Read full text: https://hereticalinsights.substack.com/p/iq-and-talebs-wild-ride
Note: This is the revised/updated version of the original post.
It’s been five years since Nassim Nicholas Taleb had published his infamous attack on IQ, and his article has generated quite a few responses. For the most part, Taleb’s arguments have been thoroughly taken apart, but there always remains a few people who bring up the same talking points he did over and over again. I ended up getting roped into it around a month ago when I was at a meeting, but rather than do the responsible thing of actually paying attention at the meeting, I instead found myself in an argument about IQ and being confronted with the same arguments that were originally made by Taleb. Because of this, I figured it’d be useful to synthesize a lot of the rebuttals that were made into this single article, as well as add additional evidence supporting the validity of IQ.
CONTENTS
Taleb Does Not Understand “Variance Explained”
Taleb starts off his attack on IQ on the grounds that it supposedly has a ‘low variance explained’ with various performances (which ones specifically he’s referring to, he doesn’t specify):.......
.....Taleb doesn’t provide any sources for where the 2-13% claim comes from, which is odd for someone who is so seemingly confident about his stance on IQ. Besides that, he makes a very basic mistake of misinterpreting anything that has a ‘low variance explained’ as being meaningless. The problem is that r2 is a flawed interpretation of effect size because it doesn’t tell us anything about the real world effect that one variable is expected to have on another variable. As an example, let’s assume that IQ explains ‘only’ 9% of the variance in income, so does it matter whether or not a one-point increase in IQ predicts a $10 increase in income or a $10,000 increase in income? Of course it does, and this is obvious to anyone who’s honest, but here lies the problem: r2 does not tell us how large the real-world effect is, because it expresses the effect size in a statistical sense, which is neat and all, but not in a real-world sense, so it’s not very meaningful. An r2 of 0.09 for IQ and income in the real-world means an r of 0.3, or that a 1 SD shift in IQ predicts a 0.3 SD shift in income, which is certainly not trivial......
......What this means is that in the real world, using our example, IQ would be closer to explaining 30% of the variance in income, not 9%. Another problem is that since the relationship between r and r2 is nonlinear, it gives a misleading picture when comparing the predictive validity of one variable (X) versus another variable (Y) for the same outcome (Z) when using r2 (ex: If X explains 64% of the variance in Z and Y explains 16% of the variance in Z, then X is twice as predictive as Y, not four times as predictive, because the actual correlations would be 0.8 and 0.4, respectively). Taleb seems to think that saying IQ has a low variance explained with certain performances allows him to disparage its utility, reality seems to disagree. According to reality and not Taleb, a small ‘variance explained’ can have immense consequences......
....So, in the real world, even things that only explain a small percentage of the variance can have large effects. Thus, a low variance explained is not valid grounds for dismissing the utility of IQ. But the truth is, we don’t even need to think that hard to find instances where a small variance explained has a large effect. Ryan Faulk gave more intuitive examples in one of his response videos which I will list here:
Mayonnaise only explains 5% of the variation in the tastiness of sandwiches. Therefore, hugely increasing mayonnaise levels will have no effect.
Salt explains little variation in steak tastes. Therefore, triple the salt!
In America, the proportion of people starving to death doesn’t explain much variation in health. Therefore, people don’t have to eat!
Stabbings explain little variation in physical health. Therefore, increased stabbings will be fine.
Oil production only explains around 5% of the variance in GDP per capita. Therefore, having 100 times more oil per capita won’t matter.
Ethnic diversity only explains 5% of the variance in GDP. Therefore, replacing Norwegians with Ethiopians won’t have any effect!
Sounds stupid, right? That’s because it is.
The other problem when it comes to the ‘variance explained’ argument is we need to put into context what exactly we’re comparing. On its own, IQ’s predictive validity might not seem too impressive, but what about when we compare IQ to other variables?......READ IT ALL FULL TEXT - https://hereticalinsights.substack.com/p/iq-and-talebs-wild-ride
