Arguments about affirmative action have often been based on misinformation about how applicants are selected in the “real” world. Susan Sturm and Lani Guinier are doing important work in developing an argument for affirmative action that is based on more accurate information about how selection works. Their argument focuses on the selection decisions used in hiring, in admissions to education, and in promotions, and it begins by challenging the implicit assumption that those who are selecting applicants know what they are doing.

It is certainly reasonable to argue that organizations have a right to select the best and most capable applicants. But in many cases, they make no rigorous attempt to do so. Most organizations (including university faculties) make hiring decisions based on unstructured interviews, despite the documented fact that interviews are about the worst method for determining which candidates will make good employees. Perceptions of applicant ability based on the typical method of unstructured interviews have roughly zero correlation with subsequent performance in jobs. Many employers appear to have no systematic arrangements for selecting among applicants.

What may be more surprising is that even systematic attempts to determine who will be a good employee or a successful student are far from accurate. Selection tests and college admission tests are limited to predicting performance in first jobs and first-semester grade point averages, respectively. Even the very best of these tests correlate at levels less than 50 percent with actual performance and most predict far less (even 50 percent correlational relationships translate into predicting only 25 percent of the variance in performance). If employers or higher education institutions cannot tell which applicants are going to be good, then the common objection to affirmative action–that giving preference in admissions or hiring to members of a protected group will in some way hurt the organization by excluding better applicants–is substantially weakened.

The strongest argument against this common objection to affirmative action comes from empirical results about what affirmative action actually does in practice. A recent and important survey of this literature by Harry Holzer and David Neumark reaches several noteworthy conclusions.1 First, affirmative action seems to work in the sense that it promotes distributive justice–that is, it increases employment or participation among women and minorities in the organizations that use it. Second, employers who engage in affirmative action seem to recruit and select more carefully–that is, they look more broadly for applicants and evaluate them against more criteria. Finally, and most importantly, these employers do not appear to pay any price in terms of employee performance for engaging in affirmative action. In fact, the job performance of women and minorities hired by these employers is, if anything, higher than that of white males. Despite the fact that the credentials on characteristics like education are lower for the women and minority employees of firms that pursue affirmative action, their employers seem to have uncovered other attributes about them, through more careful recruiting, that are associated with better job performance.

The argument that affirmative action is at a minimum no worse than the alternative from the perspective of employers seems on solid ground. If the methods that employers are using now have an adverse impact on protected groups, as Sturm and Guinier assert, and employers are not getting much from those methods, then the argument for affirmative action, which has the considerable advantage of enhancing distributive justice, is very powerful.

But what if the alternative here is not typical recruitment and selection practices? What if employers got more sophisticated and adopted the best selection procedures? Many of the best selection tests in terms of predictive power, such as cognitive ability tests, also seem to be ones that have the most adverse impact. Sturm and Guinier have a point in arguing that many paper-and-pencil tests seem to discriminate against minorities in particular. A recent study of higher education admissions and test scores confirms this point in showing that blacks score disproportionately lower on SAT tests than on other indicators of college performance, such as high school grades.2 The same may be true for employment selection tests. Here they offer a sensible alternative: if other practices, perhaps in combination, can achieve the same level of predictive power while avoiding adverse impact, then those other practices should be preferred.

Their other recommendations may be harder to swallow, at least for most employers, because they do not seem to be based on how most employers operate. For example, it is no doubt true that some individuals may be able to perform jobs well in ways other than recruiters and employment-test designers envision. But it is also difficult to ask organizations to let people who do not appear to be qualified try out jobs to see if they can succeed at them in different ways. The costs of failure for most jobs are big, not only the turnover costs if someone does not work out but also the potential costs of accidents and mistakes. The law also makes employers liable for hiring unqualified employees, for example, who cause accidents, if the employers did not take reasonable steps to screen them. Arrangements where otherwise unqualified applicants are hired and then trained and developed to handle jobs are a great idea for expanding opportunity, but are also a significant operating expense that not all employers are equipped to bear, especially in a just-in-time economy. Nor is there any reason to think that team-based, self-managed work systems will result in less discrimination. If employers discriminate, despite economic incentives to hire the best people and human resource departments trained to do otherwise, then there is no reason to think that individual employees will behave better. And while it is true that there are some advantages to a diverse workforce from the perspective of performance, there are also costs, and it is not obvious that the net effect is such that it will necessarily encourage decision-makers to act in the interests of diversity.

It is worth bearing in mind that predicting who is going to be a good employee or a successful student or a good citizen is an enormously complicated exercise. It is difficult to do even after the fact, as anyone who has ever struggled with writing a performance appraisal can attest. There are other issues associated with hiring employees and admitting college students in addition to predicting who will succeed, of course, and these concern the issue of who gets ahead in society. As Sturm and Guinier make clear, it is much easier to argue for affirmative action to advance distributive justice if it does not conflict with the right of employers or other institutions to select individuals that serve the institutions purpose. And more valid arrangements that also advance affirmative action would be highly desirable.

It does still leave us, however, with a vexing problem of procedural justice, the right of applicants to be treated fairly. The standardized tests and civil service exams that Sturm and Guinier rightly criticize on validity grounds at least have the advantage of being objective and consistent for the applicants, if not necessarily valid from the perspective of the institutions administering them. Applications for sought-after and politically sensitive positions like civil service jobs or state university admissions are based on objective tests precisely to quell the procedural justice complaints–and lawsuits–that such decisions otherwise raise. A more diverse set of selection criteria that allow applicants to demonstrate competencies in different ways has the disadvantage of making the process appear less objective. One might well prefer a system that advances diversity while producing competent individuals to one that does neither but seems fair to the applicants, but it would obviously be best to achieve all three goals.

Notes

1 Harry Holzer and David Neumark, “Assessing Affirmative Action,”Journal of Economic Literature 38 (2000): 483-595.

2 William T. Dickens and Thomas J. Kane, “Racial Test Score Differences as Evidence of Reverse Discrimination: Less Than Meets the Eye,”Industrial Relations 38 (1999): 331-363.