So I'm commenting on this post: http://cscs.umich.edu/~crshalizi/weblog/523.html. I'm hoping to understand factor analysis stuff better. Don't know where to find the time. But if I do find the time, I'm assuming the understanding gained will somehow be useful.

"The way we can unambiguously tell that it had falsifiable empirical content is that it was, in fact, falsified. Looking at larger and more diverse data sets, it became clear that the partial correlations among scores on mental ability tests were not zero, or even close enough to attribute the difference to chance."

Don't really understand this. Maybe we did discover that Spearman's data was falsified. But assuming we did, I first want to know... just how do you "control for the common factor" and produce partial correlations?

OK, now, assuming we established that Spearman's original factor analysis shit was wrong... So I have a question: Does it matter whether it is really a combination of separate abilities or one single ability that we're measuring, if in the end, we know that the score we get out correlates well with grades, chance of landing in prison, ...? Cites example http://www.insidehighered.com/news/2007/06/19/admit.

"The same study suggests that the SAT adds little predictive value to admissions decisions and is hindered by a high link between SAT scores and socioeconomic status — a link not present for high school grades."

Wha- wha - what? I would think there's a definite link between socioeconomic status and high school GPA... think about that rich kid getting tutors to do his homework... to explain every concept 4 times over... to practically hand him the test forms and ask him to memorize the answers... does that leave any room for actual on-the-spot analysis, or are we just parroting answers now?... Of course, rich parents never bribe teachers (I'm sure shit like that never happens, because I've never heard it, and I have such a sunny view of human nature)... well, nevermind. Point is, SAT scores do add predictive value to admissions decisions, despite these decisions being near impossible to make. That article proves that you can admit a smart kid only to have him decide that he'd rather stay in a constant state of drunkenness and pee on the chancellor than make any real academic progress... you can't control for that with any statistical tool.

Imagine there are 100 unique abilities (i.e. completely statistically independent in some way I don't care to elaborate on). If we use abilities 10-40 predominantly in English class and abilities
20-50 in math class (all weighted differently) and I designed a test that tested for abilities 0-60, but that nevertheless predicted grades in English and in Maths classes very well... does it matter that I didn't test exactly abilities 10-40 and 20-50 with two different tests (as a sidenote: the advantage of 1 test over 2 is obvious when you look at price, convenience, etc.)? Does it matter that it doesn't seem fair and that some students whom you might like a lot might be denied admission if you used these tests? Yeah, a little bit. But a college that has 100,000 applicants doesn't have time for that. And there are contingency plans (which got Bush admitted to Yale).

In fact, I advise performing statistical tests in an attempt to correlate just about every observable characteristic of a person with their chance of succeeding in college. Then combine the results of these tests and only then decide whether you want to admit students (I hear that's why Ivy leagues like kids that play sports).

Argh, what's the point. I think one comprehensive high school exit exam (which tests everything you supposedly learned in high school), with a mandatory year off to have the results be graded and analyzed would be best. Complaints of lack of freedom are just annoying and seem to stem from laziness, above all else. This is what countries other than USA are doing, and they completely dominate the US in secondary education. The US is known for post-secondary education and its grad-schools (partly because it imports its talent from abroad).