"Killing Jensen": A preliminary comment on Chanda Chisala’s environmentalist argument

Chanda Chisala, a visiting Fellow at Stanford University, has developed what he considers to be a devastating argument against Jensenism (racial-IQ-hereditarianism). He develops this in his 2014 blog post, “Killing Jensen — part I“. Pithily put, the reasoning runs:

(a) A hereditarian model predicts differential regression to the mean. Were one to match U.S. Black and White parents (or sibs) on IQ, the Black offspring (or co-sibs) should regress to a mean lower than that of the White offspring (or co-sibs), specifically to the proposed Black genotypic mean.
(b) The offspring of some subpopulations of Black Americans do equally well, if not better than, Whites on average.
(c) In principle, the unexpectedly high performance of the Black offspring, from these subpopulations, could be due to parental selection. But the magnitude of selection needed, given by breeder’s equation*, is more than what observation indicates.
(d) This situation disproves a hereditarian model of average differences for these subpopulations.
(e) (unclear hidden premise)
(f) Given (d) and (e), we can infer that the U.S. Black-White gap, in general, is also due to non-genetic factors.

Chanda has since applied the logic to immigrant-native differences in the U.K. Here, I will restrict my critique to the author’s argument presented “Killing Jensen — par1”. The subgroup Chanda picks, in defense of his point (b), is that discussed in “Our Kind of People”. Regarding this curious community, Chanda informs us:

In fact, this very thing has happened in the black “community” of America with regards to IQ. A fascinating book published in 1999 entitled “Our Kind of People” by bestselling black author, Lawrence Otis Graham, described in great detail how some black families have been building some sort of secret elite society for many generations, running all the way back to the Emancipation…Granted, this elitist project was apparently started by blacks who were progenies of white slave masters and black slave women working in their homes, which admits a white gene component. They were treated differently from other black slaves – the “field negroes” – and this gave them a sense of being different from (and superior to) other blacks. They thus deliberately refused to associate closely with the less privileged blacks or to live like them, preferring the culture of their white masters. They even rejected their musical tastes and other forms of black cultural expression. However, they also “realized” that they were never going to be accepted by whites as their equals and in fact were legally forbidden from marrying white people, which made them develop an even stronger animosity against whites and to fiercely embrace their compromised “black identity,” while also being determined to prove themselves to their white colleagues…Beginning with an intensive children’s mentoring program they call “Jack and Jill,” these families constantly expose their children to black role models who are doctors, lawyers, dentists and so on – they are not allowed to become entertainers or sportsmen and are never exposed to any such “role models” no matter how rich they are. These children then qualify to very good universities before proceeding to elite professions that demand a pretty high IQ while also paying very well. They use their wealth from these professions to continue supporting the activities of Jack and Jill and their own adult versions of the elite children’s group!…The writer of the book was himself a product of this enduring “secret” project and his family is a classic example of this defiance of Jensenian regression to the supposed black mean[.]

Chanda reasons that this group, taken as a whole, is culturally, not genetically, advantaged, since the offspring do not appear to regress to a mean lower than that of Whites — though no actual IQ data are provided.

The argument is open to a number of critiques. For one, as mentioned, no IQ data are presented. For another, Cochran and Harpending (2015) have shown that assortative mating can readily produce hereditary castes in just several generations. And, it so happens, that the African American subpopulation discussed by Chanda appears to have been relatively endogamous, thus they “deliberately refused to associate closely with the less privileged blacks or to live like them”. Individuals from such castes would regress to the mean of their caste, since we are no longer dealing with individuals drawn from a common population, but with separate populations.

These types of counter-criticism miss a more fundamental problem. A while back Meng Hu and I replicated Jensen’s and Murray’s findings of differential regression using the nationally representative NLSY79 and NLSY97 surveys. Aware of claims by Flynn (1980) and Brody (2002) that such results could be accounted for by an environmental hypothesis, we simulated various ones. Lo and behold, we discovered that it was impossible to distinguish between a genetic model and a mostly shared environmental one. Brody (2002) clearly erred in claiming that virtually any environmental model could produce such differential regression results, but Jensen (1998) blundered worse when he asserted that, “[n]o purely environmental theory would have predicted such [regression] results”.

That is, differential regression — and a lack thereof — can readily be produced by shared environmental advantage/disadvantage. How is this so? Imagine that we have an unstructured population and that we add structure to it. We divide it into groups, A and B, and depress B’s traits by 0.90 standard deviations (SD) in shared environmental effect and 0.10 SD in unshared effect with a 0.25 standard deviation of total depressive effect. What do we get? More or less what differential regression plots depict. How does this work in practice? Imagine B parent-offspring dyads before the introduction of the depressive effect. Assume perfect assortative mating and thus that the B parent IQ is the same as the B mid-parent IQ. Imagine that the B parent had an IQ of 120 relative to the A = B mean of 100. The offspring would be expected to have an IQ of (120 – ((120-100)*0.4)) = 112. Now add our depressive effect. The B parent’s IQ will become 120 – 0.9*15 = 106.5. And the B child’s IQ will become 98.5. Were we to match depressed B and undepressed A parents on IQ, we would find our differential regression when looking at the offspring scores.

Many self-proclaimed race realists would object to what is being suggested here on the grounds that shared environment is impotent. But large meta-analyses suggest otherwise; the correlation between IQ and shared environment (c)  is generally found to be a non-trivial 0.40 (for example: Polderman et al., 2015: table 2). Moreover, some rather sophisticated models, which take into account the covariance between genes and environment, suggest that much of the purported additive genetic variance for IQ may, in reality, be gene-environment covariance (de Kort et al., 2015). If so, the potential malleability of ability across both times and groups could be greater than previously anticipated. As I noted in my Nature of Race paper, these CovGE models allow for a more complex form of “hereditarianism”, one which can more readily accommodate unexpected fluxes in group differences across time and place. With such a model, genotypically induced phenotypic differences would be better conceived of as elastic rather than as fixed and the epistemic function of a genetic model would be to account for the tendency for groups to differ.

Now, as pointed out by ardant hereditarians, shared environmental variance seemingly drops to zero by adulthood and, indeed, meta-analyses bear out this claim (for example: Polderman et al., 2015; figure 3). And if c^2 drops to zero then covG(shared environment) must likewise (Sesardic, 2005; de Kort et al., 2015). So there are real epistemic and quantitative problems with shared environmental models of group differences in latent ability. This assumes, of course, that one does not resort to so called x-factor accounts, ones according to which environmental factors more or less uniformly depress or elevate a group’s trait values, accounts which are both theoretically and empirically implausible in the case of the U.S. B-W differential**, though not when it comes to, for example, secular and international differences. No one appears to have solved the riddle of high heritability-shared environmental malleability-and ingroup variability, but many nonetheless believe that it is a non-problem.

Whatever the case, we need not resolve this heritability-malleability issue. For the sake of this critique, it is sufficient to point out that Chanda Chisala believes that his argument demonstrates that the temporally and longitudinally stable, highly predictive, g-loaded, psychometrically unbiased U.S. B/W IQ gap is environmental in origin. This commits him to the following claims: (a) environmental factors can cause (the found pattern of) differential regression and (b) group differences of this magnitude can be readily shared environmentally produced (after all, contrary to what our sociologists assert, Black Americans do not live in an environment radically different from that of White Americans).

Of course, if an environmental depressive effect can induce differential regression (of the type found), than an uplifting effect, of the same sort, can eliminate it. Thus one does not need to posit genetic structure in a population to explain a lack of differential regression. Likewise, one need not take recourse to super selection to explain unexpectedly high trait scores of offspring. One can just maintain that these individuals along with their families are environmentally uplifted, just as, on average, White Americans are said to be relative to Black Americans. (Yes, “it’s just bee-cause of culture” goes both ways.) To demonstrate that this is not the case, one would have to experimentally control for environmental differences e.g., by employing a cross fostering research design.

And if large phenotypic differences can be environmentally induced despite genotypic equality in conjunction with high heritability, it follows that large phenotypic differences can be reduced despite low environmentality and inequality in allelic risk scores. The logic is symmetrical. If an African American subpopulation in fact (environmentally) closed “the gap”, this entails, as Chanda rightfully argues, that differences are not “genetically determined”. But it also entails that genetic differences do not determine, but rather incline, phenotypic ones — and consequently that variability in the magnitudes of phenotypic differences can not constitute dispositive evidence against a genetic model. So what was Chanda’s error? It was to not think through the logic of his own argument. At best his argument kills Jensenism in one sense only to free it up in another.

As noted prior, Chanda Chisala has expanded his argument to immigrant-native differences, specifically in the U.K and the U.S. The immigrant-native issue is more complex and it deserves a separate critical analysis. But the points made above apply to that case.

*It is often said that sibs or offspring regress owing to all but the variance in a trait explained by additive genetic factors: strictly speaking this is not correct. Consider MZ cotwins reared together. They share both genes and shared environment. On account of both, their scores will be similar. The scores will differ — will exhibit regression — only as a result of non-shared environment and measurement error, since it is only in these latter that the twins differ. Thus, in the DeFries-Fulker formula, the degree of regression for MZ co-twins acts as a direct measure of unshared environment. To put this point another way, if shared environment explained 100% of the variance in a trait, no regression to the mean would occur, just as if additive genes explained the same portion.

**This issue was discussed in the blog post, “The facts that need to be explained” (link). Search :”x-factor”.

Reference

Brody, N. (2002). Jensen’s Genetic Interpretation of Racial Differences in Intelligence: Critical Evaluation. In: Nyborg, Helmuth, ed. The Scientific Study of General Intelligence: Tribute to Arthur Jensen. Pergamon.

Chisala, C. (2014, July, 16). Killing Jensen. Part 1. Accessed at: https://chandachisala.wordpress.com/2014/07/16/killing-jensen-part-1/

Flynn, J. (1980). Race, IQ and Jensen, London and Boston: Routledge & Kegan Paul.

Harpending, H., & Cochran, G. (2015). Assortative Mating, Class, and Caste. In The Evolution of Sexuality (pp. 57-67). Springer International Publishing.

Jensen, A. R. (1998). The g factor: The science of mental ability.

de Kort, J. M., Dolan, C. V., Kan, K. J., van Beijsterveldt, C. E., Bartels, M., & Boomsma, D. I. (2014). Can GE-Covariance Originating in Phenotype to Environment Transmission Account for the Flynn Effect?. Journal of Intelligence, 2(3), 82-105.

Polderman, T. J., Benyamin, B., de Leeuw, C. A., Sullivan, P. F., van Bochoven, A., Visscher, P. M., & Posthuma, D. (2015). Meta-analysis of the heritability of human traits based on fifty years of twin studies. Nature genetics. (Pdf)

Sesardic, N. (2005). Making sense of heritability. Cambridge University Press.

[In response to incisive comments, this blog post was revised, with modifications made, on October 12.]

12 Comments

  1. “Many self-proclaimed race realists would object to what is being suggested here on the grounds that shared environment is impotent. But large meta-analyses suggest otherwise; the correlation between IQ and this variance component is generally found to be a non-trivial 0.39 to 0.44 (for example: Polderman, et al., 2015).”

    And I would put zero stock in such a finding. The limitations of meta-analyses need to be kept in the forefront whenever considering their findings. Essentially, they are only good for correcting random sampling error and for checking for publication bias. That’s it. Other forms of measurement error or otherwise biased/unreliable measurements carry over into meta-analyses.

    In general, measurement error tends to bias down heritability estimates. Shared environment estimates are inflated by assortative mating. And of course, shared environment effects are often observed in children – an effect that disappears in adults.

    In short, one large, well-designed study trumps any meta-analysis.

    • Chuck

      October 12, 2015 at 11:06 am

      (edited)
      JayMan, regarding c variance, see my reply below. I will upload the paper for the sake of transparency. My critique is not fundamentally damaged by this, though, since I stated, “We need not resolve this heritability-malleability issue, though…” The argument should be read as conditional: If you maintain that — or if — the US B-W gap is due primarily to shared environment, then… Given the psychometric nature of the differential (e.g., g-loadedness, longitudinal stability, differential regression, etc.), the B-W gap can only be due to (mostly) shared environmental and/or genetic factors. That this is the case can readily be shown. And if — if — it’s due to the former then high heritability can not preclude extensive malleability — the shared environmental differences in this case, as you would agree, are just not that great. And if this is the case, then finding exceptions — in this case, very high performing legacy African American subpopulations — proves nothing in itself, since it could always be the case that the relevant group’s phenotype is evaluated relative to it’s genotype — to demonstrate otherwise, one would have to experimentally control for environmental differences e.g., cross fostering studies. Now, my feeling is that there is something amiss with Jensen’s malleability contention, though I haven’t been able to put my finger on the exact flaw — nor have others. CovGE doesn’t actually help since if c^2 falls to 0 by adulthood, then covG(shared environment) can not be greater than nil. Whatever the case, as noted, my argument is conditional.

      Now let me note that this is why the U.S. B-W gap is so interesting, and why it’s unfortunate that (genetically informed) research about it is so neglected. It’s a real puzzle — unless one adopts your arguably naive hereditarianism. And the possible solutions have important ramifications — for example, If one can environmentally condition such a gap, given the types of environmental differences abound in the U.S., extensively shared environmentally raising familial and national psychometric intelligence should be feasible — one merely has to find out what’s depressing African American latent ability.

  2. First, if measurement errors have an effect, it’s by upward biasing the non-shared environment component. Second, I don’t know how Chuck obtained his estimates of ~.40 for c² in the Polderman study, since from Figure 2, I eyeball that c² is about 0.20. And given their numbers in Table 2, the c² component is 2*DZ-MZ, i.e., 2*.371-.646=.096. But as I said in another article, it is advisable to use c rather than c² (the latter is not really an effect size). But even with c, you obtain just a modest correlation with IQ (0.309).

    • Meng Hu, I stated that the “correlation between IQ and this variance component is generally found to be a non-trivial 0.39 to 0.44”. I suppose that this was incorrectly/misleadingly phrased. I should have written that the correlation between IQ and “shared environment” is… Yes, I agree that one should use “c”. I checked the paper. As for the estimate, you were looking at the “general cognitive” domain. I was looking at “higher level cognitive function”, which referred to IQ, thus: “for example, for higher-level cognitive function, the original studies included the trait names total IQ score, cognitive ability, intelligence or ‘g'”. In table 2 the MZ was 0.71 and the DZ was 0.441, so SQRT(2*.441-.71) = 0.41. However, which I hadn’t noticed prior, Figure 2 lists values by age: 0-11 c^2 = 0.31: 12-17 c^2 = 0.07: 16-64 c^2 = 0 (or negative): 65+ c^2 = 0 (or negative). So I (effectively) stand corrected. I will make the necessary changes.

  3. Thank you for tackling this, Chuck. I’m a layman, but you explain the statistical arguments so well that I think I’m following. I can’t make statistical arguments, but my impression is that the UK data is not very compelling as a refutation of the 1SD or so (genotypic) w-b difference generally found. I’ll describe the problems with it I see.

    For one, the UK Africans are clearly atypical of blacks in general. They were a relatively academically accomplished group, are still fairly new immigrants, and still a tiny minority in a mostly-white nation. Some seem to be members of elite, socially assorted ‘tribes’, whose offspring wouldn’t be expected to regress very much. Quantifying the selective effect is difficult, but this uncertainty is a problem for both sides.

    Second, the test data is limited and doubtful. They are g-loaded, but not very highly. If I understand right, only the baseline GCSE pass rates are given, which restricts the score range. As Fuerst and James Thompson have discussed, the scores have much increased recently, indicating the tests were made (or became) easier. If so, then a given w-b performance gap must artificially shrink, creating an illusory black gain. More definitive tests are needed.

    Third, the students were teenagers, not full adults, which means that the gap would be expected to increase if it’s genetic-based. Environmental effects are more plausible at this age. We don’t know yet whether these GCSE pass rates will translate to high level accomplishment as adults.

    Not denying that Chisala’s line of reasoning is a problem for proponents of a genetic-based gap, but opponents have a far bigger problem with Mr. Ockham in my judgment.

  4. I assume you’ve seen this too but just in case.

    http://www.unz.com/article/the-iq-gap-is-no-longer-a-black-and-white-issue/

    There’s data for the UK I hadn’t seen before. He has a breakdown of GCSE results by African ethnic group, showing some Nigerian children such as those of the Igbo tribe do as well as Chinese children. Apparently, in 2010, an Igbo girl got the best GCSE results in the country- not one white or Asian kid beat or equalled her. I’d like to see their performance at A level but their significantly above the white mean performance at GCSE does seem noteworthy. How selected can their parents be?

    I’d also like to see what percentages got A at GCSE. Why do the stats always talk about 5 A*-C’s?

    • Chuck

      October 15, 2015 at 11:40 am

      Hi Steve,

      I commented on that post: //humanvarieties.org/2015/06/19/nature-of-race-published/comment-page-1/#comment-12462 Mind the sample sizes.

      What’s interesting is that many of these kids are 1st generation, no? Might you find an estimate as to what percent? So, these kids would not be benefiting from beneficial European natural environmental factors (e.g., low pollution, low parasite load, advanced medicine, etc.). That means that they are either very selected OR that Africans in Africa are not very environmentally biologically depressed in latent ability — and hence that the international differences are mostly hollow, perhaps just do to educational differences or to not using #2 pencils when filling out test scantrons. So there are real implications here. For example, Bill Gates and the UN spends hundreds of millions on programs to reduce parasite burden: perhaps that’s not needed.

  5. Maybe there is a way to get 2001 census data and find out the number of ‘black African’ 5 year olds and detract it from the number of black African GCSE entrants in 2011- the rest came from Africa since 2001 at at least age 5. (That’s still rough but its the best method I can think of).

    ‘Moved to Britain at least aged 5’ could be a good definition of first generation for this purpose as they’d have 5 years in Africa to be exposed to pathogens etc.

    I’ll see if I can find that data when i’ve got time but for now: there were 485,000 black Africans in the UK in the 2001 census and a little over 1 million in the 2011 census.

    Do you know anything about the age structure of (African) immigrants? Are the immigrants mostly working age or do they bring kids with them?

    If they bring a proportional numbers of kids, then it could turn out that half of the test takers were first generation and at least 5 when they moved to Britain. I imagine its less though.

    Steve

    p.s. what difference does the type of pencil make?

  6. What happened to the post on European/African/Amerindian admixture in Latin America?! Ah!

Leave a Reply

Your email address will not be published.

*

© 2017

Theme by Anders NorenUp ↑