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

Racial classification in America

Kenneth Prewitt
View PDF

KENNETH PREWITT is the Carnegie Professor of Public Affairs in the School of International and Public Affairs at Columbia University. A Fellow of the American Academy since 1979 and past Academy vice president, Prewitt was director of the U.S. Census Bureau from 1998 to 2000. He recently published Politics and Science in Census Taking (2003) and is preparing a book-length treatment of issues discussed in this essay. Among the other books he has authored or coauthored are Institutional Racism in America (1969), The Recruitment of Political Leaders (1970), and a textbook on American government.

In its first national census, the young American republic not only counted its population; it racially classified it.1 From 1790 to 1990, the nation’s demographic base changed from one decennial census to the next, and so too did the racial categories on offer. Always, however, the government held fast to two premises: First, it makes policy sense to put every American into one and only one of a limited number of discrete race groups, with the decennial census being the primary vehicle by which the counting and classifying should take place. Second, when policy treats Americans differently depending on what race they belong to, it should make use of this government classification.

The second premise depends on the first. Without a limited number of bounded groups, it is difficult to fashion policy with race as a criterion. This is easily seen in comparison. Since 1790 there have been policies based on age – who can vote, own property, be drafted, buy alcohol, and claim social security. These policies use a small number of age groupings with fixed and knowable boundaries. Though policy can draw the age boundaries differently as conditions change (eligible to vote at eighteen rather than twenty-one) there is no dispute about who is in a given age group. Using race as a criterion to define groups was never this straightforward, a fact implicitly acknowledged by the government as its census added and subtracted categories from one decennial to the next and as different federal agencies used different taxonomies.

Not until 1977 did the government bring order to the country’s racial categories. Acting under the influence of civil rights legislation, the Office of Management and Budget (OMB) directed all federal agencies to follow uniform standards in collecting racial data.2 This achievement was impressive but short-lived. Changing political considerations led to major revisions only two decades later, when the logic of identity politics, with its stress on diversity, began to destabilize the older and more deeply entrenched American division between white and nonwhite.

What do these developments mean for racial and ethnic divisions in America, both today and in the future?

In the context of census 2000, I witnessed the demographic changes and the associated political pressures that make it difficult to define and refine categories focused solely on redressing past injustices rooted in race – the policy purpose that emerged after the Civil Rights Act of 1964. In response to newer political pressures, the 2000 U.S. census was the first to permit respondents to record multiple racial origins. The 1997 revision of the OMB standards for racial classification allowed for “mark[ing] one or more” of the primary racial categories, leading to a census with sixty-three possible racial responses.3

In substantial ways the “mark one or more” option was an improvement over previous census formats, especially in forcefully rejecting the hypodescent presumption.4 At issue in this essay is whether, this improvement notwithstanding, the country has the statistical tools it needs to detect – and enable the government to redress – discrimination.

So where should we go from here? To address that question, it will be useful to recall how the United States ended up with such a complicated set of racial and ethnic categories in the first place.

The public face of America’s official racial classification is its census, and has been so since the first decennial enumeration in 1790. The initial classification was implicit in two civil status distinctions: free or slave, taxed or untaxed. Applying these distinctions in the census generated a count of three ancestry groups (European, African, and [untaxed] Native American), which set the foundation for all racial classifications to come. From that starting point, the division of the population by race has been repeated in every decennial census, down to the most recent in 2000.

Across two centuries, particular categories have come and gone in response to an ever-shifting mix of political, scientific, and demographic considerations. In 1820, the category “free colored persons” was added to the census. In 1850, influenced by a pseudo race-science, the census separately counted mulattoes, a category it retained until 1930. In 1870 Chinese were first counted, and in 1890, Japanese. In 1920 Filipinos, Koreans, and Hindus appeared on the census form. Following Hawaii’s statehood, in 1960 Hawaiians were added, though Alaskan statehood did not result in an effort to specifically identify Aleuts and Eskimos for another twenty years. Subcontinent Indians were counted as Hindu in three censuses (1920–1940), but as white in the next three censuses. In 1980 they were counted as Asian, a status they retain today. Until 1930 when they got their own census category, Mexicans were counted as white. The government of Mexico contested that change, and Mexicans went back to being counted as white until 1970, when Hispanic origin became a separate category – this time defined in terms of language and ethnicity rather than race.5

In the OMB standards first issued in 1977, there were four primary racial groups: Asian or Pacific Islander, American Indian or Alaskan Native, Black, and White. These standards held that all federal statistics on race should, at minimum, include those four groups as well as one ethnic group, Hispanic, whose members would also belong to one of the four racial groups.6

What political and policy purposes lie behind this continual shifting of the race categories?

In 1790, slaves were included in the census count (the three-fifths clause) because slaveholding states had made this a nonnegotiable condition for joining the Union. The result was a power bonus for Southern states in the new Congress and in the Electoral College. This bonus, as John Quincy Adams put it, led to “the triumph of the South over the North – of the slave representation over the purely free.”7 The nation’s first decision about how to classify the population racially had immense policy consequences that lasted well into the twentieth century.

Without discarding the three-fifths clause, a new era of racial classification began in 1820 when the “free colored” were counted separately from slaves and free whites. This modification allowed citizenship and related civil rights to hinge on color rather than on condition of servitude, a policy that heralded nearly a century and a half of race-based policies focused on making it difficult, if not impossible, for nonwhites to vote, own property, marry across racial lines, enter various professions, seek advanced education, or do much else.

Meanwhile, imperialism and immigration were radically transforming the nation’s demographic base.8  Wars and the purchase of territory added Mexicans, Native Alaskans, Caribbean Islanders, and Hawaiians to the U.S. population. Permissive immigration policies supplied factory, farm, and mine workers from China, Japan, and eastern and southern Europe. The newcomers were grudgingly tolerated, and policies were designed to keep them in their place. The low point came in the 1920s, when the eugenics movement convinced the government to stop immigration of the racially undesirable. Census data were used to design the restrictive immigration laws.9

The long practice of applying racial and ethnic categories to policies of civic exclusion began to crumble with World War II, when members of every racial and ethnic group in America fought side by side to defend democracy. Remarkably, however, this monumental policy shift from exclusion to inclusion did not alter the two premises noted at the outset of this essay. Sorting the population into discrete racial groups to make policy still made sense – the trick was to turn the classification to the advantage of those minorities who previously had suffered from its imposition.

A key early step came in a 1947 report from President Truman’s Committee on Civil Rights, which used statistics to compare health access and educational opportunities for whites and blacks, giving statistical underpinnings to the committee’s broad argument that civil rights were being denied to blacks.

Across every sector of American life two political questions began to push forward: Which racial groups are underrepresented? Does underrepresentation point to discriminatory barriers targeted at racial, ethnic, or national origin groups?

When statistical proportionality came of age in the 1960s, a new policy era was born. Social justice policies formulated in response to statistical findings were widely accepted by the end of the 1960s, as the ideal of equal opportunity fueled a demand for more equal outcomes, and as the negative goal of nondiscrimination turned into the proactive policy of redress that came to be called affirmative action.

Civil rights court cases were argued on the basis of racial differences in employment patterns, wage rates, college enrollments, and electoral outcomes. In a pivotal employment discrimination case, Griggs v. Duke Power Co. (1971), the Supreme Court reasoned that Title VII of the Civil Rights Act required the “removal of artificial, arbitrary, and unnecessary barriers to employment,” and proscribed “practices that are fair in form, but discriminatory in operation.” This reasoning shifted the emphasis in enforcement from individual motivation to statistically demonstrated consequences, from prejudice to institutional racism.10 Statistical disparity worked its way into policy and law.

Drawing on the categories employed in a 1950 government form, the Equal Employment Opportunity Commission (EEOC) in 1964 identified four minority groups: Negro, Spanish-American, American-Indian, and Asian.11 The EEOC’s record-keeping institutionalized the Civil Rights Act and in the process fixed in administrative practice a racial classification based on the four groups that had been most prominent in fighting racial discrimination for more than a century.

The 1970 census modified the EEOC classification by changing Spanish-American/Hispanic from a racial to an ethnicity designation. This was formalized by OMB when, in the 1977 Standards, it directed that Hispanic be considered an ethnicity. Hispanics were also instructed to identify on the census with one of the primary race groups, now American Indian/Native Alaskan, Asian, black/Negro, and white. Other racial, ethnic, linguistic, descent, and national origin groups (for example, Korean, Haitian, Arab) would appear in official statistics only as subcategories of the primary races (in this example, Asian, black, and white, respectively).12

The classification adopted in 1977 and used in the 1980 and 1990 censuses seemed secure and capable of discharging its purposes in policy arenas. But by the middle of the 1990s, the political landscape was transformed by demographic changes, by the rise of multiculturalism, and by the multiracial movement. New political demands called into question the existing racial and ethnic categories – and also the public purposes they were thought to serve.

As noted above, the earlier OMB Standards linked Hawaiians and Pacific Islanders with the more general Asian race. The persistent Senator Daniel Akaka from Hawaii and the constituency he led saw matters otherwise. They felt the census should recognize Hawaiian and Pacific Islanders as a separate racial category. After the OMB held public hearings and examined research showing that Hawaiian and Pacific Islanders did differ from Asians more generally, it agreed to the separate category. This decision was in keeping with the rationale that classification should facilitate racially just policies. And so in the mid- 1990s the official primary race groups went from four to five.

The ease with which this change took place was consistent with the government’s position that “classifications should not be interpreted as being scientific or anthropological in nature . . . They have been developed in response to needs expressed by both the executive branch and the Congress.”13 In the absence of science, classification decisions respond to strong voices expressing themselves in the political process. Native Hawaiians, a population group that had suffered discrimination and had the (statistical) scars to prove it, became the latest of the nation’s official races.

That being so, how can we decide on the ‘proper’ number of races? Is five the right number? Why not six or seven? And what is the right number of ethnic groups? Why only one?

Leading up to the 2000 census there was pressure to reclassify persons of Middle Eastern origin from white to their own primary race category. This effort was unsuccessful in part because the advocacy groups that engaged the issue could not agree on whether the category should be Middle Eastern, a geographic designation, or Arab American, an ethnoracial designation. (The post- 9/11 treatment of Arab Americans has since led many to doubt the political wisdom of a separate identification for this population group.)

Other advocates urged a different disaggregation of the white category, pointing out, for example, that Greek Americans and Anglo-Saxons did not belong in the same general category. The failure of various efforts (other than the Native Hawaiians / Pacific Islanders) to add to the primary racial classification can be traced to incoherent arguments, insufficient political muscle, and failure to statistically sustain claims of significant past and continuing discrimination.

In the future, however, if the advocates of such efforts make more compelling arguments and apply more muscle and more convincing data, on what grounds will the federal statistical system declare that enough is enough – that four was wrong, but five is right?

There is no science to turn to, and in its absence it is difficult to arrive at a public consensus on how many racial and ethnic groups there are in America. The edifice of racial and ethnic measurement that emerged from the civil rights period was, as social scientists like to say, undertheorized.

The increase in the number of primary racial groups in the United States by 20 percent in the 1990s went largely unnoticed because there was a noisier battle underway. The politics of affirmation marched into the middle of census taking, waving the multiracial banner. Those tidy discrete census categories, whatever their number, missed a huge sociological truth: sex occurs across as well as within racial groups. The census had recognized this 150 years ago when it first counted ‘mulattoes,’ and then, in 1890, when ‘quadroon’ and ‘octoroon’ briefly entered the measurement system in service of the policy argument that racial mixing diluted the mental and moral fiber of the nation. Later the census put the “other” category into the race question in an effort to accommodate multiracialism. But by the 1990s, multiracial rhetoric was prominent in public life, and its advocates were pressing for an explicit recognition of multiracialism in federal statistics.14

It is telling that the advocates of multiracialism barely made reference to civil rights. Instead, they brought to the fore demands for affirmation, recognition, choice, and identity. In congressional testimony, Project Race held that “not all Americans fit neatly into one little box” and that it is only right that “multiracial children who wish to embrace all of their heritage should be allowed to do so.” The Association of Multi Ethnic Americans, though recognizing that the multiple-race option would make it harder to enforce civil rights law, nevertheless insisted on “choice in the matter of who we are, just like any other community.” This testimony found it ironic that “our people are being asked to correct by virtue of how we define ourselves all of the past injustices of other groups of people.”

Of course, correcting past injustices was what the traditional civil rights organizations were all about: their mission was thus threatened by talk of choice and identity. Self-expression, they insisted, was not a good reason to revise the government’s scheme of racial and ethnic categories. In its testimony, the NAACP pointed out that the current racial classification was fashioned “to enhance the enforcement of anti-discrimination and civil rights law,” and the NAACP worried that “the creation of a multiracial classification might disaggregate the apparent numbers of members of discrete minority groups, diluting benefits to which they are entitled as a protected class under civil rights laws and under the Constitution itself.” The National Council of La Raza, the powerful Hispanic organization, weighed in. It acknowledged that though concerns about self-expression were understandable, the purpose of racial classification is “to enforce and implement the law, and to inform lawmakers about the distinct needs of special historically disadvantaged populations.”15

The issue was joined. What is the policy purpose of racial and ethnic classification – to express identity or to enforce antidiscrimination law? Perhaps reflecting the fading power of the civil rights arguments so compelling forty years earlier, “mark one or more” was introduced under the OMB’s revised standards to the racial classification system in time for the 2000 census.16

This 1997 decision put to rest the view that race is a bounded and durable trait. It challenged the basic premises of racial classification that had held sway in the United States for two centuries. And it explicitly introduced claims for expressive affirmation into ethnoracial classification.17  Though using the census to express identity was itself not new, officially accepting this as a rationale was.

At the same time, “mark one or more” created a new – and not entirely stable – statistical reality. In census data, it allowed for fifty-seven multiple-race combinations that, when added to the six single-race answers (white, the four minority races, and other), generated sixty-three possible racial identifications. Because for most purposes this classification is cross-tabulated by Hispanic/ non-Hispanic, there are 126 ethnoracial groups in the 2000 decennial census data.18

The number of categories could be expanded still further. If a future census were to allow for mixed Hispanic/non-Hispanic descent (if in the census you can have a black mother and an Asian father, why not a Hispanic mother and a non-Hispanic father?) the number of ethnoracial groupings would jump from 126 to 189.

Even at the more modest 2000 level of 126 ethnoracial groups, we now know the “mark one or more” census statistics have a reliability problem; often the same individual will give different answers at different times. This problem was demonstrated when answers to the race question in the 2000 census were matched by household with answers in a follow-up quality survey a year later. Although the overall proportion giving a multiple-race answer was reasonably constant, the internal shifting was unexpectedly high. Forty percent of those who gave multiple-race responses changed their minds by the time of the follow-up survey. And many who gave single-race answers in the census declared a multiple-race identification in the follow-up survey. For example, nearly half (45 percent) of the single-race Hawaiian/ Pacific Islanders in the census reported in the survey that they were really more than one race after all.19

From the perspective of self-expression, such shifting around is reasonable. The proponent of a “Bill of Rights for Racially Mixed People” wants “the right to change my identity over my lifetime – and more than once.”20 Popular culture daily reminds us that the blending and changing of identities has become fashionable among the young (the under-eighteen marked more than one race in the census at twice the rate of the over-eighteen). The race question in official statistics is thus being treated less as a demographic fact than as something closer to an attitude toward oneself.

Of course race has always had a subjective dimension but, as Melissa Nobles notes, “in the past, race appeared more fixed because there was a range of constraints – political, intellectual, and social. Undoubtedly, some unknown number of Americans questioned race and color as concepts and as identities, but there was not much public space for such questioning.” Race in census taking was until 1960 assigned by enumerators, whose judgment in such matters was constrained by instructions as well as by social and political realities. But today we ask individuals themselves for their views and, Nobles continues, “there are no laws, social mores, intellectual agreements, or general consensus about what constitutes a racial identity.”21

Self-classification poses potential problems within the policy arena – especially to litigation-prone race policy. Because only 6.8 million Americans (2.4 percent) gave multiple-race responses in the 2000 census, the agencies that enforce nondiscrimination law could devise collapsing rules that prevented disruptions to existing policy. Data reliability is not yet a major problem, but it will become one as the size of the multiple-race population grows. This growth will occur as rates of out-marriage among children of recent immigrants from Asia and Latin America approach those reached by Italians and Poles in the mid-twentieth century, and as multiracial identification, especially among the young, is increasingly accepted.

It is not far-fetched to expect opponents of race-sensitive policies to seize upon the low reliability of racial statistics and other data problems as a way to discredit the information that is meant to document continuing racial and ethnic discrimination.

Beyond the radical changes to measurement introduced in the 2000 census, a changing demography challenges the current classification. How will new groups of immigrants arriving in large numbers find their way into a classification system designed for a different demographic and policy moment?

Hispanic immigrants pose this question sharply. They have never found a comfortable home in the federal government’s scheme of racial and ethnic classification. Labeling them an ethnic group does not work well, particularly for Mexican Americans who blend European with Native Indian descent. Many have tried to finesse the resulting awkwardness by taking advantage of the residual “other” line on the census form. Nearly half of the Hispanics did so in 2000, most of them Mexican Americans who were claiming their nationality as a race, a race not recognized in the official statistics.22

Immigrant groups that cannot retreat to an ethnic category on the census form can be even more hard-pressed to locate themselves in the standard classification system. The recently arrived Islamic Ethiopian differs in culture, language, religion, and even skin color and facial features from those Americans who trace their origin to slaves brought from Africa’s Gold Coast. Many of today’s African immigrants have no wish to be counted as blacks, and some African American leaders do not welcome them in any case.23

The Census Bureau currently has five Race and Ethnic Advisory Committees representing the minority groups recognized in official statistics. If new immigrant groups want a say in matters of racial classification, they must either find their way into this preexisting structure or argue for their own advisory committee. To deny them their own advisory committee underlines the inconsistency between saying, as the Census Bureau does, that self-identification determines racial choice but that one’s choice has to fit into predetermined categories. New immigrants add a complexity and uncertainty to ethnoracial classification and to the policies that flow from it.

My cursory survey of American history suggests that there have been three loosely construed policy regimes facilitated by the nation’s changing schemes of racial classification.

The first used census counts to give slave-owning states extra seats in Congress and extra votes in the Electoral College, shaping power and policy for decades. The second used the data to exclude from civic life various racially defined groups. The third policy regime, fully instituted only in the 1960s, has used census data to reverse the policies of the second regime by extending civil rights to all equally, regardless of race.

Are we perhaps on the threshold of a new policy regime? The advent of the “mark one or more” option on the 2000 census suggests that the United States may well be at another historic juncture – and so does the trend of recent Supreme Court decisions.

By the mid-1980s, the Supreme Court was limiting the impact of the reasoning advanced in its 1971 decision in Griggs v. Duke Power Co. In 1987, Justice Antonin Scalia argued that statistical disparities indicating discrimination are at most evidence of “societal discrimination,” and are not remedial under antidiscrimination law.24 Although in the minority in that case, Scalia was soon to express similar views for the majority. Writing for the majority in a 1995 ruling, he asserted that “government can never have a ‘compelling interest’ in discriminating on the basis of race in order to make up for past racial discrimination.”25 And in 2003, in a case involving the University of Michigan, the Court upheld the right of universities to consider race in admissions only by ignoring remedial racial justice arguments in favor of a diversity rationale – and only after the University of Michigan had defended its policies on qualitative, rather than quantitative, grounds. In an exchange with the Court, university officers said that though critical mass advanced the educational goal of diversity, critical mass was not something that should be reduced to numbers. This ‘you know it when you see it’ claim is a long way from the ‘you know it when you’ve measured it’ argument embraced in the 1970s.

So where do we go from here?


Despite the efforts of conservatives like Ward Connerly, who in 2003 funded a California proposition to prevent that state from collecting any racial or ethnic data,26 I do not think we are headed toward a policy regime that is ‘colorblind’ and that will prevent the government from collecting data about race, ethnicity, or national origin. Powerful constituencies, notably in the public health and education fields, join with civil rights groups to contest such policy changes. They will prevail because the politics behind the color-blind movement are viewed, fairly or not, as a throwback to the policies of exclusion that the majority of Americans have firmly rejected.

At the same time, it is increasingly doubtful that policies aimed at making America more inclusive will center, as they did in the 1970s, on numerical remedies using statistical disparities as evidence of discrimination or on affirmative action. Where, then, on the continuum from no numbers to only numbers will race-sensitive policy be fashioned? Two factors feature in an answer to this question.

First, the demand for recognition, choice, and identity expression as heralded by the multiple-race advocates will continue to reverberate in statistical policy-making, especially as new immigrant groups find political voice. This will lead less to claims for strict statistical proportionality than to demands for visibility and representation. For example, if Vietnamese children comprise a quarter of a local school’s student body, parents will expect there to be at least a few Vietnamese teachers. New African immigrants will point to their growing population numbers and ask why they are not better represented in political office. And so forth.

Second, there remains a key question that reliable statistics alone can answer rigorously: How well are different groups doing? Here the focus increasingly will turn from large to smaller groups. If Hawaiians can break free from the Asian category, why can’t the new African immigrants break free from the black category, or indigenous Central Americans from the Hispanic category?

These groups are not large on the national scene, but they cluster in ways that make them noticeable in many towns and cities across the country. It is in these local jurisdictions that questions arise regarding health care, performance in the classroom, and access to the ballot box.

Whether for purposes of self-expression or to detect barriers based on race, ancestry, ethnicity, or color, the United States will continue to have a racial and ethnic classification system. But is the one now in place the right one? In my view, not exactly – though of course there is no one ‘right’ classification.

There are sound reasons to hesitate before recommending measurement changes. Disrupting statistical series, especially in an area that has just had a disruption, is no small matter. Neither is the methodological challenge of assessing the consequences for data quality of even small changes, such as how a question is worded or where it is placed on a form. Few questions are more difficult to ‘get right’ than those inquiring of race or ethnicity. There are also political consequences that at the margins could increase or decrease a group’s numbers as recorded in previous statistics. I know that it is late in the day to expect a major change for the 2010 census.

Yet neither racial measurement nor policy that relies on it is in a settled state – and this provides a historical opportunity for fresh thinking, starting with the term ‘race’ itself.

There is a strong moral case for jettisoning the term ‘race’ altogether. Relevant data can be collected without ever using the term that echoes a discredited eighteenth-century science that took physiological markers as indicative of moral worth and intellectual ability. The government doesn’t have to ask what racial group we belong to; it could simply ask what population group we belong to.27 This change, too long postponed, would break with hierarchical assumptions historically attached to fixed racial categories.

If this is considered too radical a change, the government should acknowledge that the term ‘race’ is anachronistic by using it interchangeably with ‘ethnicity.’28 The census should replace the current question on race and ethnicity with one that is subtly but significantly different:

What is this person’s race or ethnic group? Mark one or more:

     American Indian or Alaskan Native
     Black/African American
     Native Hawaiian/Pacific Islander

Such a revised question would minimally disrupt statistical series. It would retain “mark one or more” and the victory for choice that option represents. It would allow the government to enforce the Voting Rights Act and other civil rights laws that center on the 1977 classification. It would improve data quality by not forcing many millions of the nation’s Hispanics to make the kind of racial choice that has driven them to the “other” category. Commenting on the question format used in the 2000 census, the Census Bureau itself recognizes that “many Hispanics do not relate to the categories in the race question.”30

Although the Census Bureau is presently field-testing five new formats for collecting race and ethnicity data in 2010, the revision I am suggesting is not among them. I do not find the reasons given for this omission persuasive, and I strongly believe there are statistical as well as moral justifications for testing a question format that, optimally, discards the term ‘race’ altogether, or that at least does not hold to the statistically meaningless distinction between the terms ‘race’ and ‘ethnicity.’31 The OMB and the Census Bureau have a historic opportunity to back away from the presumptively immutable color-coded categories inherited from Linnaeus and his students writing in the middle of the eighteenth century.

The revised question could be paired with a second, open-ended question32 :

What is this person’s ancestry, nationality, ethnic origin, tribal affiliation?33

In the long run, this question or one similar to it should replace the race and ethnicity question altogether. That change would truly reflect that these are matters of self-identification, and that self-identification is inconsistent with forcing people into prescribed categories. But from the perspective of racial justice, it is premature to discard the official categories now used to administer antidiscrimination laws.

The open-ended question nevertheless points us to the policy frontiers of the twenty-first century.34 Details of the sort provided by the open-ended question would show whether specific groups, especially recent immigrant groups, are experiencing discriminatory barriers to jobs, schooling, or home ownership – barriers that a nation committed to a policy of inclusiveness is obligated to remove. There remain strong reasons for official statistics that can detect patterns of discrimination, and our classification scheme needs to catch up with the ways in which discrimination occurs across a very diverse population.

Many thoughtful Americans, myself included, wish that antidiscrimination law were not necessary. We want a society that is truly color-blind. But if we are ever to create such a society, we need to know what is actually happening to various population groups across the country. Accepting inclusiveness as a central policy narrative for the nation requires statistics robust enough both to keep track of whether groups historically excluded are overcoming the legacy of official discrimination and to indicate whether more recently arrived groups are being unfairly held back. More than two centuries after the Constitution started the nation down the road of racially classifying its population, there remain, unfortunately, compelling reasons to design the most policy-relevant classification scheme possible. On moral and methodological grounds, the classification used in census 2000 can and should be improved.


  • 1This essay has been prepared with support from the John D. and Catherine T. MacArthur Foundation, which provided a grant for a working group on issues of racial measurement and classification. The group includes six of the authors represented in this issue of Dædalus – Ian Haney López, Victoria Hattam, Jennifer Hochschild, David Hollinger, Melissa Nobles, and Kim Williams – all of whom critiqued this paper and, more generally, substantially shaped my thinking on the issues here discussed. Katherine Wallman and Susan Schechter, both of the Office of Management and Budget, commented on earlier versions of this essay, but have no responsibility for the recommendations advanced here.
  • 2Office of Management and Budget, Statistical Policy Directive No. 15, “Race and Ethnic Standards for Federal Statistics and Administrative Reporting,” May 12, 1977.
  • 3“Mark one or more” appeared on the census form and I use it here, but statisticians normally refer to “select one or more” to encompass phone and personal interviewing. As discussed below, in “Revised Standards for Maintaining, Collecting and Presenting Federal Data on Race and Ethnicity” (October 30, 1997), the OMB designated five primary races: “American Indian/ Alaskan Native,” “Asian,” “Black/African- American,” “Native Hawaiian / Pacific Islander,” and “White.” The Revised Standards also allow the decennial census form to include a “Some Other” option, which does not appear in other federal statistical surveys.
  • 4See David Hollinger in this issue.
  • 5For an instructive overview of racial categories in the U.S. census, see Melissa Nobles, Shades of Citizenship: Race and the Census in Modern Politics (Stanford, Calif.: Stanford University Press, 2000).
  • 6For more detail, see Victoria Hattam and Ian Haney López in this issue.
  • 7Cited in Gary Wills, “The Negro President,” The New York Review of Books, November 6, 2003, 45. This essay is drawn from Wills’s “Negro President”: Jefferson and the Slave Power (New York: Houghton Mifflin, 2003), a book that develops the “slave power” argument in impressive detail.
  • 8Aristide Zolberg, A Nation by Design (Cambridge, Mass.: Harvard University Press, forthcoming).
  • 9Margo J. Anderson, The American Census: A Social History (New Haven, Conn.: Yale University Press, 1988), chap. 6.
  • 10Alan Freeman, “Antidiscrimination Law From 1954 to 1989: Uncertainty, Contradiction, Rationalization, Denial,” reprinted in David Kairys, ed., The Politics of Law: A Progressive Critique, 3rd ed. (New York: Basic Books, 1989), 296.
  • 11John D. Skrentny, The Minority Rights Revolution (Cambridge, Mass.: Belknap Press of Harvard University Press, 2002), 101–110.
  • 12Congress has not involved itself in specifying America’s race groups, preferring instead language such as “minorities historically discriminated against.” The exception to this occurred in 1976, when Congress mandated that information on Hispanics, who were defined as an ethnic and not a racial group, be collected by government agencies in order to “assist state and federal governments, and private organizations in the accurate determination of the urgent and special needs of Americans of Spanish origin or descent” (Public Law 94-311).
  • 13From the 1977 Statistical Policy Directive No. 15, cited in the Federal Register 59 (110) (June 9, 1994): 29834. See also Katherine K. Wallman, “Data on Race and Ethnicity: Revising the Federal Standard,” The American Statistician 52 (1) (February 1998): 31–33.
  • 14See Kim Williams in this issue.
  • 15“Federal Measures of Race and Ethnicity and the Implications for the 2000 Census: Hearings Before the Subcommittee on Government Management, Information, and Technology,” April 23, May 22, and July 25, 1997. Serial No. 105-57 (Washington, D.C.: U.S. Government Printing Office, 1998), 309, 324, 382, 286.
  • 16The multiple-race option was to have been in place across the federal agencies by January of 2003, but as of this writing many agencies, including the Department of Education and the EEOC, have yet to adopt the 1997 revised standards in their compliance reporting programs. In August of 2004, the government announced a further six-month delay before it could produce reporting guidelines for how agencies were to implement the 1997 standards.
  • 17For a broad review of the census and identity creation, see Naomi Mezey, “Erasure and Recognition: The Census, Race, and the National Imagination,” Northwestern University Law Review 97 (4) (2003): 1701–1768.
  • 18Only the census is large enough to accommodate all these categories. Other government surveys – even the Current Population Survey, the largest among them – cannot provide detailed racial breakdowns.
  • 19Claudette Bennett, “Exploring the Consistency of Race Reporting in Census 2000 and the Census Quality Survey,” paper presented at the joint meetings of the American Statistical Association, San Francisco, Calif., August 3–7, 2003. The author is an analyst in the racial statistics branch of the population division of the U.S. Census Bureau.
  • 20Maria P. P. Root, ed., The Multiracial Experience: Racial Borders at the New Frontier (Thousand Oaks, Calif.: Sage Publications, 1992), 7.
  • 21Melissa Nobles, personal communication, August 30, 2004.
  • 22See Ian Haney López in this issue.
  • 23In a front-page story on August 30, 2004, The New York Times noted that Alan Keyes, a black Republican running for the Illinois Senate seat, questioned whether his opponent Barack Obama, the son of a Kenyan father and a white American mother, was really an African American: “Barack Obama claims an African-American heritage. Barack Obama and I have the same race – that is, physical characteristics. We are not from the same heritage. My ancestors toiled in slavery in this country.” Mr. Obama retorted that living under white colonialism, as his father had, was not all that different from the experience of Keyes’s ancestors, and was actually more recent. In the meantime, the wife of the Democratic presidential candidate, Teresa Heinz Kerry, who is white, on occasion referred to herself as an African American, citing the fact that she was born to Portuguese parents in Mozambique.
  • 24Discussed in Freeman, “Antidiscrimination Law,” 302.
  • 25Adarand Constructors, Inc. v. Pena, 515 U.S. 200 (1995), 239.
  • 26This proposition, known as the Racial Privacy Initiative, was defeated in California’s special election in the fall of 2003.
  • 27The Hispanic ethnic question in the census is constructed without the term ‘ethnicity.’ It reads: “Is this person of Spanish, Hispanic or Latino origin?”
  • 28In its discussion of the Standards for the Classification of Federal Data on Race and Ethnicity, the OMB notes that “There are no clear, unambiguous, objective, generally agreed upon definitions of the terms ‘race’ and ‘ethnicity.’ Cognitive research shows that respondents are not always clear on the differences between race and ethnicity. There are differences in terminology, group boundaries, attributes, and dimensions of race and ethnicity,” Federal Register 60 (166) (August 28, 1995): 44680.
  • 29This essay is not the place for technical discussion, and the exact wording of this reformed question would have to be field-tested. Alphabetizing the list would move away from current practice that lists “White” as the first option. The format also discards the many subcategories that appeared on the census form in 2000. It leaves out the residual “some other race” option, though, by congressional action in 2004, that is now a required category. This requirement was made in response to the concerns of Hispanic advocacy organizations speaking for a constituency that resists being forced to select among the other five options. However, incorporating “Hispanic” into a merged ethnicity and race question would obviate the need for “some other race.”
  • 30Phyllis Singer and Sharon Ennis, “Census 2000 Content Reinterview Survey: Accuracy of Data for Selected Population and Housing Characteristics as Measured by Reinterview,” U.S. Census Bureau, Census 2000 Evaluation B.5, September 24, 2003, xxiii.
  • 31Such a question was tested by the Census Bureau in 1996 and it performed well. The large Racial and Ethnic Targeted Test used an experimental design to test the effects of eight questionnaire formats on race and ethnicity. One of these formats combined the race and the ethnicity categories. As measured by nonresponse, a key indicator of data quality, the combined format outperformed all alternatives, and for many groups by a substantial margin. See Charles Hirshman, Richard Alba, and Reynolds Farley, “The Meaning and Measurement of Race in the U.S. Census: Glimpses into the Future,” Demography 37 (3) (August 2000): 381–393.
  • 32This question should only be included in the American Community Survey, which is a continuous sample survey administered to about 15 million households over a five-year period and designed to replace the census long form. In my view, the question should not appear on the 2010 census short form, which will go to all of America’s households. Short-form data provide block level counts used to redraw congressional and other electoral districts after each census and to enforce the Voting Rights Act pursuant to whether redistricting reduces electoral opportunities for minority candidates. Only data required for these purposes should be made available at the block level. This does not include ancestry or national origin information. Having such data available at the block level can lead to mischief, perhaps serious mischief if the government feels compelled in the war on terrorism to repeat some version of the Japanese American internment during World War II, which made use of census information from small geographic areas.
  • 33This question is presently being field-tested by the Census Bureau. It is designed to accommodate as many as nineteen illustrative categories, a slight increase over the sixteen used in the 2000 census ancestry question.
  • 34Current data-capture technology can reliably record responses to such a question. Optical scanning and intelligent character recognition were very successfully used in the 2000 census, recording open-ended written responses at exceptionally high levels of accuracy.