Garett Jones on The Hive Mind

A recent line of research demonstrates that cognitive skills—IQ scores, math skills, and the like—have only a modest influence on individual wages, but are strongly correlated with national outcomes. Is this largely due to human capital spillovers? This paper argues that the answer is yes. It presents four different channels through which intelligence may matter more for nations than for individuals: 1. Intelligence is associated with patience and hence higher savings rates; 2. Intelligence causes cooperation; 3. Higher group intelligence opens the door to using fragile, high-value production technologies, and 4. Intelligence is associated with supporting market-oriented policies. Abundant evidence from across the ADB region demonstrating that environmental improvements can raise cognitive skills is reviewed.

The paper is here and the slides are here.


So low-IQ immigration will lower human capital and lower national outcomes? You don't say.

All snarking aside, this is an impressively politically incorrect paper, bringing together a set of highly relevant data.
But studies like this, based on high IQ ~ high productivity ~ high GDP hypotheses, cannot explain why communist China, with supposedly high IQ and low time preference, could murder 40 million of its inhabitants just a few decades ago. What, if anything, changed in mainland China? Is some variable missing from these hypotheses?

Just one bad idea can bring down an entire nation.

They were/are totalitarians.

But this particular bad idea is probably not enough.

There is this disease called Bolshevism. You may have heard of it. For some reason it impacts on those who should otherwise know better. It also explains the difference between North and South Korea by the way.

The paper doesn't seem to consider that the causation might work the other way round, with stronger economies producing higher IQ test results. Maybe in prosperous nations, there are good education systems and stronger incentives for an individual to work hard on improving his test-taking ability.

Actually, it does consider that possibility. See figure 2.

This is eerily similar to the insight at an old 2002 post at La Griffe du Lion ( ) and its 2004 follow-up ( ).

If you can get past the racial focus, the main points of those articles is that IQ and GDP are closely related, but not in a direct linear correlation. Most studies using IQ simply take the linear regression, completely missing the point that the IQ represents a distribution. Most of the articles on that site use the Gaussian distribution assumption, and show how applying that instead of linear regression it leads to much closer agreement between IQ and whatever is being measured.

LGdL's thesis is the links above was that society's wealth is dependent upon the percentage of people with an verbal IQ large enough to complete secretarial work. Which makes sense, if you believe that logistics and efficiency are crucial to a developed country. Bringing it back to the post links, it means that the correlation could very well be weak per individual, because what matters it how well the institutional bureaucracy works. It's an impressive feat for something I've only seen on a pseudonymous website.

"Most studies using IQ simply take the linear regression, completely missing the point that the IQ represents a distribution. Most of the articles on that site use the Gaussian distribution assumption, and show how applying that instead of linear regression it leads to much closer agreement between IQ and whatever is being measured. "

Could you elaborate on this? What you've just said doesn't seem to make a lot of sense.

The website assumes a Gaussian distribution of IQ and then uses the mean IQs to predict the fraction (F) of people with an IQ > X (some parameter). So the independent variable is the fraction of people with an IQ greater than whatever (assuming a Gaussian distribution) and the dependent variable is GDP.

The thesis here is that rather than IQ has a non-linear relationship with GDP. It doesn't have much to do with Garrett Jones except that more IQ = good for society.

Uhh, what?

Linear regression is the maximum likelihood estimator for Gaussian residual distributions...

Doug, you are being pathetically dense here. If you know what linear regression is, then you know that there is a large difference between a linear relationship and a linear regression. Also, you clearly did not read the link. The TL:DR is that even if the black-white difference in pass rates for two tests changes in the linear sense (ie. from a 10% gap to a 5% gap), it does not mean that the IQ gap has been closed in any meaningful sense.

You're not making sense. Your thesis is that a linear regression is not the same as a linear relationship? That's exactly what a linear regression, a model of a linear relationship with a Gaussian noise component.

The link posted used proportion over some IQ. Given that the cumulative normal distribution closely approximates a logstic function this is like applying a logistical transformation. The distribution of residuals will be Gaussian if you use Gaussian predictors (IQ) against a Gaussian dependent variable (log income). If you transform the predictor away from Gaussian distribution to a logistic distribution then you're no longer dealing with Gaussian residuals.

Hence your likelihood function is different and OLS no longer produces a maximum-likelihood fit. You need to use GLS with a modified likelihood function. Or some approximation method like iterative least squares, because you residual likelihood function's so screwed up from adding a Gaussian and logistic distribution.

Okay, maybe you don't really understand linear regression. Let's posit a relationship between two variables, x and y. If we create a model with a linear relationship between y and x we have y = b*x, where b describes that relationship. This is one form of linear regression. Let us instead say that y is instead a function of the log of x. We would then have y = c*log(x). This is still a form of linear regression, even though there is a log relationship and not a linear relationship between the two variables.

Doug, I haven't looked at the paper, but....

They could have some places that have different variances in IQ. So when they want to get "number of people above IQ X" that will matter.

And if they are looking at "number of people with IQ X or above" that would not be a linear relationship anyway. Imagine one place where the mean IQ is X and the variance is wide, so there are lots of geniuses and lots of idiots, and half the population fits the criterion. Then imagine one where everybody is IQ X and everybody fits the criterion. If you need a whole lot of secretaries, it does no good to have geniuses who may not be suited for secretarial work, when half the population is too stupid to do it.

Similarly, imagine a population where everybody has IQ X+100. If what you need is secretaries, a population of geniuses may not be any better than a population where everybody is IQ X+1.

So the effect is nonlinear compared to average IQ. That's plausible, assuming the idea even makes sense in the first place which I tend to doubt.

IQ and mathematical skill are good for creating value, but they're not much correlated with the verbal and political skills necessary to capture that value for yourself. It's not surprising, then, that individuals are not much rewarded for high IQ but nations are.

The canonical example of high IQ, low political skill people are Asians, who have the capacity for hard work but are generally not aggressive enough to prevent others taking credit for it.

This was my first thought: social Intelligence correlates more with income than, well, intelligence intelligence. I have met many rich people who are very dumb and very charming. The smartest people I know make a good living, work on interesting problems, and leave it at that.

I have met many rich people who are very dumb and very charming.

They want you to think that.

If they win a game of musical chairs it's better for them if everybody thinks "But they're so dumb! It must be accident or coincidence."

This is especially true in Asia, where there are no leaders or credit-hogs, only a massive collective hive mind.

Not sure if that is sarcastic or not. In Asia there are of course leaders and credit-hogs; humans vary within as well as between groups. But in the aggregate the percent of attention directed towards these zero-sum activities is lower. Good for Asian institutions, bad for Asians trying to compete in countries with more aggressive/less productive norms.

So, higher IQ citizens are individually prone to "support market-oriented policies", ostensibly in favor of individual incentives, while their individual intelligence is really of benefit only to the collective? A lovely paradox . . .

This sounds like a liberal's wet dream.

Lower your shields and surrender your ships. Your intellectual and personal distinctiveness will be added to our own. Your talents will adapt to serve us. The money you earn from your intelligence and skills will be absorbed into the collective. Resistance is futile.

Many things are winner take all making even a small difference a great advantage.

The obvious answer is that those with high IQs don't get compensated for their labor - those who own the tools they use to apply that intelligence enjoy the compensation for that value. Of course, on a national level, the benefits tend to be rewarded regionally, though I think this is not always the case either.

How come western nations endowed with lots of high IQ individuals messed up their economies to such an extent that we may be witnessing the "end of western predominance" ( the caption of the last photograph in Niall Ferguson's book Civilization: The West and the Rest")?

Thank you Mr. Strawman. No one is arguing that IQ is a perfect predictor of all economic success for individuals or even for nations. Clearly, the organization of society also matters. The canonical example being North Korea: destitute despite a population of high-IQ East Asians.

My totally unbiased conclusion is that engineers are underpaid.

Correlation is backwards. Good childhood nutrition and primary education = higher IQ.

Decades of early intervention programs that have failed to close the black-white achievement gap in the US suggest otherwise. It would be great if we knew some reliable ways to increase IQ, but we don't.

I think it's more accurate to say we are already doing the ones we know how to do, at least in reasonably well-off countries. Getting rid of lead paint, using sanitation and vaccines and pest control to keep diseases from retarding childrens' development, making sure everyone gets enough nutrition, making sure everyone gets some kind of free education, all those things do raise IQ scores, and probably raise actual ability to think as well. But we're already doing that stuff.

Yes, but there are counter-incentives to follow an alternative path to success:

1. drop out of high school
2. have a couple kids out of wedlock
3. use those kids and low income to qualify for Section 8
4. Use that Section 8 to secure a nice house in the suburbs
5. Stick it to your white neighbors who are paying triple for rent by screaming racism if they call the cops on you

Whenever someone use IQ in a scientific Paper, I run to reread again and again Cosma Shalizi's debunking of almost all studies that use IQ...

Why do people who have probably never read a single article in Intelligence or paper on EFA/CFA insist on linking to Shalizi's "debunking"? I suppose it is a lot better than citing The Mismeasure of Man

Clearly IQ is not a very good measure of the things we want to measure. GDP likewise.

But it's the measure we use. Therefore we must do the best we can with it. No one has come up with a better measure that gets used a lot. When people do use better measures for particular purposes, they tend to call them IQ anyway, blurring the issue beyond toleration.

One thing this brings to my mind is that many high paid individuals had to work very hard at un-fun jobs and had to take large risks. High IQ people can understand this more than others. Owning a business can be a real pain and you can loose it all easily but it can also yield huge income for the successful.

"The canonical example of high IQ, low political skill people are Asians, who have the capacity for hard work but are generally not aggressive enough to prevent others taking credit for it."

Don't forget NASA. Friend of mine who does location sound described a job for those fine folks. The film crew was paid to show a bunch of VIPs socializing, sipping cocktails and whatnot. Seems if the crew had turned the gear around 180 degrees, they could have been getting footage of folks hurling objects into orbit around the planet, but those fools that launch satellites don't have top political skills. Dumb NASA nerds. Big step forward for the nation when we stop funding those losers.


High average IQ is necessary, but not sufficient for good outcomes.

BTW I think this helps demonstrate that the proper way to lift Third Worlders out of dire poverty is to trade with them, not import them into your own country.

Manoel Galdino writes "Whenever someone use IQ in a scientific Paper, I run to reread again and again Cosma Shalizi’s debunking of almost all studies that use IQ…"

I remember reading those articles earlier and being shocked by how misleading they were - Cosma's disingenuous verbosity reminds me of Stephen Jay Gould. On the other hand, if one is not familiar with statistical techniques employed, Cosma's writing appears superficially legit and "science-y"...

Better title: "If I'm so smart, why ain't I rich?"

One of the major micro foundations of this is probably that lower IQ people are only productive when supervised by higher IQ people and that the higher the proportion of lower IQ people the proportionally fewer high IQ people you have to supervise them.

The quote on the post is a clear example of the main statistical point Herrnstein and Murray were trying to make in the Bell Curve:

A relationship that has only weak predictive value for an individual can have extremely high predictive value for a group mean.

In other words, if beta is big and significant, then even tho R2 is small, you can predict the group effect with confidence, and it matters.

Or, in other words, if all Indianans are smarter than all Ohioans, we still can't say with any confidence that one Indianan is going to be richer than a particular Ohioan, but we can confidently say that the average Indiana income will be higher than the average Ohio income.

No spillovers needed, just a little stats.

Actually I'm talking about something different: I'm interested in the (unstandardized) beta, not the correlation coefficient. In Murray's AER piece, he aggregates all high-IQ Americans in the NLSY and compares them to all low-IQ Americans: he gets a small implied coefficient on IQ. So within the US, IQ has a low percentage payoff.

My claim is that it's better to be the low-IQ fish in the high-IQ pond than the reverse, a testable (and now somewhat tested) hypothesis.

The burning question: Where is the asshole who captures the monetary spillover of all of Tyler's g ?

Tyler's an economist - he's a very smart guy but no matter how smart of economist you are you aren't going to be creating much in the way of monetary spillover.

All these papers on IQ and wealth fail to adress the problem of backward correlation. I'm not convinced by figure 2. It means that 50 years ago, these countries were poor with a high IQ. The relationship between IQ and wealth does not hold thorough the period. I wonder if this relationship will hold in 50 years.

Moreover, the dataset constructed by Lynn of mean national IQ suffers from biases. See the paper of Wicherts et al 2009 that criticizes the methodology used by Lynn for mean IQ of African countries:
"We found that Lynn and Meisenberg's assessment of the samples' representativeness is not associated with any of the objective sampling characteristics, but rather with the average IQ in the sample. This suggests that Lynn and Meisenberg excluded samples of Africans who average IQs above 75 because they deemed these samples unrepresentative on the basis of the samples' relatively high IQs. We conclude that Lynn and Meisenberg's unsystematic methods are questionable and their results untrustworthy."

It's hard to believe any results based on this dataset, especially when the explanation may come from backward correlation.


As Wicherts et al. note, the possible (I note possible) biases in their African data don't change the overall story. In their words: "There can be little doubt that Africans average lower IQs than do westerners."

If we combine that with the now routine finding that East Asian tend to outscore westerners, we have now already encompassed the broad outlines of Lynn and Vanhanen's database.

If we all could start discussions of cross-country IQ databases with the candor of Wicherts et al., then we could start to make some progress. They are to be applauded for taking such a big risk.

There is likely some reverse causation--I hope there's quite a lot of it--but if IQ helps cause prosperity through the micro-level channels I set forth--channels whose importance I document through micro-level studies, not just from LV's data--then it's important to increase the brain health of the world's low average IQ countries.

Small public-health-driven increases in IQ may create higher savings rates, better polities, and access to more advanced technologies. Quite a payoff. And all these channels are backed up by ordinary, non-LV economics.

From the perspective of a small-business co-owner (we produce mathematical and cryptographic software), higher mean IQ means bigger pool of potentially good collaborators and employees.

In software production, 3 or 4 good programmers working together have productivity of about 10 times higher than a single good programmer. It doesn't scale much higher than that, but "core teams" of 3-4 people can produce outstanding work.

(But no one is a good programmer without relatively high cognitive ability.)

Having a larger pool of possibly good programmers means that you have higher chance of creating a really good team. I suspect that the same applies in other engineering and technology.

But, especially at large companies, it's the account managers, sales staff, executive managers (general soft-skill, personality based work) who capture a bigger share of the pie created by the engineers.

I think this is pretty obvious - if you look at the type of people in companies (particularly in North America) who do very well, it tends to be soft-skilled people such as top sales people, account managers, advertising executives, etc. more so then engineers, R&D people or scientists even those the latter add much more value overall. Being able to succeed at office politics gives you a bigger advantage in capturing a greater share of income then say a PhD in physics.

Comments for this post are closed