« Call for help | Gene Expression Front Page | Religious Koreans aren't stupid! »
February 04, 2004

A New Cognitive Elite?

In a couple of posts (here and here) I looked at some recent studies on social mobility. The general drift of the evidence was that social mobility in western countries had always been quite high (at least since 1900), and there was no obvious trend for mobility to increase - if anything, it might recently have declined.

A number of comments on my posts suggested that societies are becoming increasingly stratified with respect to intelligence. People with high IQ are becoming more concentrated in the upper social strata, while people with low IQ sink into a cognitive underclass. On this view, social mobility would be expected to decline, and eventually come to a stop, because all the high IQ people will be in the high-IQ occupations.

These suggestions puzzled me a bit, as they didn’t seem to tie up with the evidence I was aware of. So I decided to look into the question more closely......

At first sight it may seem obvious that if people with high IQ move up the social scale, then high intelligence will become increasingly concentrated in the higher socio-economic classes. If you shake up a bottle of milk, and then leave it to stand, the cream will rise. The process is cumulative in its effects. Eventually most of the cream will be concentrated at the top of the bottle and ‘cream mobility’ will stop.

Social mobility does not work this simply. Social class is not perfectly correlated with IQ. Many factors other than IQ influence occupational outcomes, and the correlation between IQ and socio-economic status (on a 5-class scale) is only about .5 or .6. Even these moderately high correlations leave great diversity of IQ in each social class. If diversity is measured by the standard deviation of IQ around the regression line, then a correlation of .5 only reduces diversity by about 15% of its level in the whole population. The correlation would have to be over .85 to reduce diversity by as much as half. There may be less ‘scatter’ of IQ in particular occupations than in a broad social class, but even scientists (presumably among the most intelllectually demanding of all occupations) have a wide spread of IQs (see e.g. the classic study by Anne Roe).

The inheritance of IQ is also imperfect. The correlation for IQ between parent and child is only about .5, which is about what would be expected if IQ differences within a population are mainly genetic. (Assortative mating would tend to increase it, but dominance and epistasis would tend to reduce it, compared to a simple additive model.) Parents frequently produce offspring much brighter or dimmer than themselves.

These factors tend to offset the ‘sorting’ effect of social mobility. Returning to the milk bottle analogy, we should imagine the bottle being gently shaken all the time. At some point we would expect the amount of cream falling as a result of shaking to balance out the amount rising through buoyancy. This is, roughly speaking, Cyril Burt’s theory of social mobility, and it seems consistent with most of the evidence.

However, it has been argued that the intensity of ‘IQ sorting’ has increased over the last 50 years. In Herrnstein and Murray’s terms, this leads to creation of a cognitive elite. It’s a long time since I read The Bell Curve, so I thought I would dust it off and see what H & M have to say. They argue that modern societies are increasingly efficient at identifying the brightest young people and guiding them into narrow educational and occupational channels (p. 25). More young people are getting higher education, but entry to HE has been based more on cognitive ability, as measured by SATs and similar tests (30-35). Colleges themselves have become more preoccupied with cognitive ability, leading to increased differentiation between different colleges (40). A higher proportion of high-ability youths are going to college, and the link between ability and higher education has become tighter (34-35). After college, a higher proportion of the high-ability group go into occupations (the professions) requiring high IQ for success (56). Whereas previously high-IQ people were quite widely scattered through a variety of occupations, a higher proportion are now concentrated in the high-IQ occupations (61). High-IQ people are also more likely to live in the same localities (103-4), and to marry each other (111), though H & M admit they have no direct evidence for an increase in assortative mating with respect to IQ.

This is all quite interesting and plausible, but so far as IQ is concerned it is inconclusive. It is possible that as the proportion of people obtaining higher education increases (due to economic growth) the link between intelligence and occupation is increasingly mediated through educational qualifications, without the link itself becoming notably stronger. The acid test is whether the correlation between IQ and occupation has actually increased.

I can’t find any direct evidence on this point in The Bell Curve, but on skimming through my bookshelves I find little to support the hypothesis, and quite a lot to call it in doubt.

For example, L. E. Tyler, The Psychology of Human Differences, 3rd edn, 1965, p. 336, gives army test data from World War 1 which show occupational classes already well-differentiated by IQ, contrary to the impression given by The Bell Curve (e.g. p. 61: ‘In midcentury, America was still a society in which a large proportion of the top tenth of IQ, probably a majority, was scattered throughout the population...’). N. J. Mackintosh, IQ and Human Intelligence, 1998, p. 114, quotes data in which the difference in mean IQ between the highest and the lowest social classes was very similar in studies spanning 60 years. This conflicts with H & M’s hypothesis, since ‘ceiling’ and ‘floor’ effects, combined with a wider spread of IQs in each class, would affect the mean IQ of the highest and lowest classes.

Explicit correlation data also seem to have been broadly unchanged over the period in which IQ has been studied (roughly from 1920 onwards). The correlation between IQ and occupation can be measured directly, by the correlation of between the IQ of individuals and their own occupational status as adults, or less directly, by the correlation of children’s IQ with the occupation of their fathers. (The latter correlation is naturally somewhat lower than the former). Studies carried out over a very long period seem to produce consistent figures, namely about .5 for the direct correlation, and about .35 for the indirect. For example:

1. P. E Vernon, Intelligence and Attainment Tests, 1960, p. 143, ‘The correlation of father’s occupational level with child’s IQ is consistently found to be about .35’.
2. N. J. Mackintosh, IQ and Human Intelligence, 1998, p. 114: ‘A meta-analysis of a substantial number of largely American studies estimated that the average correlation between parents’ SES and their offspring’s IQ was .33’.
3. Mackintosh, p. 145 ‘The correlation between adults’ social class and their IQ scores is about .5 to .6’.
4. A. R. Jensen, Educability and Group Differences, 1973, p. 151: ‘...substantial correlation, averaging between .4 and .6 in various studies, between indices of SES and phenotypic intelligence is one of the most consistent and firmly established findings in psychological research, and it holds true in every modern industrial society in which it has been studied’.
5. A. R. Jensen, The g Factor, 1998, p.491: ‘The population correlations between SES and IQ for children fall in the range .3 to .4; for adults the correlations are .5 to .7, increasing with age as individuals approach their highest occupational level’.
6. L. E. Tyler, op. cit., p. 343 reports longitudinal results from 1937 for men originally tested in 1923 and 1918: the correlation of earlier IQ with their occupation in 1937 was .57 for the 1923 testees and .71 for the 1918 testees. As Tyler says, ‘both these figures show a substantial relationship, more pronounced where the interval was longer’. These 1937 correlations are much the same as those given by Jensen 60 years later!

Nor could I find any direct evidence that assortative mating for IQ is becoming stronger. However, one well-known implication of stronger assortative mating is that the variance of children’s IQ in the population as a whole would increase. I don’t think psychometricians have noticed such a trend, and in fact Teasdale and Owen’s Danish studies suggest that the variance may have fallen, though this is probably due to greater educational equality rather than any changes in assortative mating.

I conclude provisionally that the hypothesis that societies are becoming increasingly stratified with respect to intelligence is not supported, and is prima facie refuted, by direct evidence. As so often happens: nice theory, shame about the facts.

Posted by David B at 11:08 AM