-wing sociologists to question findings which were consistent with their own prejudices.

The substance of these paragraphs are repeated, though without the reference to Burt, in Saunders' 2010 book Social Mobility Myths also published by Civitas. The association, in this context, of  David Glass' name with Cyril Burt's seems to me, to say the least, distasteful.  Why  would you invoke the two names in the same  sentence unless you wanted to connect them with the one thing that most people  remember Burt for - the allegation that he fabricated some of his data. The caveat to the contrary looks to me quite jesuitical.

So fashionable half-truths and non-truths quickly become the conventional wisdom. The dead don't defend themselves and history is rewritten. Look, all those seventies lefties made it all up. It's a fact.


Well, sometimes the facts bite back. The political ideologists have cloth ears, but maybe there are are still a few people, even some sociologists, who care enough about the truth to listen .

1) The evidence that Burt deliberately falsified his data is far from conclusive. I don't know whether he did or or he didn't but the evidence can be read in a number of ways. It appears to be the case that a number of things he was accused of - for example making up the names of  non-existent research assistants - were false. He may have done a number of things, especially as a journal editor, that would not now be regarded as entirely ethical, but as far as deliberate falsification of data goes, what a dispassionate review of the evidence reveals  - for example Gillian Sutherland's Ability, merit and measurement: Mental testing and English education, 1880-1940 - is  best expressed in the Scottish legal formula: not proven.

2) Geoff Payne et al.'s 1977 article does not show that the distribution of father's jobs in Glass' 1949 survey was "simply impossible". (1) They certainly claim something like this is the case (though to be fair, don't put it quite as strongly) but that is not what their evidence shows.  The Census data they use to calibrate the Glass data are  in fact largely irrelevant to the question of interest, which is how to estimate the distribution of fathers' occupations (social origins) conditioning on the occupations of the sons (social destinations). In a standard mobility table the observations on the fathers do not give a snapshot of the father's occupations at any particular point in time.  Career mobility (of the fathers), differential fecundity and  variation in the timing of marriages and births are all confounded in ways which make it impossible to draw clear cut conclusions by comparing the marginal distribution of the fathers' occupations in the 1949 survey with Census distributions. Moreover, even if we could learn something from the Census distributions there is a serious gap in the evidence: there was no Census between 1931 and 1951. Serious students of social mobility know all of this; they learned it from Glass himself and if not from Glass from O. D. Duncan who  in 1966 published a classic exposition of the problems involved in trying to make inferences of the sort Payne et al. attempt. (2) Duncan's paper is not cited by them, which is, to the say the least, peculiar. Maybe they were unaware of it.

3) Rather than engaging in the hopeless task of  trying to infer something about the expected social origin distribution in a mobility table from  Census data, a more direct way of evaluating the plausibility of the Glass data is  to compare it to other survey data collected around the same time and, crucially,  coded in a similar way. In 1959, Social Surveys (Gallup Poll) Ltd conducted the Marriage Survey on behalf of the Population Investigation Committee. All copies of the original data were thought to be lost. However thanks to  the generosity of Professor David Coleman who still possesses printouts  and coding sheets for parts of the  original data matrix,  I, along with a colleague from the LSE , with the assistance of one of my doctoral students, have managed to piece together an important part of it.  The figure below presents some evidence of relevance to the Payne/Saunders story about the plausibility of 1949 data. In it I plot the proportion  (with 95% confidence intervals) of fathers of survey respondents in 1949 and 1959 in each of the 7 Hall-Jones (H-J) occupational status categories. The samples refer to men who were 25-55 in 1949 and 35-65 in 1959 ie the second survey pertains to roughly the same population of men ten years further on in their life-course - reporting on about roughly the same population of fathers. Because the extant data from 1959 refer to the population of ever married men I've selected the 1949 sample using the same criterion.


It is important to note that the confidence intervals surrounding the estimated proportions are "optimistic". They take no account of the complex sample design used in each survey and, of course, do not reflect "nuisance" sources  of variation introduced by the fact that the data were collected by two different organizations which inevitably would have had different interviewing procedures, coding procedures etc. We also have to accept that protocols for coding occupational information were not at that time as standardized as they are today.  The surviving records on how to code occupations to the Hall-Jones scale suggest that the protocols were far from complete and the procedures were in fact probably never completely codified. It is likely that some reliance was made of tacit knowledge - which is of course now lost. 

The main plank of the Saunders/Payne criticism of the 1949 data is that the origin distribution contains an implausible proportion of men with father's from white-collar backgrounds (roughly H-J values 1-4). My impression from looking at the graph is that there is a fair measure of agreement between the two surveys with regard to these proportions.  Certainly the differences are minor and could easily be accounted for by the considerations I mentioned above. Where there is divergence is in the proportions allocated to categories 5 and 6 - very roughly skilled manual and semi-skilled manual workers. This is scarcely puzzling. Without a carefully standardized set of protocols the allocation of occupations to these two categories must have been difficult.  They had 10 more years of experience and no intention of comparing the 1949 and 1959 surveys  and therefore no compelling reasons for coding occupations in exactly the same way. Nevertheless, roughly speaking, the two sources agree about the manual/ non-manual split as a whole, at least with regard to the non routine non manual grades. The latter qualification prompts a further caveat  which is that H-J category 5  (in the 7 group version of the scheme) is actually  "skilled manual and routine grades of non-manuals". This further vitiates the kind of Census comparison attempted by Payne et al.

Data collected in the middle of the Twentieth Century inevitably looks imperfect if we apply the standards of today. We have to accept that in the past they did things differently. This does not imply, as Saunders and Payne appear to believe, that the 1949 data is  fatally flawed and thus gives an unreliable portrait of origin to destination social mobility for the period it was intended to cover. Before you draw such sweeping conclusions you have to do the hard work of looking, not just at the data, but at the right data. That is  an important part of what real scholarship involves. Cheap shots and smears we should leave to the tabloid pundits. They are the professionals.

(1)  Payne, G. G. Ford and C. Robertson (1977)  'A Reappraisal of Social Mobility in Britain', Sociology, 11, pp. 289-310.
(2) Duncan, O. D. Methodological issues in the study of social mobility."pp. 51-97 in Neil J. Smelser and Seymour M. Lipset (eds.), Social Structure and Mobility in Economic Development. Chicago: Aldine.

No comments: