« Wiki | Gene Expression Front Page | Bohmian Mechanics »
January 20, 2004

Intelligence and education

In comments on my recent post on intelligence and social mobility one commentator drew attention to a recent discussion paper:

Fernando Galindo-Rueda and Anna Vignoles: Class ridden or meritocratic? An economic analysis of recent changes in Britain.

This is apparently not yet published in print, but is available as a PDF file here. (Warning: 2.63 Mb download.)

This is one of the most interesting and impressive studies I have seen on the subject. (Many thanks to ‘Economist’ for pointing it out.) I would urge all psychometricians, sociologists and economists to read it. It is based on comparison of two longitudinal studies in the UK with very large and representative samples: the National Child Development Study, which follows a cohort born in 1958, and the British Cohort Study, which follows a cohort born in 1972. The paper examines the importance of cognitive ability and parental social class in determining educational and economic success. It also includes some discussion of the effects of social class on cognitive ability itself.

For British readers, the most interesting feature may be the authors’ claim, which appears well-supported, that cognitive ability became relatively less important, and social class more important, in determining educational achievement between these two cohorts. As the authors put it, the educational system actually seems to have become less meritocratic, despite educational changes intended to reduce social advantage and disadvantage. Perhaps the most striking finding is that for children with high ability but low SES, the probability of obtaining a higher education qualification actually fell during the relevant interval, despite a general expansion of higher education.

Personally, I do not find this as surprising as the authors. These two cohorts (aged 11 in 1969 and 1981 respectively) straddle the period in which the 11-plus examination was abolished and selective grammar schools largely replaced by comprehensive high schools. The 11-plus was not a perfect instrument of meritocracy, but it was better than nothing. There were plenty of people who warned at the time (late 60s) that comprehensivisation would damage the prospects of poor bright kids.

For non-British readers, the most interesting part of the paper may be section 5, which considers the determinants of cognitive ability (alias IQ) itself. The paper provides some evidence that social class became a stronger influence on IQ at age 10/11 in the later cohort (born 1970) than the earlier one (born 1958). This seems to provide some support for an ‘environmentalist’ intepretation of IQ differences. However, I think this section of the paper needs to be scrutinised carefully by expert psychometricians. Even for a committed ‘environmentalist’, the findings should be problematic, as it is difficult to see what environmental changes could have made a large difference to the importance of social class over a period of only 12 years. (Changes in the British education system are unlikely to be the explanation, as these did not affect primary schools (age 5 to 11) as much as high schools.)

The authors themselves recognise a technical problem with the data, in that the earliest cognitive ability test in the NCDS was at age 7, whereas in the BCS it was age 5 (with no test at age 7). They therefore have to compare the predictive power of a test at age 7 (in predicting IQ at age 10/11) with the predictive power of a test at age 5. This introduces a source of measurement error variance which might affect the results. The authors attempt to allow for this (see pages 30-31), but I am not sure they have made enough allowance. It is well known that IQ tests on young children have low reliability, and there is a large increase in predictive power between ages 5 and 7 (see e.g. Robert B. McCall ‘Childhood IQs as predictors of adult educational and occupational status’, Science (1977), vol. 197, 482-3). We might therefore expect the predictive power of the early-IQ variable to decline between the NCDS and BCS cohorts, and the predictive power of parental SES in a partial correlation analysis might increase, since less of it would be ‘partialled out’. It is also interesting that the total explanatory power of all the variables, as measured by R-squared, seems to have fallen substantially between the two cohorts (see the last line of Table 8). Since the change from testing at age 7 to testing at age 5 is the main difference in the explanatory variables, this may suggest that the apparent decline in the importance of early IQ, and the increase in the influence of parental social class, is a statistical artifact. (Warning: I may have misunderstood the technicalities in this discussion, so politely correct me if I’m wrong.)

To keep the matter in perspective, the paper should not be interpreted as showing that the British system has become 'unmeritocratic'. It appears to have moved from a position which was quite highly meritocratic to one which is in some respects less meritocratic. In other respects (e.g. the economic return to cognitive ability, especially for women), it has become more meritocratic over the period studied.

But with all reservations, this is clearly an important study. As the authors point out, in some respects the British data are more useful than the analogous ones in the US, so they ought to be widely noticed.

Posted by David B at 05:38 AM