[dragonwriter]

> It's true that AA is a non-meritocratic hack on top of an otherwise meritocratic system

I'm not sure that's true, and am fairly certain its not supportable as true. While the exact models used in admission have evolved, even fairly recent studies have shown that many of the objective factors considered in admission aside from AA and contributor/alumni preferences have fairly weak predictors of success, and some of them (particularly standardized admissions tests) are measures which are very strongly tied to race/ethnicity, disadvantaging traditionally-disadvantaged groups when they are used. (Now, some schools have adapted their admissions processes to eliminate some of the more problematic factors, but that leaves them with systems that are unproven -- which is an improvement over systems that are proven to be biased rather than meritocratic, but not the same as demonstrably meritocratic.)

[yummyfajitas]

The fact that some races/ethnicities perform worse on SAT or other explicitly race-neutral metrics does not mean they are not meritocratic. It just means those groups have lower average merit.

Similarly, the fact that any individual component of a predictor has little predictive power is a truly terrible critique. Similarly, no individual pixel is particularly predictive of the content of an image. Therefore image recognition is impossible!

[dragonwriter]

> The fact that some races/ethnicities perform worse on SAT or other explicitly race-neutral metrics does not mean they are not meritocratic.

That's true, and even the fact that taken together the main objective admission criteria aren't (even considered together) strong predictors of performance taken together with that doesn't mean that the systems using them aren't meritocratic, its just strongly suggestive of that.

However, there is plenty of reason to believe that a system that doesn't account for demographics -- including race and income and possibly other factors -- can't be effectively meritocratic or (perhaps surprisingly) race-blind. For instance, studies have shown that the relationship between expected performance (in terms of college grades) -- both in terms of predictive power and expected results -- of SAT scores, college grades, advanced coursework, etc. is not consistent across different racial and income-based demographic groups. [0]

So a system which ignores those differences and just applies the measures by a one-size fits all formula is not adopting a race (etc.)-blind measure of merit.

[0] e.g., this analysis http://ftp.iza.org/dp8733.pdf which itself also references a study identifying that what predictive power SAT scores have is mostly as an indirect measure of the high school the student attended, and that within-school variations in SAT scores have almost no predictive power.

[yummyfajitas]

You seem to be suggesting that we should actively use racial and other demographic characteristics as part of a predictor.

Suppose we run our linear regression or random forest, and it turns out holding all else equal, black people underperform others. I.e., same SAT, same GPA, the black guy is likely to perform worse. (Please read your source and note the sign on the black/hispanic coefficients.)

You seem to be advocating that we should then penalize the black guy even though his grades are identical. Is that a fair statement of your post?

I don't have a strong opinion on this, though I definitely have a negative emotional reaction to it.

[dragonwriter]

I'm saying that to be actually meritocratic, where the merit to be assessed is expected college performance, using such factors (or, better, identifying the underlying reasons that existing measures do not work consistently across race, etc., and correctly using the underlying factors that research identifies) may be necessary. But I'm not arguing that admissions should be meritocratic, so by suggesting that something would be necessary for meritocracy, I'm not suggesting that it should be done. What I am arguing is that the claim that, but for AA, the admissions processes in use actually are meritocratic was, and remains, unsupported.

[yummyfajitas]

According to your source, SAT, GPA and AP/IB courses are all positive and statistically significant predictors of performance.

Am I correct that according to you, the main gap between an SAT/GPA based metric and meritocracy is applying the appropriate penalty to blacks/hispanics and the appropriate bonus to females, as per the regression table in your source?