IQ & malleability, again

Someone pointed me to a recent study — Ramsden et al (2011) — which seems to have gotten a bit of press (e.g. “Congressman Fattah Cites Study Showing Malleability of IQ in Teenage Brains“). [To summarize the study: 33 kids had their intelligences measured and brains scanned at ages 14 and 18; the test-retest correlation was 0.8; the range in IQ change between intervals was 20 points; and the IQ change correlated modestly with brain differences.]

As for the study, the authors’ comments notwithstanding — ” Neuroimaging allows us to test whether unexpected in measured IQ are related to brain development” — the found IQ changes are in line with what is typically found. Here is a table out of Brody (1992 p. 233; Intelligence, Chapter 8: Continuity and Change in Intelligence). As can be seen there can be considerable change in IQ scores between years, though the volatility both in terms of mean and range decreases with age.

Likewise, as would be expected from the above, the age to age reliability of test scores increases with age.

So the “unexpected longitudinal fluctuations,” per se, are of little interest. What is is that the score changes correlated with brain changes. As such, the fluctuations can not solely be attributed to measurement error and non-ability/developmental differences. Of course, this implies that measures of IQ have more validity than is sometimes thought (i.e. the known fluctuations in scores across development actually track fluctuations in ability, as indexed by cortical volume). Imagine the press if these results were not found (e.g. “Congressman Fattah Cites Study Showing the lack of validity of IQ in Teenage Brains”)!

But what about malleability? The paper clearly shows the changeability of IQ and the brain in this sample, but it’s not clear how much of this runs from the outside in. A while back I pointed to a few studies which found that the causality between IQ and brain volume largely runs from the inside out:

In a more recent paper by van Leeuwen et al. (2009, also see De Moor et al. 2008), the authors argue that if the causal path runs from cognition to brain, then there should be both environmental and genetic correlations between cognition and brain, since there are significant environmental and genetic effects on cognition, which would then be passed onto brain in the causal chain. Since they found only genetic but not environmental correlations between brain volumes and cognition in this study, they argue that only a causal path from brain to cognition or pleiotropy (possibilities 2 and 1 above) are consistent with their data. In the current results, since we also found significant genetic but not environ- mental correlations between brain and cognition, the most likely causal models are either a causal path from brain to cognition or pleiotropy. The same argument could be made from Posthuma et al. (2003)’s study with adults, which also did not find both environmental and genetic correlations between any cognitive and brain volume measures. (Genetic Covariation Between Brain Volumes and IQ, Reading Performance, and Processing Speed)

At least for g. Which is another issue, as the authors of the Ramsden et al (2011) note:

The locations of the grey matter changes associated with VIQ and PIQ changes do not correspond to the anterior frontal and parietal regions associated with general intelligence7 (g factor). It may therefore be the case that g remains relatively constant across ages, but changes in the ability to perform individual subtests depend on changes in sensorimotor skills. It is also notable that although completion of the subtests comprising verbal and performance measures must implicate a network of brain regions, only structural changes in regions associated with sensorimotor skills showed correlations with changes in VIQ and PIQ.

It would have been informative if they extracted g out to see if or to what extent the structural change correlated with it. General intelligence (and genes for it) is largely what gives stability to IQ scores, after all. In an analysis of the Capron and Duyme (1989) study, a study which is frequently cited in support of claims of IQ’s malleability, Jensen found that the adoption effect was not a Jensen effect. One likewise probably would find that the changes here were not g changes.

The take home message of this study (or message worth taking home) is that while cognitive abilities, at least lower level ones, are stable, they are not fixed and can fluctuate, especially during the development period. But there is really nothing new here.

Advertisements
This entry was posted in Uncategorized. Bookmark the permalink.

18 Responses to IQ & malleability, again

  1. B.B. says:

    On the subject of the malleability of IQ, I’d like to get your opinion on this article. Apparently it claims that based on NLSY data, black IQ scores are boosted four times as quickly as white IQ after attending college. At least that is how it has been summarized in the environmentalist literature, and looking over articles that have cited it on Google Scholar, I didn’t notice any hereditarians that have discussed it.

    • Chuck says:

      So that’s the paper. I would hear references to how Blacks had massive IQ gains in college but never know what the source was. The major problem with the findings is that Blacks don’t close the IQ gap in college. If they did, there would not be a huge graduate school gap. So you can cite the GRE/LSAT/MCAT/etc gap as counter evidence.

      As for the common interpretation, the problem with this an other multiple regression studies is that when controlling for college attendance and SES, you are effectively controlling for g and genes to some unknown extent. The IQ of a subset of Whites is compared to that of a subset of Blacks with the same SES and Education; but both subsets are phenotypically and genotypically unrepresentative of their populations.

      • Cognition says:

        I know this is an old discussion, but I just want to clarify:

        Is the reason for the supposed Black IQ increase in that study of college students largely if not entirely the result of the unrepresentative sample being studied? (survivor’s bias)

  2. B.B. says:

    The major problem with the findings is that Blacks don’t close the IQ gap in college. If they did, there would not be a huge graduate school gap. So you can cite the GRE/LSAT/MCAT/etc gap as counter evidence.

    As for the common interpretation, the problem with this an other multiple regression studies is that when controlling for college attendance and SES, you are effectively controlling for g and genes to some unknown extent. The IQ of a subset of Whites is compared to that of a subset of Blacks with the same SES and Education; but both subsets are phenotypically and genotypically unrepresentative of their populations.

    It didn’t close the gap, but reducing it by roughly 1/2 is nothing to sneeze at. Of course we are dealing with a non-representative subset of the black population, but that doesn’t really deal with the central issue, which is why did the black subsets scores increase significantly more than whites?

    Something to consider is that Rushton & Jensen have pointed out that on tests like the WISC, the narrowing of the black/white gap is most apparent on the subtests that are the least g-loaded. Since the AFQT consists of multiple sections (arithmetic reasoning, word knowledge, paragraph comprehension, numerical operations), couldn’t we look at how g-loaded each individual section is and in which section the scores increased the most to determine whether the gains represent an increase in actual g?

  3. Chuck says:

    “It didn’t close the gap, but reducing it by roughly 1/2 is nothing to sneeze at. Of course we are dealing with a non-representative subset of the black population, but that doesn’t really deal with the central issue, which is why did the black subsets scores increase significantly more than whites?”

    The data here was not longitudinal. Each of the 723 (see table 2) subjects in the second analysis was given the AFQT once (in 1980). Their scores were plotted with their level of education at the time the test was given. This produced the graph you see in Figure 2.
    So there is no direct evidence that the individual “Black scores increased.” What you are seeing is an association of increasing IQ with increasing education for the college sample.

    The numbers are actually rather small. In table 2, it says that the sample size was 723 with 16% of the sample being Black — If I’m reading it right. Assuming the 115 Blacks were evenly divided across the 9 grades, each point has N=13. So we’re really comparing about 50 future college grads who took tests in college to 65 who took tests in pre-college. (The 0.5 SD difference between the blacks who took the test in grades 13+14 and grades 15+16, (assuming n=26,26), doesn’t reach statistical significance.)

    Anyways, so what could be the cause of the gap between blacks who took the test in college and Blacks who took the tests in high school? One interpretation, which the authors mention, is that these were a genotypically selected bunch, relative to the Blacks population at large, all being future college grads, but that the IQs of Black who took the test in High School were artificially depressed by environmental factors (i.e. Blacks High schools). To investigate that hypothesis, we could compare, as you mention, the g-loadings of the High school group with the college group.

    Mine would be that the results were largely spurious. Again if Blacks closed the gap in college to .5 SD, there wouldn’t be a gap of the magnitude there is. We could check this by looking at the NLYS 97 data. Or we could look at longitudinal based studies like this one (Flowers and Pascrella, 2003. Cognitive Effects of College: Differences between African American and Caucasian Students. Research in Higher Education, Vol. 44, No. 1 (Feb., 2003), pp. 21-49) which have much larger sample sizes (and show no gap decrease — but rather increases — with years attended (assuming the CAAP correlated with IQ).

    • JL says:

      You beat me to it regarding the sample sizes. The NLSY sample may have more black college grads these days so the robustness of their results could be tested.

    • Chuck says:

      Apparently, those supposed 0.5 SD IQ gains in college don’t reflect on other measures. In a more recent study, Roth and Phillip (2000) found a .8 SD GPA gap (“College grade point average as a personnel selection device: Ethnic group differences and potential adverse impact” http://psycnet.apa.org/journals/apl/85/3/399/). The IQ-GPA correlation is about .5, so other factors much to involved.

  4. JL says:

    If I’m reading it correctly, the Myerson et al. study shows that black college grads (or future grads) tested at age 22 have slightly lower AFQT scores, on average, than 14-year-old whites who will later attend college and graduate. I wonder if the black means at the nine different ages in Fig. 2 are based on adequate sample sizes. The total black college grad sample size is 116, which is < 13 students for each age, and I doubt they're evenly distributed. The high school grad sample has a larger number of blacks and fewer checkpoints, and the results are quite different from the college grad sample. That paper is terse to a fault, and lacks many relevant analyses that could have shed more light on the plausibility of its claims.

    Another thing that might have some effect is that the AFQT was not designed to test high-ability individuals, so many (white) people in the college sample may have hit the ceiling of the test.

  5. statsquatch says:

    Does Congressman Fatah understand that he is lauding a study that validates the IQ test? If changes in IQ are correlated with real changes in the brain then “g” is not some statistical artifact or tool of white heterosexist male domination.

    • Chuck says:

      So, two typical spins/interpretations were that 1) IQ scores are invalid and that 2) IQ scores are unstable. For example, The Washington Post states:

      “The varying IQ scores could also indicate the test itself is flawed. “It could be a real index of how intelligence varies or it could suggest our measures of intelligence are so variable,” said neuroimaging pioneer B.J. Casey at Cornell University’s Weill Medical College, who wasn’t involved in the study.”

      And Science Daily notes that:

      “Across our lifetime, our intellectual ability is considered to be stable, with intelligence quotient (IQ) scores taken at one point in time used to predict educational achievement and employment prospects later in life. However, in a study published October 20 in the journal Nature, researchers at the Wellcome Trust Centre for Neuroimaging at UCL (University College London) and the Centre for Educational Neuroscience show for the first time that, in fact, our IQ is not constant.”

      The point about construct validity seems not to have struck many; I’m not sure why.

      What I was surprised about was that even the authors seemed to interpret the findings as showing (2) (that IQ scores are unstable and subject to “unexpected fluxes.”), despite the 0.8 test-retest correlation. They seemed to be unduly fixated with the range of changes. It’s a good think they didn’t test and retest at 1 and 5!

      • statsquatch says:

        I think the authors were being disingenuous and were trying to hype their study. They know 33 subjects is not enough (particularly since they started with 44) to overturn stability research. What they what to point out is that it is not 95% stable and that lack of stability is not measurement error. I never would have heard about it if they had been more honest.

  6. Chuck says:

    statsquatch :
    I think the authors were being disingenuous and were trying to hype their study. They know 33 subjects is not enough (particularly since they started with 44) to overturn stability research. What they what to point out is that it is not 95% stable and that lack of stability is not measurement error. I never would have heard about it if they had been more honest.

    My point was that their findings were not inconsistent with the stability research from the start. The year to year test-retest correlation is only .95. And there is an age dependent factor here.

  7. Doug1 says:

    she has much to now answer for.

  8. Meng Hu says:

    This might interest you, Chuck.
    http://books.google.fr/books?id=sMSWbI23RMUC&printsec=frontcover#v=onepage&q&f=false
    See pages 66-68.

    See also
    http://edpsychassociates.com/Papers/WISC3LongStability(1998).pdf

    “As expected, test-retest reliability coefficients for the WISC-III subtests were generally lower than the IQ and Factor Index scores, ranging from .55 (Symbol Search) to .78 (Block Design) and resulting in a median r = .68. As with the IQ and index score correlations, all subtest stability coefficients were statistically significant, p < .0001 (see Table 2)."

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