Tuesday, 22 April 2014

Nature and nurture: Rosalind Arden replies


Dr Rosalind Arden has investigated whether the concept of general intelligence, usually seen as exclusively human, can be found in apes and dogs, and finds positive results. This is interesting in its own right, and also useful in refuting the argument that “g” is just an artefact of the intelligence tests used in factor analytic studies. I will try to get her to report more about her work on dogs next week.

In the mean time, here is her note replying to the general question as to how researchers in the field see the topic of nature and nurture, and whether it is worth continuing to research it.

Nature and Nurture

The standard statistical methods in behavioural genetics (the field in which nature and nurture are empirically tested) quantify the proportion of differences among a measured population (at a specific point in time) on a given behaviour, characteristic or trait. This gives results such that differences among people are (for example) 60% genetic and 40% environmental.

What does the average reader make of this 40%? What do we mean by the ‘environment? Most people I work with seem to assume that: we don’t know exactly, but it is something to do with the world out there. That may be true, and there may be significant elements of the world out there that do contribute to such differences (such as acute infections in childhood). But we have done a woeful job of finding out just how much of this ‘environment’ could be random biological developmental noise, rather than the environment 'out there in the world'.  One place to start with this question is to measure the variability of specified traits within a population of organisms that share the same genes.  Work like this has been done - researchers have examined variability within such 'isogenic' populations, but their findings have not been much discussed in relation to the meaning of the 'unique environment' in human studies.

The potency of Nature is often misunderstood. Knowing that the differences in trait X between people are caused 60% by genes does not tell you that genes determine 60% of trait X in Jane or Jimmy. As others have said, a heritability estimate is not a ‘gene-o-meter’. Heritability is a population level statistic, not an individual metric. Nor is knowing the population-level estimate of genetic influence on psychological traits (such as intelligence) very informative about limits (such as Jane/Jimmy could never get A levels’ or ‘would be assured A levels’) because there are so many other determinants of getting A levels. These include aspects of the external environment – and other traits in Jane/Jimmy (such as propensity to be excited by work, having friends who encourage work, and so on).

It’s rarely useful to ‘know less’ (the alternative to knowing more); so learning more about nature and nurture is a good thing. There is great consensus among scientists who conduct empirical work in this area about the value of knowing more - which is encouraging. Pretty much everyone I know: shares the view that learning about ‘the causes of traits’ is a work in progress, has some humility about what we know, and is keen to learn more.

Rosalind Arden


  1. "Knowing that the differences in trait X between people are caused 60% by genes does not tell you that genes determine 60% of trait X in Jane or Jimmy."

    But h^2 does allow one to compute, for example, the probability that a random Jane's deviation from the population mean is more genetically conditioned than not, no? That is, variance coefficients are not completely uninformative on the individual level. (This was discussed by Omri Tal in 'From Heritability to Probability'.)

    "Nor is knowing the population-level estimate of genetic influence on psychological traits (such as intelligence) very informative about limits ... because there are so many other determinants of getting A levels."

    But h^2 allows one to estimate the relative influence of environment of the sort that is variable in the population in which h^2 was measured. As such, it tells one how limited the kind of environment which can be measured and manipulated -- the variable kind -- is when it comes to inducing differences. Of course, since h^2 allows one to make probabilistic claims concerning genetic influence on the individual level, it should allow one to make similar claims concerning environmental malleability (on this same level.)

    I will do the math for you, but I first need to raise my g score 1.5 sigma -- just give me a moment....



    sincerely yours,

    bgi participant

    1. I find that I'm much less likely to post impulsive comments -- with sentences upper-cased and half dashed out profanity -- when I post under my actual name. You might try this, too.

      g -- or, when at the beginning of the sentence, should it be "G"? -- either denotes the positive manifold or the latent variable which is theorizes to induce this. I'm not aware of any study which has demonstrated the non-existence of this manifold -- maybe you could provide a reference.

      Whether or not a causal latent variable underlies this manifold is still debated -- the data is at least consistent with this position -- see, for example: Panizzon, et al. (2014). Genetic and environmental influences on general cognitive ability: Is g a valid latent construct?

    2. i agree. the manifold is positive, always.

      but the point is that spearman's g doesn't explain all of the test-test correlations.

      the factors aren't just aliases for g.

    3. and it's true for all "psychological traits".

      psychologists, AT BEST, parrot their critics. they don't understand the criticism.

      the most heritable or most concordant psychological trait is SCZ. yet it simply doesn't exist in primitive societies, EXCEPT so far as ideologically motivated "scientists" "see" it. http://www.greenmedinfo.com/article/schizophrenia-prevalence-correlated-gluten-grain-consumption-0

      THE EVIDENCE is that BLOOD PRESSURE is as heritable as IQ, especially when both are adjusted for reliability.

      YET anyone, at any age, who doesn't have kidney disease or an adrenal tumor can effect a bp of 100/60. http://www.pnas.org/content/101/17/6659.full.pdf

      just fast until your bp stops dropping and start eating a low protein diet. BUT needless to say, this is too extreme an environmental intervention for most.

      and, i hope, needless to say, environmental interventions DO increase IQ...as long as they last!!!

      but diet and exercise ALSO are of no benefit after you're through with them.

      behavioral genetics has yet to have its newton or even galileo. at present it's a pseudoscience.

    4. “spearman's g doesn't explain all of the test-test correlations”

      That was Spearman’s one factor model from the first half of last century.

      “racists like jensen, are NOT VERY BRIGHT.

      I’m an unapologetic racist myself -- just the other day, I microaggressed an Asian chick.

      "and, i hope, needless to say, environmental interventions DO increase IQ...as long as they last!!!"

      Interventions increase IQ scores, surely -- as does re-testing. Do these methods increase latent ability, though? And "as long as they last" is quite a stipulation. This reminds me of when, in Cambodia, I was told by a slippery hostel manager that the internet was fee; after I began using it, I was told, though, that I needed to pay. "Free", it turned out, just meant "available". But sure, with permanent interventions you probably can permanently increase IQ scores and certain narrow abilities.

      “the most heritable or most concordant psychological trait is SCZ”

      It’s true that you can often easily raise highly heritability traits. Clowns on stilts and height come to mind. Regarding IQ scores, you can back to back retake a test. I agree that the relation between environmentality (1-heritability) and environmental malleability requires clearer exposition.

  3. are caused 60% by genes


    what this means is that ASSUMING a bivariate normal distribution for THE MODEL P = G + E one finds that...

    the phenotype for one person with G = g will 95% of the time be between .6*P +/- 2*sqrt(1-.6^2) where P is the phenotype of someone else with G = g.

    so, in the case of IQ, if one twin has an IQ of 145, the other will have an IQ >= 145...ONLY Phi(-1.5) = 6.7% of the time.

    This absurd model says that at the high end ENVIRONMENT MATTERS A LOT!!!

    But the model P = G + E is ONLY a linear approximation, a tangent plane, to the GxE to trait SURFACE!!!

    Such approximations ONLY work when BOTH genes and environment are limited to a VERY small range.