Friday, October 10, 2008

International student assessments (Rindermann) and L&V IQ scores compared

In the previous post, I looked at the relationship between international student assessment results and IQ scores as gathered by Lynn and Vanhanen in response to concerns the L&V figures are too unreliable. Running L&V and PISA (math, reading, science for 15 year-olds) '06 yields a correlation of .81.

Bettering that, Heiner Rindermann provides data aggregating results from multiple international student assessments given in various years to various students in several countries. This gets around potential isolated data errors and statistical anomalies. Jason Malloy pointed me to a site run by Volkmar Weiss presenting Rindermann's numbers, from which I sourced the nearly perfect correlation of .94 between the national IQ estimates of Rindermann and L&V.

But Weiss made some transcription errors. For Somalia, for example, Rindermann doesn't report any student assessment scores at all and assigns an adjusted IQ of 68 to the country. Weiss shows him reporting a 58 and V&H reporting an 84. But in IQ and Global Inequality, L&V give Somalia 68, the same score Rindermann assigns, as he has no assessment results for the country. I'm not sure what L&V estimated for Somalia in IQ and Wealth of Nations, which is where Weiss gets his V&H numbers, but such an enormous gap seems impossible, since in both cases Somalia is estimated from surrounding countries). There are other instances of similar errors on Weiss' site.

Knowing that Rindermann factored L&V's numbers into his own, it was clear the real relationship could not be that strong (the problem with Weiss' transcriptions aside). But how strong is it? I threw my hands up and said in lacking access to Rindermann's paper, I didn't know. Thankfully, my readers are more resourceful than I am. Bruce G. Charlton and Jason Malloy each sent me Rindermann's actual paper. The German professor explains how much weight L&V are given:
I calculated one total score for all cognitive ability studies (IQ, student assessment studies). Student assessment studies [PIRLS, PISA, and TIMSS in aggregate] were given a double weight in this total because they have newer and larger samples and they consist of more cross-national studies.
The next step is to expand the 1-r gap by 50%, yielding a correlation of .91. It had initially struck me as curious (I don't quite miss everything) that Rindermann had estimates for as many countries as L&V did, since the assessments Rindermann used are administered in few third world countries and only a handful of middling countries.

As it turns out, for more than 100 countries (60% of the national total) Rindermann's only data source is L&V. There are no recorded results from any of the student assessments in these places. So most of Rindermann's figures are identical to the estimates L&V present in IQ and Global Inequality. Using only the student assessment converted aggregate scores for the countries where they are available and comparing them to L&V's estimates for the same countries (a total of 74) reveals a correlation of .85 (p=0).

This isn't surprising, as Lynn has shown that IQ results correlate with various international student assessments in the range of .81-.89. Well, Rindermann's aggregation of those assessments correlates with IQ scores at exactly the middle of that range.

Rindermann also presents his estimates with adjustments (five points subtracted from L&V estimates for countries without any actual data, age and participation rate adjustments for international student assessments). For the same 74 countries where assessment results are available, the correlation with L&V is a marginally stronger .86.

This increases my comfort in using L&V IQ scores. International student assessments, where available, are vigorously correlated with L&V. There is no apparent reason to presume L&V's IQ scores are less accurate than assessment results converted to IQ estimates are, and L&V provide numbers for more than twice as many countries as the international assessment scores do.

The data are here.

14 comments:

Peter said...

I find the 84-I.Q. estimate for Somalia highly dubious not necessarily because it's so different from other estimates, but because it's a reasonably decent number and it's for, well, Somalia!

Audacious Epigone said...

Peter,

Right. The Somalia number flagged me. Weiss showed that L&V had reported it as being 84, but that's a standard deviation too high. They report it to be 68, much more inline with the rest of sub-Saharan Africa.

Jason Malloy said...

There is no IQ data for Somalia. L&V estimated an IQ from neighboring countries in both books, and gave the same number. Rindermann gave the same number in his paper. Weiss simply reported fallacious numbers for both.


IQ&tWoN, p 79: Somalia 68 (Average of Ethiopia 63, Kenya 72)

IQ&GI, p 60: Somalia 68 (Average of Ethiopia 64, Kenya 72)

Rindermann 2007, p 705: 68 (Taken from IQ&GI) Adjusted: 60.

Weiss webpage: 84 (said taken from IQ&tWoN), 58 (said taken from Rindermann 2007).


If you think 84 seems too high, the latest Lynn paper for the Sudan reports an IQ of 83.5 in 1964 on the Goodenough test, and 95.6 in 2006. (for 4-10 year olds)

http://dx.doi.org/10.1016/j.paid.2008.05.016

Audacious Epigone said...

Jason,

Thanks for the affirmation. You really are at the forefront.

Audacious Epigone said...

Jason,

Re: Somalia, I meant Weiss' score putatively from Rindermann was a standard deviation higher than what Rindermann actually reported. I wasn't trying to make a judgment about the plausibility of an average score in the eighties being discovered.

Peter said...

If you think 84 seems too high, the latest Lynn paper for the Sudan reports an IQ of 83.5 in 1964 on the Goodenough test, and 95.6 in 2006. (for 4-10 year olds)

If indeed it's true that Sudan has an average IQ of 95.6, it's bad news for anyone who believes in the importance of IQ. European IQ levels wouldn't have prevented the country from turning into an impoverished genocidal rathole.

Audacious Epigone said...

Peter,

Black children develop faster, and the tests forming that astoundingly high average IQ were performed on children between the ages of 4 and 10. I wonder if Jason is able to offer an insight here...

Audacious Epigone said...

That is, black children develop faster than white children, who develop faster than Asian children do.

Stopped Clock said...

HA HA HA HA HA HA HA HA!

Sudan beats Israel!!! and Ireland! and Greece! And half of Asia!

HA HA HA HA HA HA HA HA!

But seriously, before we relocate our computing industry to Darfur, I must ask if anyone has more information. I think that the thing about Africans maturing faster might be part of the reason, but looking at the IQ table on http://www.isteve.com/IQ_table.htm seems to show that even on that very same test, other African nations' children tend to score in the 70s. So this is either an outlier (a bad sample), or a

Havent there been IQ tests in the USA which showed blacks scoring above 100? I believe the 85 is just an average of many many tests of all different types, with some of them showing blacks scoring in the 60s and others in the 100s. There's got to be some random upspurts in the data here and there.

Stopped Clock said...

unfinished sentence, sorry:

... or a very unreliable method of testing (but hey, the testing industry seems to love it, and Im sure they know what theyre doing)

al fin said...

Good point, AE. That's the problem with using IQ and achievement test scores of children to demonstrate narrowing of gaps between population groups as a whole. Liz Spelke tries to do that with male-female math scores, and J. Flynn tries the same thing with ethnic IQ gaps.

You have to follow the child through puberty then test them.

AE, are you familiar with cognitive trainings such as FastForword that claim to be able to raise the IQ of children. Perhaps Jason is familiar with such programs. I am wondering if anyone knows of data demonstrating that the IQ improvements last beyond puberty?

Stopped Clock said...

And (I truly apologize for posting three times in a row, I truly do) by "more information" I mean if anyone here who has access to the dx.doi.org paper linked above can tell me if the sample size seems big enough to avoid random sample error and if the authors of the paper (particularly Lynn) think that it is a good measurement overall.

Audacious Epigone said...

AF,

Not other than HeadStart. The gains disappear within a few years. If a reliable method of permanently raising IQ scores was extant, I think it would be widely known.

SC,

The Sudanese test also strikes me as an outlier that strains credulity.

I'm only aware of one widely referenced test of American blacks (from northern states following WWI) outscoring whites (from southern states during the same time period), from Sowell's Black Rednecks and White Liberals.

Audacious Epigone said...

AF,

Maybe iodine supplementation?