Population with high IQ predicts real GDP better than population
Summary
I estimate there is a correlation of 85.4 % across countries in 2019 between the real GDP and population with IQ higher than the global mean. This is stronger than the correlation of 78.4 % between the real GDP and total population.
Among the 20 countries with the largest real GDP, the population with IQ higher than the global mean will increase the most for India (53.5 M), the United States (39.4 M), and Indonesia (15.7 M).
Acknowledgements
Thanks to alsilverback from Batonics AB.
Methods
I determined the population with an IQ above a given z-score of the global distribution assuming (see tab “Population by IQ” of this Sheet):
2019 population data by country from Our World in Data (see tab “Population”).
The mean IQ globally and of each country from Table 16 of Lynn 2019, and coefficient of variation of 15 %[1] (see tabs “Global IQ” and “IQ by country”).
To my knowledge, Lynn 2019 is the best source of IQ data by country, but it arguably does not differ much from other good sources. The correlation coefficient for the relationship between the mean IQ by country from Lynn 2019 and WordData is 88.7 %[2] (see M2 of tab “IQ by country”).
I selected data from The World Bank for the total and per capita real GDP in 2019 by country (see tab “Real GDP”).
Results and discussion
The table below contains the correlation coefficients for the relationships across countries in 2019 between the real GDP and the real GDP per capita, population, mean IQ, total IQ[3], population with an IQ higher than various z-scores of the global distribution, and total IQ of top half.
| Relationship between the real GDP in 2019 and… | Correlation coefficient (R) (%) | |
|---|---|---|
| Real GDP per capita in 2019 | 13.1 | |
| Population in 2019 | 78.4 | |
| Mean IQ | 24.3 | |
| Total IQ in 2019 | 82.7 | |
| Population in 2019 with an IQ higher than… | 3 standard deviations below the global mean | 78.5 |
| 2 standard deviations below the global mean | 80.0 | |
| 1 standard deviation below the global mean | 84.9 | |
| The global mean | 85.4 | |
| 1 standard deviation above the global mean | 80.0 | |
| 2 standard deviations above the global mean | 76.1 | |
| 3 standard deviations above the global mean | 73.3 | |
| Total IQ of top half | 84.0 | |
The population with IQ higher than the global mean is the strongest predictor of the real GDP (R = 85.4 %), better than the population alone (R = 78.4 %). Consequently, I think it may be interesting to analyse how the former will evolve. The following table presents, for the 20 countries with the largest real GDP[4], the absolute and relative variation from 2019 to 2050 of the population with IQ higher than the global mean, assuming the same parameters from Lynn 2019 for the IQ distributions.
| Country[5] | Variation from 2019 to 2050 in the population with IQ higher than the global mean | |
|---|---|---|
| Absolute (M) | Relative (%) | |
| China | -27.5 | -1.92 |
| United States | 39.4 | 12.0 |
| India | 53.5 | 3.92 |
| Japan | -18.9 | -14.9 |
| Germany | -2.85 | -3.41 |
| Russia | -7.65 | -5.25 |
| Indonesia | 15.7 | 5.82 |
| United Kingdom | 5.31 | 7.87 |
| Brazil | 7.47 | 3.54 |
| France | 1.89 | 2.90 |
| Mexico | 15.2 | 11.9 |
| Turkey | 7.18 | 8.61 |
| South Korea | -3.76 | -7.35 |
| Spain | -2.21 | -4.72 |
| Canada | 6.75 | 18.0 |
| Saudi Arabia | 2.05 | 5.99 |
| Thailand | -2.15 | -3.09 |
| Poland | -3.50 | -9.25 |
| Australia | 6.19 | 24.5 |
| Iran | 6.22 | 7.50 |
Amongst these 20 counties, the 3 whose population with IQ higher than the global mean is predicted to have the largest:
Absolute variation are India (53.5 M), the United States (39.4 M), and Indonesia (15.7 M).
Relative variation are Australia (24.5 %), Canada (18.0 %), and the United States (12.0 %).
- ^
I set the coefficient of variation to 15 % because, according to Wikipedia:
For modern IQ tests, the raw score is transformed to a normal distribution with mean 100 and standard deviation 15.
- ^
However, note the values from WordData rely on many studies from Lynn:
The intelligence quotients by countries are taken from the studies conducted by Richard Lynn and Tatu Vanhanen (2002), Heiner Rindermann (2007), Khaleefa and Lynn (2008), Ahmad, Khanum and Riaz (2008), Lynn, Abdalla and Al-Shahomee (2008), Lynn and Meisenberg (2010), as well as the PISA tests in 2003, 2006 and 2009. More recent results were weighted higher. The studies are not entirely uncontroversial as they often consider only specific population groups or a few individuals per country. If, on the other hand, an average is obtained from all the tests and studies, a usable overview will be obtained.
- ^
The total IQ is the product between the mean IQ and population.
- ^
See tab “Population by IQ” for the full results.
- ^
Ordered from largest to smallest real GDP in 2019.
You should definitely say more about what you understand from this exercise, because it’s not clear to me at all.
Hi Karthik,
As mentioned in the 1st bullet of the summary, I understand there is a strong correlation across countries in 2019 between real GDP and population with IQ above the global mean. Some more details are in this comment.
I… understand that. I’m asking what we should learn from this correlation.
Unfortunately, I have not thought about that. I just guessed some people might find it interesting, and revealing it did not seem harmful to me, given correlations between IQ and income (and other metrics) are usually believed to exist.
So my thoughts on this exercise is that interpreting correlation magnitudes is flawed unless they’re quite far apart (which these are not).
In a regression framework, if you regressed income on population and compared it to the coefficient when regressing income on population with high IQ, you would get confidence intervals on these coefficient estimates and I’m pretty confident that each coefficient would be within the other’s confident interval.
(Edit: to make that concrete, this magnitude is so small that you can easily imagine the sign reversing with different data on IQ such as WordData.)
Maybe a better way to investigate this would be to regress income per capita on the fraction of people who are high-IQ. Then if the coefficient on that term is significant and positive, you could infer something is going on.
I think this is not as simple, but I agree with the trust of your argument. The closer the magnitudes of the correlation coefficients (or anything for that matter), the more resilient the estimates have to be for one to conclude one is robustly higher/lower than the other.
Meanwhile, the article National Intelligence and Economic Growth: A Bayesian Update has been brought to my attention. It claims that:
I have only read the abstract, and so do not have any views on how trustworthy their conclusion is.
Can you say a little bit more about the point you are trying to make here? This could be an interesting, or on the other hand trivial, finding for a number of reasons, but I’m curious as to why you don’t speculate on these reasons at all. I’m not sure this correlation is particularly compelling without discussion of why it exists.
Given you favourably cite Lynn, perhaps it is the case you believe in his conclusions regarding race itself determining IQ, and think this explains (at least part) of the effect. Perhaps you don’t believe this. I think it behoves you, given the controversy regarding this work, to state your commitment (or lack thereof) to these ideas since your data source involves work conducted, arguably, to motivate this hypothesis.
Regardless, I think without further context it’s very hard to interpret this evidence as favouring any particular theory: whether it be regarding race, education, culture, nutrition or any number of other factors.
It’s also unclear to me why one would be interested in associating total IQ with GDP rather than average IQ with GDP per capita. Perhaps you could say something about that, if you’ve had any thoughts?
Hi Kieran,
Thanks for your questions.
The point I am trying to make is in the 1st bullet of the summary. Across countries in 2019, the population with an IQ above the global mean IQ was a good predictor of real GDP. I think there is a correlation between IQ and income within countries, so I was not surprised to find a correlation between population with high IQ and real GDP across countries.
Before reading this comment from Stephen, I did not know Lynn was a contentious figure. Lynn being the primary author of Lynn 2019 and other studies about IQ were what I knew about Lynn. I cited Lynn 2019 on the merits of its methodology, and tried to check the validity of the data:
Regarding:
I do not know which specific conclusions you are referring to, and I am not very familiar with the arguments and studies supporting the claims that race determines IQ. I am not sure what you mean by “these ideas”, but, to be clear, I believe race has a negligible effect on moral weight. My motivation for not including disclaimers is captured in the following paragraphs from this post from Emrik:
As for:
I agree. This is why I have not made any claims regarding implications of the correlation.
The analysis also focus on country trends (see 2nd bullet of the summary), and I suppose total metrics are more informative than per capita metrics to study the evolution of the global distribution of resources by country. Based on the data I have collected, there is a correlation of 60.6 % between real GDP per capita and mean IQ across countries in 2019.
I take your word for it that you’re naïve about Lynn’s work on race and IQ. I don’t fully buy into the idea that defensive writing is bad per se, but I won’t litigate that here. I don’t think the central errors I was criticising are relevant to this. Briefly:
If you’re going to present data, you should critically engage with the source of that data. Correlation with another source without critically engaging with that source either is meaningless. For instance, that website states: “Often surprisingly but scientifically proven, a warmer climate badly affects the intelligence quotient.” This is not an honest interpretation of the literature nor a coherent account of the scientific method.
Presenting data without any context around that data strikes me as a strange choice at best. I don’t really believe that you think there is no interesting conclusions that might be drawn: why else did you post it? Surely you think there is something interesting to be said about it? Data itself is inherently meaningless, but from its interpretation we can make interesting observations. I therefore think you should at least present relevant context around data, and state why you think it is interesting, so discussion can proceed. The alternative, especially around such a controversial topic, is the result here: confused commenters trying to tease out why you would post this data but not seem inclined to share what conclusions you are drawing from it.
To be totally clear, I believe you that you don’t find this interesting for issues relating to race, I just don’t think given that debate informs an understanding of both the data and its context, that it should be ignored. Unfortunately, information doesn’t exist in a vacuum.
Thanks for clarifying.
I agree further engaging with the quality of the data would improve the post. However, this does not necessarily imply that I should have engaged more with the data. I had done this analysis for other purposes, and thought that posting this may be better than nothing. I should have tagged the post as “Personal blog” instead of the default “Front page” (I was unaware of the tab “Personal blog”, but will have it in mind for similar posts in the future).
I would say data is inherently meaningful. For example, I think the data explorers and graphs from OWID convey lots of useful information. However, interpretation is also valuable to e.g. explain data in plain language, clarify what we can and cannot infer from them, pose further work questions, etc. (so reading articles from OWID besides just checking graphs leads to a better understanding).
I don’t know why you reported results for the correlation coefficient as a percentage. The range for R is [-1, 1] and it cant’ be interpreted as a percentage.
It feels disingenuous to use Lynn’s data or write about this topic without providing more context. Both are extremely controversial to say the least.
I strong downvoted this post and don’t think it’s a good fit for this forum.
Hi Stephen,
Thanks for your comment. I upvoted it because I think it contributes for an important discussion, and think positive/negative feedback is how things improve.
From Wikipedia, the correlation coefficient is:
This means the correlation coefficient between X and Y can be interpreted as the covariance between X and Y as a fraction (or percentage) of the product between the standard deviations of X and Y. More generally, the symbol “%” in non-dimensional number can be interpreted as meaning “per 100″ or “10^-2” (in the same way that “k” means “10^3”).
To be honest, I did not know Lynn was a contentious figure. However, I think the data from Lynn 2019 should be mostly assessed on the merits of its methodology.
Apparently naively, I thought it was fine to do an analysis involving IQ without making any claims about its implications, which are outside of scope, and would require much further thought (namely for the reasons you are alluding to). My motivation for not including disclaimers is captured in the following paragraphs from this post from Emrik: