4 ms·
>* Intelligence is a statistical construct defined based on a group of rather arbitrary tests. >That one is true. In what sense are they "rather arbitrary"?
by nokcha 9y ago
>* Intelligence is a statistical construct defined based on a group of rather arbitrary tests.
>That one is true.
In what sense are they "rather arbitrary"? It sounds to me that you might be suggesting that there is little correlation between performance on different IQ tests or different parts of the same IQ test. But the authors of the paper claim (with cited evidence) that "a common g factor accounting for about half of variance on cognitive tasks has been found across many human cultures (e.g., Carroll, 1993; Dolan, 2000; Dolan & Hamaker, 2001; Frisby & Beaujean, 2015; Gurven et al., 2017; Reuning, 1972)".
- mcguire 9y ago"Arbitrary" in the sense of not being chosen based on an understanding of what "intelligence" is. Please correct me if I'm wrong, but g is defined by noting that there is a correlation in performance on loosely defined cognitive tasks and assuming that there is a single important variable explaining that correlation.
- thraway180306 9y agoWhat you call an assumption actually sneaks in at a slightly different place. When you design multiple tests that can only correlate positively or not at all, you get a matrix with nonnegative entries. Of course when you factor it you find a dominant factor, there has to be a dominant factor (or some, if matrix has zeroes, unlikely in a noised case). Frobenius–Perron theorem guarantees that dominant factor for you. You can make up any set of traits, test for positive correlations this way and always get some dominant factor. Mathematics guarantees there will always be a g factor whether from real influence or as an artifact of factorization, if your tests can't correlate negatively.
- mcguire 9y agoHow does one choose between a single g factor and multiple g factors?
- thraway180306 9y agoThey do not appear in the real data, you have one g stronger than all others. This g has as much explanatory power as noise. Frobenius–Perron guarantees it would always appear whether there is one strong underlying influence on the data or hundreds unrelated ones. Factor analysis used to establish g doesn't and can't have the power to tell. There are numerous and intense attempts to paper over this mathematical fact by circumstantial evidence, I'm not up to date with all the contortions. I don't think they can make linear algebra a causal theory: when you want to replace your data with linear combination of factors there will happen to be a strongest factor. To somewhat answer the direct question: how to check factors. There is Confirmatory Factor Analysis that checks how factors decay. You take out factors you see as accounting for most of the variance in the data and check if indeed not much partial correlation is left. This has most power to negate an influence but next to none or very little power to confirm it not being spurious. These wikipedia pages don't look any good, but I don't want to evaluate that right now or chase better material https://en.wikipedia.org/wiki/Confirmatory_factor_analysis https://en.wikipedia.org/wiki/Confirmatory_factor_analysis https://en.wikipedia.org/wiki/Partial_correlation https://en.wikipedia.org/wiki/Partial_correlation