Heritability doesn’t measure nature vs nurture
Q: Have you read the latest issue of PLoS One?
A: What do you mean? PLoS One doesn’t have “issues”, it’s online-only and publishes papers as soon as they are ready.
Q: Well, have you read the study that led to the headline ‘exam results are influenced by genes, not schools‘, which the Herald says is in the latest issue of PLoS One?
A: Perhaps they mean this paper about heritability of exam results.
Q: Yes, that one. Couldn’t they have just linked to it?
A: My sources tell me linking is harder than it looks.
Q: Whatever. How did they find out that schools don’t matter?
A: That’s not quite what they found out.
Q: Well, they found 60% of education was due to genes, not schools, didn’t they?
A: What would that even mean?
Q: I thought I got to ask the questions.
A: <sigh>
Q: Ok, what did they find?
A: They found that identical twins had a correlation of about 0.8-0.9 between their exam results, and non-identical twins had a weaker correlation, about 0.5-0.6. If you assume that the only difference between identical and non-identical twin pairs is that the identical twins share more genes, and that the genetic and non-genetic contributions just add, you can estimate how much of the variation between twins was due to genes and how much is due to environmental factors. And they end up with estimates from 40% to 60% for the genetic part.
Q: How is that different from the headline?
A: The effects aren’t going to be additive in reality — genetics isn’t going to let you pass a British history test if you haven’t ever studied any British history — so the heritability is just a summary of variation in the current population under the current conditions. What the study really finds is that British schools currently differ from each other less than British kids do. If you made a lot of kids from Beijing or Buenos Aires take the British GCSE tests (or vice versa) they’d probably do really badly, and that would definitely be due to their schooling, not their genes.
Q: And what do the researchers say?
A: Pretty much what I said. The Herald quotes them further down in the story
“Since we are studying whole populations, this does not mean that genetics explains 60 per cent of an individual’s performance, but rather that genetics explains 60 per cent of the differences between individuals, in the population as it exists at the moment.
“This means that heritability is not fixed – if environmental influences change, then the influence of genetics on educational achievement may change too.”
Q: If schools were improved, would the heritability of exam results increase or decrease?
A: That’s a very interesting question. We don’t know. It could be that better schools would have more benefits for people who currently do poorly for genetic reasons, and would reduce heritability; it could be that they would have more benefits for people who currently do poorly for environmental reasons, and would increase heritability; it could be that they would have more benefits for people who currently do well for genetic reasons, and would increase heritability; it could be that they would have more benefits for people who currently do well for environmental reasons, and would decrease heritability. Or it could be that they wouldn’t use exams.
Q: What does this research tell us about the UK’s falling position on the PISA international education comparison? Or is that the fault of Facebook and email, like the UK Schools Minister says?
A: Well, it’s not genetic. The genes of the UK population don’t change fast enough. It’s probably not due to Facebook, either, but at least that’s conceivable.
Q: Are the results surprising?
A: Not especially. Similar correlations are seen in IQ test results, where we do know from changes over time that environmental differences can have a large impact. It’s a bit surprising that test scores are slightly more heritable than IQ test results.
Q: Could you give a less controversial example, perhaps something like height?
A: Height is an excellent example. The fact that siblings have similar heights shows the large genetic component when environments are similar; the fact that most people are taller than their parents or grandparents shows the large environmental component when genetics are similar. If you look in a wealthy Western country, height is about 80% heritable. In medieval Europe it would have been much more sensitive to environment, since the nobility were much healthier and better-nourished than the peasants. And if you mixed together people from medieval Europe and modern Europe, about half the variability would be due to which era they came from.
Thomas Lumley (@tslumley) is Professor of Biostatistics at the University of Auckland. His research interests include semiparametric models, survey sampling, statistical computing, foundations of statistics, and whatever methodological problems his medical collaborators come up with. He also blogs at Biased and Inefficient See all posts by Thomas Lumley »
And a question from the audience:
Do you think the authors should have transformed their data the way they did?
11 years ago
I don’t think the scaling is unreasonable, and to the extent that there is a multivariate Normal model on some latent scale, the optimal scaling will be the one that maximises the correlations.
Am I missing something?
11 years ago
This reasoning makes some sense, though the authors used another argument: “Because GCSE scores are negatively skewed, which is generally interpreted as a ceiling effect, in subsequent analyses, we applied a van der Waerden transformation to all GCSE scores, which normalized the distribution.” I am not familiar with these kinds of studies, but it sounds like being concerned about the normality of the dependent variable rather than the normality of the residual term.
11 years ago
Yes, the evil Normal quantile transformation. I this case I think it’s ok by the latent-variable argument. I don’t like it at all in genetic association studies, and have spent quite a lot of effort fighting it there.
11 years ago
Twin studies have always driven me a bit nuts. Sure, they’re one of the only ways we can disentangle biological and environmental influences on development, but they’re far from perfect. So I’m always skeptical about statements like ‘x% of nose picking is hereditary.’
Here’s another one I found a wee while back:
http://grumpollie.wordpress.com/2013/03/23/twin-studies/
Disclaimer: Unfortunately, as I no longer work in academia, I can’t easily access journals. For this reason my blog post is based on a few assumptions about how that particular study carried out.
11 years ago
This paper is open access, so you actually can read it. They do talk sensibly about the assumption that the difference between monozygotic and dizygotic twins is all genetic, but it is a potential problem.
There are two reasons to think it might not be too bad. The first is that there’s very little different in correlation between same-sex and opposite-sex dizygotic twins. I would have thought the same sort of processes that make identical twins more similar than dizygotic twins would also make same-sex dizygotic twins more similar than opposite-sex ones.
Perhaps more importantly, in the case of IQ test results, recent research using shared genotype fractions among basically unrelated people comes up with broadly the same answer as the twin studies. Before that research I wasn’t at all impressed by the twin studies, but now I think they might actually be ok.
11 years ago
And what difference there is is probably down to the number of shared chromosomes:
Same-sex dizygotic twins would share about 24/46 chromosomes while opposite-sex would share about 22/46 chromosomes, making something like 4% difference in correlation based on chromosome count.
However the effectual genes on the Y chromosome are quite scarce (only a few dozen), negligible compared with the other chromosomes, so there would be even less disparity in correlations than a straight chromosome count would indicate.
11 years ago
It’s even more complicated than that, because of X-chromosome inactivation.
11 years ago
I think the main reason that the English don’t do as well at Pisa is that they study lots of subjects for their GSEs. The average student sits 10 GSE exams*. That make English kids more generalists than NZ kids who have to study English, maths and science (?) plus two others at NCEA level 1. And English, maths and science are the three exams in PISA.
Anyway, the English are on to it by setting up a new accountability measure*. It will probably improve their scores on PISA but I don’t know that it will improve their education system (and the latter seem to matter less in govt initiated school reforms).
*http://www.telegraph.co.uk/education/educationnews/10478876/Children-sitting-too-many-GCSEs-exam-board-boss-says.html
11 years ago
Could be. Is this already true at age 15? I thought the US, for example, had pretty spread-out curricula at that stage.
11 years ago
And that could also be the reason that American doesn’t do that well also (although they also have huge funding disparities between the public schools that the rich and poor go to).
11 years ago
The cut-over date is roughly 1 June so you’d expect about 50% of the kids to be 15 at the start of the year they enter their NCEA L1 year. However, the “roughly” is at the schools discretion.
11 years ago
“A: Well, it’s not genetic. The genes of the UK population don’t change fast enough. It’s probably not due to Facebook, either, but at least that’s conceivable.”
Untrue. Prior to the ‘Windrush’, Britain was essentially 100% British. Now, it is 80% British at best. Naturally, PISA and other intelligence scores decline, along with population quality.
11 years ago
The Windrush was in 1948, as I assume you know. We’re talking about the past half-decade. Gross immigration from all countries is less 1% of the UK population per year. The genetic changes are not fast enough.
11 years ago
I don’t think Britain has ever been 100% British – Britain has always been awash with immigrants – nowadays they just arrive more peacefully.
11 years ago