Classificatory volatility and (in)consistency of racial inequality

Abstracts

Monitoring racial inequalities, whether socioeconomic or health-related, assumes stability in racial classification. Otherwise, the dynamics of these inequalities could result from racial reclassification rather than from processes related to socioeconomic and health inequalities per se. The study proposes a typology of uncertainty in racial classification (contextual - temporal, geographic, procedural - and sampling) and draws on the literature and nationally representative secondary data to discuss the magnitude of racial variability in Brazil according to these five dimensions. The results show that at least two of these uncertainties - geographic and procedural - are substantial, but have little influence on the racial gap in income. We address the impacts of these results on the existence and extent of racial inequalities in health and conclude that the structure of inequalities between whites and blacks is consistent, although skin color classification is volatile.

Social Inequity; Race or Ethnic Group Distribution; Socioeconomic Factors

O monitoramento de desigualdades raciais, seja num plano socioeconômico ou em termos de desfechos de saúde, pressupõe que a declaração da raça apresente estabilidade. Caso contrário, a dinâmica dessas desigualdades poderia resultar da reclassificação racial, e não de processos vinculados a iniquidades socioeconômicas e de saúde. Este estudo propõe uma tipologia da incerteza racial classificatória (contextual - temporal, geográfica, procedimental - e amostral) e discute, com base na literatura e dados secundários nacionalmente representativos, a magnitude da variabilidade racial segundo essas cinco dimensões. Os resultados demonstram que, pelo menos, duas dessas incertezas - geográfica e procedimental - são substanciais, mas têm pouca influência sobre o hiato racial de renda. Abordam-se os impactos desses resultados sobre a existência e a extensão das iniquidades raciais em saúde e conclui-se que a estrutura das desigualdades entre brancos e negros é consistente, ainda que a cor da pele seja volátil.

Iniquidade Social; Distribuição por Raça ou Etnia; Fatores Socioeconômicos

El monitoreo de desigualdades raciales, sea en un plano socioeconómico o en términos de desenlaces de salud, presupone que la declaración de raza presenta estabilidad. En caso contrario, la dinámica de estas desigualdades podría resultar de una reclasificación racial, y no de procesos vinculados a inequidades socioeconómicas y de la salud. Este estudio propone una tipología de la incertidumbre racial clasificatoria (contextual -temporal, geográfica, procedimental- y muestral) y discute, a partir de la literatura y de datos secundarios nacionalmente representativos, la magnitud de la variabilidad racial, según estas cinco dimensiones. Los resultados demuestran que, por lo menos, dos de esas incertezas -geográfica y procedimental- son sustanciales, pero tienen poca influencia sobre el hiato racial de renta. Se abordan los impactos de esos resultados sobre la existencia y la extensión de las inequidades raciales en salud y se concluye que la estructura de las desigualdades entre blancos y negros es consistente, aunque el color de la piel sea volátil.

Inequidad Social; Distribución por Raza o Etnia; Factores Socioeconómicos

Problematizing the depictions of Brazil’s racial inequality

Racial inequalities in Brazil are notorious. Whatever the measure of inequality - Gini, Theil, index of dissimilarity, ratio, or difference - or outcome of interest - life expectancy, fertility, schooling, access to higher education, homicide rate, occupational prestige, or per capita income - whites consistently possess more resources and occupy hierarchically better positions than blacks (pretos and pardos) ^{1,2,3,4,5,6,7}. Whites’ advantages in relation to blacks also persist over time, although recent evidence suggests a trend towards reduction in this difference in Brazil ^8,9. Until the 2000s, the per capita household income of white was, on average, 2.4 times higher than that of pardos (literally “browns”) and pretos (literally “blacks”) (hereinafter referred to jointly as blacks, unless specified otherwise). In 2008, this ratio had decreased to 2.1. If this reduction were to maintain its pace, racial equality in income would not be achieved in Brazil until 2029 ⁸.

Does this reduction in inequality result from greater economic approximation between whites and blacks? Is the situation of blacks improving, or at least approaching that of whites? Is stability in racial self-classification a condition for inferring that inequality has decreased? The uncertainties of racial classification, widely noted in the sociological literature, corroborate the imprecision of socially constructed and “politically oriented” racial data ^10,11. However, such imprecisions only become problematic in light of their consequences and uses. The race/color variable in itself is not a static category, nor is it useless for the perception of realities that are statistically constructed and used in the allocation of compensatory benefits. The fluidity and multidimensionality of race, perceived as inconsistencies, indicate the construct’s dynamism, but are of little assistance in understanding their causes and consequences ¹². Race, in addition to having become a criterion in public policies, is also responsible for defining our perceptions of stratification and social injustice. After all, the notion of racial inequality is dependent upon the unequal distribution of resources among groups of color, which are statistically variable.

If racial groups are unstable, that is, if those previously classified as whites are reclassified as blacks (and vice-versa), racial inequality would change based on the reclassification of persons among the categorized racial groups rather than by redistribution of income between whites and blacks, assumed to be racially stable. For example, racial inequality in income would increase if pardos (who form a large racial category that traditionally expresses a high degree of classificatory ambiguity) and especially poor pardos began reclassifying themselves as pretos or if wealthy pardos began classifying themselves as whites ^13,14. In other words, such a classificatory dynamic would lead to an increase in racial inequality without any reallocation of resources among the respective categories. The resources would not be redistributed, but skin color would be. In the presence of racial reclassification, the understanding of inequalities thus depends not only on the differential distribution of resources among racial groups, but also on their classificatory stability.

Is this the case in Brazil? How inconsistent are the measures of inequality as a result of racial reclassification? How stable is the classification of racial groups? The answer to these questions depends partly on the dimension according to which one examines the fluidity of skin color. The current article addresses five dimensions: contextual, temporal, geographic, procedural, and sampling. These dimensions are not mutually exclusive and are observed simultaneously, but they provide a theoretical framework for systematizing the factors linked to a population’s racial variations. Such a typology aims to orient the agenda for future research on the impacts of race classification uncertainty on measures of racial inequality. The proposed conceptual structure launches the question in the public health field; by emphasizing the macro-structural components of racial reclassification and their impact on inequality, it transcends the multidimensional approach to race as experienced individually, as recently examined by social sciences scholars ¹². Although the typology highlights uncertainty in racial classification, we argue that it could be equally useful in the study of other types of social classification, including those based on religion, ethnicity, political orientation, social class, etc. Figure 1 summarizes the interrelationship between the five dimensions of race classification uncertainty proposed here and their impact on racial inequalities.

Figure 1
Dimensions of racial uncertainty.

First, we will define the dimensions linked to the uncertainty of racial boundaries, reviewing recent evidence on skin color reclassification in specific contexts (e.g., universities) and at the temporal, regional, procedural, and sampling levels. Next, we will show that although racial uncertainty or ambiguity may be quantified, it is not necessarily relevant for determining the magnitude of inequalities. Importantly, the expressions “race”, “color”, and “skin color” are used interchangeably throughout the article, since they represent emic categories and constitute analytical categories for the current discussion. In addition, we adopted the term “blacks” to include both pardos and pretos, in keeping with the references consulted during the literature review. Although the latter procedure may occasionally prove problematic and questionable ^15,16, it is frequently backed by similarities between pretos and pardos in terms of income and schooling, attempts at increasing statistical power in quantitative analyses, and by the fact that public policies in Brazil currently treat the two population segments jointly.

Racial uncertainty: contextual

The first dimension, contextual, refers to situations in which there is racial reclassification as a function of people’s macro-social context. Context refers to the influence of institutional, socioeconomic, interactional, and cultural factors that affect the process of racial identification (consciously or not). This macro-social dimension encompasses and influences the others and is shown as the gray area in Figure 1. For example, racial variations in time and space are subject to specific institutional influences that operate to some extent regardless of individual preferences. Institutional influence is manifested diffusely through the consolidation of power structures that reproduce and leave legacies of superiority of one race over the other, thus contributing to racial reclassification of those discriminated against or who lack preferential access to the benefits generated by the state and organizations ^17,18. Interaction between the interviewee and interviewer, permeated by their respective attributes, is also part of the contextual dimension. Name, intonation, accent, verbal skill, ancestry, social class, and mutual phenotypical perception also influence skin color self-classification and interviewer-classification ^12,15. The macro-social context is broad and encompasses all the influences that go beyond the individual sphere, but which permeate it at the same time. The macro-social context is thus the whole that transcends the parts, and whose effect is perceived across both time and space.

According to Telles ¹⁹, in the 1990s, whites and pretos increased their relative share of the Brazilian population as a whole: while pardos decreased their share from 42.1% to 38.9% between the 1991 and 2000 population censuses, whites and pretos increased from 52.1% to 53.4% and from 5% to 6.1% during the same period, respectively. However, the relative share of pardos began to increase again in the 2000s, resembling a process referred to as empardecimento (“browning”) observed from the 1940s to the 1990s, which has also been accompanied by a slight increase in the share of pretos in recent years ²⁰. Telles interprets the Brazilian population’s increasing tendency to self-classify as pardos as the result of this group’s recent social valorization in Brazil. According to him, these individuals may have classified themselves as whites in the past in an attempt to avoid identifying with a stigmatized racial category; currently, they would not hesitate to self-classify as blacks in response to certain social incentives, including affirmative action policies launched in the 2000s.

Along this same line, studies at the Brasília University (UnB) show that the adoption of racial quotas beginning in 2004 had a significant effect on applicants’ self-classified race ^21,22. Based on prospective data, the authors of these studies observed that after the university reserved 20% of vacancies for self-classified black applicants, the tendency to identify with the darker skin color categories increased significantly, thus evidencing the inductive effect of affirmative action on constructing racial boundaries ²³. There was also evidence of a similar process at the Minas Gerais Federal University (UFMG). A policy of bonus points was adopted in 2009, according to which applicants that had graduated from public secondary schools and self-classified as blacks received an extra 15% on their admissions scores; the policy appeared to have had a similar effect on racial classification. From 2008 to 2009, the total number of applicants registered in the admissions exams remained virtually stable, but the racial composition changed. There was a percentage decrease in self-classified whites and an increase in self-classified blacks. According to one study, this “may have resulted from induction in racial self-classification generated by the possibility of obtaining bonus points on the admissions exams at UFMG” ²⁴ (p. 323). However, an absolute increase in applications by blacks and a simultaneous reduction in whites is also a hypothesis that is unlikely, but would have to be tested.

Also from a contextual perspective, but limited to the sphere of interpersonal relations as conditioning factors of racial classification, a study in the South of Brazil showed that men tended to self-classify more as pardos than as pretos when interviewed by black women interviewers compared to white women interviewers ¹⁵. The racial categories that are selected can be as diverse as the number of observers and the contexts in which they are observed. Although race is typically recorded once, one can imagine that it is specific to each moment, to each observational process ¹².

Racial uncertainty: temporal

The temporal dimension of racial uncertainty includes, within a single context, the classificatory variations observed over time that extend beyond the respective groups’ reproductive dynamics. Fluctuations due to reclassification appear here as a function of prevailing contextual conditions (politics, the economy, culture, media influence etc.).

Empirically, the temporal dimension involves racial classification of the same individuals at different points in time. Thus, this implies drawing on longitudinal studies that follow participants for a minimum period and that are capable of subjecting study participants to two or more instances of racial classification. The study by Penner & Saperstein ²⁵ in the United States is a good example, since it was based on racial classification conducted at regular one- or two-year intervals among 12,686 individuals, from 1979 to 2002. The authors observed that overall, 6% of the interviewees had their racial classification changed from one wave of data collection to another, for example changing from white to black as a function of a series of markers of social status, such as unemployment, incarceration, and impoverishment. This percentage was considerably higher than that for other characteristics of the participants; for example, only 0.3% (or 20 times fewer) changed their sexual classification during the same follow-up period.

In the Brazilian case, according to the Monthly Employment Survey (PME), 23% of the population in metropolitan areas changed their self-reported skin color over the course of nine months ²⁰. Longitudinal (cohort) studies in Brazil ^20,26 thus offer researchers an interesting opportunity to explore the temporal dimension of racial uncertainty in the country. Unlike studies elsewhere ^27,28,29, until we can determine the trends in interracial transition in Brazil, it will be hard to determine the impact of reclassification over time on inequality indicators, life expectancy, or any other socioeconomic or health outcome.

Racial uncertainty: spatial

The geographic or regional dimension of racial uncertainty refers to the differential propensity towards reclassification of persons living in different regions of the country, even assuming that all other conditions in the classification process remain stable. This dimension includes the influence of the local racial composition on the classification process. Areas with a high proportion of whites are expected to have individuals with greater potential to change their racial classification (for example, “brown” children born to interracial couples), tending to classify themselves as whites. Meanwhile, in areas with a high proportion of pretos, there might be a greater tendency for pardos to classify themselves as black.

This type of uncertainty appears when examining the tendencies towards racial reclassification in different areas of Brazil. If different regions of the country show different probabilities of racial reclassification, this would suggest that race is affected by spatial uncertainty. To test this hypothesis, we followed the ideas presented by Vitor Miranda in his doctoral thesis ²⁰ and matched the data from the PME of the Brazilian Institute of Geography and Statistics (IBGE) over time ³⁰. The rotating-panel design potentially allowed for the same individual to be interviewed at two different points in time: at the first interview and nine months later. These data allow estimating the propensity (and proportion) of individuals that changed their skin color classification between the two interviews and in different places in Brazil.

Figure 2 illustrates this process, showing the likelihood that individuals self-classified as browns (pardos) in the first interview later reclassified themselves as whites (brancos) or blacks (pretos) in the second interview in each of the six metropolitan areas covered by the PME. In general, the proportion of pardos that whitened from one interview to the next tended to be higher in metropolitan areas with a higher proportion of whites. The exception to this tendency was Rio de Janeiro, where the proportion of pardos that reclassified was less than 5%, although Rio has the third highest proportion of whites and the second highest proportion of pretos among metropolitan areas studied.

Figure 2
Predicted probabilities of pardos reclassifying as whites or pretos in major metropolitan areas of Brazil.

Meanwhile, Porto Alegre was the metropolitan area where pardos were more prone to change their race to white (about 40%) or preto (about 15%) between the two interviews. Porto Alegre is the metropolitan area with the highest proportion of whites (90%), but also the lowest proportion of pretos (5%). The higher reclassification of pardos in Porto Alegre may reflect more rigid racial boundaries and thus lower adherence to the category of pardos. Therefore, racial boundaries in Brazil thus appear more permeable and prone to influence by the local racial composition in some regions than in others. The impact of this spatial dimension on the magnitude of racial inequalities will be addressed later in this paper.

Racial uncertainty: procedural

The fourth dimension of racial uncertainty refers to the way information on skin color is collected. Different data collection methods, such as open, semi-structured, or categorical questions, or with self-classification versus interviewer-classification, lead to different racial compositions. We refer to this type of racial uncertainty as procedural uncertainty, resulting from the measurement system used.

How do people change their skin color classification, and how would the country’s racial composition change if information on race/color were produced in different ways? How consistent are the racial groups when race is measured with different questions from those used by the IBGE? Muniz ³¹, Bailey et al. ³², and Loveman et al. ³³ addressed these questions using data from the Brazilian Social Survey (PESB) of 2002, in which racial information was collected four different ways: (1) based on the interviewee’s racial self-classification using the IBGE categories; (2) based on photographs, where interviewees had to select the photo they most identified with; (3) assessed by the interviewer; and (4) based on the skin color of the individual’s darkest direct ancestor. According to the results, “half (27+19+3) of the interviewees classified themselves or were classified consistently, that is, independently of the data collection methodology used or the construction of the race/color variable” ³¹ (p. 261). Meanwhile, the other half consisted of individuals whose race varied according to the methodology used to measure it.

Data from a population-based study in a municipality in southernmost Brazil ^13,15 - Pelotas, Rio Grande do Sul State - also illustrate this type of uncertainty. In 2005, a probabilistic sample of 3,136 adults 20 years or older living in the municipality’s urban area were interviewed at home in the context of a large-scale epidemiological survey conducted periodically in the municipality. Interviewers were asked to classify the study participants according to the five official census categories for skin color (branca, parda, preta, amarela, and indígena, or white, brown, black, yellow, and indigenous, respectively) during face-a-face interviews using standardized, pre-coded questionnaires. Participants were also asked to classify their own color according to the same census categories. The sample’s color distribution according to the interviewer was: 84% whites, 4.5% pardos, 11.3% pretos, and 0.2% indigenous ¹³. Meanwhile, self-classification showed 81.6% whites, 6.6% pardos, 10.8% pretos, 0.4% amarelos, and 0.6% indigenous ¹³. Such differences between the interviewer’s and interviewee’s classification are attributable to procedural variability or uncertainty.

Racial uncertainty: sampling

The fifth and last dimension, sampling variability, is purely statistical. Some variability in racial composition is due to the use of different samples. Even if time and place do not change, the population outcomes inferred from random samples change. The idea is simple and refers to one of the basic principles of frequentist statistics: different samples lead to different results. This requires measuring the uncertainty associated with this process in order to have some confidence in the results.

Traditionally, the most widely used statistical method for measuring this type of uncertainty is estimation of confidence intervals for the target values calculated from the samples. Confidence intervals are a range of values within which the population parameter is expected to be represented with some degree of certainty. When the researchers find the proportion of self-classified pretos in Pelotas in 2005 ¹³ to be 10.8% with a 95% confidence interval from 8.7% to 12.8%, they are saying the following: in the study sample, the proportion of pretos was 10.8%, while there was a 95% likelihood that the true population proportion was between 8.7% and 12.8%. Furthermore, if it were possible to select numerous samples from the same population of Pelotas in 2005, we should expect each of them to produce a slightly different proportion of pretos, but that 95% of the time the confidence intervals generated by the samples would include the population parameter. This same reasoning applies not only to a direct measure such as this - the proportion of pretos in Pelotas -, but also to estimates of inequalities from comparisons between groups. Therefore, the principal function of sampling uncertainty is to have the researcher acknowledge that the measures of racial inequality vary as a function of the samples from which they were calculated. Although such sampling fluctuation exists, it is understood to be random in probabilistic samples and can be quantified using confidence intervals.

Racial volatility and the construction of inequalities

In this section, we demonstrate that the uncertainties involved in two of the five proposed dimensions are not sufficient to affect the magnitude of racial inequality. We will address procedural uncertainty and its effect first, and then geographic uncertainty.

Procedural uncertainty has little effect on the size of racial inequalities, whether for consumption, income ³⁴, or health outcomes ³⁵. When individuals whose race changes according to the data collection methodology (photo, interview, interviewer, or ancestry) are excluded from the calculation of interracial inequality in consumption, the latter increases by only seven percentage points, showing that inequality between whites and non-whites is not influenced to a large extent by the data collection method. Interracial inequality in consumption as a percentage of total inequality increases from 11% to 18% when we exclude from the sample all the individuals whose race, as collected, was not the same according to all four methodologies used by the PESB survey in 2002. The same is true for interracial inequality in income, increasing from 13% to 19% of total inequality. However, these increases are not statistically significant. This result corroborates evidence from previous studies ^32,33. However, the irrelevance of procedural uncertainty for measures of inequality is not sufficiently robust to be a consensus. The importance of the mode of classification for measures of inequality appears to depend not only on the methodological choices, but also on the time period, place, and target variable. In the United States, young people perceived as blacks have higher odds of being arrested when compared to those perceived as non-blacks ³⁶. The racial gap in income was also reported as higher when race was classified by interviewers when compared to self-classified, both in the United States ³⁷ and in Brazil ^13,38.

For specific health-related racial inequalities, the data from Pelotas ^13,15 are also useful for demonstrating that procedural uncertainty is important, but has a small effect on the gap between whites and blacks. As mentioned, in addition to being classified by the interviewers, participants themselves were asked to choose one of the five race/color categories used in the Brazilian census. This created a unique opportunity to compare the distribution of health conditions and behaviors according to racial self-classification and interviewer-classification of the same interviewee. For example, in 2005 the smoking rate was 25.9% in whites and 31.2% in blacks when color was classified by the interviewer. Based on self-classification, the smoking rate was 25.8% in whites and 31% in blacks. Prevalence of tooth pain showed similar results, with 16.5% and 16.2% for interviewer-classified and self-classified whites, and 24% and 24.1% for interviewer-classified and self-classified blacks, respectively. In short, the change in racial classification procedures (self-classification versus interviewer-classification) did not affect the size of racial inequality in health conditions or behaviors, contrary to findings in another setting ³⁹.

Meanwhile, the role of geographic uncertainty in racial inequalities can be measured with the following question: How would interracial income equality vary if Brazilians were classified according to standard racial perceptions in the South versus North of the country? To answer this question, we used a specific technique ⁴⁰ to simulate how income distribution would change if Brazilians were racially classified according to the standards from the South versus North of the country. Figure 3 shows the results of this simulation.

The left-hand column in Figure 3 shows that 88.6% of total inequality can be attributed to income differences between individuals belonging to the same racial group. The other component (11.4%) shows the percentage of inequality attributable to the mean income difference between whites and blacks. The other columns represent statistical simulations in which the mean individual attributes are kept constant, and only the target characteristics or coefficients (in this case, region of the country) vary. If self-classified race were only a function of the target attributes (parents’ skin color, for example), the interracial component of inequality would increase from 11.4% to 16.4%. These percentages are the equivalent of mean per capita incomes that are 2.2 and 2.5 times higher for whites than for blacks. If race were treated as a latent variable, i.e., only as a function of the country’s and the individual’s socio-demographic and economic attributes, interracial inequality would be slightly higher than actually observed in 2010.

Figure 3
Intra- and interracial components of per capita income inequality. Brazil, 2010.

Simulations performed with the specified models show that if individuals were classified according to racial classification standards in the South of Brazil, the weight of the interracial component in total inequality would be slightly lower (15.7%) than if the racial classification standard from the North were used (18%). These percentages lead to mean per capita incomes 2.5 and 2.6 times higher, respectively, for whites than for non-whites. In short, the simulations show that if racial classification in Brazil followed the classification standards from the North of the country, racial inequality would be slightly higher than that actually observed. Meanwhile, if the race classification standard from the South of the country were adopted, racial inequality would tend to be slightly lower than expected. However, although the differences shown by the simulation indicate that racial inequality is also a function of the classification standard in a given region of the country, they do not substantially alter the persistent income gap between whites and blacks.

Conclusion

Since the race variable is socially constructed, it is permeated by five uncertainties: four linked to the context and one to the samples chosen to measure the skin color. Contextual uncertainty includes time, space, data collection methods, and all the other institutional and interpersonal factors linked to skin color reclassification. Meanwhile, sampling uncertainty refers to random fluctuations that are intrinsic to the sampling process.

In addition to suggesting a theoretical framework for the dimensions involved in racial variation, the results provide answers to the following questions: How variable is an individual’s race or color? What are the consequences of this racial variation for the interpretation of inequality? If these variations did not exist, or if the racially inconsistent individuals were excluded from estimates of inequalities, would the latter change substantially?

Between 1940 and 2010, the combined share of pretos and pardos in the Brazilian population increased from 35% to 50%. However, the dynamic of the composition of these groups is not due solely to their reproductive differentials, but also to the process of racial reclassification. For example, between 1990 and 2010 the population of pretos increased by more than 30% due to processes linked to skin color reclassification ^19,41. Evidence thus illustrates the contextual and temporal uncertainty associated with race.

Based on data from the PME, individuals initially self-classified as pardos are more prone to reclassify as whites in areas with a higher proportion of whites. Over a nine-month period, the likelihood of pardos reclassifying as whites in Porto Alegre was at least four times greater than in Salvador (Bahia State) or Recife (Pernambuco State). The variation in inter-regional classificatory tendencies shows the size of geographic racial uncertainties.

With regard to procedural racial uncertainty, evidence from the PESB 2002 shows that about half of the individuals had their self-classified skin color changed when the method for capturing the race variable changed. For example, only 3% of the total population consisted of pretos (“blacks”) if one simultaneously considers the IBGE census method, interviewer’s assessment, photographs, and direct ancestry. Meanwhile, consistent race classification according to all four methods as a proportion of the total population was 27% for whites and 19% for pardos ³¹.

Of the types of racial uncertainty examined here - contextual (temporal, geographic, procedural) and sampling - we evaluated the impact of two types on inequalities. The results suggest that neither geographic uncertainty nor procedural uncertainty (involved in the classification method) substantially affects the magnitude of racial inequalities in income, schooling, or consumption. However, further investigation is needed on the impact of contextual uncertainties (specifically those linked to the temporal dimension) on the dynamic of racial inequalities.

The study’s key message was to have shown that racial uncertainties exist in different dimensions and with different magnitudes. The existence of these uncertainties is what qualifies race as a socially constructed variable, i.e., as the product of ideals and interactions rather than a permanent and immutable characteristic ¹⁵. However, the racial uncertainties do not appear to be sufficient to change the understanding of another socially constructed reality, namely inequalities by race or color. In two of the five dimensions of racial uncertainty examined here, we found no changes worthy of note in inequalities in income, schooling, consumption, or health.

Since race is a contextual putative characteristic, it may not be “a reliable marker for assigning benefits and identifying differences” ⁴² (p. 278). However, the current study showed that although race is permeated with uncertainties that vary according to context, time, space, the sample, and the method, such variations appear not to compromise the overall picture of racial inequality. Given that the racial inequalities examined here are linked to socioeconomic and consumption dimensions, it is reasonable to assume that such disparities are maintained when analyzing health conditions and behaviors. Since economic, educational, and consumption disadvantages are at the basis of social inequalities in health, there is no reason to suppose that the privileges of whites are not also reproduced consistently in health. The article addressed a demonstration of this, showing the negligible impact of procedural uncertainty on disparities between whites and blacks in relation to smoking, alcohol abuse, and tooth pain.

It is still necessary to investigate the effect of the other classificatory uncertainties, especially those that occur over time, on racial inequalities in health. It should also be mentioned that there is some overlapping of the dimensions proposed in this typology, which probably hinders the empirical and isolated examination of the effects of each on racial inequalities. No less important is that the results presented in the article may not be readily generalizable to contexts other than Brazil. At any rate, the study showed that racial inequalities prevail and are perceived in Brazil, not so much as a function of racial uncertainties and methodological oscillations, but as the result of pernicious mechanisms in the maintenance of differences whose existence and effects we would like to mitigate.

Acknowledgments

This study was supported by the Brazilian National Research Council (CNPq), through research grants 307375/2014-5 (J.O.M.) and 303857/2015-3 (J.L.B.).

Referências

• ¹
  Chadarevian PC. Para medir as desigualdades raciais no mercado de trabalho. Revista de Economia Política 2011; 31:283-304.
• ²
  Hasenbalg C. Discriminação e desigualdades raciais no Brasil. 2ª Ed. Belo Horizonte: Editora UFMG; 2005.
• ³
  Henriques R. Desigualdade racial no Brasil: evolução das condições de vida na década de 90. Rio de Janeiro: Instituto de Pesquisa Econômica Aplicada; 2001.
• ⁴
  Osório RG. A desigualdade racial de renda no Brasil: 1976-2006. Brasília: Universidade de Brasília; 2009.
• ⁵
  Ribeiro CC. Classe, raça e mobilidade social no Brasil. Dados Rev Ciênc Sociais 2006; 49:833-73.
• ⁶
  Soares SD. Perfil da discriminação no mercado de trabalho: homens negros, mulheres brancas e mulheres negras. Brasília: Instituto de Pesquisa Econômica Aplicada; 2000.
• ⁷
  Silveira LS, Muniz JO. Variações intra e intermetropolitanas da desigualdade de renda racial. Cadernos Metrópole 2014; 16:265-89.
• ⁸
  Soares SD. A trajetória da desigualdade: a evolução da renda relativa dos negros no Brasil. In: Theodoro M, editor. As políticas públicas e a desigualdade racial no Brasil. Brasília: Instituto de Pesquisa Econômica Aplicada; 2008. p. 119-30.
• ⁹
  Soares SD. A demografia da cor: a composição da população brasileira de 1890 a 2007. In: Theodoro M, editor. As políticas públicas e a desigualdade racial no Brasil. Brasília: Instituto de Pesquisa Econômica Aplicada; 2008. p. 97-117.
• ¹⁰
  dos Anjos G. A questão "cor" ou "raça" nos censos nacionais. Indicadores Econômicos FEE 2013; 41:103-18.
• ¹¹
  Nobles M. History counts: a comparative analysis of racial/color categorization in US and Brazilian censuses. Am J Public Health 2000; 90:1738-45.
• ¹²
  Roth WD. The multiple dimensions of race. Ethn Racial Stud 2016; 39:1310-38.
• ¹³
  Bastos JL, Peres MA, Peres KG, Dumith SC, Gigante DP. Socioeconomic differences between self- and interviewer-classification of color/race. Rev Saúde Pública 2008; 42:324-34.
• ¹⁴
  Schwartzman LF. Does money whiten? Intergenerational changes in racial classification in Brazil. Am Sociol Rev 2007; 72:940-63.
• ¹⁵
  Bastos JL, Dumith SC, Santos RV, Barros AJ, Del Duca GF, Goncalves H, et al. Does the way I see you affect the way I see myself? Associations between interviewers' and interviewees' "color/race" in southern Brazil. Cad Saúde Pública 2009; 25:2111-24.
• ¹⁶
  Kabad JF, Bastos JL, Santos RV. Raça, cor e etnia em estudos epidemiológicos sobre populações brasileiras: revisão sistemática na base PubMed. Physis (Rio J.) 2012; 22:895-918.
• ¹⁷
  Jaccoud L. A política de promoção da igualdade racial no governo federal em 2006 e o programa de combate ao racismo institucional. In: Jaccoud L, editor. A construção de uma política de promoção da igualdade racial: uma análise dos últimos 20 anos. Brasília: Instituto de Pesquisa Econômica Aplicada; 2009. p. 147-70.
• ¹⁸
  López LC. O conceito de racismo institucional: aplicações no campo da saúde. Interface Comun Saúde Educ 2012; 16:121-34.
• ¹⁹
  Telles E. Demography of race in Brazil. In: Saénz R, Embrick DG, Rodríguez N, editors. The international handbook of the demography of race and ethnicity. New York: Springer; 2015. p. 151-67.
• ²⁰
  Miranda V. Measuring racial self-identification over the life course in Brazil, 1940-2013. Philadelphia: University of Pennsylvania; 2013.
• ²¹
  Francis AM, Tannuri-Pianto M. Using Brazil's racial continuum to examine the short-term effects of affirmative action in higher education. J Hum Resour 2012; 47:754-84.
• ²²
  Francis AM, Tannuri-Pianto M. Endogenous race in Brazil: affirmative action and the construction of racial identity among young adults. Econ Dev Cult Change 2013; 61:731-53.
• ²³
  Francis-Tan A, Tannuri-Pianto M. Inside the black box: affirmative action and the social construction of race in Brazil. Ethn Racial Stud 2015; 38:2771-90.
• ²⁴
  Aranha AS, Pena CS, Ribeiro SR. Programas de inclusão na UFMG: o efeito do bônus e do Reuni nos quatro primeiros anos de vigência - um estudo sobre acesso e permanência. Educ Rev 2012; 28:317-45.
• ²⁵
  Penner AM, Saperstein A. How social status shapes race. Proc Natl Acad Sci U S A 2008; 105:19628-30.
• ²⁶
  Aquino EM, Barreto SM, Bensenor IM, Carvalho MS, Chor D, Duncan BB, et al. Brazilian Longitudinal Study of Adult Health (ELSA-Brasil): objectives and design. Am J Epidemiol 2012; 175:315-24.
• ²⁷
  Doyle JM, Kao G. Are racial identities of multiracials stable? Changing self-identification among single and multiple race individuals. Soc Psychol Q 2007; 70:405-23.
• ²⁸
  Liebler CA, Rastogi S, Fernandez LE, Noon JM, Ennis SR. America's churning races: race and ethnic response changes between Census 2000 and the 2010 Census. Washington DC: Center for Administrative Records Research and Applications, U.S. Census Bureau; 2014. (CARRA Working Paper, 2014-09).
• ²⁹
  Saperstein A, Penner AM. Racial fluidity and inequality in the United States. Am J Sociol 2012; 118:676-727.
• ³⁰
  Ribas RP, Soares SD. Sobre o painel da Pesquisa Mensal de Emprego (PME) do IBGE. Rio de Janeiro: Instituto de Pesquisa Econômica Aplicada; 2008.
• ³¹
  Muniz JO. Preto no branco? Mensuração, relevância e concordância classificatória no país da incerteza racial. Dados Rev Ciênc Sociais 2012; 55:251-82.
• ³²
  Bailey SR, Loveman M, Muniz JO. Measures of "race" and the analysis of racial inequality in Brazil. Soc Sci Res 2013; 42:106-19.
• ³³
  Loveman M, Muniz JO, Bailey SR. Brazil in black and white? Race categories, the census, and the study of inequality. Ethn Racial Stud 2012; 35:1466-83.
• ³⁴
  Muniz JO. Inconsistências e consequências da variável raça para a mensuração de desigualdades. Civitas 2016; 16:e62-e86.
• ³⁵
  Travassos C, Laguardia J, Marques PM, Mota JC, Szwarcwald CL. Comparison between two race/skin color classifications in relation to health-related outcomes in Brazil. Int J Equity Health 2011; 10:35.
• ³⁶
  Penner AM, Saperstein A. Disentangling the effects of racial self-identification and classification by others: the case of arrest. Demography 2015; 52:1017-24.
• ³⁷
  Saperstein A. Capturing complexity in the United States: which aspects of race matter and when? Ethn Racial Stud 2012; 35:1484-502.
• ³⁸
  Telles EE, Lim N. Does it matter who answers the race question? Racial classification and income inequality in Brazil. Demography 1998; 35:465-74.
• ³⁹
  Boehmer U, Kressin NR, Berlowitz DR, Christiansen CL, Kazis LE, Jones JA. Self-reported vs administrative race/ethnicity data and study results. Am J Public Health 2002; 92:1471-2.
• ⁴⁰
  Mookherjee D, Shorrocks A. A decomposition analysis of the trend in UK income inequality. Economic Journal 1982; 92:886-902.
• ⁴¹
  Miranda V. A resurgence of black identity in Brazil? Evidence from an analysis of recent censuses. Demographic Research 2015; 32:1603-30.
• ⁴²
  Muniz JO. Sobre o uso da variável raça-cor em estudos quantitativos. Revista de Sociologia e Política 2010; 18:277-91.

Publication Dates

• Publication in this collection
  2017

History

• Received
  15 May 2016
• Reviewed
  21 Nov 2016
• Accepted
  09 Jan 2017