[fsloth]

The applicable experiment for IQ would be to take two ”equal” populations (say,1000 ppl with middle-class background each), sort by IQ, cut the sample set at the middle to lower than median, and higher than median group.

The statistics so far show that the upper median group will do better on average. One might end up in jungle but it does not really matter for our experiment.

For individuals, IQ is sort of statistical proxy for lots of things if your daily life is lived in a first world country.

But it’s insane to hold it as some sort of key indicator of fundamental human potential.

In population statistical situations, like when hiring, however, imho it does make sense to prefer high iq individuals. Not because of what it tells of a single candidate’s potential, but it acts as a sort of maxwells demon for the workforce as total. So you end up with a employee pool closer to above-median group in our experiment which may or may not provide better business outcomes.

[tptacek]

This is basically false; you might make it true if you cut somewhere other than the middle, but famously IQ has non-uniform reliability (it's heteroskedastic? is that the way to say it?). Pop science and nerd culture have fixated on it as a global ranking of people by cognitive capability, but it's not that at all; past a threshold, the actual numbers (and test-test reliability) get really noisy.

This should make intuitive sense, because IQ was designed as a diagnostic, one in a battery of diagnostics, for people with cognitive dysfunction. It's a useful tool to deploy when you have a patient who, for instance, can't seem to progress in reading class or whatever. It's broadly misapplied in studies like this (but then, this study has deeper faults than that).