Tuesday 31 March 2015

Income, brain, race, and a big gap

Usually I do not set my readers puzzles. It is not seemly. However, the recent coverage of a paper published in Nature has set me a puzzle which I would like you to help me solve. Are the authors of the paper, the reviewers and the editors of Nature Neuroscience aware of what has been left out of this study? Did they spot the gap which calls into question the conclusions, or just choose not to mention it? Let me tell you the story, and then you can judge for yourselves.

Family income, parental education and brain structure in children and adolescents. Nature Neuroscience (2015)doi:10.1038/nn.3983 Published online 30 March 2015

https://drive.google.com/file/d/0B3c4TxciNeJZT2EzTjV5UEZKRGM/view?usp=sharing

The paper has multiple authors, but I think it kinder not to list all their names. Here is their abstract, which is what most people will read:

Socioeconomic disparities are associated with differences in cognitive development. The extent to which this translates to disparities in brain structure is unclear. We investigated relationships between socioeconomic factors and brain morphometry, independently of genetic ancestry, among a cohort of 1,099 typically developing individuals between 3 and 20 years of age. Income was logarithmically associated with brain surface area. Among children from lower income families, small differences in income were associated with relatively large differences in surface area, whereas, among children from higher income families, similar income increments were associated with smaller differences in surface area. These relationships were most prominent in regions supporting language, reading, executive functions and spatial skills; surface area mediated socioeconomic differences in certain neurocognitive abilities. These data imply that income relates most strongly to brain structure among the most disadvantaged children.

The two principal authors have given a statement, and these are instructive because they tend to reveal what the authors regard as the real implications of their findings.

Dr Elizabeth Sowell, of the Children’s Hospital Los Angeles (last named author, theory development and interpretation of results), is reported as having said: ‘Our data suggest that wider access to resources likely afforded by the more affluent may lead to differences in a child’s brain structure. Access to higher-quality childcare, more cognitively stimulating materials in the home and opportunities for learning outside the home likely account for some of these effects.’

If correctly reported, she reveals that she thinks that material resources (within the range experienced by US citizens) lead to a difference in the child’s brain, and presumably thereby to intelligence. This is a strong claim.

Dr Kimberly Noble, of Columbia University in New York (first named author who developed the theory, conducted analyses, wrote the introduction, results, discussion and methods, which in my view makes her virtually the sole author) said that despite the clear impact of socio-economic status on the young mind, it would be wrong to think that the changes are fixed. She said: ‘This is the critical point. The brain is the product of both genetics and experience and experience is particularly powerful in moulding brain development in childhood. This suggests that interventions to improve socioeconomic circumstance, family life and/or educational opportunity can make a vast difference.’

Her view is that by improving socioeconomic circumstances brain development can be improved, and thereby intellectual ability. She mentions that the brain is a product of genetics, which leads to the assumption that this has been considered in the paper but, despite that, “experience (my emphasis) is particularly powerful in moulding brain development in childhood”.

Curious about these dramatic claims, I read the paper. Here is a representative part of the introduction:

It is critical to examine socioeconomic factors such as education and income separately, as these correlated factors represent distinct resources that may have different roles in children's development. For example, income may best represent the material resources available to children, whereas parents' educational attainment may be more important in shaping parent-child interactions.

I would have added: parents’ educational attainment is also a proxy measure of their intelligence, and a good indicator of their children’s inherited intelligence, so we need to control for that, ideally by testing parents’ intelligence.

Their main findings (picked out from the paper) were:  Parental education was significantly associated with surface area independent of age, scanner, sex and GAF (racial ancestry) (β = 0.141, P = 0.031, F(22, 1076) = 31.67, P < 0.001, R2Adjusted = 0.381). Multiple regression showed that, when adjusting for age, age2, scanner, sex and genetic ancestry, family income was significantly logarithmically associated with children's total cortical surface area, such that the steepest gradient was present at the lower end of the income spectrum (β = −0.19, P = 0.004). We next constructed a model that included both education and income to assess whether these socioeconomic factors uniquely accounted for variance in surface area. Only the income term accounted for unique variance (β = 0.105, P = 0.001, F(22, 1076) = 32.52, P < 0.001, R2Adjusted = 0.387).  We next investigated associations between SES factors and cortical thickness. Initial analyses of thickness revealed that models were best fit using a quadratic function for age. When adjusting for age, age2, scanner, sex and GAF, multiple regression analyses indicated that parental education was not associated with cortical thickness, whether considering a linear, logarithmic or quadratic model.

In the discussion section the authors say: We found that parental education was linearly associated with children's total brain surface area, implying that any increase in parental education, whether an extra year of high school or college, was associated with a similar increase in surface area over the course of childhood and adolescence. Family income was logarithmically associated with surface area, implying that, for every dollar in increased income, the increase in children's brain surface area was proportionally greater at the lower end of the family income spectrum. Furthermore, surface area mediated links between income and children's performance on certain executive function tasks.

Notice that it is assumed that an extra year of education might increase the surface area of the brain. In fact the linear slope with parental education is relatively slight, as shown in their figures. 

Here is their version of the traditional required “we cannot be absolutely sure” paragraph in the discussion:

Of course, strong conclusions concerning development are limited in a cross-sectional sample. Furthermore, in our correlational, non-experimental results, it is unclear what is driving the links between SES and brain structure. Such associations could stem from ongoing disparities in postnatal experience or exposures, such as family stress, cognitive stimulation, environmental toxins or nutrition, or from corresponding differences in the prenatal environment. If this correlational evidence reflects a possible underlying causal relationship, then policies targeting families at the low end of the income distribution may be most likely to lead to observable differences in children's brain and cognitive development.

You will note that inherited characteristics are not mentioned in this important section. Not a single word. It seems to have escaped notice that the apparent SSE/brain link might both be driven by a common factor of inherited intelligence. Cross-sectional studies are particularly weak when the sample is not randomly drawn from a specific population, so we have, in my view, a sampling issue as well as a cross-sectional issue. However, in the next paragraph genetics makes an appearance, but in a slightly different context, that of race being a confounder of SES.

SES, cultural differences and genetic ancestry are often conflated in our society. To the best of our knowledge, this is the first study of SES and the brain to include as covariates continuously varying measures of degree of genetic ancestry. Notably, our results can only speak to the effects of GAF, a proxy for race. Thus, although the inclusion of genetic ancestry does not preclude the possibility that these findings may reflect, in part, an unmeasured heritable component, it reduces as far as possible the likelihood that apparent SES effects were mediated by genetic ancestry factors associated with SES in the population. Furthermore, associations between SES factors and brain morphometry were invariant across ancestry groups.

Pause a moment here. There is a mention of “an unmeasured heritable component” but it is then dismissed because the SES and brain measure relationships were invariant across racial groups. That is a different matter. SES and brain size can have the same relationship in all racial groups, and yet still be driven by inherited intelligence. What we need to see is means and standard deviations of the brain measures by racial group, so that we can see what has been adjusted in absolute terms. The paper has done well to include a genomic version of race, but that does not cover the major factor of intelligence being heritable in all genetic groups.

The authors conclude: many leading social scientists and neuroscientists believe that policies reducing family poverty may have meaningful effects on children's brain functioning and cognitive development. By elucidating the structural brain differences associated with socioeconomic disparities, we may be better able to identify more precise endophenotypic biomarkers to serve as targets for intervention, with the ultimate goal of reducing socioeconomic disparities in development and achievement.

Does this study allow us to reach that conclusion?

The sample is a bit of a mess from an epidemiological point of view, being composed of volunteers: Participants were recruited through a combination of web-based, word-of-mouth and community advertising at nine university-based data collection sites in (the US). Participants were excluded if they had a history of neurological, psychiatric, medical or developmental disorders. All participants and their parents gave their informed written consent/assent to participate in all study procedures, including whole genome SNP genotype, neuropsychological assessments (NIH Toolbox Cognition Battery).

It seems that the children (and perhaps also the adults) were given the The NIH Toolbox Cognition Battery, but I cannot find any results in the data set. The toolbox includes Vocabulary Comprehension, Reading Recognition and Pattern Comparison (processing speed) task from which an IQ estimate could be drawn and there are 5 other tests which can be looked at for a broader picture.

Income data and educational level were collected not as actual figures but in categories, so are cruder than required, particularly when fine details about lower income effects are being discussed. I have looked at the Excel sheet kindly provided, and Education is measured in years to the nearest two years. That is, all the scores are in even numbers. Years of education is a crude measure anyway (ignores education quality) but this silly restriction in the previous data collection makes it cruder still. It is not a fatal problem, but reduces data quality.

The study makes much of controlling for genetic ancestry, which is a good thing. However, they report none of their results on these differences. They say that the associations are the same in all genetic groups, which is not surprising, but no means or standard deviations for the brain measures by genetic background are given. These differences, if any, could be compared with differences in SES between racial groups, as another test of the hypothesis being examined by the authors. In terms of absolute levels how well do SES, education and race fit the data?

The sample was composed as follows: African 12%, American Indian 5%, Central Asian 2%, East Asian 16%, European 64%, and Oceanic 1%. The authors do not test whether this is representative of the USA. The European figure seems close to the  White, non-Hispanic or Latino population which make up 62.6% of the nation's total. African figure is spot on. The national Asian population is given as 4.4% so Asians seem over-represented four times. American Indians are 0.8% of US population so they seem over-represented here 6 times. That oddly American category, Hispanic was either not sampled or otherwise described.  Of course, the genetic techniques used in the study need not match perfectly with the US census classifications, but the authors could have sorted this out for us by commenting on the representativeness of their sample.

To put it at its kindest, the authors have missed a trick. They could have given the parents the psychometric test battery for good comparability, or even the very quick Wordsum test as a crude estimate, and then they could have contrasted ability with education and social status. As far as we can deduce from large scale genetic samples, both intelligence and social class have a significant heritable component. To avoid measuring that in parents when all the children have genetic and psychometric measures in place is a great pity. Perhaps all this has been done, and is being held over for another paper, but presented in this way, and described to the Press in the manner the authors have done, is very likely to mislead the average reader about the relative power of genetics and social status in brain development.

The paper and the comments will lead readers to believe that lack of money is stunting the brains of poorer children. This is possible, but not proved by this study because of obvious genetic confounders. The authors should have made it clearer that although they had the opportunity, they did not test the obvious and well established fact that different families have different abilities, and that within families siblings differ in their abilities (by about two thirds of the population variance). These differences in ability, even within families of a particular social class, lead to jobs which are more or less well paid, and thus people of different abilities achieve different social status. We know from proper epidemiological samples (Nettle, 2003) that intelligence at age 11 has more effect on achieved social class than does the original social class of origin (which is what is being measured in this study).

http://drjamesthompson.blogspot.co.uk/2012/11/social-class-and-university-entrance_28.html

Absent Third world malnutrition (itself increasing rare across the world), brain development, intelligence, health and social status contain a large genetic component. The 1099 brain scans of this study could have told us some very interesting things if coupled to data about the abilities of the parents.

What do you think?

33 comments:

  1. After a while, don't you just feel like this?

    ReplyDelete
  2. 1. Do wet streets cause rain, or is it the other way around? I forget sometimes.

    2. Common sense suggests that inheritance and environment influence cognitive development. The extreme example of the latter is hitting a kid's head with a ball-peen hammer (not recommended). Since this paper was designed to include the lowest SES fractions of the US population (except those with specified disorders), some mention should have been made of the known bad actor: childhood exposure to lead. I am unsure whether a good marker exists for older children and adults; blood lead level likely isn't the right proxy. Perhaps there are non-invasive way to measure lead integrated into bone. Census tract would likely be informative, ZIP code might well be.

    ReplyDelete
  3. "many leading social scientists and neuroscientists believe that ...": especially Comrade Lysenko, perhaps. How pathetic, how absurd, how disgraceful, to appeal to Fathers of a secular Church instead of intelligently gathering data. Shame on them.

    ReplyDelete
  4. Hey I don't know if anyone noticed yet, but the excel bits of their supplements contain what looks like a large part of their raw data (as in, family income, education, brain volume measurements, working memory, flanker scores, genetic admixture data (!!!), and perhaps more). Unfortunately I don't see the reading or vocab scores yet and you'll need to merge across multiple worksheets here, but this could make for some interesting analysis.....

    Obviously brain volume measurements would need to be adjusted for age and sex, but doing admixture analysis here could prove pretty powerful imho. I plan on looking at this soon myself.

    - FranklinDMadoff (randomcriticalanalysis)

    ReplyDelete
    Replies
    1. Franklin thank you for alerting me to this. have now found them and, as you say, open to be reanalyzed. Would you like to go first, please? Happy to give you space to post up your results here.

      Delete
    2. Great observation! I have downloaded all the data. This allows us to compute racial and gender means, SDs. The data has income, various volumes and age too.

      Too bad it does not contain the racial admixture data, then we could compute whether European admixture% in Africans predicts brain size as expected by genetic theory.

      I also think they did not mean to share this. So download it quickly before they realize their error.

      Delete
    3. Emil & James: You're welcome. It *does* contain racial admixture data! The gene categories are rounded to the nearest whole number/integer... if you increase the number of decimal points you'll see it.

      - FranklinDMadoff

      Delete
    4. I mean, they're "rounded" via Excel formatting, but the raw data still have admixture data in the the columns to some level of precision -- you just need to increase the number of decimal points in the excel formatting or make sure that you preserve the decimals when you export it out....

      -FDM

      Delete
    5. And yes, I would guess this is most likely an oversight on their part :-)

      That said, perhaps if someone with the "proper" academic credentials requested the other cognitive measures (reading and vocabulary) for replication purposes, they *might* share it. It's a long shot, of course, but why not?

      - FDM

      Delete
    6. Look forward to your findings on this data set. I had already downloaded copies. By the way, the roundings in the "bins" on education and income are quite crude, and probably affect the interpretation of the lower scores. About to post on that.

      Delete
    7. ALERT check the data quality before doing any statistical analysis. Post follows shortly

      Delete
    8. In fact, I am still working through things, so post will not follow shortly!

      Delete
  5. @James - Most published research, and almost published research in the human sciences, is worse than worthless - because it is actively misleading. This means that (unless we know and trust the honesty and incompetence of the author/s - which usually means authors within our own invisible college) our default approach to research should be disbelief and ignoring. Sometimes this is easier, when the authors advertise their own dishonesty - as in this case; but life is too short and precious to expend time on demonstrating exactly *why* the work of self-advertized liars cannot be trusted - and should not be even passed-through the mind (for fear of the unnoticed damage which it may do).

    ReplyDelete
    Replies
    1. Sadly, much social science research is replete with unexamined assumptions, but I try to examine them, and hope that such enquiry becomes more general. Some researchers do not tell the truth, the whole truth, and nothing but the truth. The problems, I believe, are in telling "the whole truth". That is the bit that gets corrupted, for supposedly noble reasons.

      Delete
    2. These tendencies are not unique to the social sciences. Here is a 2010 essay on analogous shortcomings in the field of climatology. Climategate: Not Fraud, But ‘Noble Cause Corruption’. Unsurprisingly, the field has not healed itself in the ensuing half-decade.

      Delete
  6. The measure of family income seems not so precise, from the description in the paper. Total yearly family income--does that include such things as Medicaid, Medicare, and other state and federal benefits? The cost of living and the local welfare benefits provided vary greatly in different states and towns.

    "Parents were asked to report the level of educational attainment for all parents in the home." Biological parents? Stepparents? Educational attainment for biological parents not in the home was not collected?

    The statements made to the press in interviews about the paper seem much less careful than the paper's text.

    ReplyDelete
    Replies
    1. The authors say that the income estimate was not divided by the number of people in the household (I think they inherited all this from the Ping study) so they cannot be sure how much money was available for each target child who was later scanned.

      Delete
    2. The correlations observed could be the consequence of seventy years of using standardized tests as gatekeepers for admission to higher education, in an economy which rewards intellectual accomplishments.

      Delete
    3. Well, admission to higher education should be based on standardized tests of scholastic ability, and the best and most fair procedure, bar none.

      Delete
    4. "admission to higher education should be based on standardized tests of scholastic ability, and the best and most fair procedure, bar none." Not so. Admission occurs at an age when youngsters, particularly males, are still changing quite a lot. So I recommend tests not only of scholastic ability but also of its first derivative. Seriously.

      Delete
  7. to be honest I find the article quite upsetting for many reasons. First the authors have a unique and ideological way of framing poverty. Poverty is not a universal concept. On the contrary is a relative notion that highly varies according to culture and geographic areas. Measuring wealth using monetary income is totally misleading. the same income implies big difference in quality of life if you live in Manhattan or in the countryside! Second, I don't see how over a certain level of income can make the difference. If I earn 100k and you 200k how can your childcare be better than mine?! In this sense the study is seriously affected by an ideological bias, a neoclassical tendency to utilitarianism i.e. the more I get the better I'm... Then it seems that they totally ignore the 'quality' of childcare... of course this can be influenced by income, but not always... many studies reveal that some cognitive functions of people living in the jungle are well developed, they can solve some problems much more easily than urban people. How do we measure their income?! Finally I really feel depressed to notice the obsession to find 'scientific' explanations to class/gender inequality. Why Anglo-Saxon scholars are so obsessed to find a scientific explanation to inequality?

    ReplyDelete
  8. The empirical project is to find scientific explanations for everything. Anglo Saxon scholars have many interests, and if they are obsessional it has served them very well. Difficult problems require sustained enquiry. A highly plausible hypothesis is that people differ in achievements because they differ in abilities, and that is worth testing.

    ReplyDelete
  9. A highly plausible hypothesis is that people differ in achievements because they differ in their starting conditions, possibility, class, power etc. but this is an inconvenient evidence. It's better to run 'neutral experiments' and prove with scientific method that rich are more intelligent :)

    ReplyDelete
    Replies
    1. Neutral experiments have to be run on all hypotheses, not to prove a prior assumption but to test it against other explanations

      Delete
    2. there is no such as thing as 'neutral experiments'. The study is fundamentally biased by the myth, especially very popular in the anglo-saxon societies, that people earn more because they deserve it, they have better capabilities and so on. This is a blatant fallacy as proven by the almost total absence of social mobility. Most of rich people are doing better because they have more power, stronger social networks and easier access to resources and.... even influence on 'independent' researchers who can justify they are rich because they are 'better'. very old pre-enlightenment ideology Mr Thompson.

      Delete
    3. If there is no such thing as a neutral experiment, would you reject the finding that the adoption of a poor child into a wealthy family boosts their intelligence?

      Delete
  10. "These differences in ability, even within families of a particular social class, lead to jobs which are more or less well paid, and thus people of different abilities achieve different social status."
    I am certainly very late in adding to this string but this line in Thompson's commentary on the study caught my attention because it does not for a moment consider racial or class discrimination affecting where the children of these families end up in certain educations and jobs.

    ReplyDelete
  11. Within families, not between families. That is the point of the observation.

    ReplyDelete
    Replies
    1. Yes, I see your point. However, isn't the point that individuals with differing abilities coming from those families in discriminated social groups are at a disadvantage? How do you take account of the effects of racism?

      Delete
  12. Analysis of variance. A family suffering from racism may have an overall lower achievement score, but the within family variance will still give us a measure of intelligence and personality differences. Similarly, a family enjoying wealth will still show considerable within family differences in outcomes. Adoption studies also contribute relevant findings. Intervention studies as well. You take account of things by conducting research, and comparing results.

    ReplyDelete
  13. *Household* income potential depends on different factor for men than for women. I once counseled a young lady with all of two years of high school to her name who had never made more than $40,000/year on doing a career change from the stressful, on-your-feet occupation she currently had. Her new gig affords her an 8-bedroom coop on Park Avenue and a private jet! My advice?

    "Marry Well".

    There is no similar option for us males, alas...

    ReplyDelete
  14. I think that studies like this need to go deeper and get at both beliefs about mating and actual practices. Some cultural backgrounds require compulsory marriage and breeding and base their social status system on kinship achievements whereas other groups follow assortative mating practices and delay marriage and breeding which permits them to have accumulated greater capital, all forms considered, which their offspring inherit or acquire through forms of interaction, linguistic and social, which relate to economic achievement more that kinship achievement. Another way to put it is that kinship terminology may have a more powerful role in some cultures over economic role terminology.

    ReplyDelete