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.
So my thoughts on this exercise is that interpreting correlation magnitudes is flawed unless they’re quite far apart (which these are not).
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.
We find national IQ to be the “best predictor” of economic growth, with a higher average coefficient and average posterior inclusion probability than all other tested variables (over 67) in every test run. Our best estimates find a one point increase in IQ is associated with a 7.8% increase in GDP per capita, above Jones and Schneider’s estimate of 6.1%.
I have only read the abstract, and so do not have any views on how trustworthy their conclusion is.
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.