I had several articles planned for this weekend, but I wanted to comment on a recent article by Chanda Chisala (hat-tip to commenter Race Realist for telling me about the article and helping me to locate the references). I admit that I haven’t read the entire thing, nor did I read the article’s prequel in its entirety, but the bottom line seems to be that the large number of African scrabble champions proves that HBDers are severely underestimating the intelligence of black Africa.
My argument is therefore not against the low IQ score estimates for African nations (by Richard Lynn, et al), but whether this reflects some restrictive racially linked genetic cause. If it is indeed basically genetic, it should practically be impossible to find any area of relative cognitive performance of Africans that is inconsistent with this large IQ deficit with whites and other groups…We can ignore all the statistical arguments and actual testing evidence indicating that the world champion level players would exceed IQ 140 or perhaps even IQ 150 (since Putnam Fellows have won, but no women have won), and conservatively assume that only IQ 130 is needed for such extreme distinction.
The self-reported SAT scores of scrabble experts is 1408 which translates into an IQ of about 136 (white norms) or 144, depending on whether they took the new SAT or the pre-recentered SAT in the early 1990s . Given that their median birth-year in the above linked study was in the mid-1970s, probably an equal mix of both, so let’s split the difference and assume IQ 140. But this might be a large overestimate given that people with lower scores tend not to self-report them.
Nonetheless, assuming their average IQ is 140, the minimum IQ to be a scrabble champ would be dozens of points lower than the mean, just like the minimum height to be an NBA player is more than a foot shorter than the mean height on the NBA, as 5’3″ “Muggsy” Bogues can tell you.
There should still (statistically) be no single person from African countries like Gabon. And yet they exist, constantly outperforming math professors and computer scientists from the developed world. That’s a statistical problem for the racial hypothesis but it is not a problem at all for the alternative hypothesis: the African nominal national IQs are artificially depressed by more than 30 IQ points due to an extremely deficient cognitive environment.
I don’t doubt that the average sub-Saharan IQ of 67(as estimated by Richard Lynn) is substantially supressed by environment. The environmental suppression probably contains both a biological component (malnutrition and disease) that lowers real intelligence, and a cultural component which lowers performance on tests without harming functional ability, so not only is their real biological intelligence below its genetic potential, but their test scores underestimate their real biological intelligence.
The black standard deviation is NOT 12!
But even if their real biological intelligence were indeed only IQ 67, and even if it were 100% genetic, which not even Richard Lynn has ever come close to arguing, there would still be many sub-Saharan Africans with IQs of 130+.
Chanda’s assumption to the contrary is based on his mistaken belief that on a scale where white Americans have a mean IQ of 100 with a standard deviation (SD) of 15, blacks have a standard deviation of only 12. Under this assumption, the black African mean IQ of 67 and population of say, 988 million, would predict only about 99 people in all of black Africa who have IQs of 130+.
While Chanda cites sources as impeccable as Arthur Jensen himself to claim that the black American SD is only 12 (which Chanda takes the liberty of generalizing to black African countries), this figure is based largely on studies that were more interested in determining the average black IQ, not the SD.
It’s much harder to estimate a population’s variability than it is to estimate its mean. A mean can be estimated by finding a sample of average folks conveniently located in an average part of town or average class, but estimating an SD requires either random sampling or carefully making sure that every segment of a population is included in your sample in the same proportion that they exist in the census. Black populations are notoriously difficult to sample because they have a large underclass that is inaccessible to testers and who score close to the floor of the test, masking their full variation.
The gold standard for IQ testing and norming are the Wechsler intelligence scales which (at least since the 1990s) carefully make sure that the educational distribution of blacks and whites in their norming sample, match the educational distribution of blacks and whites in the census. This is crucial for getting a correct SD because if educational extremes (college grads vs 8th grade dropouts) are not correctly represented in your sample, you will miss some of the cognitive extremes in the general U.S. population which contribute to the variance. When calculated on a scale where the white U.S. mean is set at 15, the black SD was 13.48 for the 1995 norming of the WAIS-III and 15.43 for the 2006 norming of the WAIS-IV. Both figures are higher than 12.
And of course all this assumes that Chanda is correct in applying black America’s SD to black Africa. Direct data from black Africa tells a different story. In one of the largest studies ever done on black African IQ, an astonishing 1,093 black South Africans drawn from 28 schools were compared to 1,056 white South Africans drawn from 20 schools, on the Raven Progressive Matrices. The White South Africans correctly answered 45.27 (SD = 6.34) of the 60 items, while the blacks averaged 27.65 (SD = 10.72).
If we convert the white mean and SD to IQ 100 and 15 respectively, than the black mean and SD become 58 and 25.4 respectively!
Despite the extremely low mean of 58, the colossal SD of 25.4 implies black Africa would have more than two million people with IQs of 130+. Of course such a low and wide distribution probably suggests there’s something wrong with the data (Raven raw scores may not be normally distributed because of large and uneven jumps in difficulty between test items) but at the very least it shows the black African SD is higher than 12 and is quite possibly higher than the white SD.
Regression to a much lower mean
But the larger problem with Chanda’s thesis is that scrabble is simply not an IQ test. Even if the average American scrabble champ has a mean IQ of 140, that tells us very little about the IQs of black African scrabble champs because scrabble is just a crude proxy for IQ, not a valid IQ test itself. Research suggests performance at intellectual games like chess (and presumably scrabble) only moderately correlate with IQ because of the huge role of practice, study, and special talents.
We also don’t know the minimum IQ to be a scrabble champ, because that minimum is a function of all the variables that can influence scrabble skill which are very different in a county like Nigeria where scrabble is state sponsored and incredibly prestigious than in the United States where only a few thousand people play competitively.
To make an analogy, the average height of male weight champions (people over 980 lbs) is an impressive 6’1″, but the average height of female weight champions is only 5’5″. So even though weight is a rough proxy for height within both sexes, because the worldwide height gap between men and women is so large, even at the highest extremes of weight, men are an astonishing 3 SD taller!
For all we know the same could be happening with scrabble. Despite the fact that scrabble is a rough proxy for IQ within populations, because the U.S. IQ is so much higher than black Africa’s IQ, the average American scrabble champ (IQ 140?) might be dozens of IQ points higher than the average black African scrabble champ.
One can’t assume that Americans and Africans matched on scrabble skill will be matched on IQ anymore than one can assume men and women matched on weight will be matched on height. Because weight and scrabble are only rough predictors of height and IQ respectively, people who are extreme on the former variable will regress precipitously to the mean on the latter variables, and if they come from very different populations, they will regress to very different means.