Monday, June 26, 2006



I recently learned of an article by Richard Sander, a law professor at UCLA (who was actually part of my dissertation committee), documenting the costs of adopting the binary worldview in a continuous world. The article (go here for a pdf version) claims to show using various statistical analysis that the negative effect on quality of law-school affirmative action is gigantic. Further, the unit in the language used below functions poorly, in that we may not be creating many new black lawyers. If you are familiar with statistical arguments and can tolerate the conventions of law-journal writing it is worth a read.

Recently medical students in India engaged in angry public demonstrations over the possibility that affirmative action – “reservations,” as they are known there – for lower castes and tribes would be expanded. Some of their argument was couched in terms of individual unfairness, but much of it indicated that the performance of India’s elite institutions, including its famed technical institutes, would be severely handicapped by substituting such criteria for qualifications in determining admission.

Is affirmative action – the use of hardwired tribal criteria such as "race" or "Hispanic" status, caste, and in some contexts religion or sexual orientation – costly? Few issues are as divisive in contemporary American life, and in international terms we are actually lucky. The use of such preferences is far more pervasive in other countries (e.g., India and Malaysia). And there the criticism is even louder than it is here that the outright reservation of places for or, in the U.S. case, the giving of an edge to designated groups in competition for scarce spaces in business or higher education causes people who “deserve” to be there to be displaced by people who don't.

The way people answer this question is always revealing of the prior beliefs they bring to the table. Note first that both proponents and opponents often talk in terms of how well the unit potentially subject to affirmative action will function – how high-quality the university or business will be in creating the best-educated citizenry or in maximizing profitability, how well the legislature will represent the function and enact valuable laws, etc. (To be sure, opponents sometimes talk in terms of the injustice to individual applicants denied a place because of their unacceptable tribal identity.)

With that in mind, proponents of affirmative action don’t support admitting just anyone. Instead, they argue that as long as people are “qualified,” why not admit historically disadvantaged groups, either as an act of cosmic justice or as a way to “diversify” the body for everyone’s benefit? (People forget now, but Lewis Powell’s controlling opinion in Bakke v. Regents of University of California held that such diversity in the student body was the only justification for race-based admissions to the medical school at UC-Davis.) There are thus two kinds of people in the world – "qualified" and "unqualified" – and if one "qualified" person is merely exchanged for another, the overall unit suffers no harm.

To opponents in contrast, "qualifications" is not a binary variable. Instead, it is continuous – some people are more qualified, some are less. To admit someone on affirmative-action grounds is to risk damaging the overall ability of the unit to do whatever it is supposed to do.

So which is it? Clearly it cannot be literally true that people are merely qualified or not. In principle, with full information, we could rank every applicant from most to least qualified. Of course we do not have full information. Instead all we have are indicators, some noisier than others. An employer does not have knowledge of the employee’s future work behavior. Instead he has only reference letters, some record of previous employment history, college transcripts, etc. Similarly, a university has references, high school grades, extracurricular and public-service activities, and standardized-test scores.

And how good are these instruments? In education, is the available information enough to rank candidates, and hence document that the institution will be damaged by substitution of less for more qualified candidates? (I abstract from the possibility that tribal identity itself is a qualification, i.e. the benefits of diversity.) Critics of standardized testing often argue that it reveals little about how applicants end up doing in college, with the admission of students who do poorly at the expense of those who would do better presumably being a major cost worth avoiding. The pressure group Fairtest is a longstanding critic of all standardized testing. They claim in places such as this that the SAT is a near-useless predictor of college success:

Validity research at individual institutions illustrates the weak predictive ability of the SAT. One study (J. Baron & M. F. Norman in Educational and Psychology Measurement, Vol. 52, 1992) at the University of Pennsylvania looked at the power of high school class rank, SAT I, and SAT II in predicting cumulative college GPAs. Researchers found that the SAT I was by far the weakest predictor, explaining only 4% of the variation in college grades, while SAT II scores accounted for 6.8% of the differences in academic performance. By far the most useful tool proved to be class rank, which predicted 9.3% of the changes in cumulative GPAs. Combining SAT I scores and class rank inched this figure up to 11.3%, leaving almost 90% of the variation in grades unexplained.

Interpreted properly, this suggests that of the variation currently measurable, the combined SATs probably explain somewhere between five and ten percent. It is true that "almost 90% of the variation in grades" is still “unexplained,” but so what? Studies of wages typically use years of experience and of education as predictors, and these studies routinely leave over 80 percent of the variance unexplained, and yet no one would content that education and experience are unrelated to earnings. In both cases the variables - the SAT on the one hand, education/experience on the other - are highly significant predictors, but leave much to be explained. But the unexplained variance is by definition not something that admissions committees can use rigoroulsy, particularly in light of the undoubtedly very flexible way they define such other factors as "life experience." And so the whole argument that SATs aren't close to completely predictive is amiss.

It is also frequently argued that graduation rates and college GPAs are not thoroughly explained by SAT scores. But this does not control for what sorts of majors low-SAT students admitted on, for example, affirmative-action grounds choose, and indeed whether such majors are substantially populated by such students. One study (Elchanan Cohn et al., Determinants of undergraduate GPAs: SAT scores, high-school GPA and high-school rank, Economics of Education Review 23 (6), 2004) attempted to assess the predictive power of SAT scores by using the students in a single course – principles of economics. They find that high-school class rank and GPA basically double-count, and find that if SAT scores were dropped from the university they studied undergraduate GPAs would drop substantially, something not true for the other measures.

And so it does appear to be true that in the educational arena, there is “more” and there is “less” qualified. The available measures for qualifications are far from ideal, but the ones that exist allow a fairly precise ranking of applicants from most to least qualified. And that in turn suggests that the costs to going beyond the tribally neutral set of qualifications (again, ignoring tribal identity as a qualification in itself) comes at a substantial cost. If it is true for universities, where the penalties for failure are for a variety of reasons (including, often extensive state subsidy) relatively modest, it is undoubtedly all the more compelling true in the case of affirmative action in business, where the penalties for hiring less qualified workers are considerably greater (as Southern entrepreneurs knew when they were busy integrating businesses in the interval after the Civil War but before Jim Crow). That legally required (as opposed to purely voluntary) affirmative action is costly seems almost beyond dispute. That so many groups favoring it speak in terms of only the "qualified" and the "unqualified" is linguistic trickery, designed to evade the hard truth that qualifications differ and are measurable. The only remaining justification for affirmative action is then one of preserving social harmony among tribal groups, an objective that raises disturbing questions of its own.


Post a Comment

<< Home