I'll have to read this later since I'm at work but is there a layman's explanation for how these studies distinguish environmental and genetic influence? I don't see how you could accurately separate the two.
Not entirely sure what you mean: genes cause environments, and interact with environments.
If you mean something like whether it's just picking up ancestry markers for families which happen to be rich etc, there are a lot of methodological components which try to minimize the more obvious ways of population stratification affecting results.
The most iron-clad part of showing direct gene->trait causation is the 4 sibling-comparison studies, where you see whether you can predict the difference between 2 siblings; as they have identical ancestries, families, locations etc, their environments are near-identical in the first place, and since meiosis at conception randomizes genes between siblings, success in sibling comparison shows the genes must be causal. IQ GWASes have always passed the within-family test since Rietveld et al 2013, and this is no exception.
> The most iron-clad part of showing direct gene->trait causation is the 4
I would not say it's iron-clad gene->trait causation. Even wth 4 sibling-comparisons and the same environmental factors, it can be gene-environment interaction -> trait
Their study finds that genes associated with attainment are active in neurons. These genes might just protect brain from environmental damage. For example: from malnutrition by using nutrients more effectively, infections, heavy metals, or mother smoking or using alcohol, ...
>If you mean something like whether it's just picking up ancestry markers for families which happen to be rich etc, there are a lot of methodological components which try to minimize the more obvious ways of population stratification affecting results.
Yeah this is actually what I'm interested in.
>The most iron-clad part of showing direct gene->trait causation is the 4 sibling-comparison studies, where you see whether you can predict the difference between 2 siblings; as they have identical ancestries, families, locations etc, their environments are near-identical in the first place, and since meiosis at conception randomizes genes between siblings, success in sibling comparison shows the genes must be causal. IQ GWASes have always passed the within-family test since Rietveld et al 2013, and this is no exception.
Couldn't this still lead to an overstating of the effect? So two brothers John and Jim have different builds due to genetics, John is a bit shorter and stockier and Jim is a bit taller and leaner. When they play with each other as children John usually wins when they wrestle and Jim usually wins when they play basketball. They both like winning so in highschool John joins the wrestling team and Jim joins the basketball team. Now if I do some gene testing wizardry on them as adults and try to determine the difference in basketball-skill I'll predict a large gap due to height genes and I'll get one. But part of that skill gap is due to environment (4 years of basketball practice) yet still getting attributed to genes.
> ...When they play with each other as children John usually wins when they wrestle and Jim usually wins when they play basketball...
No, this is still genetic causation (and something like this story is usually what is invoked for why the heritability of intelligence increases with age: more intelligent people seeking out niches which then encourage intelligence further rather than preferring to veg out in front of the boobtube, leading shared-environment to disappear while the permanent ongoing influence from genetics leads to heritability increases).
What's the definition of causation? It's a 'difference which makes a difference', or in Pearlean terms, if you reach in and do surgery on one node in a causal network, the counterfactual difference.
So in your basketball example, what if we reached in with CRISPR, say, and edited just the genes for builds in John to make him taller/leaner? What is the counterfactual of our intervention on the node of genes? As he grows up and becomes taller/leaner, now he still likes winning and playing with his brother is an even challenge, so he joins the basketball team along with Jim and they do well and keep playing and so on and so forth and your basketball PGS predicts less or no difference between them (since we caused them to have the same relevant genes) and indeed there is less or no difference; and if we had instead edited the other way, it'd go the other way etc. Thus, causation. That the causal pathway of the genes goes 'outside the skin', as some people put it, makes no difference that makes a difference for this particular question.
(A similar point applies to the 'nature of nurture' findings mentioned in Lee et al 2018 and the two earlier papers. Are they genetic causation? Yes, because if you counterfactually changed the genes and nothing but the genes, you would get different outcomes. They are differences which make a difference. They are, however, causal on things you didn't expect - eg edits in children would cause gains in grandchildren rather than the children.)
If what we're interested in is genetic causation then the relevant comparison is to John if he was still shorter but joined the basketball team anyways. In that case the variance between their skills would be less and yet no genes were changed. That illustrates that the original difference wasn't due to genetic but rather environmental factors.
>That the causal pathway of the genes goes 'outside the skin', as some people put it, makes no difference that makes a difference for this particular question.
When genes cause changes 'outside the skin' the way those outside changes affect your skill are environmental changes by definition. For example you're saying that in our current society, where you can choose your extracurricular activity, the difference in their basketball-skill that's attributable to genetics is x. But in a different society where high school basketball practice was state mandated the difference in their basketball-skill that's attributable to genetics would be y where y < x. The difference between x and y can't be caused by genetics because we haven't changed any genes so it shows that we were just misestimating genetic causation to begin with.
You are using a definition of genetic causation that is not used by anyone in the field of genetics. It sounds like your definition is "Is there any hypothetical environment where the phenotypic expression of a gene could be changed?" The answer is almost always yes.
Growth hormone influences height, imagine a world where everyone's level of growth hormone is regulated. Suddenly all of these growth hormone influencing genes no longer have any effect on height. By your definition growth hormone does not genetically cause height because of this hypothetical. You could see how using a definition is both incredibly restrictive as well as almost impossible to measure, and therefore not a very useful definition.
May that be because this isn't field of genetics at all, but unrelated one called "behavioural genetics" (using quantitative methods of the social sciences, with similar levels of signal, and none of the certainty of actual genetics) wherein Turkheimer formulated the so called "three laws" not unaligned with GP's point?
>You are using a definition of genetic causation that is not used by anyone in the field of genetics.
That's worrying.
>It sounds like your definition is "Is there any hypothetical environment where the phenotypic expression of a gene could be changed?" The answer is almost always yes.
Sure but we don't even need a hypothetical environment where high school basketball practice is state-mandated, we just need any hypothetical environment where John and Jim both go to basketball practice and their genes stay the same to see that what is labeled as genetic causation is actually environmental. I mean are you telling me the entire field has no way to distinguish a skill->environment->skill feedback effect from a genes->skill effect?
I'd say that genes seem essentialist, they're something that defines you and can't be changed but environment isn't that way, it's malleable and can be changed. So take this example, we're gonna say at least 11% of IQ is genetic, right? But if we pass some laws so that we have better teachers and smaller class sizes and better remedial programs then what IQ is only 8% genetic? Or even in the case that the laws help the top more and IQ expands to 14% genetic. In that case it doesn't seem that our measurements of genetics effects were accurately measuring the genetics at all, just environmental effects that are hard to account for.
>You could see how using a definition is both incredibly restrictive as well as almost impossible to measure, and therefore not a very useful definition.
Yes, I agree that what I'm saying would make genetic causation very difficult to measure. But it's also the truth about what genetic causation is. Instead we're just opting for what is easy to measure and labeling it as the thing that is hard to measure. That's not right.
Can you describe how you would even hypothetically measure your definition of "genetic causation"? Could you give an example of a "gene causing" the expression of an interesting trait in a way that is immune to environmental changes?
You're looking for a gene that causes an outcome completely independently of all environments. I don't think this exists for any interesting traits. Not to mention how much effect a change has on a given trait is dependent on the genetic make up of the population.
> But it's also the truth about what genetic causation is.
No it's not, everyone who talks about genetic causation seriously use a different definition. I don't know why your own personal definition is better than the definition used by every single person who studies genetics.
I would ditch the idea of widespread genetic correlations on complex traits, switch to a method that describes what genes (and their interactions) actually do to the human body. For example, this specific gene does x to a neurotransmitter which can interact with y or z to cause changes like this or that. (And I know it's actually more complicated than just a straight mapping of gene X -> trait Y.)
With my method you can still say that no matter the environment gene x does this thing which can interact with gene y or gene z or environmental factors a, b, c in certain ways.
I think that proving intelligence is XX% genetic is basically meaningless since we can't actually remove environmental effects and if I'm really honest I think it's popularity in the culture is usually driven by spurious motives.
>No it's not, everyone who talks about genetic causation seriously use a different definition. I don't know why your own personal definition is better than the definition used by every single person who studies genetics.
Why do you think this is at the top of Hacker News? Because people's common understanding of a title like:
New study shows that IQ is 11% genetic
Is that at least 11% of their intelligence comes straight from their DNA, not that actually we could change 11% to 5% or 8% or 17% or most any number we want just by changing some laws or fortifying food or any number of environmental changes. The popularity of genetics research in popular media requires this bait-and-switch about what is actually being (mis) measured.
> I would ditch the idea of widespread genetic correlations on complex traits, switch to a method that describes what genes (and their interactions) actually do to the human body. For example, this specific gene does x to a neurotransmitter which can interact with y or z to cause changes like this or that. (And I know it's actually more complicated than just a straight mapping of gene X -> trait Y.)
> With my method you can still say that no matter the environment gene x does this thing which can interact with gene y or gene z or environmental factors a, b, c in certain ways.
If you arguing that it is better to understand all of the casual pathways a gene influences a trait than to not understand that, then I strongly agree. If you are arguing that understanding which genes casually influence which complex traits is useless until we know all of the casual pathways that gene affect a trait then I'll have to disagree just as strongly.
> I think that proving intelligence is XX% genetic is basically meaningless since we can't actually remove environmental effects and if I'm really honest I think it's popularity in the culture is usually driven by spurious motives.
First proving what % of intelligence is heritable is not the purpose of GWAS's. Twin studies are used for this, and they've found that intelligence is highly heritable.(far higher than 11%)
Ways that GWAS's are useful without removing environment effects.
- Embryo selection
- Tests to figure out risk for certain diseases for preventative purposes
- Used to figure out what are important pathways in complex traits/disease. For instance if everyone with a certain defective enzyme has a 50% reduction in heart attacks this could lead to a drug candidate
> New study shows that IQ is 11% genetic
> Is that at least 11% of their intelligence comes straight from their DNA, not that actually we could change 11% to 5% or 8% or 17% or most any number we want just by changing some laws or fortifying food or any number of environmental changes. The popularity of genetics research in popular media requires this bait-and-switch about what is actually going on.
If you are arguing that some people(maybe even a lot) misunderstand the definition of genetic causation in a similar way that you previously did, then that might be true. But when any experts, or any study talks about genetic causation they are using the definition the rest of us are using on the this thread.
>First proving what % of intelligence is heritable is not the purpose of GWAS's.
And yet here we are on a thread about using a GWAS to prove what % of intelligence is heritable.
>If you are arguing that some people(maybe even a lot) misunderstand the definition of genetic causation in a similar way that you previously did, then that might be true.
My understanding isn't past tense. Here, allow me to illustrate why this makes no sense to me. Let's take what Gwern said above:
Are they genetic causation? Yes, because if you counterfactually changed the genes and nothing but the genes, you would get different outcomes. They are differences which make a difference.
But in my basketball example I could also say this:
Are they environmental causation? Yes, because if you counterfactually changed the environment and nothing but the environment, you would get different outcomes. They are differences which make a difference.
So we are going to take factors where both of those statements are true and still assert that they are genetically causal? That's incorrect.
>But when any experts, or any study talks about genetic causation they are using the definition the rest of us are using on the this thread.
The 'rest of us' in this thread are definitely not using the same language you use. Let's take a look at just the post author's comments:
---
>The most iron-clad part of showing direct gene->trait causation is the 4 sibling-comparison studies
except it's not iron-clad at all and it doesn't show direct gene->trait causation.
>where you see whether you can predict the difference between 2 siblings; as they have identical ancestries, families, locations etc, their environments are near-identical in the first place
Their environments are not near-identical.
>No, it doesn't. It merely tells us that with the data up to now, we've reached 11% (up from, incidentally, ~0.3% just 5 years ago). It provides a very loose lower bound on genes.
Except you're telling me that not only is a GWAS not useful for this purpose but that even this number shouldn't be treated as a 'lower bound' but as a potential identifier to further study the genes identified. It would seem that this
>> Observational studies of this type cannot show causality
>They can: sibling comparisons.
Again, this quote appears to conflict with what you believe GWASs are capable of.
---
Again, why are these falsehoods being spread around? Because of an (mis)understanding that 'genetic causation' as measured in the paper is not manipulatable by the environment. Except that it actually is. And a misunderstanding of what GWASs are good for.
>- Embryo selection
No it's not. GWASs don't understand what they're measuring and label the environmental as genetic biasing their measurements. Embryo selection is done the old-fashioned way looking for specific genetic variants that we understand through the use of other types of studies.
>- Tests to figure out risk for certain diseases for preventative purposes
Look at the use of GWASs in identifying schizophrenia. They typically fail to replicate. GWASs find genetic connections where none exist.
Take schizophrenia. It could be that we run a GWAS on schizophrenia and find some genetic similarities. Then we look at these similarities and find one of them is a cuddle gene. A cuddle gene what the heck? Well the cuddle gene makes you want to cuddle which makes their mothers more likely to own pets to cuddle which makes them more likely to own cats which increases their risk of exposure to toxoplasmosis which increases their of schizophrenia. Is that a genetic causation? Because the way genetic causation is actually being used in this thread really doesn't lead to that conclusion. It leads to the idea that the identified genes make you a little more unstable in a certain way, not that we could totally eliminate that element o...
Relevant, from the article FAQ:
"A recent study of Icelandic families showed that the parental allele that is not passed on to the parent’s offspring is still associated with the child’s educational attainment, suggesting that GWAS results for educational attainment partly represent these intergenerational pathways (Kong et al. 2018). Our sibling analyses yield results that are consistent with this conclusion (see FAQ 2.4)."
So, it might just be both environment and the genetic influence both correlating to the same thing, and they admit as much.
You do a twin study comparing monozgyotic to dizygotic twin pairs. The variance of the trait between the two pairs is the genetic effect. Determining environment effect is done by doing twin-reared-apart studies (to separate shared vs non-shared environment).
taking all of the associations into account and making a polygenic predictor score "does a lousy job predicting the outcome for any specific individual, but it can explain 11 percent of the population-wide variation in years of schooling."
only 11 percent. this tells us that the remainder is more likely to be attributable to environmental factors.
in other words, even with a view of the genes, we would need to encourage education more strenuously if we wanted to improve educational outcomes for society.
> only 11 percent. this tells us that the remainder is more likely to be attributable to environmental factors.
No, it doesn't. It merely tells us that with the data up to now, we've reached 11% (up from, incidentally, ~0.3% just 5 years ago). It provides a very loose lower bound on genes. Saying that is like saying, 'our software testing up to now has only revealed bugs in 11 places, therefore, every other line of code must be bug-free!' Even if you knew nothing about the actual heritability estimates, one would think that finding so much so quickly would make one suspect that there's a lot more to find...
there's always more to find, but consider that this model is leaving out the environment entirely and its predictive value is even less than you might think at first glance.
we know the environment impacts gene transcription, and we know that genetic determinism (as far as outcome phenotypes) is not a thing. likewise, there may be epigenetic factors at play which wouldn't be captured in a GWAS yet still might (or might not) have an impact on outcomes.
it isn't a "the other lines of code are bug free" situation so much as a "we expect there to be an abundance of bugs with undefined consequences outside of the subset which we have examined which most likely outweigh the stuff we found"
I wouldn't write off genetics. GWASes, as a technique, are still pretty new and the number of genomes available for GWASes is still relatively small. Plus, twin studies point to traits like educational attainment being >11% heritable.
GWAS are great, don't get me wrong. but there's always a lot of jumping to conclusions that is done whenever a GWAS finds something interesting. afterwards, you find the people wishing that they could CRISPR themselves to fit the pattern found in the GWAS, etc.
you can see that in the article the authors of the study specifically went out of their way to try to caution people against doing that.
Full disclosure: I have attended multiple presentation on this paper by some of these authors, and I'm doing a PhD in quantitative Genetics. So I like to think I have something to add. But I'm not an expert on educational attainment, so correct me if I'm wrong.
The 11% of variance that can be predicted using the current information of a fairly bad genetic predictor. The predictor is bad because many cohorts used in this study do not share all their information, but only share so called summary statistics. Which are combined into a prediction model. If all the individual level data is available, it becomes possible to get a much better prediction model.
In addition there are ways to estimate how much can be predicted from genetic information, a term called heritability.
Heritability can be estimated from so called twin studies. Where twins who are genetically identical (monozygotic) are compared to twins who are not genetically identical (dizygotic). In this way you can measure the environmental effect (the variation seen in the monozygotic twins) and the genetic effect (the variation that is seen in the dizygotic twins compared to the monozygotic twins). This is, of course, a gross oversimplification, but can provide reasonable measures of an upper bound of what can be explained genetically
For most traits this upper bound is around 0.5 R^2, and I believe for educational attainment a bit higher, but I could not find it in the paper or a quick google search.
Another factor in this case is that there are unmeasured genotypes which can explain some extra variation. There are also solutions for this. Which are able to explain the upper bound of what can be predicted from the data at hand. For instance meaning that you can find 80% of all the genetic variation using the genetic measurements that you have.
But you often need individual level data for this. Something which cohorts are not often prepared to share.
Another complicating factor is the heterogeneous nature of the phenotype: educational attainment is different per country. So it's hard to say what's really being measured. It's likely very different in France, compared to say The Netherlands. Due to the differing nature of their educational systems.
Personally I'm morally conflicted by these studies which measure intelligence.
It's very interesting to see the practical outcomes from the paper. I've seen an example where it's well powered to stratify high educational attainment and low educational attainment individuals for social science studies. But I'm of the opinion that this may be a slippery slope towards misinterpretation, as we currently see in the whole IQ business. As well as too strong an emphasis on being 'smart' in certain social groups.
We know Intelligence (or whatever is being measured) is heritable. But I'm not fond of the implications which people attach to these studies. Even though I know people are going to act on it.
For the earliest examples of this, I'm going to look at certain east Asian countries where ethics may be very different compared to the western views.
> In addition there are ways to estimate how much can be predicted from genetic information, a term called heritability. Heritability can be estimated from so called twin studies. Where twins who are genetically identical (monozygotic) are compared to twins who are not genetically identical (dizygotic). In this way you can measure the environmental effect (the variation seen in the monozygotic twins) and the genetic effect (the variation that is seen in the dizygotic twins compared to the monozygotic twins). This is, of course, a gross oversimplification, but can provide reasonable measures of an upper bound of what can be explained genetically For most traits this upper bound is around 0.5 R^2, and I believe for educational attainment a bit higher, but I could not find it in the paper or a quick google search.
Just wondering, have you ever encountered any of the serious critiques of twin studies.
The comparison of dizygotic (DZ) and monozygotic (MZ) twins relies on the Equal Environment Assumption (EEA) - which assumes that the environment experienced by MZ twins is the same as that of DZ twins. This assumption has been shown empirically to be false - parents, siblings, teachers, friends, and the twins themselves treat MZ twins very differently than they treat DZ twins.
There are good reasons to think that heritability results derived from twin studies are highly inflated.
> However, theoretical projections that failed to consider heterogeneity of effect sizes were optimistic Our and others’ findings suggest that imperfect genetic correlation across cohorts will be the norm for phenotypes, such as educational attainment, that are environmentally contingent.
So it is somewhat worrying that genetics predict the clearly environmentally contingent like educational attainment better than "cognitive performance."
> Our results also highlight two caveats to the use of the polygenic scores in research. First, our within-family analyses suggest that GWAS estimates may overstate the causal effect sizes: if educational attainment-increasing genotypes are associated with parental educational attainment-increasing genotypes, which are in turn associated with rearing environments that promote educational attainment, then failure to control for rearing environment will bias GWAS estimates.
That is a very complicated way of saying that some genes express themselves by black skin, and these genes have a strong effect on funds available to the school.
> So it is somewhat worrying that genetics predict the clearly environmentally contingent like educational attainment better than "cognitive performance."
I don't agree that schools are any more 'clearly environmentally contingent' than intelligence, especially with the mandatory public schooling applicable to all of the cohorts used.
> That is a very complicated way of saying that some genes express themselves by black skin
That's a remarkable assertion, considering that this dataset is made of 100% white people, works on white siblings like Swedish twins, and the PGS is still predictive when applied within a separate African-American dataset.
> I don't agree that schools are any more 'clearly environmentally contingent' than intelligence,
Clearly I can change all the rules governing schools without changing the genetics of any of the students. So you are claiming that it is possible that if intelligence could be as environmentally contingent like school system, then one would need to show that this type of study can produce interesting results in the first place.
> That's a remarkable assertion,
I read the sentence as a general qualifier for this kind of studies. But the objection is, that these genes would produce a clear signal in this kind of studies without any of the causality taking place inside of a brain. (And unlike 'intelligence,' I can actually define what I mean by 'black skin' 'school funding' and 'average neighborhood income.' )
So for the study to have any kind of explainative effect, it needs to show that the effect is created by anything that can be reasonably called intelligence, instead of something that takes place completely outside of the brain.
It would be remarkable if there was not some predictive value when applied to AA datasets given the level of European ancestory in the AA population is around 30%.
(Closer to 20%, and the attenuation of 85% is about what you would expect; we already know that that's the usual level of decrease when you test PGSes in AAs, and it goes to near-zero in African populations. It looks like it's not an issue of entirely different causal variants or different environments, but mostly LD tagging decay.)
It would be remarkable that someone would fall on their sword like that today.
It'll be another couple of years before "The Flippening", i.e. scientists looking to cement their legacy doing the political calculation and deciding they'd rather risk some short term fallout vs joining Lyskenko and Lamarck in the history book.
That elaborate kabuki we saw from David Reich at Harvard a few months ago? It'll get even more entertaining in the coming months and years as that fig leaf continues to shrink.
My interpretation of this FAQ is that it provides a great deal of context that would be apparent to people in the field (and therefore not included in the paper itself) but is critical background for anyone else who is trying to make sense of it.
It therefore provides a bridge between the highly technical paper itself and journalism which too often misunderstands the true meaning of research results.
I think this is so valuable. I hope that this kind of thing will catch on for more research that attracts public interest.
I think the FAQ has a strong self-defence purpose. See the last question covered, which is probably the first one people care about:
> 3.7. Could this kind of research lead to discrimination against, or stigmatization of, people with the relevant genetic variants? If so, why conduct this research?
I read the entire FAQ. It's gold. Every major paper that makes strong claims should be required to write as much explaining. For example, after reading the FAQ, I understand now that most GWAS papers overstate everything about their findings. The FAQ is much more measured and explicates the severe restrictions on interpreting the findings.
Interesting discussion of genes and pathways involved around page 84.
"Results: Causal Genes and Enriched Gene Sets"
I'll just type in the pathways. ( Supplementary Figure 22.) "Brain Specific Expression of Significantly Enriched Gene Sets across Development. (page 196)":
Chromatic modification, protein binding transcription factor activity, npBAF complex, Central nervous system neuron differentiation, forebrain development, partial posttnatal lethality, abnormal cerebral cortext morphology, endometrial cancer , GAB1 signalosome, regulation of nervous system development, telencephaol cell migration, protein tyrosine kinase activity , neuron recognition, axon guidance, small cerebellum protein phosphatase regulator activity, signaling by eGFR, dendrite morphogenesis, signaling by NGF, regulation of neuron projection development, behavioral fear response, DAG and IP signaling, associative learning , voltage-gate calcium channel activity , ataxia, extracellular-glutamte-gate ion channel activity, post NMDA receptor activation events, regulation of neurotransmitter levels, regulation of synaptic transmission, synapse part ,gate channel activity.
Single Nucleotide Polymorphisms (snps) discussed in supplement are:rs10189857 rs1167898 rs11678980 rs18235539 rs182355396 rs186456786 rs186456786 rs186456786 rs5951458 rs61734410
It is worth noting that these SNPs aren't causal, they are just associated.
The resolution of GWAS isn't at the level yet where we can pick out individual changes that cause a phenotype. Instead we can detect that a chunk of the genome is associated with a phenotype. These snps are like a street name, they identify the region associated.
We need whole-genome sequence on a million-person scale to actually find the causal variant. That will come with time.
> Observational studies of this type cannot show causality
They can: sibling comparisons.
> the results have not been confirmed by independent methods.
They're tested out of sample many times (4 for sibling comparisons alone!), and you can find many other papers successfully using the IQ GWASes. Belsky et al 2018 just came out a few weeks ago, for a particularly good example testing OP's PGS. (The PGS has been out since like March, so several papers have used it.)
> “If you did a study like ours 100 years ago, the strongest genetic predictor of education would be how many X chromosomes you had, because society was set up in a way that it was much harder for women to get educated than men,” says Benjamin. Likewise, many of the genes that are associated with education today are likely important “because of how today’s educational system is set up. It requires people to sit at desks for hours, and listen to instructions from a teacher. People who get restless, or are less obedient to authority, will fare less well in that environment.”
> The current study further confirmed the finding from our earlier work that the effects of individual genetic variants on educational attainment are extremely small. The average effect size across the 1,271 genetic variants was just 1.8 weeks of schooling per allele; even the SNPs with the strongest associations only predicted around 3 weeks of additional schooling per allele. Taken together, these 1,271 SNPs accounted for just 3.9% of the variation across individuals in years of education completed.
> Here is another way to think about this result. Imagine that we used the results for these 1,271 genetic variants (not the ~1 million SNPs across entire genome we discuss in FAQ 2.3) to predict the educational attainment for a new group of people (separate from our discovery sample). We could then compare each individual’s predicted educational attainment to their actual educational attainment. If we did so, our results suggest that we would find that the predictions and actual outcomes correlate only very modestly (at about r = 0.20). That, in turn, means that if someone were predicted to complete an above average number of years of schooling (i.e., to be in the top half of educational attainment), that person would have about a 58% chance of actually being in the top half of educational attainment. Fifty-eight percent is better than chance (i.e., 50%), suggesting that a prediction based on these 1,271 SNPs has more power to predict educational attainment than a coin flip—but only a bit more power. By contrast, a prediction based on a polygenic score that combines ~1 million SNPs that we studied (see FAQs 1.5 & 2.3) has more predictive power: r = 0.33, corresponding to 11% of the variation across individuals.
> The contrast between the 3.9% of the variation predicted by the 1,271 SNPs and the 20% known to be explained by common SNPs (see FAQ 1.7) implies that there are many other SNPs that have not yet been identified. Even larger sample sizes will be needed to identify them.
They only included people of European ancestry to eliminate sources of noise and cofounding factors. The effect size is very small, yet it does exist.
This seems to prove that intelligence influenced by genetics but is affected by many thousands of genetic variations. There is no "smart" gene, just lots and lots of randomness.
This makes sense. People use different combinations of skills and physical traits to excel in arts, sports, and business. It seems natural then that “intelligence” would encompass a wide range of abilities and traits.
>This seems to prove that intelligence influenced by genetics but is affected by many thousands of genetic variations. There is no "smart" gene, just lots and lots of randomness.
Not quite how I would put it. There is no one smart gene, just thousands and thousands of smart alleles.
That's as much as Turkheimer's three laws of behavioural genetics are saying.
Contrary to how this field styles itself it is a quantitative social science having little to do with actual genetics, using next to none of its methods but a lot of quantitative social science's statistics with levels of signal similar to social science.
>This seems to prove that intelligence influenced by genetics
Only if the educational attainment difference predicted by these genes is because of "intelligence." It could just as easily have to do with "stick-with-it-ness" or "I-have-parents-that-insist-I-go-to-college" or "I-do-what-is-expected-of-me."
71 comments
[ 3.3 ms ] story [ 146 ms ] threadFAQ: https://www.thessgac.org/faqs
After high correlates are found, further experiments will disentangle the relationship.
If you mean something like whether it's just picking up ancestry markers for families which happen to be rich etc, there are a lot of methodological components which try to minimize the more obvious ways of population stratification affecting results.
The most iron-clad part of showing direct gene->trait causation is the 4 sibling-comparison studies, where you see whether you can predict the difference between 2 siblings; as they have identical ancestries, families, locations etc, their environments are near-identical in the first place, and since meiosis at conception randomizes genes between siblings, success in sibling comparison shows the genes must be causal. IQ GWASes have always passed the within-family test since Rietveld et al 2013, and this is no exception.
I would not say it's iron-clad gene->trait causation. Even wth 4 sibling-comparisons and the same environmental factors, it can be gene-environment interaction -> trait
Their study finds that genes associated with attainment are active in neurons. These genes might just protect brain from environmental damage. For example: from malnutrition by using nutrients more effectively, infections, heavy metals, or mother smoking or using alcohol, ...
Yeah this is actually what I'm interested in.
>The most iron-clad part of showing direct gene->trait causation is the 4 sibling-comparison studies, where you see whether you can predict the difference between 2 siblings; as they have identical ancestries, families, locations etc, their environments are near-identical in the first place, and since meiosis at conception randomizes genes between siblings, success in sibling comparison shows the genes must be causal. IQ GWASes have always passed the within-family test since Rietveld et al 2013, and this is no exception.
Couldn't this still lead to an overstating of the effect? So two brothers John and Jim have different builds due to genetics, John is a bit shorter and stockier and Jim is a bit taller and leaner. When they play with each other as children John usually wins when they wrestle and Jim usually wins when they play basketball. They both like winning so in highschool John joins the wrestling team and Jim joins the basketball team. Now if I do some gene testing wizardry on them as adults and try to determine the difference in basketball-skill I'll predict a large gap due to height genes and I'll get one. But part of that skill gap is due to environment (4 years of basketball practice) yet still getting attributed to genes.
No, this is still genetic causation (and something like this story is usually what is invoked for why the heritability of intelligence increases with age: more intelligent people seeking out niches which then encourage intelligence further rather than preferring to veg out in front of the boobtube, leading shared-environment to disappear while the permanent ongoing influence from genetics leads to heritability increases).
What's the definition of causation? It's a 'difference which makes a difference', or in Pearlean terms, if you reach in and do surgery on one node in a causal network, the counterfactual difference.
So in your basketball example, what if we reached in with CRISPR, say, and edited just the genes for builds in John to make him taller/leaner? What is the counterfactual of our intervention on the node of genes? As he grows up and becomes taller/leaner, now he still likes winning and playing with his brother is an even challenge, so he joins the basketball team along with Jim and they do well and keep playing and so on and so forth and your basketball PGS predicts less or no difference between them (since we caused them to have the same relevant genes) and indeed there is less or no difference; and if we had instead edited the other way, it'd go the other way etc. Thus, causation. That the causal pathway of the genes goes 'outside the skin', as some people put it, makes no difference that makes a difference for this particular question.
(A similar point applies to the 'nature of nurture' findings mentioned in Lee et al 2018 and the two earlier papers. Are they genetic causation? Yes, because if you counterfactually changed the genes and nothing but the genes, you would get different outcomes. They are differences which make a difference. They are, however, causal on things you didn't expect - eg edits in children would cause gains in grandchildren rather than the children.)
>That the causal pathway of the genes goes 'outside the skin', as some people put it, makes no difference that makes a difference for this particular question.
When genes cause changes 'outside the skin' the way those outside changes affect your skill are environmental changes by definition. For example you're saying that in our current society, where you can choose your extracurricular activity, the difference in their basketball-skill that's attributable to genetics is x. But in a different society where high school basketball practice was state mandated the difference in their basketball-skill that's attributable to genetics would be y where y < x. The difference between x and y can't be caused by genetics because we haven't changed any genes so it shows that we were just misestimating genetic causation to begin with.
Growth hormone influences height, imagine a world where everyone's level of growth hormone is regulated. Suddenly all of these growth hormone influencing genes no longer have any effect on height. By your definition growth hormone does not genetically cause height because of this hypothetical. You could see how using a definition is both incredibly restrictive as well as almost impossible to measure, and therefore not a very useful definition.
That's worrying.
>It sounds like your definition is "Is there any hypothetical environment where the phenotypic expression of a gene could be changed?" The answer is almost always yes.
Sure but we don't even need a hypothetical environment where high school basketball practice is state-mandated, we just need any hypothetical environment where John and Jim both go to basketball practice and their genes stay the same to see that what is labeled as genetic causation is actually environmental. I mean are you telling me the entire field has no way to distinguish a skill->environment->skill feedback effect from a genes->skill effect?
I'd say that genes seem essentialist, they're something that defines you and can't be changed but environment isn't that way, it's malleable and can be changed. So take this example, we're gonna say at least 11% of IQ is genetic, right? But if we pass some laws so that we have better teachers and smaller class sizes and better remedial programs then what IQ is only 8% genetic? Or even in the case that the laws help the top more and IQ expands to 14% genetic. In that case it doesn't seem that our measurements of genetics effects were accurately measuring the genetics at all, just environmental effects that are hard to account for.
>You could see how using a definition is both incredibly restrictive as well as almost impossible to measure, and therefore not a very useful definition.
Yes, I agree that what I'm saying would make genetic causation very difficult to measure. But it's also the truth about what genetic causation is. Instead we're just opting for what is easy to measure and labeling it as the thing that is hard to measure. That's not right.
Can you describe how you would even hypothetically measure your definition of "genetic causation"? Could you give an example of a "gene causing" the expression of an interesting trait in a way that is immune to environmental changes?
You're looking for a gene that causes an outcome completely independently of all environments. I don't think this exists for any interesting traits. Not to mention how much effect a change has on a given trait is dependent on the genetic make up of the population.
> But it's also the truth about what genetic causation is. No it's not, everyone who talks about genetic causation seriously use a different definition. I don't know why your own personal definition is better than the definition used by every single person who studies genetics.
With my method you can still say that no matter the environment gene x does this thing which can interact with gene y or gene z or environmental factors a, b, c in certain ways.
I think that proving intelligence is XX% genetic is basically meaningless since we can't actually remove environmental effects and if I'm really honest I think it's popularity in the culture is usually driven by spurious motives.
>No it's not, everyone who talks about genetic causation seriously use a different definition. I don't know why your own personal definition is better than the definition used by every single person who studies genetics.
Why do you think this is at the top of Hacker News? Because people's common understanding of a title like:
New study shows that IQ is 11% genetic
Is that at least 11% of their intelligence comes straight from their DNA, not that actually we could change 11% to 5% or 8% or 17% or most any number we want just by changing some laws or fortifying food or any number of environmental changes. The popularity of genetics research in popular media requires this bait-and-switch about what is actually being (mis) measured.
> With my method you can still say that no matter the environment gene x does this thing which can interact with gene y or gene z or environmental factors a, b, c in certain ways.
If you arguing that it is better to understand all of the casual pathways a gene influences a trait than to not understand that, then I strongly agree. If you are arguing that understanding which genes casually influence which complex traits is useless until we know all of the casual pathways that gene affect a trait then I'll have to disagree just as strongly.
> I think that proving intelligence is XX% genetic is basically meaningless since we can't actually remove environmental effects and if I'm really honest I think it's popularity in the culture is usually driven by spurious motives.
First proving what % of intelligence is heritable is not the purpose of GWAS's. Twin studies are used for this, and they've found that intelligence is highly heritable.(far higher than 11%)
Ways that GWAS's are useful without removing environment effects.
- Embryo selection
- Tests to figure out risk for certain diseases for preventative purposes
- Used to figure out what are important pathways in complex traits/disease. For instance if everyone with a certain defective enzyme has a 50% reduction in heart attacks this could lead to a drug candidate
> New study shows that IQ is 11% genetic > Is that at least 11% of their intelligence comes straight from their DNA, not that actually we could change 11% to 5% or 8% or 17% or most any number we want just by changing some laws or fortifying food or any number of environmental changes. The popularity of genetics research in popular media requires this bait-and-switch about what is actually going on.
If you are arguing that some people(maybe even a lot) misunderstand the definition of genetic causation in a similar way that you previously did, then that might be true. But when any experts, or any study talks about genetic causation they are using the definition the rest of us are using on the this thread.
And yet here we are on a thread about using a GWAS to prove what % of intelligence is heritable.
>If you are arguing that some people(maybe even a lot) misunderstand the definition of genetic causation in a similar way that you previously did, then that might be true.
My understanding isn't past tense. Here, allow me to illustrate why this makes no sense to me. Let's take what Gwern said above:
Are they genetic causation? Yes, because if you counterfactually changed the genes and nothing but the genes, you would get different outcomes. They are differences which make a difference.
But in my basketball example I could also say this:
Are they environmental causation? Yes, because if you counterfactually changed the environment and nothing but the environment, you would get different outcomes. They are differences which make a difference.
So we are going to take factors where both of those statements are true and still assert that they are genetically causal? That's incorrect.
>But when any experts, or any study talks about genetic causation they are using the definition the rest of us are using on the this thread.
The 'rest of us' in this thread are definitely not using the same language you use. Let's take a look at just the post author's comments:
---
>The most iron-clad part of showing direct gene->trait causation is the 4 sibling-comparison studies
except it's not iron-clad at all and it doesn't show direct gene->trait causation.
>where you see whether you can predict the difference between 2 siblings; as they have identical ancestries, families, locations etc, their environments are near-identical in the first place
Their environments are not near-identical.
>No, it doesn't. It merely tells us that with the data up to now, we've reached 11% (up from, incidentally, ~0.3% just 5 years ago). It provides a very loose lower bound on genes.
Except you're telling me that not only is a GWAS not useful for this purpose but that even this number shouldn't be treated as a 'lower bound' but as a potential identifier to further study the genes identified. It would seem that this
>> Observational studies of this type cannot show causality >They can: sibling comparisons.
Again, this quote appears to conflict with what you believe GWASs are capable of.
---
Again, why are these falsehoods being spread around? Because of an (mis)understanding that 'genetic causation' as measured in the paper is not manipulatable by the environment. Except that it actually is. And a misunderstanding of what GWASs are good for.
>- Embryo selection
No it's not. GWASs don't understand what they're measuring and label the environmental as genetic biasing their measurements. Embryo selection is done the old-fashioned way looking for specific genetic variants that we understand through the use of other types of studies.
>- Tests to figure out risk for certain diseases for preventative purposes
Look at the use of GWASs in identifying schizophrenia. They typically fail to replicate. GWASs find genetic connections where none exist.
Take schizophrenia. It could be that we run a GWAS on schizophrenia and find some genetic similarities. Then we look at these similarities and find one of them is a cuddle gene. A cuddle gene what the heck? Well the cuddle gene makes you want to cuddle which makes their mothers more likely to own pets to cuddle which makes them more likely to own cats which increases their risk of exposure to toxoplasmosis which increases their of schizophrenia. Is that a genetic causation? Because the way genetic causation is actually being used in this thread really doesn't lead to that conclusion. It leads to the idea that the identified genes make you a little more unstable in a certain way, not that we could totally eliminate that element o...
So, it might just be both environment and the genetic influence both correlating to the same thing, and they admit as much.
only 11 percent. this tells us that the remainder is more likely to be attributable to environmental factors.
in other words, even with a view of the genes, we would need to encourage education more strenuously if we wanted to improve educational outcomes for society.
No, it doesn't. It merely tells us that with the data up to now, we've reached 11% (up from, incidentally, ~0.3% just 5 years ago). It provides a very loose lower bound on genes. Saying that is like saying, 'our software testing up to now has only revealed bugs in 11 places, therefore, every other line of code must be bug-free!' Even if you knew nothing about the actual heritability estimates, one would think that finding so much so quickly would make one suspect that there's a lot more to find...
we know the environment impacts gene transcription, and we know that genetic determinism (as far as outcome phenotypes) is not a thing. likewise, there may be epigenetic factors at play which wouldn't be captured in a GWAS yet still might (or might not) have an impact on outcomes.
it isn't a "the other lines of code are bug free" situation so much as a "we expect there to be an abundance of bugs with undefined consequences outside of the subset which we have examined which most likely outweigh the stuff we found"
you can see that in the article the authors of the study specifically went out of their way to try to caution people against doing that.
The 11% of variance that can be predicted using the current information of a fairly bad genetic predictor. The predictor is bad because many cohorts used in this study do not share all their information, but only share so called summary statistics. Which are combined into a prediction model. If all the individual level data is available, it becomes possible to get a much better prediction model.
In addition there are ways to estimate how much can be predicted from genetic information, a term called heritability. Heritability can be estimated from so called twin studies. Where twins who are genetically identical (monozygotic) are compared to twins who are not genetically identical (dizygotic). In this way you can measure the environmental effect (the variation seen in the monozygotic twins) and the genetic effect (the variation that is seen in the dizygotic twins compared to the monozygotic twins). This is, of course, a gross oversimplification, but can provide reasonable measures of an upper bound of what can be explained genetically For most traits this upper bound is around 0.5 R^2, and I believe for educational attainment a bit higher, but I could not find it in the paper or a quick google search.
Another factor in this case is that there are unmeasured genotypes which can explain some extra variation. There are also solutions for this. Which are able to explain the upper bound of what can be predicted from the data at hand. For instance meaning that you can find 80% of all the genetic variation using the genetic measurements that you have. But you often need individual level data for this. Something which cohorts are not often prepared to share.
Another complicating factor is the heterogeneous nature of the phenotype: educational attainment is different per country. So it's hard to say what's really being measured. It's likely very different in France, compared to say The Netherlands. Due to the differing nature of their educational systems.
Personally I'm morally conflicted by these studies which measure intelligence. It's very interesting to see the practical outcomes from the paper. I've seen an example where it's well powered to stratify high educational attainment and low educational attainment individuals for social science studies. But I'm of the opinion that this may be a slippery slope towards misinterpretation, as we currently see in the whole IQ business. As well as too strong an emphasis on being 'smart' in certain social groups.
We know Intelligence (or whatever is being measured) is heritable. But I'm not fond of the implications which people attach to these studies. Even though I know people are going to act on it. For the earliest examples of this, I'm going to look at certain east Asian countries where ethics may be very different compared to the western views.
Are you, or your cohort fighting on the side of the angels, as morally conflicted by the treatment received by Arthur Jensen, Charles Murray et al?
It's less a "moral conflict" than the dawning realization that 50 years of New Age psychobabble (aka New Left "social sciences") are coming to an end.
Just wondering, have you ever encountered any of the serious critiques of twin studies.
The comparison of dizygotic (DZ) and monozygotic (MZ) twins relies on the Equal Environment Assumption (EEA) - which assumes that the environment experienced by MZ twins is the same as that of DZ twins. This assumption has been shown empirically to be false - parents, siblings, teachers, friends, and the twins themselves treat MZ twins very differently than they treat DZ twins.
There are good reasons to think that heritability results derived from twin studies are highly inflated.
> However, theoretical projections that failed to consider heterogeneity of effect sizes were optimistic Our and others’ findings suggest that imperfect genetic correlation across cohorts will be the norm for phenotypes, such as educational attainment, that are environmentally contingent.
So it is somewhat worrying that genetics predict the clearly environmentally contingent like educational attainment better than "cognitive performance."
> Our results also highlight two caveats to the use of the polygenic scores in research. First, our within-family analyses suggest that GWAS estimates may overstate the causal effect sizes: if educational attainment-increasing genotypes are associated with parental educational attainment-increasing genotypes, which are in turn associated with rearing environments that promote educational attainment, then failure to control for rearing environment will bias GWAS estimates.
That is a very complicated way of saying that some genes express themselves by black skin, and these genes have a strong effect on funds available to the school.
I don't agree that schools are any more 'clearly environmentally contingent' than intelligence, especially with the mandatory public schooling applicable to all of the cohorts used.
> That is a very complicated way of saying that some genes express themselves by black skin
That's a remarkable assertion, considering that this dataset is made of 100% white people, works on white siblings like Swedish twins, and the PGS is still predictive when applied within a separate African-American dataset.
Clearly I can change all the rules governing schools without changing the genetics of any of the students. So you are claiming that it is possible that if intelligence could be as environmentally contingent like school system, then one would need to show that this type of study can produce interesting results in the first place.
> That's a remarkable assertion,
I read the sentence as a general qualifier for this kind of studies. But the objection is, that these genes would produce a clear signal in this kind of studies without any of the causality taking place inside of a brain. (And unlike 'intelligence,' I can actually define what I mean by 'black skin' 'school funding' and 'average neighborhood income.' )
So for the study to have any kind of explainative effect, it needs to show that the effect is created by anything that can be reasonably called intelligence, instead of something that takes place completely outside of the brain.
0. https://en.wikipedia.org/wiki/African_Americans
It'll be another couple of years before "The Flippening", i.e. scientists looking to cement their legacy doing the political calculation and deciding they'd rather risk some short term fallout vs joining Lyskenko and Lamarck in the history book.
That elaborate kabuki we saw from David Reich at Harvard a few months ago? It'll get even more entertaining in the coming months and years as that fig leaf continues to shrink.
My interpretation of this FAQ is that it provides a great deal of context that would be apparent to people in the field (and therefore not included in the paper itself) but is critical background for anyone else who is trying to make sense of it.
It therefore provides a bridge between the highly technical paper itself and journalism which too often misunderstands the true meaning of research results.
I think this is so valuable. I hope that this kind of thing will catch on for more research that attracts public interest.
> 3.7. Could this kind of research lead to discrimination against, or stigmatization of, people with the relevant genetic variants? If so, why conduct this research?
Interesting discussion of genes and pathways involved around page 84. "Results: Causal Genes and Enriched Gene Sets"
I'll just type in the pathways. ( Supplementary Figure 22.) "Brain Specific Expression of Significantly Enriched Gene Sets across Development. (page 196)": Chromatic modification, protein binding transcription factor activity, npBAF complex, Central nervous system neuron differentiation, forebrain development, partial posttnatal lethality, abnormal cerebral cortext morphology, endometrial cancer , GAB1 signalosome, regulation of nervous system development, telencephaol cell migration, protein tyrosine kinase activity , neuron recognition, axon guidance, small cerebellum protein phosphatase regulator activity, signaling by eGFR, dendrite morphogenesis, signaling by NGF, regulation of neuron projection development, behavioral fear response, DAG and IP signaling, associative learning , voltage-gate calcium channel activity , ataxia, extracellular-glutamte-gate ion channel activity, post NMDA receptor activation events, regulation of neurotransmitter levels, regulation of synaptic transmission, synapse part ,gate channel activity.
Single Nucleotide Polymorphisms (snps) discussed in supplement are:rs10189857 rs1167898 rs11678980 rs18235539 rs182355396 rs186456786 rs186456786 rs186456786 rs5951458 rs61734410
Links to dbsnp for snps discussed in supplement: https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=101... https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=116... https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=116... https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=182... https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=182... https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=186... https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=595... https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=617... https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=624...
dbSNP Frequency Allele 1 Effect size
https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=235... rs2352974 0.4875 -0.0319
https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=190... rs1906252 0.4873 0.0307
https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=131... rs13107325 0.0750 -0.0543
https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=938... rs9384679 0.3717 -0.0282
https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=312... rs3128341 0.2052 -0.0334
https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=242... rs2426132 0.4576 0.0270
https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=121... rs12128707 0.7333 -0.0303
https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=117... rs11793831 0.4158 0.0269
https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=148... rs148696809 0.8828 -0.0406
https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=272... rs2726513 0.4069 -0.0261
https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=124... rs12448902 0.6061 0.0260
https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=374... rs3740422 0.3506 -0.0255
https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=805... rs8054299 0.6834 -0.0256
https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=111... rs11123820 0.6057 -0.0243
https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=108... rs10874938 0.5103 0.0236
https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=108... rs10875914 0.5993 -0.0237
The resolution of GWAS isn't at the level yet where we can pick out individual changes that cause a phenotype. Instead we can detect that a chunk of the genome is associated with a phenotype. These snps are like a street name, they identify the region associated.
We need whole-genome sequence on a million-person scale to actually find the causal variant. That will come with time.
Observational studies of this type cannot show causality, and the results have not been confirmed by independent methods.
They can: sibling comparisons.
> the results have not been confirmed by independent methods.
They're tested out of sample many times (4 for sibling comparisons alone!), and you can find many other papers successfully using the IQ GWASes. Belsky et al 2018 just came out a few weeks ago, for a particularly good example testing OP's PGS. (The PGS has been out since like March, so several papers have used it.)
>They can: sibling comparisons.
But then they get much lower concordance so they come up with various ad-hoc hypotheses to explain it, instead of acknowledging the negative result.
This is an hypothesis generating paper, using a highly suspect methods. Not a discovery.
I think the above snip (ha) from https://www.theatlantic.com/science/archive/2018/07/staying-... is useful in interpreting the results.
> Likewise, many of the genes that are associated with IQ today are likely important “because of how today’s IQ tests are set up.”
See also The Mismeasure of Man by Stephen Jay Gould.
[0] https://en.wikipedia.org/wiki/The_Mismeasure_of_Man
https://www.nationalaffairs.com/storage/app/uploads/public/5...
https://futurism.com/intelligence-changing-what-you-think-yo...
Performance in IQ tests.
> The current study further confirmed the finding from our earlier work that the effects of individual genetic variants on educational attainment are extremely small. The average effect size across the 1,271 genetic variants was just 1.8 weeks of schooling per allele; even the SNPs with the strongest associations only predicted around 3 weeks of additional schooling per allele. Taken together, these 1,271 SNPs accounted for just 3.9% of the variation across individuals in years of education completed. > Here is another way to think about this result. Imagine that we used the results for these 1,271 genetic variants (not the ~1 million SNPs across entire genome we discuss in FAQ 2.3) to predict the educational attainment for a new group of people (separate from our discovery sample). We could then compare each individual’s predicted educational attainment to their actual educational attainment. If we did so, our results suggest that we would find that the predictions and actual outcomes correlate only very modestly (at about r = 0.20). That, in turn, means that if someone were predicted to complete an above average number of years of schooling (i.e., to be in the top half of educational attainment), that person would have about a 58% chance of actually being in the top half of educational attainment. Fifty-eight percent is better than chance (i.e., 50%), suggesting that a prediction based on these 1,271 SNPs has more power to predict educational attainment than a coin flip—but only a bit more power. By contrast, a prediction based on a polygenic score that combines ~1 million SNPs that we studied (see FAQs 1.5 & 2.3) has more predictive power: r = 0.33, corresponding to 11% of the variation across individuals. > The contrast between the 3.9% of the variation predicted by the 1,271 SNPs and the 20% known to be explained by common SNPs (see FAQ 1.7) implies that there are many other SNPs that have not yet been identified. Even larger sample sizes will be needed to identify them.
They only included people of European ancestry to eliminate sources of noise and cofounding factors. The effect size is very small, yet it does exist.
This seems to prove that intelligence influenced by genetics but is affected by many thousands of genetic variations. There is no "smart" gene, just lots and lots of randomness.
Not quite how I would put it. There is no one smart gene, just thousands and thousands of smart alleles.
Contrary to how this field styles itself it is a quantitative social science having little to do with actual genetics, using next to none of its methods but a lot of quantitative social science's statistics with levels of signal similar to social science.
Only if the educational attainment difference predicted by these genes is because of "intelligence." It could just as easily have to do with "stick-with-it-ness" or "I-have-parents-that-insist-I-go-to-college" or "I-do-what-is-expected-of-me."