Some of you may have seen this story in the Guardian. I have been avoiding commenting on it until after the comments section was closed, because I hope to do some discourse analysis on both the newspaper report and the comments section, to try to find out how people are using the terms intelligence and heritability in everyday(ish) language.
The Guardian also usefully made the source material available. What I want to do in this blog entry is discuss the substantive points Dominic Cummings makes with regard to education, genetics, heritability and IQ.
IQ as a measurement of Cognitive Ability
On page 68 of Some Thoughts on Education and Political Priorities Cummings suggests that if you ask the question of people what proportion of kids with an IQ of X could master integration, and then goes on to say that research on children with different cognitive ability should be done.
IQ, of course, isn't a measure of cognitive ability. If one wanted to do this research then measuring actual cognitive ability (in mathematics) would be the thing to do, not measuring IQ. One reason for this is that IQ tests are made up of several different abilities, another is that the nature of standardisation of the IQ tests obscures the underlying cognitive abilities.
It may be that the people that Cummings has asked this question are just being polite by saying that no empirical research has been done, rather than pointing out to him the stupidity of the question.
IQ not the only thing that affects academic performance
Cummings cites Lewis Terman on the idea that IQ is not the only thing that affect academic performance. This is of course true, but given that IQ was developed to predict educational attainment I tend to see this as a proof of failure of concept rather than anything else.
The extremely limited capacity of short term working memory (p72)
In talking about the way that basic research is being ignored in education Cummings talks about the extremely limited capacity of short term working memory being a reason why information may not be retained from a lecture. I often use short term memory as a great example of the difference between academic and lay understandings of psychology concepts. Short term memory refers to the few seconds after being exposed to, say a list of words, and the capacity of short term memory is normally seen as 7 plus or minus two things.
Working memory is a different topic it is a theoretical framework that refers to structures and
processes used for temporarily storing and manipulating information.
So short term memory doesn't include memory of what happened over the last hour.
And short term working memory appears to be a conflation of two separate things.
On page 72 Cummings begins his confused ramblings about genetics.
The heritability of educational outcomes
Cummings cites with approval some work by Plomin on using heritability statistics on things such as phonics tests (70% heritability), reading and maths tests (at ages 7,9 and 12) as showing 60-70% heritability and GCSE as showing approximately 60% heritability.
Cummings is careful to talk about heritability as being about the amount of variance as due to genetic factors, but, like Plomin, seems to have a willfully ingenuous view on what, if anything, heritability statistics might reveal.
Like that other greatly misused statistic regression there appears to a tendency to conflate the assumptions of the test with what the test can tell us.
With regression one of the assumptions of a regression model is that the predictor variables cause the dependent variable; however it is often used as if the regression equation reveals whether or not a predictor variable (in a battery of tests) is causal.
With heritability there are a whole bunch of assumptions that need to be met before the statistic can be used.
The heritability statistic was created by Ronald Fisher who is also responsible for creating the ANOVA statistic so beloved in social sciences. A lot of Fisher's work related to problem particularly in crop breeding, where it could be know with a high degree of certainty the precise environment crops were exposed to, and a good idea of what we would now call the genetic background of the crops.
While there is no need to know which gene (or genes) are responsible for any trait that may be open to heritability analysis if it is assumed there is more than one gene they need to act upon each other in an additive way, rather than interacting with each other.
The other assumption is that 'the environment' acts in an additive way rather than interacting with the genetic component.
These assumptions seem far fetched when it comes to IQ tests. They are stretched beyond any credible breaking point when it comes to testing educational outcomes.
I have no doubt that genetics and environment play an important role in educational outcomes, but I am sceptical that we can reduce the complexities of their interactions to a single number. Even if we can then heritability is not that number for any trait where there is interaction between genes or interaction between genes and the environment.
