In a previous post looking at the relationship between a country's average intelligence and the ideal age advantage of men in the eyes of the country's women, Agnostic wonders if there aren't more reliable sources for IQ data than Lynn and Vanhanen. Specifically, he points to some digging by Dienekes that seems to show Lynn inexplicably adjusted Greek IQ downard a couple of points in one of the two samples used to compute the nation's average. Dienekes argues that the other result L&V use (showing an average IQ of 87) should be given little weight because it was conducted in 1961 on 9-14 year olds before universal education had been adopted in Greece.
Other aptitude testing results lend credence to the L&V estimate, however. The Programme for International Assessment (PISA), tests 15 year-olds every three years in the areas of reading, math, and science. PISA results convereted to IQ scores on a scale with a 15-point standard deviation puts Greece at 91 in '06, one point lower than L&V's estimate. In '03, PISA scores suggested an IQ of 92, identical to L&V's most recent data, so large fluctuations don't appear to offer an explanation.
L&V's numbers hold up pretty well against PISA results. For the 50 countries where both L&V and PISA estimates are available for '06, the correlation between the two is a statistically significant .81 (p=0). This relationship is similar to the relationships found between the figures Lynn lays out in Race Differences in Intelligence (p173-175) and the International Studies of Achievement in Mathematics and Science, for which 52 countries provide results. Lynn's estimates correlate with these scores in the range of .81-.89, depending on the year and subject of the test.
Unfortunately, these tests only extend to OECD countries and a handful of middling countries that are trying to get in, like Russia and Brazil. L&V's estimates allow for far more extensive international correlations to be searched for.
GNXP's Jason Malloy suggested I look to a site by Volkmar Weiss for additional numbers from German professor Heiner Rindermann, who composed his own estimates by combining L&V estimates with results from PISA, PIRLS (a reading assessment), and TIMSS. I'm not sure where Rindermann's numbers come from for many third-world countries, as only around 50 or 60 mostly developed countries participate in them, but I may be missing something (the paper is not open access). Whatever his methodology, these results correlate at .94 (p=0) with V&H's estimates. That would be more telling if I knew how much weight Rindermann gave to V&H's data in coming up with his estimates, but they're essentially interchangeable regardless.
Weiss makes the case that the small variances between L&V and PISA suggest the existence of dysgenic effects in some countries*. His assertion is based on the differing periods of time the respective tests are measuring. L&V are using figures derived from people who are today somewhere between midlife crisis and the grave. PISA and others like the aforementioned PIRLS or TIMSS, in contrast, are looking at contemporary teenagers. I've suggested something similar may be happening in the US, specifically in the Southwest, where estimates several decades old vary most greatly with contemporary estimates based on NAEP scores (they are stable over time in the rest of the country).
Given their consistent relationship with other international testing results, the L&V estimates remain among the most thorough and wide-ranging available, so I'll continue to use them.
I could substitute Rindermann's figures, which are as expansive as L&V's are, but the extremely low values given to sub-Saharan Africa make me less comfortable with their reliability. For example, Sierra Leone is given a 59, and that's with L&V's 63 pulling it up to some extent. If L&V are given half weight, that'd mean the other tests put the country's average at an almost unheard of 55 (that'd put the average Sierra Leonean at about the 0.1 percentile in the US). Somalia strains credulity to the greatest extent of all. L&V put it at 84 to Rindermann's 58. At a half-weight, that's 32 (!) from non-L&V data. Even at one-fourth weight, it's only 51.
At almost non-existent levels of education in many of African countries, scholastic aptitude tests are probably going to tend to understate actual IQ relative to tests like Raven Progressive Matrices. Anyway, I see no compelling reason to assume Rindermann's are necessarily superior. I suppose I could run both sets to try to better distill real relationships, but as we're looking at estimates not absolutely precise values, and they are already so similar, that won't add much value.
The data are here.
* Among those said to be undergoing dysgenic changes in average intelligence: Hungary, Luxembourg, Germany, Greece, Poland, Spain, Italy, the US, and Mexico.