Jay S. Kaufman on The Race Variable
We live in a world of profound inequality, and a central responsibility of the social sciences, whether in government agencies or academic institutions, is to document, track, and explain these inequalities in the service of broadly held egalitarian ideals of fairness. Of the many axes of inequality in modern societies, racial and ethnic differences are among the most ubiquitously encountered and widely reviled. This motivates us to monitor disparities in health, education, criminal justice, and economic well-being, and to continually seek to understand and alleviate any gaps that arise from injustice.
The process of collecting data on different groups and making contrasts of outcomes to document disparities is caried out by thousands of professionals every day, and the results fill our newspapers and guide the formation of our public policies. Government agencies issue reports and university press offices publicize new findings, and we are all bombarded with the results. But do we really understand these numbers? And are the procedures used to generate the reported disparities giving us the information that we think we are getting?
The Race Variable is about this process of generating disparities statistics, where it can go wrong, when it leads us astray, and how we can do it better.
The Race Variable is about this process of generating disparities statistics, where it can go wrong, when it leads us astray, and how we can do it better. The book is about statistics, but fundamentally the problems we encounter are not mathematical in nature, they are about our notions of fairness. Confusion arises largely because the social and biomedical scientists generating these figures don’t publish the “raw” numbers; instead they offer a modeled version that is meant to highlight the aspects of the disparity that are supposedly most relevant or consequential, or for which extraneous imbalances have been removed. These adjustments can change the reported numbers dramatically, but the details are buried in the footnotes, and few lay readers pay attention to how the numbers are processed before being disseminated.
A recent example: National Affairs is a quarterly journal of the American Enterprise Institute and has an online blog called Findings. The blog post from December 15, 2025, highlights an article published the same month in American Journal of Preventive Medicine by Thomas McAdams and colleagues, in which they reported on racial/ethnic disparities in the use of sedation and/or physical restraint among the roughly 12 million emergency room visits in the United States during 2022. The blog cites the authors as finding that about 7 percent of emergency behavioral health encounters involved restraint or sedation, and that racial and ethnic minorities were more likely to be treated in this way. For example, the blog reports that Black patients had 30 percent higher odds of being jointly restrained and sedated than white patients.
The effect of an adjustment is to “excuse” that factor, so differences attributable to it are not conflated with the racial disparity.
But why this list of adjustments? Are there confounding factors like sex and age that are not on the list but should be? Are there factors that are on this list but should not be? What is the rule for selecting adjustment factors, and how do you know if you have too few or too many? The effect of an adjustment is to “excuse” that factor, so differences attributable to it are not conflated with the racial disparity. Should all such differences be excused? If urban and rural patients are sedated at different rates, is that entirely independent of issues of race and ethnicity? And what about alcohol and drug use? This was present for 19 percent of whites, but 33 percent of Native Americans. The adjustment asks what the racial disparity would look like if this substance abuse disparity were absent, but we live in a world where it is very much present. Which world should we care about?
These are the questions I explore in The Race Variable, along with many other issues in design and analysis that arise in these settings. A major premise of the book is that these adjustments are statistical, but their motivations are ethical, even if the ethical discussion is hidden from the reader—and often also from the analysts. Readers of the blog post or the press release are rarely even informed of the specific adjustments, let alone offered any explanation of how they were chosen. In this example, the real-world disparities were shrunk to a fraction of their original sizes, but in many instances the disparity can even be reversed, making one group look better off in the adjusted analysis even if it is worse off in the real world. If we really care about these disparities, we need to become more critical readers and question such adjustments: whether they are made when they should not be, or not made when they should be. As always, the devil is in the details.
Jay S. Kaufman is a professor in the Department of Epidemiology, Biostatistics, and Occupational Health at McGill University, and the author of The Race Variable: How Statistical Practices Reinforce Inequality.