Replying to Crook’d Dope:
Thanks for your sane, rational, and well-written reply.
If you’re willing, for now, I’d like to confine the discussion of AA to two groups which do not have a long history legacy in the US. For there are (at least) three rationales for AA. The first is an argument that AA is necessary to correct for some kind of mismeasurement of potential. (The argument I attempted to address in my first post.) The second is that AA is some kind of policy necessary for social justice due to the past and current injustices against blacks or Amerinds. (Your point D.) The third would be some kind of argument that racial diversity is a public good for the non-AA student population.
The case of blacks and Amerinds is both more complicated and more persuasive than that for other groups. So, if AA were invalid for the simpler case of the Hmong, it could still be valid for the more complex and stronger case for blacks and Amerinds. On the other hand, if AA is valid for the Hmong, it is then also valid for others.
Okay then. Suppose we have three types of data on our applicants:
A) Educational data (transcripts, tests, activities, letters of recc, etc.)
B) non-racial socioeconomic data (parent’s education, income, geographic location, etc.)
C) Sex and Self-reported Sexual Orientation Data
D) Self-reported Religious Data
E) Self-reported Race/Ethnic Data
My claim is this:
It’s only valid to use data for admissions (temporarily leaving aside reparataions-type arguments and diversity-as-public-good type arguments) that have predictive power for how well the student will perform at college.
Therefore, we develop some measure of student college performance (SCP). Maybe it 's GPA, maybe it’s how much money they contribute back to the colledge as alumni, maybe it’s how many Google references they produce 10 years after graduation. Maybe it’s some combination of all the above.
Anyway, we run some regressions, trying to explain SCP=f(A), SCP=f(A,B), SCP=f(A,B,C), SCP=f(A,B,C,D). We look at the amount of variance in SCP explained by different combinations (R-sqrd, in stat parlance), and in particular, at how much this increases as we add each incremental set of variables.
I feel perfectly fine accepting A as a criterion of admission. It bears a direct (but imperfect) relationship to the educational mission of a college. (Unlike, say, height.) It is the type of data that is very strongly related to the student’s inclinations and efforts–it has much to do with their autonomous self-definition. Suppose lower type A is correlated with lower SCP. (Hopefully, this is non-contentious.)
I feel a little less fine about accepting type B data as a criterion of admission. It bears an indirect (but plausible) relationship to the educational mission of a college. It is not the type of data that corresponds to autonomous choice and self-definition. Here we are in some sense rewarding or penalizing students for factors they had no control over. But, at least, we’re rewarding or penalizing them for factors that someone (say, their family) had some control over.
Suppose people with ‘lower’ type B data also have lower type A data. Suppose that our regression discovers that, holding type A data constant (this is what regression does), lower type B predicts higher SCP. That is, suppose being poorer, given the same GPA, indicates increased college GPA (say). Then we can say that the GPA (or whatever) underestimates poor students’ potential. Therefore, ‘economic AA’ would seem to be appropriate. If, on the other hand, we find that lower B, holding A constant, predicts lower SCP, then GPA (or whatever) overestimates poor students’ potential. Therefore, ‘reverse economic AA’ would seem justifiable. If B has no predictive power, then ‘class blind’ admissions would seem just.
Consider sex and sexual preference. Here we have a variable that most men and women, and most heteros and homos consider immutable. I’m deeply nervous about using characteristics that are immutable in college admissions. It seems to reward or penalize people for things no one had any control over. But even this would seem (ignoring the reparations issue, though not the diversity issue) preferable to using religion. The Constitution frowns on showing preference or prejudice based on religion. The Constitution doesn’t say anything policies based on sex or sexual preference. Yet.
But even so, at least religion is something we can choose. The constitution not only frowns on the government showing preference or prejudice based on race, but race is not something we can choose–leaving aside people of mixed descent and post-modernist wiseasses.
Therefore, it behooves us, IMHO, to be very leery of using race or religion as a factor in something like college admissions.