by Raji Jayaraman
Racial disparities are present in all aspects of life. In the U.S. labor market black men are 28 per cent less likely to be employed than white men, and those that are employed earn 69 cents on a white man’s dollar. Blacks and hispanics are 50 percent more likely to experience some kind of force in their interactions with police. Blacks drivers are 40 percent more likely to be stopped than white drivers. The prevalence of, and mortality from, Covid-19 is disproportionately high among blacks.
Megan Thee Stallion’s opinion piece was one of last week’s most popular articles in the New York Times. In it, the hip-hop star notes that, “Maternal mortality rates for Black mothers are about three times higher than those for White mothers, an obvious sign of racial bias in health care.” What is obvious to me is that this disparity is unacceptable. What is less evident is that it is “an obvious sign of racial bias”. Is race per se to blame for racial disparities? Maybe, maybe not. It’s possible, perhaps even likely, that healthcare workers or employers harbour racial prejudice against blacks. But it’s not obvious and yet, this type of claim—that racial bias can be inferred from the presence of racial disparities—is commonplace.
Economists have an arguably simplistic, but nevertheless useful way to think about the question of racial disparity. They break it down into two categories: racial bias and statistical discrimination. Racial bias refers to plain vanilla “racism”—I treat a black person differently than I do a white person because I harbour racial prejudice against blacks. Statistical discrimination is at play when I treat blacks and whites differently because, quite apart from race, they have substantively different underlying characteristics.
Consider racial disparities in employment. Prior to the pandemic, the unemployment rate among black workers was twice as large as that of white workers. One explanation for this is preference-based. For example, an employer may choose a white hire over a black hire because she prefers whites to blacks. This is racial bias. Another possibility is that the employer just wants the best person for the job. Unfortunately, she doesn’t get to observe a person’s talents or skills. What she does know is there exists a large educational achievement gap between blacks and whites, so she makes a valid statistical inference that a white applicant is more likely than a black applicant to have the requisite skills for the job. So, even if she harbours no personal prejudice against blacks, she still deliberately chooses a white hire over a black hire. This is statistical discrimination.
Two points about statistical discrimination are worth noting. First, race-based statistical discrimination only exists because we don’t get to observe the truth. An employer can never perfectly observe an applicant’s job aptitude, so she makes an inference based on what she can see. If unobserved truths were observed—if ability or job match quality were stamped on an applicant’s forehead—statistical inference on the basis of skin colour would be unnecessary. The second point to note is that statistical discrimination may logical, but it’s not fair. Just ask the man who couldn’t afford to attend university; would have been perfect for the job; but just so happens to be black.
Empirically, the question of whether racial disparities are driven by racial bias or statistical discrimination is tricky to answer for the simple reason that we don’t get to see the counterfactual: would a black job applicant have been hired, had he been white? Same interviewer, same job, same person,…different race. We can’t know that because the applicant is black. Still, there are ways to try to tease these two things apart and the evidence is mixed, with some studies finding that once one accounts for observable differences between blacks and whites there doesn’t appear to be racial bias, and others finding that these other differences account for part but not all of the racial disparities in labor market or law enforcement outcomes.
This distinction may seem like splitting hairs because at the end of the day, racial disparities remain and should fill us all with moral repugnance. Still, as Ken Arrow pointed out, “moral feelings without analysis can easily lead to unconstructive policies,” and this distinction matters for policy. If racial disparities are driven by racial bias, then reducing these disparities will involve either changing racists’ preferences or getting them to act against their underlying (racist) preferences. If, on the other hand, disparities are driven by statistical discrimination, then reducing racial disparities require some combination of information and opportunity provision.
Consider again the case of racial disparities in employment. If racial bias accounted for this, then potential remedies may be to ensure that there are no racists in the employer’s HR department; or to train them to overcome racial prejudice. If, instead, statistical discrimination is to blame, then the employer may want to gather more information on applicants in order to reduce their reliance on race-based statistical inference; or governments may need to do more to bridge the educational achievement gap between blacks and whites.
I am no psychoanalyst, but casual observation of my fellow human beings over the course of several decades has left me skeptical that we have any credible means to identify racists; “convert” them; or get them to behave properly in any consistent manner. Recent evidence that things like implicit bias training and diversity training are largely ineffectual lend support to my skepticism. The upshot is that studies, which find that racial disparities can be attributed to racial bias leave me feeling incredibly depressed. Human nature in adulthood just isn’t malleable enough to fundamentally change people’s preferences or behaviours.
Evidence that racial disparities can be attributed to statistical discrimination gives me no joy. But it does give me a modicum of hope that something can be done to reduce disparities. We could reduce the need for statistical inference based on race by gathering better information. We could institute policies that level the playing field and provide equality of opportunity. These policy measures have their own problems and some may prove intractable, but addressing the problems underlying statistical discrimination strikes me as a less futile endeavour than “curing racism”.
One of my main objections to left-leaning friends who cry, “Racist!” when confronted with racial discrimination is that their explanation trivializes what is in fact a much deeper problem, that warrants a complex set of solutions that can only be informed through a deeper understanding of underlying sources of statistical discrimination. Solutions that we have some ability, and hopefully the will, to implement.