Thursday, May 26, 2011

Todd, Greg and Dave ...

... discuss the reduced form relationship between test scores on college entrance exams (i.e. the SAT or ACT) and performance at university.

Lesson 1: Dave needs better research assistants, who will read the literature before his column is printed in the NYT.

Lesson 2: As they so often are, one's priors are confirmed in the data.


asociologist said...

Dear Jeff,

Are there any pieces from this literature (on race, SAT/test scores, and grades) that you would particularly recommend?


B said...

I think this is a case where writing down a model explicitly would be helpful. (As would Mankiw actually *reading* Todd's entire comment; Greg tends to cherrypick things more than most economists, and that's saying something.)

Mankiw's "test" relies on family income/background affecting college grades only through SAT scores and not directly (through college tutors, perhaps; we're expensive, I know). Todd mentions this is probably false, and, indeed, I think this was your prior, Jeff.

A test of Leonhardt's claim could be rudimentarily performed by regressing college GPA (or admittance) on SAT score---interacted with family income---along with family income levels and more extensive controls for high school environment, tutoring, and other variables in, say, NELS. Now one can test for whether the coefficient estimate on SAT score varies across income levels.