Portal de Periódicos da CAPES

Is Race Really So Fluid? Revisiting Saperstein and Penner’s Empirical Claims

2016; University of Chicago Press; Volume: 122; Issue: 1 Linguagem: Inglês

10.1086/687375

1537-5390

Richard Alba, Noura Insolera, Scarlett Lindeman,

Critical Race Theory in Education

Previous articleNext article FreeCommentary and DebateIs Race Really So Fluid? Revisiting Saperstein and Penner's Empirical ClaimsRichard D. Alba, Scarlett Lindeman, and Noura E. InsoleraRichard D. AlbaGraduate Center, CUNY Search for more articles by this author , Scarlett LindemanGraduate Center, CUNY Search for more articles by this author , and Noura E. InsoleraUniversity of Michigan Search for more articles by this author PDFPDF PLUSFull Text Add to favoritesDownload CitationTrack CitationsPermissionsReprints Share onFacebookTwitterLinked InRedditEmailQR Code SectionsMoreTo conserve space for the publication of original contributions to scholarship, the comments in this section must be limited to brief critiques; author replies must be concise as well. Comments are expected to address specific substantive errors or flaws in articles published in AJS. They are subject to editorial board approval and peer review. Only succinct and substantive commentary will be considered; longer or less focused papers should be submitted as articles in their own right. AJS does not publish rebuttals to author replies.The status of race is in flux. Once viewed as a fixed, ascriptive, and self-evident characteristic derived from ancestry but manifested in phenotype, race is increasingly seen by scholars as fluid and changeable over time, possibly during a person's life span and perhaps even from situation to situation. A number of intellectual currents have contributed to this transformation, which is due above all to the wide acceptance among social scientists of the idea of the social construction of race. The idea seems to preserve the sociological reality of race in the face of the scientific denial of a significant biological foundation for race (but see Morning 2011). Sociologists continue to demonstrate that, in the United States, race is a profound determinant of an individual's position in major social hierarchies—a critical lever, for example, in the intergenerational transmission of place and its linkage to systematic advantages and disadvantages (Sharkey 2013).Yet race is also mutable. The historical literature on whiteness in the United States shows that groups can change their racial identification over time. The immigrant groups from southern and eastern Europe, along with the Irish, were once viewed as racially inferior to white Americans, but in the post–World War II period they succeeded in gaining acceptance as fully white (Jacobson 1998; Roediger 2005; Alba 2009). And it has long been known from the phenomenon of passing that individuals too can move from one racial group to another—the case of the New York writer and critic Anatole Broyard has been frequently cited (Broyard 2007). Recent research establishes an even greater fluidity for multiracial individuals, whose racial identities can shift from one context to another (Harris and Sim 2002; Rocquemore and Brunsma 2008).The latest contribution in this vein comes from Saperstein and Penner (2012), who provide a sophisticated and appealing theoretical account of the case for fluidity in "Racial Fluidity and Inequality in the United States" (AJS 118: 676–727). To give empirical flesh to this conceptual skeleton, they analyze the 1979 National Longitudinal Study of Youth (NLSY 2006) and appear to uncover a far greater fluidity in race than had been heretofore imagined. Their study attends not only to racial identity but also to the racial perception of individuals by others, which is obviously a keystone in any system of racial stratification. The instability in racial perceptions seems remarkable, inasmuch as, according to the analysis, a fifth of NLSY respondents are classified inconsistently by interviewers over a 19-year time span. The study also loosens the linkage of unstable racial perceptions to their presumed foundation in mixed ethnoracial family origins insofar as instability seems to be found across a spectrum of groups defined by ancestry.Saperstein and Penner's findings threaten to diminish the status of race as a primary determinant of life trajectories and status. Indeed, they argue in their conclusion for a view of race at the individual level as "a flexible propensity rather than a fixed characteristic" (p. 706). They advance a Latin American–flavored lightening/darkening thesis on the basis of findings that perceptions of race are tied to dimensions of present socioeconomic position: individuals appear to lighten as they move up and darken as they move down, by for example experiencing unemployment or incarceration. In the words of Saperstein and Penner (p. 676), "These findings suggest that, in the contemporary United States, microlevel racial fluidity serves to reinforce existing disparities by redefining successful or high-status people as white (or not black) and unsuccessful or low-status people as black (or not white)." The findings also threaten a measurement system that is taken for granted by sociologists, which assumes that the declared racial identities of individuals—for example, on the census—correspond tightly with their family origins and also with the way they are perceived by others. At the end of their article, Saperstein and Penner call for new approaches to measuring and modeling race.Reanalyzing the NLSY data, however, we find that racial fluidity is less common than Saperstein and Penner assert. Overall, there appears to be a very high level of congruence between the ethnoracial backgrounds declared by individuals during their original interviews and the ways they identify and are perceived by others over a more than two-decade period. Non-Hispanic whites and blacks are to a very great extent consistent in these respects. Inconsistency of racial perceptions is highly concentrated in particular groups, Latinos above all but also among non-Latinos with mixed racial family backgrounds. Further, while we do find some evidence, albeit weak, to support the lightening/darkening thesis put forward by Saperstein and Penner, our findings concerning the inconsistency of racial perceptions point more strongly to the evolving social categorization of individuals who fit poorly into a black-white scheme. In that sense, the apparent instability of race is, to an important extent, a function of uncertainties in racial categorization, which have probably declined over the life of the survey, and of the highly constrained measurement instruments in the NLSY.NLSY DataThe NLSY79 data consists of an ongoing longitudinal set of samples that began in the base year 1979 with interviews of American youths between ages 14 and 22. About half of the original data (N=12,686) is constituted by a representative sample (N=6,111), that is, one that represents the noninstitutionalized civilian population born in the 1957–64 period. A supplemental sample almost as large is an oversample of minorities and economically disadvantaged youths (N=5,290). A third subsample contains members of the military born in the same period (N=1,285). When working with these samples, especially when they are combined, it is necessary to employ weights to achieve representativeness,—that is, to avoid distortions, such as an overrepresentation of Latinos (Frankel, McWilliams, and Spence 1983).Saperstein and Penner make clever use of these NLSY79 data, two components of which are essential to their analysis of racial consistency. At the end of each in-person interview through 1998 (except for 1987; hence, 17 occasions in all), interviewers were asked to record their perception of a respondent's race as "white," "black," or "other." This sequence of perceptions allows Saperstein and Penner to make one critical evaluation of consistency. They exploit the fact that the interviewer's perception was recorded at the end to infer that it may be influenced by information provided during the course of the interview (see, p. 688). According to their analysis, about 20% of Americans (belonging to the birth cohorts in the NLSY) are perceived inconsistently over the 19-year period between 1979 and 1998.Another indicator of consistency concerns ethnoracial identity. In 2002, respondents were asked to identify themselves by questions that are similar to the 2000 census race and ethnicity questions (thus allowing multiple responses to the race question). Saperstein and Penner observe a nontrivial degree of inconsistency between the backgrounds recorded in 1979 and the identities registered in 2002.We find a third component of the data to be critical for an understanding of fluidity and consistency. In the original 1979 wave, respondents were asked about their ethnoracial origins and were provided with a flash card that contained a list of 29 group labels (e.g., "Irish," "Cuban or Cubano"), including "other" and "American." Respondents could provide up to six responses. We will refer below to a variable we construct from this information as "ethnoracial origin" or "background." Saperstein and Penner use the information about background to some degree (especially to control for Hispanic ancestry and mixed race in their models), but we will show below that a fuller specification of ethnoracial origin is the key to understanding patterns of racial perception and identity.How Much Fluidity in Ethnicity/Race and for Whom?We find, first of all, that the level of inconsistent racial perceptions is substantially lower than Saperstein and Penner indicate. In contrast to their report of 20% for the NLSY sample, we find that inconsistency occurs for about 11% (Table 1).1 Our estimate differs from theirs because we use the NLSY-provided weights (in particular, the base year weight). Because Saperstein and Penner reject weighting, they fail to properly adjust for oversampling of minority and economically disadvantaged youths.2 For instance, as we will observe below, inconsistency is especially associated with Hispanic ancestry, and the overrepresentation of Latinos in the unweighted data elevates the apparent level of inconsistency (see the leftmost columns of Table 1).Table 1. Perceived Race (1979–98) and Declared Identity (2002) by Ethnoracial Origin (1979) Sample PercentagePerceived Race 1979Changes In Perceived RaceDeclared IdentityEthnoracial OriginUnwgt.Wgt.WhiteBlackOther01–23+Non-HispanicHispanicMixed with WhiteBlack/ American IndianWhiteBlackOtherMainly/exclusively European: European only…33.445.099.0.9.297.02.5.597.1.6.71.2.5.0 American/none…5.05.999.2.6.296.23.0.895.61.3.03.1.0.1 White/American Indian…5.66.897.6.12.294.83.22.091.7.33.71.72.7.0 White/other…7.511.599.2.6.297.62.1.398.3.1.5.9.2.0Black only…23.112.7.799.2.188.210.21.6.497.6.6.6.1.7American Indian only…2.22.388.62.19.384.48.67.084.31.310.71.91.4.4Asian only….8.725.73.770.616.520.962.611.83.080.6.54.0.0Latino categories: Latino only …13.25.371.1.928.09.118.772.21.8.2.697.5.0.0 White/Latino…1.4.988.92.58.656.721.521.839.8.0.060.2.0.0 Asian/Latino ….1.022.4.077.68.915.176.0.0.0.0100.0.0.0 Black/Latino ….0.0.0100.0.070.73.425.9.055.4.044.6.0.0 Latino/other….1.191.8.08.261.430.08.65.0.0.066.7.028.4Other only…2.94.295.92.02.088.96.54.693.9.71.52.51.4.0Other mixtures: Asian/white….4.692.0.08.076.66.317.175.7.010.93.410.0.0 Black/white….3.16.884.88.354.126.419.510.866.77.511.43.6.0 Black/American Indian….5.3.094.75.388.15.26.71.690.02.1.01.74.6 All other mixed…2.52.685.88.16.281.110.38.676.88.03.69.41.8.5No response….91.098.21.8.094.15.6.392.42.4.03.02.2.0 Total …100.0100.083.213.92.988.95.35.876.214.21.67.1.7.1Note. Except for the first column, all percentages are derived from the weighted sample (using the 1979 sample weight); changes in perceived race are measured only for individuals who appear in two or more waves. After the first two columns of the bottom row, "total" indicates the breakdown of the whole sample for "perceived race," "changes in perceived race," and "declared identity." wgt. = weighted; unwgt: = unweighted.View Table ImageInconsistency of observer racial perceptions is concentrated in specific portions of the population marked off by ethnoracial ancestry. In this sense, the conventional hypothesis that racial classification corresponds to family background is maintained, and strongly so. To begin with, there is little inconsistency to be found for individuals whose ancestry is mainly or exclusively European. Such individuals compose about 70% (69.2%) of the weighted sample. Along with respondents who report only European ancestry at first interview, we include under this rubric individuals who report no specific or exclusively American ancestry or who claim to have European and American Indian forbears or European and "other" ancestors. As is well known, many whites claim some American Indian ancestry (Snipp 1986), and the "other" category in the case of whites includes, we presume, many European ancestries that were not separately coded (e.g., Lithuanian, Swedish). (We do not, however, include in the white category those who specify only "other," even though many of them likely named one or more of these smaller European ancestries, because by itself the "other" category appears to contain some nonwhites as well.)The initial racial classification of individuals in all these categories was "white" in 98%–99% of the cases (in Table 1). Over all the waves analyzed, inconsistent racial perceptions characterize just 3.2% of respondents. In the majority of cases, a shift in perceptions occurs just once or twice; most of these individuals were classified a single time as other than white (the "count" of between-year inconsistencies in such cases is typically two because respondents are classified as other than white in one year and then classified again as white in the following interview). High inconsistency—three or more perceptual shifts—appears in just 0.7% of the cases. Another way of looking at the consistency of racial perceptions is based on the count of all interviewer perceptions rather than on the number of respondents: if we eliminate the small group who were recorded as nonwhite at first interview, most of whom were also classified as nonwhite in some or all subsequent interviews, then 99.7% of all interviewer observations saw European-descent individuals as white.3Even in these highly consistent categories, there is some patterning to the modest degree of inconsistency that in turn relates to ancestral background. The most salient example is Portuguese, an ancestry found among a quarter of all the highly inconsistent Europeans. Undoubtedly, the presence of mixed-race Cape Verdeans in the Portuguese group contributes to this frequency.African-Americans, despite the well-known history of racial mixing during slavery, have a frequency of inconsistent perception that is almost the same as the average for the sample as a whole. Respondents who report only black or only African ancestry constitute 12.7% of the weighted sample, and 12% of them are perceived inconsistently at least once. But for the great majority, shifts occur just once or twice. Only 1.6% experience a high level of inconsistency, three or more perceptual shifts between interviews. Altogether, only 1.1% of observer perceptions of African-American respondents see them something other than "black."We find inconsistency more commonly for three types of ethnoracial categories: (1) single-race individuals who are neither white nor black—especially Asians, a small a small category in the NLSY that accounts for less than 1% of the weighted sample; (2) Hispanics (including those who report other backgrounds also), who make up 6.3% of the sample; and (3) mixed-race categories other than those already named, which constitute 3.6%.Inconsistent racial perceptions are most common among individuals who report Latin American ancestry, whether in mixture with other ancestries (e.g., European) or by itself. About 70% of all the respondents with a high level of inconsistency are Latino. (The concentration of high inconsistency among Latinos is even greater in the unweighted sample analyzed by Saperstein and Penner.) The level of inconsistency is so large that it suggests uncertainty among the interviewers about the racial classification of Latinos—perhaps this is not so surprising given the limited choices available to the interviewers for coding race and also the cultural uncertainty in American society at the end of the 1970s about the racial position of Latinos (the census, it should be remembered, only introduced a question to systematically identify Hispanics in 1980). Among those who report Latino ancestry only, just 9% percent are classified consistently across all the NLSY surveys. Nearly three-quarters of the Latino-only respondents experience three or more between-year shifts in racial classification.Yet the inconsistency of racial perceptions of Latinos still operates largely within a confined categorical space, for perceptions of them as black are quite uncommon in the NLSY, occurring just 1.9% of the time. Interviewers, in other words, toggle mostly between white (73.0% of all perceptions) and other (25.1% of perceptions). Another way to think about this confinement of perceptions is that only 7.0% of Latinos are ever perceived as black. A majority of them trace at least some of their ancestry to the Caribbean, mostly to Puerto Rico.The level of inconsistent observer perceptions is substantially lower among those coming from other mixed-racial backgrounds than it is among Hispanics. Eighty percent of them are consistently perceived by interviewers, and just 10% exhibit high inconsistency. However, curiously, there is hardly any greater distribution of racial perceptions than among Hispanics, but in this case, the primary division is between white and black. More than three-quarters of all perceptions classify mixed-race individuals as white, while a fifth have them as black. Just 2.5% of the time are they seen as other.Overall, then, there appears to be a high level of congruence between the ethnoracial backgrounds of respondents, as reported at the time of the first interview in 1979, and the way they are perceived by interviewers over a nearly two-decade period. To be sure, there is some inconsistency, but it is concentrated among respondents with particular ethnoracial backgrounds, especially those who do not fit easily into the white-black division that underlay the racial categories interviewers were offered at the outset of the NLSY data collection in 1979.There is also a quite strong congruence between ethnoracial background and the way that respondents identify themselves in terms of the census category scheme, according to the data that were collected in 2002. These data were gathered through separate questions about Hispanic origin and race, and the race question allowed for multiple responses. The correspondence between background as specified in the original 1979 interview and identity in 2002 is enhanced if we also take into account the interviewer's racial perception in 1979, which recognizes the racial category of some ambiguous cases of self-report (e.g., among persons who claimed exclusively "American" origins in 1979).For instance, among those who were classified in 1979 by the interviewer as white and reported mainly American or European backgrounds then (as before, we include here persons who report no or exclusively American origins, as well as those who report American Indian or "other" origins in addition to European backgrounds), the percentage who describe themselves as non-Hispanic whites is very high—97.3%. The largest category among the remainder is Hispanic (1.3%), which may indicate no more than that some of the original reports in 1979 were incomplete; others describe themselves as white and American Indian, consistent with their 1979 ancestry reporting. Only a tiny number identify as black, either alone or in combination with white.A similar consistency can be found among those respondents who were identified by the original interviewer as black and pointed to exclusively African ancestral origins. Nearly 98% percent (97.7%) of them identify in 2002 as non-Hispanic black, and another 0.7% say that they are black and American Indian.Finally, Hispanics are also quite consistent across this time span. Of those who described exclusively Latin American ancestry in 1979, 97.5% identified as Hispanic or Latino in 2002. Those who described a combination of European and Latino ancestry in 1979 divided 40–60 between non-Hispanic white and Hispanic in 2002. Other mixed Latino categories are too small to report.In conclusion, there is a very strong congruence between the ethnoracial origins described in the original interviews in 1979, on the one hand, and the identities and perceptions of respondents collected over a nearly 25-year period. Although there are some inconsistencies, their patterns make sense in terms of respondent backgrounds.Are Racial Classifications Influenced by Socioeconomic Events?A key claim by Saperstein and Penner is that the correlation between race and socioeconomic status is created by mutual influence: just as race influences socioeconomic processes and thereby determines status to a significant degree—the classic sociological argument—so socioeconomic events, like becoming unemployed or going to prison, can influence racial classification. Thus, positive events like college graduation can "whiten" Americans while negative ones can augment the likelihood that they are perceived as minorities. In this section, we look for evidence within the framework of our analysis that might support this claim and evaluate any effects we find against other influences on perceptions of an individual's race.Given the patterns of racial perception by ethnoracial background that we described in the previous section, it seems logical to us that the determinants of perception vary by major ethnoracial group—that is, we see little reason to believe that the "social rules" of racial classification are the same for, say, Hispanics and African-Americans. In this way, our analysis departs from that of Saperstein and Penner, who generally include individuals regardless of ethnoracial background.An additional benefit of separate analysis by group is, we believe, that it reduces the risk of spurious conclusions about the connection between perceived race and events like unemployment and teen parenthood. That is, since the probability of these varies quite strongly by ethnoracial origin, there is a risk for an analysis comprising people of all origins of concluding that the events cause the perception when in fact ethnoracial status lies behind both, heightening the probability of the events and leading to the perception. This risk is potentially exacerbated in the Saperstein-Penner analytic framework because it focuses on changes in perception between interviews (going from white in one wave to other in the next, say) and because they use several cumulative variables (e.g., ever incarcerated). As we have already noted, a sizable number of shifts in perception involve returns to a classification consistent with ethnoracial origin (as when a European-descent respondent is perceived once as nonwhite and returns in the next wave to classification as white). Such return shifts are inevitably predictable to some degree in terms of events that are associated with ethnicity and race, such as being recently unemployed for several months or ever being incarcerated. (For example, a return to the white category may be predictable from prior incarceration, say, because whites are much less likely to be incarcerated.) Shifts in perception away from ethnoracial origin, as Saperstein and Penner note (p. 689), may be less predictable because at least some involve more or less random interviewer or coding error.We do not use the same statistical approach as Saperstein and Penner, though we do use most of the same (or similar) variables. Rather than employ an event-history framework, we take a hierarchical linear modeling (HLM) growth-curve approach: specifically, the variables at the lower level of the HLM model are those measured for a given individual in a particular survey year, while those at the higher level are stable characteristics of the individuals. At the lower level, racial perception at time t, the dependent variable, is a function of time-varying variables, such as unemployment at time t and interviewer characteristics. At the higher level, where the equation determines the intercept in the lower-level equation, the variables are such stable characteristics such as gender and ethnoracial origin (as declared in 1979). In equations, our models can be represented as:Ln Oit=ci+Σ bkTkit+εit,ci=ai+Σ blXli+εi,where Oit is the odds of a particular racial classification for individual i and time t, the T's are time-varying variables, and the X's are stable individual characteristics. We first develop a base model of this type, and we then test variables of interest, such as unemployment in a given year, by adding them singly to the model.Hence, unlike the analysis in Saperstein and Penner's "Racial Fluidity," our analysis does not focus on a change in perception, but on the probability of being classified as white rather than black or other, or as black rather than white or other. If changing social characteristics affect racial classification, we should be able to detect such an effect in this format. Since inconsistency is concentrated among individuals from Hispanic origins and from mixed-race backgrounds, these are the two groups we analyze. For Hispanics (unweighted N of persons=1,978), who are overwhelmingly perceived as white or other by interviewers, the dependent variable is defined as white versus the remaining categories.4 For non-Hispanic mixed-race persons (N=365), it is black versus the other two options. In both analyses, we depart from Saperstein and Penner by including dummy variables to mark key aspects of ethnoracial origin (e.g., having European ancestry), since we hypothesize that it influences both racial perceptions and significant life events and adult statuses.Table 2 presents the results. Several variables in the basic model for Hispanics have pronounced effects. Given the correspondence we have already noted in detail between ethnoracial background and consistency of racial perception, it is probably not surprising that detailed ethno-racial background has a large effect on the likelihood that a Hispanic respondent is perceived as white. Someone who mentions having European ancestry is three times more likely to be perceived as white than is another person from a Mexican background (i.e., exp (1.101-[-.023])=3.08). An Hispanic who cites African ancestry is almost 40 times more likely to be seen as nonwhite compared to the latter. Puerto Ricans are also more likely to be seen as nonwhite, and Cubans, white.5Table 2. HLM Logistic-Regression Analyses of Perception as White for Hispanics and Perception as Black for non-Hispanics of Mixed Race Hispanics (1 = White)Non-Hispanics of Mixed Race (1 = Black)Stable variables: Sample: Representative (omitted)---- Poverty…−.203**.232 Military…−.516**−1.106Ethnoracial origin (1979): European…1.101***−1.701* Black…−3.703***14.624*** Mexican…−.023NA Puerto Rican…−.293*NA Cuban….495***NAOther demographic characteristics: Gender (female) ….092−.235 Foreign born…−.059−2.238*Completed education: < high school…−.001−.376 High school (omitted)……….---- Some college………..234**−.021 College graduate….430***−.492Dynamic variables: Age…−.032***−.063* Marital status and children: Single, no child (omitted)…---- Single, child….034.854 Married, no child….130*−.311 Married, child….139**−.104Residence: Not in SMSAa (omitted)…---- Inner city…−.048−.483 Suburb….017−.250 West (omitted)…---- Northeast…−1.023***.303 Midwest…−.498***.235 South….071.224Interviewer characteristics: White (omitted)…---- Black…−1.191***.423 Hispanic…−.259***−1.826+ Other…−.2782.296 Female….049−.466 Sex missing…−.155−1.007 Young (< 38) (omitted) …---- Young middle age (38–47)….030.330 Old middle age (48–56) ….768***.948* Older (58+)….938***.262 Age missing….200**.295 Constant…1.758***−6.682***Tested variables (entered singly): Public housing…−.173*.672 Poverty….019.045 Household income ($1979) ….001+−.004 SEI (Duncan prestige index)….002+−.008 Unemployed (> 16 weeks) …−.005−.331 Welfare….0401.158 Teen parent….0451.815+ Incarcerated….399+−.396 N…1,978365a. Or SMSA not listed.+. P<.10.*. P <.05.**. P <.01.***. P <.001.View Table Image: 1 | 2Interviewer characteristics also have sizable impacts. A black interviewer is three times as likely as a white one to perceive an Hispanic respondent as nonwhite; Hispanic interviewers are also more likely to classify the respondent as nonwhite but not to the same degree.6 This relationship probably speaks to the greater sensitivity of nonwhites to race nuances (Harris 2002). The age of the interviewer plays a further role. Older interviewers are more likely to see respondents as white, possibly because they have less exposure to diversity in their social circles. Moreover, the likelihood of an Hispanic being classified as nonwhite increases over time (reflected in the age variable), which we think may indicate an evolving classification system (we will say more about this in the conclusion).Several variables potentially carry socioeconomic influences on racial perception. Education is one, though the interpretation is of necessity ambiguous. Education has been entered as a stable characteristic of the individual here—rather than as a time-varying event such as a graduation—on the assumption that the educational trajectory of an individual is often apparent well before the year of a confirming event such as a graduation. The