Arnold Kling  

IQ and Living Standards

PRINT
Innovation, Business Behavior,... Gas Tax Debate...

Randall Parker sent a link to a cross-country comparison of GDP per capita and the percent of the population with verbal IQ's above 106. The equation fit very well, which led me to wonder how such a model could explain the low income of Communist countries. The article includes this note.


We can reduce some of the scatter by noting that 12 of the 81 countries have their GDPs skewed by special circumstance. Their inclusion in the analysis obscures the relation between GDP and IQ. China's absurdly low GDP, for example, has nothing to do with IQ and everything to do with fidelity to a failed economic system. Likewise Cuba. I eliminated them both.

The author lists other countries that are excluded from the analysis. So in the overall model to explain living standards, the fraction of the population with high verbal IQ is not the only variable. Whether a country has a capitalist or Communist system also matters.

For Discussion. What other cross-country comparisons might raise doubts about the strength of the relationship between IQ and living standards?


Comments and Sharing





COMMENTS (16 to date)
Abiola Lapite writes:

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

"We can reduce some of the scatter by noting that 12 of the 81 countries have their GDPs skewed by special circumstance."

I think this little bit of special pleading illustrates the worthlessness of the methodology at work here. What is it about the word "IQ" that brings out the gullibility in so many people who ought to know better, including economists who are presumably more than passingly acquainted with econometrics?


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (MingW32) - GPGshell v3.10
Comment: My Public Key is at the following URL:
Comment: http://www.alapite.net/pgp/AbiolaLapite.txt

iD8DBQFAu2mwOgWD1ZKzuwkRAku5AJ9Y3O1eTrah4FPFuyMUhnWeQmFBHQCgh3z0
NM7XgsoRvHhkzP40Sfm9BIE=
=g4Cg
-----END PGP SIGNATURE-----

Xavier writes:

I disagree with Abiola. There was a fairly strong correlation even before those outliers were removed, and the outliers weren't at all arbitrary. They were communist or recently communist countries and countries with economies based on oil. It's not unreasonable to exclude them from the study because obviously they would blur the correlation between intelligence and GDP. It would have been better if he had a more rigorous standard for determining which countries should be removed. He doesn't exclude Iraq even though it doesn't fit the model very well. I would think that Iraq's GDP in 1998 would satisfy just about any definition of "special circumstances."

There may be other problems with the methodology, but I don't think this shows the whole thing to be worthless.

Lawrance George Lux writes:

The high verbal IQs could be the result, not the cause, of high GDP. My studies indicate that it is the proliferation of Consumer products which both raise living standards and verbal IQ. The access to both high-quality and low-quality Consumer products lead to defined Consumer choice, developing high verbal IQs. Lack of this Consumer product spread reduces economic evaluative verbal IQ. These above comments are based more on hypothesis than scientific observation, though I spend a fair amount of time on both sides of the American/Mexican border. lgl

Bruce Cleaver writes:

Lawrance George Lux beat me to the punch here. My first thought was that it is certainly a two-way street, as bigger GNP leads to better nutrition, healthcare, information availability, and the means to take advantage of all of the above.

sudhir writes:

GNP described by verbal IQ? Sounds neat. Wonder if western values like degree of openess, tolerence, democratic-inclinations etc can also be explained by IQ.
End of the day, its likely that any strong link could be the result of some yet-undefined underlying factor. The results we are seejg could merely be model equivalence results.

dsquared writes:

Lord, Arnold, you've got a PhD in economics from MIT. You don't need me to tell you why this is rubbish, do you?

My first comment -- Huh?

My priors: there is probably some link between intelligence and economic growth. However, I have some questions

a) how does one measure intelligence
b) even if one comes up with a reliable test for measuring intelligence, how on earth does one measure it "consistently" worldwide
c) how do we know the link isn't two-way?

Garett Jones writes:

I'm doing some research in this area, so I thought I'd join in....

In ongoing research with psychologist Joel Schneider of Illinois State, we've shown that this relationship between IQ and growth is highly robust. Out of 1330 regressions we run using only robust control variables, IQ is statistically significant in 99.7% of them. It easily outperforms every other education measure that we throw in.

It's not just a fluke. People who are good problem-solvers are likely to be good innovators and good imitators. Those are the kinds of folks I'd like to have working in my economy!

And since IQ can be increased by good nutrition, a solid education, and other factors, this is a ball we should keep our eye on if we care about the world's poor. References to the relevant literatures (both psych and econ) are in the paper.

Garett Jones writes:

Here's the link to our IQ/Growth paper:

dsquared writes:

Garrett, you need to bring a statistician into your project, fast. The following passage from the paper linked:

"Sala-i-Martin judges a variable to be statistically significant if the more than 95% of the mass of a growth variable's "cumulative distribution function" was greater than zero. In practice, this closely corresponds to the case where the average coefficient value across all regressions is two standard errors away from zero (where the standard error, likewise, is an average across all regressions)."

is dangerously wrong. The 95% point for the mass of a cdf is only placed two standard deviations away for a normally distributed cdf. A cdf of estimated regression coefficients is not approximately normal. You *have* to use Sala-i-Martin's nonparametric method to summarise the regression results for this reason. Simply using 1.96*SE introduces a completely unwarranted parametric assumption.

Bernard Yomtov writes:

This post is an embarassment.

All the various objections raised seemed right to me, though I confess to only a vague understanding of dsquared's statistical point.

In any case, comparing "verbal IQ" across widely differing countries strikes me as plainly absurd.

Garett Jones writes:

Responding to dsquared's and Yomtov's interesting points:

To dsquared's point first:
I considered running the nonparametric estimates, but it seemed like econometric overkill. Keep in mind that Sala-I-Martin's "robust" variables were typically statistically significant in only about *two-thirds* of his regressions.

Only equipment investment, percent Confucian, and a small handful of other variables were statistically significant as often as IQ is. IQ was statistically significant in 99.7% of the Jones/Schneider regressions.

Further, note that in my regressions, I'm only including robust control variables, while S-i-M included combinations of *all* variables, both weak and strong. So it appears that IQ does quite well, even when run through a tougher test.

If and when journal editors ask me to run extra robustness tests, I'll be glad to run them, but I think the large IQ coefficient estimates (see the final page of the paper) and the modest standard errors for IQ speak for themselves.

Regarding Yomtov's point: We don't use *verbal* IQ scores for most countries.

For example, the most widely used IQ test for non-literate peoples is the Raven's Progressive Matrices, a sort of visual pattern-finding test. Repeated studies by psychologists have consistently found that scores on these types of non-verbal tests are strongly correlated with results from more comprehensive IQ tests. Jensen's excellent book, "The g Factor," explains this at great length.

Jim Glass writes:

I'm kind of surprised nobody has mentioned the Flynn Effect.

Garett Jones writes:

Yes, the Flynn effect (which we discuss at length in our paper, linked above) is key to this discussion.

The Flynn effect is the long-term rise in measured IQ that has been discovered in every nation: An average, nationwide increase of 2 points per decade is one common figure that gets thrown around a lot.

(FYI: Our data, which come from Lynn and Vanhanen's book "IQ and the Wealth of Nations," takes account of the Flynn effect. The Flynn effect is sometimes called the Lynn/Flynn effect, mostly because it rhymes.)

One of the big areas of debate among intelligence researchers is (to use econo-jargon) how much of the IQ rise is "real" and how much is just "nominal." I.e., how much is due to people getting smarter via good education, better nutrition, and educational TV shows, versus how much is just due to teachers "teaching to the test." It's a lively debate.

My amateur judgment is that at least some of the rise is real. Part of the reason I think some of the rise is real is because the biggest increases in IQ over time come from people in the bottom half of the income distribution--so rich folks have always been smarter on average, but now the poor folks are catching up. That means that IQ inequality is falling over time in the US (in contrast to income inequality).

Also, as we note in the paper, some new research in Kenya provides tantalizing results indicating that with good nutrition, parental literacy, and a cleaner environment, much larger Flynn effects are possible in the developing world.

The possibility of *global* IQ convergence--with the lower-average-IQ nations catching up to the higher-IQ nations of East Asia--has obvious policy ramifications.

If IQ is a good measure of one form of human capital--as Schneider and I argue--and if human capital is an important driver of economic productivity--as a large theoretical and empirical literature argues--then policymakers in poor countries should be deeply concerned with whether their development agenda is raising their nation's average IQ .

This version of our paper focuses more on the statistical robustness of IQ's large impact on a nation's growth:
http://www.siue.edu/~garjone/JonesSchneApr.pdf

This (older) version has a lengthy discussion of the theoretical IQ literature written by my coauthor, a psychologist:
http://www.siue.edu/~garjone/JonesSchne.pdf

dsquared writes:

Why does the paper only consider quadruplets, btw (ie, IQ plus three other selections from the Sala-i-Martin list)? What happens if you choose all the combinations of five regressors, or three?

I also dispute that only using a subset of regressors, rather than the full set, is a "tougher test". It's only a tougher test if you are able to make precisely the kind of parametric assumptions that extreme-bounds analysis doesn't make.

Other comments: Sala-i-Martin's nonparametric tests aren't "econometric overkill"; they're more or less the only way that you can consistently estimate the cdf. The method you use of weighting estimates by R^2 doesn't look to me as if it's going to give consistent results. The data used to produce your figure IV makes it look as if you've done more or less all the number-crunching you need in order to do the nonparametric test so I don't understand why you don't just bite the bullet and do it.

Finally, I think that, given that you're trying to establish a connection between IQ and GDP, I think it's not valid to assume a constant Flynn effect across countries. Countries with higher GDP growth and better performance on development indicators would have experienced different IQ outcomes, which would bias your dataset for estimated 1979 IQ. You need some sort of two-stage approach to get starting date IQ estimates which are consistent with the overall model.

I really, really think that it would be a good idea to recruit a statistician to consult on this project, and I suspect that the editors of any decent journa will agree with me.

Garett Jones writes:

Responding to dsquared's very interesting questions:

> Why does the paper only consider quadruplets, btw (ie, IQ plus three other selections from the Sala-i-Martin list)? What happens if you choose all the combinations of five regressors, or three?

I wanted to follow S-i-M's methods as closely as possible. He used seven variables at a time: 3 fixed regressors (life expectancy in 1960, primary school enrollment in 1960, and log GDP per capita in 1960), 3 rotating regressors, and the regressor that he's testing (plus a constant, of course). That's what I did. Seven variables plus a constant in every regression, with IQ as the seventh variable.

Why did I take this approach? Because S-i-M's "Two million regressions" paper is canonicial. And since I was pushing myself outside the "academic canon" by including IQ, I thought I'd stay within the canon as much as possible for the rest of the paper. FYI, in S-i-M's newer "Bayesian Averaging" robustness tests of growth models, he finds that the optimal number of regressors is something like 7.4, if I recall correctly. So we're not too far off.

So, the short answer is that I don't know what would happen if I threw in more and more regressors into the same regression (it would be a snap to do it, though). But when I was a grad student at UCSD, Rob Engle taught us that he strongly favored smaller models (four or five regressors, at most), and I've carried that atheoretical, soft knowledge with me into my career. I have a prior weighted toward small models.

> I also dispute that only using a subset of regressors, rather than the full set, is a "tougher test". It's only a tougher test if you are able to make precisely the kind of parametric assumptions that extreme-bounds analysis doesn't make.

I always thought that if you're tossing out the garbage regressors, then the remaining robust ones are quite likely be eating up lots of the econometric space spanned by GDP growth. Sure, I can think of exceptions, but as a general tendency....But I agree with you that more thorough tests are better, if the audience's interest is unlimited.....(Which just might be the case on the subject of IQ.)

> The data used to produce your figure IV makes it look as if you've done more or less all the number-crunching you need in order to do the nonparametric test so I don't understand why you don't just bite the bullet and do it.

Well, you make a good point. Thanks for the advice. I just may do that when I get back from vacation (I'm in Southern California as I write this). It would be good to have the results on my hard drive just in case an editor asks me to present them.

>Finally, I think that, given that you're trying to establish a connection between IQ and GDP, I think it's not valid to assume a constant Flynn effect across countries.

I agree, the Flynn effect surely differs between countries--but there are two big barriers to creating country-specific Flynn effects in this particular paper.

1. Psychologists aren't close to a consensus regarding country-specific Flynn effects. For us to just jump into the debate and run some kind of 2SLS that creates estimates of such country-specific effects would probably annoy more psychologists than we were plannning to; it would be a bit imperialistic, a bit presumptuous, at this early point. I think it's a good idea for a later paper, but I wanted to get these results out there first, to soften up our audience.

2. The second problem is that, again, I'm trying to write a paper that is transparent to the primary intended audience--economists interested in growth. And the interesting 2SLS idea that you recommend would be less transparent than a straightforward linear regression. My limited experience in academia has taught me that "marketing is all," and econometric niceties are often left by the wayside in the desire to tell a good story. And the robust IQ/Growth relationship seems to be a good story.

I just have this feeling that if I had 5 or 10 pages of the paper justifying and explaining my choice of instruments (like I did in my disseration years ago), that would give the audience an excuse to just tune out the main story: "Oh, he's playing games with his choice of instruments." "The field of global IQ research is just getting started, and he already is trying to justify exogenous shocks to a nation's IQ? Sheesh!"

That said, you are exactly right that this is where the field will need to go over the next few years if it appears that national average IQ is a robust measure of one form of human capital.

As the cliche goes, "More research is needed." I'm hoping that our work encourages other people to join in on this research. Lord knows that Joel and I don't want to do it all ourselves.

I want people to come away from the paper saying, "Gee, I wouldn't have thought that IQ's relationship with growth was that robust, and that IQ was more closely related to growth than *any* basic education measure. I wonder why that is? Gosh, we economists should probably start pushing to measure national average IQ (or its politically-correct proxy, the math/science 'achievement test') much more consistently."

Clearly, we can't take answer every question in a 30-page journal article. But we can encourage other people to start answering those questions, one article at a time....

In summary, I guess I'd just have to plead guilty to reading too much Deirdre McCloskey in grad school--her criticisms of modern econometric practice and her focus on the "intended audience," especially in her book, "Rhetoric of Economics," have stuck with me over the years.

Again, many thanks for your helpful comments. I'm sure that they'll improve our IQ/Growth research agenda.

Comments for this entry have been closed
Return to top