NeoEugenics: Curiosity or Game‐changer?

A more interesting question is: a more or less egalitarian (not-so-distant) future? (Once the genetic architecture of egalitarianism itself is cracked, this question will take on an even more delicious dimension.)


Shulman, C., & Bostrom, N. (2014).
Embryo Selection for Cognitive Enhancement: Curiosity or Game‐changer?. Global Policy, 5(1), 85-92.

Abstract: Human capital is an important determinant of individual and aggregate economic outcomes, and a major input to scientific progress. It has been suggested that advances in genomics may open up new avenues to enhance human intellectual abilities genetically, complementing environmental interventions such as education and nutrition. One way to do this would be via embryo selection in the context of in vitro fertilization (IVF). In this article, we analyze the feasibility, timescale, and possible societal impacts of embryo selection for cognitive enhancement. We find that embryo selection, on its own, may have significant (but likely not drastic) impacts over the next 50 years, though large effects could accumulate over multiple generations. However, there is a complementary technology – stem cell-derived gametes – which has been making rapid progress and which could amplify the impact of embryo selection, enabling very large changes if successfully applied to humans.

neoeugenicpossibilites

About these ads
This entry was posted in Uncategorized. Bookmark the permalink.

79 Responses to NeoEugenics: Curiosity or Game‐changer?

  1. Did you just admit to paleo-eugenics? That’s my only interest in this verbal masteurbation.

  2. nikcrit says:

    My initial five-cents-worth on ‘neo-eugenics': If it somehow prevails and does take hold of the socio-cultural zeitgeist, then i’d predict the distinction between born or ‘natural’ cognitive advantage-vs.-‘post-op’ genetic-intellectual enhancement would be the new and crucial distinction delineating ‘us’ from ‘them.’

    The hierarchy of designation will go on, adapting any and all changes and updates as needed, I’m pretty sure you’d agree, yes?

    • Chuck says:

      “the new and crucial distinction delineating ‘us’ from ‘them.”

      It would be “a” new and crucial distinction. I just don’t think that e.g., race was ever “the” crucial one. There was always class, and religion, and politics, and whether-you-had -the newest-coolest-sneakers-at-school, etc. But ya, it’ll make things interesting — it will surely be a perspective changer for the race/anti-race obsessed.

  3. nikcrit says:

    Nikcrit on your blog in 2011:

    I sincerely believe there’s a better chance —– a much better chance, actually —– of putting energy, capital and resources into a scientific solution to ‘closing the gap.’ That, and perhaps a mix of policies that would directly or indirectly lead toward black self-correcting eugenic-breeding behaviors…..
    I mean, if each race has its special talents, contributions and abilities, then I say the northern Europeans should get down to their labs and start cooking up a game-changer that will make all these stale debates and polemics of the last 50 years suddently irrelevant.
    For it will only be some such unanticipated development that will put this issue to rest and make the reams of established contemporary ‘moral-theology’ irrelevant.

    So perhaps the ‘neo-eugenics’ you reference now are the first real evident truth of what I semi-seriously conceived and wished for way, way back at the start of this aging decade, eh? lol!

    Idk; it seems the older i get, the clearer it becomes that these social divides and tidal-sway bearing ‘isms’ are things society sustains and even lobbies for, rather than representations of some biologically predicated reality that must be dealt with. just biz as usual, in a way.

  4. alabastrine excellence says:

    So, the new, genetically-altered, genius Ubermensch will be rulers of the Brave New World. I’m sure that the cyborgs will help them in eliminating all of us naturally-developed sub-geniuses from their presence. Only the wealthy ‘elite’ will be able to afford these new ‘designer babies’. Maybe they’ll keep some of us ‘natural’ humans around as pets. It reminds me of that old horror movie, “Children Of The Damned”. Genocide and extinction of ‘humanness’ for all! Hey, it’s not nice to fool with Mother Nature! That’s what got us into the trouble we’re in now with the great, dusky-hued ‘minorities’ of the world. We keep meddling in their affairs, feeding them, and trying to get them to adapt to White society, White morality, and White law and order. They’re outbreeding and overpopulating us with more and more mouths to feed, more ‘diversity’ divisiveness, and more anti-White hate and homicide. We should quit while we’re behind.

  5. bob sykes says:

    In as much as the cost per embryo are on the order of tens of thousands of dollars, such procedures will be so rare as to be irrelevant.

    • Chuck says:

      “In as much as the cost per embryo are on the order of tens of thousands of dollars, such procedures will be so rare as to be irrelevant.”

      Moore’s Law applies to genomics too.

  6. John says:

    Chuck,
    Where have you been? Piffer’s turning out papers and you’re not blogging? According to his findings (factor scores) thus far, there’s about a two standard deviation difference in intelligence-enhancing alleles between British people and people of sub-Saharan African descent, including African-Americans. Assume for the moment an average British IQ of 100 and an African-American average of 85. Assume also that the African-American result does NOT reflect malnutrition or other deprivations and that the Flynn effect has stopped. Then a one standard deviation change in intelligence-enhancing alleles is worth 7.5 IQ points for genotypic IQ; African-Americans have a two standard deviation deficit in intelligence-enhancing alleles and thus a 15 point IQ deficit. However, Piffer’s results ALSO indicate no deficit for Southeast Asians; to the contrary, Vietnamese are found to be innately more intelligent than British people, and Cambodians are innately more intelligent than the French.

    Get in the game!

  7. First Ypres says:

    there’s a reason why breeding and evolution work and genetic engineering ala Prof Steve Shoe may not or rather that there is no good reason to believe it will.

    http://www.faculty.biol.ttu.edu/Rice/rice08b.pdf

    if you can understand the above paper you’ll understand a lot more than Prof Shoe, fuehrer of the BGI study, Razib Khan, Greg Cockring, or any HBD blogger except you.

    in short, additive heritability is not what one might think it is. it is merely the degree to which the P(G, E) surface can be approximated by a plane.

    when the region in GxE space changes, so does the planar approximation.

    — BGI volunteer

    • Chuck says:

      The posted article isn’t about genetic engineering, you twit; it’s about eugenics, which is selective breeding/zygote selection. For this, interactionism on the deep is irrelevant; one only cares about the surface;.GxE on that is no less relevant for breeding and evolution; yet the latter works dandy; because there’s no GxE across most of the functional environmental range.

      • First Ypres says:

        It depends on how many alleles are selected for.

        Embryo selection?

        How can the phenotype of embryos be measured?

        Sorry to say, but like all hereditists.

        YOU’RE A MORON. AND HAVE NO IDEA WHAT YOU’RE TALKING ABOUT.

        — BGI volunteer http://tinypic.com/view.php?pic=28s67ih&s=8#.VDuHCtfn9dg

        • Chuck says:

          You select for polygenic allelic score, dipshit. Don’t comment on this blog again.

        • Chuck says:

          I didn’t ban, in the sense of block; rather, I asked you to not comment; as I don’t anticipate anything remotely informative. The GxE interaction critique (in context to Iq) is hardly novel. It’s stuck in apriori mode i..e, no evidence. Or do you think that it’s not detectable in the case of e.g., IQ? Let’s start here: What percent of IQ variance do you attribute to GxE? Evidence? No one rejects GxE, as such; it’s part of the standard biometric model; a significant GxE variance component just doesn’t fit the data, when it doesn’t (as in the case of IQ).

      • First Ypres says:

        GxE on that is no less relevant for breeding and evolution; yet the latter works dandy; because there’s no GxE across most of the functional environmental range.

        Yet more misunderstanding.

        A curve can be approximated by a lot of line segments. Yet the whole curve is very poorly approximated by a single line.

        • Chuck says:

          GxE interaction — meaning that e.g., for different genotypes the same environment has different effects on phenotype — is detectable using classical and more contemporaneous biometric methods. For certain traits e.g., diabetes it may explain a non-trivial portion of variance. With regards to intelligence this hasn’t been found. If there is such interaction, it’s not in the normal range of environment. Thus, not relevant. As Plomin noted 25 years ago: “There is no conspiracy against interaction: If an interactive model could be shown to fit the data better than the traditional model, researchers would be quick to use it.” Where the evidence (in the case of IQ)? I thought so.

        • Chuck says:

          First Ypres. Not interested in wasting time.

          Here were you points summarized:

          1. “the sense of “GxE” in “GxE space” is totally unrelated to the behavior genetics (i.e. moron) sense of “GxE”.”

          GxE space describes norms of reaction. Biometric GxE describes a region in phenotypic space which transverses norms. To be clear, biometric GxE is:

          What constitutes a good environment for one genotype in terms of the development of the phenotype may constitute a bad environment for some other genotype in terms of the development of the phenotype OR Environmental advantage, through acting in some phenotypic direction for all genotypes may have unequal phenotypic effects on different genotypes

          This, of course, is statistically detectable. And the variance for a trait in a given population owing to GxE can readily be quantified. For IQ it has been found to be trivial. Interactive models simply don’t fit the data well.

          Now, of course, outside the ordinary range of environment, the range we don’t currently live in, biometric GxE might be of importance. But this is presently irrelevant.

          So to locate the point of disagreement, let me know if you agree thus far. Perhaps you consider the above “moronic common sense”. Honestly, I can’t tell, as you seem to be conceptually confused.

          2. “the polygenic score is meaningless when it’s an extrapolation far outside the range of G for which the linear fit was made.”

          It’s perfectly meaningful given the same norm of reaction — which, granting my point above, encompasses the ordinary range which we happen to currently live in. Thus, for example, the allelic scores based on the known handful of well replicated alleles allows one to predict individual IQ scores to a small degree. See, for example, Zhu et al. 2015, if I remember correctly. Now, it may be that if one moves “far outside” the current range of environment one will bump into another reaction norm in which a polygenic score from this range would be predicatively meaningless. It may. But it may not and we don’t know. More importantly, that’s a range of environment which is currently far outside the one we know. Generally, this locality argument is a sweeping nihilistic critique of all research. One doesn’t know, for example, what effect environmental factor X (e.g., nutrition, or schooling, or breastfeeding) will have on such a trait “far outside” the current range, therefore we can’t know anything for certain and therefore all current research must be useless. Oh my god, maybe far outside our current range of experience, gravity runs reverse! Better hold onto your hat!

          Silly. But probably that’s not what you meant. But some e.g., Lewontin have made this type of argument — so I can’t be sure. For clarity, let me know if the area of disagreement lies elsewhere.

          3. “quoiting Plomin. that guy is a joke; you and Plomin simply lack mathematical sophistication; I’m an actuary. I am mathematically sophisticated.”

          These are the types of claims which puzzle. Plomin was lucidly discussing Biometric GxE and the lack of evidence for it (in the case of IQ). It’s really a simply matter — which you should be able to grasp. Thus, there must be conceptual confusion. You must be talking about something other. What that is, I’m not yet sure.

          4. “there’s nothing other than GxE, but within a tiny range of the (G, E) plane the P(G, E) surface can be approximated by a plane without any GxE.”

          Sounds metaphysical. Refer back to points 1 and 2. Perhaps we live in a tiny range of the metaverse were the physical laws are regular. Outside, anything goes. Ok .. but as far as we can tell, tiny we are stuck in this tiny range. So those local laws generalize just fine for us. But this statement makes your position less opaque. You’re now not claiming that GxE explain a significant amount of IQ variance, but rather that it could possibly. Well, a lot of things could be possible. I read Sci Fi, too.

          4. “besides GxE has been found in four cases I can name. and has been found in innumerable cases whenever the sample h^2 varies significantly from one study to another, that is from one small region of (G,E) space to another.”

          Name the four cases. Also, environment x h^2 (i.e., E x G/P) isn’t the same as G x E. For example, in all of Turkheimer et al.’s studies additivity held. Sounds like you’re confusing issues. Yet you said: “in order to get significantly different sample h^2s the P(G,E) surface must bend at some point, aka GxE.” No, because h^2 is G/P. So you have E x G/P, which can be caused by differential P. For example: Hart, et al. (2013). Expanding the environment: gene× school‐level SES interaction on reading comprehension. Journal of Child Psychology and Psychiatry, 54(10), 1047-1055.

          5. twice by Turkheimer and in the case of eveningness. night owls have higher IQs yet get much worse grades. if school were held at night? and here’s another http://www.ncbi.nlm.nih.gov/pubmed/6609726. and here

          I don’t get the point regarding Turkheimer. I’ve read quite a bit of his work and I don’t recall him ever uncovering significant biometric GxE, not to be confused with E x h^2. And the link was for schizophrenia. Yet this is one of those traits for which biometric analysis finds evidence of GxE e.g.,

          Gene-Environment Interactions in Schizophrenia: Review of Epidemiological Findings and Future Directions

          http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2632485/

          Your arguments are all over the place. Which is it?

          There is evidence of significant GxE for IQ, given the present range of environments? (Uh, No.)
          In some possible future range of environments, there could be significant GxE for IQ. (Uh, so what?)
          Environment x h^2 interactions evidence G x E interactions. (Uh, no.)

          6. “there is to date not a single study made up of twins like these: http://www.pbs.org/independentlens/twin-sisters/
          and therefore ALL of behavior genetics is meaningless.”

          Segal, N. L., & Cortez, F. A. (2014). Born in Korea-adopted apart: Behavioral development of monozygotic twins raised in the United States and France. Personality and Individual Differences, 70, 97-104.

          7. “here’s that paper again. http://www.faculty.biol.ttu.edu/Rice/rice08b.pdf
          or is it over your head?”

          Nope. But no evidence of significant IQ GxE

          • Chuck says:

            On the other hand, First Ypres does provide yet more anecdotal evidence against the existence of a g-factor.

  8. Revolver says:

    ‘ So, you have to agree that it could have been found if it was like that for other traits e.g., Schizophrenia’

    I missed this. First, the fact that schizophrenia has a high h^2 but has also been found to show significant G x E should tell us all something; namely, that high h^2 does not always imply a high independent G.

    Further, it could simply be that IQ research lags mental disorder research in this area. But it could be catching up: “Emergence of a Gene × Socioeconomic Status Interaction on Infant Mental Ability Between 10 Months and 2 Years.” http://pss.sagepub.com/content/22/1/125.

    • Chuck says:

      “I missed this. First, the fact that schizophrenia has a high h^2 but has also been found to show significant G x E should tell us all something; namely, that high h^2 does not always imply a high independent G.”

      It would depend on what you modeled for. Some analyses just model for E + G. One needs special designs to disentangle other variance components such as COV(GE), VARGE, Epigenetic Variance, Dominance. The first point is that this can be done. The second point is that when this is done, for IQ, GE is generally insignificant; there’s 50 years of research on this. The third point is that even for trait for which GE turns up as significant, one can still meaningfully interpret heritability estimates. I uploaded Tal’s paper: https://occidentalascent.files.wordpress.com/2014/11/the-impact-of-genee28093environment-interaction.pdf Thus, I don’t see the fuss.

      “Further, it could simply be that IQ research lags mental disorder research in this area. But it could be catching up: “Emergence of a Gene × Socioeconomic Status Interaction on Infant Mental Ability Between 10 Months and 2 Years.”

      Again, that’s E (in the form of SES) x h^2. It’s funny that you cite that example because I used some of the authors data, which they generously provided, in a meta-analysis on (the non-existence) of race x h^2 interactions. The authors assumed an additive model, thus “Each latent factor described above (initial level and change in mental ability) was modeled as a linear combination of three standardized (z-scored) biometric components: an additive genetic component (A), a shared environmental component (C, i.e., all environmental influences that make twins similar), and a nonshared environmental component.

      • Revolver says:

        ‘The second point is that when this is done, for IQ, GE is generally insignificant; there’s 50 years of research on this’

        GxE modeled as “sensitivity?” The problem I refer to is a ranking problem. Point me to one of those studies where they have modeled GxE to do away with this ranking problem.

  9. Revolver says:

    Hm.

    Well let me try to reduce this into 2 dimensions.

    First assumption, we have captured a general population genotype. That way we don’t need to worry about intersecting norms.

    First we take a measurement in the unstratified environment.

    We come up with a line…let’s say it’s .9x or some high correlation.

    Now we stratify. We separate environment into 3 levels. In each level we check the correlation.

    What we want to know is whether the genetic effect looks like a curve or a line. So we then take our heritability from the correlations. Then we aggregate the lines back into the original unstratified space. If it looks like a curve (changing rate increase), then each new increment of environment is adding many more genetic differences. If it looks more like our original line, then not so much. But the former would be g x e.

    Re: uniform factors, I’m unsure how “black” in America doesn’t function as an immediate signal for a certain mental image which in turn prompts certain treatment over a wide range of within black skin color, culture, etc.

    The Y factor would be something cultural.

    Regarding the study, I know it’s old, but I haven’t seen any other studies done on admixture. I’ve seen many inferences from biracial test scores and whatnot, but I consider a study specifically on the matter more informative.

    • Chuck says:

      You: “f it looks like a curve (changing rate increase), then each new increment of environment is adding many more genetic differences. If it looks more like our original line, then not so much. But the former would be g x e.”

      You keep bouncing back and forth. Do you mean that the former would necessarily be g x e sensitivity or that it would be consistent with this? My point has been that you can get the same effect without G x E in the sensitivity sense. I gave a plethora of examples. Just imagine if we measured the heritability of correct scoring on a multiple choice math test in a population and then introduced a series of environment moderator — e.g., randomly confiscating calculators — and kept re-estimating the h^2. Our moderators would (imaginably) lead to an inflation of e^2 and decrease in h^2 but it would be unrelated to sensitivity sense GxE. You’ve been trying to dismiss these types of hypotheticals arguing that this would represent psychometric measurement error, but that’s meaningless, since biometric analysis doesn’t care about the trait in question, it doesn’t care if there’s bias with respect to a latent variable. We have a genuine conceptual discordance here. We have three phenomena and their relations are not clear:
      (a) crude h^2 x sub-population interactions, (b) h^2 x e^2/c^2 interactions, (c) sensitivity sense GE. Now it’s obvious that (a) and (c) can be discordant. The issue is (b) and (c) when sophisticated model are used. Numerous studies have looked at (b) and they haven’t found consistent evidence of this. See for example: Molenaar, et al. (2013). Genotype by environment interactions in cognitive ability: a survey of 14 studies from four countries covering four age groups. Behavior genetics, 43(3), 208-219. Now, they would have quantified the overall meta analytic effect so when I write them I will ask for the variance component and we can plug it into Tal’s equation and generate a probability for you. I’m just not certain that (b) ~ (c).

      You: “uniform factors, I’m unsure how “black” in America doesn’t function as an immediate signal for a certain mental image which in turn prompts certain treatment over a wide range of within black skin color, culture, etc. The Y factor would be something cultural.”

      Because, of course, “blacks” aren’t a uniform group. And they aren’t uniformly affected by the IQ depressing factors. For example, many persons with one White and one Black parent identify as “Black” and these individuals perform intermediately to the parental groups. And they happen to have intermediate histories, phenotypes, and so on. You can predict the scores of these “Black” people using the education level of their White parent, controlling for that of their Black parent. The relevance of this situation regarding mixed race individuals was explicated by Baker (1974):

      Baker (1974)

      You: “Regarding the study, I know it’s old, but I haven’t seen any other studies done on admixture. I’ve seen many inferences from biracial test scores and whatnot, but I consider a study specifically on the matter more informative.”

      I discussed Witty and Jenkins here. The results weren’t actually inconsistent with a genetic/shared environmental hypothesis. Anyways, numerous other studies have been conducted demonstrating an association between indexes of ancestry and outcomes:
      studies reviewed by Shuey 1966; analyses using more recent nationally representative samples; and others which I reviewed here in sections O through Q. And of course there are admixture mapping studies which use education, income, and job prestige as outcomes. I don’t think that a more or less linear association between racial ancestry and cognitive ability is seriously disputable at this point.

      I find it bizarre that anyone would earnestly defend a uniform factor explanation, since in practice no one really takes this possibility seriously. Sociologists, for example, recognize that differences are being inter-generally transmitted on the family level; they just attributed this to shared family environments; hence they try to statistically explain differences using regression analysis. It would be nice if someone could list the specific factors that are said to be uniformly distributed e.g., all Blacks eat Watermelons, therefore…

    • Revolver says:

      I didn’t bounce back and forth. In one case I said it would be g x e if in fact, every environmental increase in turn added an increasing rate of genetic effects. In the other, I said that this wasn’t a smoking gun; stratifying the population reduces sample size (and if the overriding E itself has a genetic component, then you’d be restricting varG with each successive stratification), etc. etc. etc. So your measurement may be off. But if it is in fact a curve, then that would be g x e.

      “Then introduced a series of environment moderator — e.g., randomly confiscating calculators — and kept re-estimating the h^2.”

      This would look like a h^2 line, though, assuming the depressive effect of not having a calculator was a non g x e; of course, if there are certain individuals who have G: IQ sensitive to working arithmetical memory and G: IQ not sensitive to working arithmetical memory, you may find some g x e here.

      ‘You’ve been trying to dismiss these types of hypotheticals arguing that this would represent psychometric measurement error, but that’s meaningless’

      I dismissed one type of hypothetical on that ground. And it’s not because it was measurement error, it’s also because the h^2 would probably look like a line there too.

      ‘ (b) h^2 x e^2/c^2 interactions, (c) sensitivity sense GE’

      At all times h^2 is the proportion of phenotype variance attributable to genotype variance. And typically, an index of causation. I don’t know what the confusion is. If h^2 grows at an inconstant rate through an environment space, that is g x e.

      H/h^2 is just varG, which makes it the derivative of G, the effect. So, if h^2 is a curve, then it stands to good reason that G itself is in fact a linear function and increases. If h^2 is linear, then G remains the same, even if it accounts for more or less of the trait proportionally.

      ‘Genotype by environment interactions in cognitive ability: a survey of 14 studies from four countries covering four age groups’

      This paper found a lot of G x E. When they said ‘consistent’ they meant that they couldn’t find any identifiable trend to the G x E. It went in several different directions. Of course, that’s not surprising, considering the critique raised by your other commenter.

      ‘Because, of course, “blacks” aren’t a uniform group.’

      If they are recognizable as a race, then I’m not sure why they can’t be roughly uniform.

      ‘White and one Black parent identify as “Black” and these individuals perform intermediately to the parental groups. And they happen to have intermediate histories, phenotypes, and so on’

      They’re also a very tiny segment of the “black” population, which would make this differential, if conceded, inconsequential.

      ‘I discussed Witty and Jenkins here. The results weren’t actually inconsistent with a genetic/shared environmental hypothesis. Anyways, numerous other studies have been conducted demonstrating an association between indexes of ancestry and outcomes.’

      Your chart undoes a lot of what you try to do in that post. You give many plausible reasons for why the study does not sink your position. At the same time, your efforts are speculative.

      ‘Studies reviewed by Shuey 1966. Analyses using more recent nationally representative samples. And others which reviewed here in sections O trough Q. And of course there are admixture mapping studies which use education, income, and job prestige as outcomes. I don’t think that a linear association between racial ancestry and cognitive ability is seriously disputable at this point.’

      The reviewed studies note performance differences. Last, you present to me two blog posts. What of the black-german soldier study? What of the Willerman, Naylor, & Myrianthopoulos study re the difference between white mother-black father/black mother-white father children? Scarr, Pakstis, Katz, & Barker, 1977 correlation between admixture and IQ almost nothing; Loehlin, Vandenberg, and Osborne (1973) same.

      I beg to differ on that “serious dispute.”

      ‘I find it bizarre that anyone would earnestly defend a uniform factor explanation, since in practice no one really takes this possibility seriously.’

      Do you not believe blacks, as a group, are systemically treated differently than whites? I am not here to peddle absolutes, but the contribution from this variable is significant. Or at least a reasonable case sits exists for it being so.

      ‘It would be nice if someone could list the specific factors that are said to be uniformly distributed’

      It’d be like listing the many different specific factors affecting the life of an ugly man.

      • Chuck says:

        I had a lengthily reply, but I won’t post it, because it just tracks this discussion which is headed off-track. Your concern is about ranking people. You argue that GxE interactions in the sensitivity sense makes this impossible. I replied that GxE interactions in the biometric sense are not large in the case of IQ such to preclude probabilistically ranking genotypes. I noted that the math was explicated by Omri Tal:

        Tal, O. (2009). From heritability to probability. Biology & Philosophy, 24(1), 81-105.
        Tal, O. (2012). The Impact of Gene–Environment Interaction and Correlation on the Interpretation of Heritability. Acta biotheoretica, 60(3), 225-237.

        And I pointed to some attempted assessments of the variance due to GE interactions. From there on it seems that we got lost: You argued that there was “significant GE interaction” and pointed to some studies showing h^2 x subgroup interactions. I disputed the inferences, pondered about the connection between biometric GE and sensitivity sense GE and that between h^2 x E and varG x varE/varC, and then I complained about sloppy semantics in the literature.

        Let’s start again with a secure (and narrow) conception of biometric GE: “Simply stated, GxE designates a contribution that some non-additive function of the hidden variables G and E makes to the phenotypic value, independently of the main effects of these variables. (Tal, 2012)”. So, P = G + E + GE. And then we note that biometricians look for this in the form of VarG x VarE/C.

        We also note that developmental GE could exist in the population in a cryptic form without actually being manifest as discussed by Tabery and Griffiths (2010): “But in the context of the experimental study of behavioral development interaction is a causal-mechanical phenomenon, not just a statistical one. Genetic and environmental factors causally interact in the processes that give rise to phenotypes. (Perspectives on Behavioral Genetics and Developmental Science. Handbook of Developmental Science, Behavior, and Genetics, 41)”

        Thus, the non substantiality of biometric GE (varG x varE/C) does not necessitate that developmental sensitivity GE is not lurking in the wild waiting to render genotypic predictions invalid. Yet, on the other hand, the possibility of cryptic developmental sensitivity GE does render biometric decompositions invalid either, since they provide local causal information. And, for all we know, the local range is our universe. One can compare the situation to the 18th century debate about the indubitability of basic principles such as causation. Hume argued, for example, that all causal inferences are tentative and local, since it could always be otherwise. Others, such as Kant, tried to prove the principles a prior, to secure science. Over time, though, we ended up more or less granting Hume’s position but adding Bayesian probability.

        So, it seems that there is only the empirical matter of determining varGxE and of clarifying the relation between h^2 x subgroup interactions and varGxE.

        (1) Regarding a paper I cited, you said: “This paper found a lot of G x E. When they said ‘consistent’ they meant that they couldn’t find any identifiable trend to the G x E. It went in several different directions. Of course, that’s not surprising, considering the critique raised by your other commenter””

        As for finding a “lot” I didn’t notice a quantification of the relative magnitude. They were clear that they didn’t find evidence of systematic h^2 x a^2/e^2: “In the present study, it appeared that results on the aggregated data are difficult to interpret in terms of a genuine GxE effect because results differed considerably with respect to presence, as well as direction of the GxE effect, across countries and ages. The possibility that differences in measurement instruments across countries caused artificial GxE in different directions cannot be ruled out (see Eaves et al., 1977)…Taking all together, using the heteroscedastic ACE model we did not find a consistent pattern of results between age groups, datasets, and countries.” As discussed, the unsystematic nature of the results makes it unclear if the apparent effects were real or were artifacts of measurement error.

        Whatever the case, one could take the meta-analystic value at face value and input this into a probability function and thus probabalistically rank genotypes. This would give us a lower bounds range, since our suspected GxE is only suspected. To move beyond this broad claim, we would need a specific varGE estimate. Email the authors and ask.

        (2) Regarding the point about h^2 this is both silly and a distraction. You said:

        “So your measurement may be off. But if it is in fact a curve, then that would be g x e….This would look like a h^2 line, though… If h^2 grows at an inconstant rate through an environment space, that is g x e.”

        (a) If the environmental moderator itself has a curvilinear effect on the phenotype, then it could have a curvilinear effect on the intraclass correlations and so on the h^2 estimates independent of sensitivity sense GxE. If you want, we could model the normally distributed effect of three hypothetical conditions e.g., malnourishment with an average depressing effect of 1.5 SD, poor nourishment with an average depressing effect of 0.5 SD, and well-nourishment with an average depressing effect of 0 SD. This almost surely would produce a curvilinear h^2 x E (nutrition), (b) As you noted, there can be varC x varE effects and these can result in h^2 x E effects. For example:

        “How can the present finding of SES moderation of the shared environmental effect on IQ, be reconciled to the reports of SES moderation of the genetic component of IQ? An increase in the contribution of C in lower-SES families would seem to require a reduction in the relative contribution of A because environmental and genetic variance components are complementary, and explain 100% of the variance. However, this is only the case for standardized components that are forced to sum to 100% regardless of total variance differences. Our most consistent finding is that total IQ variance is greater in lower-SES families, which must be caused by greater A, C, or E components of variance in lower-SES families. Although the power demands are daunting to disentangle A and C sources of this increased variance in lower-SES families, data from our large sample suggests that the source is C rather than A. The genetic effect does not differ for low- and high-SES groups using unstandardized estimates (A, C, and E) that take into account the greater total variance in the low-SES group, but the relative contribution of genes – heritability or h2 = A/(A+C+E)) – is lower in low-SES families because the shared environmental effect increases.”

        http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0030320#pone-0030320-g007

        The above would be similar to the examples which I offered, insofar as one is inflating the total phenotypic variance by increasing the environmental.

        (c) A good example of a curvilinear h^2 x subgroup interaction is the age effect for height or IQ. The major explanations for this are COVGE and .genetic amplification. According to the latter, the effects of genetic differences are related to the amount of sensory input. This, I suppose, would be a form of GxE interaction in the sense that genetic differences interact with environmental ones to produce phenotypic differences. Since I think that this is a more correct model, I guess that here again I’m a GE interactionist. Nonetheless, since the phenomenon is open to a COVGE interpretation, we can conclude that we can’t conclude that there is GE interactionist simply on account of there being nonlinear h^2 x subgroup interactions.

        I won’t presently comment on the B/W IQ issue, because that’s even more of a distraction. I clearly don’t agree with your assessment.

        • Chuck says:

          Another issue to keep in mind is that varG x varE interactions show at as VarE in twin studies. Only VarG x VarC would be conflated with VarG (as discussed in e.g., Purcell, 2002). Now recall the findings from the 14 study analysis which I pointed to: “As none of the AxC interactions were significant and at most linear AxE interactions were found, we focused on the results of the ACE model with a linear AxE interaction in all age groups for the remaining analyses”. Thus you see in their table 6 parameters for E and AxE. So in all of these studies A is not being overestimated. Rather E is. So I imagine that you can simply input the A estimate into Tal (2009) — that is you can safely treat ExA as E given the interest in genotypic not unshared environment ranking.

          AxEinter

      • Revolver says:

        ‘If the environmental moderator itself has a curvilinear effect on the phenotype’

        You equate the VarP in each strata. I already said this earlier…you can’t stratify a population successfully on the basis of the metric unless you account for these parameters. If each strata has different VarP, then there’s a multiplier xVarP from lowest to highest strata. If h^2 increases/decreases by this multiplier, then we can be sure that the moderator drives this relationship and the increase in VarG is 0. If it increases/decreases > than this multiplier, then the increase in VarG is > 0.

        On the basis of that amount >< multiplier, we can gauge how much varGh^2 rises or falls.

        'Your concern is about ranking people'

        Not in this instance. H/h^2 always equals simply varG/varP. That is all it can be. You are looking at studies which assess h^2 versus c^2 (shared environment over variation). The one I mentioned noted that h^2 itself did not significantly change over the space. So, it’s consistent with what I’ve been saying.

        Look at it another way. What is the heritability of the trait’s heritability? If it is constant throughout, then there is no g x e. If it changes, there is g x e. Hence, a curved H/h2 indicates that heritability is heritable, which is genetically controlled sensitivity to environment.

        Your other examples likely would have a line for h^2, indicating an unchanging sensitivity to the environment.

        ‘I replied that GxE interactions in the biometric sense are not large in the case of IQ such to preclude probabilistically ranking genotypes’

        Those papers do not stand for that proposition. They simply allow an individual to calculate the probability of G >E given certain conditions. It actually breaks h^2 down as a fact at any given time or place. Instead of “this is 100% certainty at this time and place,” instead, depending on the h^2 found, it’s “this is a 100 x (1-p) certainty at this time and place.” This does not rank genotypes. Ranking still remains as an assumption.

        ‘it seems that there is only the empirical matter of determining varGxE and of clarifying the relation between h^2 x subgroup interactions and varGxE.’
        ‘If you want, we could model the normally distributed effect of three hypothetical conditions e.g., malnourishment with an average depressing effect of 1.5 SD, poor nourishment with an average depressing effect of 0.5 SD, and well-nourishment with an average depressing effect of 0 SD. This almost surely would produce a curvilinear h^2 x E (nutrition)’

        If the heritability of the trait’s heritability increases, then we have different sensitivities to nutrition in our chosen population. That is G x E. However, the interaction has limitations. For example, there can significant G x E over an interval that slows down or speeds up past certain points. The surface of P is complicated.

        ‘ For example:’

        I ALREADY cited this paper and explained why this explanation is unconvincing, in terms of a generalization and also consistent with their being inconsistent g x e in the literature. In either case they DID NOT find non-linear h^2. The effect remained constant.

        ‘Nonetheless, since the phenomenon is open to a COVGE interpretation, we can conclude that we can’t conclude that there is GE interactionist simply on account of there being nonlinear h^2 x subgroup interactions.’

        GE correlation is just a way of explaining GE interaction. It’s an organism’s propensity to seek out environments that it is sensitive to in a good way or avoid environments that it is sensitive to in a bad way. They need to be separated only on paper to avoid confounds, but it’s all describing the same phenomenon in reality.

        The stratified population example should help explain why the additive model is just an approximation. Fitting the additive model over a wide E range (even assuming G was small), showed that the original line would likely be a very inaccurate depiction of the true surface. In reality, G is not small and there is no one dominant genotype in the population.

        ‘I won’t presently comment on the B/W IQ issue, because that’s even more of a distraction. I clearly don’t agree with your assessment.’

        You brought up the issue, but fair enough.

        • Chuck says:

          You said: “You equate the VarP in each strata. I already said this earlier…you can’t stratify a population successfully on the basis of the metric unless you account for these parameters.”

          We agree then that h^2 interactions per se don’t entail GxE interactions. That’s the only point which I wanted to make in this regards. I don’t necessarily agree that holding VarP constant would do the trick, because there are other variance components that could fluctuate and give the mirage of GE interactions. For example CoVGE(unshared) mimics additive genetic effects in basic twin designs, while CoveGE(shared) mimics environmental effects. Imagine the basic underlying model:

          VarP = A + C + E + CovGE(shared) + CovGE(unshared)

          Now just oscillate the CovGEshared/unshared components and you will get what appears to be an A x E/C effect. So, I think that it’s more complex than what you suggest. Though, granted this would be a generic problem.

          “Look at it another way. What is the heritability of the trait’s heritability? If it is constant throughout, then there is no g x e. If it changes, there is g x e. Hence, a curved H/h2 indicates that heritability is heritable, which is genetically controlled sensitivity to environment,”

          Talk of the heritability of heritability is nonsense. We agree that if one detects VarGxVarE/C one has biometric GE. We agree that fluctuations in h^2 don’t logically entail biometric GE. There should be nothing more to discuss on this point.

          You: “I replied that GxE interactions in the biometric sense are not large in the case of IQ such to preclude probabilistically ranking genotypes’

          “Those papers do not stand for that proposition. They simply allow an individual to calculate the probability of G >E given certain conditions. It actually breaks h^2 down as a fact at any given time or place. Instead of “this is 100% certainty at this time and place,” instead, depending on the h^2 found, it’s “this is a 100 x (1-p) certainty at this time and place.” This does not rank genotypes. Ranking still remains as an assumption.”

          You’re mixing up issues. First, when it comes to individual one can only make probabilistic statements in the first place. You can only make statement such as the following::
          ………….
          If the distribution of P is approximately normal and we assume no G–E
          interaction and zero or only slight correlation, we can also make the following quantitative probabilistic statements:
          (3) For a particular individual P = p (in standard deviations from the mean), we
          can quantify the probability that genetic factors contributed more than
          environmental factors to its deviation from the mean.

          ‘‘So, if the heritability of IQ in a population is 80%, one can say of an individual in that population who has an IQ of 120 that the deviation from the mean (20 points) is with probability 84% due more to genetic factors than environmental ones. (Tal, 2009)’’
          ………………….

          I took this as granted. This lets you rank people, but the rankng is probabilistic. The above, for example, roughly translates to: “There’s an 84% change that a person with an IQ of 120 has a genotypic IQ >110″. But the probability that someone with an IQ 90 will have a genotypic IQ >110 will be much lower; thus ranking. This is about all one can do — knowing just phenotype and heritability — so if you don’t like this, then you should have said so from the start. Now as indicated in (3), GE interaction is a potential problem. When this is too high — or at least when h^2 is too entangled with 1-h^2 — tt’s impossible to meaningfully move from population to individual in the probabilistic sense. But when it’s low to intermediate, one can conditionally probabilistically :
          .
          “The framework outlined in this paper allows incorporating estimates of gene–environment interaction and covariance within a probabilistic interpretation of heritability (Tal 2009; see Tal et al. 2010, for an extension that includes a putative epigenetic variable). Specifically, given estimates of heritability and the variance components associated with GxE and rGE, a method based on the standard quantitative model generates the conditional probability that genetic factors had a greater effect than environmental factors on a deviation from the population mean”

          The same would apply to h^2 SNP. So the question was: “Was the GxE such to preclude such ranking?”

          You said: “If the heritability of the trait’s heritability increases, then we have different sensitivities to nutrition in our chosen population. That is G x E. However, the interaction has limitations.”

          This is just nonsense. You’re confusing issues. At a given time in a given population variance component varG x VarE/C has some value. If it’s modest to low then hypothetical developmental sensitivity interactions don’t obstruct the ability to partition variance and so on. You can predict IQ fine based on genetic similarity.

          So what’s your point? I think that your trying to escape into the possible worlds argument. I already discussed ad nauseam.

          You: I ALREADY cited this paper and explained why this explanation is unconvincing, in terms of a generalization and also consistent with their being inconsistent g x e in the literature. In either case they DID NOT find non-linear h^2. The effect remained constant.

          Ridiculousness. I see what’s happening, now. The only “evidence” you have for substantial GxE interaction, specifically of the shared environmental type — the type that’s gobbled up by varG — are a couple of studies that show h^2 x SES. So you have to adamantly defend this tenuous link. So to be clear: no h^2 by SES is not necessarily a form of h^2 x c^2 and the studies have been inconsistent and all over the place.

          You: “GE correlation is just a way of explaining GE interaction. It’s an organism’s propensity to seek out environments that it is sensitive to in a good way or avoid environments that it is sensitive to in a bad way. They need to be separated only on paper to avoid confounds, but it’s all describing the same phenomenon in reality.”

          Now, you’re drifting into silliness. Biometrically, no — and there are a number of important differences that are germane to this discussion, for example COVGE non-shared is gobbled up by varG; while GE interaction is gobbled up by environmental variance..

          Developmentally, they are crucially different where it counts. For example passive CovGE is causally environmental (genes are spuriously correlated)
          while active GE is causally genetic in the indirect sense. What interesting about GxE interaction is that it’s truly causally tangled.

          You: “I won’t presently comment on the B/W IQ issue, because that’s even more of a distraction. I clearly don’t agree with your assessment.’

          You brought up the issue, but fair enough.”

          We drifted into the issue.

          • Chuck says:

            You said:

            ““Look at it another way. What is the heritability of the trait’s heritability? If it is constant throughout, then there is no g x e. If it changes, there is g x e. Hence, a curved H/h2 indicates that heritability is heritable, which is genetically controlled sensitivity to environment,”

            Perhaps what you mean is “variability” in heritability estimates. Are you thinking (but not wanting to say outright):

            “Well, ok, I don’t want to get metaphysical like First Ypres and just posit possible worlds were phenotype can’t well be predicted by genotype and I’m not sure about
            actual A x E/C within given population, so I’ll try to sneak in an argument for A x E/C by way of variable h^2 estimates between populations. So, for example, I can point to the difference between this low estimate made in 1980 in the US and that high one made in 2010 in Japan and try to imply A x E/C from this is the sense that it might have been found if one looked at a meta-population including 1980 US and 2010 Japan siblings.”

          • Revolver says:

            There is nothing metaphysical here. At this point I must just be failing to correctly communicate to such a great extent that you’re here reading my thoughts. Let’s just call it a day. Appreciate the discussion.

        • Revolver says:

          ‘We agree then that h^2 interactions per se don’t entail GxE interactions’

          h^2 interactions as such in a properly stratified population do entail G x E interaction. We disagreed on what was meant by “stratified.”

          ‘So, I think that it’s more complex than what you suggest’

          It simply involves expanding the model to account for the genetic effect attributable to the moderator and doing the same thing — stratification.

          ‘Talk of the heritability of heritability is nonsense’

          No it is not. This represents the heritability of the effect of genes. Genetic sensitivity to the environment is the heritability of genetic effect.

          ‘thus ranking’

          To SAY a genotypic IQ of 110 IS TO rank genotypes and assume they are ranked.

          ‘This is just nonsense’

          No it is not. Here, let me clarify — by limited I mean that the interaction over P varies. The surface varies. It’s smoother in some places and the approximations there will be good.

          ‘So to be clear: no h^2 by SES is not necessarily a form of h^2 x c^2 and the studies have been inconsistent and all over the place’

          There’s no “possible” worlds. The fact that these studies are inconsistent, with several showing curved h^2 AS A, shows that in some populations there are differential sensitivities. In other populations, less so. This is consistent with the rest of biology.

          ‘Developmentally, they are crucially different where it counts’

          No they aren’t. Passive describes a risk factor — g x e. Evocative describes a feedback loop between an individual and his or her environment — g x e. Active describes a propensity — g x e.

          ‘Biometrically, no’

          There is a difference between model and reality. Like I said, both describe the same phenomenon regarding the surface of P.

          As I said before, we are both wasting our time. I apologize.

          • Chuck says:

            You said: “h^2 interactions as such in a properly stratified population do entail G x E interaction. We disagreed on what was meant by “stratified.”

            I’m not interested in debating points, just clarity. And while you don’t seem to think so, I found this discussion to be somewhat informative. I had a much fuzzier sense of the issue, prior. You forced me to do some reading and reflecting. Now to make progress, Let’s try to figure out the major remaining areas of disagreement.

            1. I see your point about properly stratifying. Do you see mine that (a) this is often not done and (b) this means that we need to be cautious when interpreting h^2 variability?

            You said: “No it is not. This represents the heritability of the effect of genes. Genetic sensitivity to the environment is the heritability of genetic effect.”

            hmmm…perhaps you could clarify. I anticipated that you were going to make an argument, silly in my opinion, from between population variability in h^2 to A x E/C– but, perhaps, you were intending something else. I agree that “genetic sensitivity to the environment” is a genetic effect in the sense that it’s under potential selection. This is well recognized.

            “The very fact that G x E exists at all means that sensitivity or reaction to the
            environment is itself under genetical control, and may consequently be altered by
            selection just as selection may change or stabilize any other aspect of the phenotype
            which. is under genetical control”…..Although we sometimes include G x E with the non-heritable agencies, for predictive purposes we should recognize that the existence of G x E mdicates that sensitivity to the environment is to a greater or lesser degree under genetical control. This enables expectations for G x E to be written in terms of gene effects. Furthermore, the existence of G x E indicates that sensitivity to the environment is potentially subject to the influence of selection. To affirm, as many have done, that the certainty of genotype-environmental interactions undermines genetical analysis, is both to exaggerate their significance relative to the overall variation in a population and to miss their potential biological significance.:

            https://occidentalascent.files.wordpress.com/2014/11/eavesjinks1977.pdf

            I don’t see the point, though. Isn’t the question: Are we so selected that in the local or realistic global range A x E/C makes genetic predictions untenable? Articulate.

            You: “To SAY a genotypic IQ of 110 IS TO rank genotypes and assume they are ranked.”

            Good. We agree on this point.

            You: “By limited I mean that the interaction over P varies. The surface varies. It’s smoother in some places and the approximations there will be good.”

            Agreed. (And my point was that you can still (locally) probabilistically rank so long as the surface isn’t to0 warped.)

            You: “The fact that these studies are inconsistent, with several showing curved h^2 AS A, shows that in some populations there are differential sensitivities. In other populations, less so. This is consistent with the rest of biology.”

            Disagree. Let’s put aside the contentious h^2 x SES studies for now. Let’s just note that in a 14 study sample some A x E interactions were found and ignore the fact that A x C weren’t. As noted prior, if interactions were all over the place, which they were, this might indicate either unsystematic (and therefore unpredictable and hard to detect) true A x E/C or false positive A x E/C due to measurement error. I don’t know why you expect unsystematic A x E/C — or why you would expect me to be over concerned about the possibility. Here is a quote from Eaves et al:

            ” The practical importance of such systematic interactions is that they allow a measure of prediction about the relative efficiency of the same degree of environmental manipulation at different points of the scale of measurement….From another viewpoint, however, the demonstration of widespread G x E involving cultural factors, especially if such interaction were of the unsystematic variety, would argue against any general procedures for the environmental modification of behavior, since individual genotypes would respond in quite unpredictable and specific ways to changes in their cultural environmental.”

            Would you agree that such GxE would be as epistemically and practically problematic for interventionist environmentalists? And would you agree that they generally are not concerned? If so, why should I be more concerned? Or do you pester them too? I ask because I am trying to take your critique seriously and I can’t if it comes across as too one sided.

            Now, that asked, I don’t think that this is a problem for h^2 estimates. Eaves et al. note the appropriate method:

            “Thus, although G x E may pass undetected. if it is completely unsystematic, and
            although G x E will bias our estimates of genetical and environmental variance in such
            circumstances, it will none the less bias equally our estimates of DR and Es in the same
            direction. In the light of this discussion, therefore, it seems appropriate to compute
            the estimate of the (narrow) heritability which would be obtained if a simple model were
            fitted in which the contribution- of G x E was mistakenly ignored, and to compare the
            value with that computed from the true parameter values. We do this, not because we
            believe a knowledge of ‘heritability’ is fundamental to the problem of individual
            differences, but because certain authors (e.g. Moran, 1973; Layzer, 1974; Feldman &
            Lewontin, 1975) have focused much of their criticism of the analysis of human
            behaviour on estimates of ‘heritability’. In writing an expression for the ‘true’
            heritability, given G x E, we include any variation due to G x E in the denominator only.”

            One can still simply treat GxE as non-genetic. So the only potential problem is with our ranking.

            That said, to discuss further, we would need:
            (a) to determine if there is a way to distinguish between true but unsystematic A x E/C and false positive A x E/C
            (b) if not, what the range of the specific interaction components were

            Edited: (I will have to look over the figures below later.)

            2012

  10. Meng Hu says:

    Those who argue that GxE interaction explains a big chunk of IQ variance are certainly poorly informed. For what I know, the evidence for this claim is weak. I have an article in which I talk about that, right here. If the article is too long, just type CTRL+F and “GxE” (without marked quotations).

    So, in short, van Leeuwen et al. (2008, pp. 78, 85-87) report large effect of GxE interaction (due to nonshared environmental influences), -0.30, which is different from the small effect reported by Jensen (1973, pp. 173-174), Plomin et al. (1988, pp. 240-249) or Finkel & Pedersen (2001). Hoekstra et al. (2007, p. 112) found no strong GxE interaction with regard to verbal and nonverbal IQ.

    I have also noted :

    Unlike heritability-by-IQ interaction, the research on heritability-by-SES and GxE interactions produce conflicting results. Thus the question remains unresolved. Molenaar et al. (2013) explain that the differences in the GxE interactions (where the environmentality is unmeasured) may be due to differences in the methods, e.g., tests used, their numbers, length, reliability, and domain measured, verbal, nonverbal, others, or full IQ, but also (un)representativeness of samples, notably at the extremes, which will cause data non-normality. A more serious threat is the portion of measurement error included in the non-shared environmental component, and reducing the former will likely converge towards more consistent patterns.

    In the Molenaar’s meta-analysis, there is no clear evidence of such GxE interaction, but they say that the variability between the samples is so large that aggregating the data sets makes no sense. Dolan (by email) told me that the scaling issue (any trivial source of non-normality) dominates the issue: any source of poor scaling will result in non-normality.

    Also, “Revolver” seems to mistakenly interpret GxE as SES-heritability interaction. I also made that mistake before. However, they are close concepts but still different things. The study of GxE is about an unmeasured (nonshared) environment. Refer to Molenaar’s article. And Chuck already answered that here.

    More generally, concerning the subject of interest which was GxE vs G+E. Here’s what James J Lee (2010) said : “the simplest additive model predicts that first-degree relatives should be half as similar as MZ twins, and this prediction does not seem far from the truth”. For what I see from his table 1, the numbers show no strong evidence that rMZ>2rDZ or that rMZ<2rDZ. Lee also noted :

    A large genetic variance in the absence of additive genetic variance is a rather peculiar case even in theory. Consider a trait influenced by a single locus with two alleles. In order for the genetic variance to be completely non-additive, there must be equality between the means of the two homozygotes and also the frequencies of the two alleles. In the absence of such special disordinality and symmetry, substantial additivity must be the rule. … Even in the presence of substantial non-additive gene action, population-genetic theory predicts that most of the genetic variance in a polygenic trait should be additive in nature. Because random fluctuations in allele frequency will lead eventually to the loss of one allele, the long-term expected frequency of a mutable, weakly selected DNA variant in a population of small effective size is very near either zero or one. The rarity of one allele at many loci tends to prevent the kind of symmetrical situation leading to non-additive genetic variance. For example, even if a given pair of loci show a strong non-additive interaction, a low frequency of an allele at one locus means that an allelic substitution at the other occurs against a nearly uniform genetic background and thus exerts a predictable effect.

    Generally, what I said has been covered in Chuck’s comment.

    When Revolver says that “the model is set up has G capturing all G x E effects” I don’t understand the sentence, because he is not explaining what he meant by that. Recall that the common way of detecting GxE has been described here (pages 173-174). Also, he continues saying GWAS studies show most genetic variance is due to non-additivity, without any need to provide the proof. Here’s Trzaskowski et al. (2013) : “As noted above, our GCTA estimates of genetic influence account for 74–94% of our twin-study heritability estimates, which implies that most of the missing heritability can be found with additive effects of common SNPs”. Stop the stubbornness.

    When Chuck says “h^2 of IQ is lower in first generation Hispanics than second generation ones because the first gen ones don’t all speak the language of the test” and Revolver replies “Seems to reflect measurement error, in that case” it is not only wrong and an ad hoc argument, but it is more accurate to say that this is the consequence of equating the groups in terms of environments. If IQ is biased against 1st generation hispanics due to language barriers and other environmental factors, the remaining variance (after removing environmental variation due to language bias) becomes more genetic.

    I also recommend Revolver should stop saying stuff like “Probably because the “reared apart” studies don’t tend to do a good job of quantifying “apartness.”” when he doesn’t care to provide any explanation and proof of what he says, because it renders his comments extremely difficult to read. Less mystification, and more clarification, please.

    Finally, regarding the issue of differential h2-SES with respect to within and between-groups, you can think about the ecological fallacy and individualistic fallacy. The former generalizes group differences to individual differences, while the latter does just the reverse. I have lot of examples where you can find correlations in between-country analysis, but in within-country analysis, no such correlation holds when examining trend over time. For the equal h2 of IQ between blacks and whites, I can assume that the higher h2 in higher SES is true within the white and black populations, but not when you compare blacks with whites. If you think otherwise, you’ll be making the so-called individualistic fallacy.

    There are two things I haven’t covered here : sensitivity with respect to GxE and ranking genotypes problem. The reason is because I don’t understand what you are talking about.

    • Revolver says:

      ‘the model is set up has G capturing all G x E effects’

      G + E. If G increases as a function of E, it will be represented simply by G.

      ‘Also, he continues saying GWAS studies show most genetic variance is due to non-additivity, without any need to provide the proof. ‘

      I said that the simplest explanation to reconcile the extremely low variances explained by particular alleles and the high heritability estimates was non-additivity.

      ‘Probably because the “reared apart” studies don’t tend to do a good job of quantifying “apartness.”” ‘

      They treat it as a binary variable 1/0 rather than quantifying the degrees of apartness.

      ‘Seems to reflect measurement error, in that case” it is not only wrong and an ad hoc argument,’

      A) it isn’t wrong. If you administer an IQ test in English to spanish speakers, you will not get an accurate measure of IQ.

      B) I already explained this. Incidentally, these “scenarios” that were dreamed up could show g x e interaction.

      ‘The reason is because I don’t understand what you are talking about.’

      The ranking problem assumes that one genotype in all environments gives the highest possible trait value before we consider environment. It also ignores differential sensitivities to the environment within the population.

      ‘the simplest additive model predicts that first-degree relatives should be half as similar as MZ twins, and this prediction does not seem far from the truth’

      Meaningless.

      ‘Stop the stubbornness.’

      I realize now that this is a waste of time. I apologize for wasting your and Chuck’s time as well. I actually only wanted to hear more of the interlocutor’s thoughts. Have a nice day.

    • Meng Hu says:

      Revolver, you say “G + E. If G increases as a function of E, it will be represented simply by G.” and I’m not sure what you are talking about. In the presence of rGE, such effect is represented by G, but in case of GxE, this effect is treated as E.

      You also say “I said that the simplest explanation to reconcile the extremely low variances explained by particular alleles and the high heritability estimates was non-additivity.” but you ignore other possibilities. For example, Yang et al. (2011) wrote :

      For most traits, the associated SNPs from GWAS only explain a small fraction of the heritability. There has not been any consensus on the explanation of the “missing heritability.” Possible explanations include a large number of common variants with small effects, rare variants with large effects, and DNA structural variation. We recently proposed a method of estimating the total amount of phenotypic variance captured by all SNPs on the current generation of commercial genotyping arrays and estimated that ~45% of the phenotypic variance for human height can be explained by all common SNPs. Thus, most of the heritability for height is hiding rather than missing because of many SNPs with small effects.

      And that’s what most scientists actually believe. If you think they are wrong, you have to explain why.

      You say that apartness, in twin studies, is treated as binary variable. And I think you should read this. When you say “If you administer an IQ test in English to spanish speakers, you will not get an accurate measure of IQ.” you are obviously showing you don’t understand the difference between random and systematic measurement error. The first is what we usually call measurement error, and the second is what we call psychometric bias. You referred to the second, although you thought you were talking about the first thing. You consider Lee’s comment meaningless, but you don’t even care to explain why. That’s not an argumentation.

      • Revolver says:

        ‘but you ignore other possibilities…’

        It is only a problem so long as we are ranking genotypes versus all environments, and consequently, if the causes of heritability are the same for all environments. However, if different genes and different interactions produce the results, the problem goes away. The rest of biology operates this way. It’s the simplest explanation.

        ‘And that’s what most scientists actually believe’

        You cite a passage stating that there “is no consensus” and positing an explanation as evidence that “most scientists” believe that explanation.

        ‘You are obviously showing you don’t understand the difference between random and systematic measurement error.’

        What a meaningless comment, to admit that they are both types of measurement error and then quibble with my “failure” to use words that you wished I would have used, then to argue that me correctly terming the phenomenon is evidence of my “failure” to understand that there are different types of measurement error. Twice in one post you use evidence that contradicts your proposition while insulting me; like I said, this is a waste of time.

        • Chuck says:

          None of this matters. In table 6 of the 14 study analysis, the average E + ExA variance component is 0.17 (i.e., 1- A+C ). Of that, based on the reported parameters, ExA is only a fraction of that. In the extreme cases, it looks to be 50% of the E variance; in others, it’s ~0%. This degree of ExA is relatively trivial given our concern. If GxE is ubiquitous it is largely cryptic. Since it’s at best cryptic, it’s not attenuating much the relation between genotype and phenotype in existent populations. You then are forced to make the possible worlds argument. Maybe possibly we could find an environment in which phenotype couldn’t well be predicted by genotype. Maybe — but since such hasn’t been found yet, there is reason to suspect that it will not be anytime soon and to ignore the possibility until it is.

      • Meng Hu says:

        What’s interesting with the implication of GxE interaction research is that these people all expect lower genotypes to be more responsive to improved environments. See Rowe et al. (1999, p1159) for a discussion of this. For what I know about preschool intervention programs, it is not clear at all that education improves intelligence. That was true for poor people in the white and black groups. I can’t see the interaction in these effects. I don’t see whites improving less or more than blacks. Both groups show null effects in g gains. You can say that the poorest environments in USA were not enough for showing these effects, and that we need even poorer environments, such as in Africa. Well, this is shifting from explaining the 1SD black-white gap in the USA to explaining the 2SD black-white gap between USA/Europe and Sub-Saharan Africa.

        Furthermore, if you don’t even acknowledge your error, after what I said about the difference between noise and bias, I think you’re definitely beyond salvation. I’m also curious to know “your” definition of “insulting people”.

        • Revolver says:

          ‘Both groups show null effects in g gains’

          This is far less impressive than you believe it to be. It’s just orthogonalization.

          Also, systematic error and random error are both types of measurement error. Your comment is as ridiculous as saying that me referring to a boy playing as a “child” playing proves that I “obviously” have no understanding that boys are different from girls. What’s bizarre is that you concede my correctness and how asinine the entire point is in referring to them as “random and systematic measurement error.” And more bizarre is that you continue to insist that my correct but general description is at once an error that “proves” I am beyond salvation.

        • Chuck says:

          Could both of you refrain from derogatory comments and undiplomatic tones, please. They emotive discussions and this precludes rational deliberation. Take a step back, identify areas of disagreement and reformulate your positions. Thanks.

          Also, try to keep on discussion. If you want to discuss the B/W gap, perhaps you could do that on another thread e.g.,

          http://occidentalascent.wordpress.com/2012/06/10/the-facts-that-need-to-be-explained/

          • Revolver says:

            I will respond regarding a curved h^2. VarG represents the extent of the genetic effect on the trait. VarGVarG represents the extent of the genetic effect on the extent of the genetic effect.

            A steady VarGVarG implies equal sensitivity to the environment. Changing VarGVarG implies different sensitives to the environment. That is G x E.

            Stratification that detects this effect lets us know over how wide our intervals can be for the model to remain accurate.

            You say that unsystematic G x E is cryptic. This is untrue. It tells us about the applicability of our h^2 estimates.

          • Chuck says:

            “You say that unsystematic G x E is cryptic. This is untrue. It tells us about the applicability of our h^2 estimates.”

            I cited a 14 study analysis that directly, not indirectly, looked for G x E. The G x E found was unsystematic and the authors suggested that it could be due to measurement bias. Importantly they found zero G x E of the A x C type, the types which gets lumped into A estimates when not modeled for. Importantly also the magnitude of the G x E they found was trivial — a fraction of the E components which itself was only 0.17. I noted that this magnitude of possible G x E does not complicate ranking — thus, if you want to make an argument you have to make one similar to that of your abusive friend First Ypres/ HughCipher/jorge videla/Mugabe. You have to make a possible worlds one. I noted that such an argument would apply equally well to environmental rankings. Since you never never know if extensive cryptic G x E will manifest, you can’t rank environments, saying some are better than others and thus all environmental intervention based research is meaningless. This is SILLY. No one takes it seriously because we don’t live in the possible world but in this one. What do you disagree with? Please be clear because I very much dislike discoursing with people who don’t commit to or concede positions.

            To be clear, I understand First Ypres/ HughCipher/jorge videla/Mugabe ‘s argument but I take it about as serious as the average environmental interventionalist does. Please explain why we, meaning both naturists and nurturists, should do otherwise, why we should write off the possibility of improving IQ or other such traits through either genetic or environmental means because of the possibility of possible developmental sensitivity G x E becoming manifest. If you can’t explain why, note this so that we might move on.

          • Revolver says:

            No one has said that h^2 is useless, which is what you seem to be arguing against. Again, if you have different individuals in the population with different sensitivities to the environment it will make your h^2 less accurate.

        • Meng Hu says:

          Systematic measurement error is better (and should be) called measurement bias. Not “error”. I know of no one else who think of “bias” when writing “measurement error”. But you can invent whatever definition that suits you, I don’t care.

          Also, I didn’t want to talk about B-W gap. It just happened that it provides a good illustration of what I’m trying to say.

          Revolver is just like all the other guys who worship GxE interaction and GE correlation. When they refer to these things, they (almost) never talk about the implication of these “nonlinear” models. For rGE, people fail to understand that the model makes sense only if we can assume that the environmental effect is causal. But this causality is rarely demonstrated in the discussion section of most (if not all) papers. They don’t even want to talk about the consequences for the rGE model. When it’s found that the intensive intervention fails to yield meaningful and sustainable IQ gains, some rGE theorists (e.g., Dickens & Flynn 2001) argue it’s due to such rGE effects, but this is pretty much irrelevant when the question is “what can we do to boost IQ”. For GxE model, researchers believe that public intervention is needed and can be useful for the poorly educated children. And yet, I can’t see the positive effect predicted by all these models in terms in cognitive improvement for the poor children living in advanced countries. What these models predict is not supported by experimental studies.

          Generally, the debate on heritability ultimately is one which must lead to malleability. As I said, most researchers don’t talk about malleability so much. When they do, they cite the Abecedarian and, rarely, the french adoption studies of Capron/Duyme; both of which provide only a weak/meager evidence for the environmental hypothesis, especially the GxE model.

          Unless I missed something important here, I will continue to say that I don’t understand why no one seems to bother about talking heritability from the perspective of malleability.

          • Chuck says:

            “Revolver is just like all the other guys who worship GxE interaction and GE correlation. When they refer to these things, they (almost) never talk about the implication of these “nonlinear” models”

            For a while I thought that it was important to refute the GxE/ COVGE arguments. But then I realized that they were motte-and-bailey doctrines. Strong claims are made such as “because of GxE you can’t predict outcomes/biometrics is worthless”. It’s then pointed out that you can predict them quite well, given the existent range of environments. The clams are then modified, walked back to, to “strong developmental sensitivity GxE could exists which could become manifest therefore you can’t trust predictions”. It’s pointed out that this possibility worlds argument applies to all scientific predictions (e..g, Hume on causality) and just as much to environmental ones as genetic ones. There then is either silence or scientific nihilism and an end of discussion. But after it’s ended the strong claims re-emerge “because of GxE you can’t predict outcomes/biometrics is worthless”. Similar happens with COVGE. When a practically insignificant amount is detected and when this is found to behave just as G (i.e., active COVGE) the argument is walked back. When the discussion is over, the old strong claim re-emerges.

          • Revolver says:

            ‘The child who is a boy is better (and should be) called lad. Not “child”. I know of no one else who think of “lad” when writing “child”. But you can invent whatever definition that suits you, I don’t care.’

            Pedantism at its best or worst.

            ‘I don’t understand why no one seems to bother about talking heritability from the perspective of malleability.’

            That can be implied from the extent to which the actual extent of genetic effect is inherited.

  11. Chuck says:

    Revolver: “No one has said that h^2 is useless, which is what you seem to be arguing against. Again, if you have different individuals in the population with different sensitivities to the environment it will make your h^2 less accurate.”

    Unless I’m missing something, (for a give sample) the h^2 will be accurate so long as the G x E is of the A x E sort, since this gets treated as environment. But, if I understood the Tal (2012) paper correctly, the probabilistic ranking will nonetheless be less so because of the curvature — is that the correct term? — introduced into the genotype-phenotype plane.

    I’m not sure that it’s meaningful to speak of a cross population accurate h^2. What you might say is that because of GxE — but in what sense? –the h^2 of a given population will less accurately predict the h^2 of another or of the theoretical meta-population. Ok, but so also with variance in phenotypic variance or test construction or sample size, etc. I don’t have expectation that the h^2 in one population would be the very same as in another. I highly doubt that the h^2 in e.g., Rawanda would be the same as that in Japan.

  12. Chuck says:

    Cease diarrhea-ing on my blog; I’m not going to keep sifting through your hershey squirt comments.

    Let’s deal with one issue at a time. Do you agree that gene x environment and environment x heritability interactions are conceptually different? The existence of heritability x environment interactions just means that heritability estimates are population specific — a point which everyone agrees on. Now, as argued by Lewontin (1976?), high non-additvity of a trait should correspond with high locality (poor generalizability) of h^2 estimates. It doesn’t follow however — rather, this is affirming the consequent — that environment x h^2 interactions presuppose biometric GxE ones, since h^2 is G/P; just agree or disagree with this local point and then we will move on.

  13. Chuck says:

    So, you agree that you were wrong and that you really don’t know what you’re talking about. Probably just read a couple of papers by Lewontin. Thought so.

  14. Revolver says:

    You’re obviously just censoring him….
    A lot of your rejoinders are assertions about how g x e can be easily or readily modeled without explanation.

  15. Revolver says:

    ‘Sounds metaphysical. Refer back to points 1 and 2. Perhaps we live in a tiny range of the metaverse were the physical laws are regular’

    It’s not metaphysical at all. The idea is that the model that most closely fits norms of reaction would not be a plane; it would be a curved surface. The G + E model therefore would simply approximate a small region of this surface, meaning that it would only be accurate across a certain limited snapshot of genetic variation and environmental variation.

    ‘This, of course, is statistically detectable. And the variance for a trait in a given population owing to GxE can readily be quantified. For IQ it has been found to be trivial. Interactive models simply don’t fit the data well.’

    This is the assertion I’m talking about. I’m not sure how current models will successfully detect G x E interaction, if current models come laced with the presumption that there is ONE SET ranking of genotypes.

  16. Chuck says:

    That happens when you call the admin a moron, imbecile, and retard and, worse, spam a bunch of micro non-replies. If he posts something rational, I’ll let it through. If you have a question, I will answer it — unless it’s overly silly. Generally he was confusing issues. The locality of heritability is not the same of G x E interaction. Imagine a simple additive model: p = 10, G=10, E=0. H^2 = G/P = 100%. Now increase environmental component and with it the total variance. P=100, G=10, E=90. H^2 = 10%. This is what Turk. (2003) found, roughly. But it’s not G X E interaction, as out friend imagined. There is little evidence for biometric G X E in the case of intelligence. This is often tested using Jinks & Fulkner method but also newer ones. References: Plomin et al. 1988; Cattell 1982; McGue 1989; Plomin 1990; Brody & Crowley 1995; McGue & Bouchard 1998. This is just silly. My interlocutor replies, though: statistical power. But, you only need massive sizes to detecting very small G x E; the fact that it hasn’t been detected — or when it has that it is of little significance — means that it doesn’t account for a large portion of the variance give the range of environments for the tested samples. This includes, for example, the whole of Scotland in the early 1900s.

  17. Revolver says:

    [Admin: I condensed replies]

    ‘The locality of heritability is not the same of G x E interaction’

    No, one causes the other is his point. G x E interaction makes the surface curved and the effect of G x E interaction grows as both genotype variation and environmental variation grows. Thus, the G + E = P model is only an accurate approximation of the surface over small variation in both.

    ‘Imagine a simple additive model: p = 10, G=10, E=0. H^2 = G/P = 100%. Now increase environmental component and with it the total variance. P=100, G=10, E=90. H^2 = 10%.’

    Yes that is an obvious consequence of the equation.

    ‘But, you only need massive sizes to detecting very small G x E; the fact that it hasn’t been detected’

    Yes, that is because the model itself only tells us the effect variation has on a trait, and from that, how much of the trait’s variation is associated with variation in genotype. If G x E interactions exist, the current model swallows them up in G. Generally, the studies I have seen (esp of twins) do not meaningfully vary the environments.

    For example, GWAS studies are finding minimal additive effects from the SNPs associated with genotypic variation. Some simply dismiss these results for various ad hoc reasons, but these results are actually very consistent with earlier heritability estimates. The effects from these SNPs are probably not additive. That is the simplest explanation — and that explanation reconciles the data.

    Well, the reason it hasn’t been found is because the way that the model is set up has G capturing all G x E effects. G x E is more a measure of causation, whereas the model only estimates based on variation. The assumption in the model is that that variation represents contributions to an additive model — G + E. So again, a trait could have 100% heritability — in the “associated with geneotypic variance” sense — and still be 100% controlled by G x E.

    ‘It’s possible that in a broader range of environment, it would show up’
    ‘But it doesn’t occur to any practically significant degree in the known range, so it is not an issue’

    If you had a broader range of environments, I’m not sure how this model would detect or capture G x E without subsuming G x E into G.

    *effects on the phenotype are probably not additive is probably a more accurate way to say this.

  18. Chuck says:

    “It’s not metaphysical at all.”

    I mean that substantial G x E has not been found for IQ. It has, though, for other traits, which indicates that the statistical methods for detecting the phenomenon are sound. It’s possible that in a broader range of environment, it would show up; it’s also possible that in such a range times runs reverse. A lot of things are possible. Maybe ankylosauruses still exist on the earth. Maybe. But it doesn’t occur to any practically significant degree in the known range, so it is not an issue. You can now estimate variance components based on whole country cohorts — thus greatly expanding the environmental range — it’s still not showing up. The inability to detect non-trivial G x E (in this case) implies that other norms of reaction are not lurking around the corner, just as the continual inability to find the Loch Ness monster implies that it also is not. I fail to see why this is difficult.

  19. Chuck says:

    “The G + E model therefore would simply approximate a small region of this surface, meaning that it would only be accurate across a certain limited snapshot of genetic variation and environmental variation.”

    So, as I noted, there are multiple arguments.

    1) the current environmental range, for those on which h^2 estimates are made, does transverse different norms of reaction; G x E is known to explain a significant chunk of IQ variance.
    2) the methods for detecting G X E are not sufficiently robust; in the real world G x E explains a significant chunk of IQ variance in the samples were h^2 is estimated, but it isn’t detected.
    3) if h^2 estimates were based on samples from an expanded range of environments, G x E would be detected because then different norms would be crossed

    No for (1) and (2). (3) is cryptic.. If bigfoots were plentiful there would be more evidence of them; they would more frequently be run across. they aren’t, so they must be very uncommon if they exist at all; thus practically they are not much of a concern. Where do we disagree?

  20. Revolver says:

    For example, GWAS studies are finding minimal additive effects from the SNPs associated with genotypic variation. Some simply dismiss these results for various ad hoc reasons, but these results are actually very consistent with earlier heritability estimates. The effects from these SNPs are probably not additive. That is the simplest explanation — and that explanation reconciles the data.

  21. Chuck says:

    [Admin: a commenter reminded me that, for clarity, P,G,E,and GxE should be prefixed by Var, when discussing variance components]

    “No, one causes the other is his point. G x E interaction makes the surface curved and the effect of G x E interaction grows as both genotype variation and environmental variation grows. Thus, the G + E = P model is only an accurate approximation of the surface over small variation in both.”

    He was arguing that ExG/P is evidence of ExG. I noted that he was affirming the consequent. That is, you tend to get E x G/P if you have a lot of ExG, but the former does not presuppose the latter. Thus, the former can’t be used as proof of the latter. I noted, for example, Turk. et al. (2003) who found additivity as a best fit yet substantial ExG/P. Do you agree or not? This relates to my point that there is no evidence for substantial GxE in the case of IQ (a lot of negative results), though it has been found for other traits.

  22. Revolver says:

    [Admin: I added my replies]

    ‘G x E explains a significant chunk of IQ variance ‘

    Let’s hone in on this right here. Heritability does not explain variance. Heritability tells us that X amount of genetic variation is associated with X amount of Phenotypic variation. When you say explain, it does seem like you mean ’cause.’

    [Chuck: G isn’t heritability; G is genetic variance; heritability is Genetic/Phenotypic variance; “explain” can either mean statistically or causally explain. In context to biometric decomposition it means “causal”, in the sense that you are decomposing the cause of population variance into genetic, environmental, gene x environment, Cov(gene,environment), etc. components. Kinship studies are considered to be causally informative natural experiments.]

    Like I’ve said. The main problem here is an additive model that itself is only indirectly assessed by measuring variance.

    [Chuck: I don’t know what you mean here. Non-additivity, defined as a statistical interaction between G and E, is assessed by looking for a statistical interaction between G and E; why would one do otherwise?]

    ‘if h^2 estimates were based on samples from an expanded range of environments, G x E would be detected because then different norms would be crossed’

    Maybe, but also maybe not. It would depend on the genotypes you used. You’d need several different groups from differing environments with the same a) general genotype and b) general phenotypic value relative to their environment. Then, you would need to mix them all up and match them in random environments. Then if the environments are normally distributed, we make no assumption about the group’s environment (i.e. at the center of the norm), and we assume that each group represented one general genotype ranking, we still wouldn’t be able to aggregate the results — some groups would have been put in environments that lift their trait values while others would have been put in environments that depressed their trait values. So we’d have to simply note the average difference in trait value for each group. That’d be the way to test for G x E.

    [Chuck: This is ridiculous. Biometric GxE refers to a population, not individual, level statistic. It refers to how much trait variance is (causally) explained by the interaction between genes and environment. When GxE is low it simply means that not much of the variance in the population is caused by this. While there is a probablistic relation between population parameters and individual ones, generally, the question of whether GxE is significant for a particular individual is separate from whether it is significant for a population. Maybe try to think about the matter in another context. Most people would agree that better education and good family increase at least some cognitive domains. Most people assume that the effects are additive, but in principle they could be interactive in weird ways. Perhaps better education is deleterious for people raised in nice families. Perhaps, but in fact no such statistical interaction has been found. Thus this is not a concern for policy makers. However, for a given, very atypical individual, it might be — but this is another matter.]

    Heritability studies tend to feature multiple genotypes across diverse (though not very) environments. If we aren’t even set up to gather the evidence, it’s not surprising that we wouldn’t run across the evidence.I point out the lack of environmental diversity to suggest potential flaws in associative variance, not to add to the G x E critique. Two different issues — external to model vs within model.

    [Chuck: As noted, GxE has been found for other traits using the same methods and the same types of samples. So, you have to agree that it could have been found if it was like that for other traits e.g., Schizophrenia. Also, I don’t know what mean by “not very” diverse environments. A number of bio-metric estimates are based nationally representative samples. They are also conducted in developing countries, in which there are massive socio-environmental differences, such as China.

    On another note, I should point out that contrary to the original interlocutor’s claims, the presence of substantial GxE does not undermine the meaningfulness of heritability statistics. This point was well articulated by Tal (2012). In the presence of strong GxE or Cov(G,E) one can simply express relations probabilistically Quote:

    “The framework outlined in this paper allows incorporating estimates of gene– environment interaction and covariance within a probabilistic interpretation of heritability (Tal 2009; see Tal et al. 2010, for an extension that includes a putative
    epigenetic variable). Specifically, given estimates of heritability and the variance components associated with GxE and rGE, a method based on the standard quantitative model generates the conditional probability that genetic factors had a greater effect than environmental factors on a deviation from the population mean. Previous approaches of reformulating heritability arise from the application of more complex mixed quantitative models (see Oakey et al. 2007, for a ‘‘generalized heritability’’ that incorporates pedigree information to form an extended pedigree model).”

  23. Revolver says:

    G/P is not h^2. G + E is a genetic component plus an environmental component equaling the trait value. H/h^2 is the variance in genetic component/variance in phenotype. They are not the same thing. H/h^2 is an indirect way of assessing G. You are attempting to unmask the underlying function by assessing variation.

    Over VERY SMALL variations in G and E, you may well approximate the effect of genes and environment by simply assuming additive effect from the variation. However, to assume the same over a wider range of genotypes and environments may be incorrect.

    Can you link me to Turk et al.?

  24. Chuck says:

    Yes, for clarity, I should have written VarG/VarP and for that matter VarGE but I made it clear elsewhere that I was discussing variance and since the context is biometric variance decomposition, this should be obvious. But maybe this is part of the confusion. I am discussion variance: GxE = VarGE, G=VarG, P=VarP, E=VarE. What have you been discussing?

    [It doesn’t change the point, though, since the original interlocutor was still arguing that E x h^2 evidences varGE, which it doesn’t.]

    [G x E in the developmental sense, of course, is trivially true; when people argue about GxE they mean in the statistical biometric sense, thus VGE.

  25. Revolver says:

    ‘[Chuck: G isn’t heritability; G is genetic variance; heritability is Genetic/Phenotypic variance; “explain” can either mean statistically or causally explain. In context to biometric decomposition it means “causal”, in the sense that you are decomposing the cause of population variance into genetic, environmental, gene x environment, Cov(gene,environment), etc. components. Kinship studies are considered to be causally informative natural experiments.]’

    I should have showed my work. To talk about whether G x E explains the variance in IQ in the context of the additive model may fundamentally miss why that particular model cannot capture G x E. Heritability itself does not causally explain variance unless you assume the additive model — that is the problem.

    So it is not that h^2 lacks robustness….it’s that the G + E additive assumption/model may fail to capture G x E.

    ‘ This is ridiculous. Biometric GxE refers to a population, not individual, level statistic. ‘

    My groups above were populations that represented a particular genotype. Varying one group through different environments is not enough to detect G x E because it could simply be that the varying environments are either good or bad for all genotypes. Therefore, you’d need different populations representing different genotypes rotating through different environments.

    ‘Perhaps better education is deleterious for people raised in nice families. Perhaps, but in fact no such statistical interaction has been found.’

    What’s been found is only that genetic variation mostly associated with trait variation. Nothing more. So the fact that ‘no such statistical interaction has been found’ represents the fact that the model is not really set up to catch it.

    ‘the presence of substantial GxE does not undermine the meaningfulness of heritability statistics.’

    Of course it doesn’t. As I noted earlier — a trait may have 100% heritability and also be 100% controlled by G x E.

    ‘ a method based on the standard quantitative model generates the conditional probability that genetic factors had a greater effect than environmental factors on a deviation from the population mean’

    This still doesn’t seem to address the criticism. In fact, this just seems to be a way to have multiple h^2’s depending on the conditions, which would further underscore h^2’s limited utility in this endeavor.

  26. Revolver says:

    I have been discussing G x E vs G + E. G + E is under attack — not H/h^2.

    And it seems like you have misinterpreted his argument as E x h^2. He’s saying that significantly different h^2 with different samples (that are not so radically different in genotypic/environmental variation) suggests that there not a set genotypic ranking which suggests that the underlying function is curved rather than planar (for lack of a better word).

  27. Revolver says:

    As an aside…I probably haven’t delved into it to the depth that the “interlocutor” has…so you should uncensor him and let him defend his own point.

    Roughly though…with a set ranking, the surface may look like this

    without the ranking, the surface may look like this

  28. Chuck says:

    You: “I have been discussing G x E vs G + E. G + E is under attack — not H/h^2.”

    The standard biometric model (a) is (population mean) P = (mean) G + (mean) E + (mean) GxE + etc. Biometricians concern themselves with population variance, They decompose the variance in phenotype into (b) variances due to genes, environments, interactions, co-variances, etc, thus our VarP = VarG + VarE + VarGE + etc. which I lazily call P, G, E, GE**. One line of argument against VarG/VarP estimates is that they are confounded by other variance components, VarGE, COV(G,E), VarEpig. And this is where the biometric interaction argument comes in. The argument is either that VarGE is neglected (which is false, it’s just not detected) or that VarGxE is lurking out there in the population and will show up as VarGE in the next study or when the sample is more representative, perhaps even globally. Thus if you are attacking (b) VarP = VarG + VarE, you are attacking h^2. And if you are attacking (a) P = (mean) G + (mean) E you are attacking (b) VarP = VarG + VarE, because (mean) GxE is granted in theory, but when it is tested for***, it — and so VarGE — turns up as negligible.

    *phenotype attributed to differences in genes
    **you are right that we should distinguish between population means and variance. But we also shouldn’t get confused on the other end. Biometric analysis deals with variance around the mean. From this perspective, there isn’t GE independent from VarGE, since the concern is variance. .
    ***the traditional way to test for GE and determine VGE was to compare the mean and absolute score difference of MZ twins reared apart

    You: “And it seems like you have misinterpreted his argument as E x h^2. He’s saying that significantly different h^2 with different samples (that are not so radically different in genotypic/environmental variation) suggests that there not a set genotypic ranking which suggests that the underlying function is curved rather than planar (for lack of a better word)”.

    I asked for clarification. He said:

    “besides GxE has been found in four cases I can name. and has been found in innumerable cases whenever the sample h^2 varies significantly from one study to another, that is from one small region of (G,E) space to another.”

    As I said, this isn’t true. GxE is but one of several possible explanations for variation in h^2. Show me the evidence of significant GE/VarGE in the case of IQ.

  29. Chuck says:

    To keep things civil — myself also — I blocked a number of derogatory terms. If he can’t pass this simple civility test, he can’t post. I appreciate that those are high standards, but this ain’t a democracy

  30. Revolver says:

    ‘They decompose the variance in phenotype into’

    another additive effect model, yes. But as far as I can tell, the G x E model does not presuppose intersecting norms of reaction. Instead, it’s setup to describe special sensitivity to certain environments, so still seems to suffer from assuming an objective ranking.

    ‘Thus if you are attacking (b) VarP = VarG + VarE, you are attacking h^2. ‘

    No I am not. You are assuming that putting GxE in the context of another additive model is satisfactory. It may not be. If there is no set genotypic ranking, then I’m not sure how G + E + GxE sheds light on anything.

    Further, h^2 often assumes 0varGxE and focuses solely on varG/varP.

    The way they seem to suss out the effect of G x E (in the “sensitivity” sense) using H/h^2 is to stratify the population in one way or another and then assess H/h^2 on the basis of whatever stratification metric and look for the H/h^2 relationship among each “level.” Of course, this also comes laden with assumptions (equal variance of G at each level, smaller sample sizes in each population, etc.)

    But if there’s a sensitivity in one direction, I’m not sure why there wouldn’t be genotypic sensitivities in the other direction.

    ‘Show me the evidence of significant GE/VarGE in the case of IQ.’

    The simplest explanation of the GWAS results in the context of current heritability estimates is that the majority of genetic effects on the trait are non-additive and that there are intersecting norms of reaction.

    I also cited a recent study regarding G x E in the context of SES. You countered that this represented E x h^2 (I understand now you mean varE x h^2). Have we found G x E through any means other than E x h^2?

    We take a sample moderator — an environmental stimulus — and note h^2 in relation to that stimulus.

    Here is another for mathematical ability: http://link.springer.com/article/10.1007/s10519-010-9405-6.

  31. Revolver says:

    If SES is an example of E x h^2, and SES itself has high h^2, does this not suggest that the sensitivity has a genetic component?

  32. Chuck says:

    You: “The way they seem to suss out the effect of G x E (in the “sensitivity” sense) using H/h^2 is to stratify the population in one way or another and then assess H/h^2 on the basis of whatever stratification metric and look for the H/h^2 relationship among each “level.” Of course, this also comes laden with assumptions (equal variance of G at each level, smaller sample sizes in each population, etc.) …I also cited a recent study regarding G x E in the context of SES. You countered that this represented E x h^2 (I understand now you mean varE x h^2). Have we found G x E through any means other than E x h^2?”.

    I don’t see how VarE/E x h^2 can show developmental sensitivity GxE. Maybe you could explain. Just think of examples. h^2 of IQ is lower in first generation Hispanics than second generation ones because the first gen ones don’t all speak the language of the test. Is this really .G x E in the “sensitivity” sense, as you mean it? See my comments below.

    You: “The simplest explanation of the GWAS results in the context of current heritability estimates is that the majority of genetic effects on the trait are non-additive and that there are intersecting norms of reaction.”

    “Missing heritability” refers to one of two things: (a) lower GCTA than kinship heritability estimates; (b) low predictive power of specific alleles. I don’t see how (b) in absence of (a) suggests a developmental sensitive GxE model. Imagine that GCTA h^2 = 1, but specific alleles explain no variance. In this case a person’s phenotype can be predicted by the genotype of random members of the population. The more genotypically similar to smarter individuals, the smarter. In what way does this suggest G x E in the “sensitivity” sense?

  33. Chuck says:

    “Have we found G x E through any means other than E x h^2?”
    “If SES is an example of E x h^2, and SES itself has high h^2, does this not suggest that the sensitivity has a genetic c
    omponent?”

    …..

    Hmmm….Perhaps we are running into a semiotic problem here. When I use GxE I mean it in the traditional biometric sense of:

    “what constitutes a good environment for one genotype constitutes a bad environment for another genotype”
    or
    “environmental advantages have unequal phenotypic effects on different genotypes”

    Let’s call this Chuck’s biometric GE.

    An example: For most of the population (i.e., most of the genotypes) increased nutrition is associated with increased height/IQ/etc., but for some, it may be associated with decreased height/IQ/etc. Thus, one can’t speak of generally good environments or generally good genotypes.

    Other people use the term in a much broader sense. Some use it to refer to developmental interactions i.e., genes act with environment to produce traits. This is trivially obvious and is called developmental GE. Additionally, an increasing number of researchers, including some well respected behavioral geneticists, use GE in an inclusive biometric senses such that heritability x environment interactions are ipso facto a GE interactions. For example. Hofer and Shannahan, in “Social context in gene-environment interactions: retrospect and prospect”, tell us:

    “Broadly conceived social control refers to any social structure or process that maintains social order…Many biometric studies of heritability demonstrate a GE interaction that can be explained by social control model. These studies typically show that in settings marked by high levels of social control, h^2 attenuates, whereas in contexts marked by low levels of social control, h^2 increases. In other words, in circumstances marked by high levels of social control a large percentage of the sample — irrespective of their geentic diversity — exhibits the same phenotype; in settings marked by low social control, people choices and behaviors are more apt to reflect their genotype. Thus, social control mechanism reflect norms and other social forces that canalize (i.e., restrict variability in the phenotype of) genetically diverse people. As these canalizations forces increase (i.e., norms are more effective and choices are minimal), genetic differences are of diminishing consequences.”

    The authors give the example of the differential heritability of drinking alcohol in religious and non-religious Dutch families.

    This type of GE is different from what I am talking about, since we are not dealing with the effects of the same environment on different genotypes, but the effects of different environments on the same genotypes. Going back to the original example: For all of the population (i.e., all of the genotypes) increased nutrition is associated with increased height/IQ/etc., but some of the population experiences low levels of nutrition and some high — because of this, in those with very low levels, the association between genotype and phenotype is attenuated relative to those with very high levels.

    In retrospect, I can see why one might call this GE interaction. There’s a symmetry between the two situations. I just find the latter uninteresting. By this there is, for example, a “GE interaction” between PKU (genotype) and low phenylalanine diet (environment). Those not on a diet exhibit MR, those on none. Ok. But the point is that this is different from a situation in which some variants of PKU react positively to a phenylalanine diet while others act negatively.

    Now when you argue for substantial biometric GE interaction do you mean Chuck’s biometric GE or something else?

    I agree that there is often substantial GE interaction in this other, loose sense. Some environments are deleterious to the realization of one’s genetic potential. In fact, one of my arguments for a racial hereditarian hypothesis is that while low SES, at least in the US, tends to be associated with lower heritability (higher discordance between genotypes and phenotypes) being African American isn’t (which arguably suggests that African American IQ isn’t environmentally depressed.)

    I just don’t see much evidence for Chuck’s biometric GE in the case of IQ.

    Regarding the issue of ranking genotypes, Chuck’s biometric GE seems to be more of a problem given the prevalent conditions. Of course, an attenuated association between genes and phenotypes owing to a massive inflation of environmental variance complicates genotypic ranking, too. And one can readily think of situation where this is of practical significance. For example, as environmental circumstances improve, inbreeding, due to consanguineous marriage, in the Middle East/South Asia explains increasing amounts of malformities; thus, in a sense, these lineages are increasingly undesirable. There is then a cross temporal — and presumably spatial (Pakistanis in the UK) and class — ranking complication. Nonetheless, most people would recognize that a better/worse genotype is still better/worse even if the betterment/deleteriousness is only fully realized in certain environments. No one, for example, wants PKU even if it can mostly be treated.

    Concur that the issue is the degree of Chuck’s biometric GE and we can then proceed.

  34. Chuck says:

    “Other people use the term in a much broader sense. Some use it to refer to developmental interactions i.e., genes act with environment to produce traits. This is trivially obvious and is called developmental GE”

    I should have stated this differently. There is a really trivially true sense in which genes and environment interact to produce phenotype. After all our genes function in an environment, not a vacuum. Some people point to this and argue that therefore population variance can not be decomposed. This is silly. But then there is another sense which corresponds to what you mean by environmental sensitivity. Here we are saying that different gentoypes are sensitive to different environments. A classic example is Cooper and Zubek’s rat experiment, in which the performance of different rat breeds depended on the environment. This is related to Chuck’s biometric GxE in that it will manifest as a statistical interaction on the population level; it will show up as Chuck’s biometric GxE. For a discussion of the relation, refer to: Tabery, J. (2007). Biometric and developmental gene–environment interactions: looking back, moving forward. Development and psychopathology, 19(04), 961-976.

    Now, according to Tabery biometric and developmental GE are closely related in the sense that the latter evidences the former and the former manifests as the latter. A problem that we are running into is that these terms are used in other senses.

    Let’s call what you mean = developmental sensitivity GE and let’s call what I mean Biometric genotypes GE, where we emphasize that we are dealing with different effects of environments on different genotypes.

    As best I can tell, developmental sensitivity GE would lead to Biometric genotypes GE and so Biometric genotypes GE indexes the presence of developmental sensitivity GE. And then I would say that little evidence has been found of Biometric genotypes GE (for IQ) and, thus by extension, for developmental sensitivity GE. One way to test for Biometric genotypes GE is to correlate the mean scores of MZT (a measures of genotype) reared apart with there differences scores (a measure of environmental difference). This should show a significant correlation in the presence of Biometric genotypes GE, but such hasn’t been found.

    Now you keep pointing to h^2 x E interactions. And I know that Turkheimer and others call these GE interactions. But my point has been that these need not represent Biometric genotypes GE ~ developmental sensitivity GE. There seems to be continual disagreement on this point. I go back to the example of the heritability of alcohol consumption by Dutch religious versus non religious, which has been called a form of GE interaction. The authors account, in fact, wasn’t a developmental sensitivity one, in your sense, or a Biometric genotypes one in my sense — in that of different genetoypes reacting differently to different environments — but rather it was a canalization explanation (to use their term), a restriction of the phenotypic expression of genes in a similar genotype subset of the population. Now, I consider this above type of “GE interaction” to be trivially true. The subpopulation of kids hit on the head with hammers or who can’t speak the test language probably have lower IQs and show less heritability than that of Kids reared in a non-hammer hitting or language of the test environments. Duh. But, unless I am missing something, this isn’t evidence of developmental sensitivity GE.

    Let me know if you more or less agree with these clarifications.

  35. Revolver says:

    ‘When I use GxE I mean it in the traditional biometric sense of’

    As far as I know, your definition is just a subset of the traditional definition: “the phenotypic effects of different genotypes are differently affected by a given environmental change.” That would cover “environmental advantages” having a different effect and it would also cover the crossing reaction norm you describe.

    ‘For most of the population (i.e., most of the genotypes) increased nutrition is associated with increased height/IQ/etc., but for some, it may be associated with decreased height/IQ/etc. ‘

    “Nutrition” is very vague. Part of the entire problem is using vague metrics. If “nutrition” means a set of 10 foods, 3 of which generally encourage development of P in half the population and 7 of which generally encourage development of P in the other half, we will still miss the G x E.

    ‘But the point is that this is different from a situation in which some variants of PKU react positively to a phenylalanine diet while others act negatively.’

    Considered in isolation, sure. But as time goes on, research finds other, similar effects. For example, the FADS2 gene mutation’s connection to breastmilk and IQ. In one study, a CC or CG mutation was noted to have a positive effect on this relationship versus the insignificant relationship of the GG mutation, but in another study the GG mutation had the largest positive effect on this relationship. The main difference between the two studies was the overall genetic makeup/similarity in the studied populations. So…..in one genotypical group, the effect by mutation type is up, up, 0. In another group, the effect is small up, small up, big up. The latter study noted that GG children, on their own, had lower average IQs than the other CG and CC children. This difference disappeared when controlling for breastmilk. The above describes a G x E interaction, as you have defined it.

    There also many connections between several micronutrients and IQ. You could say that these are just “general effects up or down” but it seems like we are picking up differential sensitivities to different environmental stimuli. The small average effects we see are likely the result of small subsections of the populous being heavily affected versus the rest.

    ‘In fact, one of my arguments for a racial hereditarian hypothesis is that while low SES, at least in the US, tends to be associated with lower heritability (higher discordance between genotypes and phenotypes) being African American isn’t (which arguably suggests that African American IQ isn’t environmentally depressed.)’

    Well, within group and between group heredity would function like that anyway. If African Americans uniformly suffer a bad environment, then we would expect to find exactly that — high heritability among African Americans.
    A good setup would be to inquire as to the heritability of a trait between black twins who have markedly different skin tones.

    Lighter skinned full siblings are more likely to be enrolled in school and attend higher quality schools: http://faculty.ucr.edu/~jorgea/econ261/colorblind.pdf

    Of course, that’s chicken-egg…is it lighter-skin therefore possess more high IQ admixture or lighter-skin therefore receive differential treatment? Seems uncertain at this point.

    ‘Nonetheless, most people would recognize that a better/worse genotype is still better/worse even if the betterment/deleteriousness is only fully realized in certain environments. No one, for example, wants PKU even if it can mostly be treated.’

    Yes, but evolution is a story of tradeoffs. It’s probably not ‘I would like the genotype less PKU.’ It’s probably more like ‘in this environment I will take the genotype less PKU but with X, Y, Z additional sensitivities, because X, Y, Z will not be present in this environment.’ That is why objective ranking may be flawed.

  36. Chuck says:

    “As far as I know, your definition is just a subset of the traditional definition: “the phenotypic effects of different genotypes are differently affected by a given environmental change.” That would cover “environmental advantages” having a different effect and it would also cover the crossing reaction norm you describe.”

    The point is that we are dealing with different genotypes. We are not just dealing with the differential expression of a genotype in different environments or with a non-linear effect of environment on phenotype for a given genotype.

    So ya, unit change in overall nutrition has more of an impact on height/IQ at lower nutritional levels. At a certain point you hit a plateau or threshold and the benefit of unit increase in overall nutrition diminishes. This situation should and at times does correspond with an increasing h^2 with environmental moderator. But this isn’t GE in the sensitivity sense, not if you mean different genotypes.

    I’m trying to distinguish between this traditional sense and e.g., Hofer and Shannahan’s social control sense where a h^2 x E interaction is said to necessarily be a GE interaction.

    As for Blacks and IQ…Ya, but who believes that the depressive effect acting on the African American population is uniformly distributed. What Y-factor is maintaining this uniform effect? As for “colorism”, I’ve looked into this. The association (in the US) is mostly between families, disconforming a discrimination model. The sample size wasn’t large enough to decompose the IQ-color correlation. But all of this is another topic.

  37. Chuck says:

    “Yes, but evolution is a story of tradeoffs. It’s probably not ‘I would like the genotype less PKU.’ It’s probably more like ‘in this environment I will take the genotype less PKU but with X, Y, Z additional sensitivities, because X, Y, Z will not be present in this environment.’ That is why objective ranking may be flawed.”

    In these circumstances, you can locally rank if you know the specific environmental reaction ranges. And you can probabalistically rank if you know the variance in the trait due to biometric GxE effects (refer back to Tal, 2012). Thus, I don’t see the problem, so long as your concern is pragmatic.

  38. Revolver says:

    ‘I don’t see how VarE/E x h^2 can show developmental sensitivity GxE’

    When you stratify the populations and look at h^2, if you find a something like a curvelinear h^2 relationship between the populations then that would be evidence of G x E. If these are all the same genes in degrees of the same environment, then you are noting a patterned increase in not only the genetic effects but the magnitude of the genetic effects. It’s not a smoking gun or anything, but that’s why I used the term “suss out.” I mean it in the sense of investigating, not completely explaining or figuring out.

    I know there’s talk about inferring environment-environment interaction from such a relationship, but that sounds unconvincing a) in the context of an additive model and b) in the greater context of biology, where environment-environment interaction (as far as I know) is not a well-known term.

    http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0030320

    That’s an example. They note that because the moderator has an effect on C, then that must mean that there is an environment-environment interaction. However, that may be to simply leave the likely equation unsimplified. The direction could be G: if low SES, then sensitive to C, and if high SES, then not sensitive to C.

    ‘ h^2 of IQ is lower in first generation Hispanics than second generation ones because the first gen ones don’t all speak the language of the test.’

    Seems to reflect measurement error, in that case.

    ‘This should show a significant correlation in the presence of Biometric genotypes GE, but such hasn’t been found.’

    Probably because the “reared apart” studies don’t tend to do a good job of quantifying “apartness.”

    ‘But my point has been that these need not represent Biometric genotypes GE’

    It depends on what the h^2 relationship looks like. If it’s more linear, then that’s more reason to believe that the trait is malleable vis x environment. If it’s more curved, then that’s more reason to believe that the magnitude of the genetic effect grows.

    ‘I go back to the example of the heritability of alcohol consumption by Dutch religious versus non religious, which has been called a form of GE interaction’

    It may be as you say — a single environmental stimulus having a singular effect across all genotypes. However, if you plotted out religiosity in degrees and noted differential h^2 and found a curve, you could reasonably find evidence for developmental GE. I don’t know if the authors did that or not.

    ‘The subpopulation of kids hit on the head with hammers or who can’t speak the test language probably have lower IQs’

    One on account of environmental impact on the trait but the other because of measurement error.

    ‘In these circumstances, you can locally rank if you know the specific environmental reaction ranges. And you can probabalistically rank if you know the variance in the trait due to biometric GxE effects (refer back to Tal, 2012). Thus, I don’t see the problem, so long as your concern is pragmatic.’

    Okay. But in most cases it doesn’t seem like we know the specific environmental reaction ranges. Nor do we know the variance in traits due to “biometric G x E.”

    ‘But this isn’t GE in the sensitivity sense, not if you mean different genotypes.’

    At the population level you could simply be seeing G x E. There are several individuals who maintain high IQ in the face of malnourishment. However, it’s likely that more individuals are sensitive to nutrition. This sensitivity would be genetic. And what complicates the “nutrition” metric is that “nutrition” means different things to different genotypes. In the case of one “nutrient,” breastmilk, for example there is a G x E interaction.

    ‘As for Blacks and IQ…Ya, but who believes that the depressive effect acting on the African American population is uniformly distributed. What Y-factor is maintaining this uniform effect? ‘

    A unique history. A unique skin color. A unique culture. Any number of uniformly unique things.

    ‘As for “colorism”, I’ve looked into this. The association (in the US) is mostly between families, disconforming a discrimination model.’

    Everything I have seen makes an assumption about lighter skin color possibly correlating with > IQ because of > European Admixture, accounts for this possibility, and then makes their between/within conclusion. The only study on this I know of is Witty, P. A. and Jenkins, M. D. (1936). Intra-Race Testing and Negro Intelligence. Journal of Psychology, 1, 179-192. And they found that greater European admixture had nothing to do with higher intelligence in blacks.

  39. Chuck says:

    You: “You said: When you stratify the populations and look at h^2, if you find a something like a curvelinear h^2 relationship between the populations then that would be evidence of G x E.”

    When I get a chance, I’ll email a couple of experts for clarification regarding the differences between h^2 environmental moderation, h^2 x e^2/c^2 interactions, and developmental sensitivity GE. There’s no point in discussing this further, until we can get a better grasp of the conceptual and statistical differences.

    I don’t really want to waste my time one this, but…

    You: “A unique history. A unique skin color. A unique culture. Any number of uniformly unique things.”

    The problem is the uniform part. The point is that one would expect some large chunk of the Black population to escape the depressive effects holding the average down. Just as a large chunk of hispanics manage to become proficient in English or just as after a couple of generations most slave Blacks caught up with free Blacks in outcomes a large chunk of African Americans should be able to realize their genetic potential for cognitive ability, Otherwise you need to propose some Y factor which is homogenizing the differences, None of the factors you mentioned, of course, are uniformly distributed in the Black population.

    You: “Everything I have seen makes an assumption about lighter skin color possibly correlating with > IQ because of > European Admixture, accounts for this possibility, and then makes their between/within conclusion. The only study on this I know of is Witty, P. A. and Jenkins, M. D. (1936)”

    Sounds like you haven’t well followed the debate,

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s