Debate about genetics and school performance

Jag Bhalla points us to this article, “Differences in exam performance between pupils attending selective and non-selective schools mirror the genetic differences between them,” by Emily Smith-Woolley, Jean-Baptiste Pingault, Saskia Selzam, Kaili Rimfeld, Eva Krapohl, Sophie von Stumm, Kathryn Asbury, Philip Dale, Toby Young, Rebecca Allen, Yulia Kovas, and Robert Plomin, along with this response by Eric Turkheimer.

Smith-Wooley et al. find an association of test scores with genetic variables that are also associated with socioeconomic status, and conclude that “genetic and exam differences between school types are primarily due to the heritable characteristics involved in pupil admission.” From the other direction, Turkheimer says, “if the authors think their data support the hypothesis that socioeconomic educational differences are simply the result of pre-existing genetic differences among the students assigned to different schools, that is their right. But . . . the data they report here do nothing to actually make the case in one direction or the other.”

It’s hard for me to evaluate this debate given my lack of background in genetics (Bhalla shares some thoughts here, but I can’t really evaluate these either), but I thought I’d share it with you.

89 thoughts on “Debate about genetics and school performance

  1. Thanks Andrew for sharing. “Gene effects” in these kinds of studies pose a key test for philosophy of stats vs practice of stats.
    Genes (and SNP variant allelles) can look “independent” but their effects rarely are. And they can’t be disentangled from environmental & cultural & behavioral factors. All of which makes standard stats practices very iffy…

    And the %genes vs %environment framing is great example of how…
    Analysis-of-variance training encourages fallacy-of-division miscalculations.
    Many phenomena are emergently co-caused and resist meaningful decomposition. What % of car speed is “caused’ by engine or fuel? What % of drumming is “caused” by drum or drummer? What % of soup is “caused” by its recipe?
    from https://bigthink.com/errors-we-live-by/judea-pearls-the-book-of-why-brings-news-of-a-new-science-of-causes

      • D Kane:

        What information from that the Genomic Prediction site are asserting supports the argument in the paper discussed? Because I don’t see anything related.

        • edit:

          What information from the Genomic Prediction site are you asserting supports the argument in the paper discussed? Because I don’t see anything related.

        • Bhalla claims (corrections welcome!) that “polygenic scoring” is useless/wrong/misleading, that “standard stats practices [are] very iffy” in the context of genetics.

          The folks at Genomic Prediction thing that this is nonsense, that standard stats works very well in genomics and that polygenic scores work. In fact, they will provide such scores for your embryos, allowing you to select the one to implant, on the basis of things like forecast educational attainment and height.

        • D Kane:

          Where are you finding the research at the Genomic Prediction for educational attainment? I only see the research on diseases and height. Can you provide a link to the study they did on educational attainment?

        • “association ≠ prediction”
          see Cecile Janssens
          ‏”Too many studies report that X ‘predicts’ or ‘is predictive it’s Y when all that is demonstrated is statistically significant association. That’s too confusing when others study the predictive ability of X. Let’s call it what it is.”
          https://twitter.com/cecilejanssens

      • What does prediction do with causation?
        Im pretty sure you can predict US test scores based on individual’s genes. Does not mean that scores are caused by genes.

        • I agree with the point that (I think) Mikhail is making.

          Contrary to Jag Bhalla, very often associations can be used to make predictions: “association = prediction”.

          Whether the associations are causal has little bearing on whether they can be used to make “good” predictions.

  2. Challenge for Turkheimer (and those who agree with him): What (conceivable) evidence would cause you to believe that “socioeconomic educational differences are simply the result of pre-existing genetic differences among the students assigned to different schools?”

    If there is no evidence that would convince you, then your belief is religious rather than scientific.

    • D:

      Without commenting one way or another on your substantive point, I want to push back against your use of the term “religious” for a belief that is not based on good evidence. As I wrote several years ago, maybe it’s time to retire use of the term “religion” to mean “uncritical belief in something I disagree with.”

      Kind of amazing to think that post appeared nearly nine years ago; I remembered it as being much more recent.

      • Hmmm. I don’t want to use “religious” to mean “a belief that is not based on good evidence.” I want to use it to mean “a belief that no amount of (conceivable) evidence would change.”

        But I am always in favor of better communication! What word would you recommend using to denote such resistant-to-evidence views? Perhaps “ideological?”

        • D:

          “Religious” is typically associated with a connection to the supernatural, which doesn’t seem quite right here. One of the commenters at the above link suggested “dogmatic,” which seems like a better fit.

        • Isn’t a frequently used term for that “vampirical rather than empirical – unable to be killed by mere evidence”? Not that that term necessarily helps with clear communication.

    • I think that Turkheimer has a perfectly good answer that he (and others who share his perspective) have given repeatedly. We need to know the biological mechanism that causes the outcome. Then, we can see what part (if any) of the biological mechanism is associated with a set of genes. Until then, we just have associational studies that raise hypotheses and nothing more.

      • This strikes me as a bit odd.

        Darwin had zero knowledge of DNA and genetic mechanisms, yet some people believe Darwin made a significant contribution to science.

    • To me, the most important aspect of your quoted statement is not whether it is true but what would *not* follow if it were true. You may think that its truth would have substantial implications, but it wouldn’t.

      In particular, what does not follow from any study showing (in whatever way) that “socioeconomic educational differences are the result of pre-existing genetic differences among the students”:

      1. that genetic differences are the only differences that could make a difference to socioeconomic status/education

      2. that because genes are involved, the outcome is hard or impossible to change

      3. that having certain of the relevant genes is the only cause, or “the” cause, or a privileged cause, of having a certain socioeconomic status/education. This follows from the nature of organismic development, where genes always interact with many different other causes to produce a certain outcome. Neither genes nor any other of these causes can be considered a priori privileged or more “determinative”, and neither genes nore any other causes can claim temporal priority (because at no time in development do genes exist prior to other developmental causes)

      4. that a change of the relevant genes is the only, or most effective, or “simplest”, or “cleanest”, way to change socioeconomic status/education. Again, this follows from the necessary involvement of many other causes in development and the fact that other causes might well be more proximate (and therefore more specific) to the outcome, may be easier to change, may produce fewer side-effects, or all of the above.

      These points address popular misconceptions surrounding issues of genetic determinism (point 3), malleability of traits (point 2), and treatment/therapy/interventions to change traits (point 4). What *does* follow is that knowing that an observed phenotypic difference is caused by a genetic difference gives us very little useful information. Even less does it demarcate some categorical difference in causation.

      • > that genetic differences are the only differences that could make a difference to socioeconomic status/education

        Why waste our time with such an obvious straw man? No one involved in the debate believes this.

      • > What *does* follow is that knowing that an observed phenotypic difference is caused by a genetic difference gives us very little useful information.

        Who is “us” in that sentence? Genomic Predictions expects to sell polygenic score information to hundreds, and then thousands, of customers, all of whom are paying precisely because they thing “genetic difference” gives us a great deal of information about “observed phenotypic difference.”

        • Unless I’m misunderstanding polygenic scores, no, they’re not based on genetic differences giving us a great deal of information about observed phenotypic differences. They’re based on an absolute numer reflecting an individual’s genetic make-up predicting another absolute number reflecting an individual’s phenotype.

          Also, this prediction (to the extent it works) does not work *by virtue* of genetic differences causing phenotypic differences. You can see this simply by considering a case where there is no variance in genetic make-up, but the genes nevertheless cause a certain phenotype. Here you would never observe genetic differences and still a polygenic score might be highly predictive.

        • EDIT: My statement “What *does* follow is that knowing that an observed phenotypic difference is caused by a genetic difference gives us very little useful information” was too strong. If we actually know *which* genes cause a phenotypic difference, then we have a useful clue as to which genes are causally involved in creating that phenotype, and we can use these to create a polygenic score from.

          I was thinking in the framework of classical behavioral genetics, where the evidence of genetic differences comes *without* any information about which genes are involved.

  3. Just read the Bhalla piece. It is deeply unimpressive. Example:

    Presumptive causality strains the statistical sense of sexy “polygenic scoring” methods, whose fans hope “predictions can operate in complete ignorance of the biological basis.” This data-fueled folly ignores “ginger-gene” complications.

    Let’s make some predictions! Bhalla’s view suggests:

    1) Start up using polygenic scores won’t raise any money. Why would VCs invest in such obvious folly?
    2) They may raise money (VCs are stupid) but they won’t come up with a product. How could they given their “complete ignorance of the biological basis.”
    3) They may come up with a product, but no one will buy it. Why would the public be fooled when it has access to smart critics like Turkheimer and Bhalla?
    4) People may buy their products, but the products won’t work. How could they, given ginger-gene complications?

    I bet that Bhalla is wrong on all four points. Any takers?

    • Oh, come on. With regard to #4, people buy all sorts of products that don’t actually work. There are whole industries whose business model is nothing but parting fools from their money.

      • > An odd line of argument

        Perhaps, and the Theranos counter-example is well-chosen.

        But then how do you recommend discussing this topic with someone like Bhalla? My impression is that there is no conceivable counter-evidence that would cause him to change his views. (If there is, would someone specify it.) Although it is true that VCs often fund things that fail — like Theranos — and people often (?) buy things that have no (?) change of succeeding — like homeopathy, I still think that VCs and consumers provide some counter-evidence. Surely they don’t provide zero . . .

        • The utility *to the VCs* is strictly in terms of how much money they can make. If you can make money by providing homeopathy pills, they’ll fund it. For the most part they care not a whit about utility *to the end consumer*

          Sure, plenty of stuff funded by VCs makes money because it provides utility to consumers (say, flash-drives for example) but plenty of it doesn’t. That VC’s will fund it provides us information about whether they think it will make them money, not whether they think the end user gets any value.

        • I just think the actual scientific arguments seem more relevant. Especially for the audience at this blog, which is, in my estimation at least, quite open to changing beliefs given evidence.

      • Theranos collapsed because they hadn’t actually made the advances they claimed. Homeopathy is a better example, as that persists despite having no effectiveness.

  4. Since most of as subscribe to the materialistic world view, this is trivially true: since our mental states, thoughts, and everything is contingent on our physical states, then of course it is dependent on our genetics. What I don’t get is why any of this matters. Ok, so success in school is related to genetics–so what? Why does that matter? And mind you, I’m an academic, someone who has always done well in school–I just can’t understand why so much weight is put on that. Why don’t people study blacksmiths? People who make clothes? Someone who knows how to make houses? Why do we value abstract number-crunching and memorizing things so much?

    Being a good blacksmith depends on your genetics!

    This is ridiculous. That we in the first place are able to continue our little debates of no material consequence at all is dependent on the fact that other people, who know how to make the world turn, who grow crops and get it to the shops near us, who collect our garbage, who manufacture computers–yet we have the audacity to define intelligence as the thing we do, or how well we do in school. The people who make this discussion possible might’ve been horrible in school–yet here we are. And still, it seems that that is all anyone cares about. School, numbers…

    • “Being a good blacksmith depends on your genetics!”

      Actually, that is entirely possible, even likely. There are genetically determined variations in muscle cell metabolism and protein structure. It is quite plausible that some of those variations are associated with better performance at certain tasks. I’m not going to commit to any particular position about blacksmiths in particular, but the example is not the least bit ridiculous, and it might well be worthy of study.

    • Some times ago there were a post at this blog on teaching evaluation. (could some one help me to find it?)

      In this post, everybody agreed that you cant treat numerical evaluation of teachers seriously. Number alone do not give you _any_ information, you have to have written comments and textual explanation. But yet we are still willing to assign our students a single number.

    • I react similarly to the entire genetics/evolution wing of the social sciences.

      Either we observe a certain behavior in a representative portion of humankind, or we don’t. As this blog shows day after day, it’s damned hard enough even to figure that much out!

      So I don’t see why it’s interesting to add yet another layer of complexity, by theorizing that this thing we think we see in data about human behavior is “associated with” something “genetic.” As C.D-h. says, sure, I bet it is—and it isn’t. Weakly. Whatever. Of what possible use could that information be? (Only as fuel for racism and sexism among people who don’t understand the data, I fear.) Honestly, what body of useful knowledge do the academics in this field think they are building?

        • It’s odd how so many people fall down on the ground helplessly when confronted with rather straightforward implications of genetics/evolution/heritability. Some can’t bring themselves to agree with even the most obvious empirical regularities. Some suddenly require complete, absolute, and irrefutable proof of every single aspect before they can venture even the slightest agreement. Others get all tangled up in terminology and hopelessly befuddled to the point where they cannot form coherent sentences. Some claim to not see why there is even the slightest importance in any of it.

          We see all this in an academic environment where researchers often struggle mightily to unearth the tiniest and most tenuous of empirical regularities.

          Perhaps Kyle C gives the game away in the second-to-last line of his post.

    • The sense of “depends” that people care about is the degree to which (relative to our current society) we expect who your biological parents are to be predictive of your performance in school in a way that is robust against (non-genetic correlated) modifications to the environment. In other words if we give all the public schools the same resources as elite prep schools (and even throw money at kid’s home life) should we expect the next generation of great minds to still be the children of the intellectual elite in this generation or not?

      The answer to that question bears on questions of whether or not we should look harder for educational interventions and offers information about whether fancy prep schools are worth the money. It’s indirect information but it bears on many important questions.

      • And my view is, this is uninteresting from a policy perspective because of all the confounders. Until we equalize people’s life chances so that “who your parents are” only matters along one dimension, the “genetic” — and we are very, very far from such an egalitarian society — I see no point in speculating about “genetic” and “inherited” factors behind educational success, especially since (1) all evidence suggests this will turn out to be a moderate but variable effect if properly isolated and measured, and (2) throughout human history, most of our intuitions about “inherited merit” have been wrong. My VERY strong sense is that a lot of people are interested in this debate because they want reassurance that structural inequalities really don’t matter and that liberals are silly and unscientific to think they do.

  5. I didn’t see anything in Bhalla’s piece involving start-ups, so I found this comment confusing. Then, I realized that you were saying that because VCs were investing in companies that are using polygenic scores, that was some type of evidence against Bhalla’s views. But, as far as I can tell, Bhalla has said nothing that would imply any of your four points. Someone can find it perfectly reasonable to use polygenic scoring and still not believe that a question about the causal link betweeen a set of genes and a particular outcome has been established. In fact, I can hardly understand how it would be reasonable to think otherwise. Polygenic scoring like other statistical methods might be useful in looking at a set of data and generating hypotheses about the biological mechanism. Some of those hypotheses are going to yield real answers, but we will only know that when we have interventions that we can test. So, some VC may have invested in a company that uses polygenic scoring to find the next big think. That doesn’t mean polygenic scoring gives us cause and effect. It pretty clearly doesn’t.

    • some VC may have invested in a company that uses polygenic scoring to find the next big think. That doesn’t mean polygenic scoring gives us cause and effect.

      A lot depends on your demands for specificity about how much we need to know about cause and effect. Does the moon “cause” — or at least effect — the tides? Of course it does! Do you know the precise mechanism by which this occurs? No. Similar, genes effect educational outcomes. To use this information, you don’t need to know the precise mechanism.

      • Yeah, I am of the position that if we didn’t know anything about gravity, then we couldn’t say that the moon causes the tides. And, yes, we have to have a specific theory about how genes cause educational attainment or it is just an association that could be spurious., i.e., people who have tended to come from wealthier or more educated backgrounds may tend to be more closely related.

        • > I am of the position that if we didn’t know anything about gravity, then we couldn’t say that the moon causes the tides.

          Really? So, sailors for five thousand years couldn’t say that the moon causes the tides? Even though they noticed, every single month, a perfect correlation between the moon and the tides? Would you really have urged them to ignore the phase of the moon?

          What do you think they would have replied to such advice?

        • You do realize, I hope, that for almost nothing in genetics do we have a “specific theory” about how genes map onto phenotypes, but that doesn’t stop us from using genetic information (nor should it). Even for something as glaring as Down syndrome, we don’t know how, in any specific sense, an extra copy of chromosome 21 leads to the manifestations of the syndrome. We don’t, however, throw up our hands in despair, nor do we wonder if it’s all a strange coincidence that people will three copies of chromosome 21 have Down syndrome, nor do we prevent genetic testing to take place.

  6. I’m astonished to see Toby Young as a co-author. Young is a British journalist, troller, and self-publicist. He has no academic qualifications and few educationalist ones. He is a supporter of a form of eugenics and has a ideological prejudice for strongly inherited IQ – “prejudice” is the sense of belief regardless of the evidence. He is puerile and recently had to resign a high profile administrative post because of his sexist comments – whch wern’t a one-off, by the way, but part of a life-long pattern. He has been caught lying and cheating several times in his life, and what he’s achieved is largely down to powerful connections and family. I cannot imagine what he’s doing in Nature, and why the Smith-Woolley team thought he was in any way an appropriate person to be involved with, let alone grant co-author status. Young’s own account of it (at https://www.spectator.co.uk/2018/03/genetics-and-the-truth-about-private-schools/) just says he had a minor role, without saying what that was. Young is one of those figures that have sprung up recently that wallow in heir own political incorrectness and who go out of their way to offend. They turn notoriety into political promotion, with current success in both Britain and he United States. Regardless of such people’s role in politics, they should have no place in science or academia.

    • That IQ is strongly heritable is extremely well-supported. If Young believes in it “regardless of evidence”, he’s being stupid because he could hold exactly the same views while relying on truckloads of evidence. His ousting from the administrative post seems to have been due to his opposition to the pseudoscientific blank slate views that are entrenched in the educational establishment. As to Young coauthoring a study in Nature with one of the world’s leading behavior genetics teams, good for him! You’re a strange and jealous man to begrudge him for his accomplishments.

      • I suspect George Blot was imprecise in his language: some people use ‘heritable’ synonymously with ‘determined by genes.’ Many strongly heritable characteristics have little or no genetic component. The language people speak is highly heritable, as is their religion and whether they say tomayto or tomahto. Everybody realizes these things are highly heritable but that they have very little or no genetic component.

        Where things get fraught is things like educational attainment, socioeconomic status, and so on. These are highly heritable, but how much of that heritability is due to genetics? Some people are utterly convinced that most people of high attainment have better genes than most of those of low attainment — at least, better in our current environment. Others are utterly convinced that this is not the case.

        I took a hobbyist’s interest in this stuff about twenty years ago, but fairly quickly concluded that it wasn’t worth my time to read the studies. Perhaps unsurprisingly, most of the people who go into studying these issues are people who care deeply about the social implications of them…which is a kind way of saying that I thought most of the researchers have biases that can be, and I think often are, reflected in their work. I’m not talking about outright misrepresentation or deception, just that the ‘garden of forking paths’, as Andrew calls it, provides many ways to analyze any dataset. And it’s just about impossible to do an experiment — at least, an ethically acceptable one — that would get at these questions. There is some interesting work with twin studies — twins adopted into different families — but those studies have some pretty extreme limitations too.

        And finally, I don’t think it takes a “strange and jealous man” to resent what you believe to be unjustified success by someone you find despicable. I think most of us would feel that way. I certainly would. I suppose you can argue that that means most of us are ‘jealous’, and perhaps it does, but I don’t think something can be fairly characterized as ‘strange’ if it is an expected response by normal people. (By the way, I am not making any comment on Toby Young here, I don’t know him and as far as I know have never previously heard of him; but if we take George’s word for how he feels about Toby, then I don’t find George’s feelings strange at all.)

        • “some people use ‘heritable’ synonymously with ‘determined by genes.’ Many strongly heritable characteristics have little or no genetic component. The language people speak is highly heritable, as is their religion and whether they say tomayto or tomahto. Everybody realizes these things are highly heritable but that they have very little or no genetic component.”

          Your use of “Everybody” is highly optimistic.

          In any case, I’ve written about different concepts of heritability, including your point about equating “heritable” with “genetic”. Behavioral genetics is using a concept that has little to do with what heritability means in its original, biological-evolutionary context, creating endless confusion. I hope to clear up some of that here:

          https://osf.io/ea7h3/

        • Alex, you’re saying some people think your spoken language is determined by your genes?

          I just read the article you mentioned; it’s interesting and I agree it does clear up some confusion.

          For me an interesting bit of understanding came thirty years ago when I saw a talk related to dental care. I have long forgotten what the talk was about but I remember the speaker saying that when he was young there was very little relationship between people’s genes and the condition of their teeth, but now there’s a very strong relationship. Someone asked how that’s possible and the speaker said “When I was young some people brushed their teeth and some didn’t, some had fluoridated water and some didn’t, some got dental education and some didn’t, some got regular dental checkups and some didn’t. So there was a huge amount of environmental variability and it pretty much swamped the genetic variability. Now just about everyone has all of those things so there’s a lot less environmental variability, so a much higher percentage of the variability is genetic.” [That’s not a quote, but it’s the gist.] I had never thought about that before, but I think about it any time someone is interested in what fraction of the variation in something is genetic vs environmental. Even if we agreed on what the question means and how to measure it, it’s not the right question in a lot of contexts.

          And in your paper you make the point about genes and environment being intertwined to an extent that what fraction of the variation is environmental isn’t even a well-defined question.

          Other readers of this blog might want to click on your link, I recommend it.

        • Phil,

          I’m happy to hear you find the article useful.

          I haven’t heard of the dental health example but it’s a good illustration of the fact that heritability in the behavioral genetic sense is not a property of some given trait (like tooth health), but a property of the population in which heritability it is estimated. This simply follow from the fact that it’s defined in terms of variances, which will be different for different populations at different times and in different locations.

          >you’re saying some people think your spoken language is determined by your genes?

          No, I’m saying that most people, in my experience, would not consider language to be heritable, because they equate heritable with genetic.

        • That’s a very confused article. The worst confusion is perhaps your failure to understand that (narrow-sense) heritability can be equivalently thought of as either “percentage of phenotypic variance explained by genetics” or as the regression of offspring phenotype on the midparent phenotype. These are the same quantity.

          Your confusion may stem from the fact that human subject studies rarely use the latter method because parent-child comparisons in humans (as opposed to, say, in cattle) cannot disentangle genetic and environmental effects. Nevertheless, if environmental effects and epistasis are ignorable, offspring-midparent regression produces an unbiased estimate of narrow-sense heritability in (randomly mating) humans as well.

          Heritability is not something that behavior geneticists have come up with. It’s a core concept in the neo-Darwinian synthesis. Evolution cannot be understood without it.

        • M,
          I’m not sure you realize that ‘heritability’ means different things to different groups of people, even people who use the term in a way that is technically correct in their field.

          For instance, nobody disputes that people tend to be like their parents in terms of academic success; that is an empirical fact. But if you say ‘academic success is highly heritable’, to some people that is an assertion that it has a substantial genetic component; to other people it is not an assertion about genes at all.

          The name of the concept is not the concept. “Heritability” is a word that different people use for _different_ concepts. I think Alex’s article helps clarify that. Evidently you disagree, which is fine, if you didn’t even pick up on the fact that the article doesn’t take issue with the _concept_ of heritability in the Darwinian sense then clearly you didn’t understand the article…which is not necessarily your fault at all, maybe it’s just not written clearly.

        • My point is that they they purport to criticize the concept of heritability (as used in behavior genetics and evolutionary biology) without actually understanding what the concept means. They claim that “it is clear that the process by which heritability_BG estimates are arrived at does not even cast so much as a glimpse at the process of genetic transmission from parents to offspring”, when in reality heritability_BG is, given such and such model assumptions, a direct estimate of the strength of genetic transmission from parents to offspring.

        • The problems with the interpretations of heritability in the behavioral genetics literature are well known since at least Lewontin’s criticisms in the 70s. What I find confusing about the paper that Alex linked to is the very first sentence of the abstract “The concept of heritability in behavioral genetics is different from the more standard concept used in biology” — In my mind the “standard” concept of heritability in biology *is* the concept used by behavioral genetics. What Alex calls the “standard” concept in biology I would say is the “folk” concept with the added modern biological understanding that the phenomenon is due to transmission of genetic information. And, the behavioral genetics concept goes back well before behavioral genetics…it started with animal breeding. See for example the initial work of Sewell Wright in about 1920 on guinea pigs, who measured h^2. It might have been Lush a little later (1930s) in his work on animal breeding that first defined h^2 as heritability. I’ll even throw out as a working model that any use of “heritability” by a community of biologists that is not the sense developed by and coined by animal breeders is due to ignorance of this prior use, which happens a lot with many other terms in biology — language is fluid and biology is a big field. I’d be happy to see any literature that established the term “heritability” prior to the animal breeding usage of the term.

        • Hi M,

          as I try to make clear in the article, the concept of heritability get its significance from being one (of three) requirements for Darwinian evolution (these being variability in traits, differential reproductive success for the bearers of some trait variant, and heritability of traits). And it is quite easy to see that evolution by natural selection does not require any particular mechanism by which traits are “passed on” from parents to offspring. All that is really required is that offspring and parents show a higher-than-average resemblance. It doesn’t matter what makes this happen, as long as the process works reliably for multiple generations.

          Now, in the triumphant and self-serving history as told by Neo-Darwinians and geneticists, it turned out post-Darwin that it was genes, and only genes, that were responsible for the transmission of traits across generations (“DNA, the hereditary molecule!”). However, that is simply not true, not for the animal world, and even less so for the world of humans. In the human case, transmission of culture is the one big thing besides genes that supports the resemblance of parents to offspring. Cultural inheritance is meeting the heritability requirement just as well as genes do, although it does (of course) not have the exact same properties as genetic reproduction.

          The behavioral genetics concept of heritability was derived from this original, evolutionarily relevant concept in the 1930s/40s and later assumed to to be a quantitative measure of it, because, again, heritability / inheritance was seen as an exclusively gene-based process. But again, nothing in the theory of evolution requires inheritance to be genetic, and at least some people realized at some point that we should be looking out for other causal implementations of trait heritability. Somewhat ironically, it was a psychologist/philosopher who produced the earliest comprehensive formulation of a generalized concept of heritability / inheritance already in the 1980s. This was the real eye opener (or red pill, if you will) for me: Susan Oyama’s book “The Ontogeny of Information”.

          OK, I hope I managed to make it clear that the “evolutionary” concept of heritability is very very different from the behavioral genetic one. And to address one final point, the process by which classical (that’s what the article deals with) estimates of heritability are arrived at does indeed not examine the process of genetic transmission across generations in any way. I’m not sure how you could think otherwise.

        • OK, I hope I managed to make it clear that the “evolutionary” concept of heritability is very very different from the behavioral genetic one. And to address one final point, the process by which classical (that’s what the article deals with) estimates of heritability are arrived at does indeed not examine the process of genetic transmission across generations in any way. I’m not sure how you could think otherwise.

          As Jeff Walker points out above, what you call “biological heritability” is a folk notion rather than a biological one (you don’t even cite any sources for this definition), while the behavior genetic notion is the same as the biological notion. Nevertheless, the central aspect of your concept–parent-offspring resemblance due to genetics–is captured by additive heritability. Heritability is not usually estimated from parent-offspring resemblance in human studies–except sometimes in adoption studies–but this doesn’t matter because the same heritability can be identified from many kinds of relationships and even from samples of “unrelated” people.

  7. I tend to believe that genetic difference contributes substantially to cognitive capability difference. That said, Turkheimer’s criticism of the article seems largely fair. The article shows that the performance gaps between school types are mainly attributable to the types of students they have (I didn’t check the supplements but assume that the most relevant results are in the article). The evidence simply does not support the conclusion re genetics.

  8. The subtext of this nature/nurture study remains the same: As long as nurture is immaterial, government programs such as education are a waste of time and money; any intervention will ruin the Pareto optimality.

  9. Worth repeating: “association ≠ prediction”
    —>
    ‏”Too many studies report that X ‘predicts’ or ‘is predictive it’s Y when all that is demonstrated is statistically significant association. That’s too confusing when others study the predictive ability of X. Let’s call it what it is.”
    Cecile Janssens
    Epidemiologist UNC who applies “genomics to… clinical and public health practice”

    • I’d like to understand what you mean by “association ≠ prediction”.

      If X is associated with Y can’t one typically use X to make some kind (possibly weak, but better than random) of prediction for Y?

      • +1
        for me it is “(prediction == association == structure) ≠ causality”

        Maybe what was implied here is that “statistical significance ≠ association”, or, as Andrew put it, “correlation does not even imply correlation”.

        • Couple of points, misused (or abused) by Plomin

          a) Statisticians often use “predict” and “explain” differently than lay people… and it causes confusion.
          See Cosma Shalizi on over-causal phrases that mislead, item 5
          https://bigthink.com/errors-we-live-by/dumb-genes-dumber-stats-and-epistemic-comedy

          b) Typical analysis of variance practices… often mis-allocate due to non-causal non-random factors
          (can only eliminate if how the causality works is to some degree known)
          see Red Brake Risk
          http://bigthink.com/errors-we-live-by/can-cars-show-us-how-some-gene-stats-mislead

          c)Behavior & traits are often too entangled to be statistically separable.
          see Ginger Genes Scenario in a) above.

        • Hi Jag, apart from the evidence in this article do you think the monozygotic twins-reared-apart evidence is good evidence for the heritability of intelligence and several other traits? If not why not?

        • The universal stability of heritability claims is the main issue. There are no generational replications of twins to support a stability claim. The IQ phenotype effect sizes of MZT, MZA, DZA and DZT treatments, and their differences matter more than their within group correlations. A statistically `significant’ effect size is not a big deal. What matters is how big this effect size is. Flynn effect sizes does not appear to be temporally stable either.

        • George:

          Does Raj’s answer to your question satisfy you? It strikes me as evasive, but I’m not an expert on this area.

          Your question seemed pretty straightforward, but Raj doesn’t seem to have answered it. Instead, he appears to avoid the issue you raise by jumping to another issue he feels is more important. If, indeed, stability is more important as he claims, wouldn’t a responsive answer to your question be something like: “yes, the MZT evidence is good evidence for heritability of intelligence, BUT, stability is more important because …, and we don’t know the answer to that.”

        • Hi Terry, yes evasive. Also, I never said anything about statistically significant. As you say Raj could have made a “Yes, but..” response. Plomin and colleagues are confident (perhaps a little over confident) because of the evidence in support of their ideas that goes beyond this specific study.

        • Heritability is rather intrinsically tied to its stability.

          There are always questions for which the answer may not be clear at this point of time. There are many theories of causality. Physicists are happy with `entanglement` theory, and they do not accept causal theory per se. Mathematicians are happy with `non-anticipative` causal theory (fixed point, optimal control theory etc.). Casual theories such as cybernetics, top-down; bottom up, butterfly effect etc. are nothing new. We may be in search for ‘stable associations’ and we may never know the ‘true causality’ when it comes to behavioural effects.

          Most twin studies are not free from identifiability issues. We are not studying `twins of twins` over generations. Not all phenotype responses are the same. Stability in IQ phenotype is bit of a stretch because we cannot demonstrate its stability. This stability may be demonstrable for non-behavioural phenotypes say for example using plant experimental trials. The path in behavioural disciplines is rather opportunistic if we go by the arguments of Stuart Kauffman. We may not be able to find a stable set of factors for IQ determinism.

      • I guess in these crap small sample studies where authors are p-hacking the correlations they would like to see, we would have association =/= prediction. That’s just overfitting though. If you do a good ML exercise (e.g. cross validation) then it seems reasonable to say association = prediction. Coming from an economics background, the Lucas Critique would perhaps say otherwise. Basically, Lucas’ critique requires that we are still operating in a system with the same “rules” (e.g. economic policy). So we can’t use historical data to predict what will happen when we change policy if we don’t understand the underlying “structural” mechanisms (this led to economists building micro-founded macro models). I’m not sure that this is the point Jag is making. It seems like you could pretty easily predict outcomes with genes without having an understanding of the mechanisms, as long as the underlying processes that determine the outcomes (i.e. human biology) are constant over time, which presumably they are..

        • After reading Jag’s article, his example with ginger kids certainly fits the Lucas critique paradigm. As long as the “rules of the game” are unchanged (i.e. ginger kids keep getting sent to bad schools) then yes it’s true that ginger genes will predict intellectual performance successfully. It does seem like Jag is being too critical here. While these studies that attempt to control for everything environmental and then find substantial associations between genes and some outcomes may not definitively prove that genes cause the outcomes, they certainly are evidence in favour of that view. Finally, I assume many readers here adopt the potential outcomes framework to causality? That framework would seem to not require understanding the structural mechanisms. If a person with gene set A has an IQ of 100 (sps we are in deterministic world) and that same person counterfactually with gene set B has an IQ of 80, why can’t we conclude that moving from gene set B to gene set A causes higher IQ? We don’t understand the mechanism.. maybe gene set A -> interacted with the environment -> better outcomes. But we can still say A caused higher IQ, right?

        • To my last sentence I should add also that the ways genes are interacting with environment also needs to stay constant, if that’s part of the associative nexus (was going to say causal nexus, but refrained).

  10. I guess Jag Bhalla implies situations such as the absence of local control, relevant prior knowledge or the typical iid requirement that can lead to “association ≠ prediction”. Simpson’s paradox is another scenario.

  11. While the article you link at the end is interesting it plays into the unfortunate dynamic that we see all too often in this subject. The side arguing against IQ heritability plays dirty by changing the discussion from the normal scientific question (given theory, priors, direct and INDIRECT evidence) which account is overall more probable. Instead, the question is reframed as “Does the side arguing for this undesirable view have clinching evidence that proves their view.” In effect the attitude is to convince the reader that they lack the expertise, and even the experts lack the clinching evidence, that would excuse seriously considering the possibility that IQ is strongly heritable, there is a strong racial relationship in that heredity etc.. (I don’t personally find the later convincing and have mixed feelings about the former but I they should stand or fall on their merits).

    A fair attempt to judge the strength of arguments supporting more hereditarian positions on IQ (or similar controversial points) would start with (formally or informally) some kind of discussion about what our priors should be, e.g., how hereditary are relevantly similar properties in other mammals, what does theoretical work on natural selection suggest etc..

    I know I’m being a bit harsh on this author since it’s not really his fault given the lack of true experts willing to dive into the issue and evenhandedly consider the best arguments for both sides in science popularization forums.

Leave a Reply to Steve Cancel reply

Your email address will not be published. Required fields are marked *