The Efficiency of Race-Blind Affirmative Action Plans

By Glenn Ellison

Posted on June 6, 2017


In public secondary education and university systems it is common to offer some programs only in a few elite schools. Flagship public universities offer programs and research opportunities not found on other campuses. And in many cities only a few high schools offer the highest level STEM courses. In the 1970s many public and private K-12 and university systems adopted race-based affirmative action policies. These plans gave talented minority students access to these high-level courses. And systems also often view affirmative action plans as benefitting all students by creating richer educational communities.


More recently, race-based plans have come under attack on two fronts: some states have passed laws or constitutional amendments banning affirmative action; and the US Supreme Court has set a high bar for the use of race. In 2003 it ruled in Grutter vs. Bollinger that “strict scrutiny” must be applied to race-based plans. And its 2007 ruling in PICS vs. Seattle striking down assignment plans in Seattle and Louisville made clear that the same standard applied to public K-12 systems.


Many public K-12 and university systems have as a result shifted from race-based affirmative action plans to alternatives that do not explicitly consider race or ethnicity. Some examples are Texas’s “top 10%” rule which initially guaranteed admission to any state university to students graduating in the top 10% of their high school class, and a plan adopted by the Chicago Public Schools which reserves a number of places in the incoming class at each of its magnet schools for students living in low-SES neighborhoods.


Race-blind affirmative action policies seem appealing on multiple dimensions. They target disadvantaged students and hence should increase socioeconomic diversity. And, given the correlation between race and disadvantage in the US, they will also increase racial diversity. Much less, however, is known about how well race-blind policies work in practice. In a recent working paper, “The Efficiency of Race-Neutral Alternatives to Race-Based Affirmative Action: Evidence from Chicago’s Exam Schools,” Parag Pathak and I take advantage of unique data on the plan implemented by the Chicago Public Schools (CPS) to provide some of the first evidence on the topic. We provide a number of new observations. Among our conclusions are that existing race-blind programs may not work nearly as well as was hoped, and that it is challenging to find effective substitutes for race-based policies.


Legal decisions have held that race-based plans are only permissible if the same objectives cannot be effectively achieved in a race-neutral manner. Accordingly, many have focused on the question of whether race-neutral plans can achieve the same level of minority representation as race-based plans. Our view is that this method of evaluation is misguided. The objective of an affirmative action plan at an elite school cannot be just to increase the diversity of the class. Indeed, the more important objective of any elite school’s admissions process must be to admit highly talented students who will benefit from the school’s unique programs. With this perspective we propose that an important criterion for assessing affirmative action plans is the relative “efficiency” with which they achieve some goal. We define efficiency as the fraction of the decrease in the preparation of admitted students which was necessary to achieve the goal. For example, if the reduction in average entrance exam scores that is necessary to increase minority representation by one percentage point is always twice as large under a race-neutral program as it is under a race-based program, we would say that the race-neutral program is 50% efficient as a means to increase minority representation.


Our empirical analysis focuses on the two most elite high schools in the Chicago Public School (CPS) system, Walter Payton College Prep and Northside College Prep. In 2009 CPS replaced a system which had relied on racial quotas with what is perhaps the most sophisticated race-neutral system in existence. It begins with a fine-grained division of the city into roughly 800 neighborhoods. Census data on six proxies for disadvantage—for example, median household income, the percentage of single parent families, and test scores in the local elementary school—are used to classify each neighborhood as belonging to tier 1, 2, 3, or 4, which can be thought of as neighborhoods with high, medium, low, or very low socioeconomic status. The admissions policy for the elite schools then uses quotas for students living in neighborhoods classified as belong to each socioeconomic tier to ensure that students from neighborhoods of all types are represented. Thirty percent of the seats are allocated to the highest-scoring applicants. And the remaining 70% of the seats are then allocated to the highest-scoring students from each tier. The proxies for disadvantage are also correlated with race and consequently the plan has some of the same effects as the former plan in terms of boosting racial diversity. Under a purely score-based admissions policy, only about 20% of the students admitted to Walter Payton and Northside would be underrepresented minorities. The CPS plan boosts this to about 35%. (Underrepresented minorities are about 85% of the overall CPS population.)


Although the CPS plan boosts minority representation, it is easy to provide anecdotal evidence that it does not always do what one would like. For example, we identify free-lunch eligible students with composite entrance exam scores of 99.0 out of 100 and minority students with scores of 98.8 who are denied entry to Walter Payton. Meanwhile, Payton admitted non-free-lunch-eligible majority students with scores as low as 93.9. Some such examples are inevitable when one relies on proxies for disadvantage rather than data on whether a particular student is personally disadvantaged. Our analysis of relative efficiency can be thought of as reflecting how common such mistakes are.


Our most basic finding about the CPS race-neutral affirmative action plan is that it is only 20-25% efficient as a means to increase minority representation. By this, we mean that the composite scores of admitted students are being reduced by four to five times as much as would have been necessary to achieve the level of minority representation that policy achieves if one were allowed to explicitly consider race. We also note that the extent of minority representation that can be achieved with plans like the CPS policy is quite limited. Even if one were to turn the dial up to 100 percent and allocate all seats by SES tier, one could not increase minority representation at Payton beyond 50%.


Race-neutral affirmative action plans like that of CPS should not be judged purely on the efficiency with which they increase minority representation. These plans have grander ambitions—they aim to increase the number of disadvantaged students of any race—and a plan that is simultaneously pursuing multiple goals could not possibly be as efficient as the best means to achieve any one of the goals. For this reason, we also evaluate the CPS plan as a method for increasing the number of low-income students admitted to Payton and Northside. Here, we obtain an even more disappointing finding: the CPS plan is only 10-15% efficient as a means to increase low-income representation. This reflects both that the admissions process passes over many high-achieving, low-income students whose parents have chosen to live in high-SES neighborhoods (perhaps so their children can attend a high-achieving public school), and that some admitted students from low-SES neighborhoods are not themselves poor.


Indeed, a particularly striking observation on this topic is that, holding entrance scores fixed, CPS’s explicitly SES-focused plan actually admits fewer low-income students than would a purely race-based plan. This may seem paradoxical. How can a plan that explicitly weights factors like neighborhood median income not admit more low income students than a plan that just looks at race? Our main insight here is that in thinking about these plans one should always keep in mind that high-scoring students are inherently unrepresentative. It is easy to predict whether a student chosen at random from a given neighborhood is likely to be poor. But when we are told that a student from a low-SES neighborhood achieved a 98th percentile score on a standardized test we learn that the student is not representative of his or her neighborhood. We have to throw out much of what we thought we knew about him or her. A high scoring minority student is also much less likely to be poor than a randomly chosen minority student. But apparently in the population of high-scoring students, minority status is a better predictor of having low income than is living in a low-SES neighborhood.


One aspect of race-based affirmative action plans that sometimes attracts criticism is that they could create within-school racial achievement gaps, thereby contributing to stereotype formation. The hope that race-blind policies may be preferable on this dimension is another part of their appeal: the students from low-SES neighborhoods they favor can be of any race, so achievement gaps could be smaller and not clearly aligned with race. Our empirical analysis of this issue again turns out to be disappointing. We only have data on student preparation, not achievement post-admission, but note that the CPS plan widens racial achievement gaps in entrance exam scores. Intuitively, while the CPS plan does admit some less-prepared majority students, it also misses many high-achieving minority students (living in medium and high-SES neighborhoods). Apparently, in Chicago this latter effect is more relevant.


Students applying to CPS magnet programs submit rank-ordered lists indicating their preferences over schools. Using this data we can simulate how students would have been assigned to schools under alternate admissions policies that CPS could have implemented. For example, we can compare the CPS policy with alternate race-blind procedures along the lines of the Texas Top 10% policy: we can say what would have happened if seats in the elite schools were allocated by neighborhood or census tract rather than by SES tier. Here, we find that allocating students by neighborhood area would be somewhat more efficient as a means to increase minority representation than the current CPS policy. But it would do so in part by admitting fewer low-income students.


Moving beyond this, we push further on the question of how the CPS plan might be improved. The CPS plan defines low-SES neighborhoods as neighborhoods that rank low on six dimensions, including having low median per capita income and high percentage of single-parent families. Obviously, disadvantage could alternatively have been defined in many different ways. We note that it is easy to make the CPS plan a little more efficient. For example, one can do this by replacing some of the proxies for disadvantage that CPS is now using with others, or even by just reweighting the variables they are already using. But the gains from these simple modifications are limited. And we find that it is hard to make race-neutral affirmative action plans much more efficient even if one uses many more variables than Chicago is using and exploits them in statistically sophisticated ways.


We are left with the message that we can improve on existing plans, but race-blind alternatives to affirmative action will not work as well as one might have hoped: they will miss many highly talented minority students; they may not actually admit more economically disadvantaged students; and they may worsen within-school achievement gaps. Whether race-blind or race-based plans are preferable is a matter of society’s preferences. But, policy makers should recognize that there is no win-win solution and abandoning the use of race in elite-school admissions will have efficiency costs.


In the near future, it seems unlikely that public school systems will return to assignment policies in which race place a substantial role. Accordingly, a practical implication of our research for the tech industry is that many talented students from underrepresented minority groups will have more difficulty gaining access to elite schools than they would have a few years ago. Firms seeking to diversity their workforce will need to be aware of this change and the challenge it presents.



Note: In a related topic, Professor Glenn Ellison examines efforts to diversify the IT workforce.
Read more: “Diversity in the IT Workforce: Precursors in Education and Challenges



About the Author

  • Glenn Ellison
  • Massachusetts Institute of Technology
  • 50 Memorial Dr.
    Bldg. E52-424
    Cambridge, MA 02142

Recent TAP Bloggers