Wednesday 30 September 2015

Does culture cultivate, or do you need a good plough?


The search for culture-free or culture-fair tests has proved endless, because “culture” can be used so broadly as to encompass virtually anything a human does. People live in society, and societies transmit the habits of previous generations. There was a time in the debates after Jensen’s 1969 paper when psychologists believed that they could estimate the cultural loading of a test by inspecting the items. Indeed, my very first published paper attacked Jensen for arguing that the Wechsler subtest of Block Design was relatively culture-free, such that black-white differences on that test were probably genetic, whereas I felt it depended on access to constructional toys.

How does one determine the cultural loading of an intelligence test item? A Dutch team have plunged into these waters (strictly speaking they are below sea level, but no matter) and have rated subtests thus: Cultural load was operationalized as the average proportion of items that were adjusted in each subtest of the WAIS-III when the scale was adapted for use in 13 countries (Georgas et al., 2003).  To my eye that is certainly a language adjustment, though I wonder whether it allows for the different availabilities of artefacts in the home (not that I can think of an easy way to measure that).

Kees-Jan Kan, Jelte M. Wicherts, Conor V. Dolan, and Han L. J. van der Maas. On the Nature and Nurture of Intelligence and Specific Cognitive Abilities: The More Heritable, the More Culture Dependent. Psychological Science 24(12) 2420–2428




On this measure, Vocabulary is the most culture dependent subtest. On first glance that makes sense: the easiest way to learn a language is to be immersed in the particular culture that speaks it. However, that merely covers translating words from one language to another. Even within a culture, even the most ethno-centric citizens do not learn all the available words: intelligence is required for that.

From an early post on Vocabulary: Some people have the simplistic notion that vocabulary must be determined by mere exposure to spoken language. That is necessary, but far from sufficient, as even children work out. They notice patterns, informal rules, and the contexts in which communication takes place.  “The acquisition of meaning is based on the eduction of meaning from the contexts in which the words are encountered”. (So, even if the word “eduction” in the quotation from page 146 of Jensen’s “Bias in mental Testing” is unfamiliar, you will not be surprised to deduce that it means “To assume or work out from given facts; deduce”).The meaning of a word is acquired in some contexts which permits at least some partial inference as to its meaning. By hearing or reading the word in different contexts, through a process of generalization, discrimination and eduction one can guess at the essence of the meaning of the word, so as to use it (experimentally) oneself the next time a similar context presents itself. Words move from being unfamiliar to familiar, from familiar but not really understood to being familiar and partly understood (at which stage the explanations given about the meaning of the word are threadbare and inaccurate), and from there to being explained by use of synonyms (though those can range from partial to full understanding as shown by the power of the explanations and definitions).

The Methods section is explicit about how things were calculated, one step at a time: a model approach to be commended.


WAIS subtest heritabilities

The culturally loaded tests have higher heritabilities.

The authors conclude:

Each subtest’s proportion of variance in IQ shared with general intelligence was a function of cultural load: The more culture loaded, the higher this proportion. In addition, in adult samples, culture-loaded tests tended to have greater heritability coefficients than did culture-reduced tests, and there was a relationship between subtest’s proportion of variance shared with general intelligence and heritability. In child samples, these relationships were in the same direction, but correlations were small and insignificant.

They sound a cautionary note about the data, but their substantive point is:

A correlation between, for instance, g loading and heritability coefficient is in line with the hypothesis that the g factor is the most heritable factor (Jensen, 1998), but a test of the significance of this correlation does not provide the means to test whether the g factor is indeed the most heritable factor1 (Dolan & Hamaker, 2001). The method merely serves to evaluate competing theories of intelligence (Rushton & Jensen, 2009): A significant correlation denotes that a phenomenon exists that is in need of theoretical explanation. Theories that account for the correlation are stronger (with respect to this correlation) than are theories that do not account for it or are silent about it. The same line of reasoning holds for the correlations of cultural load with g loadings and heritability coefficients.

Having given their conclusions, the team then go against normal sequence and start a discussion.

Our result showing that culture-loaded knowledge tests (crystallized tests) are more strongly related to general intelligence than are culture-reduced cognitive processing tests (fluid tests) fits better with the idea that g loadings reflect societal demands (Dickens, 2008) than that they reflect cognitive demands (Jensen, 1987). Furthermore, in adult samples, our finding that the heritability coefficients of culture-loaded tests tend to be larger than those of culture-reduced tests calls for an explanation, given that this result does not follow from the subtest-complexity and investment hypotheses of g theory and fluid-crystallized theory.

After discussing some options they plump for genotype-environment covariance.

Because the acquisition of knowledge depends on cognitive processing, individuals who develop relatively high levels of cognitive-processing abilities tend to achieve relatively high levels of knowledge. High achievers are more likely to end up in cognitively demanding environments that encourage and facilitate the further development of a wide range of knowledge and skills. The contents and organization of these environments largely reflect societal demands. These societal demands thus influence the degree of dynamical interaction among cognitive processes and knowledge and, hence, their intercorrelations. In this way, the societal demands determine IQ-subtest loadings on the general factor of intelligence and, eventually, the degree to which broad-sense heritability coefficients of IQ subtests include the effects of (growing) genotype-environment covariance. In view of theoretical parsimony, we conclude that the assumption of a true causal g can be incorporated but that this is not required.

This paper presents interesting, counter-intuitive findings, which deserve replication on other samples and other psychometric tests. As to their favoured genotype-environment effect, I don’t see how bright people can obtain high levels of knowledge without being bright in the first place. They don’t develop intelligence, they have that ability in varying degree and use it to develop their knowledge to varying degrees. I am still working this out, but I think that ability is prior, and therefore more likely to be causal.

See what you think.


  1. Lion of the Judah-sphere1 October 2015 at 14:51

    Still mind-boggling to me that Information, and especially, Vocabulary, have such high g-loadings. Those areas seem to be easy to artificially boost through rote memorization. I know a lot of people do this very thing.

  2. We also doubt that our Western society creates a homogeneous learning environment, given that schools and school systems differ strongly.

    What of learning which takes place in the family before students enter preschool?

    I spent a great deal of time singing and reading traditional children's rhymes to my children. This little piggy went to market/who has seen the wind/under the name of Sanders. I presume that practice, as well as talking to my children, helped them acquire large vocabularies. Causation? Correlation? If we didn't all enjoy the process, we wouldn't have continued.

    How in the world do you separate "heritability" in a biological sense from "heritability" of knowledge taught directly or indirectly from the biological parents?

    You can't deduce traditional children's rhymes or vocabulary. Once you know some basic forms, you can understand that "educe" bears some relation to "education"--but it's very different from reciting a string of numbers.


      In our current research, Naomi Slonim and I are finding that large vocabulary differences are present by the end of grade 2—amounting to more than 3,000 root words between high and low quartiles in a normative population (Biemiller and Slonim, in press). After grade 2, cross-sectional data indicate that the lowest-quartile children may actually add root word vocabulary faster than the higher-quartile children. However, by grade 5, they have only reached the median for grade 2 children. Thus, if we could find ways of supporting more rapid vocabulary growth in the early years, more children would be able to comprehend "grade level" texts in the upper elementary grades. (Note that the "reading grade level" of texts is in fact almost entirely determined by the vocabulary load of those texts (Chall and Conard, 1991; Chall and Dale, 1995). Thus early vocabulary limitations make "catching up" difficult even though once in school, children appear to acquire new vocabulary at similar rates. To "catch up," vocabulary-disadvantaged children have to acquire vocabulary at above-average rates.

    2. How in the world do you separate "heritability" in a biological sense from "heritability" of knowledge taught directly or indirectly from the biological parents?
      Adoption studies.

    3. Thank you for your link on teaching Vocabulary, which I have read with interest. I would like you to send me your paper, in press, when it is available. I agree that we ought to teach Vocabulary and read to our children, but I don't think that differences in children's vocabularies are usually primarily due to mothers speech and parental reading, so long as children have general access to language in society.
      I presume you have read my posts on vocabulary. These make points slightly different from yours, I think. That is, other than in severely deprived children, I think comprehension is largely driven by intelligence.
      On the Hart and Risley (1995) study, I wonder if you have seen Heiner Rindermann's paper
      Rindermann, H. & Baumeister, A. E. E. (2015). Parents’ SES vs. parental educational behavior and children’s development: A reanalysis of the Hart and Risley study. Learning and Individual Differences, 37, 133-138. (-PDF- password-protected)

    4. In fact, I covered the paper here:

      You will see that Rindermann (as you do) stresses "educational behaviour" but of course cannot exclude genetic effects in a biological parents teach biological children model.

    5. Dear Sir: I am not the researcher who wrote the piece on teaching Vocabulary; I had intended the italics to signify a quote from the piece.

      Comprehension is driven by intelligence, I would agree, but if the children from less advantaged backgrounds learn new words at a faster rate, once they reach school, it seems odd that the children from more advantaged backgrounds would seem to learn more slowly, if they were more intelligent? Logically, if I have not been exposed to a word, I cannot learn it.

      In the issue of adoption studies, presumably of identical twins, I think it unusual these days for twins to be deposited in families of greatly differing SES. Indeed, the adoption agencies in this country seem very stringent about placing infants in functioning families.

    6. I think that adoption shows little effect on language development.


    7. "but if the children from less advantaged backgrounds learn new words at a faster rate, once they reach school, it seems odd that the children from more advantaged backgrounds would seem to learn more slowly, if they were more intelligent?"

      Low hanging fruit seems a possibility.

    8. Yes, of course that's possible. However, hasn't speed of learning new things generally been show to correlate with intelligence? As we're considering early elementary schooling, the "more advanced" words aren't that much more advanced.

      It is possible, of course, that teachers tend to focus instruction on the children who need to catch up, rather than the students who are already ahead. Although that would also emphasize that vocabulary is taught and learned, not an automatic birthright.

      Our children heard two languages in infancy. I noticed that they seemed to learn the word for an item or concept in both languages simultaneously. However, they didn't seem to tie either word to its pair in the other language; they associated both words with the underlying concept. So, for example, the sight of a chair (or the action of sitting) could elicit the words in either languages, but the word "chair" did not pull up the foreign language equivalent at the same early stage. For my children at least, there was an ideal, Platonic chair, which precedes vocabulary.

  3. Language is fluid and convenient because it accompanies the ephemeral emergence (or not) of our ideas and thoughts. We all have our favorite words long or short term.

    Words that are moderately hard (which tend to be synthesized) are more efficient to make a clear and understandable or didatic communication.

    Exposure won't enough without having mutual interest / curiosity for the occurrence of internalizing and learning.

    It would be better if we measured the conceptual knowledge of persons in respect of abstract words that most use in their daily lives.

    1. ''Accompanies'' or (better) follow, ;)

  4. Difficult words were invented for avoid ''repetitions'' of common words, ;)

    Language is quasi-always about the beauty of their sounds and not exactly by its efficiency.

    Verbally smart people hate when they repeat a lot some common words.

    1. More likely they appeared as a measure against loading the text with attributes and adjectives.

    2. Good joke!!! :)

    3. He who lives in a house of glass... should cut out on the snide.

  5. The culturally loaded tests have higher heritabilities.

    They also seem to have some of the lower Flynn Effects as well. According to Flynn's 2007 book, the lowest Flynn Effects were seen on Information, Arithmetic, Vocabulary, and Comprehension, which are four of the five most culturally loaded subtests:

    1. My impression is that a sizable part of the Flynn Effect is due to cultural differences that exist over time rather than space. The whole world is changing culturally, and global culture is becoming more like what IQ tests were like.

    2. My idea is that IQ tests were ahead of their time in being much like dealing with electronic intelligence.

      I realize I haven't spelled my theory of the Flynn Effect out very well, but when I look at the output of Silicon Valley, say, as representative of the way global culture has been trending, it strikes me as not coincidental that the two individuals most often called the Father of Silicon Valley -- William Shockley and Fred Terman of Stanford -- were friends, and that Terman's father, Louis Terman, was the father of IQ testing in America.

  6. Dear Steve, Thanks for your interesting comments. I think that Heiner Rindermann and I would argue that culture has always been led by the smart fraction, and silicon valley had a very smart fraction, leading to the creation of tools which were very smart. The total power of our culture has increased through computation. Paradoxically, this has made life easier for the average person. They can now control a complicated device just by pointing their finger at a pictogram. Smartphones help the least smart most, because they use intuitive pointing, and not the high demand of computer programming. Were IQ test ahead of their time in being like dealing with electronic intelligence? Most of them were not, but I am still thinking through the main thrust of your suggestion.

    1. "culture has always been led by the smart fraction": maybe. But one of the Roman writers pointed out that when the Germans invaded Gaul, the leading Germans adopted much of Roman culture, and the poor Gauls adopted much of German culture.

  7. Do animals would know estimate the intelligence of human beings **
    Pretty stupid question a priori, but who knows ... it never hurts to ask.

  8. "This paper presents interesting, counter-intuitive findings, which deserve replication on other samples and other psychometric tests."

    Personally, I don't fee that the idea is worth the hassle of a write up.

    1. Let me clarify this last statement. I invited Kees-Jan Kan to use our 46 batteries -- a data set which Meng Hu and I meticulously built for Jan te Nijenhuis -- to try to replicate his analysis. I did not want to do this myself, since I was unsure about my cultural load codings and since I enjoy writing up papers (or blog posts) as much as I do getting my blood drawn. For the heck of it, I had run the analysis based on my codings and found that while g mediated the r(CL x h^2), h^2 did not fully mediate the r(CL x g). This seemed not to support Kan et al. CovGE interpretation, as I understood it. Curiously, I also found, if I recall correctly, that cultural load did not correlate with shared environment, as one might expect. Whatever the case, if anyone cares to replicate and extend the findings, a massive data set is readily available. The original findings might deserve a replication, but it's not a project that would be worth while for me, given my particular circumstances.

    2. Thanks for this explanation, and the invitation extended to readers.

    3. I don't see why a finding while g mediated the r(CL x h^2), h^2 did not fully mediate the r(CL x g) would go against the Coc G-E interpretation. Apart from the MCV not being a mediation TEST, mediation of h and/or of Cl by g says nothing about the correlation between Cl and g. Second, h is never a mediator; it's a proportion of variance.

    4. I don't see why a finding while g mediated the r(CL x h^2), h^2 did not fully mediate the r(CL x g) would go against the Coc G-E interpretation. Apart from the MCV not being a mediation TEST, mediation of h and/or of Cl by g says nothing about the correlation between Cl and g. Second, h is never a mediator; it's a proportion of variance.

  9. The Problem is this: If vocabulary is the result (the consequence) of fluid reasoning (the cause), how come individual differences in measures of the consequence are HIGHER heritable than individual differences in measures of the cause?

    Note: The question addressed in the paper is not 'How come individual differences in vocabulary are highly heritable?'. No, the question is 'How come individual differences in vocabulary are HIGHEST heritable'?

    Note 2: The explanation can't be 'due to differences in measurement error variance', because those differences were corrected for.

    Note 3: The explanation can't be 'due to a homogeneous environment'. That would only explain EQUALLY high heritability for the consequence as compared to the heritability of the cause. (unless one important assumption, namely that the cause, g/Gf, is the highest heritable variable, is violated).

    One possible explanation is that the environmental influences on vocabulary are getting lined up with genotype. In other words, the environmental influences on vocabulary also become heritable. So, whereas the environmental influences on fluid reasoning are relatively independent from fluid reasoning, the environmental influences on vocabulary are not; they will reflect in substantial part these differences in fluid intelligence. If so, the heritable part of the environmental variance adds up to the genetic variance of the underlying cause.

    So in equations:

    var(Gf) = var(Gfgenetic) + var(Gfenvironment)

    var(Voc) = var(Gfgenetic) + var(Gfenvironment) + var(VocEnvironment)

    = var(Gfgenetic) + var(Gfenvironment) + c*var(Gfgenetic) + d*var(VocResidual)

    So, note 4: The paper does not provide a cultural explanation; it provides an explanation in terms of 'genetics through environment'. The role of culture is to provide the conditions through which environmental differences are lining up with genotypic differences. We send out smart kids longer to school and to better schools than not so smart kids, so there definitely is G-E covariance. Imagine a society in which kids are placed in schools randomly, so not based on their intelligence, heritability of vocabulary will definitely drop.
    Or imagine a society in which the not so smart kids are the one that get selected to go the best schools and smart kids leave school sooner. Heritability of vocabulary will definitely drop even more.

    For the rest, please stop the MCV analyses. The MCV is crap.