More than one million people reported their race as American Indian/Alaska Native (AIAN) in the 2000 U.S. census but did not do so in the 1990 census. We ask three questions. First, which subgroups had the greatest numerical growth? Second, which subgroups had the greatest proportional increase? And third, are the 2000 single-race AIANs and the 1990 AIANs the same set of people? We use full-count and high-density decennial census data; adjust for birth, death, and immigration; decompose on age, gender, Latino origin, education, and birth state; and compare the observed subgroup sizes in 2000 with the sizes expected based on 1990 counts. The largest numerical increases were among adolescent and middle-aged non-Latinos, non-Latino women, and adults with no college degree. Latinos, women, highly educated adults, and people born in Eastern states had the largest proportionate gains. The ability to report multiple races in 2000 and the new federal definition of “American Indian” may have especially affected these groups, although personal-identity changes are probably also involved. We find that thousands of new Latino AIANs reported only one race in 2000, but many 1990 AIANs reported multiple races in 2000. Thus, the 1990 AIANs and 2000 single-race AIANs are not always the same individuals.
Between 1990 and 2000, the size of the enumerated American Indian/Alaska Native population doubled. In 1990, about two million people marked the “American Indian or Alaska Native” box for the U.S. census race question.1 Ten years later, that number jumped to more than four million. Setting aside births, deaths, and immigration, about one million new American Indians/Alaska Natives (hereafter American Indian or AIAN) remain in the Census 2000 data. These one million individuals reported a non-American Indian race (e.g., white) in 1990 when single race responses were required, and in 2000 either added American Indian as an additional race (e.g., white and AIAN) or reported only American Indian race.2 In this research, we explore the characteristics of “new” American Indians and discuss the theoretical and practical implications of this large-scale intercensus racial reclassification.
The relative population increase among American Indians was much larger than that of other groups.3 The non-Latino white population decreased by 8 %, as more people left the group —through mortality, emigration, or change in race response—than joined it in 2000 by adding (or switching to) “white” as a race.4 Non-Latino blacks and Japanese also decreased in population size. In contrast, there was a 65 % increase in non-Latino American Indians between 1990 and 2000, and Latino American Indians increased by 153 %. Only Pacific Islanders increased as much, with 77 % and 108 % increases in non-Latino and Latino Pacific Islanders, respectively.
Group sizes change as a result of new identity decisions by new immigrants or for mixed-heritage newborns (cf. Waters 1999) or shifts in established identity among those with longer tenure in the society (cf. Loveman and Muniz 2007; Sturm 2011). Either type may be permanent (Nagel 1996) or constantly negotiated (Khanna and Johnson 2010). These personal identity decisions and shifts are important sources of change in the meanings of socially defined race or ethnicity categories.
Changes in responses to questionnaires can also be caused by more mundane methodological issues, such as wording, formatting, and inconsistencies between a questionnaire and a person’s self-conception (Bates et al. 2006). For example, Hispanic/Latino status is assessed in a separate question—not in the race question—leading many Latinos to express confusion and disagreement about how to properly answer the race question and to give inconsistent answers when they answer at all (cf. Bates et al. 2006; Berkowitz 2001; Dowling 2014; Rodríguez 2000).
Our research centers on changes in responses to the race question—whether resulting from shifts in established identity or from methodological issues—that lead to changes in net American Indian group size as measured in adjacent censuses. Such changes have been documented in other groups. Some people drop their minority status labels, including later-generation Mexican Americans who have achieved socioeconomic parity with whites (Alba and Islam 2009; Duncan and Trejo 2011), Puerto Ricans in the early twentieth century (Loveman and Muniz 2007), and young adults in the Soviet Union (Anderson and Silver 1983). At the same time, white Americans selectively highlight certain ancestral origins (Hout and Goldstein 1994; Waters 1990) or are redefined as minority by themselves or others (Saperstein and Penner 2010, 2012). Short-term variation in race responses has been particularly noted among part–American Indians (cf. Harris and Sim 2002; Martin and Gerber 2005).
The million-person surge in the American Indian population in 2000 continues a long-standing pattern among American Indians.5 Each census since 1960 has shown hundreds of thousands of new American Indians—people who joined that population through response changes rather than birth or immigration. There is substantial potential for even more growth; millions report American Indian ancestry yet do not report this as their race.6 In all, the jump between 1990 and 2000 greatly exceeded population projections but was not entirely unprecedented.7
We address three research questions. First, we measure numerical increase: which subpopulations gained the most people? This helps us understand whether the new American Indians are similar to the 1990 population, which is useful for policy and governance purposes. Second, we look at proportionate increase: which subgroups showed the greatest propensity to identify as American Indian for the first time in 2000? These results point to social locations favorable for racial identity change or those especially affected by changes in the census form. Third, we ask whether the 1990 American Indian population is the same set of people as the 2000 single-race American Indian population. This is relevant to social scientists and policy evaluators hoping to assess trends in education, income, location, and well-being.
As in most prior research, we identify net changes in population size by comparing two cross-sectional data sets; longitudinal population data are not available for these years. To identify subgroups with the largest increases, we calculate the expected population in 2000 (by age, gender, Latino status, educational attainment, and birth state) and compare each number with the observed subgroup size in 2000. Through the Census Bureau’s Research Data Center network, we use the full-count short-form data (~100 % of the United States) and the complete long-form data (a 17 % sample) from the 1990 and 2000 censuses. Our work extends prior studies to a new census year using better data and in previously unexplored directions.
Theoretical and Practical Importance
The existence of more than one million new American Indians in 2000 relates to sociological theories of group boundaries. Social theorists describe how group boundaries are constructed, negotiated, maintained, and moved by people and institutions both inside and outside the group (Alba and Nee 2003; Barth 1969; Loveman and Muniz 2007; Omi and Winant 1994; Wimmer 2008). Alba and Nee (2003:60–61) identified three ways in which individuals can move from one group to another: boundary crossing (changes in individuals that make them newly qualified for membership in a different group), boundary blurring (in which “the social profile of a boundary becomes less distinct”), and boundary shifting (“the relocation of a boundary so that populations once situated on one side are now included on the other”).
The boundaries of “American Indian” have been crossed, blurred, and/or shifted for decades. However, not all these processes are equally evident between 1990 and 2000. Most new American Indians have not crossed a social boundary by changing their language use or residential location; for example, speaking an indigenous language was less common among American Indians in 2000 than in 1990.8 Rather, evidence suggests that the American Indian race boundary is sufficiently blurry that people who have a serious identity awakening (cf. Sturm 2011), or a more ephemeral moment of connection to their American Indian ancestry, feel qualified to claim American Indian race.9 Boundary shifting through the federal redefinition of “American Indian,” discussed later, may also be involved.
Racial formation theory (Omi and Winant 1994) indicates that the blurring or shifting of boundaries can transform racial categories, which has the potential to reorganize racial dynamics and redistribute resources accordingly. A person’s self-defined identity and how they are seen by others are consequential for many aspects of life, including where they live, who they marry, their education and occupation, and their health and longevity. New American Indians who are perceived by others as non-Indian (e.g., white) may have social, health, and economic outcomes more similar to their first race (Saperstein 2012), especially if the identity change is recent, short-term, or not meaningful. Conversely, changes in self-presentation accompanying a serious identity change could affect lived experience.
By joining en masse, new American Indians increase the size, geographical distribution, and average education of the American Indian population (Eschbach et al. 1998). Practically speaking, this undermines efforts to evaluate the effects of programs and to develop policies relevant to the needs of lifelong American Indians and tribal communities. It also complicates population projections and estimates, confuses measures of undercount, and likely dilutes the average strength of ties to a tribal community, homeland, or culture.
Large-scale population shifts undermine the (usually implicit) assumption that birth and immigration are the only sources of new members. This problem cannot be sidestepped by selectively ignoring respondents, as is done, for example, when analysts focus only on single-race American Indians. Nor can we ignore American Indians as a group; they must continue to be included in policies, programs, population estimates, and social research. A researcher who takes into account characteristics of new American Indians is less likely to draw inaccurate conclusions about the group as a whole (e.g., interpreting group advances in economic standing as evidence of individual improvements). Similarly, someone who understands that the single-race population in 2000 is not the same group of individuals as the 1990 population (as shown here) will not make misleading comparisons between the two populations.
The American Indian Population Boom
There was a steep, steady decline in the American Indian population from first contact with Europeans to the turn of the twentieth century (Thornton 1987), with a population nadir in 1900 at about 240,000. The slow population increase between 1900 and 1950 reflects competing forces of high fertility and reduced mortality on the one hand, and powerful federally backed assimilation programs on the other (Thornton 1987). By 1950, there were about 350,000 American Indians enumerated in the U.S. census.
An accurate count of the American Indian population has long been a struggle (Census Office 1894). More than 60 years ago, Census Bureau employee Calvin Beale (1958:537) wrote, “Although there is little Indian immigration or emigration, no notion of the biological natural increase of Indians can be gained from the data for successive censuses.” Before 1960, enumerator inconsistencies and Census Bureau policies tended to minimize the American Indian count. When enumerators discerned each person’s race through observation or assumption (before 1960), American Indians living in nonstereotypical places were frequently coded incorrectly (cf. Census Office 1894). When they were found, enumerators were to mark American Indian race only if they were enrolled in a tribe or recognized as American Indian in their community (Thornton 1987).
In 1960, the census transitioned to a mail format, and the American Indian population boom began. Self-identified American Indians came forward and, along with births, caused a 46.5 % population jump between 1950 and 1960 (Thornton 1987). If prior enumerator error were the only issue, the corrective jump would have appeared only in 1960. This was not the case, though. In each census from 1960 to 1990, the American Indian population jumped by hundreds of thousands. Researchers responded with demographic error-of-closure studies highlighting identity change as a major cause (Eschbach 1993; Harris 1994; Passel 1976, 1997; Passel and Berman 1986) and qualitative documentation of formerly non-Indian adults transforming into people with heartfelt American Indian identities (Fitzgerald 2007; Liebler 2001; Nagel 1995, 1996; Quinn 1990; Sturm 2011).
Prior work suggests at least three reasons for the large net increase between 1990 and 2000. First, negative stereotypes and cultural repression that previously curbed identification have waned such that lifetime American Indians are more willing to publicly embrace this identification (Cornell 1988; Nagel 1995, 1996; Thornton 1990). This is likely part of the increasing population since 1960 but probably not a large part of the increase from 1990 to 2000.
Second, there may be subtle changes in the appeal of American Indian identification among whites with American Indian ancestry who see it through an “optional” or “symbolic” lens, as is common for those with European heritages (Alba 1990; Gans 1979; Waters 1990). For some, embracing American Indian heritage—as opposed to other heritages or none at all—may be motivated by hopes of gaining political, financial, or social benefits (cf. Quinn 1990). For others, it may be a profound identity awakening, made possible by the freedoms of self-definition afforded whites (cf. Fitzgerald 2007; Sturm 2011). The private self-labeling involved in modern census race questions as well as the ability to add a second race in 2000 could fit well with the way whites of Indian descent conceptualize their heritage. A small increase in the appeal of one prevalent ancestry (e.g., Irish or American Indian) can greatly increase population size (cf. Hout and Goldstein 1994).10
Third, changes in census design, advertising, instructions, and definitions (Bates et al. 2006) may have affected the enumerated Latino American Indian population. The federal definition of “American Indian” became more inclusive in 1997, changing from “A person having origins in any of the original peoples of North America, and who maintains cultural identification through tribal affiliation or community recognition,” to those with origins in “. . . original peoples of North America and South America (including Central America). . . .” (Office of Management and Budget 1978:19270, 1997:58789). Although the definition of “American Indian” may not have fit them in 1990, many Latinos officially qualified as American Indian in 2000.11 Also, other questionnaire changes successfully increased Latino response rates on the Census 2000 race question (Martin 2007).
Areas of Uneven Growth
The extent of race response changes is likely to be uneven across subpopulations. We focus on variation by age, gender, Latino status, educational attainment, and birth state.
Identity changes are more likely at some life course and developmental stages (e.g., during adolescence). Most previous research on large-scale change in identification has included age as a primary line of investigation, with two consistent results. Young adults are the most likely to change race or ancestry categories between censuses (Anderson and Silver 1983; Lieberson and Waters 1986; Loveman and Muniz 2007), and substantial numbers of people in all age groups change their identification (Alba and Islam 2009; Anderson and Silver 1983).
Identity, socialization, and cultural scripts differ for men and women; thus, gender variation in patterns of race response can be expected. Loveman and Muniz (2007) found an especially large increase in the population of young white women (ages 20–24) in early twentieth century Puerto Rico, and suggested that women’s races are reconsidered (and reclassified) after marriage. Other researchers have not found clear gender differences.
The U.S. Latino population has grown markedly, and more Latinos are reporting American Indian heritage (Decker 2011). In 1990, about 383,000 Latinos indicated American Indian race or ancestry, whereas about 956,000 did so in 2000 (data from Ruggles et al. 2010). Previous studies have not explored this dynamic. Analyzing racial identity change among Latino American Indians is complicated by simultaneous changes in Latino identification; persons of Mexican origin are less likely to report Latino status as personal income or education increase (Alba and Islam 2009; Duncan and Trejo 2011). This countervailing trend away from Latino identification may dampen the observed net increase in Latino American Indians.
Racial self-perception and perception by others can be influenced by a person’s achieved education and can also affect a person’s educational attainment. More-educated people give more complex responses to ancestry questions (Lieberson and Waters 1993), although well-educated Latinos can be unlikely to report Latino origin (Duncan and Trejo 2011) and may be perceived as nonminority by others (Loveman and Muniz 2007). Unfortunately, we cannot separate these dynamics. Instead, we identify net patterns by disaggregating the population by education level. Eschbach and colleagues (1998) did this for 1980 and 1990, finding higher average education among new American Indians in 1980 (but not 1990) than among previously identified American Indians.
Characteristics of a person’s location are related to the development and maintenance of their racial identity (Eschbach 1992; Harris and Sim 2002; Kanaiaupuni and Liebler 2005). We use birth state as a measure of the context of identity development.12 Features of one’s birth state may affect childhood racial-identity development, especially if there is a concentration of American Indian people or homelands (Liebler 2010b). Quantitative (Eschbach 1993) and qualitative (Sturm 2011) evidence highlight a preponderance of new American Indians outside a 10-state “old Indian region.”13 A federal program of the 1950s and 1960s relocated thousands of reservation-based American Indians to urban areas in the old Indian region and in California, New York, Texas, and Michigan (Tax 1978).
Identity Flows Across Censuses
Although our first two research questions focus on changes in group size, the third question asks about identity flows between types of responses. We ask whether some new American Indians are identified as single-race American Indian in 2000 (rather than reporting multiple races) and also whether some 1990 American Indians reported multiple races in 2000. These possibilities are sometimes overlooked, perhaps because the 2000 single-race American Indian population is about the same size as the 1990 American Indian population.
Quantitative and qualitative evidence suggest that the 1990 population would not be the same individuals as the 2000 single-race population. All previous quantitative research on population increases among American Indians (covering 1960 to 1990) documented hundreds of thousands of single-race new American Indians. We see no reason for the 2000 census to be different. Qualitative evidence (Liebler 2001; Sturm 2011) has shown that single-race responses are common among people who have become American Indian through an identity awakening and that multiple-race responses are common among lifelong American Indians.
In this research, we provide information about the one million new American Indians in 2000: which subpopulations grew the most, which grew the fastest, and likely source populations. Extending a long tradition, our study of the census error of closure for American Indians uses full-count nonpublic data and covers a previously unstudied decade that saw unprecedented growth. These en masse changes in self-perception and group membership are relevant for sociological theory and have practical implications for policy evaluation and the study of socioeconomic inequality.
Data and Methods
For our primary analyses, we use the 1990 and 2000 full-count nonpublic decennial census microdata available through the Census Research Data Center network.14 These data contain all census respondents, as opposed to public-use samples. Full-count data provide maximally accurate estimates of each subpopulation.15 For education and birth-state analyses, we use nonpublic long-form data, which contain records for all long-form respondents (a sample of about one in six households in both years).
The Census Bureau creates weights to make long-form data nationally representative, but population estimates in weighted long-form data do not exactly match the full-count data (Hefter and Gbur 2002; Schindler et al. 1992). Compared with the full-count data, the weighted long-form data give slightly larger estimates of the non-Latino American Indian population in 1990 and 2000 as well as the Latino American Indian population in 1990, but smaller estimates of the Latino American Indian population in 2000. To account for the different base population sizes, we deflate/inflate all numerical estimates derived from the long-form data to match the Latino origin-specific totals from the full-count data.16 Only after this adjustment do we use weighted long-form counts in the sample selection adjustments of the full-count data described later.
Calculation of Expected 2000 Population Using 1990 Full-Count Data
To calculate the expected number of American Indian responses in the 2000 full-count data, we begin with the 1990 full-count microdata and adjust for mortality, immigration, and other factors as described later. These full-count data contain information for all people in the United States in 1990, including 1,967,100 American Indians.17
Before calculating mortality estimates, we exclude cases in the 1990 data based on three criteria. First, to eliminate population differences resulting from varying imputation practices, we exclude 68,700 cases in which the Census Bureau imputed missing race responses. Second, to account for immigration, we estimate the number of foreign-born individuals (53,700) by age, gender, and Latino status, using the deflated long-form data. Then in the 1990 full-count data, we subtract the estimated foreign-born American Indian population from their age/gender/Latino category. Third, some West Indians and Asian Indians mistakenly marked the American Indian category on the census form (Liebler 2004). We use long-form information on language and ancestry to identify likely Asian Indian or West Indian respondents (700) and subtract the deflated estimates (by age/gender/Latino status category). These steps reduce the 1990 American Indian base population to 1,844,000.
We then estimate mortality in the base population between 1990 and 2000 by applying single-decrement life tables with race-, gender-, and age-specific mortality rates (National Center for Health Statistics 2003).18 This leaves 1,751,000 people; in other words, if no one changed their race response, we would expect 1,751,000 U.S.-born American Indians, ages 10 and older, in Census 2000.
Calculation of Expected 2000 Population Using 1990 Long-Form Data
For changes in education and birth-state characteristics, we use weighted long-form data to calculate both observed and expected population sizes. We calculate the expected population size in the 1990 weighted long-form data by applying listwise deletion to cases with imputed race, those born abroad, and those who appear to be West Indian or Asian Indian, and then applying mortality adjustments. Based on these calculations, we would expect 1,841,700 U.S.-born American Indians (ages 10 and older) in Census 2000.
Case Selection for Observed Population Size Using 2000 Data
There were 4,010,100 American Indians enumerated in the 2000 full-count data. In order to make clean comparisons to the expected population, we exclude 746,300 children born between the censuses and 167,900 cases with imputed race responses. (For more about imputation, see U.S. Census Bureau (2004).) We use the Census 2000 long-form data to estimate the foreign-born American Indian population in 2000 (236,200) and likely West/Asian Indian respondents (3,400) and then subtract the deflated/inflated number from each age/gender/Latino category. This leaves 2,856,300 observed U.S.-born American Indians, ages 10 and older, in the Census 2000 full-count data—1,105,300 more than expected.
To calculate observed populations by birth state and education, we use the Census 2000 long-form data and remove children under age 10, imputed race, foreign-born, and likely West/Asian Indian respondents. The result is 3,090,800 observed U.S.-born American Indians, ages 10 and older, in 2000; this is 1,249,100 more than expected based on the 1990 long-form data.
Our method assumes that individuals who reported American Indian race in 1990 and/or 2000 were consistent in their reported birth year, gender, Latino status, and birth state across censuses, although a small number of changes would not affect our results. We restrict the study of education to adults who were at least age 25 in 1990. Nevertheless, some increases in the population’s education—especially among women—are due to society-wide increases in personal educational attainment (Freeman and Fox 2005).
Analytic Strategy and Multiple-Race Responses
To find the largest and fastest-growing subpopulations of new American Indians, we compare the expected size with the observed size of each group without regard to the number of races a person reported. For the third research question, we unpack multiple-race responses to investigate identity flows. Multiple-race responses are common among American Indians; about 40 % of the observed American Indian population reported multiple races in 2000.19 Unfortunately, our data are cross-sectional, so we cannot be sure of individuals’ previous race responses. Nevertheless, we make fruitful comparisons of aggregate population sizes to discern changes in aggregate population characteristics.
Our first research question asks which subpopulations had the largest net increases between 1990 and 2000. In Fig. 1 and Table 1, we show the expected and observed American Indian populations by age, sex, and Latino status. We present Latino and non-Latino numbers on the same scale to give a sense of each population’s relative size. Notably, the observed population is substantially larger than the expected population in every age group, among Latinos and non-Latinos, and among men and women. Joining the American Indian race was widespread in 2000. Still, some subpopulations dominate the new American Indian population; teens, young Baby Boomers, and non-Latino women are numerically dominant.
In Fig. 2 and Table 2, we compare expected and observed educational attainment among American Indians ages 35 and older in 2000. Although the bulk of American Indian adults have no college degree, American Indians in 2000 are notably more educated than American Indians in 1990. Whereas the modal education category in 1990 was no high school diploma, the 2000 population’s modal categories are high school diploma (among non-Latinos) and some college (among Latinos). Without new American Indians (or personal gains to education), there would have been twice as many high school dropouts as people with a college degree in 2000 (Table 2, column 1). With new American Indians, however, there are almost as many with a college degree as there are high school dropouts (Table 2, column 2), and there are twice as many with at least some college experience as there are high school dropouts.
To provide geographic context to these population increases, Table 2 also shows observed and expected population sizes by birth state. More than one-half of new Latino American Indians were born in California, Texas, or New York. The birth-state distribution of non-Latino American Indians is more dispersed, so we provide a map in Fig. 3. Using stars, we denote “old Indian” states—areas with long-standing American Indian populations and little previous population change resulting from shifts in identity (Eschbach 1993). Like Eschbach (1993), we find that most new American Indian people have roots outside the “old Indian” states. The seven most common birth states—with more than 35,000 new non-Latino AIANs each—were California, Oklahoma, Texas, New York, Michigan, Ohio, and Illinois. Among these, Oklahoma is the only “old Indian” state; as “Indian Territory” in the 1800s, Oklahoma was the main destination for people forced out by the 1830 Indian Removal Act. Notably, the other common birth states received thousands of American Indian migrants from reservations in the 1950s and 1960s as part of the Indian Relocation Act of 1956 (Tax 1978).
In Fig. 4, we summarize characteristics of the new American Indian population. The top panel shows that many new non-Latino American Indians under age 35 are teens in 2000; their parents are probably responsible for the change in race response. We see evidence of few new American Indians in their 20s even though many experienced major life-course transitions likely to affect identity; perhaps the aggregate numbers are masking countervailing trends. About one-half of new non-Latino American Indians were ages 35 or older in 2000 (many with 12 to 14 years of education). In the middle panel, we show the age and education distribution of new Latino American Indians. In contrast to the top panel, Latinos are especially young, and Latino adults have relatively little education. The bottom panel of Fig. 4 combines groups to show the distribution of the whole set of new American Indians in 2000. About one-sixth are Latino, one-third are non-Latino and under age 35, and almost one-half are non-Latino ages 35 and older. The birth-state distribution shows that about one-sixth of new American Indians were born in California, one-seventh were born in “old Indian” states, and almost one-half were born in other states in the South or Midwest.
Our second research question asks us to identify which subpopulations have large proportionate increases. People in these groups may have particularly compelling reasons for marking American Indian for the first time in 2000.20
In Fig. 5, we express the observed population size as a percentage of the expected population size by age, sex, and Latino status. Underlying numbers are in Table 1. For example, 153,350 non-Latino American Indian boys ages 10–14 were observed in 2000, which is 174 % of the 88,050 expected. Thus, Fig. 5 shows 174 % for this group. Among non-Latino American Indians of both genders, most observed age groups are about 150 % of their expected size. Although women and girls are numerically dominant among new non-Latino American Indians, their proportionate increase is similar to men’s. The proportionate increase among Latino American Indians is remarkably larger than that among non-Latinos; Latino American Indian groups are 200 % to 250 % of what was expected. Proportionate increases were especially large among Latino teens and Latino women under age 50.
Adults with higher educational attainment are disproportionately overrepresented among new American Indians, as shown in Fig. 6. There are more than three times as many non-Latino and Latino American Indians with graduate degrees in the observed population as in the expected population, and 2.5 to 3.5 times as many with bachelor’s degrees. In contrast, the observed size of the non-Latino American Indian population with less than a high school education is near the expected size.
The birth-state maps shown in Fig. 7 reveal several states in which the observed population is more than three times the size of the expected population; these states are darkest on each map. There is strong draw toward American Indian identification in most East Coast states, the lower Midwest, and Hawaii; people born in those places were particularly likely to newly select American Indian race in 2000. Birth states with long-standing American Indian populations did not see substantial proportionate increases.
Multiple Race Responses
In our third research question, we focus on identity flows (to the extent allowed by our data) to discern whether all new American Indians gave multiple race responses in 2000. We also ask whether it is plausible that all 1990 American Indians reported single-race American Indian in 2000. We find that neither is a safe assumption.
In Tables 1 and 2, positive numbers in column 6 (in boldface type, for clarity) denote subpopulations with more new American Indians than multiracial American Indians; some new American Indians must have marked only one race. For example, in Table 1, we see that there were expected to be 88,050 non-Latino boys ages 10–14 in 2000 (column 1), but the observed number was 153,350 (column 2), of whom 58,400 were reported as multiracial (column 4). The difference between the observed and expected number of boys is 65,300 (column 5); there are about 6,900 more new non-Latino boys in the data than there are multiple-race responses in this same subcategory (column 6). Thus, we have evidence that 6,900 boys were reported as non-American Indian in 1990 (when they were ages 0–4) and as single-race American Indian in 2000. This crude number, gained by subtracting aggregate group sizes, gives a minimum estimate of the number of new, single-race American Indians. Most highlighted cells are among Latino American Indians of all ages and education levels, especially those born in California and Texas. Among non-Latinos, we see evidence of new single-race American Indians who are children (ages 10–14); adult women (ages 30–44); people with a bachelor’s degree; or born in Louisiana, New Mexico, South Carolina, or Texas. The positive numbers in column 6 sum into the tens of thousands.21
We also find evidence that some 1990 non-Latino American Indians reported multiple races in 2000. There are many instances in which the 1990 subgroup size (the black bar in Fig. 8) is greater than the 2000 single-race population (the light bar in Fig. 8); this is true in “old Indian” states, among men and women in their 20s, and among less-educated adults. Although this direction of response change is not particularly surprising, it again illustrates that the single-race American Indian population in 2000 is not the same group enumerated in 1990. In sum, many non-Latino 1990 American Indians reported an additional race in 2000, and many Latinos became new single-race American Indians in 2000.
Discussion and Conclusions
Identity change and measurement issues contributed to the largest net increase in American Indians yet. About two million were enumerated in 1990, and an additional one million new American Indians were enumerated only a decade later. Casual observers might attribute this jump solely to changes in the race question in the 2000 census, but net growth in the American Indian population has been 400,000 to 800,000 in each census since 1970. Using dense nonpublic data, we have decomposed the numerical and proportional net increase along five dimensions and given evidence of race response changes both into and out of the single-race American Indian category.
Who are the one million new American Indians? Our analyses show two patterns. First, the population increase occurred across the board. Men as well as women, adults as well as children, the highly educated and less educated, and Latinos and non-Latinos all joined the American Indian population between 1990 and 2000. Generally, the points of largest numerical increase in the American Indian population were in subpopulations that were already large: teens and middle-aged people, those with high school or some college education, non-Latinos, and people born in Oklahoma or populous states with long-term urban Indian populations.
Second, the 2000 population has a higher proportion of females and highly educated people than the 1990 population. New American Indian women outnumber men by almost 85,000, perhaps because of uneven interest in genealogical research or gender differences in social rules defining the American Indian race boundary. In the youngest group (ages 10–14), boys were as likely as girls to become new American Indians; maybe parents who changed their race responses also changed their children’s. Education levels attained by new American Indians are higher than the averages of U.S. minority groups, hinting that whites are the population from which most new American Indians are drawn. The education profile of previously identified American Indians could only be matched if new American Indians were pulled from among the least-educated whites or other people of color with comparable education profiles.
What types of people were disproportionately drawn into the American Indian population? In most cases, the number of American Indians born in East Coast and lower Midwest states doubled or tripled. Qualitative research has documented real identity changes, although not always widespread acceptance of these changes, among former whites living outside the old Indian region (Fitzgerald 2007; Sturm 2011).
The number of Latino American Indians in most age groups more than doubled between 1990 and 2000, probably as the result of three enumeration issues (see Martin 2007): (1) increased Latino response rate to the race question; (2) confusion about a “proper” race response for Latinos, given no Latino category on the race question; and (3) a relevant expansion of the definition of “American Indian.” Noting these issues, Census Bureau methodologists have recommended that the race and Hispanic-origin questions be combined into a single inclusive question (cf. Compton et al. 2012). Experimental results show that if implemented, this questionnaire change could probably address these enumeration issues effectively (Compton et al. 2013). Not all new Latino American Indians would necessarily rescind their newfound race response, however; claiming group membership for any initial reason can have long-standing effects on self-perception and interaction.
The number of American Indians with a bachelor’s degree or graduate degree increased by more than 250 %; many people with college experience joined the American Indian population. Contrary to assimilationist expectations, this pattern might indicate that people strategize when making identity decisions, aiming to maximize material and symbolic benefits. Follow-up research could reveal that exposure to racial diversity in college, increased academic knowledge of American Indian history and cultures, or interest in scholarships affect these individuals’ sense of what is beneficial and why.
Are the 1990 American Indians and the 2000 single-race American Indians the same people? We found two types of evidence that these two groups are not equivalent. First, thousands of individuals reported a non–American Indian race in 1990 and reported American Indian single race in 2000. This was common among Latinos as well as among middle-aged women and the college educated. This provides evidence that race is indeed socially constructed and that race group boundaries, like other social boundaries, are crossed by many. Whether race response changes are due to questionnaire issues (as may be true for Latinos) or real identity changes (as may be true for non-Latino women), current methods of collecting and analyzing data on race do not take this possibility into account. False assumptions of equivalency give misleading information that might suggest, for example, that tens of thousands of American Indians went to college or that they migrated from “old Indian” states to settle in California. A researcher could mistakenly conclude that education policies are effective (and stratification is being alleviated) or that programs are needed to help California’s new residents.
Second, we find evidence that a number of 1990 American Indians reported multiple races in 2000, especially non-Latino American Indians who were older, less educated, or born in an old Indian state. These American Indians are likely to be long-term American Indians—they fit the profile of the 1990 population—but they are excluded by researchers who ignore multiple-race responses in favor of comparing the 1990 population with the 2000 single-race population. Both of these results sound a warning to analysts hoping to use census data for cross-time comparisons: these data must be used with caution and caveats.
In short, our key findings are threefold. First, the one million new American Indians come from all age, gender, Latino, and education groups, including many non-Latino adults with 12–14 years of education. Second, disproportionate increases in the number of American Indians who are very well educated, Latino, or born outside an old Indian state point to effects of changing measurement and personal changes to identity. Third, any comparison between 1990 and 2000 American Indians must be made with great care because these are not the same individuals. New American Indians are probably distinct along most social and economic dimensions.
This research has theoretical and practical implications. Subpopulations with especially large jumps are ripe for future investigations of why these race boundaries are particularly blurred. For example, the social definition of American Indian race group boundaries—who can be socially considered American Indian and who cannot—is being stretched by new Latino American Indians and by new American Indians who once self-identified as white. Many new American Indians are probably formerly self-identified whites, according to qualitative studies of people with American Indian ancestry (see footnote 6). Abandoning a privileged (white) racial identity in favor of a minority identity may seem counterstrategic, considering the socioeconomic and health disparities between the groups. Ironically, this choice may be made possible by whites’ freedom of self-definition (cf. Fitzgerald 2007; Sturm 2011; Waters 1990), with social or symbolic benefits outweighing costs for former whites whose income, education, and health are already well established.
Even Latinos who identify as indigenous only for the sake of the awkward two-question census format may affect race boundaries through resulting shifts in self-perception and feelings of cohesion with North American Indians. Out-group members may redefine who is considered a “real” American Indian to match the people they meet who claim this identity. Notably, tribe-specific classification and analysis (Liebler and Zacher 2013) is less possible for Latino American Indians and Latinos self-identifying as mestizo22 because the historical sociopolitical systems of Latin America have emphasized villages, regions, and language groups rather than “tribes” per se (cf. Bates et al. 2006; Berkowitz 2001).
To the extent that the American Indian race boundary is changing over time in irregular ways, statistics on education, poverty, intermarriage, and health status for American Indians will also show uneven patterns of change. For example, new American Indians’ higher education hints that they also may have better health, higher income, and more stable housing, especially if they are former whites who are seen as such by others. Survey designers should ask about (self-defined and other-defined) race at each wave of a longitudinal study, and analysts should be careful not to analytically or conceptually assume that race responses are stable over time.
Importantly, our results give information about net changes in race responses, rather than gross changes. The large net increase in the number of American Indians could be masking substantial churning in racial identification. We cannot know from these data whether 0 or 50,000 or 500,000 people changed their race response away from American Indian in this decade. The history of American Indian assimilation (Thornton 1987) and evidence from Latinos (Duncan and Trejo 2011) suggest that this number is larger than zero.
Speculation abounds about the meaning of these race response changes for the individuals involved. One group of researchers, including Anderson and Silver (1983:482), claims that “change in ethnic self-labels is generally not made lightly and that it typically implies a serious change in ethnic attachments.” In this scenario, the boundary changes have real implications for the redistribution of resources through organized race projects (Omi and Winant 1994) and ethnic mobilization (Nagel 1994; Sturm 2011). If they have made a serious change in their identities, new American Indians would be (or become) similar to previously identified American Indians in terms of life experience and policy needs. Low levels of churning would point to this interpretation.
Other scholars see this change in race response as less personally meaningful, more likely to be short term, and less likely to affect the future actions of individuals. In this scenario, the new American Indians will not adopt the segregation, education, marriage, and health patterns of previously identified American Indians. Instead, as Eschbach (1995:103) suggested, “Some people with a very low degree of Indian descent may continue to identify as Indians as long as the symbol is available and socially meaningful . . . . Ethnic identification for most will be ‘costless’ and voluntary.” If the change is not made for serious reasons or for long periods of time, new American Indians may not pass this identity to their children and may identify differently when asked about their race in a slightly different way. Future research showing extensive churning in racial identification would lend support to this understanding of new American Indians as not fully attached to an American Indian identity.
Regardless of the meanings of changes in race responses, the changes themselves fundamentally affect how these data can be used. Analysts using race-specific data must keep in mind that these race categories are socially constructed and that race is not a fixed biological feature of an individual. Researchers must take this into account in conceptualization, measurement, and modeling strategies that use data on race. An explicit or implicit assumption that race is permanent ignores theory, research, and reality.
Acknowledgments and Disclaimer
This is a posthumous publication for Dr. Ortyl who died suddenly in 2013. His contributions to this article and sociology’s broader intellectual community were substantial. The research was conducted in the Minnesota Research Data Center, which receives funding from the National Science Foundation (SES-0851417 and ITR-0427889). Funding was provided by a Grant-in-Aid-of-Research from the College of Liberal Arts, University of Minnesota. We also gratefully acknowledge support from the Minnesota Population Center, which is funded by a center grant from the National Institutes of Health (R24-HD041023). This research was presented at the 2011 annual meetings of the Population Association of America and the Research Data Center Annual Research Conference, and was published by the U.S. Census Bureau as Center for Economic Studies Working Paper 13-02. For helpful comments, we thank J. Trent Alexander, Caren Arbeit, Julia Rivera Drew, Catherine Fitch, Liying Luo, Ann Meier, Sonya Rastogi, C. Matthew Snipp, John Robert Warren, and Meghan Zacher. Any opinions and conclusions expressed herein are those of the authors and do not necessarily represent the views of the U. S. Census Bureau. All results have been reviewed to ensure that no confidential information is disclosed.
Anyone may mark the “American Indian or Alaska Native” box on the census form, but official enrollment in a federally recognized tribe is required for federal legal recognition as an American Indian. Tribal enrollment numbers are much lower than census counts: for example, 900,000 versus 1.37 million in 1980 (Thornton 1997:37) and 1.8 million versus 4.1 million in 2000 (Bureau of Indian Affairs 2001; Grieco and Cassidy 2001).
We see census answers as indicators of identification or attachment because they usually are self-reports or provided by someone in the home. For ease of exposition, we write as though they are self-reports.
Here, we are comparing U.S.-born populations in 1990 with U.S.-born populations ages 10 and older in 2000 (data from Ruggles et al. 2010). See Online Resource 1 for details.
Throughout, we use “Latino” to mean someone who answered “yes” to the census question “Is this person Spanish/Hispanic/Latino?”
Although intercensal estimates were lower (Perez and Hirschman 2009), 5.2 million people marked AIAN race in the 2010 census (Humes et al. 2011).
The open-ended ancestry question in 1990 and 2000 was, “What is this person’s ancestry or ethnic origin?” The first two responses were coded. People who report AIAN ancestry but not race usually report white race and are generally socioeconomically, culturally, and geographically distinct from racially identified AIANs (Liebler 2010a; Snipp 1989). Their number is growing; it was 8.9 million in 1990, 9.1 million in 2000, and 19.8 million in 2010 (data from Ruggles et al. 2010).
Census Bureau employees making population projections in the 1990s expected the AIAN population to reach 4.3 million in 2050 rather than in 2000 (McKenney and Bennett 1994).
About 16 % of AIANs in 1990 spoke an American Indian language; only 11 % of single-race and 0.5 % of multiple-race AIANs did so in 2000 (data from Ruggles et al. 2010).
Former non-Indians have been (re)claiming American Indian race for over a century (Census Office 1894:131), but the number of people doing so has increased over time (cf. Passel 1976; Passel and Berman 1986; Nagel 1996; Sturm 2011).
There are also many blacks with AIAN heritage who could be part of this increase (cf. Naylor 2008), although optional ethnicity has not been documented among blacks.
Some Latinos dislike the instruction to report an “enrolled or principle tribe” (cf. Berkowitz 2001; Crowley 2004), and definitions of race groups were not listed on the form. However, Census Bureau procedures follow contemporary federal guidelines when recoding write-in responses into federally defined race groups.
Birth state is probably reported consistently across censuses, whereas data on current location confounds response change with interstate migration. In 1990, 63 % of the U.S. population and 72 % of AIANs lived in their birth state. In 2000, those numbers declined to 61 % and 63 %, respectively (data from Ruggles et al. 2010).
This “old Indian region” (Eschbach 1993) includes states with many AIANs before the identity-related AIAN population boom. Its 10 states (Alaska, Arizona, Minnesota, Montana, North Carolina, North Dakota, New Mexico, Oklahoma, South Dakota, and Wisconsin) held 77 % of AIANs in 1930.
To protect respondent confidentiality, the nonpublic data can be used only in a secure data enclave with explicit permission from the Census Bureau by researchers with federal security clearance. All results are reviewed and approved by the Census Bureau before dissemination. To further protect against disclosure risk, we present our results in rounded numbers.
The nonpublic data list two-race responses in 1990 and write-in responses in both years, some of which indicate American Indian as part of the response; we include these in our samples.
The deflation/inflation quotients were: 0.994 for non-Latino AIANs in 1990, 0.780 for Latino AIANs in 1990, 0.938 for non-Latino AIANs in 2000, and 1.072 for Latino AIANs in 2000.
The Census Bureau estimated an undercount rate of 4.5 % among AIANs in the 1990 census (http://www.census.gov/dmd/www/techdoc1.html) and a slight net overcount of AIANs in Census 2000 (1.16 % overcount off reservations and 0.3 % undercount on reservations) (U.S. Census Bureau 2009: Table 10-7). These estimates are based primarily on individual reinterviews but are complicated by changes in racial identification and enumerator intervention (Bentley et al. 2003; Schindler et al. 1992).
We use AIAN mortality rates. Because AIAN race is underreported on death certificates (Epstein et al. 1997), leading to artificially low mortality-rate estimates for AIANs, we may be overestimating the number of 1990 AIANs surviving to 2000. Eschbach and colleagues (1998) conducted a sensitivity test on this issue and found that it did not affect their results.
Estimates of the number of multiracial AIANs differ substantially between long-form and full-count data because of weights used with long-form data, which account for estimated nonresponse (Hefter and Gbur 2002). In the long-form data, 1,347,500 of 3,090,800 are multiracial AIANs (43.6 %). In the full-count data, 1,141,000 of 2,856,300 are multiracial AIANs (39.9 %).
We do not interpret high proportionate increases if the expected population was very small.
The positive numbers in column 6 add to about 76,000 in Table 1 but only about 56,000 in Table 2. This disparity is mainly due to large weights assigned to multiracial Latino American Indians in the long-form data.
In Latin America, many people claim a “mestizo” identity, which is conceptualized as a fusion of European, African, and indigenous heritages (Miller 2004).