Coase Lecture Examines the Enduring Harms of Statistical Discrimination

In 2016, the National Football League settled a class action lawsuit on behalf of former players who said they had suffered traumatic brain injuries. In order to access the settlement funds, these players had to undergo a medical assessment that included submitting cognitive test scores. What they didn’t know at the time was that each of their scores had been “corrected” based on their race.

This “race-norming” practice is one high-profile example of how statistical generalizations can be used to rationalize discrimination, Professor Sonja B. Starr said during the 2022 Coase Lecture in Law and Economics.

“The way [the scores were] corrected was that each player's performance on the cognitive test was being evaluated relative to the norms for their race based on a past study,” said Starr, who is the Julius Kreeger Professor of Law & Criminology. “Because Black people had scored lower than White people in that past study, it meant that Black players had to basically perform in a way that demonstrated more cognitive impairment […] in order to qualify for a payout.”

In other words, Starr said, scores that were considered cognitively impaired for White people were deemed normal for a Black person.

The NFL’s race-norming scandal isn’t an isolated incident, Starr added. Despite the US Supreme Court’s numerous decisions stating that discrimination cannot be justified by statistics, there are plenty of examples where racism and sexism, supposedly backed by empirical evidence, go unchecked, she said.

“We have a pretty sharp incongruity between what I think the formal law is in our country about statistical discrimination, and the way some of these practices play out on the ground,” Starr said. “Why does that persist? I think in some ways, things that we would never have allowed to happen without those statistical tools are being allowed to happen.”

The idea of statistical discrimination came from a subfield of economics launched in the 1950s by the work of the late University of Chicago economist Gary Becker, which sought to explain the persistence of discrimination in markets. Becker’s original model focused on the effects of discriminatory “tastes” (pure prejudice against a group), while subsequent work in the field developed the alternative “statistical” theory: decision-makers use race or other group membership as a proxy for correlated characteristics that they don’t have individualized information about. Starr observed that this distinction between taste-based and statistical discrimination, despite its centrality to the economic literature, does not exist in the law, where both fall under the category of disparate treatment. At the same time, this hasn’t stopped parties accused of discrimination from citing statistics as part of their defense.

“We see it a lot in sex discrimination cases, where the defendant has openly discriminated based on sex,” Starr said. “And they try to justify it by pointing to some difference between men and women that they say that some studies support. In those cases though, the government offering those justifications loses again and again.”

Starr offered the 1974 case of Weinberger v. Wiesenfeld as an example. According to the Social Security Act, because married women were most likely financially dependent on their husbands, widowed men were not entitled to social security benefits. The Court held that, even if this assumption about financial dependence was true most of the time, it was an overbroad generalization that could not be tolerated under the Constitution.

“The Court articulates a crucial principle, which is that otherwise unlawful discrimination can’t be justified by resorting to statistical generalizations about groups,” Starr said. “Doing so would be unfair to the individuals within those groups who aren't well described by the generalization.”

Nonetheless, statistical discrimination perseveres in various legal proceedings, Starr said, offering an example from wrongful death lawsuits. Often in these lawsuits, forensic economists calculate the lost earnings of a deceased individual using actuarial tables that project their would-be earnings and life expectancy—and these projections are race- and gender-specific. As a result, the award of damages for the hypothetical death of a Black woman would end up lower than the award for the death of a White man in an otherwise identical case, Starr said.

“The contribution to accurate forecasts for an individual is actually very low, because there's a huge amount of individual variation within race and sex groups in lifetime earnings and in life expectancy,” Starr said. “And it may be that relying on the race- and gender-specific tables makes accuracy worse […] because it takes the disparities that affected past generations and extrapolates them forward into the future.”

Another example involves race-norming similar to the NFL’s but with even higher stakes. In 2002, the Supreme Court held in Atkins v. Virginia that defendants below a certain intelligence level could not be executed under the Constitution. Since then, at least eight states have used intelligence tests, normed by race, to determine whether defendants facing the death penalty meet that threshold.

Prosecution experts have justified the upward adjustment of scores for people of color by citing testing bias, Starr said. That seems extremely perverse, she added, given the many other instances in which life opportunities are determined by standardized tests that are not normed by race.

“I do not think that where a person's life is at stake, like in the death penalty context—in a context where, by the way, the Supreme Court has emphasized the importance of individualized determination of everything—that relying on that kind of upward adjustment is enough to justify an execution,” Starr said.

Starr pointed to a number of explanations for why this statistical discrimination has persisted, among them a tendency to defer to experts without digging too much into the quantitative research.

“I think that people without technical training don't necessarily try very hard to understand things that are quantitative,” Starr said “You hear the words, ‘regression model’ or ‘statistical adjustment’ and your eyes start to glaze over a little bit.”

Starr urged law students attending the lecture to interrogate the statistics cited in these types of cases.

“If you don't understand something that an expert says in a courtroom, for instance, ask somebody and keep asking questions until you do,” Starr said. “Believe that you are capable of this, because if you turn the other way and just always defer to what the normal practice is in some scientific field, you may miss injustices that are happening in front of your nose. And you also may miss ways to help your client.”

A full video of the Coase Lecture can be viewed above.