An Exam-Grading Hypothetical for Affirmative Action’s Supreme Court Test

[This post originally appeared at the University of Chicago Law Faculty Blog on December 8, 2015.]

It’s the end of the fall term on university campuses across the country, and so professors are gearing up to grade final exams. Meanwhile, the Supreme Court is gearing up for oral arguments this Wednesday in a case brought by Abigail Fisher, a white student who claims that the University of Texas at Austin’s race-based affirmative action program is unconstitutional. End-of-term exam grading gives rise to a thought experiment with potential implications for Fisher’s case—a case likely to be among the most consequential of this Supreme Court term.

Imagine that a professor keeps a Microsoft Excel spreadsheet with all his students’ names, as well as demographic information (race, gender, etc.) and the students’ scores on the midterm and final exams. The professor doesn’t intend to use the demographic information in the grading process; he just wants to be able to know at the end of the term whether there are substantial racial or gender disparities in students’ scores. But there’s a snag in his plan: Unfortunately, the professor makes a mistake in writing the Excel formula that will spit out final grades. The professor meant for the formula to spit out the weighted average of midterm and final exam scores, but absentmindedly and accidentally, he subtracted 10 points from the grades of all the female students in the class.

We might forgive the absentminded professor for his error. Anyone who uses Excel often enough has probably made the mistake of adding the wrong columns at one point or another. But all would agree that the error should be corrected. The professor should add 10 points back to the women’s scores so that the final grades reflect each student’s actual performance in the class.

Might the male students in the class argue that the professor improperly engaged in gender-based affirmative action? I doubt it. I think we would all say that once the professor recognized his unfortunate error, the right thing for him to do was to correct it. Perhaps some would say that the professor shouldn’t have been tracking gender in the first place, though we can also imagine good reasons why the professor would want to know whether men had systematically outperformed women in his class or vice versa. (Maybe the data might lead him to adjust his teaching style or his exam questions next term.) In any event, we wouldn’t say that the 10-point correction reflected a “preference” for female students. Rather, adding 10 points back to the women’s scores made the grading system more meritocratic.

The Excel example is only hypothetical; the possibility of educators accidentally subtracting points from students of a specific demographic group is not. Indeed, we have reason to believe that educators across the country are making a mistake similar to the absentminded professor’s formula-writing error. Some of the strongest evidence comes from the Implicit Association Test, pioneered by psychologistsMahzarin Banaji and Anthony Greenwald. One version of the test asks subjects to classify children’s faces as “African American” or “European American” and to classify words as “pleasant” or “unpleasant.” In some stages of the test, subjects are asked to press the “e” key when they see an unpleasant word (such as “disaster” or “agony”) or an African-American child’s face, and to press the “i” key when they see a pleasant word (“smile,” “honest”) or a white child’s face. In other stages, subjects are asked to associate the “e” key with unpleasant words and white children’s faces, and the “i” key with pleasant words and African-American children’s faces. Researchersrepeatedly find that white adults more quickly associate pleasant words with white faces than with African-American ones. (Evidence of an implicit in-group preference among African-American adults is much weaker.)

If white adults are more likely to link pleasant words with white children’s faces, then that might lead us to suspect that white teachers are more likely to associate positive feelings with white students. And that, in turn, might lead white teachers to assign higher grades to white students than to students of color. This isn’t to suggest that white teachers are closet racists: they might be unaware of their implicit bias, and might strive to correct their own bias if made aware. Here at UChicago Law, professors blind-grade exams in part so that we can prevent implicit biases from affecting the results. But students applying for undergraduate admission to the University of Texas most likely were not blind-graded throughout high school, and so there is good cause to believe that implicit bias has affected their grades and class rank.

Banaji and law professor Jerry Kang have previously argued that implicit bias findings strengthen the case for affirmative action in general, but the findings are especially relevant to the particular facts of the Fisher case. In 1996, the University of Texas at Austin introduced a new undergraduate admissions program that guarantees spots for Texas high school students who finish in the top 10% of their graduating class. More than half of Latino students in Texas and 40% of African-American students attend high schools with minority enrollment of 90% or more; this de facto segregation makes it a mathematical certainty that the top 10% at those high schools will include minorities. Indeed, more than half of African-American and Latino students admitted under the Top 10% Plan hail from high schools with minority enrollment of 90% or higher. One of the reasons why University of Texas says it wants to use race-based affirmative action is so that it can assure that African-American and Latino students from racially diverse high schools are represented on campus too. (Since Fisher’s lawsuit was filed, Texas has amended the plan so that only the top 7% to 8% of students in each high school graduating class are guaranteed admission, but otherwise the details remain essentially the same.)

Abigail Fisher’s lawyers say that the University of Texas has no good reason for “favoring” minority students who have been educated at racially integrated high schools. In Fisher’s view, the set-aside for the top students from each high school already assures that minority students from predominantly minority high schools will be represented on the UT Austin campus, and the university has no additional interest in ensuring that minority students from integrated high schools will be represented as well. But the results of the implicit bias research suggest otherwise. For an African-American student at a high school that is virtually all black, implicit bias is unlikely to affect class rank; the student is being compared to classmates of the same race. But for African-American students at mixed-race high schools, the effect of implicit bias is potentially more pernicious. If white teachers unconsciously favor white students when grading, then an African-American student’s class rank at an integrated high school won’t be an accurate measure of the student’s performance. Rather, the African-American student will find himself in the same position as the female students in the Excel hypothetical who had points arbitrarily deducted from their score.

If you thought the professor should add back 10 points to the female students’ grades in the Excel hypothetical, and if you accept the overwhelming weight of the implicit bias research, then you should probably support the University of Texas policy as well. An admissions preference for minority students from integrated high schools is a way for the university to add back points incorrectly deducted by earlier graders—that is, by the white high school teachers who gave grades reflecting this own implicit bias. In other words, giving a preference to minority students from integrated high schools is a way to measure merit more accurately—which is to say, it’s not a “preference” at all.

Not everyone will be convinced by this argument. Perhaps some students in integrated high schools didn’t have white teachers (though that seems unlikely—65% of public school teachers in Texas are white, despite the overall student body being majority minority). Not all white adults exhibit a pro-white implicit bias (though in some studies, more than three-quarters do). And it is difficult to know whether implicit bias bleeds over into grading (though there is evidence that it affects trial judges as well as employers evaluating resumes). Moreover, the affirmative action policy at the University of Texas gives a boost to Latino students as well as African-Americans, whereas the implicit bias findings are strongest with respect to anti-black bias (though there is evidence of an anti-Latino bias too). And unlike the Excel hypothetical, where it was clear how many points the female students lost due to the spreadsheet error, it’s harder to know how many GPA points to add back for minority high school students.

So a closer hypothetical might be as follows: Let’s say that the absentminded professor can’t be sure whether he accidentally deducted 10 points from all the female students’ scores. He strongly suspects that he did, but he has deleted the spreadsheet and lost the original exams. Should the professor add 10 points to the female students’ scores based on the belief that he probably (but not certainly) subtracted those points by accident? In other words, should the professor take a step that will probably (but not certainly) make the final outcome more meritocratic? That, in a nutshell, is the quandary facing the University of Texas. The university wants to add those points back. The question now is whether the Supreme Court will let it.