Interrogating Race

Racial Categories Versus Color Scales

As a window into many of the issues surrounding the color scale versus racial category debate, we begin with more detail on the dueling perspectives offered in the ERS symposium. Banton’s focal article argues strongly against the use of “race” as a concept in the social sciences and against the use of racial categories for the study of social inequality in Latin America. Some of his reasoning against “race” includes the following: it is a collectivist construct based largely on a U.S. folk vision of human diversity, it has no basis in biology, and it hampers cross-national comparison due to its varying localized meanings. Other scholars express related analytic concerns (e.g., Bailey et al., 2013; Bourdieu & Wacquant, 1999; Hirschman, 2004; Loveman, 1999; Monk, 2013, 2014). Instead of racial categories, Banton argues in favor of using a skin color scale.

Color scales and racial categories are both often considered ascribed characteristics. However, the two may not be easily interchangeable or equally efficient in capturing inequality structured by phenotype. According to Banton (2012), skin color along a continuum is “the primary form of differentiation on the basis of outward appearance”; however, it is just “one of the constituent elements that is taken into account” in structuring social inequality (p. 1111). In contrast, he argues that racial categories mark “a breach” in the skin color continuum (p. 1111), thereby possibly weakening analyses of the effects of outward appearance on social inequality. According to Banton (2012), then, an important sociological problem arises when scholars of Latin America “assume that the attribution of social significance to differences of color is better described as racial distinction” and move from an analytic focus on individual-level skin color variation to collective racial constructs (p. 1116). What is more, beyond breaching color scales and thereby possibly hampering the analysis of inequality, he adds that the “idiom of race” can “create” the discontinuities or racial categories it claims to simply measure (p. 1112); in contrast, “the idiom of color” merely “recognizes” a continuum of differences among individuals (p. 1112).

Latin America’s Trend Toward Racial Categorization

Due to his preference for color over racial category for the analysis of social inequality in Latin America, Banton takes a negative view of the important trend among Latin American states, and among scholars of that region, of movement away from color as a meaningful descriptor of phenotype diversity and toward racial categories. This shift may be most evident in the increasing use of racial categories in national censuses across the region (Loveman, 2014). Scholars attribute this shift toward an embrace of racial categories in Latin America to various factors: social movement mobilization, requirements of international funding agencies (such as the World Bank), the intervention of international human rights organizations, U.S. academic and cultural influence, and state intervention via race-targeted policy (Bailey, 2008; Banton, 2012; Bourdieu & Wacquant, 1999; Loveman, 2014; Nobles, 2000; Telles, 2004, 2012).

Although Banton posits that the scholarship of Edward Telles and Peter Wade has contributed to the state embrace of racial categories in Latin America, the core of his critique involves what he views as unsound analytic reasoning in the scientific community. Beginning here with Wade (2009), Banton goes straight to the crux of his issue: he charges Wade with the wholesale adoption of racial categories as more appropriate than color scales for scientific analysis in Latin America. This is the mirror opposite of Banton’s position. In his contribution to the symposium, Wade (2012) vigorously defends this analytic preference for racial categories. Perhaps most importantly, he views color as “too narrow” of a measure (p. 1171). According to Wade, racial categories do involve color, but also other physical features (like hair), as well as cultural traits. Hence, racial categories are “more powerful that skin color” as analytic measures (p. 1171). He elaborates:

a focus on skin color might well alert us to the fact that a darker skin color may be the target of discrimination in Latin America, but it would not alert us to the differences between people classed as índio and negro (or indeed mestizo or pardo), who happen to have the same skin color. To achieve this, we need to be alive to what I would call racial categories. (p. 1172)

Furthermore, in response to Banton’s criticism that the scientific use of race uncritically mimics a U.S. approach and folk concept, Wade (2012) counters that “race” is in fact “global in scope.”

Banton’s more extensive critique of Telles (2004) targets his tendency to “conflate” racial categories and skin color, treating these as interchangeable. Telles (2004, p. 79), for example, writes, “color [in Portuguese] captures the Brazilian equivalent of the English language term race.” Likewise, as part of the ERS symposium, he adds, “We could exchange the words race and color and we could come to the same conclusions” (Telles, 2012, p. 1166). Although some of Telles’s writings appear to lend themselves to Banton’s characterization of his scholarship on the interchangeability of race and color, Telles’s position is more complex, as he argues in his reply to Banton (Telles, 2012) and illustrates in much of his work. Telles’s scholarship in the past two decades demonstrates the utility of multiple ethnoracial measures for understanding social inequality in Brazil (and for most of Latin America). He first modeled this approach in his 1998 exploration of how self- versus other-classification by racial category produces contrasting estimates of social inequality (Telles & Lim, 1998). In doing so, he moved the field toward leveraging multiple measures for untangling some of the complexity of how racial stratification operates.

Moreover, because of his interest in skin color as a core element in discrimination dynamics, Telles recently collaborated in a large survey project that adopted a skin color scale for the first time in national samples across Latin America. Based on analyses of these data, the principal conclusion in his new book, Pigmentocracies (Telles, 2014), is not only that racial categories and skin color are not interchangeable, but that skin color in Latin America does a better job of capturing social inequality than do racial categories. In this respect, then, Telles and Banton appear to both endorse the primacy of color scale as a first-order approach for the analysis of racial inequality in Latin America.

Social Class and Outward Appearance

Banton’s view on the analytic primacy of skin color over racial categories is consistent with research on Latin American positing 1) the relative fluidity of racial category distinctions in the region, and 2) the larger role of social class over, and in mediating, skin color effects on social entitlement (Telles, 2004; Torche, 2014; Wagley, 1952). The centrality of social class in Latin America may reflect its history as one of the world’s most unequal regions in terms income and wealth (Torche, 2014), while the fluidity of categorical distinctions may follow the abovementioned relative absence of de jure race category assignment and discrimination. Whether racial categories and color work differently in the United States and Latin America in terms of their relationships to social class for determining social entitlement is understudied due in large part to the lack of comparable data.

“Social origin” may be a particularly important lens for understanding of the role of social class in relation to ethnoracial markers. Social origin generally references the socioeconomic standing of parents (origins) in determining the socioeconomic standing of their adult children (destinations), that is, intergenerational social mobility (Mare, 2011; Torche, 2014). According to Hout (2015), “We ask about mobility because we care about opportunity and fairness and think if society provides opportunity and does so fairly, then more people will move up as time goes on” (p. 28). Illustrating that maxim through a negative case, early studies on inequality in the United States revealed that upward social mobility was thwarted for African Americans through the absence of opportunity and fairness specifically for that subpopulation (via de jure racial discrimination), thereby washing out the effect of parents’ social class (Blau & Duncan, 1967).

Current research suggests that the United States and Latin America may not differ much regarding the centrality of social origins in structuring social inequality in the past two decades. According to Torche (2014), there is a consensus in studies on social mobility in Latin America that “parental education is the most influential circumstance; ethnicity and region of birth have smaller roles” (p. 633; see also Torche & Spilerman, 2009). In the United States, some scholars posit that the intergenerational effect of parental education has replaced the dominance of race for determining social mobility (Marrero & Rodriguez, 2011).

Against this backdrop, we build on research exploring whether racial category or color scale best fits patterns of social inequality in the United States versus Latin America (cf. Bailey et al., 2014) by examining the degree to which social inequality structured by these racial markers is accounted for by the legacy of parental class (an individual’s social origin). To interrogate these questions, we estimate the gross effect of both racial categories and color scales on per capita household income across 19 countries of the Americas, including the United States. We then compare these results with estimates from models that control for the social origin of respondents in the 19 countries, allowing us to observe and compare how much change there is in color scale and racial category inequality when we account for social origin.

Data and Method

Our data are from the 2012 General Social Survey [GSS] in the United States and the 2014 AmericasBarometer [AB] surveys in Latin America. The 2012 GSS is a nationally representative sample of U.S. adults, aged 18 years and older, living in households; 1,974 people were interviewed. ² AB is part of the Latin American Public Opinion Project. Our analysis focuses on 18 countries from the following regions: Mexico/Central America, Andean/Southern Cone, and the Spanish-speaking Caribbean. The country surveys are nationally representative, face-to-face interviews of voting age adults. Sample sizes are approximately 1,500, with the exception of Bolivia, which is approximately 3,000. Samples in each country were developed using a multistage design and were stratified by major regions of the country, size of municipality, and by urban and rural areas within municipalities.

In all surveys, interviewers rated respondent skin color after concluding their interview using similar 10-point (GSS) or 11-point (AB) scales with visual color referents. Interviewer field manuals included a color card in the United States with 10 circles of varying skin colors, each corresponding to one of the points on the color scale. Similarly, the AB surveys used a color card with 11 color shades corresponding to points on a color scale. The color cards were not shown to respondents.

With regard to racial categories, GSS respondents were asked to self-identify their race in response to the question: “What is your race? Indicate one or more races that you consider yourself to be.” Up to three responses were recorded using the 15 categories used in the U.S. Census. Immediately prior to reporting their race, respondents were asked: “Are you Spanish, Hispanic or Latino/a?” Based on responses to these questions, we recoded respondents into the most commonly used, mutually exclusive racial categories for the United States.

The racial categorization variable in the AB survey was also based on self-identification. Respondents were asked: “Do you consider yourself White, mestizo, indigenous, negro, mulato, or other?” In all countries, the first part of the question (“Do you consider yourself . . . ”) was the same, but the response categories differed according to country schemes. For example, the Brazilian survey (in Portuguese) used national census categories: White, pardo (Brown or mixed), preto (Black), amarelo (Yellow or Asian origin), and indigenous. Other variations include Venezuela, which uses the term moreno, and Guatemala, where only the categories Ladino and Indigenous were used.

To capture social origin, we focus on mother’s education (Hout, 2015; Torche, 2014). Class status, like racial status, is often considered ascribed at birth. Using mother’s education to capture class status allows us to control for some advantage (or disadvantage) accumulated from parents’ class position. While we would ideally include other indicators of parental class as well, such as father’s education and parents’ occupation (cf. Hout, 2015), we focus on mother’s education because it allows us to retain a broad international perspective. Information about parental occupation, for example, was only available in two of the countries in the AB data, and only partially.

We chose per capita household income as our dependent variable. Although measures like hourly wage (net of overtime pay) can better address issues around discrimination, household income is often considered a better measure for understanding inequality more broadly as it includes differences resulting from factors like assortative mating (see Torche, 2011). Hence, according to Torche (2011), these attributes make total family income “perhaps the best measure of economic well-being” and specifically oriented toward capturing the “the transmission of advantage across generations” (p. 774). Household income is self-reported in the GSS using a list of 25 categories. Respondents are instructed to answer based on their “total family income, from all sources” for the previous calendar year. We recoded each category to its midpoint with the exception of the open-ended top category, which we assigned a value of $160,000 based on the same formula, described below, that we applied to each country in the AB data. (This coding will understate self-identified race or color inequality to the extent that “Whites” and lighter skinned Americans are overrepresented in the top category and have incomes substantially greater than $160,000.) We then used the count of persons in the respondent’s household to calculate per capita income. Household sizes ranged from 1 to 10. ³

The income measure in the Latin American surveys is self-reported using 16 intervals based on each country’s currency. Respondents were instructed to answer based on “the total monthly income of this household, including remittances from abroad and the income of all the working adults and children.” For each country, we assigned midpoint values to the first 15 intervals, and assigned the open-ended top category a value corresponding to the top value of the penultimate category plus half of the penultimate category’s range. To calculate per capita income, we used the count of persons in the respondent’s household, ranging from 1 to 20. ⁴

To examine per capita household income by racial categories and by skin color scales, as well as the degree to which social origin mediates these effects, we use ordinary least squares regression and four models across our 19 countries. Our outcome variable in all models is the natural logarithm of per capita household income. ⁵ The predictor variables in our models are as follows: color scale alone (Model 1), color scale and mother’s education (Model 2), racial category alone (Model 3), and racial category and mother’s education (Model 4). We present our results in four graphs. In the first, we graph average predicted values of per capita household income for each point on a country’s skin color scale, for points with 30 or more cases. ⁶ Skin Color Category 5 serves as our benchmark in each country; we present all other average incomes in relation to the value for that midrange color point. For each country, we plot not only predicted income by color point but also a second set of color points representing predicted values for color controlling for mother’s education. Comparing these two provides information about the degree to which color differences in income are independent of maternal education. In Figure 2, we highlight the degree to which social origins account for color inequality by graphing the change in the size of the differences when we examine color scale alone and controlling for social origins. Figures 3 and 4 follow the same format as our first two, but provide information on racial categories instead of color scale points.

Our approach is intended to highlight the overall level of per capita household income inequality and its relationship to skin color, racial category, and class at a broad comparative level across 19 countries. Hence, we do not control for other factors through which racial inequality might be mediated or reproduced in a given setting. Those considerations, such region, education, and marital status, are important for the purposes of identifying additional intervening factors, but our aim in this study is not to isolate a complete list of country-specific mechanisms through which racial inequalities arise. Introducing additional variables would cloud the analyses and make our broad comparison difficult to interpret (cf. Hout, 2015). For example, controlling for the respondent’s education would mean that we would be unable to differentiate between countries where there is no racial inequality and countries where there is substantial racial inequality that is mediated by education.

Findings

The graphed results of our regression models reveal several interesting aspects of how color scale and racial category operate in Latin America and the United States and the possible mediating role of social origin. To facilitate interpretation, we begin with a brief look at results for Argentina in Figure 1. The countries in all of our graphs are labeled at the top of columns and ordered by percentage White: largest (United States) to smallest (Nicaragua). Argentina is listed third. The column for each country contains two sets of results. First, on the left, we present the value of color points (relative to Color Point 5) from models using color alone; on the right, we present results from models controlling for maternal education. Examining the color points on the left for Argentina shows the lightest color point is at the highest position on the income scale, and that Argentines with this skin tone have average incomes that are nearly 75% greater than Argentines rated 5 on the color scale (the reference color point). In contrast, Argentines represented by the darkest color point have average incomes roughly 25% less than Color Category 5. The results in the right subcolumn for Argentina show that these differences are virtually identical when controlling for mother’s education, suggesting that the role of color in structuring income inequality is largely independent of mother’s education.

OPEN IN VIEWER

Figure 1.

Skin color, social origin, and income inequality across the Americas.

Looking broadly at the first subcolumn for each of the countries in Figure 1 reveals many striking similarities, but also differences. All but three countries (Costa Rica, Honduras, and Panama) provide evidence of the association of lighter color with higher per capita household income. ⁷ Argentina and Mexico exhibit a strikingly linear relationship between color and household income. In other countries, some color points are clustered, suggesting that there are parts of the color scale where variation matters less than others. The cases of the United States and Brazil illustrate clustering, as several darker color points have similar household incomes.

The second subcolumn in Figure 1 presents predicted household income by color point controlling for mother’s education. If social origins play an important role in mediating the effect of skin color in Latin America, we would expect these color points to be substantially closer to the zero line than those in the first subcolumn. Contrary to that expectation, Figure 1 suggests that controlling for maternal education does not substantially reduce skin color differences, so that the color differences with and without this control are relatively similar. The United States does not seem to be an exception to this broader Latin American pattern.

Figure 2 more succinctly summarizes how controlling for social origins affects color inequality in household income by graphing the relative change in the coefficients for color points when maternal education is introduced as a control. The country that shows the smallest impact of social origin on the relationship of color to income inequality is Argentina, with an average of roughly 2% difference between the gross effect of each color point and the effect of each color point controlling for maternal education. In the United States, there is generally a 4% difference between the results with and without controls for mother’s education. This indicates that color inequality in the United States operates largely independent of maternal education, and it is important to note that the United States is similar to many Latin American countries in the degree to which maternal education reduces the color gradient in household income. In particular, Brazil, Dominican Republic, Uruguay, and Nicaragua all see reductions in the magnitude of their color gradient when controlling for maternal education that are relatively similar to those observed in the United States. At the other end of the spectrum, Figure 2 shows that social origin has a greater impact on color inequality in El Salvador, Peru, and Panama. Countries such as Chile and Costa Rica have relatively little color-based inequality to begin with; so the changes that are observed in these countries when maternal education is introduced should be interpreted cautiously.

OPEN IN VIEWER

Figure 2.

Coefficient change for skin color after controlling for mother’s education.

Turning our attention to how racial categories operate across the Americas, in Figure 3 we plot each racial category’s average predicted per capita household income relative to the mestizo category (multiracial in the United States; pardo in Brazil; and Ladino in Guatemala). In examining the first subcolumn of category income estimates for each country, many of the color scale patterns we saw in Figure 1 are well reflected in racial category hierarchies in Figure 3. This suggests that color is a primary element in ethnoracial category construction, but with exceptions. Overall, White racial category privilege relative to other racial populations is clear in all but seven countries: the United States, Chile, Honduras, El Salvador, the Dominican Republic, Mexico, and Nicaragua; and in many of these, Whites are still near the top. ⁸

OPEN IN VIEWER

Figure 3.

Race categories, social origin, and income inequality across the Americas.

For those countries where the color scales estimated in Figure 1 do not seem to closely match an expected racial category hierarchical configuration in Figure 3, one plausible explanation is that this disconnect is evidence of the broader meaning of those social categories, and the more narrow meaning of skin color as an individual-level marker with regard to outward appearance (cf. Wade, 2012; Banton, 2012). In the United States, for example, the average skin color of Asian Americans is slightly darker than the average of White Americans, but the Asian racial category may capture more than color and stands in a position of privilege relative to Whites in the United States (although in our sample, the difference in per capita household fincomes between these two populations is not statistically significant). Likewise, the lower status of Whites in El Salvador, Mexico, and Nicaragua evinces the broader character of racial categories compared with skin color in relation to the income gradient. ⁹

Regarding the degree to which racial category inequality can be accounted for by social origin, Figure 3 echoes Figure 1 in that it suggests that racial category differences remain substantial even after accounting for maternal education. That is, the second subcolumn of racial category coefficients mirrors fairly closely the first subcolumn across all countries. Figure 4, though, shows that there are some, albeit somewhat limited, reductions in the racial category gaps when we control for maternal education. Similar to Figure 2, Figure 4 plots relative change in the coefficients for each racial category by country when controlling for social origin.

OPEN IN VIEWER

Figure 4.

Coefficient change for race categories after controlling for mother’s education.

We also see in Figure 4 that in many countries (e.g., Guatemala, Ecuador, Mexico, Peru, and Bolivia), accounting for social origin matters less for indigenous than for other categories. In general, though, controlling for maternal education in most countries does little to change racial category gaps: on the order of 3% and potentially up to 5% or 7%. One notable exception is Panama, where maternal education plays a substantially larger role in accounting for the racial gaps that we observe, at about 12% to 15%. This suggests that in Panama, more so than in other countries, racial category differences in household income are due to social advantages (and disadvantages) passed on through maternal education.

Discussion and Conclusion

We began this article by recognizing long-standing debate on the analytic utility of racial categories versus color scales in the study of social inequality. Our analyses address a key question at the core of this debate: whether and to what extent social class proxies mediate the relationship between both racial categories and color scales to income. In doing so, we also highlight the somewhat different understandings of inequality that are provided by examining racial categories and color scales in certain contexts.

At the broadest level, we provide evidence for the utility of both racial category and color scale in analyses of social inequality in both regions. It can no longer be easily argued that there is a simple dichotomy, that for interrogating social stratification, color scales are best in one region and racial categories in the other. Rather, our results suggest that instead of generalizing about two broad regions, we must go to the country level to discern the utility of these social constructs for understanding inequality (cf. Bailey et al., 2014). For example, a color scale maps onto household inequality in Argentina and Mexico, but does not in Costa Rica or Honduras. Racial categories clearly reveal the disadvantage of most countries’ indigenous, while in other countries, racial categories fail to capture meaningful variation in household income (e.g., in Colombia for Whites, mestizos, and mulattoes). Still in other national contexts, the use of racial categories, instead of capturing White privilege, reveals apparent non-White racial category advantage (e.g., in the United States and in Nicaragua). ¹⁰ These patterns are both variegated and revealing, and we suspect that future research disaggregating even further to examine within-country regional variation will reveal similar heterogeneity and highlight the importance of understanding racial categories and color scales in both broad as well as more localized contexts.

Regarding social origins, overall, our results show that they matter in the relationship of both color scales and racial categories to social inequality, though perhaps less than one might have expected. Given that we have only one measure of social origin, mother’s education, due to data limitations, our results should be considered conservative estimates of social origin effects. Here, it is perhaps most interesting to note that indigenous differences were particularly poorly explained by maternal education. This finding highlights the importance of understanding the contrasting dynamics that lead to the advantages and disadvantages among diverse social categories or subpopulations.

Do our findings suggest support for Banton’s critique of Latin American scholarship, or do they suggest support for the suppositions of the two scholars of Latin America he engages? Banton’s assertion of the appropriateness of a color scale over racial category in Latin America depends on what country you examine and for what purpose. If it is to capture meaningful variation in per capita household income, our findings suggest he is correct in a large number of contexts in Latin America (e.g., in Mexico) but clearly not in a few others (e.g., Honduras). Wade’s (2012) assertion that racial categories in Latin America are more powerful than skin color for understanding social stratification receives support only in a limited number of cases. For example, our results suggests he is right regarding racial categories in Costa Rica, where Figure 3 shows a clear hierarchical ordering of White racial category privilege, and where Figure 1 shows that a color scale fails to map that inequality. Telles’s (2014) recent work appears to place his position on the efficiency of color scales over racial categories for the analysis of social inequality in Latin America close to that of Banton (2012). Nonetheless, both Banton and Telles assert that in certain contexts, racial categories (or “color lines” in Banton’s idiom) may be appropriate analytic measures in tandem with color scales for the analysis of inequality by phenotype. The appropriateness of a categorical approach may stem in large part from de jure racial category definition and discrimination, as in the United States and South Africa. The rejection of equating color lines with “race” and “racial” categories, however, sets Banton apart from Telles, as well as from most scholars of racial stratification. Most importantly Banton’s core critique rests on the premise of the logical primacy of first-order over second-order abstractions in sound analytic reasoning. Hence, only through establishing the centrality of the first abstraction, skin color scale, could it be reasonable to move to the second, color lines (i.e., “racial” categories).

Several issues, however, are still pending empirical analysis, and this is where we hope to contribute to further debate. First of all, our analysis offers a wide view of the effects of skin color and racial categories in Latin America in explicit contrast to the United States. Hence, we have been able to engage the debate comparatively. Banton (2012) posits, for example, that the U.S. system of social stratification is becoming more similar to that of Latin America. In this, he appears to echo Bonilla-Silva’s (2004) thesis regarding the Latin Americanization of U.S. racial stratification. Whether the two systems of social stratification are converging over time is beyond our analysis (see, e.g., Daniel, 2006); our data and methods do not directly engage change in the U.S. or Latin American systems. Instead, we demonstrate that currently these two broad regions have much in common, and that there is no homogeneous “Latin American model” (cf. Bailey et al., 2014).

In addition, our analyses explore the specific question of the contribution of the intergenerational transmission of social class to contemporary inequality relative to both racial categories and color scales in the United States and in 18 countries of Latin America. That is, are racial category effects independent of social class proxies, such as social origin, in structuring contemporary inequality in the United States, whereas color is tied to a more complex interrelation with social class in Latin America? Put simply, we find little evidence that the United States is unique, either in the racial category gaps or color gradients.

Hence, our research can contribute to reorienting the field in at least two ways. First, instead of centering on assumed differences between the United States and Latin America regarding racial categories versus color scales, we should focus on expanding our analytic tool kit to clarify the persistent and cross-national connection of multiple ethnoracial markers to social inequality. Second, while it is clear that ethnoracial markers and social origins are intertwined, our results suggest that it remains nonetheless challenging to understand how they combine to create the patterns of inequality that we observe. Hence, it may be that expanded attention to this specific area is where scholarship can produce substantial new insights. In addition, greater attention to the conceptualization of “racial origin,” akin to social origin, and to empirical modeling for capturing its specific effects may be key.

In sum, employing a multidimensional approach with comparative data, we highlight that the United States not only falls well within the range of Latin American racial category and color inequality but also is generally similar in terms of the degree to which mother’s education helps explain these differences. Perhaps our most surprising finding, that we often think about racial inequality as being an intergenerationally accumulated (dis)advantage in both Latin America and the United States, is the relatively little change in the levels of racial category and color inequality observed after accounting for maternal education. Thus, while we robustly document the negative effects of darker skin color in the United States and in most of Latin America, as well as patterns of White category advantage in most contexts, our results suggest that there is much about patterns of racial inequality that we do not yet understand.