Abstract
Racial distinctions in the United States have long been characterized as uniquely rigid and governed by strict rules of descent, particularly along the black-white boundary. This is often contrasted with countries, such as Brazil, that recognize “mixed” or intermediate racial categories and allow for more fluidity or ambiguity in racial classification. Recently released longitudinal data from the IPUMS Linked Representative Samples, and the brief inclusion of a “mulatto” category in the U.S. Census, allow us to subject this generally accepted wisdom to empirical test for the 1870–1920 period. We find substantial fluidity in black-mulatto classification between censuses—including notable “downward” racial mobility. Using person fixed-effects models, we also find evidence that among Southern men, the likelihood of being classified as mulatto was related to intercensal changes in occupational status. These findings have implications for studies of race and inequality in the United States, cross-national research on racial classification schemes in the Americas, and for how demographers collect and interpret racial data.
Introduction
More than 40 years ago, historian Carl Degler outlined a provocative comparison of race relations in Brazil and the United States. The crux of his argument about then-contemporary differences between the two countries rested on the relative status of “mulattos.” Specifically, Degler claimed that the progeny of unions between black and white Brazilians were accorded an intermediate position in the social and racial hierarchy: “The mulatto in Brazil represents an escape hatch for the Negro, so to speak, which is unavailable in the United States” (Degler 1971:107). More controversial, still, is the related and oft-repeated assertion that Afro-Brazilians can avail themselves of this “escape hatch” not only across generations by marrying lighter-skinned spouses but thanks to “the ability of wealth and education to whiten” within a single generation. As Degler put it: “Once ‘whitened’ by money, a ‘Negro’ becomes a ‘mulato’ or ‘pardo,’ regardless of his actual color” (Degler 1971:107–08; emphasis in the original).
The ensuing scholarly debate has focused on whether Degler’s notion of an escape hatch was an accurate description of the Brazilian racial hierarchy, with its absence in the United States largely taken for granted. Researchers have come to varying conclusions regarding whether the situation of lighter-skinned or mixed-race Afro-Brazilians represents a meaningful improvement, materially or otherwise, compared with that of their darker-skinned counterparts (Loveman et al. 2012; Sheriff 2001; Telles 2004). Consensus regarding the claim that “money whitens” has also been elusive because of the lack of nationally representative, longitudinal data on race and socioeconomic status (SES) in Brazil (although, see Schwartzman 2007). In the United States, some “passing”—that is, when people with African ancestry hide their full family history to take advantage of their “white” appearance—was and is publicly acknowledged (e.g., Gates 1997; Johnson 1925), but it has generally been considered the exception rather than the rule of racial classification and social mobility. Yet, nationally representative, longitudinal data on the racial classification and SES of individuals do exist in the United States that could provide direct, systematic evidence on these issues. Research using the 1979 National Longitudinal Survey of Youth (NLSY) shows that social status and racial fluidity are linked in contemporary America: increases in status increase the odds of being classified as white and decrease the odds of being classified as black, and decreases in status decrease the odds of being classified as white and increase the odds of being classified as black (Saperstein and Penner 2012). Thus, regardless of whether the “mulatto escape hatch” is—or ever was—an accurate description of racial stratification in Brazil, it has become pertinent to ask whether increases in social position ever led to increases in racial position among Americans of African ancestry.
Recently released historical linked census samples from the Minnesota Population Center allow us to answer this question. These data provide fresh insight into the era of racial retrenchment following the Civil War and Reconstruction, and bracketing the turn of the twentieth century—a period when “Jim Crow” laws and the “one-drop rule” dictating racial classification were slowly building up steam in the South,1 even as the U.S. Census was going to great lengths to count the mixed ancestries of Americans. In this context, we find substantial fluidity in mulatto classification between censuses. We also find evidence for a recursive relationship between racial and social status, from person fixed-effects models estimating whether shifts in racial classification between “black” and “mulatto” are associated with changes in occupational position or other changes in individual- and county-level characteristics.
Our analysis indicates that early in the post-Reconstruction South, men with African ancestry could “escape” blackness through occupational upward mobility. The results also suggest that this “hatch” did not operate only in one direction, as the original Brazilian formulation of the concept implies. In the United States during this period, social and racial mobility was a revolving door that maintained a relatively small number of more privileged positions that could be gained, as well as lost, over the course of one’s life. These findings call into question much of the received wisdom regarding the rigidness of racial distinctions in the United States, and further complicate comparisons of the moving targets that are race relations in the United States and Brazil. They also imply that the pursuit of “reliable” racial data may come at the expense of validity and a deeper understanding of social inequality.
The Mixed Status of “Mulattos” in the United States
Although it has been out of official use for nearly 100 years, a categorical black-mulatto distinction was once salient in the United States, in both the census and popular perceptions. In the mid-nineteenth century, an influential group of American scholars and politicians believed that racial mixture between blacks and whites created inferior, sickly, and potentially sterile progeny. To test the theory that the extinction of such “racial hybrids” was immanent, the category “mulatto” was added to the census forms in 1850 for both the slave and free populations. The category continued to be included as an official racial designation in six of the next seven decennial counts, although the validity and reliability of the resulting population estimates were hotly debated, even among high-ranking census administrators. (See Nobles (2000) for more details on changing racial classifications in the U.S. census.)
The instructions and category descriptions provided to census enumerators during this period defined the difference among white, black, and mulatto Americans as a matter of descent, or “blood,” divisible into fractions: one-half for each parent, one-quarter for each grandparent, and so on. In 1870 and 1880 (and similarly in 1910 and 1920), mulatto was defined as a “generic” mixed-race category that included “. . . all persons having any perceptible trace of African blood.” Enumerators were cautioned that “important scientific results” rested on the proper reporting of this classification (U.S. Census Bureau 2002:14). The Congressionally mandated study of the “amalgamation of human species” culminated in 1890 with a detailed classification of “half breeds,” which included the categories “quadroon” (one-fourth African ancestry) and “octoroon” (one-eighth African ancestry), along with “mulatto” and “black” (Nobles 2000:55–56). That year, enumerator instructions carefully delineated that mulattos had “from three-eighths to five-eighths black blood” (U.S. Census Bureau 2002:27).
At least initially, these racial distinctions were thought to be “immediately accessible and knowable,” with even the specific blood quanta assumed to be visible to enumerators (Nobles 2000:62). Over time, however, criticisms of data inaccuracy mounted—perhaps not coincidentally, given that the counts showed no evidence of population decline. In 1880, the full count of mulattos was not included in published census tabulations. The categories of “quadroon” and “octoroon” did not survive beyond the 1890 census, and “mulatto” had a brief hiatus in 1900. The category returned in 1910, and an unexpected doubling in the mulatto population, presumed to be impossible because of widespread anti-miscegenation sentiments and the extinction hypothesis, became attributed to an increase in the number of black enumerators employed that year (Nobles 2000:68). This perceived lack of reliability in the counts, along with the eventual widespread adoption of the one-drop rule prompted the Census Bureau to remove the category of “mulatto” after the 1920 count.
Shifting Boundaries
In contrast to historical assertions of mulatto inferiority, contemporary scholarship has highlighted the relative privilege of mulattos in the United States, compared with other Americans with African ancestry. This work also identifies our study period—namely, the decades between the end of Civil War (1865) and the early twentieth century—as a period of flux in U.S. racial boundaries, particularly those distinguishing mulattos from blacks.
Distinctions among Americans with African ancestry were more widely recognized in the Lower South than in the Upper South, particularly in port cities such as New Orleans (Louisiana), Charleston (South Carolina), and Mobile (Alabama), where close economic ties to the Caribbean and colonial ties to France and Spain may have contributed to the early acceptance of a mixed-race category (Sowell 1978; Williamson 1995). Mulattos were also overrepresented among the “free colored” population prior to the American Civil War (Berlin 1974), and tended to occupy more advantaged positions than blacks in the postwar labor market (Bodenhorn and Ruebeck 2007; Ruef and Fletcher 2003). Following emancipation, free people of color had a clear incentive to distinguish themselves from former slaves in order to maintain their social position, marginal though it may have been. As a result, the mulatto-black divide became a partial substitute for the former division between free and slave. Perhaps the most well-known example of this calculated differentiation during the Reconstruction period was the proliferation of elite “blue-vein” societies, whose alleged criterion of membership was skin light enough to reveal the veins on one’s arm (Russell et al. 1992).
However, these ostensibly color-coded distinctions were not only recognized among Americans with African ancestry. Without the slave institution to maintain white supremacy, the black-mulatto distinction became particularly salient and potentially problematic for whites (Hodes 1997). Southern white governance strategies eventually coalesced around racial segregation enforced by the one-drop rule, but unanimity developed gradually and was not fully established until the early twentieth century (Williamson 1995). During Reconstruction, mulattos were seen alternately as a threat to white domination and the developing system of legal exclusion, or as a potential buffer class, occupying a functional middleman position, capable of defusing racial conflict.
An analysis of 1880 census data explains this seeming contradiction as a function of local demographic variation (Gullickson 2010). Mulattos held a greater occupational advantage over blacks in areas where whites were both fewer and more privileged. Thus, Gullickson (2010) argued, where whites were more advantaged, emphasizing a black-mulatto boundary created an effective buffer against the threat of a large population of undifferentiated blacks; where the status of whites was lower, the intermediate position of mulattos threatened the status of whites and overrode the value of a more variegated racial hierarchy.
The turning point in support for upholding distinctions among Americans of African ancestry might have been the Supreme Court verdict in Plessy v. Ferguson (1896). The plaintiff, Homer Plessy, was described in court documents as an “octoroon,” and case histories suggest that the New Orleans native was chosen as the test case for Louisiana’s separate car law because he was “white enough to gain access to the [whites only] train [car] and black enough to be arrested for doing so” (Medley 2003:17). The court’s ruling against Plessy gave implicit constitutional support to racial segregation in public facilities, services, and transportation across the United States for the next half-century (Lofgren 1988).
The Plessy decision also exemplifies the argument that the United States drew uniquely “bright” racial boundaries (cf. Alba 2005), especially the one separating “blacks” from “whites.” Not content to let appearance guide racial classification, white Americans began to seek stricter guarantees of racial purity. The dawn of the twentieth century saw Southern states adopt increasingly expansive descent-based definitions of blackness: from one-half, to one-quarter, to one-eighth, and finally to “any perceptible trace of black blood” (Davis 2001). As white support for the one-drop rule increased, lighter-skinned elites within the African-ancestry population shifted their political tactics to emphasize unity, downplaying categorical distinctions between mulatto and black (Williamson 1995). Thus, whites, mulattos, and blacks all came to support the one-drop rule governing white-black racial classification because it offered a partial overlap of interests (Wimmer 2008). For whites, it served as a primary bulwark of the Jim Crow system and continued white supremacy; for blacks and mulattos, the combination of a more inclusive category marked by a more rigid racial boundary helped to generate stronger group solidarity and resistance to that very same system.
Analytic Approach
One of the many consequences of the one-drop rule of racial classification is that it creates a false sense of security—or, in the case of demographic measurement, stability. In theory, if one’s race is determined by one’s ancestry or descent, it can be ascribed at birth and fixed. In practice, racial perceptions and classification were, and are, much messier.
We take advantage of the brief inclusion of the mulatto category on the U.S. Census to explore the existence of fluidity in racial classification between 1870 and 1920, using linked microdata from sequential censuses that capture change over time in the racial and social status of individuals. These unique data allow us to explore the possibility of a “money whitening” effect, whereby increases in social status facilitated upward racial mobility from the black to the mulatto category.2 Put another way, we ask whether the higher status of mulattos in the postbellum United States was only the result of inherited privilege, or whether it also could be earned through successful social mobility.
Williamson (1995:80–82) suggested that a causal relationship between social and racial status might have existed in the early years after the Civil War—when educated mulattos from the North moved south to teach, govern, and minister to the newly freed slaves, and “freedom generated prosperous blacks” who were then absorbed into the mulatto elite. However, Williamson provided no direct empirical support for the claim. As in Brazil, an absence of longitudinal data has precluded a rigorous investigation of the hypothesis that social mobility facilitated racial mobility in the United States during this time period—until now.
With pairs of observations for the same individual, we can begin to untangle the direction of causality. If, for example, being lighter-skinned gave Americans with African ancestry advantages only through inherited privilege and color preferences, then increases in social status should not be related to changes in racial status between censuses. However, if higher social status also made one more likely to be perceived as mulatto, then an increase in social status between censuses would be related to an increase in the likelihood of being classified as mulatto in the next enumeration.
Data limitations, discussed in the next section, prevent us from tracking a representative sample of Americans across the black-white color line and examining the full range of racial fluidity and its relationship to social status. However, previous scholars (e.g., Gullickson 2010; Williamson 1995) have argued that something like a “triracial order” (cf. Bonilla-Silva 2004) took shape during this era, particularly in the South. Thus, we assume that “mulatto” was a middle ground between “white” and “black”—both in terms of racial or ancestral status and social status. Of course, if we find that increases in social status lead to higher odds of mulatto classification, this would support the idea that mulattos were not only seen as the products of racial mixture—an aspect of one’s background that would not change from one census to the next—but as somehow “better” than their black counterparts. That is, although poor mulattos and rich blacks certainly existed, such evidence would suggest that the cognitive schema or stereotype that Americans used to define racial group membership during this era (cf. Brubaker et al. 2004) placed mulattos higher on the racial hierarchy than blacks.
Data
The IPUMS Linked Representative Samples match individuals from the 1850–1930 IPUMS census samples to the full-count data from the 1880 census, allowing for longitudinal analysis of census data (Ruggles et al. 2010). We focus on the linked samples between 1870 and 1880, 1880 and 1910, and 1880 and 1920. The 1850 and 1860 samples do not include slave populations and so are of limited use to us, and the “mulatto” option was removed from both the 1900 and 1930 census forms. The census data from 1890 are not part of the IPUMS samples because the original files were destroyed in a fire.
Linkage Procedure
Individual record linkages between censuses were made on the basis of five variables: birth year, place of birth, given name, surname, and race. Although race was used in the linking algorithm, coding distinctions were not made between those who were classified as black and those who were classified as mulatto. Thus, the linkage procedure used in the IPUMS samples effectively eliminates any potential switching between white and black/mulatto (by defining it as a false link), but allows for switching between black and mulatto. This would bias our results to the extent that mulattos were either periodically or eventually classified as white. Scholars at the time speculated that tens and possibly hundreds of thousands of “negroes” and “mulattoes” were passing into the white category during this era (Johnson 1925), although later research suggested the number might have been much lower (Burma 1946; Eckard 1947). Either way, if being reclassified as white was associated with gains in status, our conclusions regarding the existence of a “mulatto escape hatch” and the potential for money to “whiten” in the United States should be considered conservative.
The Minnesota Population Center (MPC) used the Freely Extensible Biomedical Record Linkage software (Christen 2008) to generate scores for each potential link, depending on the similarity of the five linking variables in the two records. The MPC dropped any cases in the sample data with more than one potentially “true” link in the 1880 census data, so the final data consist only of links with one unambiguous “true” link. This decision severely reduces the size of the linked data set but also reduces the probability of making an erroneous “false positive” linkage. It also creates some bias in the linked data toward individuals with less-common names born in less-populated areas. Analysis of the association between name commonness and occupational scores in the 1870 data revealed little correlation (Goeken et al. 2011). However, the linkage procedure is more problematic for women than men because women with a surname change between time periods as a result of marriage or remarriage are effectively eliminated from analysis.
Table 1 shows the sample size for each linkage by gender, along with the percentage of the population classified as mulatto in each census year. About 1 in 5 African-ancestry individuals were classified as mulatto, and the percentages are similar across linkages, although they are slightly higher for women than men in almost every sample. The smaller number of cases for women overall reflects the missing links from women’s surname changes. For the later linkages, 1880–1910 and 1880–1920, we see a sharp decline in the number of linked cases for both men and women. Such declines are to be expected as a result of mortality. Sample size declines are even more dramatic for women because almost all women who were never married in the 1880 data are unobserved in the later linkages as a result of surname changes.
Because of the smaller size of the 1880–1910 and 1880–1920 samples, we combine these two linkages to create an 1880–1910/1920 sample. This pooling increases our statistical power to detect changes from the earlier 1870–1880 period to the later 1880–1910/1920 period, at the expense of being able to examine changes between 1910 and 1920. However, because of the small size of each of the later samples, our statistical power to detect changes of either sort is low.
To better assess the potential for bias in the linkage procedure, we compare the means of variables used in our analysis for the 1880 panel of each linkage sample against the full-count 1880 census data. More details of the comparison are available in Online Resource 1. Discrepancies between our samples and the 1880 census data are minimal for men, but some notable biases exist for women (see Table S1). Because of concerns regarding selection, as well as conceptual differences in measuring the social status of women in a breadwinner family system, we model only the relationship between social status and racial classification for men.3 However, because the data are relatively new and the frequencies for women might be useful as points of comparison and hypothesis generation for future research, we provide descriptive statistics by gender in the following analyses.
Measuring Mulattos
From 1790 until 1960, the U.S. Census Bureau conducted its decennial counts using enumerators. Initially, the task of going door-to-door collecting information fell to U.S. Assistant Marshals. Starting in 1880, enumerators were hired temporarily and generally drawn from the local community (Anderson 1988; Magnuson 1995). Thus, our analysis relies on racial classifications recorded by someone other than the individuals in question and likely determined without their direct input.
As noted earlier, enumerators were given specific definitions of the categories “black” and “mulatto.” However, they were not given explicit guidance as to how to apply these definitions to the people they came across. From 1790 to 1880, census returns were posted in public, ostensibly so people could ensure that they had been counted (Magnuson 1995). This requirement, along with the instructions for other racial categories, such as “Indian” (see Nobles 2000), suggests that the default would have been to classify someone according to his or her status in the community. Similarly, a study of court proceedings on questions of racial classification during this era found that when race could not be definitively established by ancestry or appearance, evidence was often admitted that spoke to how the person behaved in public: for example, where one worked and lived, whether one voted, and who one’s friends were (Gross 2008).
We assume, then, that being classified as “mulatto” between 1870 and 1920 was a reflection of how people were perceived by others: not the direct measure of their ancestry it was purported to be, nor necessarily how they would have identified themselves. (See Roth (2010) and Saperstein (2008) for discussion of conceptual and empirical differences between different measures of race.) Of course, these classification decisions likely varied across individuals and contexts. Ideally, we would have individual-level information about the enumerators, as well, allowing us to control for any differing perceptions. However, previous studies of racial classification fluidity find that controlling for interviewer characteristics does not alter the underlying relationship between social status and racial perceptions (Saperstein and Penner 2012). Further, local variations in classification that were at odds with one another should bias our results toward finding no effect in the aggregate.
Measuring Occupation
Our primary measure of social status derives from the individual’s occupation. During our study period, occupations were recorded in an open-ended format by census enumerators. A response of “none” was often used for individuals who were unemployed as well as those who were out of the labor force for other reasons, so we limit our analysis to men between the ages of 16 and 65 to minimize age-related explanations for not working. This restriction applies to both halves of the linkages: for example, individuals must be between ages 16 and 55 in the 1870 data (thus, ages 26 to 65 in 1880) to be included in the 1870–1880 models. In sensitivity analyses (not shown), we reduced the lower age bounds to 10 years of age and found substantively identical results.
The MPC recoded the open-ended occupation descriptions into 1950 occupational codes. This translation of the write-in responses is unproblematic for most cases, particularly when 1950 occupational categories are aggregated into the larger umbrella categories of professional, managerial, clerical, sales, craft, operative, farmer, service, and labor (Ronnander 1999). Figure 1 shows the distribution of these large occupational categories for black and mulatto men separately by region and year for our linked data sets. The most notable feature of this figure is the concentration of individuals in lower-tier occupations, particularly unskilled labor, and the rarity of individuals in nonmanual occupations. In all our samples, and in line with previous research, mulattos were more likely to hold prestigious jobs.
We use two different specifications to model occupational status. First, we use a categorical scheme based on the criteria of generating sufficient, theoretically meaningful variability without creating data sparseness that would hinder our analysis. The primary distinction we employ is between skilled and unskilled labor, with additional categories for farmers and the non-employed. Unskilled labor is used as the reference category and includes both agricultural and nonagricultural unskilled labor, as well as service work. Skilled labor includes all other occupational categories except farmers.4
The farmer category presents a special challenge for the South because of the rise of tenant farming and sharecropping during this time period. These data do not allow us to distinguish tenant farmers and sharecroppers from individuals who truly owned their farms. Thus, we cannot capture an important status distinction between farm ownership and tenancy. The farmer category increased substantially between 1870–1880 and 1910–1920 (see Fig. 1), and this growth almost certainly reflects the development of the sharecropping system in the intervening period. As a result, we do not assume movement into and out of the farmer category represents either upward or downward mobility, but rather, we focus our analysis on mobility into and out of unskilled labor.
We also measure occupational status using scalar occupational income scores. Occupational income scores are based on the median income of individuals who held those occupations in 1950 and are scaled between 0 and 100. Although using a score developed from 1950 data on data from 1870–1920 is imperfect, the overall ranking of occupations using such scales has been shown to be highly consistent across time and place (Hout and DiPrete 2006). An advantage of the scalar approach is that it allows us to avoid some of the problems of data sparseness associated with the use of a categorical variable.
Methods
To estimate fluidity between the mulatto and black categories across linkages, we calculate the odds ratio of consistent classification from the cross product of the 2 × 2 table formed by racial classification for the two time periods. The higher the odds ratio, the more likely that individuals will be consistently classified across the two time points. Conversely, lower odds ratios indicate greater fluidity in racial classification.
An advantage of this approach is that the odds ratio formed by the table is independent of the marginal distribution of race in either time period, and thus we are able to isolate the fluidity of racial classification from overall changes in the percentage of the population that is classified as black and mulatto. We estimate separate odds ratios by gender (as discussed earlier) and region because of potentially important differences in the size, origin, and development of the mulatto populations in different parts of the country (Reuter 1918; Sowell 1978).
Errors in the linkage procedure can generate some bias in these odds ratios. For erroneous linkages, racial classification in the later year will be independent of racial classification in the early year, leading to overreporting of reclassification between mulatto and black. Thus, the odds ratios of consistent classification we present should be considered lower-bound estimates of the stability of the black-mulatto boundary during this era.
Multivariate Models
The ability to control for unobserved time invariant variables is a powerful and necessary feature of our analysis. However, fixed-effects models also have a high cost in terms of efficiency, particularly for binary outcomes. Only observations that vary in racial classification across time periods are included in our analysis, and only observations that vary across time periods for a given independent variable contribute to the estimation of β for that variable. This limitation reduces the size of our effective sample, particularly for the non-South, where we have 23 and 14 cases of racial reclassification for the 1870–1880 and 1880–1910/1920 linkage data, respectively. We estimated models separately by region, but we limit our discussion to the results for Southern men because of a lack of statistical power for the non-South.5 Individuals who migrated between regions between censuses are necessarily excluded from analysis.6
Our baseline fixed-effects model includes only occupational status. We also estimate a second model that includes controls for urban versus rural residence, marital status, and a county-level contextual variable measuring the relative size of the African-ancestry population. Urbanicity is an important control because the urban labor market provided more opportunity for upward occupational mobility. Histories of the mulatto population also suggest a positive relationship between urban residence and mulatto classification (e.g., Reuter 1918). We include marital status, coded as ever married versus never married, to capture an alternative form of social status. We estimate supplementary models, restricted to men who were unmarried in the initial time period, that include their new wife’s racial classification as well. These models allow us to test whether the racial classification of a man’s spouse had an additional effect on his own classification, net of his occupational standing.
The county-level contextual variable incorporates insights from Gullickson (2010), who found that county-level occupational differentiation between blacks and mulattos corresponded to county-level characteristics, particularly the relative size of the African-ancestry population and the occupational status of whites. We cannot estimate the overall occupational status of whites directly for all our linked samples, but we can use published census data to calculate the percentage of the total population that is recorded as having African ancestry in each year of our linked data. This variable should have a positive effect on mulatto classification, if growth in the relative size of the African-ancestry population made whites more likely to acknowledge or promote an intermediate buffer class.
Erroneous links have the potential to generate spurious effects on the coefficients in the fixed-effects regression models. To determine the potential strength of such bias, we used the full-count 1880 data to perform a Monte Carlo simulation, detailed in Online Resource 1. Even when we generated a fairly high proportion of erroneous linkages (20 %), the positive bias was small relative to the size of the occupation status effect that we actually find in our data. Given that the MPC linkage procedure intentionally minimized the potential for erroneous linkages, we are confident that our results are not driven by this source of bias.
Results
We begin with cross tabulations showing the correspondence between an individual’s racial classifications across the pairs of census years for the 1870–1880 and 1880–1910/1920 linkage data, respectively. We examine this correspondence separately by region and gender. For each cross tabulation, we calculate the corresponding odds ratio and its 95 % confidence interval. We then provide an illustration of the relationship between changes in occupational status and changes in racial classification for Southern men, before moving to the results of our fixed-effects models.
Racial Classification Consistency
In all years, and for both men and women, there is little aggregate change in the population proportions for blacks and mulattos. For example, 17 % of men were classified as mulatto in 1870, and 16 % of men were classified as mulatto in 1880 (see Table 1). However, this aggregate stability hides significant individual switching between categories. As Table 2 shows, of the 370 men who were classified as mulatto in 1870, less than one-half (177) were classified as mulatto in 1880. Of the 1,811 men who were classified as black in 1870, 9 % (171) were classified as mulatto in 1880.
Because the overall percentages were very similar between the early and late census years within each linkage, there was little opportunity for racial upgrading as a result of relative growth in the size of the mulatto category—a phenomenon often referred to as structural mobility in studies of occupational mobility (Featherman and Hauser 1978). In other words, the absolute number of people who were downwardly mobile from mulatto to black roughly counterbalanced the absolute number of people who were upwardly mobile from black to mulatto. This lack of structural racial mobility is consistent across the South and the non-South as well as for both men and women (see Tables 2 and 3).
Yet, the circular patterns of change between censuses are not random, as might be suggested by the Census Bureau’s insistence on removing the “mulatto” category because of poor data quality (Nobles 2000:69). The odds ratio for men in the 1870–1880 period is highly statistically significant (p < .001). The odds of being classified as mulatto in 1880 were 8.8 times higher for those who were classified as mulatto in 1870 than they were for those who were classified as black in 1870. Higher odds ratios in the South compared with the non-South also suggest that racial classification was “stickier” in the South, although the differences do not reach statistical significance. This is consistent with the longer history of status distinctions between blacks and mulattos in the South that likely made that boundary more salient. We also calculated odds ratios separately for the Lower and Upper South but did not find any statistically or substantively significant differences in levels of fluidity (results available upon request).
Racial Fluidity and Occupational Mobility
Although the patterns of consistent classification are interesting in their own right, the test of whether a “mulatto escape hatch” existed in the United States lies in whether reclassification was related to SES. Figure 2 shows the average change in occupational income scores for Southern men, by racial classification and time period. On average, occupational income scores increased as men aged. However, the differences in the increase by racial classification are telling. First, focusing on the results for men classified as black in 1870, in the upper-left corner of the figure, we see that the average increase in occupational income score was much higher among those who were reclassified as mulatto in 1880 than among those who were consistently classified as black. Conversely, as shown in the upper-right corner of the figure, individuals who were classified as mulatto in 1870 and reclassified as black in 1880 had substantially less change in their occupational income score than individuals who were consistently classified as mulatto. The average changes for the 1880–1910/20 linkages, shown in the lower panels, are in the same directions but with more noise. Overall, the results indicate that occupational mobility can contribute to racial reclassification and that the effect of occupational mobility works in both directions.
Of course, these results are only suggestive because this approach does not account for potentially confounding individual-level characteristics. For example, an individual with light skin classified as black in 1870 might be more likely both to be classified as mulatto in 1880 and to have upward occupational mobility than a similar person of darker skin. Using the fixed-effects models outlined earlier, we can account for these individual confounders by looking at within-person change.
Fixed-Effects Models
Table 4 shows changes in the log odds of being classified as a mulatto as a result of shifts in occupational status for Southern men, across a series of fixed-effects models. Panels A and B show estimates for the 1870–1880 and 1880–1910/1920 data, respectively. The first model in each panel provides the bivariate association between occupational categories and mulatto classification. The second model adds controls for urban residence, marital status, and the relative size of the African-ancestry population. Models 3 and 4 follow the same format but use occupational income scores instead of occupational categories.
In the 1870–1880 data, changes in occupational status have substantively strong and statistically significant effects on racial classification. Model 1 shows that being out of the labor force is associated with decreased odds of mulatto classification and that skilled labor is associated with increased odds of mulatto classification (χ2 = 13.68, df = 3, p = .0084). Model 3 similarly shows that a single point increase in the occupational income score is associated with a 7 % increase in the odds of being classified as mulatto. Thus, positive occupational shifts are associated with upward racial mobility from black to mulatto, while negative occupational shifts are associated with downward racial mobility from mulatto to black. These effects only increase when we control for other factors in Models 2 and 4. In the categorical occupation models, we find little evidence that shifts into or out of the farmer group are associated with racial classification in the early Jim Crow South.
For the 1870–1880 data, we also tested models using occupational income scores separately for the Lower South and the Upper South (not shown). These models indicated a stronger relationship between occupational status and mulatto classification in the Lower South (b = 0.116, p = .045) than the Upper South (b = 0.042, p = .301), but the difference in the coefficients was not itself statistically significant (p = .285). These results suggest that the relationship between occupational mobility and racial mobility might have been stronger in areas with a longer history of recognizing mulattos as a distinct class, but the finding remains largely speculative.
Panel B of Table 4 shows estimates for the 1880–1910/1920 linkage data. The results for these models are inconclusive. The effects for skilled labor and occupational income score are in the expected direction but are of somewhat lower magnitude than the 1870–1880 data, and none of the results are statistically distinguishable from zero. Interestingly, in the categorical model, the positive effect of being a farmer on the odds of mulatto classification increased and is marginally statistically significant (p = .104). We can only speculate about what is driving this result. As shown in Fig. 1, the number of farmers increased in the later sample, but the number of unskilled laborers—including farm labor—declined. This shift was partly driven by the rise of sharecropping during this time period, but it also may reflect changes in how occupations were recorded by enumerators.
Models 2 and 4 of each panel also include the relative size of the African-ancestry population as a predictor of mulatto classification. In the 1870–1880 data, the relative size of the African-ancestry population had a positive effect on mulatto classification. An increase in the relative size of the African-ancestry population in a county by a single percentage point increased the odds of mulatto classification by about 3 %. These results are consistent with a “buffer class” argument in which the perceived threat of a large African-ancestry population made whites more willing to grant an intermediate status to certain members of that population. This effect is nonexistent in the later sample, suggesting that the utility of treating mulattos as a buffer class declined, coincident with the spread of the one-drop rule.
Urban residence also produced strong but contradictory effects across the two time periods. In the 1870–1880 linkage data, migration to the city is associated with increased odds of being identified as mulatto. However, in the 1880–1910/1920 data, moving to the city is associated with decreased odds of being identified as mulatto. The reasons for this shift are unclear but suggest a changing role for Southern cities in supporting black-mulatto distinctions, perhaps tied to the growth of a new Southern industrial system outside the old centers of Creole life.
The results in Table 4 indicate that the act of marrying, in and of itself, had little effect on men’s odds of being classified as mulatto. However, additional analysis suggests that it mattered whom one married. As Table 5 shows, the odds of mulatto classification for Southern men with African ancestry changed depending on the racial classification of their wives. To isolate this relationship, we examine the 1870–1880 linkage data for men who were unmarried in 1870.7 The results of Model 2 indicate that having a new wife who was classified as black (in 1880) significantly reduced the husband’s odds of being classified as mulatto in 1880. The results for having a new mulatto wife in 1880 are smaller and not statistically significant, but suggest a parallel process in which marrying a woman who was classified as mulatto increased the husband’s odds of being classified as mulatto. The effect of occupational income score on mulatto classification also declines between Model 1 and Model 2. This suggests that occupational mobility and marriage prospects were linked, as one might expect. One possible mechanism is that a better occupation enabled men to “marry lighter” and thus increase their odds of being classified as mulatto to match their wives. Another possibility is that racial status—like social status more generally—tended to accrue to the household as a whole. In any case, the results in Table 5 indicate that in the South, a man’s racial classification depended both on where he worked and with whom he shared his house during this period.
Discussion
Our results demonstrate not only that the racial boundaries between Americans of African ancestry were in flux in the decades after the Civil War, but also that higher social status could be leveraged into higher racial status—at least for black men in the early Jim Crow South. Using unique large-scale, individual-level data, this study provides historical support for contemporary claims that racial classifications are not static, and that fluidity is related to racial inequality in the United States (see Saperstein and Penner 2012).
The results also support the contention that money did “whiten” in the United States, although the phrase is more commonly used to describe racial classification and inequality in Brazil. The fact that we find evidence of intragenerational racial change following an increase in occupational status for African ancestry men in the nineteenth century United States further complicates long-standing claims regarding the rigidness of race in the United States, as well as debates over whether, how, and when racial hierarchies and classification schemes differ between the two countries (see Telles (2004) for a review). In the sections that follow, we offer both micro- and macro-level implications drawn from our results that can enrich these discussions and comparisons.
“Passing” and Slipping
Perhaps the most surprising aspect of our results is the complementary nature of racial fluidity in the United States. When changes in classification occurred, Americans with African ancestry not only moved “up” in the racial hierarchy, but they also moved “down.” Thus, in addition to an escape hatch allowing high status blacks to become mulatto, our results indicate there was a corresponding trapdoor into which mulattos who lost status could slip and become black. Recognizing the existence of both upward and downward racial mobility has important implications for understanding racial fluidity at the individual level and its relationship to inequality. Yet, accounts of “passing” in the United States (and “whitening” in Brazil) have focused attention on the potential for individual upward racial mobility in otherwise persistent systems of racial inequality, leaving the possibility of analogous downward mobility unexplored (but see Loveman and Muniz 2007).
Given the dearth of information on the phenomenon, it is notable that in our data, downward racial mobility—that is, from mulatto to black—is as common, in absolute terms, as the more typically discussed upward variety (see Table 2). Further, the associations in our multivariate models indicate that a decrease in occupational status decreases the odds of mulatto classification, all else being equal, just as an increase in occupational status increases those same odds (see also Fig. 2).
So it is not simply that one could leverage higher social status for higher racial status in the post-Reconstruction era, but that losing social status came with the risk of losing racial status, as well. This “slipping,” like passing, serves to preserve the aggregate racial/status hierarchy even while individuals experience mobility across category boundaries.
Structural Versus Circulation Mobility
The possibility of “slipping” also introduces a new wrinkle in comparisons between the racial hierarchies in the United States and Brazil. Typical accounts of Brazil have emphasized the recognition of a more variegated spectrum of color, the celebration of racial mixing, the intertwining of race and class distinctions, and the potential for phenotypically distinct full siblings to be racially classified differently (Harris 1964; Telles and Sue 2009). This is generally contrasted to the more rigid and dichotomous white-black racial schema of the United States, with its emphasis on ancestry over appearance (Davis 2001).8 Our results suggest that although the rigidity of the U.S. racial hierarchy has likely been overstated, an important distinction remains regarding the consequences of historical racial fluidity in the two countries.
Brazil, along with several other Central and South American nations, explicitly promoted “whitening” as part of its early twentieth-century nation-building, which led to an increasing percentage of the population being classified into lighter racial categories over time (Loveman 2009; Skidmore 1974). To borrow the language of studies of occupational mobility again, up until recently, Brazil cultivated strong structural racial mobility that lightened the overall distribution of race.9 This structural mobility can be distinguished from circulation racial mobility, which reflects the level of fluidity in racial classification for individuals (irrespective of aggregate shifts in the distribution of race). Our data suggest that the key distinction between the United States and Brazil is not a lack of racial mobility in the United States per se, but the lack of similar structural racial change.
These two different racial mobility regimes are related to the differing development of mixed-race categories in the two countries, as well. Any official recognition of a separate racial status for “mulattos” was eliminated in the United States after 1920. However, even during its heyday, the level of fluidity we observe combined with the lack of state-sanctioned structural racial mobility likely contributed to instability of the mixed-race category because individuals were frequently shuffled between categories and thus had less attachment to an intermediate position. In Brazil, on the other hand, the combination of high levels of individual fluidity and upward structural racial mobility may have provided some stability to the mixed-race category, making it more of a destination than a way station. We lack comparable longitudinal data from Brazil to explicitly test this hypothesis, but we suggest that increased attention to the interplay among racial classification, circulation racial mobility, and structural racial mobility would be a fruitful new direction for cross-national research on race in the Americas.
Conclusion
U.S. Census Bureau officials have long spoken with scorn about the late nineteenth and early twentieth century data on racial mixture in the United States. A recent article on the history of census racial distinctions reiterated the general belief in the lack of validity and reliability in the categorization of mulattos, quadroons, and octoroons: “. . . the attempt to classify individuals into these multiracial groups failed all of our statistical criteria. The distinctions were not recognized by either society or the individual, nor did they predict opportunity” (Humes and Hogan 2009:114).
We submit that, at least for a time, the categorical distinction between mulatto and black was meaningful, both socially and statistically, although not for marking the purely ancestral differences the census definitions intended. Early in the Jim Crow South, the two categories also marked a status distinction, as captured by our data on shifts in occupational status between censuses. During this period, opportunity predicted racial classification, and vice versa.
Yet, racial boundaries that draw their significance from demarcating status categories and not simply color or ancestry categories also appear in the contemporary United States (Saperstein and Penner 2012), in the wake of the Civil Rights Movement, and along with increased immigrant diversity and renewed attention to multiraciality. The timing of evidence for U.S. racial fluidity, both historical and contemporary, raises the possibility that clear links between social and racial mobility emerge only when the overall racial order is in flux. Another possibility is that the relationship emerges when racialized inequality is at its most stark, and widespread stereotypes link social status with racial membership, lending more incentive to reclassification and individual boundary crossing (cf. Wimmer 2008). Alternatively, perhaps racial circulation mobility has been an unrecognized feature of U.S. race relations all along, acting as a kind of individual-level safety valve on racial tensions, not unlike Degler’s (1971) description of the purpose of the “mulatto escape hatch” in Brazil.
With no longitudinal data currently available between the 1930s and the 1970s to test these theories, they remain entirely speculative. However, if and when such data become publically available for the past or are collected in the future, we hope that researchers interested the dynamics of race and inequality will heed the more general implication of our results: although measures of racial classification might be unreliable from a technical standpoint, it is precisely their fluidity, or lack of consistency, that reveals substantively interesting insights about the relationship between race and inequality in the United States.
Acknowledgments
A previous version of this paper was presented at the 2011 annual meeting of the Population Association of America, Washington, DC. We are grateful to Ann Morning and Roy Mill for their helpful comments and suggestions, and to Krystale Littlejohn for research assistance.
Notes
Jim Crow laws were the various legal mechanisms used to support white supremacy in the South, most notably the “separate but equal” statutes that enforced racial segregation. The name “Jim Crow” itself comes from a popular black caricature performed in minstrel shows during this time period. Under the one-drop rule anyone with known African ancestry would be considered exclusively black, regardless of physical appearance or any other known ancestry.
We use the term “whitening” throughout this article, as Degler (1971) did, to signal the direction of change in racial perceptions, not necessarily the ultimate categorization. According to this definition, classification changes from black to mulatto, mulatto to white, or black to white would all count as “whitening,” although we can capture only the first type of racial fluidity in our data.
We tested models using a woman’s own occupational status or her husband’s occupational status as the independent variable, and neither produced statistically significant results.
We also experimented with coding service workers, craftsman, and operatives as separate categories. The resulting models fit poorly and did not change our conclusions. Most importantly, the categories of craftsman and operative, which comprise the vast majority of skilled laborers, had similar effects on racial reclassification. These findings support our decision to use a single category of skilled labor.
Results for the non-South sample are available upon request. All models fit poorly with no statistically significant coefficients and with inconclusive substantive results in terms of the direction of the estimated coefficients.
This restriction results in a loss of 2 % to 5 % of Southern men for each of our time periods because of their outmigration between censuses. We do not drop more than 30 total cases because of migration in either time period. As a sensitivity analysis, we also estimated models in which anyone who moved between counties between time periods was excluded, and the results are similar, although the statistical power is reduced because of the smaller sample size.
We exclude already married individuals from this analysis to reduce endogenous explanations for any observed changes. Shifts in racial classification among previously married couples could be prompted by changes in either spouse’s characteristics, or by unobserved household characteristics that triggered the reclassification of both spouses.
Recently, scholars have suggested that the two countries may be moving closer to one another, at least along the dimension of continuous versus categorical distinction, with Brazil adopting a dichotomous categorization of whites and negros for its affirmative action policies and the United States beginning to recognize multiraciality and becoming increasingly diverse through immigration (Bonilla-Silva 2004; Skidmore 1993).
This overall lightening of the Brazilian population—above and beyond the effects of birth, death, and immigration—can be seen in the changing census racial distribution between 1950 and 1980 (de Carvalho et al. 2004). However, there is evidence that the “whitening” trend has been reversing, especially among highly educated Afro-Brazilians (Marteleto 2012; Schwartzman 2007).
