Hacker News Reader

Comment by cltby | Hacker News Reader

> IQ is somewhat effective as a diagnostic tool for significant impairment and not much else

This places exactly zero constraints on the utility of IQ tests. Instead of asking why group A scores higher than group B, I can simply ask why group B is impaired relative to group A. Apparently that should be enough to convince you of the validity of the comparison.

> The variance of measured IQ for the same individual over time is very high.

The test-retest correlations of IQ batteries are typically in the 0.8-0.9 range. For example, the Wechsler test [1]. This makes IQ probably the single most reliable measure in the social sciences. Perhaps you're referring to the Wilson effect, but that's not at issue here.

[1] https://images.pearsonclinical.com/images/pdf/wisciv/WISCIVT...

View on HN

Correlation being 0.8 means that R^2 is 0.64. A test that can only explain 64% of the variance in its result based on the fact that I am still me is suspect.

> This places exactly zero constraints on the utility of IQ tests. Instead of asking why group A scores higher than group B, I can simply ask why group B is impaired relative to group A. Apparently that should be enough to convince you of the validity of the comparison.

I said significant impairment. I don't mean significant in the statistical sense but in the sense of "wow that's really noticeable". Basically any test of aptitude can be used to detect lead poisoning, or the impacts of brain tumor removal, for example.

Please answer my request though, in which contexts do you think IQ suffers from denialism?

> A test that can only explain 64% of the variance in its result based on the fact that I am still me is suspect.

The WISC full scale test correlation is 0.97, so R^2 0.94. As a reference point, you might compare to the 100m sprint [1], where the correlation is something like 0.9. If you believe that a timed run is a reliable way to determine speed, you should strongly believe an IQ test is a reliable way to determine mental ability.

> I said significant impairment

This is basically a statement about statistical power: you claim IQ is subject to large measurement error, so given typical sample sizes, we're not able to detect anything but the largest effects. But no part of this is true.

1. Good IQ proxies are reliable, i.e. they give consistent answers for a given individual.

2. Good IQ proxies are usually measurement-invariant (certainly for US samples), i.e. they measure the same thing for different groups [2]

3. Sample sizes are massive (the entire pool of SAT takers, AFQT takers, NLSY participants, etc).

> in which contexts do you think IQ suffers from denialism?

Basically all of them! You yourself just denied its validity in all but a handful of narrow situations.

[1] https://www.sciencedirect.com/science/article/abs/pii/S09608... [2] https://www.tandfonline.com/doi/full/10.1080/13854046.2016.1...

Reply on news.ycombinator.com