The R^2 vs r debate in IQ research
lol
In 2018, Taleb dropped the biggest piece of coal his thoughts on IQ, basically saying that the construct was worthless, the research was flawed, and IQ didn’t predict anything. The article, in general, is awful and full of conceptual and factual errors1.
There is, however, one point he made that did hold water. That, when you correlate IQ scores with other outcomes of interest that aren’t test scores, the relationships observed aren’t strong — 25% of the variance in productivity and lifelong income; less of others like divorce rates or political beliefs.
Taleb’s critics pivoted to arguing that he was using the wrong statistic (R^2), and that if he really wanted to know how much IQ matters, he should have used the correlation coefficient (r). Converting from r to R^2 is just a matter of multiplying r by itself, squaring it, so to say.
I think this is an awful argument. Let me put this mathematically: if IQ explains 25% of the variance in productivity, then 75% of the variance is explained by other factors. 75% is a lot bigger than 25%, so it would be fair to conclude that there is a lot more that goes into productivity than just IQ. It’s also literally just the same statistic, just on a different scale.
Sometimes, neither of these statistics are the correct ones to use. Let’s say you want to test how much malaria harms your health. If you were to correlate people’s malaria infection status to their health in the South Africa, the correlation would be close to zero because there’s very few people who are infected with it at a given point. But malaria absolutely does mess you up.
Where to go from here?
When people realise that IQ doesn’t explain that much of the observable universe statistically, this could be due to two reasons.
One of them is that intelligence cannot be measured accurately by IQ tests, and that intelligence is best defined as something else. A popular definition I’ve frequently seen is that “intelligence is the ability to get what you want out of life”. The spirit of the phrase is that your intelligence doesn't matter if it doesn't lead to any outcomes, which I agree with, but it is still false by definition. The realisation helps prevent people from coping with failure by thinking "oh, I am intelligent haha", but that doesn't personally resonate with me.
I think that this definition, covertly, sneaks IQ determinism back into the equation, just with the aside that it can’t be measured by a test, but with what you do and get. And those things definitely matter, I just don’t think they are that related your thought processess.
The definition of intelligence I've arrived to is the ability to infer information from information. The inference can be difficult because it is far-reaching, unintuitive, requires background knowledge, or mentally tracking lots of involved parts. IQ tests can measure how intelligent people are, because they measure the amount of knowledge they have already accumulated (vocabulary, general knowledge tests), proxies for cognitive ability (e.g. reaction time, processing speed), and making conclusions that are far-reaching and involve tracking lots of parts.
Funny that Taleb’s critique involves arguing that there is no right answer to the question ‘ 1 → 2 → 3 → X ‘ , because the pattern could be something besides 1+1+1+1 that looks like 1+1+1+1 for the first three numbers. In the real world, a lot of conclusions people have to make look like this: no 100% logically correct right answer, but something that “feels right” and works.
I’ve entertained the perspective that IQ tests do not measure intelligence before, and came to the conclusion that they do so pretty well. It’s not really possible to determine how well exactly because intelligence is an abstract, unmeasurable concept, but I think the true correlation between intelligence and IQ is something like .8.
The other reason that intelligence could not be a good statistical predictor of anything besides test scores is that it just doesn’t matter that much. It doesn’t factor in, well, everything else about a person: their personality, physical capabilities, appearance, physical health, or circumstances.
Notes:
I haven’t posted in a while, partly due to having the flu from Jan 2 - 15; and also because I have been cooking two really long pieces: an attempt to explain all personality psychology, and an attempt to build a framework that explains all differences in income between people.
Said flu also led me to reflect on the time I got typhoid fever when I was 19. I did it on my new substack publication Selective Contrarianism, where I just post whatever I think in the moment without regard to what is acceptable, what most people would find interesting, or whether it makes sense.
There’s no point in addressing them individually, other people have already done it, and the main reason this argument wrt IQ exists has more to do with emotions and values than facts and logic.
I suspect that high IQ / high g is particularily important in understanding the natural world - which is rather more consistent than people. The STEM fields seem to have field specific minimum intelligence levels to master a given field - but 'success' in general tends to be more defined by mastering interactions with people, particularily in social groups. And people are irrational and inconsistent enough that facility with natural patterns is of limited utility (useful, but not dominant).
A challenge I proposed for Taleb was for him to come up with a better measure than IQ which has 1) at least as much predictive validity for various important life outcomes as IQ, and 2) fulfills all of his absurd statistical standards that no one else has.