An open access publication of the American Academy of Arts & Sciences
Winter 2005

Why race still matters

Ian Hacking

Ian Hacking, a Fellow of the American Academy since 1991, holds the chair of Philosophy and History of Scienti½c Concepts at the Collège de France. His work spans the philosophy of science, the philosophy of language, the theory of probability and statistical inference, and the socio-historical examination of the rise and fall of disciplines and theories. His most recent books are “Mad Travellers” (1998), “The Social Construction of What” (1999), and “Historical Ontology” (2004).

Why has race mattered in so many times and places? Why does it still matter? Put more precisely, why has there been such a pervasive tendency to apply the category of race and to regard people of different races as essentially different kinds of people? Call this the ‘first question.’ Of course there are many more questions that one must also ask: Why has racial oppression been so ubiquitous? Why racial exploitation? Why racial slavery? Perhaps we tend to think of races as essentially different just because we want to excuse or to justify the domination of one race by another.

I shall proceed with the first question by canvassing five possible answers to it that variously invoke nature, genealogy (in the sense of Michel Foucault), cognitive science, empire, and pollution rules.

One final preliminary remark is in order. Most parts of this essay could have been written last year or next year, but the discussion of naturalism, medicine, and race could only have been written in November of 2004, and may well be out of date by the time this piece is printed.

Why has the category of race been so pervasive? One answer says that the distinction is just there, in the world for all to see. Superficial differences between races do exist in nature, and these are readily recognized.

The naturalist agrees at once that the distinctions are less in the nature of things than they once were, thanks to interbreeding among people whose ancestors have come from geographically distinct blocks. Racial distinctions are particularly blurred where one population has been translated by force to live in the midst of another population and yet has not been assimilated–slaves taken from West Africa and planted in the Southern United States, for example. The naturalist notes that traditional racial distinctions are less and less viable the more children are born to parents whose geographical origins are very different.

Sensible naturalists stop there. The belief that racial differences are anything more than superficial is a repugnant error. John Stuart Mill was the wisest spokesman for this position.

Here, in modern terminology, is his doctrine: (1) Nature makes differences between individuals. These differences are real, not constructed. (2) We classify things according to differences we observe. Classifications are made by people and encoded in social practices, institutions, and language. (3) Some classes are such that their members have little in common except the marks by which we sort them into those classes–call those superficial kinds. (4) Other classes have members with a great many things in common that do not follow from the marks by which we sort them into classes. These are “real Kinds.”1

Examples? “White things,” he wrote, referring not to race but to the color itself, “are not distinguished by any common properties except whiteness; or if they are, it is only by such as are in some way dependent on, or connected with, whiteness.” But horses, to use one of his other examples, have endless properties in common, over and above whatever marks we use to distinguish them from other animals or other kinds of things. Horses form a real Kind, but the class of white things is a superficial kind.

The contemporary philosophical concept of a ‘natural kind’ is a descendent of Mill’s notion. Nonphilosophers who have come across this phrase may suppose it refers to a well worked out, technical, and stable concept. I argue elsewhere that it does not.2

Mill himself was as notable a profeminist and antiracist as can be claimed for a white nineteenth-century man. Although he argued that real Kinds exist, he at once went on to ask whether the races and sexes are real Kinds, or if they are merely superficial, like the classifications “Christian, Jew, Musselman, and Pagan.” The religious confessions are not real Kinds, he argued, because there is no property that Christians have and Muslims lack, or vice versa, except whatever follows from their faiths.

What about race? Most anthropologists of Mill’s day held that there were five races, named geographically but recognized by color: Caucasian, Ethiopian, Mongolian, American, and Malayan. According to Mill, color and certain other physiological traits are the marks by which we distinguish members of the different races. Races would be real Kinds if there were endlessly many other differences between the races that did not follow from the marks by which we distinguish them. Are there endlessly many such differences?

Well, you cannot rule that out a priori, Mill thought. “The various races and temperaments, the two sexes, and even the various ages, may be differences of Kind, within our meaning of the term. I say they may be; I do not say they are.” Mill believed that only empirical science could determine whether the various races, as distinguished by color and a few other features, pick out classes that are distinct in a great many unrelated ways. “If their differences can all be traced to climate and habits [or, he added in later editions, to some one or a few special differences in structure], they are not, in the logician’s view, specifically distinct.” He would have been pleased by Anthony Appiah’s careful discussion of very much the same question using more recent terminology. Science might have revealed an endless number of differences between the races that are not consequences of the marks by which we distinguish them, namely color and physiognomy. But science has not done so, and almost certainly will not. Mill, like Appiah, thus concludes that the races are not real Kinds.

This conclusion, however, does not answer, or aim at answering, the specific question I raised at the outset, of why there is such a pervasive tendency to apply the category of race. Maybe Mill thought the answer was obvious. The desire of one racial group to dominate, exploit, or enslave another demands legitimacy in societies that, like modern Europe and America, are committed to versions of egalitarianism. Race sciences were devised to discover a lot of differences between races that do not follow from the marks of color and structure by which we distinguish them. You do not have to treat people equally, if they are sufficiently different.

Although it takes us some distance from the ‘first question,’ some recent events force us to clarify the naturalist position on race. In an important editorial on the U.S. census published in the year 2000, Nature Genetics stated: “That race in this context is not a scientific term is generally acknowledged by scientists–and a message that cannot be repeated enough.” An editorial in 2001 observed that “scientists have long been saying that at the genetic level there is more variation between two individuals in the same population than between populations, and that there is no biological basis for ‘race.’”3 Now–in November of 2004–this selfsame journal has produced a special supplement on the medical and genetic uses of racial and ethnic classification. And the November 11 issue of The New England Journal of Medicine highlights the news of the ‘race-based’ drug targeted at African Americans suffering from certain types of heart failure. All this is breaking news. Hence what follows cannot be definitive, but one may hope that a perspective somewhat distanced from media discussion can be useful even in the midst of it.

We must first update Mill with a little logic. When he wrote about differences between classes, he had in mind properties that serve to distinguish members of one class from another in a uniform way. A uniform difference between cows and horses is something that is true in the main of any cow but not true in the main of any horse–digestion by rumination, for example. There are ever so many such differences between horses and cows; hence they are real Kinds. Call them uniform differences. There are a great many uniform differences that distinguish horses from other kinds of animals, but almost no uniform differences that distinguish white things from green things, except their color, or Muslims from Christians, except their faith.

Writing in 1843, Mill had little occasion to think about statistical differences, which were only just beginning to loom large on the scientific horizon. We need some new concepts: I will use the words ‘significant,’ ‘meaningful,’ and ‘useful.’ All three go with the dread word ‘statistical.’ Since we are among other things talking about so-called races, namely, geographically and historically identified groups of people, we are talking about populations. And we are talking about some characteristic or property of some but not all members of a population.

‘Significance’ was preempted by statistics early in the twentieth century. It is completely entrenched there. Here I use it for any major difference detected by a well-understood statistical analysis. A characteristic is statistically significant if its distribution in one population is significantly different from that in a comparable population. Let us say that a characteristic is statistically meaningful if there is some understanding, in terms of causes, of why the difference is significant. For example, in the early days no one knew why smoking was associated with lung cancer, but now we understand that quite well, although not completely. The correlation used to be merely significant, but now it is meaningful.

Finally, a characteristic is statistically useful if it can be used as an indicator of something of interest in some fairly immediate practical concern. Take an example from another topic nowadays much discussed. A body mass index (BMI) over 31 is a statistically useful indicator of the risk of type 2 diabetes, and is therefore useful in epidemiology and preventive medicine. (There are much better indicators involving the distribution of mass and muscle in the body, but at present such indicators are expensive to measure, while BMI measurement costs almost nothing.)

Classes that are statistically significant, meaningful, or useful are not thereby real Kinds. There is no reason to believe that there are a great many independent and uniform differences that distinguish obese persons from those whose BMI is in the recommended range of 18 to 25.

‘Significant’ in the end relies on technical notions in applied probability theory. ‘Meaningful’ has no resort to viable technical notions in any discipline (all claims to the contrary are spurious). There do exist clear, although often abused, criteria of statistical significance. There are no clear criteria for being statistically meaningful. In practice the distinction is often easily made. For a long time, the class of people who smoke was known only to be statistically significant with respect to lung cancer. One had no idea of the causal mechanisms underlying the correlation. Now we think we understand the connections between nicotine and death, although these connections are still merely probable. We cannot say of a young man beginning to smoke that if he continues with his vice he will succumb to lung cancer if nothing else gets him first. But we can say that many such young men will die of lung cancer, and oncologists know enough to be able to explain why.

Unlike statistical significance, the idea of being statistically meaningful is a hand-waving concept that points at the idea of an explanation or a cause. Imprecise hand-waving concepts are dangerous when they are given fancy names. They can be put to wholly evil ends. But if we do not give them phony names and are well aware of their imperfections, they can be useful when we need them.

We do need this concept. Many people–as evidenced by debates going on at the time of this writing, in November of 2004–are scared of the idea that the traditional list of races employed by traditional racists might be statistically significant classes. With good reason!

Ten years ago The Bell Curve by Richard Herrnstein and Charles Murray attracted a great deal of attention. The authors claimed that the Gaussian distributions of IQ scores establish a natural distinction of some importance between different races. They forcefully argued that the class of African Americans is a statistically significant class–significant with respect to a property they called intelligence, and which they measured with IQ tests. They did not imply that the races are real Kinds. That is, they did not state that there is a host of uniform differences between Caucasian Americans and African Americans. Readers not unreasonably assumed, however, that the authors meant exactly that. At any rate, the authors clearly were not talking about mere correlations, namely, disparities between IQ scores within different racial groups. But they did not establish that these disparities are statistically meaningful to any biological understanding.

About the same time that The Bell Curve was published, ogre naturalists, such as Philippe Rushton in Race, Evolution, and Behavior, made more sweeping claims to biologically grounded racial differences. They claimed that the races are distinguished by many properties rightly prized or feared for different strengths and weaknesses. If that were true, then races would exactly fit Mill’s definition of a real Kind.

One deplores both Rushton and The Bell Curve, but there is an absolutely fundamental logical difference between what the two assert. Rushton claimed that the races are real Kinds. One imagines that Herrnstein and Murray thought so too, but what they claimed was that the races are statistically significant classes. And they implied that this is statistically meaningful. Despite the fact that his doctrines have a centuries-old pedigree, we can dismiss the egregious Rushton. We can also refute Murray and Herrnstein.4 Mill’s type of naturalism has contempt for both doctrines. Loathing of these quite recent doctrines and their predecessors has, not surprisingly, produced revulsion against any sort of naturalism about race. Today there is some consternation over the appearance of what is called race-based medicine.

The science of medicine was for quite a long time the science of the European male body, with footnotes for non-European or female bodies. All that has changed: those footnotes are now chapters. But the current situations for the groups that had been relegated to the footnotes are quite different. Many medical differences between males and females are uniform, but medical differences between races are almost always only statistical.

We have long known that some ailments are restricted to some gene pools. Tay-Sachs is a hereditary disease (in which an enzyme deficiency leads to the accumulation of certain harmful residues in the brain and nerve tissue, often resulting in mental retardation, convulsions, blindness, and, ultimately, death) that almost exclusively affects young children of eastern European Jewish descent. ‘Ashkenazi’ is a valuable geographical, historical, and social classification. It is geographical because it indicates where members of this class, or their near ancestors, came from, namely, eastern Europe. It makes a contrast with Sephardic Jews, whose roots are in Spain. In modern Europe and North America, social differences between the Ashkenazi and Sephardic hardly matter to most people, but they remain significant in North Africa and West Asia. Until further interbreeding makes it totally obsolete, Ashkenazi is a statistically significant and a statistically meaningful class with respect to Tay-Sachs disease.

There are similar geographical-historical indicators for lactose intolerance and for an inability to digest fava beans. West African ancestry is an indicator for being a carrier of the sickle-cell anemia trait, which confers some immunity against malaria. This trait was often stigmatized as simply ‘black.’ In fact, it is primarily West African, although it shows up in Mediterranean populations where malaria was a major selector for survival. The indicator was abused for racial reasons in widespread screening.

“Drug approved for Heart Failure in African Americans”–headline on the first business page of The New York Times, July 20, 2004. Here we go again? Quite possibly. “The peculiar history [of this drug] on the road to the market presents a wide array of troubling and important issues concerning the future status of race as a category for constructing and understanding health disparities in American society.”5 For a stark reminder of the commerce, the Times reported that the previous day the stock of the drug’s maker, NitroMed, rose from $4.31 to $10.21, and had reached $16 at midday. This story has been ongoing for a decade in medical, commercial, and regulatory circles.

There are real problems about the racially targeted heart drug. BiDil is a mixture of two well-known heart medications. Scientific papers assert, first, that other medicines are not as good for African Americans with heart failure as they are for other Americans with this problem, and, second, that BiDil works better for African Americans with certain specifics than any other drug on the market.6 In fact, randomized trials were discontinued because the drug was manifestly effective on black patients. Nobody well understands why. The reasons could be at least in part social and economic (including dietary) rather than hereditary. The correlation is strongly significant, but it is not statistically meaningful at present from a genetic or other biological point of view.

Even if one is a complete skeptic about, for example, a genetic basis for the differential efficacy of the drug, the drug does appear to be statistically useful in treating the designated class of patients. That means that race may be a useful indicator to a physician of the potential effectiveness of this rather than another drug–under present social and historical conditions.

Now turn to leukemia. Bone marrow transplants help an important class of patients. Donors and recipients must have matching human leukocyte antigens (HLAs); at present, doctors try to match six different types of them. If a patient has no relative to serve as a donor, matches are hard to come by. The relevant antigens are unevenly distributed among ethnic and racial groups.7 There exist registries of possible donors –truly generous persons, for at present donation of bone marrow is quite harrowing. Happily, free-floating stem cells in the blood also help, but the donor must take a lot of drugs to boost those stem cells. Another source of cells is umbilical cord blood. But this, like all the other options, requires antigen matching.

In the United States, the National Bone Marrow Program maintains the master registry. Most people in existing registries have tended to be middle-aged and white, which means that whites have a good chance of finding a match. Hence there have been racially targeted programs for Asian and African Americans. In the United States and Canada there is also the Aboriginal Bone Marrow Registries Association, and in the United Kingdom there is the African Caribbean Leukemia Trust. Asians for Miracle Marrow Matches has been very successful, especially in the Los Angeles region. The African Americans Uniting for Life campaign has been less successful, for all sorts of historical reasons. An African American with leukemia has a far worse chance of finding a match in time than members of other populations have. That is a social fact, but there is also a biological fact: there is far greater heterogeneity in the human leukemia antigen in persons of African origins than in other populations.8 (This fact fits well with the hypothesis that all races are descendants of only one of many African populations that existed at the time that human emigration began out of Africa–populations whose characteristics have continued to be distributed among Africans today.)

If you go to the websites for the organizations that maintain the registries, you will see they do not shilly-shally in some dance of euphemistic political correctness about race. For them it is a matter of life and death. Without the Asian registries there would have been many more dead Asian Americans in the past decade. For lack of more African Americans on the registries there will be more dead African Americans in the next few years than there need be.

We certainly lack a complete understanding of the distribution of human leukemia antigens in different geographically identified populations. But we do have some biological understanding of the underlying causal differences. And race is a very useful quick indicator of where to look for matches, just as the BMI is a useful quick indicator of potential health problems.

So when, if ever, is it useful to speak in terms of the category of race, on the grounds that the races in some contexts are not only statistically significant but also statistically useful classes? To answer this question, we can use our distinctions:

  • The Bell Curve may show that IQ is a statistically significant characteristic of some American subpopulations, but it is neither meaningful from a biological point of view nor useful for any well-defined purpose.
  • Some medications may be less effective, and BiDil may be more effective, for African Americans with certain types of heart failure. If so, this is statistically significant and statistically useful for helping patients, but (in my opinion) it is at present not statistically meaningful.
  • The relationships between human leukemia antigens and race are statistically significant, statistically meaningful for a biological understanding, and statistically useful in making marrow matches possible for minority groups.

It is not a good idea, in my opinion, to speak of BiDil as a race-based medicine, as do The New York Times and other media. The drug is not in the least based on race. It is quite possible that the reason it is more useful for African Americans than for other large and loosely characterized groups has less to do with the inherent constitution of their cardiovascular systems than with a mixture of social factors. If we had reliable data on the relevance of diets shared by a subclass of white and black Americans, we might be able to help whites with similar diets. The drug would not then be ‘diet-based’ but ‘diet-targeted.’ If you find it useful to use the word ‘race,’ say ‘race-targeted’ medicine.

I should have thought that the differential distribution of human leukocyte antigens would be esoteric enough to escape notice. Not so. The Stormfront White Nationalist Community, whose best-known figure is the neo-Nazi David Duke, is having a good time on one branch of its website discussing HLA diversity. In my opinion, the correct strategy is not to play down the differential distribution of HLA, but to make it common knowledge that specific differences among peoples may be used in helping them–in much the same way that white Australians, given their socially induced tendency to overexpose themselves to the sun, should be targeted to cut down on the rate of death due to skin cancer.

I have introduced these remarks to make plain that naturalism about race, far from being an atavistic throwback to an era well left behind, is a topic for today, one about which we have to become clearer. Not because the races are real Kinds, denoting essentially different kinds of people. But because already we know that the races are not only statistically significant classes for some diseases, but also statistically useful. Some correlations are statistically meaningful. There is every reason to believe that more statistically meaningful correlations will be discovered.

Every time such a phenomenon is found useful, the racists will try to exploit the racial difference: witness the neo-Nazi use of differential antigens. Hence we need to be fully aware of what is involved.

A historian may well despise the complacency of naturalism. Differences between the races have seemed inevitable in the West, it will be argued, because of a framework of thought whose origins can be unmasked only by a genealogy. Classification and judgment are seldom separable. Racial classification is evaluation. Strong ascriptions of comparative merit were built into European racial classification and into evaluations of human beauty from the beginning. And so the Caucasian face and form were deemed closest to perfect beauty.

That is the vein in which Cornel West has sketched a genealogy of modern racism.9 Though his is not exactly a deep genealogy in the spirit of Nietzsche and Foucault, it is an excellent résumé of events. I wish only to comment on his starting point, less to correct it than to encourage rethinking the connection between race and geography.

.  .  .


  • 1His own words are old-fashioned but lovely. The differences between members of classes “are made by nature . . . while the recognition of those differences as grounds for classification and of naming is . . . the act of man.” However, “we find a very remarkable diversity . . . between some classes and others.” Only superficial resemblances link members of one type of class, while members of classes of the other type have a vast number (he said an endless number) of properties they share. Those that share an almost endless number of properties are his real Kinds. From John Stuart Mill, A System of Logic: Ratiocinative and Inductive, first published in 1843. The discussion of racial classification is found in bk. 1, chap. 7, sec. 4. The changes Mill made in later editions of the book involved sex, not race–doubtless because Mill hoped to get the questions about sex exactly right for Harriett Taylor. See chap. 7, on Millon classification, in my forthcoming book, The Tradition of Natural Kinds (Cambridge University Press).
  • 2This is one of the conclusions urged in my book The Tradition of Natural Kinds.
  • 3“Census, Race and Science,” Nature Genetics 24 (2000): 97; “Genes, Drugs and Race,” Nature Genetics 29 (2001): 239.
  • 4There is a tendency among proper-thinking people to dismiss The Bell Curve cavalierly, as both wrong-headed and refuted, without actually saying why. Many things wrong, and one has an obligation to say what. My own ‘genealogical’ objections are stated in a piece in The London Review of Books, January 26, 1995.
  • 5Frederick Kahn, “How a Drug Becomes ‘Ethnic’: Law, Commerce, and the Production of Racial Categories in Medicine,” Yale Journal of Health Policy, Law, and Ethics 4 (2004): 46.
  • 6Anne L. Taylor, “Combination of Isosorbide Dinitrate and Hydralazine in Blacks with Heart Failure,” New England Journal of Medicine 351 (2004): 2049–2057.
  • 7This also matters to renal transplants. See Pauline C. Creemers and Delawir Kahn, “A Unique African HLA Haplotype May Identify a Population at Increased Risk for Kidney Graft Rejection,” Transplantation 65 (1998): 285–288.
  • 8For HLA differentiation, see T. D. Lee, A. Lee, and W. X. Shi, “HLA-A, -B, -D and -DQ Antigens in Black North Americans,” Tissue Antigens (1991): 79–83. For maps, see, for example, one of the essays in the November Nature Genetics issue referenced in the text: Sarah A. Tishkoff and Kenneth K. Kidd, “Implications of Biogeography of Human Populations for ‘Race’ and Medicine,” Nature Genetics Supplement 36 (2004): 521–527.
  • 9Cornel West, “A Genealogy of Modern Racism,” in West, Prophesy Deliverance!: An AfroAmerican Revolutionary Christianity (Philadelphia: The Westminster Press, 1982), 47–65.
To read this essay or subscribe to Dædalus, visit the Dædalus access page
Access now