May 10, 2004

The meritocratic margin of error

Posted by Arcane Gazebo at May 10, 2004 4:50 PM

The Caltech discussion board on orkut has a thread running on affirmative action. I'm crossposting here the argument I made in favor of "affirmative action within the margin of error". Feel free to punch holes in it so that I can refine the argument (or discard it if it's just flat-out wrong). (I think in a strict sense I am misusing the term "meritocratic" which is supposed to refer to government.)

Some [participants in the orkut discussion] seem to be arguing for a totally meritocratic admissions process. One could imagine a process in which the admissions committee distills each application into some single metric of qualification, something like a "predicted GPA". Then they could just take the top N applicants.

This hypothetical process has its appeal, but the obvious practical problem is the uncertainty associated with any particular metric that might be chosen. Since the college application provides very incomplete information about the applicant, this uncertainty will be rather large.

The University of California found that a linear combination of high school GPA and SAT II scores was the best predictor of freshman GPA, but it's still not very good, explaining only 20% of the variance. The College Board's own data (pdf) shows a .69 correlation for their metric (a combination of SAT I and HS GPA) in the category relevant to Tech. The point is that there is information about a student's qualifications which is not in the application, and this leads to substantial uncertainty in any ranking by admissions.

Which raises the issue of tie-breakers. A "tie", after all, isn't a pair of identical applications, but two students who both appear to be qualified within the margin of error. The margin of error being rather large, a certain number of ties are certain to occur in any admissions cycle, and admissions has to choose some non-meritocratic way of resolving them. I don't have a problem with their choosing affirmative action as this method.

To add a bit (based on feedback I got in the original discussion), actual colleges obviously do use a lot of information in the application apart from the SAT/GPA numbers. My sense is that the additional data doesn't add much to the determination of which applicants are qualified to attend the school, but are useful to admissions in other ways. I didn't address these things because the concern of the affirmative action opponents seemed to be that applicants were being held to different standards based on gender or ethnicity, and I'm pretty sure they mean things like SAT scores and not community service.

Post a comment