Welcome all, and thank you for coming. In terms of qualifications (list them) I am probably the least important speaker at the whole festival, but the good news is that rather than just recycle stuff I’ve said previously like all the busy eminent persons will largely do, you get to hear something new. Today you’re going to hear to a general talk - less focused on than my researchEd talk on behaviour genetics - on the extent of human variation, how we measure it, and the probable limits of educability.

Now, you’re all going to sit an intelligence test. Don’t worry, it’s only a short one. You can all blame Martin Robinson for this idea. On your chairs you should have all found a sheet of paper with 10 words on it. If you would be so kind as to indulge me, I’d appreciate it if you could you all please write a definition for each word down. Don’t worry, I’m not collecting any data here. You have two minutes.

The vocabulary test you have just completed is known as WORDSUM. It is administered to a large population-representative sample of ADULTS as part of America’s General Social Survey, run by the University of Chicago. I’ll assume that nearly all of you knew what all ten words meant, or at least 9 out of the 10. They seem straightforward, do they not?

Here is the national American mean on WORDSUM. If you knew what all ten words meant, this gives you a rough minimum IQ of 125, or at the 95th percentile, to put it another way. I hope you all feel better about yourselves now for winning the genetic lottery - do feel free to ring your parents up later for thanking them for having sex. But more about that later.

WORDSUM is a pretty lousy intelligence test, in psychometric terms. Only having ten items means that it lacks the ability to differentiate between people with roughly similar intellects, and it suffers having a relatively low ceiling - too many people get perfect scores. Not enough of the items are of moderate difficulty. While it may not be the best measure of individual differences, it is a cheap and cheerful way of assessing population capabilities, with scores correlating at a surprisingly high .7 with more standard full IQ test batteries such as the Weschler.

Let’s think a little more about what the scores mean. 1 in every X American adults, after a full childhood and adolescence of education, cannot provide an adequate definition of A and B. Similar examples of remarkable ineptitude are found in responses to the NAEP test questions. My thanks go to Charles Murray for permitting me to borrow these from his excellent book Real Education.

I argue that the fundamental problem that people who fail these tests face is not a lack of high-quality education, but a lack of intelligence. This deficit comes with large consequences, both for the individual and for society as a whole. At this point it is customary to begin a debate over what intelligence does and does not mean. I am not the Academy of the English Language, so I propose to skip this futile debate and simply identify intelligence with the construct that is measured by IQ and scholastic aptitude tests. If you do not like calling that construct intelligence - perhaps because your personal definition of “intelligence” includes emotional intelligence or other abilities - then try thinking of “general mental ability” instead.

The key word is general. It is a commonly held folk belief that mental ability is specific: if Tarquin is good at maths then he’s more likely to be bad at English. Models such as “multiple intelligences” propagate this idea by arguing, explicitly or implicitly, that cognitive aptitudes are uncorrelated. The most replicated result in all of psychology, and the greatest triumph of psychometrics, is the definitive proof that this is false. In fact, if Tarquin is above-average in maths, he is likely to be above-average in English, and all other subjects as well. It has proved impossible, in fact, to devise a pure test of a specific ability - and eminent psychometricians spend decades trying. Every possible mental test is, to some degree, measuring the same thing - a very general ability that underpins all advanced cognition. The importance of GMA to success in life was described in typically amusing fashion by the English polymath and genius Sir Francis Galton:

“People lay too much stress on apparent specialities, thinking that because a man is devoted to some particular pursuit he would not have succeeded in anything else. They might as well say that, because a youth has fallen in love with a brunette, he could not possibly have fallen in love with a blonde. As likely as not the affair was mainly or wholly due to a general amorousness”

From this it follows that if your purpose is to measure an individual’s level of general mental ability, it does not matter much which tests you use. The English psychologist Charles Spearman, who provided the first proofs of the existence of general mental ability, called this principle “the indifference of the indicator”. Some tests may have practical or psychometric advantages over other tests. Conventional IQ tests are carefully constructed so that test-retest reliability is very high: much higher, in fact, than the reliability of some medical tests administered by skilled physicians (SEE SLIDE). On one level, asking a person to tie their shoelaces is a mental test, but a bad one because of the very low ceiling, and it is also contaminated by physical, non-cognitive ability. WORDSUM, the test you all sat earlier, is a much better test, but again it has ceiling problems. Advanced test batteries, such as the Weschler, are designed to avoid this and other problems: the Non-verbal tests, such as Raven’s Progressive Matrices, the Porteous Maze, and the Cattell Culture Fair, were created for cross-cultural work, and for use with groups who will struggle with the native language through no fault of their own, such as recent immigrants and the Deaf. The forms of these tests are all very different (see SLIDE). But they were all created to measure the same construct - general mental ability - and despite the dissimilarity of form, the correlations between the tests are high. Note that some tests with extremely dissimilar forms actually correlate more highly with each other than they do with tests with more ostensibly similar forms (see intercorrelations on slides - here in the talk I pointed out that Forwards and Backwards digit span actually correlate less highly with each other than do Vocabulary and Block Design)

Over time, as samples got larger and analysis became more sophisticated, researchers arrived at a consensus as to the broad structure, on the psychometric level, of human mental ability (SEE SLIDE). In Spearman’s original formulation, each test tapped into both general ability (his famous g factor) and specific ability required for that test only. Over time, because some related but separable tests correlate more highly with each other than they do with other tests, it was acknowledged that group factors of mental ability exist (visual/verbal/mathematical-spatial). When Howard Gardner wrote about his multiple intelligences theory, he had, to a substantial degree, simply relabelled group factors and ignored the general factor that arises out of their intercorrelations, which is a bit like a modern-day playwright deciding to rewrite Hamlet and leave out the role of the Prince. This is particularly so because it is the general factor that has the strongest predictive validity across domains.

The first IQ tests were atheoretical, constructed by the French researcher Alfred Binet so that there would exist a method of identifying children who would probably struggle in school. These children could then be targeted for remedial education. The tests became popular purely because they worked. Unsurprisingly, perhaps, the predictive validity of IQ tests in the domain of academic achievement remains extraordinarily strong. In a population-representative sample of 70,000, eminent Scottish researcher Ian Deary and colleagues found a correlation coefficient of .8 between IQ measured at age 11 and a composite scores derived from GCSEs taken at age 16, 5 years later. In the realm of biosocial science this is an extraordinarily high correlation, and suggests that, after aggregating over subjects, most variation in GCSE achievement is due to variation in intelligence. (here in the talk I pointed out that this data also shows that IQ is not all that matters, since boys and girls are equal in IQ but girls do significantly better at GCSE 5 years later)

It also suggests that ability, not income, is a better predictor of need. We allocate pupil premiums on the basis of free school meals eligibility, which is itself based on parental income. Yet the income-GCSE correlation is a relatively lowly .3 - fairly unimpressive compared to the far more substantial IQ-GCSE correlation. I strongly suggest that a more sensible solution would be simply to measure ability at age 11, when children leave primary education, and allocate pupil premiums accordingly. In a world of limited resources it does not seem to make huge sense to spend extra money on the poor but highly able. Of course, because IQ and income are themselves correlated (again at around .3) under my proposed system a disproportionate share of pupil premium money will still go to the economically underprivileged.

At this point I hope to have convinced you that in general, by far the strongest single predictor of an individual’s educational performance is their ability. Incidentally it is also the best single predictor of job performance across domains. Earlier I implied that you could thank your parents having sex for your ability. What did I mean by that? In a research program that dates back - again - to Galton, scientists have used a variety of methods from twin studies to adoption studies to genotype-phenotype correlations in large samples of unrelated individuals to show that IQ is highly heritable. What does this mean? It means that a lot of the variation in the phenotype - IQ - is due to naturally occurring variation in the genotypes of the population. I spoke a lot more about this in my talk at researchED last year, which you should have come to, but for those I didn’t, I’ll post the script again on my twitter after this is over. Briefly, for the uninitiated, here’s an overview of the main methods used to assess heritability: my apologies for those who also went to Plomin’s talk yesterday

(slides: identical twins raised apart - how different are they)

(slides: twins raised together - differences in similarities)

(slides: adoption)

(slide: GCTA)

In fact one of the more surprising findings is that Everything is Heritable, not just IQ but personality, willpower, effort, whatever educationally-relevant phenotype that you’re interested in, genetic variation explains a pretty fair chunk of the phenotypic variation. Also families don’t matter very much when you get older. The home environment matters most for younger children, but over time the percentage of variance attributed family upbringing tends to get smaller and smaller, in many cases going to zero by adulthood. In the long run, parents and children resemble each other because of their shared genes, not their shared experiences. This is not JUST true for IQ, it’s true for almost every trait. So remember that whenever you see some article in the newspaper saying that parents matter because XYZ just remember that parents share genes with their kids, not just a home. Take the kid OUT of the home and put them with some other parents and they still wind up looking quite similar to their real parents. Does this ever happen? Oh wait, yes it does, it’s called adoption. Separate identical twins for their entire lives and you find out they still share a ton of even very minor and bizarre personal quirks!

So to finish off, because I want lots of time for Q and A, here’s some more public policy proposals. Do I expect any of them to be taken up any time soon?

Stop worrying about achievement gaps so much: income and ability are correlated. If you are smart you likely earn more money than someone who is less smart than you. In turn you pass your smart genes onto your kids, so QED of course there’s a relationship between parental income and child educational performance. It would be a miracle if there was not. In some places this relationship probably gets really artificially limited because the parents are recent immigrants from poor countries who are perfectly average in GMA - in fact probably above average, at least in their native country, but haven’t had a chance to earn much money over here. So think really careful before going on too much about achievement gaps. You aren’t going to abolish them any time soon. Education needs to have achievable goals.

Speaking of achievable goals, here’s another public policy proposal: just teach kids stuff. Pretty much every child can learn to read and write with reasonable proficiency, with some exceptions (hi Jarlath). Stop expecting education to solve your social problems. This is my problem with the vogue for character education, be it the liberal character education of the Freire school that aims to train little soldiers for the bloodbath revolution, or the Tory character education of the Morgan school that aims to produce good little consumers for the machine. As measured by crime rates, there are definitely societal trends in douchebag behaviour - apparently in large-scale decline since the 90s, after having spiked suddenly - massively - from the 1960s onwards. No one knows why in the 60s people turned into assholes. No one knows why they stopped and keep on getting less assholish even when the economy tanks. For sure I’ve never seen any good evidence that schools have anything to do with it. Assholes will always be with you, so they might as well be well-educated assholes.

Also you can’t educate people out of lack of intelligence. There are no miracles to be had. Even the most miraculous of educational interventions that I have come across generally only raise scores by about half a standard deviation, or one grade, or 15 percentage points, and that was in a fairly small pilot - I’d expect the true, replicable parameter value to be lower. Any time a genuine educational miracle happens, I guarantee you cheating is at the bottom of it somewhere. This DOES NOT MEAN that intensive educational interventions for low-ability, at-risk youth are not worth doing. We can and are devising good ones that are worth it, in both economic and human terms. But miracles will not happen.

Until the transhumanist future arrives, in the days when embryo selection for GMA comes to save us all and elevate the human race to a new plane of brilliance, we are more or less stuck where we are. So until then, I have one final thing to say, and I’m going to take a leaf out of Martin Robinson’s book - you haven’t read it? You should. Martin says, as only Martin can in his manner that takes you fifty readings to figure out what he is saying before the meaning beautifully floats itself to the top, that teachers need to stop saying sorry for teaching. Stop justifying education in terms of character development or social justice or economic value or any of this crap. It doesn’t really work and never will. Education is the virtue of civilized man. It is its own alpha and omega. It needs no apology. Savages and infants learn. They are not educated.

\