How do you measure the value of racial diversity? The most common approach over the past generation has been to try to make the business case for it. This directs our focus to classical economics and allocative efficiency, and subordinates both the moral case for equality of opportunity and the reality that efficiency arguments often ignore the distribution of wealth and income. A more robust assessment of racial diversity’s value should be able to, if only imperfectly, answer multiple philosophical approaches.
The business case for racial diversity has qualified support. Herring (2009) establishes a correlation between greater racial and gender diversity and larger business size, market share, and profitability; he suggests this at a minimum conflicts with the perspective of “diversity as process loss” due to added organizational conflict. Andrevski et al. (2015) and Richard et al. (2007) find evidence for cognitive diversity gains in competitive markets with higher growth potential. Nathan (2016) suggests small ethnic diversity benefits are driven by a minority of larger knowledge-intensive firms. Smulowitz et al. (2018) detect a positive relationship between law firm profits and racial diversity, but the most pronounced diversity occurs at junior employee levels.
Noon (2007) persuasively asserts that the business case for ethnic or racial diversity is flawed on moral grounds as it makes equality of opportunity subject to economic efficiency. Applied to the U.S., it subverts the Civil Rights Act of 1964 equal employment protections if no clear economic benefit for racial diversity can be ascertained. The law will not matter much if corporate management norms center around the business case and demote the concept of equal employment opportunity to a filing compliance process. From a social justice perspective, equal opportunity is an unconditional right. Therefore, any measure of racial diversity in a firm that ignores this right fails an important test.
The Global Reporting Initiative’s diversity and equal opportunity sustainability standard (GRI 405) attempts to address diversity of corporate governance and employment on the basis of age, gender, and minority or vulnerable group status. It calls on firms to disclose the percentage breakdown of employees or directors by various groupings (e.g., age, gender, race or ethnicity), and encourages distinctions between management and broader employee diversity as well as references to sector or regional benchmarks. In addition, it asks companies to disclose ratios of pay between women and men for each employee category. This information may provide a helpful starting point, but it is lacking in two important aspects.
First, apart from pay ratios for women and men, GRI 405 does not address the distribution of employee compensation. The percentage of non-White management employees may provide some indication of racial diversity, but it does not account for the wide dispersion of income levels inherent within broad employee categories — any degree of wage discrimination would go unnoticed.
Second, the comparison of the proportion of a firm’s employees that identify with a particular race and their representation in the general population implies that a firm’s people should look like its locale. But most companies specialize in one or a few industries that require very specific skills. Differences of educational attainment and areas of study between racial groups might explain most of the observable differences in employment percentages. For example, the Federal Reserve Bank of Cleveland finds that Asian students are roughly three times as likely to major in STEM subjects as Black students. Therefore, it would be unsurprising for firms to employ a greater percentage of Asian engineers, despite the fact that Asian-Americans account for a smaller percentage of the U.S. workforce.
Educational attainment differences remain a social injustice, but they are better addressed by public education than corporations. For business, the norm should be equal opportunity for a given level of education and experience. And the best reflection of equal opportunity is the outcome of equal compensation distributions by race or other demographic measure. If an employee has a college degree plus 15 years of experience and is in all other respects an average performer given those characteristics, their compensation should fall in the middle of the distribution of college-educated employees with 15 years of experience — whether I choose a distribution of all employees, a distribution of White employees, a distribution of Hispanic employees, etc.
Some may try to argue that pay differences between employees with equal skills don’t exist, making such analysis superfluous. They might posit that some skills can’t be measured except by the “wisdom” of the market, and therefore any pay differences must be justified. Yet skill measurement challenges should not be systematically different by race, age, or gender unless there is an element of discrimination. Pay deviations from what one would expect based on the résumé should cancel out when looking at large groups of employees. Bayer and Charles (2018) find, however, that Black college-educated males are paid systematically less than White college-educated males when examining the entire pay distribution. A Black college-educated male at the median income level for his race would only be at the 39th percentile of income among White college-educated men. A Black man at the 90th percentile for income within his racial group would be at the 86th percentile among White-male college graduates. Bayer and Charles’ study further suggests that these relative pay disparities have been persistent since the 1970s.
Building on the method of Bayer and Charles, I recommend that corporations examine the differences between their pay distributions by race for employee groups with similar education and experience. At each percentile, the average compensation across all races and deviation from this average should be calculated. If equality of opportunity were realized, the sum of deviations for a particular race should be close to zero when the entire distribution is considered, especially when the employee population is large. However, if systematic discrimation is present, we might expect even small differences at each percentile to cumulate to sizable disparities.
If cumulative differences in pay are present, the next task is to value them. A $5,000 difference of income between racial groups at the 25th percentile of their respective distributions is likely to be far more meaningful than the same dollar difference at the 99th percentile where annual compensation may be measured in millions. What we really care about is the disutility caused by systematic discrimation between groups based on race and other identities. Accordingly, we should apply some utility function to the income differences calculated above. For example, take a firm that divides its employees into three groups: White, Black and Other. Assume that the pay at the 25th percentile is $25,000 for White workers, $23,000 for Black workers, and $27,000 for workers identifying with other races. The average pay among the three groups at the 25th percentile is $25,000. The deviation from the average is zero for Whites, negative $2,000 for Blacks, and positive $2,000 for Other. Because any utility function we choose will be convex, it follows that the utility loss suffered by Black workers will be greater than the utility gained by those in the Other group. Therefore, any difference between groups at each percentile will result in a net loss of utility across all groups, at least for that percentile of the distributions.
To determine the net loss of utility for each group, we average the gains and losses of utility at each percentile. If no systematic discrimation is present, utility differences should average out to zero. The sum of the averages for each group yields the average utility loss per employee — again, this total can never result in a net gain given the convexity of utility. The next task is to redollarize this utility figure. To create comparability with corporate financials, this can be accomplished by determining the dollar value loss by shareholders that would result in the same level of utility sacrificed. We can arrive at a cumulative figure by multiplying this dollar value by the total number of employees.
Going back to our example, let’s assume the net utility loss is 0.008 at each employee percentile. Let’s further assume that the average income of shareholders weighted by the value of their shareholdings is $250,000, and that at this income level a loss of utility of 0.008 equates to about $250. If the firm employs 100,000 people, the disutility of discrimination would have a negative social value of $25 million per year from the perspective of a socially responsible investor. We would then apply a social discount rate (SDR) to this annual figure to estimate the present social value lost. Using a SDR of 2% (and assuming no real growth in the disparity of income by race) would translate to a social loss of approximately $1.2 billion.
One potential pushback against this type of analysis from corporations would be “competitive” concerns. Companies don’t want to disclose their compensation information for an obvious reason: they want to pay everybody as little as possible and don’t want employees negotiating their way up the pay scale within or across the industry. But this concern can be obviated with a standard that only asks them to disclose a single dollar figure for chosen employee categories using the method above along with detail on any assumptions made.
Perhaps the greater impediment to adoption is that this figure can only highlight an ill — it will only yield a social cost. In that respect, it is no different than carbon emissions. Hence, it may require a similar level of social muscle to make this type of disclosure a reality.