Note however the caveat in Note 1, i.e. the restriction of range problem. The same issue is discussed in the paper that is the source for most of these data:
One caveat concerning the associations in these studies relates to the restriction of range in our sample. Certainly, it is true that individuals with average and above average cognitive ability are overrepresented in samples composed entirely of university students. [...] [O]ur participants attended a moderately selective state university. The SAT total means of our samples are roughly .60 of a standard deviation above the national mean of 1021 (College Board, 2006). The standard deviation of the distribution of scores in our sample is roughly .55–.70 of the standard deviation in the nationally representative sample.
These numbers mean that they're dealing with the upper half of the distribution in cognitive ability among SAT takers, the mean being somewhere around the boundary between the third and the highest quartile. I strongly suspect that the correlations would be much higher if they dealt with a sample from both sides of the general population bell curve.
That's really useful. It would be nice if someone would expand it into a post, with an explanation of all the terms which need explaining for a general audience, like "Proportion dominance effect" and so on.
I'm confused. If there are cognitive taks tasks that aren't correlated with cognitive ability, doesn't that just mean that the conventional definition of cognitive ability is wrong? (Probably insufficiently multidimensional)
Yeah, it's bad wording. "Cognitive ability" generally refers to g. Calling g "cognitive ability" is valid - if a bit confusing - since it more or less correlates with most of the things we would consider cognitive, though as this chart shows, not all. But if you want to be exact, you should say just g and not "cognitive ability".
If you read the paper, you'll understand what is meant by the terms used. For example, 'cognitive ability' refers to things like fluid intelligence but does not include the possession of probability theory.
Interesting, sunk cost bias does not vary with intelligence?
One study cited in the above is a small study of 21 'gifted' children and 82 poor children:
It too found no effect by age, nor any effect by gifted/poor group (pg 10). On the sunk-cost question, the sunk cost response was a (large) minority. I wonder whether the adults who answered that question original (in Arkes & Blumer) were more or less prone to sunk cost than the children?
Here. From this 2010 book chapter by Stanovich, Toplak, and West. (Here is the book.)
See also Baron's table of cognitive biases, the normative models they violate, and their explanations.