For anyone still in doubt, a deeply statistical analysis by the National Bureau of Economic Research— complete with “Epanechinikov kernels” and “Silverman bandwidths” — of the effect of banning racial preferences in admission to the the UC Berkeley (Boalt Hall) and UCLA law schools demonstrates that eliminating racial preference reduces the numbers of previously preferred minorities admitted, at least to the degree that the preferences were in fact eliminated. (The study, “Affirmative Action Bans and Black Admission Outcomes: Selection-Corrected Estimates From UC Law Schools,” can be found here, free for some and $5 for others; the abstract here; and short discussions by the Chronicle of Higher Education and Inside Higher Ed here and here. )
Most academic readers of the study, by Danny Yagan, an assistant professor of economics at UC Berkeley, will probably be impressed by the sharp reduction found in minority admissions. Others of us, however, will marvel at what the analysis reveals of how successful those two law schools have been at evading the ban. As Yagan puts it in a phrase that is either quite droll or, more likely, an example of the often euphemistic awkwardness that characterizes much social science writing, “affirmative action ban avoidance is far from complete.”
Specifically, Yagan finds that the ban on racial preferences “reduced the black admission rate from 61% to 31%” at the two schools, but “black admission rates would have fallen to 8% (not just 31%) had all races been subjected to observed pre-ban white admission standards based on LSAT, GPA, and inferred strength.” Even after the “ban,” in short, Yagan finds that the two law schools “used unobserved black-correlates … to admit blacks at a 63 percentage point higher rate than observably similar whites.”
Yagan concludes that the primary cause of the reduced admission rate for the previously preferred minorities was that “the black share of the applicant pool permanently declined by half,” and thus “the post-ban recovery in black admission rates” — raising to 31% what with racially neutral standards based on credentials would have been 8% — resulted from the two schools “learning to sustain most of [their] pre-ban black admission advantages.” Although it was clearly not his intent, Yagan’s study provides strong support for the conclusions reached by UCLA Professor Tim Groseclose in his book, Cheating: An Insider’s Report on the Use of Race in Admissions at UCLA as well as the conclusion of Richard Sander and Stuart Taylor in Mismatch that outright bans will simply lead to cheating, a problem I discussed in The Cheating Defense Of Affirmative Action.
Far from suggesting that the two schools were scofflaws, however, Yagan seems disappointed that even more of those “pre-ban black admission advantages” were not preserved. Post-ban, he finds, “he estimated maximum black-white admission rate difference was 57 percentage points at Berkeley and 69 percentage points at UCLA,” but if they had placed more weight on “those black-correlates relative to observed credentials like LSAT and GPA, UC schools could likely have admitted black applicants at substantially higher rates.” In short, as he states in the abstract, “UC schools were technologically able to sustain substantially higher black admission rates after the ban but were either unwilling or legally unable to do so.”
As mentioned, this paper is statistically dense, and I am not qualified to evaluate its methodology. (Richard Sander, please come forward.) I do, however, want to offer an observation on one of its key points. For reasons I confess I do not completely understand, Yagan finds it essential to “hold the applicant pool constant at pre-ban levels along three directly observed characteristics — test score, undergraduate grade point average, and race — that together correctly predict over 89% of admission decisions.” That was presumably easy enough with objective criteria, but now note his additional step:
I further hold constant a powerful summary measure of not-directly-observed applicant strength, inferred from admission decisions at non-UC schools. This inferred strength measure is based on the intuition that if an applicant was consistently admitted at non-UC schools in spite of weak observed credentials, the applicant was likely strong on commonly valued unobserved credentials like recommendation letters. The presence of thousands of independent screens at non-UC schools is a major advantage of these data.
Yagan’s “intuition” here is either naïve or disingenuous. He assumes the fact that these black law school applicants had been previously admitted to non-UC, often “elite schools” as undergraduates means their non-observable credentials were unusually strong, but it is equally if not more likely that they were simply the beneficiaries of preference based on their race, preferences that seem to have survived the “ban” in California quite nicely.