Do schools matter less than we think?

//Do schools matter less than we think?

Disturbingly for all of us involved in education, it seems as if schools and teaching may matter a lot less than we would like to believe. Before setting out the arguments I want to make it clear that this is a struggle for me and I really don’t want it to be true. That said, being professionally sceptical requires that we doubt what we want to believe as much – more – than the stuff that’s obvious guff.

In order to understand what comes next, I’m going to take the liberty of providing a quick refresher on the mechanics of behaviour genetics. When we want to find out to what extent a trait like intelligence can be attributed to genes or the environment, we usually start by looking at identical twins because we know they share the same genetic material. Therefore, we assume that the extent to which twin pairs have similar IQ scores is due to heritability while the extent to which their scores are different can be put down to differences in the environment. Simple.

Or not so simple. Most identical twins are raised by the same parents in pretty much the same environment and yet, although they’re very similar, they’re not exactly the same. For instance, about a fifth of identical twins don’t share handedness – one will be left-handed while the other is right-handed, or sometime, twins can have different sexualities – with one being gay, the other straight. If both their genes and environments are the same, what could be causing these differences? The truth is, we don’t really know, but act as if we do. We label the cause of any differences between identical twins raised together as ‘non-shared environment’. 

In my last post I showed that shared environmental factors, like parenting, have pretty much no longterm effect on adults and that pretty much every human trait can be attributed fairly equally between heritability and non-shared environment.

For the past thirty years or so, researchers have decided that non-shared environment probably includes things like the unique experiences each twin has at school, or the different peer groups each belongs to.* This is plausible in most cases, but is harder to accept in the case of identical twins. Although I’m sure there will be many identical twin pairs who buck this trend, many – perhaps most – will end up in the same class at the same school with the same teacher and the same peer group. Of course each twin will have an ever increasing variety of unique experiences as they grow up, but can we plausibly conclude that these small and irregular differences really account for so much difference.

However, non-shared environment is, in reality, a slops bucket into which researchers stuff everything that isn’t nature or nurture. We don’t really know what it contains. This post by Scott Alexander suggests various possible factors that might make up ‘non-shared environment’ that aren’t anything to do with schools or peers. The factors he considers are measurement error, luck, and various small biological differences. Let’s briefly consider how each of these might contribute to the differences that show up when identical twins are tested for different traits.

Measurement error

Most human characteristics are tricky to measure and even stable ones like intelligence as measured by IQ are subject to predictable measurement problems. Alexander gives this example:

Imagine a world where intelligence is entirely genetic. Two identical twins take an IQ test, one makes some lucky guesses, the other is tired, and they end up with a score difference of 5 points. Then some random unrelated people take the test and they get the 5 point difference plus an extra 20 point difference from genuinely having different IQs. In this world, scientists might conclude that about 80% of IQ is genetic and 20% is environmental. But in fact in terms of real, stable IQ differences, 100% would be genetic and 0% environmental.

Most other traits are much harder to measure than intelligence and rely on much fuzzier tests, self-reports and other error-prone tools. As a result, it’s hard to know if what shows up as ‘non-shared environment’ is actually the effect of genes in disguise. This study apparently shows that when all these factors are taken into account, it looks like the effects of non-shared environment tend to average out at 15% rather than 50%.


It’s theoretically possible for heritability to account for all the differences between individuals and yet for each to be subject to the slings and arrows of outrageous fortune. Alexander gives the example of identical twins who get identical jobs in the same industry and work equally hard. One has a great boss who recognises his talents, the other has the bad luck to be supervised by an unprincipled fool. One gets promoted, the other doesn’t and any researchers studying their experiences might draw the conclusion that the unlucky twin may have done something to deserve his lack of promotions. They start looking and notice that the twins were in different Year 7 maths classes and decided one must have had an inspirational teacher. Or maybe they see one twin had a slightly different friendship group who were a little less academically inclined. Post hoc explanations are troublingly easy to find.

Biological differences

It turns out that identical twins don’t actually have 100% identical genetic material. Although they both come from the same zygote, each may be subject to slightly different random mutations. These tiny difference can magnify over time resulting in 100s of tiny genetic differences. Add to this the fact that all identical twins have subtle biological differences – fingerprints, freckles etc. – and you can imagine that these tiny deviations might add up to something greater than its parts. The point is, we don’t really know. We also don’t really know what difference epigenetics might make. The environment interacts with our genes in all sorts of subtle and profound ways. Identical twins may show substantially different epigenetic effects which can lead to genes being expressed in different ways. We’re on the fringes of discovering more about how all this effects heritability estimates – as we learn more we may find some of what we think of as environmental differences have as much to do with different gene expressions. Time will tell.

The point is, because we don’t know, we can’t really be sure whether the effects of schooling are as important as we want them to be. Just like parenting, it seems obvious that our experiences in school must account for some of the differences between us, but can we take that for granted? There are very obvious ways in which schooling and education matters: if you don’t learn to read then your life chances will be effected. Of course some children will find it harder to learn than others, but pretty much all children can, with sufficient support learn. I’ve argued before that if a child leaves school unable to read it is the school’s fault. The counter-intuitive truth is that as access to education becomes more equitable, children experience less and less environmental differences. In a truly equitable system, all of the differences between children would be due to heritability.

But beyond the brute force logic underpinning what we teach – if a child doesn’t know something he or she cannot then think about it – I wonder whether how we teach might be far less important to how children turn out as adults than we like to think. As I said at the outset, I’m not sure about all this and I certainly don’t want it to be true, and it’s not popular to go about attributing children’s success or failure to who they are rather than what they experience. That’s by the by: what we want to be true, or what’s popular should never be the foundation for our thinking.

To be clear, none of this is an argument for genetic profiling or eugenics or any other lunacy. Who we are is wonderfully unknowable and in some ways it’s reassuring to think we’re less at the mercy of our environments than is often supposed. maybe it’s a good thing our biology cannot be so easily derailed? Whatever the case, the universe is profoundly uninterested in what we want to be true. As ever, all we can do is our best; that will have to do.

* The arguments in favour of peer effects are made cogently and persuasively in Judith Rich Harris’s masterpiece, The Nurture Assumption

2017-08-12T17:44:15+00:00August 12th, 2017|research|


  1. Grumpywearymathsteacher August 12, 2017 at 6:45 pm - Reply

    You say “in a truly equitable system…” But what we have now is nowhere near that, so schools (and parents) DO matter.

    • David Didau August 12, 2017 at 10:03 pm - Reply

      Yes they do. But much less than we think. Or at least, that’s what I’m coming to believe

  2. Sara Hjelm August 12, 2017 at 8:19 pm - Reply

    I must agree with that comment .. And, even if not cleverer, if we by that mean intelligence they become knowledgable, get around more easy as they know. What, where, who, why, etc, etc
    The huge advantage would be for the individual to be able to be more strategic and make better and above all quicker choices. For what means is another question.

  3. Elizabeth Nonweiler August 14, 2017 at 12:17 pm - Reply

    As well as slightly different random mutations, twins have different environments from the start. For example, one is probably more squashed in the womb than the other.

  4. kesheck August 14, 2017 at 5:45 pm - Reply

    If I understood one of Michael Anderson’s points in his book “After Phrenology: Neural Reuse and the Interactive Brain,” the benefit humans gained due to the lack of “hard-wired” instincts found in other animals is our brain’s ability to forge neural networks in response to environmental stimuli.

    In other words, animals can take care of themselves very shortly after birth because they are essentially born with much of the programming that allows them to survive. The range of environments in which most animals can survive, however, is severely limited.

    Humans, on the other hand, are pretty much helpless until they’re old enough to learn to drive. The survival benefit we gain from the lack of cognitive pre-programming is our brain’s neural flexibility.

    Reading is an example of this ability. We can read because our brains, under the right stimuli, connect groups of neurons that have, in Anderson’s words, “dispositional tendencies” to respond to phonological inputs like speech sounds and visual inputs like various types of lines. I’ve taught enough struggling 13 and 14 year-old readers of varying academic proficiency to believe that how reading is taught is crucial.

    I could bore you with my take on Fauconnier and Turner’s idea of conceptual blending, and how I think that idea reveals the importance to imagination and creativity of what’s stored in long-term memory and how that knowledge is organized. But I won’t bore you with that, other than to say what’s stored in LTM is a function of environmental input – most importantly, what we’re taught.

    I absolutely believe that much of ability is heritable, and I agree with much of what you write here. I’m just more hopeful about the importance of schools – but perhaps that’s just my bias at work. 🙂

    • David Didau August 14, 2017 at 5:53 pm - Reply

      Oh, I’m hopeful too – I just don’t think hope should blind us.

  5. Debaser August 15, 2017 at 2:35 pm - Reply

    ‘But beyond the brute force logic underpinning what we teach – if a child doesn’t know something he or she cannot then think about it – I wonder whether how we teach might be far less important to how children turn out as adults than we like to think.’

    Ok, but surely this is another argument to suggest that there should be much more focus on content, knowledge and curriculum design in schools and a lot less on pedagogical gimmicks.

    Perhaps the extent to which a school ‘matters’ should be measured by the extent to which they equip their students with the knowledge necessary to fulfil their genetic potential and everything else – character, ‘skills’ etc – is less important than we might think.

  6. […] Twitter following David Didau’s recent blog posts that question the effect that families and schools have on children. I suspect that many teachers go into teaching in order to make a difference. So […]

  7. […] Do schools matter less than we think? (The Learning Spy) Disturbingly for all of us involved in education, it seems as if schools and teaching may matter a lot less than we would like to believe. […]

  8. TheOtherDrX August 18, 2017 at 7:51 pm - Reply

    An interesting ‘thought experiment’ to test one’s views on the importance of hereditary vs family/schooling is as follows:
    Scenario 1: Take an ‘intellectually average person’ and clone this person (Dolly the sheep-style) into 100 identical clones and place these resulting offspring into 100 random families and schools, ranging from affluent to some of the worst sink estates and failing comps, but to parents representing each of the individual centiles of the population, so as not to skew the data too much.
    Scenario 2: Take 100 random people at birth, give them the the best possible education and family life, of the sort that produces the most able, employable and successful in our society.
    Draw the resulting bell curves after termination of the experiments, lets say, after 40 years, for some measure of success. This could be IQ if you wish, but probably something more reflective of overall ability, such as qualifications gained. Compare these bell curves to 100 random people in born into their respective families, environment and schools etc.
    I have run scenario 1 as a thought experiment several times, and I keep getting different results, in terms of how narrow/broad the curve is and whether there would be a shift from the mean vs the random population, and how many geniuses would there be, and how many would end up in prison. You can the repeat the experiment with cloned geniuses if you wish. Scenario 2 is interesting, and clearly there is a shift to the right. But how far? Would it even be a bell curve, or show considerable skewing peaking around the top quartile for the 100 random people? Would the top end not shift all that much? I don’t know, but the by writing the justification for why you think the curves would be as predicted, you get an idea of what is shaping your views on the relative importance of genetics vs environment. Or not, if you keep getting different results.

    • David Didau August 21, 2017 at 10:28 am - Reply

      You’re asking me to guess how 100 identical people would fare if placed in wildly varying environments? What would this prove? That environment matters? If so, I completely agree. Scenario 2 is, I hope, what an equitable education system aspires to – we can’t get rid of ‘gaps’ per se and the distribution of the curve should be roughly the same as now. What would (might?) be different is that the gap wasn’t caused by SES, or any other environmental factor.

      • TheOtherDrX August 21, 2017 at 11:33 am - Reply

        Yes. That’s what I’m asking. Not guess numbers exactly but in general terms. What would it prove? Nothing, just illustrate one’s thoughts on the relative contribution of genes and environment. That environment matters, yes, but just how far to the right of the curve can environment take you if you are average? If ‘all the way’, then environment is almost everything given a genetic many-up of a representative human. If only up to the 80th centile, then the genetics component is illustrated clearly, and everyone with offspring in Oxbridge can feel smug with their genes. Just how damaging can environment be to centile is just as important, and this maybe where we have the best evidence. A more difficult question is how to fairly identify that person on the 50th centile to start with!
        In scenario 2, yes it is equitable. If genetics is irrelevant, should all be about the same, at the far right of the present curve (barring pathological genetic changes)? It is the magnitude of shift to the right that interests me, and the spread. Would the distribution be much the same as now, only shifted to the right with a perfect bell curve, or would the majority in an equitable system do similarly well and well above the current mean? A more skewed curve to the right suggests stronger environmental beneficial effect. A curve with a normal distribution and a broad and similar spread to now, just moved to the right, suggests strong genetic component giving that wide spread of individuals of varying ability. Gaps in attainment that cannot reasonably be filled.

        • David Didau August 21, 2017 at 11:36 am - Reply

          I believe it’s possible to make the curve steeper, but I don’t have empirical evidence for this. It’s what I want to be true 🙂

  9. Stephen Foster September 22, 2017 at 4:19 pm - Reply

    A significant reason why people people learn is linked to their motivation to do so. I believe teachers have a responsibility to discover and cultivate that motivation, something that I believe can only be done when a trusting relationship is cultivated and this would be difficult with less contact. It has been shown that what students can learn depends to a large degree on what they already know. It is easier to learn something that we can relate to in some way. You made the point about the importance of “what” teachers teach but suggest that “how” we teach is less important, however I disagree. Developing strategies and techniques that help the student utilize information they already know and make it relevant to the class will greatly improve their learning.
    I also believe that people learn in the interests of involvement in communities that are most relevant them. They learn so they can recognize how to be useful in their community, attain access to important aspects of community participation, and ultimately be accepted by that community. Again, a good teacher can recognize this and use an approach that improves the learning process for the student. Simply presenting information to people in a general manner will of course produce different results as we all learn differently.

  10. Dylan Wiliam (@dylanwiliam) September 27, 2017 at 1:50 am - Reply

    I have used both PISA data and the DfE’s value-added data on schools, and both datasets, independently, suggest that in England around 7% of the variation (variance) in student test scores is attributable to the school they attend. What this means is that only around 10% of students are close enough to an achievement threshold for the choice of school to make a difference to whether they reach the threshold or not. For the other 90%, they are either so far below the threshold that going to the best school in their area won’t get them over the threshold, or so far above it, that they still clear the threshold in a school chosen at random. There are many reasons to choose one school over another (sport, drama, music). Academic achievement is not one of them.

    • David Didau September 27, 2017 at 2:13 pm - Reply

      Thanks Dylan, that’s fascinating. Have you published anything on this?

  11. […] Do schools matter less than we think? (The Learning Spy) Disturbingly for all of us involved in education, it seems as if schools and teaching may matter a lot less than we would like to believe. […]

Constructive feedback is always appreciated

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: