I’m probably unusual for Phi Beta Cons in that I support scrapping the SAT — not just scrapping it, like Wake Forest did, but replacing it with some SAT II tests. Charles Murray brought me around to this way of thinking through an article in The American.

I read Salins’s piece, and with all the respect due to a former provost, I’m highly doubtful of many of the things he says.

My main objection is to this:

Among the countless studies done on this subject over the years, not a single one has failed to find a high correlation between SAT scores and academic performance in college, as measured by grades or persistence. On a personal note, during my ten years as Provost of SUNY, I had my institutional research staff repeatedly review the relationship between SAT scores and academic success among our 33 baccalaureate campuses and their 200,000 + students, and found – as all the national research has confirmed – a near perfect correlation.

It’s debatable what a “high” correlation is in the social sciences. For the purposes of testing, I would throw out the words “high” and “low,” and replace them with “better/worse than the alternatives.” As Murray’s article demonstrates, the numbers prove that “better” is simply not true of the SAT, relative to the SAT II.

It’s not debatable what a “near perfect correlation” would be — that would be a relationship where if you have one variable (in this case, SAT score), you can almost perfectly predict another (first-year GPA) by running it through a formula. This is measured by what statisticians call r-squared, which ranges from 0 to 1 — a 0 means there’s no relationship, and a 1 means that if you have one variable, you can perfectly predict (or predict 100 percent of the variation in) the other. You virtually never see anything close to a 1 in social science. [UPDATE: On a technical point, a reader points out that to have an r-squared of 1, the variables must not only predict each other, but must do so in a linear fashion.]

I have no idea what his institutional research found, but “all the national research” has most definitely not concluded there’s a “near perfect correlation.” An ABC News story puts the range at 10 to 20 percent of the variation, and that sounds in line with numbers I’ve heard before. That’s a decent correlation in social science, but whether it’s “high” depends on the alternatives, and there’s no way you could describe it as “near perfect.”

Salins also takes issue with those who claim high-school grades predict college success better than the SAT does. However, he simply says why this shouldn’t be the case, and doesn’t use his “institutional research,” or any other research for that matter, to answer the question empirically. At least one academic study disagrees with him.