Fred — very good points, and I agree the article can be a bit shaky at times. The reason it convinced me, though, wasn’t Murray’s high-minded rhetoric but his “blizzard of data.” You’re right that he presents SAT I and SAT II as rough equals, and argues for the SAT II on various other grounds; I’m not much more sympathetic than you to the other grounds, but I think it’s important that they’re only rough equals. The SAT II is “slightly” better than the SAT I at predicting college performance, and in my opinion, that alone ought to tip the scales in its favor.

I’d have been just as convinced had Murray’s article consisted solely of these two paragraphs (taking at Murray’s word that the study was a good one, and had not been contradicted by others):

Achievement tests did slightly better than the SAT in predicting freshman grades. High school grade point average, SAT scores, and achievement test scores were entered into a statistical equation to predict the grade point that applicants achieved during their freshman year in college. The researchers found that achievement tests and high school grade point each had about the same independent role — that is, each factor was, by itself, an equally accurate predictor of how a student will do as a college freshman.
But the SAT’s independent role in predicting freshman grade point turned out to be so small that knowing the SAT score added next to nothing to an admissions officer’s ability to forecast how an applicant will do in college — the reason to give the test in the first place. In technical terms, adding the SAT to the other two elements added just one-tenth of a percentage point to the percentage of variance in freshman grades explained by high school grade point and the achievement tests.

I see college admissions as a very simple process. You have measures that predict how well someone will do in college (test scores, high-school GPA, grades on application essays), and you have measures of how someone actually does (in both college and life; freshman-year college GPA is not ideal, as you point out, but right now it’s one of the better ones we have available). The goal of the admissions process is to figure out which former factors, when combined, have the highest possible correlation with the latter.

You’re right that to do this, you’ll typically want to get as much data as possible. But in statistics, returns diminish quickly: High-school GPA plus SAT score is better than high-school GPA alone, but not as much better as some might expect. The reason for this is that high-school GPA and SAT score are themselves correlated — they’re measuring some of the same things, like academic ability. The question is where to draw the line: How much predictive power should a test add (not just have itself) in order to be worth making a whole bunch of people pay to take it?

I looked up the paper Murray was referring to, and here are the precise numbers on SAT I/SAT II/high-school GPA. These numbers indicate how well each measure predicted (or explained variance in) freshman-year college GPA. Alone, high-school GPA explained 15.4 percent, SAT I 13.3 percent, and SAT II 16.0 percent. Judging by this alone, it looks like it would be worth taking account of all three in the admissions process.

But when you put the numbers together, the picture changes significantly. The SAT I and SAT II together are no better than the SAT II alone is: 16.0 percent. If you want to find the best combination of two variables, your best bet is the SAT II and GPA: 22.2 percent. If cost were no object, you could go for all three — 22.3 percent — but is a tenth of a percentage point really worth putting everyone through another test?

Bottom line: You can point out why the SAT I and SAT II should measure different things, but the numbers say otherwise. The numbers further say that the SAT II better measures what these tests do measure. Thus, if it’s one or the other, the SAT II should win out.