In 1989, 25 years after the passage of the Civil Rights Act of 1964, Supreme Court Justice Harry Blackmun asked “whether the majority [of the court] still believes that race discrimination – or, more accurately, race discrimination against nonwhites – is a problem in our society, or even remembers that it ever was.” Prompting Justice Blackmun’s question was the decided shift in the late 1980s in how the court’s majority treated evidence of bias, especially numerical/statistical evidence. Another 25 years have now passed and the question remains apt.
When President Lyndon Johnson signed the Civil Rights Act on July 2, 1964, and a year later the Voting Rights Act, our society was transformed and a period of dramatic movement against an array of discriminatory practices began. The workplace, in particular, was remade by Title VII of the Civil Rights Act, which banned employment discrimination based on “race, color, national origin, sex, and religion,” while the Voting Rights Act significantly altered the makeup of elected bodies across the nation.
Blatant acts of job discrimination (“colored need not apply”) were plainly illegal, but establishing more subtle and indirect forms of bias required statistical evidence. A unanimous court recognized this implicitly in 1971 in Griggs v. Duke Power when it ruled that the Civil Rights Act barred not only “overt discrimination, but also practices that are fair in form, but discriminatory in operation.”
How to determine whether a practice is “discriminatory in operation?” Look at the numbers; the data on hiring, wages and promotions of employees carry the relevant information. Similarly, in 1986, the court held that the Voting Rights Act banned voting practices that had the effect of preventing minorities from electing candidates they preferred. The way to show that was to collect and analyze election data.
Since 1987, the Supreme Court has issued a series of closely decided rulings on capital punishment, employment discrimination, and voting rights that jeopardize these legislative milestones. This is not news. What has been less noticed is the marked change in the court’s treatment of numerical/statistical evidence. The result: a chilling effect on the advancement and protection of civil rights in elections and employment.
In October 1986, the court’s composition changed, with William Rehnquist replacing Warren Burger as chief justice, and Antonin Scalia filling the vacancy left by Burger’s departure. This court took an entirely different view of “the numbers.” In McCleskey v. Kemp , a death penalty case, the “Baldus study,” an extraordinarily detailed statistical analysis, showed a high risk of racial prejudice in capital sentencing in Georgia. Defendants charged with killing white victims were, by the study’s most conservative measure, 4.3 times as likely to be sentenced to death as defendants charged with killing blacks. The court accepted this number, along with many others, but said they only established a risk that racial prejudice played a role in death penalty cases. And, in a 5-4 decision, that risk was not unconstitutional, because “[t]he statistics do not prove that race enters into any capital sentencing decisions or that race was a factor in petitioner’s case.” The dissent by Justice William Brennan differed: “Close analysis of the Baldus study, however, in light of both statistical principles and human experience, reveals that the risk that race influenced McCleskey’s sentence is intolerable by any imaginable standard.” What degree of risk would be unacceptable went unmentioned in the majority’s opinion.
The majority’s growing hostility to statistics over the next few years took clearer shape in a 1988 employment discrimination case cautioning against an “inevitable focus on statistics” and assumptions that “statistical evidence is reliable.” A 1989 decision that effectively overruled Griggs (and prompted Justice Blackmun’s question cited above) led Congress in 1991 to amend the Civil Rights Act to explicitly incorporate the Griggs standards and restore “disparate impact” (discriminatory in operation) as a means for identifying unfair employment practices.
In Wal-Mart v. Dukes (2011), the court found a new way to undercut the impact of statistical evidence. In another 5-4 decision, it held that a class of 1.5 million female employees should not have been certified despite statistical evidence finding large promotion and pay disparities unfavorable to women. The data showed, for example, that women were promoted to key positions at a much lower rate than men (women’s rate of promotion was less than 60 percent of the men’s rate for some positions) and salaried women employees were paid about 10 percent less than men. Such disparities were, moreover, remarkably uniform over the 41 regions maintained by Wal-Mart.
The court’s view: “Merely showing that Wal-Mart’s policy of discretion has produced an overall sex-based disparity does not suffice [to permit certification of the class].” The opinion emphasized the varied impacts of discretionary decisions by individual managers on individual women, ignoring the policy’s effect on the class of women. The implication is that the members of the class cannot be treated together whatever the data show. It may be too early to judge the ultimate significance of Wal-Mart but this much is clear: The decision has made it much harder to bring a broad-based job discrimination class action lawsuit.
More recently, the decision in Shelby County v. Holder (2013) gutted a key element of the Voting Rights Act and may spell the end of the process requiring some states (mostly Southern) to obtain prior federal approval of redistricting or other electoral changes. The requirement was intended to protect minorities’ electoral rights, that is, to inoculate voting practices against the discrimination virus. In 2006 Congress overwhelmingly approved, for the fourth time, reauthorization of the Act after hearings produced a “voluminous and extensive legislative record” of data showing persistent efforts by state and local governments to harm a minority’s electoral chances. President George W. Bush then signed it into law.
The court, in yet another 5-4 decision, chose to ignore the data relied upon by Congress, including studies showing Justice Department objections to more than 600 proposed election law changes over 20 years. Instead, the court invoked its own findings that times had changed and the data used in determining which jurisdictions were covered were out-of-date. The reasoning: because of the Voting Rights Act, “[v]oter turnout and registration rates” in covered jurisdictions “now approach parity,” so the process requiring them to obtain prior federal approval was unnecessary. Inoculation had worked, so it was no longer needed. Fortunately, public health officials do not follow such logic.
The court has stated that statistics pointing to effects on minorities or women cannot prove discrimination. By its nature, statistical evidence cannot remove all uncertainty, but in such arenas as medicine, business, the environment and climate change we live with uncertainty and put statistics aside at some peril. Recent Supreme Court decisions notwithstanding, civil rights also count and the numbers must matter.