Review by Scott Alexander. Review by Stuart Ritchie. Interview with Jones.
Alexander makes the point that I tried to make, which is that aggregation can produce higher correlation in noisy data. Ritchie says because the correlation between self-control and IQ is only 0.4, it is a bit of a swindle to say that self-control is a big factor explaining why nations with high average IQ do well economically. Among other interesting things in the interview, Jones says
Years of education is terrible measure of human capital. Look at broad-based test scores to get a sense of where a country’s economic future is heading.
Years of education isn’t even a good proxy for education.
Arnold:
That’s part of the reason why my core argument is about effect sizes, not about correlations. I used correlations in the book to some degree because it’s the one statistical tool a good number of readers will be exposed to. But my core story, most formally explicit in my paper with Schneider, “IQ in the Production Function: Evidence from Immigrant Earnings” is that 1 IQ point appears to cause about 1% higher wages for individuals, but 6% higher GDP for nations.
I’ve learned in recent months that quantitatively-informed people outside of economics just understand effect sizes little if at all. You’ll see that the economists who review my book tend to see the importance of effect sizes, but others are routinely confused on the issue.
Magnitude matters, effect size matters. I hope that my readers and my critics will pay attention to effect size, and pay particular attention to equations 1 and 2 of “IQ in the Production Function”. Gamma is perhaps 1 for individuals, perhaps 6 for nations. This is the Paradox of IQ.
Aggregation has nearly nothing to do with that finding under mainstream econometric assumptions. Magnitudes matter, coefficient sizes matter.