So what do I think about what should be done about the “affirmative action” problem? I’m sorry you asked that question, but I suppose I ought to answer it.
I think that in some cases diversity serves legitimate organizational interests, but I doubt that “diversity” is a good shorthand for the goals to be served by racial preference programs. Racial preferences and other forms of affirmative action ought to be evaluated, in my view, as means to the end of shrinking the inter-ethnic gaps (and most of all the gap between African-Americans and the rest of the society) on various measures of income, wealth, well-being, and social influence, toward the goal of creating a society in which being black is no more important in determining one’s social position and opportunities than being, say, of Irish descent is now. [Which, let’s not forget, is substantially less important that was the case one or two or three or four generations ago: being Irish in 1870 was probably as important as being black today.]
I would say the same thing about social class: Children whose parents never attended college ought to get some sort of leg up in the elite college admissions process, both because I think the yuppie puppies would benefit from meeting as equals some people who chose their ancestors less wisely and because the children of the poor have had, on average, less than an even break, and increasing the rate of intergenerational social mobility is consistent with American ideals.
Given that as a goal — which I acknowledge not everyone shares — then the question is how best to accomplish it. Here’s how I would define the problem formally:
Assume we have Y applicants for X slots at, let’s say, an elite university, and that we can rate those applicants according to some scoring system in which group membership is not a factor. Call that “quality.” (Calling it “merit” suggests that persons of the highest “quality” deserve preference over those of lower quality, which may or may not be the case.) If “quality” is the only thing that matters, then the optimal choice rule is to pick the applicants in descending order of measured quality until all the slots are filled.
We might also care about the composition of the group: needing a critical mass of some characteristics (enough musicians to fill the college orchestra), at least a sampling of others (Sinhalese? Montanans? Conservatives?) for the sake of diversity in its literal meaning, and not too many of still others (pre-med or pre-law students, perhaps). Or we might not want a college class consisting entirely of tall girls and short boys, on the theory that such an arrangement would interfere with social life. Such considerations would complicate the choice algorithm.
Now assume that there is some group G that is “under-represented” in the class resulting from the choice algorithm we select, in the sense that we’d like to have more, rather than fewer, members of G in the final mix, other things being equal. Assuming that the “quality” measure is valid, we’d also like to have more quality, rather than less, again other things equal. Those two objectives will trade off. Once we’ve taken all the G-member applicants who rank as highly on quality as the lowest-ranked non-G-member applicant accepted, then every additional G-member we take will reduce overall quality as we measure it.
Pretending that the quality of applicants is some ineffable characteristic not subject to measurement, or that the complexity of needing enough oboe players and not too many nerds makes the whole process irreducibly arbitrary, is simply obscurantism in the service of refusing to face the hard problem. I understand why many people would prefer to avoid saying that there are proportionately many fewer blacks than whites among high-school graduates with the very best academic preparation, but if it weren’t true then we would have to re-estimate the legacy of slavery and Jim Crow. Undeserved suffering may be, as Dr. King liked to say, redemptive (though I rather doubt it) but we can’t just wish away the results of generations of poverty and discrimination.
Given the existence of the tradeoff, what we want is a process that will give us as much overall quality as possible for any achieved level of G-membership, and as much G-membership as possible for any given sacrifice of overall quality. The approach that accomplishes that is to rank-order G-members and non-G-members separately, and take from the top of each list. (This is mathematically equivalent to the approach called “race-norming” when applied to standardized testing programs.) We can set a quality cutoff, take all G-members down to that cutoff, and then fill the balance of the class the best non-G-members; we can set a G-membership floor (“quota”), take that many G-members, and fill the balance with the best non-G-members, or we can decide where to make the cut once we have the applications in hand, based on how well we’re doing on each objective and on the terms of the tradeoff as we face it.
That might mean setting an upper limit to how big the gap we’re prepared to allow between the lowest-quality G-member accepted and the highest-quality non-G-member rejected. That gap measures both the quality cost of “diversity” and the likely visibility of the “quality” difference between the two groups of students, or graduates, which is certainly one negative feature of aggressive preference programs.
(Oddly enough, I have never heard the beneficiaries of legacy admissions policies, including the one that got our current President into Yale, or of geographic diversity programs, such as the one that got our former President his Rhodes Scholarship, complaining about the resulting stigma.) [UPDATE: A reader who attended Yale College and Harvard Law School reports that the stigma was nonetheless real: “I was actually quite shocked at the level of disdain for students who got in through legacy admissions (and everyone knew who they were).” So may I now assume that the same folks who oppose racial preferences out of concern about stigma will oppose legacy admissions for the same reason? After all, it isn’t fair to the legacies who actually could have gotten in on their own to devalue their credentials by association with people who are just relatives of buildings.])
Attempts to avoid facing the terms of the quality/diversity tradeoff through “race-neutral measures” must increase the quality sacrifice required to achieve any given level of diversity, because in general they will lead to taking some lower-quality non-G-members in preference to higher-quality non-G-members merely because they share some characteristics (e.g., attendance at certain high schools) with G-members. That is, “race-neutral measures” require you to foul up your entire selection process, including the part that applies to sorting applicants from the non-target group, in order to get the “diversity” you want. (Some police departments have abandoned the use of IQ testing for candidate cops in order to avoid screening out too many African-American applicants.)
In return, race-neutral measures allow you to avoid facing either the fact that some non-G-members are being excluded because of their non-G-membership or the precise terms of the quality tradeoff. Too high a price, in my view.
A note from a reader convinced me that the above is obscurely written. Perhaps an example will help:
Imagine you run a police department that recruits cops based in part on an IQ test, which turns out to be a pretty good predictor of job performance; higher IQs correlate, among other things, with lower rates of infliction of severe injury or death on suspects. [No, I don’t want to have the IQ argument now; pick your own predictor if you’re IQ-phobic.] But the department is also concerned about having a good representation of group G among cops, and it turns out that G-members tend to underperform non-G-members on IQ tests.
You could fix that by keeping the IQ test and picking both G-members and non-G-members based on their IQ scores, but using a lower cutoff for G members. That would be a non-race-neutral “quota” system. It ensures that you get the “best” (by this partial measure) G-member and non-G-member recruits, minimizing the sacrifice of average IQ score for any given target representation level for group G.
Alternatively, you could eliminate the IQ test, or keep it but simply set a very low cutoff score for “adequate” IQ and make everyone above that level eligible to be hired. If the distribution of IQ scores is lower among G-members, then lowering the cutoff score will tend to increase the representation of group G among those selected. That’s a “race-neutral” measure. Because you’re now selecting neither G-members nor non-G-members based on your quality measure, the overall measured quality of the group you hire will be lower under this system than under the forbidden “quota.” Both the G-members and the non-G-members hired will, on average, have lower IQs than would have been the case using an explicitly race-conscious selection mechanism, and you wind up with a dumber and more brutal police force because you didn’t want to be explicit about your need to hire G-members and the costs in quality terms of doing so.
More here; Glenn Loury supplies some numbers.