Conceptualizing and operationalizing American Indian populations is challenging. Each census for decades has seen the American Indian population increase substantially more than expected, with indirect and qualitative evidence that this is due to changes in individuals’ race responses. We apply uniquely suited (but not nationally representative) linked data from the 2000 and 2010 decennial censuses (N = 3.1 million) and the 2006–2010 American Community Survey (N = 188,131) to address three research questions. First, to what extent do American Indian people have different race responses across data sources? We find considerable race response change, especially among multiple-race and/or Hispanic American Indians. Second, how are people who change responses different from or similar to those who do not? We find three sets of American Indians: those who (1) had the same race and Hispanic responses in 2000 and 2010, (2) moved between single-race and multiple-race American Indian responses, and (3) added or dropped the American Indian response, thus joining or leaving the enumerated American Indian population. People in groups (1) and (2) were relatively likely to report a tribe, live in an American Indian area, report American Indian ancestry, and live in the West. Third, how are people who join a group different from or similar to those who leave it? Multivariate models show general similarity between joiners and leavers in group (1) and in group (2). Population turnover is hidden in cross-sectional comparisons; people joining each subpopulation of American Indians are similar in number and characteristics to those who leave it.
Most demographers expect a population to increase only when births and immigrants outnumber deaths and emigrants. This straightforward balancing equation has been challenged by the American Indian1 case, which highlights another possibility: namely, population growth through changing racial identification.2 Although the American Indian population grew at a relatively conventional pace from 1890 through 1950 (see Fig. 1), a major shift occurred in 1960 when census respondents could first self-identify their race,3 resulting in a 52 % net increase in the number of enumerated American Indians (Passel 1976; Thornton 1987). This number has continued to grow remarkably. Hundreds of thousands more American Indians have been counted each census than expected based on births and migration.4 Demographers, as well as qualitative researchers investigating the phenomenon, have concluded that people have been changing their race response to include American Indian.
The difference between the number of American Indians expected each year and the number enumerated—the “error of closure”—has been followed with interest since the 1970s, but researchers have had very limited data on which to base their studies. Because censuses are cross-sectional, only net population changes could be assessed, and compositional change could be viewed only in the aggregate (Glick and Han 2015; Liebler and Ortyl 2014; Perez and Hirschman 2009). “New American Indians” were deduced to include many former whites with relatively high education and/or from areas far from large American Indian populations (Eschbach et al. 1998; Liebler and Ortyl 2014). Qualitative sociologists have interviewed former whites who began (re)identifying as American Indian (Fitzgerald 2007; Jacobs 2015; Liebler 2001; Nagel 1996; Sturm 2011). With little available evidence, the characteristics of those who left the American Indian category have not been studied. However, numerous policies and programs (e.g., allotment, boarding schools, and out-adoption) encouraged assimilation by people of American Indian descent and may have discouraged many from reporting this heritage.
Within this research tradition, we break new ground. We construct a longitudinal data set with information on approximately 3.1 million people who reported (or were reported as)5 American Indian in the 2000 census, the 2010 census, or both.6 For those who also participated in the American Community Survey (ACS) in 2006 through 2010 (N = 188,131), we have substantial supplementary data. With linked longitudinal data about individuals, we move beyond the study of net change to explore the composition of countervailing flows of people into and out of the American Indian response category.
Our research questions are threefold. First, to what extent do people change race responses to include or exclude American Indian? The answer addresses the common assumption that race responses are stable over time. Second, how are people who change race responses different from or similar to those who do not? They may have varied identities as well as distinct race-related experiences. Third, to what extent are those who join an American Indian subgroup (e.g., non-Hispanic, single-race American Indian) different from or similar to those who leave it? Differences between “joiners” and “leavers” may point to reasons why people change responses. If the groups are similar, joiners and leavers may be two views of a single group with dynamic identity experiences. Programs serving the point-in-time American Indian population can do so if joiners and leavers are similar, despite flux in identification.
We find that a large number of individuals changed their race response to include or exclude American Indian, and this is particularly true among those who also reported a Hispanic origin and/or another race. We also find that people who changed their race response to include/exclude American Indian differ from those who keep the same response across the two censuses, particularly in terms of their connections to other American Indians measured in terms of whether they reported a tribe or American Indian ancestry and whether they live in American Indian areas. Finally, we find many similarities between joiners and leavers in terms of both numbers and characteristics.
This research is important for both practical and theoretical reasons. We intend to aid analysts, policy makers, and the public in understanding American Indians in 2000 and 2010 (see Lujan 2014). We give disaggregated information about joiners, stayers, leavers, Hispanics, non-Hispanics, single-race responses, and multiple-race responses. Our multivariate analyses provide new evidence of characteristics accompanying each response pattern—evidence relevant to theories about racial identity and the social construction of race (e.g., Wimmer 2008).
American Indians: Exceptions or Forerunners?
In research on identity change and response change, part-American Indians have been shown to shift responses more often than people who report black, Asian, white, and/or Hispanic heritage (Campbell and Troyer 2007; Doyle and Kao 2007; Dusch and Meier 2012; Harris and Sim 2002; Hitlin et al. 2006; Singer and Ennis 2003; U.S. Census Bureau 1993). Are American Indians fundamentally different? In agreement with Snipp (1997), we think not. Instead, we see American Indians as representing the vanguard; other groups may well follow in their path. For example, like American Indians, people in Asian American and Hispanic American groups are increasingly forming unions across racial and ethnic lines (Wang 2012); and as greater numbers of Hispanics and Asians form successive generations in the United States (i.e., fourth-generation), racial and ethnic changes may become more common within these groups. Questions of identity and socially defined group boundaries are likely to expand for many racial/ethnic groups in coming years. How each group is measured will also continue to affect the social construction of race and group boundaries (e.g., Humes and Hogan 2015; Omi and Winant 1994; Prewitt 2013; Snipp 1997). Pacific Islanders and multiple-race respondents from all race groups already show a high level of race response change across the 2000 to 2010 period (Liebler et al. 2014).
At the same time, American Indians are not the same as other racial/ethnic groups in the United States. What it means to be American Indian is complicated by the existence of tribal governments, indigenous homelands, tribal enrollment blood quantum requirements, and political relationships with the federal government. A person deciding whether to mark American Indian as his or her race has extra dimensions to consider: “American Indian” includes sometimes knotty political and/or legal statuses (and related contested identities) that are not at issue in nonindigenous groups (see Robertson 2013; U.S. Census Bureau 2008:v).
In this complex milieu, millions of people with American Indian heritage report it to the Census Bureau in an ancestry question but not in the race question (see Fig. 1). Over the centuries, many policies and practices have strongly urged American Indian assimilation, fostering an atmosphere of stigma that may still affect some. Also, as Senator Elizabeth Warren’s experiences illustrate, part-whites who claim indigenous heritage can be heavily criticized for seeming to try to benefit from minority status (e.g., Seelye 2012). For part-black American Indians, twin pressures discourage an American Indian response: part-blacks are often seen as “just” black (Davis 2005; Khanna and Johnson 2010), and part–American Indians are much less often seen as American Indian (Gullickson and Morning 2011; Snipp 1989, 2003).
The Hispanic American Indian category contains a variety of people, including at least four groups. Some Hispanic American Indians have, for example, one Hispanic parent and one American Indian parent. Some have a mestizo identity and, in an effort to convey this identity on the census form, mark a combination of Hispanic, American Indian, white, and perhaps black.7 Some South and Central American indigenous people8 mark the “American Indian or Alaska Native” check box on the race question; most of these people were reported Hispanic.9 And some South or Central American indigenous people reported their tribe on the form without marking the American Indian check box. These responses10 were coded as American Indian in post-enumeration processing, in accordance with the federal definition of American Indian (Office of Management and Budget 1997).11
To What Extent Do People Change Responses to Include or Exclude American Indian?
Do people change their racial identification to/from American Indian? How common is this? Based on the demographic balancing equation, Passel (1976) calculated the expected net increase from 1960 and 1970 to be 202,000, but the 1970 census count of American Indians was 67,000 higher than this. Passel suggested that part of this error of closure was due to changes in racial identification from white in 1960 to American Indian in 1970. As illustrated in Fig. 1, large errors of closure continued to appear in subsequent decades: 366,000 between 1970 and 1980 (Passel 1997; Passel and Berman 1986), 181,000 between 1980 and 1990 (Harris 1994), and just over 1 million between 1990 and 2000 (Liebler and Ortyl 2014). These studies and others (Eschbach 1993, 1995; Eschbach et al. 1998; Harris 1994) have pointed to changes in racial classification as a primary factor causing these errors of closure.
Researchers using smaller longitudinal data sets have found lower rates of race response consistency among people giving American Indian responses compared with those reporting white, black, or an Asian group (Dusch and Meier 2012; Singer and Ennis 2003; U.S. Census Bureau 1993). For example, two-fifths of 1990 census American Indians reported a different race in the post-census reinterview (U.S. Census Bureau 1993). More than one-quarter of non-Hispanic American Indians and more than three-quarters of Hispanic American Indians gave a different race/Hispanic response between the 2000 census and the 2000 CPS (del Pinal and Schmidley 2005). One-third of single-race American Indian youth in 1994/1995 reported a different race six to eight years later (Doyle and Kao 2007).
Most prior researchers could not study multiple-race responses and did not disaggregate by Hispanic origin (Eschbach et al. 1998; Harris 1994; Passel 1976, 1997; Passel and Berman 1986). Liebler and Ortyl (2014) are the exception. Using cross-sections of the 1990 and 2000 censuses, they deduced that many newly identified Hispanic American Indians were relatively young. They also demonstrated that some 1990 single-race American Indians must have reported multiple races in 2000.
Not all racial reclassification is due to identity changes. Response change can also be a result of a different person within the household filling out the form, individuals making a mistake when filling out their form, individuals purposely misreporting their race, or an outsider replying for a nonresponsive household (Compton et al. 2012; Porter et al. 2015; Rockquemore and Brunsma 2002). Most of these issues could affect all groups, so the particularly high level of response change among American Indians is probably not entirely driven by these factors.
Other measurement issues may disproportionately affect American Indians because of the particularly complex set of forces (described earlier) that construct who is socially seen as American Indian. A survey instrument may not be fully able to capture complex racial identity experiences. Relatedly, a survey response category that draws responses from a heterogeneous set of people may not fully reflect the identities of those who list it as their identification, potentially resulting in response fluidity.
How Are People Who Change Responses Different From or Similar to Those Who Do Not?
What are the characteristics of people who change their race response to/from American Indian? Are they different from those who are consistent in identification and, if so, how? Prior quantitative and qualitative researchers (e.g., Eschbach et al. 1998; Sturm 2011) have provided a few answers to these questions.
Although known to be mutable, racial identity is generally thought to be central to self-conception. A consistent race and Hispanic-origin response (whatever the details)12 may indicate a relatively strong attachment to the group. A person who gives the same race and Hispanic responses in 2000 and 2010—a “stayer” in our study—may have a relatively strong sense of American Indian identity and may experience the social world accordingly. Indigenous connections to traditional or legal homelands (Liebler 2010b) and “thick ties” to race groups (Cornell and Hartmann 2007) suggest that stayers might more often report a tribal affiliation, live in an American Indian area, and report American Indian ancestry.
New Personal Circumstances
People who change their response may have experienced changes in circumstance that influenced self-conception or others’ perception of them (Cooley 1902). Given the impact of local area characteristics on race responses (Eschbach 1993; Kana’iaupuni and Liebler 2005; Liebler 2010b; Xie and Goyette 1997), we anticipate that residential mobility spurs response instability, especially if the residential locations have different racial compositions or culturally relevant meanings. Children in 2000 who moved out of their parents’ homes by 2010 may report a different race for themselves than was previously recorded. Other life changes could also potentially affect race responses, including marriage, divorce, discrimination experiences, or new information about family heritage.
People who do not read English well or who are less educated may find the census questions and federal definitions challenging or confusing, potentially resulting in response change from one census to the other. Noncitizens13 and new immigrants may be unsure of the social meaning of each race group in the U.S. context and have shifting understandings as they spend more time in the United States. Some people—such as newly immigrated indigenous Central and South Americans—might face more than one of these barriers.
Variation in Source of Information
Observed response changes could also reflect a difference in opinion about what race(s) best describe a person (Song 2003). Enumerators are common in American Indian areas (Bates 2008), and the mail-out form is usually filled out by one member of the household for everyone in the household (Sweet 1994). The multiple facets of race (e.g., observed versus self-identified race) do not necessarily align (Harris and Sim 2002; Porter et al. 2015), and thus variation in the source of the information about a person’s race(s) may cause variation in the content of that information.
How Are Joiners Different From or Similar to Leavers?
Our third research task is to understand differences and similarities between those who join and those who leave American Indian subgroups.
People who join a subgroup might be distinct from those who leave the same subgroup. For example, people who reported single-race American Indian in 2010 but not in 2000 might have newly heightened American Indian identity because they recently moved to a densely American Indian area or recently married an American Indian (Eschbach 1993; Kana’iaupuni and Liebler 2005; Lieberson and Waters 1993; Loveman and Muniz 2007). A different change in local or family context could suppress an American Indian race response.
Joiners and leavers might be similar if there are certain types of people who are likely to change race responses.
Identity Flexibility and White Privilege
Many whites in the United States experience their European ethnicities as relatively symbolic or optional,14 causing cross-time fluctuations in the number reporting each European ancestry group (Gans 1979; Hout and Goldstein 1994; Lieberson and Waters 1993; Waters 1990). After centuries of mixing, many people who identify and live as white have American Indian ancestors (Liebler 2010a; Snipp 1989; Waters 1990). If some of these people turn a symbolic or an optional ethnicity lens to their non-European ancestors, they may decide to mark American Indian race (at least for a time). Eschbach et al. (1998) and Liebler and Ortyl (2014) found that the “new” American Indians in previous censuses have an education profile similar to that of whites.
Self-Conception Mismatched With Questionnaire
Translating complex identities into answers to fixed-choice questions can be a challenge. If a person changes her opinion about the best way to convey her self-conception on a census form, this could cause response change. For example, the Census Bureau does not require tribal enrollment in order to be considered American Indian, but other federal agencies (e.g., the Indian Health Service) do require enrollment. A non-enrolled but self-identified American Indian might assume that she does not qualify as American Indian on the census but later learn that she does fit the census definition and so change her response. Also, someone with a mestizo identity might be unsure of whether to mark white, American Indian, and Hispanic, or to mark Hispanic and Some Other Race (or another combination of responses), and might change her opinion on this issue over time.
Multiple Salient Heritages
Prior research about people with more than one salient racial heritage shows that many have dynamic racial identities and relatively nonstable patterns of race response (e.g., Harris and Sim 2002; Rockquemore and Brunsma 2002, 2008; Root 1996). This may be especially true of people whose parents are of different races if they have relatively extensive experience with the race groups represented by their parents. Consistent with this, we expect some people to have fluid race reports that reflect fluid identities.
Data, Case Selection, Methods, and Measures
The U.S. Census Bureau Center for Administrative Records Research and Applications (CARRA) created the restricted-access data set we used for this study. CARRA used probabilistic record linkage techniques (Fellegi and Sunter 1969) and personal information (name, sex, date of birth, and address; see Wagner and Layne 2014) to strictly assign15 a unique identifier—a protected identification key (PIK)—to data sources including the census and ACS.16 The data were anonymized and can be used only for statistical and research purposes. We used the PIKs to link individuals’ census and ACS records into a longitudinal data set. Linking people in the 2000 census with their own responses in the 2010 census gave a data set with about 200 million people.
Our linked data do not include all people. Some people do not receive a unique PIK (e.g., if their personal information was incomplete or not unique). PIK assignment rates are lower for Hispanics and nonwhites than for non-Hispanic whites (Bond et al. 2014). A person counted in 2000 who died or left the country by 2010 could not be in these data, nor could someone who was not yet born or did not live in the United States in 2000. Some individuals who were present were not enumerated in one of the censuses (Lujan 2014; Mule 2012; U.S. Census Bureau 2003) and were therefore also left out of our data.
To select cases for this study from the linked decennial census data, we began with all people whose race response included American Indian in the 2000 census and/or the 2010 census: 4,140,582 people. We made a series of exclusions (described in Table 1) to further reduce the chances of false links and to constrain the extent of response changes resulting from enumeration issues. After case selection, our linked decennial data include 3,059,818 people whose race report included American Indian in the 2000 census and/or the 2010 census. The information we have about these 3.1 million people is limited to the few questions that were on the decennial census short questionnaires. We supplement this information by including ACS responses for 188,131 people (after case selection) who participated in the ACS between 2006 and 2010.
The people in our study are all the people in the linked data who fit the case selection criteria. Results are not weighted estimates, and our data are not nationally representative.17 Compared with the full census counts of American Indians in 2000 and 2010 (shown in Table A1 in Online Resource 1), people in our decennial linked data were more often non-Hispanic, female, reported a tribe, and/or lived in an American Indian area (as defined in Table 2). Because Hispanics have a relatively low response rate to the race question (Ríos et al. 2014), they were disproportionately excluded. Females have slightly higher response rates than males (Rastogi et al. 2014), a difference that is magnified when data sources are linked. Using an address when assigning a PIK favors people who move less often, thus reducing the PIK rates for males ages 20 to 50 (Rastogi and O’Hara 2012).
To improve knowledge and to coincide with common methods of operationalizing “American Indian,” we divide people into four subgroups for analysis: (S1) non-Hispanic, single-race American Indian, (S2) non-Hispanic, multiple-race American Indian, (S3) Hispanic, single-race American Indian, and (S4) Hispanic, multiple-race American Indian. Subgroups describe a response at a point in time. A person could be in S1 in 2000 and S2 in 2010, for example, or could be in none of the subgroups in 2000 or in 2010 (e.g., by reporting single-race white).
We use logistic and multinomial logistic regression models to address our second and third research questions, which focus on differences among joiners, stayers, and leavers. Dependent variables for all models reflect race and Hispanic responses in the censuses of 2000 and 2010 only. We explain the dependent variable for each model when introducing the results of the model. So that we can include measures of education, marital status, and labor force participation, we include only people ages 25 and older in the multivariate models.18 Descriptive statistics include people of all ages (except as noted).
We include measures of individuals’ demographic, socioeconomic, and geographic characteristics as independent variables in our analysis. See Table 2 for details about coding.
To What Extent Do People Change Responses to Include or Exclude American Indian?
Race responses are not necessarily stable across a person’s lifetime: a high proportion of people in our data changed their racial identification to/from American Indian over the 2000 to 2010 period, as shown in the off-diagonal cells in Table 3. Less than one-third of ever–American Indian people in the data had the same race/Hispanic response in 2000 and 2010. The remainder changed their race and/or Hispanic response across the decade. In other words, although their responses may reflect their identity at the time, it is not safe to assume that the race and Hispanic-origin responses of people in our study will be the same from one census to another. Response change might affect any point-in-time measure of race; this has been shown to affect multiple race responses, Pacific Islander responses, and race responses of those who report Hispanic origins (del Pinal and Schmidley 2005; Harris and Sim 2002; Liebler et al. 2014).
Four patterns of response change in Table 3 stand out. First, a large proportion of people in our data (45 %) moved between a single-race response and a multiple-race response. This type of response move is consistent with prior research on identity flux in multiracial people (Harris and Sim 2002; Rockquemore and Brunsma 2008; Root 1996).
Second, some people (20 % in our data) changed their response from one single-race response to another. This was particularly common among Hispanic American Indians; of those who were consistently identified as Hispanic, 52 % changed their race response between American Indian and either white or Some Other Race. Note that most people in our decennial linked data (94 %) reported Hispanic/non-Hispanic consistently across censuses.
Third, the number of people joining and leaving each subgroup is similar in size. For example, 1,046,000 people in our data reported non-Hispanic, single-race American Indian in 2000. Of these, 158,000 were reported as non-Hispanic, single-race white in 2010. They were replaced in the non-Hispanic, single-race American Indian category in 2010 by 173,000 people who were reported as non-Hispanic, single-race white in 2000. Without longitudinal data, the 15,000-person difference would be the only evidence of these large, countervailing flows.
Fourth, each American Indian subgroup has a low proportion who stayed in the group. Of people in our data who ever reported non-Hispanic, single-race American Indian, 47 % joined or left this group between the 2000 and the 2010 censuses.19 Among Hispanic and/or multiple-race American Indians in our data, more than 85 % joined or left over the period.20 If this pattern holds in other data, the total number of people reported as American Indian at one point in time reflects only a fraction of the number of people ever reported as American Indian.
In a closer look at these 3.1 million people, we disaggregate each subgroup’s joiners, leavers, and stayers by age and sex (see Table A2 in Online Resource 1). Hispanic, multiple-race American Indians are a young group, and many children moved into or out of this category before their teenage years. Maybe this multifaceted response reflects an early stage of identity development (Erickson 1968), or perhaps a parent reported the child’s race in 2000 but the (former) child self-reported in 2010. More broadly, we see that response changes are not the sole experience of a certain age group or sex; a wide variety of people are involved in response change.
To get a third measure of reported race, we next incorporate the race responses given by/for those who also participated in the ACS; see Fig. 2 and Table A3 (Online Resource 1). Some race response change occurred even among people we label “stayers” using census responses.21 For example, all people in rows 1, 4, and 21 of Fig. 2 (53,495 people) reported single-race American Indian in both censuses, but only those in row 1 (45,869 people) also reported this in the ACS.22 Again, we see that a point-in-time measure of race does not necessarily match measures at other points in time; race responses are subject to change.
For each of the 24 possible race response patterns in our ACS-decennial linked data, we show the proportion who (1) reported Hispanic origins, (2) lived in an American Indian area, (3) reported a tribe, and (4) were children. Hispanic American Indians predominate among those who changed from a single-race American Indian response to a non–American Indian response (rows 9 and 10) or vice versa (rows 15 and 16). Consistently reporting American Indian (including adding or dropping other race responses; rows 1–8) is associated with living in an American Indian area and reporting a tribe.23
How Are People Who Change Responses Different From or Similar to Those Who Do Not?
Our second research question asks how stayers (people who gave the same race and Hispanic reports in both censuses) compare with people whose census responses changed. Table 4 and Table A4 (Online Resource 1) show characteristics of leavers, stayers, and joiners in each subgroup of American Indians within the ACS-decennial linked data. These tables show, for example, that people who reported multiple-races including American Indian (S2 and S4) tended to have more education than those who ever reported single-race American Indian (S1 and S3).
Prior research on joiners (e.g., Eschbach et al. 1998; Sturm 2011) led us to expect some differences between those who changed responses and those who did not. We find evidence of some differences between stayers and changers. In each subgroup, residential migration was slightly more common among leavers and joiners than stayers. Those who left a subgroup tended to have left an American Indian area, and those who joined a subgroup tended to have moved to one. This is consistent with prior research relating homelands and indigenous identities (Eschbach 1995; Kana’iaupuni and Liebler 2005; Liebler 2010b; Memmott and Long 2002).
Hispanic and non-Hispanic American Indians show differences in terms of English language proficiency, education level, and citizenship status. Hispanic American Indians with low English proficiency often change responses, but the few non-Hispanic American Indians who are not proficient in English are more often S1 stayers. Similarly, in the Hispanic American Indian subgroups (S3 and S4), low education is associated with response change; in the non-Hispanic S1 group, lower education is associated with response stability. In S3 and S4, Hispanic foreign-born noncitizens often had different responses from one census to another, but no pattern is evident among the few non-Hispanic noncitizen American Indians. In sum, the four subgroups hold different types of people and should be studied separately when possible.
Besides identifying the response changers, statistics in Table 4 describe characteristics of stayers. When stayers differ from joiners and leavers, cross-sectional numbers give inaccurate estimates of stayer characteristics. For example, compared with S1 joiners and leavers, relatively few adults who stayed in S1 were married. Thus, a cross-sectional point-in-time view would show a higher marriage rate for non-Hispanic, single-race American Indians in 2010 than was true of those who had this response in both 2000 and 2010.
To learn whether differences between stayers and changers are statistically significant, we apply multivariate models in two ways. First, we use multinomial logistic regression models24 (shown in Table 5) to predict joining or leaving each subgroup, relative to staying in that group. Second, we compare stayers with people making various common response moves. In Table 6, we compare the characteristics of non-Hispanic American Indians who stayed in S1 or S2 with those who made one of three response moves: (1) between single-race and multiple-race American Indian, (2) between single-race American Indian and single-race white, and (3) between multiple-race American Indian and single-race white. We present a similar model about Hispanic American Indians in Table 7, comparing S3 and S4 stayers with those who changed (1) between single-race or multiple-race American Indian and single-race white, and (2) between single-race or multiple-race American Indian and single-race Some Other Race.
Tables 5, 6, and 7 show substantial and significant differences between stayers and those who changed responses across the decade. Measures related to nativity and group connections (noncitizen, English skills, race/ancestry reports in the ACS, tribe reported, and living in an American Indian area) are especially able to distinguish stayers from others.
Our independent variables are particularly effective at parsing non-Hispanic American Indian stayers from joiners/leavers, as shown by the relatively high values of r2 (.39 for S1 in Table 5, and .45 in Table 6). Connections to land and tribe are powerfully predictive of race response patterns among non-Hispanic American Indians, including people who switched between multiple-race and single-race American Indian race responses (but were consistently non-Hispanic). They were relatively likely to report a tribe, report American Indian ancestry, and/or live in an American Indian area compared with people who left the American Indian group entirely and also compared with S2 stayers.
Measured characteristics are somewhat less effective at distinguishing Hispanic American Indian joiner/leavers from stayers (as seen in the r2 values in Table 7 and also models about S3 and S4 in Table 5). Hispanic American Indian stayers (S3 and S4 in Table 7) were much more likely to report American Indian ancestry, report a tribe, or live in an American Indian area than those who changed responses to/from Hispanic, single-race white or Hispanic, single-race Some Other Race.
In sum, people in our data who changed their race response between 2000 and 2010 were significantly and substantively different than those who did not. Changers who gave a non-American Indian response in 2000 or 2010 (single-race white or Some Other Race) were also notably different from those who consistently reported American Indian (either as stayers or by moving between single- and multiple-race American Indian responses). Like non-Hispanic stayers (S1 and S2), people who moved between non-Hispanic, multiple-race and single-race American Indian (moved between S1 and S2) were more likely to report a tribe, live in an American Indian area, and report American Indian ancestry than were people who changed to/from non-Hispanic white. They seem to have “thicker ties” to American Indians (Cornell and Hartman 2007) than those who left the American Indian group entirely.
How Are Joiners Different From or Similar to Leavers?
People who joined a particular subgroup of American Indians over the decade may have had different experiences than those who left the same subgroup. In prior research on joiners (e.g., Liebler 2001; Sturm 2011), some joiners have spoken of an identity awakening spurred by relocation or new family history information. Prior researchers have not been able to study leavers.
Joiners and leavers appear very similar to each other in Tables 4 and 5. To explore this more deeply, we next show disaggregations of each group of response changers—for example, those moving between S2 and S1 separate from those in S2 who changed to/from non-Hispanic white—and we present descriptive statistics about each group in Tables A5–A9 in Online Resource 1. Table 8 shows 12 logistic regression models predicting joining a subgroup rather than leaving it (stayers are excluded).
Models in Table 8 show that joiners were significantly different from leavers in some ways in all groups, with cross-group variation in the specifics of these differences. At the same time, model fit for all 12 models in Table 8 is poor, with r2 ranging from .02 to .07. This means that within a particular response pattern, those who move in one direction (e.g., from S3 to Hispanic, single-race white) are very similar to those who move in the opposite direction (e.g., from Hispanic, single-race white to S3), at least with respect to the characteristics measured here. This model fit is especially poor in comparison with our other analyses using these same variables to distinguish between other types of response change.
Similarities between joiners and leavers could indicate that the census snapshots caught them at different points in a generally dynamic experience. Prior research outlined earlier suggests that joiners and leavers who otherwise report non-Hispanic white would be similar to one another. Qualitative researchers have found that people with fluid identities give multiple-race responses sometimes and single-race responses at other times (Rockquemore and Brunsma 2008; Root 1996). Based on our models’ inability to distinguish joiners from leavers, we conclude that these scenarios are plausible and bear further study.
Discussion and Conclusion
Researchers have known for decades that the American Indian population grows not only through births, deaths, and migration but also through changes in how people report their race on the census form (e.g., Liebler and Ortyl 2014; Passel 1976, 1997). Until now, it has not been possible to learn characteristics of those who joined the population, whether anyone left the population by changing race responses, and/or the level of response stability. Our research has closed these gaps. We used high-quality, large-scale linked data to study race and Hispanic response changes among people who reported American Indian in the 2000 or 2010 census or both. We addressed three questions. To what extent do people join or leave subgroups of American Indians? How are joiners and leavers different from or similar to stayers? And how are joiners different from or similar to leavers?
We found substantial changes in race responses in our data. Almost one-half of the non-Hispanic, single-race American Indians in 2000 left this subgroup and were replaced by others by 2010. A much higher fraction of Hispanic and multiple-race American Indians left and were replaced; in these groups, response change is vastly more common than response stability. Similarly high levels of response change have been found among other multiple-race groups and among Pacific Islanders (Liebler et al. 2014).
People in our data who changed race responses (joiners and leavers) had different characteristics than those who kept the same response across two measures a decade apart (stayers). Stayers were generally distinct from response changers in terms of measured connections to other American Indians, including tribe response, ancestry response, and living in an American Indian area (some changers also have these attributes). This suggests that stayers have had different race-related life experiences than those who changed responses.
By further disaggregating joiners and leavers into subgroups, we revealed multiple dynamic processes involving racial fluidity. For example, those who changed between a single-race white response and an American Indian response had a different demographic profile than those who moved between multiple-race and single-race American Indian responses. Our results support the decision to separate investigations of formerly white American Indians (e.g., Fitzgerald 2007; Sturm 2011) from studies of people who consistently report American Indian but sometimes report an additional race or races (e.g., Liebler 2001).
We found substantial similarities in the number and characteristics of people who made a particular response move (e.g., from Hispanic, single-race American Indian to Hispanic, single-race white) and others who made the inverse move. We used multivariate models to distinguish characteristics of people in inverse groups. The models have very poor fit, perhaps indicating that joiners and leavers are engaged in similar identity processes and simply are captured in our data at different points in the process. This complicates the search for reasons that people change race responses; social movements like Red Power (Nagel 1996), for example, are thought to cause mostly unidirectional response change (i.e., joining) and thus cannot give a complete explanation of our findings. Programs serving American Indians may not notice this large-scale churning of individuals into and out of the populations they serve because of similarities between those who join and those who leave the population of self-defined American Indians.
Our research has important caveats and limitations. First, response changes do not necessarily mean identity changes. Some are due to false links, differences in post-enumeration processing across the two censuses, differences in opinion about what would be a “correct” response, or mistakenly marking the wrong box(es) when filling out the form. Second, we are not able to study those with an American Indian identity that was not reported in the census race question. Third, our results overrepresent response stability in two ways. First, because of case selection and limitations of linked data, our data include relatively many non-Hispanics, tribe reporters, and people in American Indian areas—characteristics shared by people with stable responses. Second, we focus on only two measures of a person’s race over an entire decade even though further response changes are possible (and evident in the ACS data).
Our results have theoretical, practical, and policy-related implications. We contribute to conceptual understandings of racial identity and racial fluidity by identifying characteristics of three groups of American Indians who seem to have distinct identity and response fluidity experiences:
Stayers: People who keep the same American Indian race/Hispanic response.
Joiners/leavers who stay in the American Indian category: People who sometimes report single-race American Indian and other times report multiple-race American Indian.
Joiners/leavers who enter/exit the American Indian category: People who add or drop the American Indian race response entirely.
People in groups 1 and 2 show substantial cultural connections to American Indians: many live in an American Indian area and/or report a tribe. Further research is necessary to understand whether the joiners and leavers in groups 2 and 3 are captured at different points in a common identity experience; this would explain why their characteristics are so similar.
Our results show diversity within the “American Indian” population. Many American Indians already know of this diversity, and our work provides important validation and documentation. Our work can also help researchers, policy makers, and tribal leaders more effectively interpret census data about American Indians. For example, because we find similarities between people who consistently report non-Hispanic American Indian (even if they add/drop other race responses), researchers should consider including multiple-race American Indians in their analyses. Grouping multiple-race American Indians instead with other multiple-race respondents (e.g., black-whites) may result in unnecessarily separating similar individuals (single-race and multiple-race non-Hispanic American Indians) into two groups. More generally, researchers should also be careful when making claims about American Indians as a whole, given the diversity that exists within this group.
The American Indian case may show the future of race response change for people of many racial and ethnic groups (Liebler et al. 2014). We expect response change to increase for all groups as more unions are formed across racial and ethnic boundaries and as greater proportions of other groups (e.g., Hispanics and Asians) become grandchildren and great grandchildren of immigrants. In these situations, there may be more conversations and social processes defining what makes a person a “real” Hispanic or Asian, and response change could follow.
Although efforts to improve race and ethnicity questions continue (see Compton et al. 2012; Humes and Hogan 2015; Prewitt 2013), these are complex and personal social constructs, and it is possible that no point-in-time measure of race or ethnicity will be able to meet the goal of “categoriz[ing] individuals into the same groups over a long period of time” (Humes and Hogan 2009:112). Researchers designing questions to measure race and ethnicity should consider the possibility of response change, and multiple measures of race should be incorporated into data collection and analysis whenever possible. Analysts from all fields would benefit from conceptualizing and operationalizing a person’s race as having a past, present, and future (as is done for marital status, work, and residence, for example), rather than acting as if race were an unchanging trait. The dynamics of race exposed in this research lend an unfamiliar dimension of complexity to the study of groups such as American Indians, but this should not deter researchers from engaging the issue (see Espey et al. 2014). Rather, with new knowledge about the extent of these dynamics, we can employ repurposed strategies and theories to gain more realistic insights into our complex social world.
This article is released to inform interested parties of research and to encourage discussion. The views expressed are those of the authors and not necessarily those of the U.S. Census Bureau. At various stages of this research, we have benefited from thoughtful comments from many people, including C. Matthew Snipp, Amy O’Hara, James Noon, Leticia Fernandez, Sharon Ennis, Julia Rivera Drew, Catherine Fitch, Liying Luo, Caren Arbeit, Susan Mason, J. Trent Alexander, Jenifer Bratter, and Mary Campbell. We also thank the American Indian and Indigenous Studies Workshop and the Inequality and Methods Workshop, both at the University of Minnesota, for sponsoring helpful discussions of this research. The University of Minnesota’s Minnesota Population Center provided important support for the first author through programs made possible by an NIH Center Grant (R24HD041023).
We use “American Indian” to describe a person who reported “American Indian or Alaska Native” (AIAN) in the race question on the census form. Unless specified, we are referencing the entire group regardless of whether other races were also reported and regardless of Hispanic origins. Our study includes those reported as American Indian in the race question in the 2000 and/or the 2010 census. We do not assume that they always have or always will report American Indian (or be reported as this). We use the person’s time-specific race response to describe him/her. For instance, if someone reported American Indian in 2000 and white in 2010, we refer to him/her as non–American Indian in 2010.
We use “racial identification” and “race response” to mean the response given on the decennial census form. This is not necessarily the same as a person’s racial identity, although they are probably related.
We apply the terms “race” and “Hispanic origin” in congruence with the federal statistical definitions used to collect the data (Office of Management and Budget 1997). Each questionnaire used here included one question about Hispanic origin (one response allowed) and one question about race (multiple responses invited).
Responses might not be self-reports, although we simplify our prose by writing as though they are self-reports. Using case selection, we ensure that these are self-reports or reports by someone else in the household (probably the householder or his/her spouse; Sweet 1994), although enumerators visited some homes and could influence responses.
The census counted 4.1 million American Indians in 2000 (Grieco and Cassidy 2001) and 5.2 million in 2010 (Humes et al. 2011). For reasons described later herein, our linked data are not nationally representative.
Many Latin American countries recently legally recognized indigenous groups (Telles and Bailey 2013). There are about 400 indigenous groups in Latin America and the Caribbean (Montenegro and Stephens 2006), and more than 40 million of 500 million Latin Americans self-identify as indigenous (Telles and Bailey 2013). Many are bound to their indigenous heritage through language and political, social, and cultural ties (Gonzalez 1994; Montenegro and Stephens 2006). Although some live on (often remote) tribal lands or rural areas, an increasing share live in urban areas (Dahl and Jensen 2002; Del Popolo et al. 2007; Roldán Ortiga 2004). As a whole, they are relatively poor and have worse social and health outcomes (Kearney 2000; Psacharopoulos and Patrinos 1994).
Among American Indians whose only reported tribe was from South or Central America, 86 % reported Hispanic in the 2000 census, and 94 % did so in the 2010 census.
People who listed a tribe anywhere in the race question were coded as American Indian race in post-enumeration processing. Of 244,761 people reporting a Central or South American tribe in the 2010 census, only 38 % marked the “American Indian or Alaska Native” check box. Of the 3,195,538 who reported a North American tribe, 84 % marked the box. Our linked decennial sample has higher proportions who marked the check box: 46 % and 98 %, respectively.
The federal definition of American Indian or Alaska Native is “A person having origins in any of the original peoples of North and South America (including Central America), and who maintains tribal affiliation or community attachment” (Office of Management and Budget 1997:58,789).
The decision of whether to give a single-race response or to report multiple races is based on heritage and also on political and legal considerations, community connections, and other factors (see Liebler 2001; Robertson 2013).
Foreign-born individuals who have gone through the citizenship process have had considerable experience with the U.S. system and may have substantial understanding of U.S. social practices.
Relatedly, socioeconomic privilege could make a race response change from white to minority seem especially costless because the person is buffered from the harshest costs of color.
An estimated 1 % of links were to the wrong person (Layne et al. 2014).
The decennial data has not been through data perturbation. We ensure disclosure avoidance using disclosure review. The ACS data has undergone data perturbation, causing some response mismatch between the decennial and ACS data points.
We do not use ACS weights. Because they account for factors such as survey nonresponse and sampling strategies but do not adjust for record linkage and case selection, they would not make the data representative.
Alternate versions of all multivariate models with fewer independent variables but including people of all ages are available on request. Also, descriptive statistics for only people ages 25 and older are available on request.
A total of 1,365,025 people in our decennial linked data reported non-Hispanic, single-race American Indian in 2000 or 2010 (=1,045,627 + 1,042,724 – 723,326). Of these, 723,326 gave the same report both times. Thus, 723,326 / 1,365,025 = 53 % of people in S1 were stayers.
Stayers represented 13 % of the people in S2, 11 % of those in S3, 9 % of those in S4.
We use the ACS-decennial linked data in Fig. 2 and Tables 4–8. Throughout the article, we define joiners, stayers, and leavers using only information from the decennial censuses of 2000 and 2010. We take into account ACS race responses in our multivariate analyses.
A small proportion of ACS race response changes may be due to data perturbation and not the respondent.
People in rows 1–8 and 21–24 (American Indian in both censuses) can have a recorded “enrolled or principal tribe” in 2000 and/or in 2010, while those in rows 9–20 (American Indian in one census) can have a recorded tribe in only one census. We code any write-in response as a “tribe report.”
Relative risks (exp(β)) that are below 1.0 show a negative relationship. For example, in Table 5, those who were never married were significantly less likely to leave S1 than they were to stay in this subgroup (exp(β) = 0.75). Relative risks above 1.0 show the opposite: people who did not report a tribe in either census were more than five times as likely (exp(β) = 5.63) to be S1 leavers than to be S1 stayers.