The analysis of ethnic populations with census data generally presumes substantial stability in the identification of members over time. At the limit, this assumption is tantamount to the demographic change equation. Without such an assumption, it is virtually impossible to make sense of trends in such important indicators as residential segregation and educational attainment. That is, if membership is unstable, so that many former members no longer indicate membership or previous non-members now elect to belong, then what appears at first blush to be a change in the objective situation of the group, such as a decline in residential segregation, can be the product of shifts in the ways its membership is identifiable.

Although a long-standing sociological tradition views racial and ethnic memberships as ascriptive, and therefore stable, life-long, traits, we know that the reports of membership in some ethnic populations have shifted over time, in some cases according to a consistent trend, in others according to idiosyncratic fluctuations. Thus, the numbers of Americans claiming to be American Indian on the census race question rose sharply after 1960, as individuals who had once described their race in some other way changed their self-designation (Eschbach et al. 1998; Nagel 1995). It is plausible to hypothesize that this rise has occurred because of the growing acceptability of indigenous origins, which has encouraged many Americans with part Indian ancestry to claim this identity. By contrast, the numbers of claimants of various European origins on the census ancestry question have exhibited considerable flux, prompted partly by the examples listed just below the write-in question (see Farley 1991; Lieberson and Waters 1993).

In this paper, we examine the issue of stability for U.S.-born Mexican Americans, by far the largest of the ethnic groups growing as a result of contemporary immigration. In the 2000 Census, the identification of Mexicans and other Hispanic groups was made problematic by a wording shift in the Hispanic-origin question (del Pinal 2004). However, we show that an exit of some members from the Mexican group was apparent even in the 1990 Census and was accentuated in 2000. We show also that this exit is selective, so that a comparison of the characteristics of U.S.-born members over time presumably is affected by changing patterns of identification with the group. However, because the exit takes place into two distinct kinds of identity categories, one of which is pan-ethnic Hispanic while the other reflects a more mainstream assimilation, the impact of exit on the characteristics of the identifiable Mexican–American group is complex.

Identity Shift

One force that can lead to identity shifts for large numbers of individuals is assimilation (Alba and Nee 2003; Gordon 1964). It has long been posited that, in its more developed phases, assimilation brings a weakening of the secure anchoring of ethnic identification in ancestral origins (Alba 1990; Waters 1990). Substantial evidence shows that, for the descendants of European ethnics, identities have been affected by assimilation and especially by the mixed ancestry that results from intermarriage. Thus, the ethnic identities of whites evidence considerable lability, so that their specific manifestations shift over time and from one context to another. One signal of this weakening may be flux over time in the way individuals respond to questions about ethnicity or in their responses to different versions of an ancestry question (Lieberson and Waters 1993).

In principle, assimilation can lead to various outcomes in terms of identity and group membership. As Portes and Zhou (1993) point out in their exposition of the concept of “segmented assimilation,” it matters what societal sectors individuals and groups are assimilating to. One possibility is entry into the mainstream society, which is dominated by white Americans with European ancestries. In the case of Mexicans, who come from a country that prides itself upon mestizaje and who have both in the past and present displayed a relatively high rate of marriage with Anglos (Bean and Stevens 2003; Spickard 1989), there exists a large population of individuals who can see themselves as having mixed ethnic and racial origins. In this connection, it is important to recognize also that Mexicans are a racially diverse population, which ranges from those who have an indigenous appearance in North American eyes to others with a European appearance (Murguia and Forman 2003). Racial appearance may well be linked to the treatment Mexican Americans receive in American society (Lopez and Stanton-Salazar 2001; Murguia and Telles 1996). The research of Alba and Logan (e.g., Alba et al. 2000; see also Massey and Denton 1992) on locational attainment shows that Mexican Americans who declare themselves to be “white” on the census reside in somewhat better neighborhoods (as indexed, say, by the average income of their residents) than otherwise comparable Mexican Americans who classify themselves racially as “other.” The combination of a European appearance and a family background of intermarriage to Anglo Americans could encourage some individuals to identify with their more prestigious European ancestral origins than with their Mexican ones (Duncan and Trejo 2005).

This tendency could be strengthened by the economic impacts of immigration from Mexico. Economists have noted that the deepest impacts of newly arrived immigrants are on the earnings of longer resident members of the same groups (Borjas 1999). This suggests that, in areas of high immigration, Mexican Americans, even those who are U.S. born, could have reasons to distinguish themselves from recently immigrated Mexicans. An additional push in this direction may stem from the high rate of undocumented status among the immigrants and the hostility expressed towards them by many Anglos. Speculating a bit, one could infer that these pressures could spur those of partly non-Mexican ancestry to identity with it rather than with their Mexican heritage (Jasso 2001).

Another possible outcome of assimilatory processes is a pan-ethnic identity as Hispanic or Latino (Espiritu 1992; Portes and Rumbaut 2001, Chap. 7). Pan-ethnic identities have long been a feature of U.S. society, as exemplified by the acquisition of a white racial identity by the members of racially “in-between” groups such as eastern European Jews and southern Italians (Barrett and Roediger 1997; Ignatiev 1995). Some sociologists assert that they are likely outcomes for contemporary groups from Asia and Latin America because the U.S. classification system tends to slight nationality distinctions, as between Mexicans and Salvadorans, in favor of more global categories, like Hispanic. In the U.S., so the argument goes, the connection of such categories to opportunities for advancement that have been generated by affirmative action and diversity policies more generally will encourage individuals to identify themselves in pan-ethnic terms (Nagel 1994; Skrentny 2004). The nature of the political system, which favors larger groups over smaller ones, points in the same direction.

But there are arguments that would lead one to hypothesize the resilience of a Mexican–American identity. Immigration itself might strengthen that identity. That is, there are incentives for later-generation individuals to identify with a demographically and culturally powerful group, especially in areas with a salient Mexican presence. (Such incentives could account for the tendency of many later-generation, mixed-ancestry whites to identify with the Irish, for example [Hout and Goldstein 1994.]) Indeed, insofar the Mexican population is constituted by immigrants, there is a niche for members of the U.S.-born generations to serve as the interface between the group and the institutions of the surrounding society, in such occupations as teachers and lawyers. This reasoning suggests that the attractiveness of a Mexican identity should be stronger in areas of immigrant concentration and weaker where the Mexican population is small (Jiménez 2004).

Analytic Strategy

Our analysis involves tracing Mexican–American birth cohorts across the 1980–2000 censuses. We define Mexican Americans in narrow and broad ways—at the narrowest, the Mexican–American population is equated, as is conventional, with those who identify themselves as Mexican on the census Hispanic-origin question, but at its broadest we include others with discernable Mexican ancestry, who are therefore eligible to place themselves in the Mexican category on the Hispanic question but do not. We consistently restrict the analysis to U.S.-born individuals. In that way, we virtually eliminate the possibility that international migration, specifically to and from Mexico, can account for the changes in size that we will document. (The Mexican Census of 2000 shows only a small resident population—about 500 thousand—of individuals born in the United States. Since there is a sizable group of Anglo retirees living in Mexico, it is implausible that one can account for much of the size changes in the Mexican–American population by migration to Mexico. U.S.-born children living in immigrant families are something of an exception, as noted subsequently.)

The diversification of Mexican–American identities is evident in the different ways that members of the group can be identified in census data. They are shown in Table 1. The Mexican category in the census Hispanic-origin question, which has been asked since 1980 (where our data series thus begins), is the basis for virtually all tabulations of Mexican–American characteristics to be found in the literature (e.g., Bean and Stevens 2003). To distinguish this group from others, we will describe it as “Mexican Americans in the census” or “Mexican Americans as conventionally defined.”

Table 1 Distribution of the U.S.-born Mexican-ancestry population across categories of identification in census data
Full size table

One complication for the data series derived from this category lies in the changes in the wording of the Hispanic-origin question across the census years (shown in Table 2). Nevertheless, in all 3 years, the labels “Mexican,” “Mexican Amer.” and “Chicano” appear together on the census form. In 1980, they appear in connection with a question that neutrally inquires, “is this person of Spanish/Hispanic origin or descent?” In 1990, the question was the same, except for the elimination of the phrase “or descent.” In 2000, however, the question shifted to: “In this person Spanish/Hispanic/Latino?” It appears that the new question was understood by some respondents to ask whether they labeled themselves in these ways. It may have encouraged, in other words, Hispanics to identify pan-ethnically rather than with specific origins (see del Pinal 2004, pp. 9–10).Footnote 1

Table 2 Wording of question on Hispanic origin in 1980, 1990, and 2000 censuses
Full size table

Partly because of the change, a growing number of individuals with Mexican ancestry can be found by other means. Some of them place themselves in the “other” Hispanic category, consistent with the possibility that they think of themselves now in terms of pan-ethnic categories (or because they responded in unanticipated ways to the Census 2000 question). This category grew in size especially between 1990 and 2000, suggesting an influence of the change in wording of the 2000 Hispanic-origin question. Some of its incumbents can be identified as having Mexican ancestry because of their responses to the ancestry question. Such individuals are subsequently described as “other-Hispanic Mexicans,” and their percentages of the total U.S.-born Mexican-origin population are shown by census year in Table 1. In some tables we will combine the Mexican Americans in the census with the other-Hispanic Mexicans in order to account for the possibility that the change in wording in 2000 may have contributed to the apparent loss of Mexican Americans; the resulting composite category will be labeled “Mexican-origin Hispanics.”

A final category is initially the most perplexing: the “non-Hispanic Mexicans.” These are persons who refuse to describe themselves as Hispanic or Latino on the Hispanic-origin question, but then indicate some Mexican ancestry on the ancestry question. They are a relatively stable component of the U.S.-born Mexican-origin population, accounting in each census year for about 3% of the total. Nevertheless, there is some evidence that the individuals in this category can identify themselves as Mexican American under some circumstances. In particular, the Census Bureau’s reinterview study to examine consistency of reporting in the 2000 Census found a significant number of individuals who were reported as non-Hispanic in the 2000 Census but claimed in the follow-up survey to be a mix of Hispanic and non-Hispanic origins; in the main, they described their Hispanic ancestry as Mexican (del Pinal 2004, pp. 16–17).

The Magnitude of Mexican–American Loss Over Time

Table 3 traces the changing size of a single birth cohort from 1980 to 2000: Mexican Americans who were in the ages 25–29 in 1980 (and thus were born in the U.S. in the period 1951–1955). We have chosen to begin with a group that was in adult ages at the start of the period we can observe, so that the individuals involved, rather than their parents, have been able to specify their census classifications at every time point. We show the evolution of two different definitions of this birth cohort: the first, in the top panel, includes only Mexican Americans as conventionally defined; the second expands the count to include the other-Hispanic Mexicans. The cohorts are shown separately for men and women. In order to take into account losses due to mortality, a “relative” loss figure is shown; this is the loss beyond that experienced by the same birth cohort of all U.S.-born persons (regardless of ethnic origin) during the same period. This is a reasonable approach to accounting for mortality in light of the recent analysis by Palloni and Arias (2004), which shows that the Hispanic mortality advantage is essentially confined to the foreign born.

Table 3 Disappearing Mexican Americans, 25–29 years old in 1980
Full size table

If we restrict our attention only to individuals who declared themselves to be Mexican on the Hispanic-origin question, then the attrition over a two-decade period is very substantial. Net of the losses due to mortality, the fall-off in membership is 3–5% by 1990 and nearly 20% by 2000. Obviously, the bigger decline occurred during the 1990s, and this fact strengthens the suspicion that the shift in question wording is somehow implicated in it. However, the picture is not changed very much by adding in the other Hispanics who can be identified as Mexican through the ancestry question. With them included in all years, the attrition in the group during the 1980s is now somewhat larger than before, while the total loss between 1980 and 2000 is somewhat ameliorated. But at 13–15%, it is certainly not small.Footnote 2

How extensive are these losses for other birth cohorts? Table 4 summarizes the declines over time for 5-year birth cohorts, up through the 60–64 year-olds of 1980. To simplify the presentation, we drop the separate panels for men and women, and we present only the evolution of the Mexican-origin Hispanics, i.e., the combined population of Mexican Americans as conventionally defined and other Hispanics of Mexican ancestry.

Table 4 Mexican-origin Hispanics: Proportions of counted to expected, 1980–2000, by birth cohort
Full size table

Aside from the age extremes, the average decline for Mexican-origin Hispanics in the 1980–1990 period seems to have been modest, on the order of 4–5% points more than one would have expected from the size changes exhibited by all the U.S. born. An exception is the 0–4 age cohort of 1980: it does not decrease in size. Plausibly, this occurs because of the circularity of Mexican migration in the pre-IRCA period (Massey et al. 2003). Thus, some families with U.S.-born children would have been in Mexico at the time of the 1980 Census and returned to the U.S. before 1990, possibly in order to allow their children to attend school. This addition would have offset the loss the cohort might have experienced.

In general, there are large declines when the 2000 counts are compared to the 1980 base figures. Aside from the youthful cohorts of 1980, the declines are at least 10%, and usually they are more, averaging around 12% compared to the expected values. The youthful cohorts, extending through the 10–14 year-olds of 1980, do not decline as much. Though we do not show separate figures for men and women, it turns out that for males who were under 10 in 1980 and are thus 30–39 years old at the time of the 2000 Census, there is even an increase over time. Quite possibly, some of the U.S.-born children who were in Mexico at the time of the 1980 Census have returned to the U.S., where they can reside and work because of their U.S. citizenship.

Nevertheless, for the cohorts that were past childhood in 1980, there is a consistent pattern of substantial decline in size by 2000. The losses evident in the Mexican-origin Hispanic population reflect changes in identities that are specific to Mexican Americans: they are not universal among Latin American-origin groups. To demonstrate this point, we have produced Table 5, which is the equivalent to Table 4 but for U.S.-born Cuban-origin Hispanics (We have chosen Cubans over Puerto Ricans for this exercise because of the complications introduced by back-and-forth migration between U.S. states and Puerto Rico.). In Table 5, there is no consistent pattern of change in the sizes of birth cohorts over time. The Cuban group as counted by the census is not losing members in the way that the Mexican group is.

Table 5 Cuban-origin Hispanics: proportions of counted to expected, 1980–2000, by birth cohort
Full size table

Characteristics of Different Mexican-Ancestry Categories

This brief overview of the declines in the numbers of U.S.-born persons who report themselves as Mexican and Hispanic (in either the narrow or the broad sense this characterization can be given) suggests the complexity of the phenomenon. The declines cannot be simply explained as a result of the change in question wording in 2000, although that probably contributed to their magnitude. It cannot be explained either as a consequence of the shift for young people from parental reporting of race/ethnicity to self reporting. The magnitude of the decline suggests the possibility of very substantial effects on the apparent characteristics of the Mexican–American group. For instance, educational data can be affected by these losses among adult Mexican Americans. Thus, it behooves us to try to examine the characteristics of the individuals who appear to be “disappearing” from the Mexican–American group. However, an immediate stumbling block becomes evident: how can we identify them?

One way of approaching the possible selectivity of exit from the Mexican–American group is to examine the characteristics of the different categories of Mexican-ancestry individuals that we can identify in census data: Mexican Americans as defined in the census, other Hispanics of Mexican ancestry, non-Hispanic Mexicans. Table 6 therefore presents the racial self-classification, language, and educational characteristics of the incumbents of these three categories in 2000.

Table 6 Characteristics of U.S.-born Mexican-ancestry population in 2000
Full size table

The table demonstrates that the non-Hispanic Mexicans stand apart from the other two categories. In fact, they resemble what one would predict from conventional assimilation theory: that is, their characteristics suggest that their exit from the group is linked to entry into the mainstream. They are far more likely than the incumbents of the other two categories to describe themselves in racial terms as “white” only, and the great majority of them do so. They are very likely to speak only English at home. And their educational attainment is, on average, superior to that of other Mexican-ancestry individuals. In addition, we have confirmed something that the table does not show: many of these individuals report some European ancestry in addition to Mexican. It is logical to infer that many incumbents of this category are the descendants of intermarriages (see Duncan and Trejo 2005).

The differences between Mexican Americans in the census and the other-Hispanic Mexicans are less clear-cut. But the small differences that one finds suggest that the other-Hispanic Mexicans are more distant from the Anglo mainstream than the Mexican Americans in the census are. Perhaps this is the case because the category contains some individuals who misunderstood the nature of the Hispanic-origin question posed in the 2000 census, and such individuals are likely to have low levels of education. In any event, the other-Hispanic Mexicans have slightly lower education than Mexican Americans and are less likely to speak only English at home and to classify themselves as racially white.

Conclusion

Some Mexican Americans are disappearing from the group. At least, that appears to be a fair conclusion from the declines in the sizes of US-born Mexican-origin Hispanic birth cohorts over the period 1980–2000. These declines are well beyond what can be accounted for by mortality, and they are visible in both decades under observation. The change in question wording in the 2000 census may have distorted the magnitude of decline during the 1990s, but it is highly implausible that it can explain it. The inevitable conclusion is that identity shifts are taking place in the Mexican–American population and are moving some Mexican Americans into other racial and ethnic categories.

Both of the identity shifts predicted earlier are evident in the data, though only one of them does much to explain the declining sizes of Mexican–American cohorts across censuses. The outcome conventionally predicted by assimilation theory, a weakening of ethnic identity in tandem with a closer identification with the mainstream society, is visible in the category of non-Hispanic individuals with Mexican ancestry, which regularly accounts for 3–4% of the Mexican-origin population. This is not a large fraction, to be sure, and this group has represented a stable fraction over time; thus, it does not do much to explain the 1980–2000 declines. Presumably, the identity shifts involved here occur across generations and are, in significant part, the outcome of intermarriages in the past. This group resembles an assimilation outcome in several respects: its members are likely to identify racially as whites; they are more likely than other Mexican-ancestry Americans to speak only English; and their educational attainment is higher than that of Mexican-ancestry Americans in general. In addition, many report having some European ancestry.

The other shift involves the emergence of a pan-ethnic identity, i.e., Mexican-origin individuals who apparently identify mainly as Hispanics or Latinos (see also Portes and Rumbaut 2001). We can see this phenomenon in the category of Mexican-ancestry individuals who acknowledge their membership in the Hispanic-origin group but without checking the Mexican box. This category grew substantially in size in 2000; and even though a change in question wording may have contributed to this growth, it is unlikely to account for all of it.Footnote 3 The pan-ethnic category appears to be selective of individuals who, to judge by educational attainment, are less well positioned than the average Mexican American. They are also more likely to identify with the “other” race category as opposed to the white category.

We cannot address here whether any other racial/ethnic groups are undergoing the same set of dynamics that we find among Mexicans. One could argue of course that some white ethnic categories, as measured by the census ancestry question, have exhibited similar declines over time. But, in one important respect, the way that whites respond to the stimuli in the ancestry questions is different from the pattern we have documented among Mexican Americans. In writing answers to an open–ended question, whites are more likely to name the categories that appear in a list of examples (Lieberson and Waters 1993). In the Mexican–American case, the specific ethnic labels—”Mexican, Mexican Am., Chicano”—have regularly appeared on the census form, but the responses of individuals with Mexican ancestry to them have not been consistent.

The trends are of course interesting in themselves, but demographers and sociologists need to be concerned about them for other reasons: namely, they could affect analyses of the Mexican–American group. These identity shifts could interact in complex ways with such phenomena of interest as the Hispanic mortality paradox, which some research has found to apply to Mexican Americans (see Hummer et al. 2007). On the one hand, the loss of Mexican Americans from the group as measured by the standard census definition reduces the denominator used in calculating death rates for this group; on the other hand, the factors that lead some individuals of Mexican ancestry not to identify consistently with the group could also affect their classification in death records and hence the numerator.

In addition, the selective departures from the Mexican–American group are on a scale that could easily impact on the measurement of group characteristics over time. This worrisome conclusion seems especially valid if we restrict our attention to the conventional census definition of the group: individuals who describe themselves on the Hispanic-origin question as Mexican. Table 3 showed that, for a young adult cohort in 1980, the unexplained loss was on the order of 20% during a two-decade period. Granted, the patterns of identity shift we have identified in this paper exhibit different forms of selectivity: those pulled toward a pan-ethnic category are less “assimilated” in conventional terms than the average Mexican American; those identifying with the mainstream society are obviously more so. That these changes are somewhat offsetting does not negate the need for more research on the ramifications of the substantial outflow from the Mexican–American group.