| Author |
Message |
    
Tonya
"Cyniquian" Level Poster
Username: Tonya
Post Number: 2880
Registered: 07-2005
Rating: N/A
Votes: 0 (Vote!) | | Posted on Tuesday, July 04, 2006 - 02:53 pm: | 
|
NBER WORKING PAPER SERIES
COLORISM AND AFRICAN-AMERICAN WEALTH:
EVIDENCE FROM THE NINETEENTH-CENTURY SOUTH
Howard Bodenhorn
Christopher S. Ruebeck
Working Paper 11732
http://www.nber.org/papers/w11732
NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
November 2005
Corresponding author: Department of Economics, Lafayette College, Easton, PA 18042-1776
bodenhoh@lafayette.edu. Bodenhorn thanks the National Science Foundation (SES-0109165) and the Earhart
Foundation for financial support. We thank Sandy Darity, Patrick Mason, Darrick Hamilton, Deborah
Cobb-Clark, two anonymous referees, and seminar participants at Binghamton University, the Brookings
Institution, George Mason University, Florida State University, Lafayette College, Lehigh University, and
the NBER for many useful comments on earlier versions.The views expressed herein are those of the
author(s) and do not necessarily reflect the views of the National Bureau of Economic Research.
©2005 by Howard Bodenhorn and Christopher S. Ruebeck. All rights reserved. Short sections of text, not
to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including ©
notice, is given to the source.
Colorism and African American Wealth: Evidence from the Nineteenth-Century South
Howard Bodenhorn and Christopher S. Ruebeck
NBER Working Paper No. 11732
November 2005
JEL No. N3, J7
ABSTRACT
Black is not always black. Subtle distinctions in skin tone translate into significant differences in
outcomes. Data on more than 15,000 households interviewed during the 1860 federal census exhibit
sharp differences in wealth holdings between white, mulatto, and black households in the urban
South. We document these differences, investigate the relationships between wealth and the recorded
household characteristics, and decompose the wealth gaps into treatment and characteristic effects.
In addition to higher wealth holdings of white households as compared to free African-Americans
in general, there are distinct differences between both the characteristics of and wealth of free
mulatto and black households, whether male- or female-headed. While black-headed households'
mean predicted log wealth was only 20% of white-headed households', mulatto-headed households'
was nearly 50% that of whites'. The difference between light- and dark-complexion is highly
significant in semi-log wealth regressions. In the decomposition of this wealth differential, treatment
effects play a large role in explaining the wealth gap between all subpopulation pairs.
Howard Bodenhorn
Department of Economics
Lafayette College
Easton, PA 18042
and NBER
bodenhoh@lafayette.edu
Christopher S. Ruebeck
Department of Economics
Lafayette College
Easton, PA 18042
ruebeckc@lafayette.edu
-2-
Colorism and African-American Wealth:
Evidence from the Nineteenth-Century South
1. Introduction
The first documented case of a white man being punished for engaging in miscegenation
occurred in Virginia in 1630, just 10 years after the arrival of the first Africans into the colony
(Mumford 1999). Interracial sex is not new and neither are the questions of the relationship
between identity and race. Throughout the nineteenth and twentieth centuries, both whites and
blacks were uncomfortable with miscegenation and mixed-race people (Bynum 1998). Although
the number of interracial unions in the United States is currently increasing, the desire of mixedrace
individuals to seek a racial identity separate from their constituent racial ancestries is not
new. For many of African descent, black is not black—both in terms of how they view
themselves and how others view them. There are meaningful subtleties of shade, differences that
have socially and politically meaningful distinctions today—just as they have in the past.1
Despite the complex history of miscegenation, researchers have been complacent about
the content of federally defined racial categories (Snipp 2003). With a few recent exceptions
(Bodenhorn and Ruebeck 2003; Darity, Mason, and Stewart Forthcoming), economists have used
the blunt instrument of binary racial classifications to categorize individuals as either black or
white. Researchers consider all individuals within a category to be subject to similar treatment.
Relatively little attention, if any, is paid to the concordance between social constructions of race
and official categorizations. What may appear to be self-evident—whether an individual is black
1 Modern studies of the social consequences of skin shade differences include Russell, Wilson and Hall (1992),
Root (1996), Daniel (2002), Tizard and Phoenix (2002), Herman (2002), and Doyle (undated).
-3-
or white—actually masks a complex reality and likely does not capture the subtleties of racial
interactions. If different skin tones are treated unequally, the continued practice of scholarly
racial categorization based on anachronistic notions developed in the Jim Crow South may not
lead to useful or relevant insights into modern phenomena. As Trina Jones (2000, p. 1499) notes,
to understand “the nuanced ways in which discrimination operates and differentially impacts
similarly but not identically situated people, we must examine and attempt to understand color.
Such an understanding begins with history.”
Embracing Jones’ call for historical understanding, our paper turns to history to
document the economic consequences of colorism, thereby providing fresh insights into an
increasingly modern concern. Using data on more than 15,000 households interviewed during the
1860 federal census, we find sharp differences in wealth holdings between white, mulatto and
black households in the urban South.2
This study advances the literature in two ways. First, we consider the operation and
effects of colorism closer to its historical origins. Historians attribute the origins of lightcomplexion
preferences to slaveholders. By investigating data on complexion-based differences
in attainment of free African-Americans in the pre-Civil War South, our study provides a more
objective basis for such beliefs and provides a baseline against which to consider modern
outcomes. Second, recent economic and sociological research has stressed the importance of
household wealth, rather than income, for the determination of socioeconomic status (Conley
1999; Shapiro 2004). Accumulations of household wealth have been shown to insulate middle-
2 Teo (2003) notes how the use of terms such as mulatto and colored, as well as the conception of mixed-blood, are
terms heavily laden with racial, often racist, connotations. This is not the place to develop a new terminology. We
will use the term mulatto in the contemporary context to describe mixed-race and, generally, light-complected
African Americans. We recognize that some readers dislike the word, and ask for their indulgence. Some have
suggested the terms biracial or multiracial be used instead, but we resist because the modern use of these terms is
politically charged and does not capture the meaning of mulatto as it was used in the mid-nineteenth century. We
will also use the terms light-complected and dark-complected to recognize the likely factor that determined the
“mulatto” and “black” identifications and to refer to modern studies of colorism.
-4-
and working-class families from falling into poverty when they face even short-term labor
market disruption or other temporary crises (Oliver and Shapiro 1995). Previous studies of the
economic consequences of modern colorism have focused on income (Goldsmith, Hamilton, and
Darity (2004); Darity, Dietrich and Hamilton (Forthcoming)), which incompletely captures the
consequences of differential treatment according to skin shade. With its twin focus on wealth
holdings and complexion, this paper demonstrates that the economic consequences of colorism
were (and likely are) far reaching.
Section 2 summarizes the history and modern experience with colorism in the African
American community, including just a handful of recent economic studies. We then proceed
through an empirical analysis in three steps. Section 3 describes the data and takes a first look at
the three sub-populations’ differences evident in the 1860 census manuscripts. In Section 4 we
present several regressions in which we show that bivariate racial classification does not capture
the subtleties of race even after controlling for a wide variety of individual and household
influences. By separating African-Americans into blacks and mulattoes, we find meaningful
differences in outcomes between households with a black or a mulatto head. Section 5 presents
decompositions of the complexion wealth gap to identify how the wealth gaps are influenced by
differences in wealth-generating characteristics between the groups and by differences in
treatment between the groups not explained by the observed characteristics other than skin color.
The decompositions show that this treatment effect explains a substantial fraction of the
observed complexion wealth gaps among all sub-populations. Section 6 summarizes and offers
some concluding remarks.
2. Colorism and socioeconomic outcomes
-5-
History provides compelling evidence of a consistent pattern of preferences shown, by both
whites and blacks, toward light-complected African Americans. The economic advantages of a
light complexion date back to slavery. Reuter (1918) and Frazier (1957), among others, argue
that a complexion advantage appeared early in the slavery era when slaveholders
disproportionately selected light-complected blacks to work as house servants and field foremen.
Indeed, some slaveholders contended that “no man would buy a mulatto for field work” (Johnson
1996, p. 111), while others reported that light-complected slaves resisted field work by arguing
that their physical constitutions could not stand up to the demands of hard labor (Kemble 1863,
p. 193). Although few slaveholders balked at setting light-complected slaves to field work, they
preferentially selected light-complected slaves for craft training and apprenticeship.
According to Frazier (1957) a fair complexion improved a slave’s life chances by
significantly reducing his or her toil and drudgery, by improving his or her access to food and
shelter, and by exposure to the culture, as well as the manners, dress, and linguistic conventions
of whites. Light-complected, mixed-race slaves were also more likely to be educated and
manumitted. Moreover, despite legal prohibitions in many states against slaveholders providing
for slaves in their wills, masters did occasionally provide for their mixed-race offspring at their
passing, sometimes at the expense of white offspring (Horowitz 1973).
In the antebellum South, free light-complected blacks were more likely than darkcomplected
blacks to be literate. Horton and Horton (1997, p. 154), for example, report general
white opposition to black school attendance except for those of “very light complexion.” With
greater access to education, mulattoes labored at more remunerative occupations than blacks.
Neither mulattoes nor blacks labored in high prestige occupations but, among African
Americans, mulattoes dominated positions at the top of the employment ladder—merchants,
-6-
shopkeepers, and skilled craftsmen. Those with fair complexions were also overrepresented
among skilled workers and professionals. Among rural free African Americans, Bodenhorn
(2003) finds that mulattoes were more likely than blacks to work as tenants or to own their own
farms.
Gatewood (2000) shows that social preferences for light-skinned blacks persisted through
the post-Civil War period, but the increasing severity of Jim Crow and the movement toward the
one-drop rule3 worked to politically and socially align light and dark blacks, groups that were
often at odds prior to the Civil War. Nevertheless, color-based differences persisted and by 1920,
colorism and complexion-based stratification in the African American community gained the
attention of sociologists, notably E. B. Reuter (1918, 1928). A number of studies conducted
through the 1930s and 1940s, including Myrdal’s (1944) magnum opus, found support for
Reuter’s contention that light skin tones and perceptible traces of non-African heritage were
associated with material advantages for African Americans (Hill 2000). A rhetoric of
complexion egalitarianism, however, seemingly reigned through the Civil Rights era and into the
1970s, which led to fewer studies of colorism during the 1960s and 1970s (Freeman et al. 1966 is
a notable exception). Since the early 1990s, however, sociologists have become interested in
complexion-based differences in non-white communities. Modern evidence documents the
continuing, perhaps growing, significance of colorism.
3 In 1787 Virginia defined anyone with one or more black grandparents as mulatto, a definition that was reiterated in
1792 and 1819. In 1833 the Virginia legislature conceded that the one-fourth definition created a class of people that
were neither black nor mulatto nor white. An act defined such people with less than one-fourth black ancestry to
petition the local court for certificates of freedom that would establish their legal “whiteness.” It was not until 1866
that the term mulatto was eliminated from the civil code and replaced with the word colored. In 1910 the statutory
limit was changed so that anyone with one-sixteenth black ancestry was considered colored. Finally, in 1930 the law
was changed so that anyone with any ascertainable amount of black ancestry was colored (Wadlington 1966, pp.
1196, 1201). Mississippi’s state constitution in 1911 classified individuals with one-fourth black ancestry as legally
colored (Anonymous 1911). By 1948, the constitution had been amended so that individuals with one-eighth black
ancestry were colored (Bynum 1998). Legal definitions in other southern states followed similar trajectories.
-7-
Using data from the National Survey of Black Americans conducted in 1980, Hughes and
Hertel (1990), Keith and Herring (1991), and Seltzer and Smith (1991) find that dark-skinned
blacks acquire less education, work in less prestigious occupations, and earn lower incomes than
lighter complected blacks, even after controlling for other factors believed to be correlated with
achievement. Using longitudinal analysis, Hill (2000) finds that biracial men enjoyed modestly
advantaged backgrounds compared to blacks, but multivariate analysis reveals that family
background variables account for only a small portion of complexion-based differences in
outcomes. Color is the decisive factor. Light-complected men are more than twice as likely as
dark men to find high-prestige employment. Research has also documented skin-tone differences
among Latinos, which suggests that colorism characterizes a host of minority communities
(Murguia and Telles 1996; Telles and Murguia 1990; Mason 2004; Darity and Boza 2003).
Gullickson’s (2003) interpretation that the significance of skin tone differentials are declining
stands in sharp contrast to an emergent literature that finds complexion differences still loom
large in the life experiences of mixed-race youth (Russell, Wilson and Hall (1992), Root (1996),
Daniel (2002), Tizard and Phoenix (2002), Herman (2002), and Doyle (undated)).
Despite the attention that colorism has garnered among sociologists, economists have not
accounted for intraracial differences in economic outcomes that may result from colorism. One
recent exception that uses nationally representative samples of African Americans from 1979 and
1992 finds that the wages of light-complected blacks exceed those of dark-complected blacks by
about eight to ten percent (Goldsmith, Hamilton, and Darity 2004). By comparison, their
regression decompositions suggest color-based wage differentials (treatment effects) between
whites and light-complected blacks of about five percent. Thus, colorism accounts for about
twice the income gap between light- and dark-complected blacks as between whites and light-
8-
complected blacks. Thus, subtle distinctions in skin tone seemingly translate into significant
outcome differences, including educational attainment, occupational achievement and earnings,
among others. We now turn to a description of the data source that allows us to contribute to this
emergent and important discussion.
3. Data
This study uses a large, regionally representative sample of free African Americans residing in
the urban South, constructed from information in the original population manuscripts of the 1860
census. In discussing the outcomes of free African Americans we are, by definition, excluding
the experience of slaves, who represented the majority of blacks in the antebellum South and
held little or no wealth.4 The cities that are included in the sample are listed along with their
relative frequencies in the lower part of Table 1’s summary statistics. Williamson (1980) notes
that the border states of the Old South formed a sort of mulatto belt, which contained a high
proportion of mulattoes relative to blacks. The Lower South cities of Charleston, Mobile, and
New Orleans also had substantial populations of gens de couleur libre. The size or proportions of
their respective mulatto and black populations did not drive the choice of cities for this study.
Rather, cities were selected to provide a representative sample of all southern, urban free African
Americans in 1860.
Every household headed by an African American (black or mulatto) from these cities was
recorded while collecting data, along with an equal number of randomly selected white
households. Some non-white households were later dropped because the census failed to report
information on one or more of the variables of interest or because the values reported were
4 Kenzer (1997) provides a discussion of wealth accumulation among free blacks, slaves and former slaves for
North Carolina.
-9-
clearly erroneous. The resulting sample includes information on 15,861 households. Of these,
5,774 households were headed by white males, 2,121 were headed by mulatto males, and 3,697
by black males. The sample of 4,269 female-headed households includes 842 headed by whites,
1,451 headed by mulattoes, and 1,976 headed by blacks.
The 1860 census is a particularly valuable source for the study of the economic
consequences of colorism because census marshals were instructed to classify individuals as
white, black or mulatto. Specifically, marshals were directed to leave the “Color” column blank
if the individual was white, to insert the letter “B” if the person was black without admixture,
and to insert the letter “M” if the person was a mulatto or of mixed heritage. It is unclear whether
marshals inquired into a person’s racial heritage or classified him or her based on physical
appearance, but Gross (1998) contends that local residents, including census takers, used locallyaccepted
notions of race and admixture to draw conclusions about one’s racial heritage. Most
southerners believed the differences between blacks and mulattoes were obvious and involved
complexion and physiology, as well as the complexion and physiology of those with whom one
associated.5 There is no reason to believe that a census taker’s assignations of color would have
deviated from accepted community standards of racial classification. A second concern is
whether a person classified as a mulatto in Baltimore would have been similarly classified in
New Orleans. Even if there were inter-city differences in classification standard, to the extent
5 Of tangential concern to us is the possibility (in fact, the likelihood) that the identity of one’s associates also played
a part in the census taker’s determination of color. This connection between an individual’s choices and his or her
identity is important but outside the scope of this paper. It is an active area of research (see Bodenhorn and Ruebeck
(2003) for references), but here we seek to document the effects of colorism, with less concern for questions of its
source and its links to individuals’ identity.
-10-
that they existed at all, they do not vitiate the results presented here. What ultimately determined
whether an individual received differential treatment was whether he or she met the local
physiological or heritage standards, because that individual interacted with members of the local
community. Inclusion of city dummy variables in the regressions reported below also controls, in
part, for local differences in the relationship between treatment and racial categorizations.
In addition to the valuable information on complexion, federal census takers from the
mid-nineteenth century collected and recorded information on a number of demographic and
economic variables. Household structure is easily inferred from the manuscripts because the
head of each household is listed first, identified by both given name and surname. Only first
names are recorded for spouses and children. Members of the extended family were listed next,
followed by unrelated members of the household (typically servants or apprentices), both
identified by given name and surname. The ages and genders of all household members were
recorded, along with their state or nation of birth, and whether the individuals were literate or
had attended school in the past year. The census marshals also recorded economic data, including
the occupations of household heads. In a few instances, marshals recorded the occupations of the
wives of male-headed households, but no use is made of this information because it was
inconsistently reported.
Finally, census marshals were asked to provide estimates of the dollar value of each
household’s real and personal property. The value of household real estate was to be obtained by
asking the householder for his or her valuation of the real property owned by the household
ignoring any lien or encumbrance on the property. Thus, the reported figure is gross, not net, real
estate wealth and does not include the value of rental property. Marshals were also instructed to
obtain the householder’s valuation of all other property (personal estate), including financial
-11-
assets, slaves, livestock, jewelry, fixtures and furniture. The instructions recognized that an
accurate valuation may not be possible, but marshals were encouraged to obtain as “near and
prompt” an estimate as they could. The Census Bureau anticipated the reluctance of many
householders to divulge information about their wealth and instructed marshals to cajole and
reassure respondents that the information was confidential and would not be used for private gain
or any other public (i.e., tax) purposes.
It is clear that some marshals were better at cajoling and reassuring householders than
others. It was not uncommon for marshals to return a blank (nonresponse) when reporting real
and personal estate in the manuscripts. An empty cell in the real estate column is generally taken
to represent zero wealth, reflecting that the family rented rather than owned its current habitation,
but historians have long debated the meaning of nonresponses for personal wealth, and the
debate is far from resolved. Some contend that marshals left the cell blank rather than recording
zeroes. Others contend that marshals failed to report modest or hard-to-value property holdings,
so that nonresponses likely represent small but nonzero personal wealth. Conley and Galenson
(1998) and Bodenhorn (2003) review the debate and the data and conclude that marshals had
idiosyncratic, nonzero censoring points for personal wealth below which they returned a blank.
Just as there is no consensus concerning the interpretation of unreported personal wealth
for a household, there is no consensus among economic historians concerning a preferred
empirical technique. Conley and Galenson, and Bodenhorn, estimate quantile or median
regressions; others estimate Tobit models. Given the common assertion that nonresponses are
typically indicative of small, but nonzero personal property values, we attribute to each
nonresponse one-half the minimum value recorded by any marshal in each city ward, and then
correct the standard errors for ward-level clustering. We conducted a number of robustness
-12-
checks (available by request): we estimated models that excluded households with unreported
wealth; we excluded wards if nonresponses exceeded 10, or 25, or 45 percent of sampled
households; we replaced the natural log of wealth as the dependent variable with the inverse
hyperbolic sine, among other procedures. In every instance, the basic results reported below
hold. We prefer our procedure because estimating the models using Ordinary Least Squares
allows us to report the full results of standard Oaxaca-Blinder regression decompositions. Thus,
although the wealth data drawn from the 1860 U.S. census are imperfect, they are rich enough to
generate meaningful comparisons of wealth accumulation among white, black, and mulatto
households.
In addition to age and complexion, we employ a number of variables expected to explain
differences in household wealth. We label household heads Migrants if they migrated across
state lines.6 We also label as Immigrants any head of household born outside the United States.
Migrants and immigrants are often thought to be self-selected, highly motivated individuals,
suggesting that, all else equal, movers will accumulate more wealth than nonmovers (Borjas
1994). On the other hand, movers may have financed their migration by consuming previously
accumulated wealth, which may imply lower current wealth. There is no reason to expect, a
priori, one effect to dominate. Summary statistics reported in Table 1 show that while a little
over 10 percent of all male heads of households migrated across state lines, another 25 percent
immigrated from outside the United States. Women were as likely as their male counterparts to
have migrated across state lines, but a much smaller proportion of female heads of household
were immigrants. Given the restrictions on black in-migration in most southern states, it is not
surprising that blacks and mulattoes were less likely than whites to migrate across state lines
6 Given the proximity of Baltimore and Washington, D.C., individuals who moved from Washington to Baltimore
were not considered interstate migrants.
-13-
(Guild, 1969). Despite such restrictions, about 7 percent of African-American male household
heads and nearly 9 percent of African-American female heads had migrated across state lines.
Further, 1.5 percent of African-American male heads of households had emigrated from foreign,
mostly Caribbean, countries, as had 2.7 percent of African-American female heads of
households.
Another set of variables is included to control for the household’s human capital. We
include a dummy variable equal to one if the head of the household was literate, a variable equal
to the number of literate males 20 years of age or older (including the head of the household),
and a variable equal to the number of literate females 20 years of age or older in the household.
Table 1 shows that about 82 percent of male heads of household and 74 percent of female heads
were literate. Both male- and female-headed households on average had more than one literate
female (1.109 and 1.280, respectively), but female-headed households generally had far less than
one literate adult male while the mean male-headed household had more than one. Given the
absence of publicly funded education for African Americans in the antebellum South, it is not
surprising that African American households had fewer literate adults than white households.
Indeed, the level of literacy among African Americans is notable given the outright animosity
among many whites toward the education of blacks and mulattoes (Woodson 1919).
Two additional variables are constructed to capture possible education or training effects.
The first (SEI) is a Duncan-style socioeconomic index (Reiss 1961), intended to indicate
occupational prestige. It is based on wages and educational levels associated with several
hundred occupations in the 1950s. For occupations recorded in the 1860 census with the same
title (e.g., barber, blacksmith, carpenter, bricklayer, etc.) the occupational prestige index value
was taken directly from the 1950s index. For some occupations recorded in the 1860 census,
-14-
assigning a corresponding Duncan occupational prestige score required some ingenuity.
Duncan‘s index has no entry for carriage drivers (ignoring only those few still found in Central
Park and other tourist locations), but fortunately Duncan did list a prestige score for the modern
counterpart of the carriage driver, namely the taxi driver. Similarly, modern bus drivers are
roughly the equivalent of stagecoach drivers. Because Duncan’s classification includes several
hundred occupations—from bootblacks (SEI = 8) and charwoman (SEI = 10) to lawyers (93) and
dentists (96)—and nearly every major industrial category, it is possible to assign a reasonable
occupational prestige value to every occupation listed in the 1860 census.7 When a specific
occupation listed in the 1860 census had no reasonable corresponding Duncan value, that
occupation was given Duncan’s general prestige score for job class (foreman, operative, or
laborer) and industry (textiles, metals, machinery, services, etc.). Finally, just as marshals failed
to report wealth holdings for some households, they also failed to report occupations for every
household head. These observations are also not deleted from the sample. We attribute to each
of these households the average occupational prestige value by the gender of the household head
(15 for males, 14 for females), and we include a dummy variable to indicate these observations.
The dummy variable should capture any fundamental differences between reporting and nonreporting
households, but we can see from Table 1 that there may be much less concern for this
issue in the men (6% of the male sample) than females (46% of the female sample, with
representation decreasing significantly from white to mulatto to black women).
It is likely that the labor supply decisions of women and men could be very different, as
they often can be today, and as a result the determinants of wealth accumulation may have been
quite different across gender as well. This leads us, as is the usual practice, to present separate
results for male- and female-headed households. Historians and labor economists have noted
7 A detailed appendix of occupational classifications is available from the authors on request.
-15-
differential economic and occupational opportunities for women, and have discussed the
negative consequences of single parenthood (Moehling 2004). We also report our full model
regressions both with and without the occupational variables because occupations may be
endogenous: to the extent that family wealth persists across generations, greater wealth in
previous generations may imply both greater wealth and a more prestigious occupation for the
current head of household. Because we cannot account for intergenerational transfers and
because we lack any reasonable instruments to perform two-stage least squares, we ask the
reader to interpret those regressions including the occupational variables with care.
Before moving on to the regressions, we call attention to the substantial differences in
means across same-gender subpopulations (comparing among white, mulatto, and black as well
as comparing white to African-Americans as a whole). Nearly all these differences were highly
statistically significant (see the comments below Table 1 for exceptions). Note especially the
significant differences in characteristics between blacks and mulattoes. In every category
mulatto-headed households were better endowed than their black counterparts, but mulattoes
were still not as well endowed as heads of white households.
4. The determinants of household wealth
Our empirical strategy in this section is to estimate household-level wealth regressions that
include several explanatory economic and demographic characteristics as well as the complexion
of the household head, Black and Mulatto (omitting White). The parameter vectors b and g in
the equation
(1) ln(wj ) = b
0 + b
1 X j + b
2 Yj + b
3 Z j + g
1 Black j + g2Mulattoj + e
j
-16-
are estimated using ordinary least squares regression. The dependent variable ln(wj ) is the
natural logarithm of the jth household’s wealth (following the usual semi-log regression form due
to the skewed nature of the distribution of wealth in levels), Xj is a vector of individual
characteristics of the household head (age and its square, immigrant and interstate migrant status,
literacy, and occupational prestige score), Yj is a vector of household characteristics (the number
of literate males, the number of literate females, and the size of the household), and Zj is a vector
of city dummy variables (excluding Baltimore) to capture any systematic regional differences in
the ability of households to accumulate wealth. e
j is the error term associated with the jth
household. We also estimated the equation
(2) ln(wj ) = b
0 + b
1 X j + b
2 Yj + b
3 Z j + gAfAmj + e
j ,
where both subpopulations with African-American heritage are grouped into the same category,
AfAm. We only present the value of this shift variable’s coefficient in our results because the
other parameters did not differ substantively from those estimated using (1).
Table 2 reports the regressions. Probability weights are applied to account for the
sampling method described in the Section 3.8 The columns labeled (3) and (4) are our preferred
specifications because they offer the largest number of exogenous controls. A mentioned above,
we include regression (4) to account for occupational differences but recognize that occupations
may be endogenous to wealth. Before turning to the issue of colorism, we summarize and
interpret the other individual and household-level correlates and the overall performance of the
regressions.
8 These weights take into account the undersampling of white-headed households. We took the assertion that the
census included all free households at face value and used the probability weights to adjust for the relative
probability of sampling from each household subpopulation.
-17-
Household wealth for both male- and female-headed households increased with age at a
decreasing rate. The maximum for men occurs at about 65 years, an age which little more than
five percent of our sample exceeded, although a handful of ages in the 100’s were recorded for
both genders.
Households headed by male immigrants had about half (e-0.56 = 0.57) of the wealth of
households headed by native-born men, indicating that the benefits of initiative were outweighed
by the costs of moving (at least for the current generation), although the results in column (4)
indicate that immigrants’ advantage may be explained by their occupations. The results in
columns (1)-(3) are consistent with the economic studies of immigration that show that firstgeneration
immigrants are slow to assimilate often because immigrants tend to congregate in
ethnic enclaves (Borjas 1994; Borjas 2000). We have no evidence on the ethnic or racial
compositions of the neighborhoods in which southern urban African Americans resided, but
relying on crude data at the level of census wards Curry (1981) found that blacks were
considerably less likely to be crowded into proto-ghettoes in southern than in northern cities in
the antebellum era. It may also be the case that these free blacks’ forebears had not passed
significant wealth on to them. Female-headed household wealth did not have a significant
relationship with immigrant status. Unlike international immigration, internal migration has no
statistically or economically meaningful effect on household wealth once the city controls are
added.
Household size was significantly related to household wealth. The addition of one person
to a male-headed household was associated with more than a 10 percent increase in male-headed
household wealth, while the increase was somewhat smaller in a female-headed household. The
results are consistent with the hypothesis that larger extended households, which were relatively
-18-
common in 1860, brought together larger agglomerations of wealth than smaller nuclear
households.
Literacy, as expected, had an economically meaningful association with household
wealth. Households in which the male head was literate had more than double (e0.83 = 2.3) the
wealth of households with an illiterate male head (about 50% more if we control for occupational
status). There was little relationship between wealth and female heads’ literacy. Interestingly,
the addition of a literate female was associated with an increase in male-headed household
wealth of about 20 percent, and an increase in female-headed household wealth by about 25
percent, but the addition of a literate adult male did not add significantly to household wealth.
We do not have an explanation for this phenomenon.
Turning now to the issue of colorism, we consider the estimated coefficients on the Black
and Mulatto variables, as well as that on the AfAm variable in the incompletely reported
companion regressions. (See the first row of coefficients in Table 2 for the latter, and the second
and third rows for the former pair of variables.) It is immediately apparent that standard
specifications of a single race identifier do not fully capture the subtleties of race and racial
wealth differences. Simply comparing the means of the white, mulatto, and black subpopulations
implies that a household headed by an African-American male had just 30 to 50
percent (e-1.26 to e-0.76)—or less—of the wealth of the average household headed by a white male.
But when we separate African-American households into those with mulatto and black heads of
household, there are significant and meaningful differences between them. Households headed
by a black male had on average less than 40 percent (e-0.95)—probably less than 25 percent (e-
1.46)—of the wealth of households headed by white males, while the mean household with male
mulatto head had at least 40 percent – and perhaps as much as 63 percent (e-0.46) -- of white male
-19-
wealth. Even more distinct bounds can be placed on female-headed households’ wealth
differentials by racial ancestry. The fact that the separate male and female regressions generate
such similar results increases our confidence that we are identifying a real phenomenon. Further
evidence is provided in the following section by estimates of separate wealth equations for each
subpopulation where we decompose the differences into characteristics and treatment effects.
We recognize that, if colorism was as powerful a social force as we contend, ancestry and
skin shade may have influenced wealth accumulation because lighter-complected individuals
received preferential treatment at younger ages that created occupational opportunities not
available to those with darker complexions. Dark individuals may have been discriminated
against through the apprenticeship system or through other types of occupational training.
Columns (4) in Table 2 address this concern by including the Occupational Prestige (SEI) and
No Occupation variables. As we recognize above, the coefficients on the colorism variables are
smaller than in columns (3), but the colorism effect does not disappear. It remains a powerful
force even after controlling for occupation. In every specification, we reject the null hypothesis
of Black and Mulatto coefficient equality at confidence level p0.001.
The shortcoming of the analysis in Table 2 is that it imposes equality restrictions on the
non-race (non-colorism) coefficients across the subpopulations. Thus the analysis above ignores
the possibility that, in controlling for the household’s characteristics, we have missed the
differing return to characteristics across subpopulations. In the next section, we report the results
of standard Oaxaca-Blinder regression decompositions to investigate the extent to which the
colorism effect was attributable to differences in characteristics across groups versus arising
from the treatment afforded those of different complexions.
-20-
Before moving on to the decompositions, we should also note that the regressions’
adjusted-R2 values indicate that there is much variation in household wealth remaining to be
explained. Because the decomposition techniques we use below rely on an analysis of residuals,
the results must still be considered only imperfect insights into the treatment effects that follow
from colorism.
5. Did light-complected households receive preferential treatment? Evidence from
regression decompositions
Wealth differences between any two groups may be due to differences in the average features of
that group’s members (characteristic effects), or to systematic differences in how each group is
rewarded for those features by the marketplace (treatment effects). The premise of this article is
that colorism influences economic outcomes, generating greater household wealth in lightcomplected
households for a given set of observable characteristics. In order to better
understand the contributions of these two effects of colorism, we use the following methodology.
For each gender, separate coefficients are estimated for the white, mulatto, and black
subpopulations. Because there are some concerns that the occupational prestige variable may be
endogenous, regressions and decompositions are reported with and without the occupational
variables, as in columns 3 and 4 of Table 2. We omit the regression results themselves to
conserve space and because they do not meaningfully differ from those in Table 2. We then use
these regressions to further investigate the differences between subpopulations’ wealth using the
decomposition method of Oaxaca (1973) and Blinder (1973), in addition to Cotton’s (1988)
modified form.
-21-
The goal of the decomposition is to consider the counterfactual implied by the assertion
that different groups have different wealth levels, due perhaps to colorism: what wealth would
the members of each group expect to have in the absence of their identification as a member of
their group? Although we will construct a decomposition for all four pairs of sub-populations,
consider for concreteness the pair of subpopulations whose wealth differences may illustrate
colorism: mulattoes and blacks. Label the associated pair of regressions
lnwMulatto = aMulatto + XMulatto bMulatto + e and lnwBlack = aBlack + XBlack bBlack + e. To decompose the
advantages accrued by those with lighter skin, we consider the difference in predicted log wealth
evaluated at the explanatory variables’ means (indicated by “ ” notation) and parameter
estimates (indicated by “ ˆ ” notation)
(3) lnwMulatto - lnwBlack = ˆ a Mulatto - ˆ a Black + X Black ( ˆ b Mulatto - ˆ b Black ) [ ]+ (X Mulatto - X Black ) ˆ b Mulatto [ ] ,
where the first bracketed term is the unexplained, or treatment, effect and is presumed to be due,
at least in part, to colorism. The second term is the characteristic effect, the effect on mean log
wealth due to differences between the two groups. With the structure of (3) we are taking the
perspective that in the absence of colorism the advantaged group’s wealth structure would
prevail. The opposite assumption is to presume that all groups would receive the disadvantaged
group’s wealth structure in the absence of colorism, or
(4) lnwMulatto - lnwBlack = ˆ a Mulatto - ˆ a Black + X Mulatto ( ˆ b Mulatto - ˆ b Black ) [ ]+ (X Mulatto - X Black ) ˆ b Black [ ] .
These two decompositions provide a range within which we expect the actual effect of colorism
to lie. Many alternative decompositions have been proposed to predict a particular value inside
this range. Because we feel, along with other authors, that the counterfactual wealth is likely to
lie closer to the larger group’s prevailing wealth, we calculate Cotton’s (1988) decomposition as
well. That is,
-22-
(5)
lnwMulatto - lnwBlack
= ˆ a Mulatto - a*+X Mulatto ( ˆ b Mulatto - b*) [ ]+ a*- ˆ a Black + X Black ( b*- ˆ b Black ) [ ]
+ (X Mulatto - X Black ) b* [ ]
The vector of parameters a*, b* ( ) generalizes the previous two cases: a*, b* ( )= aLight , bLight ( ) in
(3), while a*, b* ( )= aDark, bDark ( ) in (4). The parameter vector that we actually use for a*, b* ( )
is a weighted sum of the previous parameters, where the weights are the relative frequencies of
the two populations9. This decomposition has the added feature that it apportions the treatment
effect into components that indicate the gain of the advantaged subpopulation and the loss of the
disadvantaged population (respectively the first and second bracketed terms above), but we do
not report these separately because our focus is instead on the total treatment effect, the addition
of those first two bracketed terms. It is also worth noting that, because the weights are the
subpopulations’ relative frequencies, when comparing the treatment of people of color (blacks,
mulattoes, or African-Americans as a whole) to the white subpopulation this decomposition
weights the advantaged population (the majority whites) more heavily, but when considering the
existence of colorism it is the disadvantaged population (the majority blacks) that receives the
higher weight.
The two panels of Table 3 first provide estimates of the subpopulation pairs’ relative
predicted wealth and the resulting raw wealth differentials, as well as the subpopulations’
predicted geometric mean wealth (the exponential of the arithmetic log mean). Note that our
discussion of the raw wealth ratios compared across subgroups from the previous section’s
results with pooled regression is similar to the results presented here from separate regressions.
Households headed by black men hold on average 13% of white wealth, while mulatto-headed
9 Due to our sampling method, the relative frequencies are actually calculated as the sum of each subpopulation’s
probability weights.
-23-
households’ average wealth is 35% of white-headed households. The results for female-headed
households are quite similar to those for male-headed households.
Next in Table 3’s two panels, the raw differential is decomposed into two parts using the
method described above and estimation of the model from either column 3 (no occupation
variables) or column 4 (with occupation variables) of Table 2. The raw differential is
decomposed into both the treatment and characteristic effects, but to save space Table 3 presents
only the treatment effect. This effect is the (imperfect) estimate of discrimination or colorism.
As well, the characteristic effect (to which we return later in the table) need not be listed
explicitly because it is simply the remainder of the raw differential. Our discussion focuses on
the third set of treatment effect estimates, those typeset in boldface and listing the characteristic
effect as C = b* (XH
- - XL ), using Cotton’s (1988) decomposition. For example, on the right
side of Table 3 where the results of pairing the white and African-American subpopulations are
presented, we see that for male-headed households the treatment effect is estimated at 69% of the
raw differential with model (3) and 42% with model (4). Thus the characteristic effect is 31%
and 58%, respectively, for the two models’ application to this pair of subpopulations. These
percentages are the focus of this portion of the table: how much of the raw wealth differential is
explained by treatment effects?
For the moment, compare the boldface treatment effects to those presented in the rows
above them for each gender. We see—as should have been expected—that Cotton’s estimate is
always between those of the traditional Oaxaca and Blinder decompositions. Its counterfactual
assumption is a weighting of the two groups’ returns to characteristics, while the Blinder-Oaxaca
decompositions take two possible extreme counterfactual assumptions: that either the advantaged
-24-
or the disadvantaged group’s returns to characteristics would obtain in the absence of differential
treatment across groups.
Our first evidence of colorism in Table 3 is simply to note the large estimate of the
treatment effect in the first column’s decomposition of the Mulatto and Black subpopulation pair
of regressions. This treatment effect is about 40 percent of the raw wealth differential, with little
difference dependent on including occupational status. Considering all sub-population pairs, the
differences between models 3 and 4 illustrate another result we should expect: controlling for
occupation dampens the treatment effect; more of the raw wealth differential can be explained by
including this characteristic of the household head to in the regressions. Duncan’s SEI is one
way in which all subpopulations are different from each other, it has a positive relationship with
wealth, and the more advantaged subpopulations have higher SEI.
On the other hand the larger white-mulatto treatment effect (85%) as compared to the
white-black treatment effect (66%), without controlling for occupation (model 3), can be
explained by mulattoes’ characteristics having greater similarity to whites’ characteristics than
blacks’ characteristics have to whites’ characteristics. Even with a higher return to mulattoes’
characteristics as compared to blacks’, more of the raw wealth differential is left unexplained by
any difference in mulattoes’ characteristics as compared to whites precisely because their
characteristics are less different. The resulting smaller characteristics effect leads to a larger
treatment effect. Once occupation is included (model 4), the difference in treatment effects is
small (41% and 43%, for males). The treatment effect for the mulatto-black pair of
subpopulations shows little change whether we include occupation or not (38% versus 42%).
The final portion of each of Table 3’s panels then uses the boldface estimate of the
treatment effect (Cotton’s method) to estimate two more comparisons of the results for the
-25-
subpopulation pairs. First is the estimated return to characteristics, the wealth ratio. It is the
ratio of the advantaged group’s wealth to the disadvantaged group’s wealth. The values are all
greater than 1, reflecting the inherent advantage in characteristics that the lighter-skinned group
has, regardless of differential treatment. Loury (1998) focuses on this as an issue that remains
even after dealing with discriminatory practices: the fact that a subpopulation has a disadvantage
in characteristics should still be of concern even if there is no treatment effect due to
discrimination. First note that including another explanatory variable (in particular, occupation)
in the regression will always increase the return to characteristics. These comparisons confirm
our earlier observation that whites had a much greater advantage in characteristics over blacks (a
factor of 2 to 3) than they did over mulattoes (always less than 2). Likewise, the advantage of
mulattoes in characteristics as compared to blacks (remarkably similar for males and females) is
about a factor of 1.8. The final comparison in each panel is Oaxaca’s (1973) adaptation of
Becker’s (1971) discrimination coefficient, the relative benefit to the advantaged group from
discrimination, assuming that the entire treatment residual can be attributed to discrimination (as
measured by Cotton’s residual). It is measured as the percentage gain in the wealth ratio as
compared to what it would be without differential treatment. Similar patterns are seen in these
comparisons as those above. The gains from treatment are large, and the differences within
African-Americans are also large.
So we see that, although differences in underlying characteristics were a driving force
behind the mulatto-black wealth gap, treatment disadvantages were substantial. The magnitude
of the treatment disadvantage estimates supports the colorism hypothesis, and reflecting on the
characteristic effect further strengthens that hypothesis. Further research needs to uncover the
extent to which pre-market discrimination led to differences in observable wealth-generating
-26-
characteristics. It is well known that light-complected slaves were more likely to receive skill
training (Margo 1992), for example. It may have been that mixed-race free African Americans,
too, received more or better education and job skills than monoracial blacks. Uncovering
whether and why this conjecture was true represents a potentially fruitful line of further inquiry.
6. Concluding Remarks
Does black always signify Black? In matters of race economists have assumed it does. Nearly
every economic investigation into the consequences of race uncritically and without pause
separates Whites and Blacks into distinct and mutually exclusive categories. For the purposes of
empirical work done during the past 50 years such a dichotomous racial classification scheme
was probably appropriate, an intellectual and academic recognition of a broad social acceptance
of Jim Crow’s “one-drop” rule. In the future, however, dichotomous racial categories may not
capture the increasing subtleties of race as race mixing spreads and becomes increasingly
socially acceptable. Empirical economics cannot continue to model race as a binary variable and
expect the resulting coefficient to meaningfully capture the subtleties of race.
Our paper shows that mulattoes were treated differently than blacks in the mid-nineteenth
century. As a percentage of white households’ log wealth, the predicted log wealth of blackheaded
households was 20 percentage points lower than that of mulatto-headed households.
Regression decompositions reveal that treatment effects played a large role in explaining the
white-mulatto wealth gap, even after including a measure of occupation; treatment effects played
a comparable role for other subpopulation pairs as well. Our results are consistent with those
from a sociological literature dating to Reuter (1917) and an emergent economics literature that
finds that colorism remains a powerful social phenomenon in minority communities. An
-27-
important line of further inquiry will be to understand how color-based identities are established
and reinforced in historical and modern populations. We intend to contribute to that developing
literature as well.
References
Anonymous. “The Negro Defined.” Yale Law Journal 20:3 (January 1911): 224-25.
Becker, Gary S. The Economics of Discrimination (Chicago: University of Chicago Press, 1957).
Becker, Gary S. “A Theory of Marriage, Part I.” Journal of Political Economy 81:4 (July 1973),
813-46.
Becker, Gary S. “A Theory of Marriage, Part II.” Journal of Political Economy 82:2 (March
1974), S11-S26.
Blau, Francine D. and John W. Graham. “Black-White Differences in Wealth and Asset
Composition.” Quarterly Journal of Economics 105:2 (May 1990), 321-39.
Blinder, Alan S. “Wage Discrimination: Reduced Form and Structural Estimates.” Journal of
Human Resources 8 (Fall 1973): 436-455.
Bodenhorn, Howard. “The Complexion Gap: The Economic Consequences of Color among Free
African Americans in the Rural Antebellum South.” Advances in Agricultural Economic History
2 (2003), pp. 41-73.
Bodenhorn, Howard and Christopher S. Ruebeck. “The Economics of Identity and the
Endogeneity of Race.” NBER working paper #9962 (September 2003).
Borjas, George. “Economics of Immigration.” Journal of Economic Literature 32:4 (December
1994), 1667-1717.
Borjas, George. “Ethnic Enclaves and Assimilation.” Swedish Economic Policy Review 7:2 (Fall
2000), 89-122.
-28-
Bynum, Victoria E. “‘White Negroes’ in Segregated Mississippi: Miscegenation, Racial Identity,
and the Law.” Journal of Southern History 64:2 (May 1998): 247-76.
Conley, Dalton. Being Black, Living in the Red. Berkeley: University of California Press, 1999.
Cotton, Jeremiah “On the Decomposition of Wage Differentials.” The Review of Economics and
Statistics 70:2 (May 1988): 236-243.
Curry, Leonard P. The Free Black in Urban America, 1800-1850: The Shadow of a Dream.
Chicago and London: University of Chicago Press, 1981.
Daniel, G. Reginald. More Than Black?: Multiracial Identity and the New Racial Order.
Philadelphia: Temple University Press, 2002.
Darity, William A., Jr. and Tanya Golash Boza. “Choosing Race: Evidence from the Latino
National Political Survey.” Unpublished working paper, University of North Carolina at
Chapel Hill, 2003.
Darity, William A., Jr., Jason Dietrich, and Darrick Hamilton. “Bleach in the Rainbow: Latin
Ethnicity and Preference for Whiteness.” Transforming Anthropology, Forthcoming.
Darity, William A., Jr., Patrick L. Mason, and James B. Stewart. “The Economics of Identity:
The Origin and Persistence of Racial Identity Norms.” Journal of Economic Behavior
and Organization, Forthcoming.
Doyle, Jamie Mihoko. “Self-Identified Multiracials and Sexual Debut: A Move Toward
Explanations.” Unpublished working paper, University of Pennsylvania Population
Studies Center, undated.
Freeman, Howard E.; David Armor; J. Michael Ross; and Thomas F. Pettigrew. “Color
Gradation and Attitudes among Middle-Income Negroes.” American Sociological Review
31:3 (June 1966): 365-374.
-29-
Gatewood, Willard B. Aristocrats of Color: The Black Elite, 1880-1920. Fayetteville, Ark.:
University of Arkansas Press, 2000.
Goldsmith, Arthur H.; Darrick Hamilton; and William Darity, Jr. “From Dark to Light: Skin
Color and Wages among African-Americans.” Unpublished working paper, University of
North Carolina at Chapel Hill, August 2004.
Gross, Ariela J. “Litigating Whiteness: Trials of Racial Determination in the Nineteenth-Century
South.” Yale Law Journal 108:1 (October 1998): 109-88.
Guild, Jane Purcell. Black Laws of Virginia; A Summary of the Legislative Acts of Virginia
concerning Negroes from the Earliest Times to the Present. New York: Negro
Universities Press, 1969.
Gullickson, Aaron. “The Significance of Color Declines: A Re-Analysis of Skin Tone
Differentials in Post-Civil Rights America.” Unpublished working paper, University of
California-Berkeley, September 2003.
Harris, David R. and Jeremiah Joseph Sim. “Who is Multiracial?: Assessing the Complexity of
Lived Race.” American Sociological Review 67:x (August 2002), 614-27.
Herman, Melissa. “The Black-White-Other Test Score Gap: Academic Achievement among
Mixed Race Adolescents.” Unpublished working paper, Northwestern University
Institute for Policy Research, 2002.
Hill, Mark E. “Color Differences in the Socioeconomic Status of African American Men: Results
of a Longitudinal Study.” Social Forces 78:4 (June 2000): 1437-60.
Hughes, Michael and Bradley E. Hertel. “The Significance of Color Remains: A Study of Life
Chances, Mate Selection, and Ethnic Consciousness among Black Americans.” Social
Forces 68:4 (June 1990): 1105-20.
-30-
Johnson, Whittington B. Black Savannah, 1788-1864. Fayetteville, Ark.: University of Arkansas
Press, 1996.
Keith, Verna M. and Cedric Herring. “Skin Tone and Stratification in the Black Community.”
American Journal of Sociology 97:3 (November 1991), 760-778.
Kemble, Frances A. Journal of a Residence on a Georgian Plantation in 1838-1839. New York:
Harper Bros., 1863.
Kenzer, Robert C. Enterprising Southerners: Black Economic Success in North Carolina, 1865-
1915. Charlottesville and London: University Press of Virginia, 1997.
Loury, Glenn C. “Discrimination in the Post-Civil Rights Era: Beyond Market Interactions.” The
Journal of Economic Perspectives 12:2 (Spring 1998): 117-126.
Margo, Robert A. “Civilian Occupations of Ex-Slaves in the Union Army, 1862-1865.” In
Without Consent or Contract: Markets and Production: Technical Papers (Volume I), pp.
170-185. Edited by Robert W. Fogel and Stanley E. Engerman. New York: W. W. Norton
& Co., 1992.
Mason, Patrick L. “Annual Income, Hourly Wages, and Identity Formation among Mexican
Americans and Other Latinos.” Industrial Relations 43:4 (October 2004), 817-34.
Moehling, Carolyn A. “Family Structure, School Attendance, and Child Labor in the American
South in 1900 and 1910.” Explorations in Economic History 41:1 (February 2004), 73-
100.
Mumford, Kevin. “After Hugh: Statutory Race Segregation in Colonial America, 1630-1725.”
American Journal of Legal History 43:3 (July 1999): 280-305.
Murguia, Edward and Edward E. Telles. “Phenotype and Schooling among Mexican
Americans.” Sociology of Education 69:4 (October 1996): 276-89.
-31-
Myrdal, Gunnar, with the assistance of Richard Sterner and Arnold Rose. An American
Dilemma: The Negro Problem and Modern Democracy. New York: Harper & Row,
1944.
Oaxaca “Male-Female Wage Differentials in Urban Labor Markets.” International Economic
Review 14 (October 1973), 693-709.
Oliver, Melvin L. and Thomas M. Shapiro. Black Wealth / White Wealth: A New Perspective on
Racial Inequality. New York and London: Routledge, 1995.
Reiss, Albert J. Occupations and Social Status. New York: The Free Press, 1961.
Reuter, Edward B. “The Superiority of the Mulatto.” American Journal of Sociology 23:1 (July
1917), 83-106.
Reuter, Edward B. “The American Mulatto.” Annals of the American Academy of Political and
Social Science 140 (November 1928), 36-43.
Russell, Kathy, Midge Wilson, and Ronald Hall. The Color Complex: The Politics of Skin Color
among African Americans. New York: Anchor Books, 1992.
Seltzer, Richard and Robert C. Smith. “Color Differences in the Afro-American Community and
the Differences They Make.” Journal of Black Studies 21:3 (March 1991), 279-86.
Shapiro, Thomas M. The Hidden Cost of Being African American: How Wealth Perpetuates
Inequality. Oxford and New York: Oxford University Press, 2004.
Snipp, C. Matthew. “Racial Measurement in the American Census: Past Practices and
Implications for the Future.” Annual Review of Sociology 29 (2003): 563-88.
Telles, Edward E. and Edward Murguia. “Phenotypic Discrimination and Income Differences
among Mexican Americans.” Social Science Quarterly 71: (1990): 682-96.
-32-
Williamson, Joel. New People: Miscegenation and Mulattoes in the United States. New York:
The Free Press, 1980.
Woodson, Carter G. The Education of the Negro prior to 1861. Washington, D.C.: Associated
Publishers, Inc., 1919.
-33-
Table 1: Summary statistics
Male-headed households Female-headed households
Full sample White Af-Am Black Mulatto Full sample White AfAm Black Mulatto
Af-Am 0.502 0.803
Black 0.319 0.463
Mulatto 0.183 0.340
Wealth 3,380 6,328 455 231 846 1,023 3,707 363 177 617
(18839) (26285) (2079) (1009) (3138) (6130) (13111) (1556) (842) (2154)
ln(Wealth) 5.074 5.924 4.230 3.876 4.846 4.176 5.424 3.869 3.450 4.439
(2.280) (2.440) (1.738) (1.565) (1.849) (1.971) (2.415) (1.711) (1.519) (1.793)
Age 40.354 40.089 40.617 40.951 40.034 42.923 45.008 42.411 43.174 41.371
(11.817) (11.358) (12.250) (12.409) (11.948) (14.054) (13.440) (14.155) (14.212) (14.016)
Immigrant 0.252 0.490 0.015 0.008 0.029 0.080 0.296 0.027 0.014 0.044
Migrant 0.105 0.139 0.072 0.058 0.096 0.098 0.147 0.086 0.070 0.108
SEI 22.320 29.756 14.940 13.446 17.545 14.892 18.487 14.008 12.895 15.524
(17.907) (20.495) (10.617) (9.377) (12.054) (8.524) (12.154) (7.093) (6.521) (7.548)
No occupation 0.058 0.050 0.065 0.057 0.079 0.456 0.684 0.400 0.332 0.492
Literate head 0.821 0.971 0.672 0.606 0.788 0.736 0.944 0.684 0.626 0.764
Literate males 1.209 1.506 0.914 0.796 1.119 0.485 0.787 0.411 0.361 0.478
(0.961) (1.009) (0.810) (0.749) (0.869) (0.879) (1.094) (0.800) (0.700) (0.914)
Literate females 1.109 1.383 0.837 0.723 1.034 1.280 1.776 1.158 1.039 1.320
(0.950) (0.954) (0.864) (0.805) (0.926) (1.076) (1.125) (1.028) (0.980) (1.070)
Adults in hhold 5.381 5.965 4.801 4.608 5.137 4.174 5.001 3.971 3.799 4.205
(2.662) (2.706) (2.485) (2.398) (2.597) (2.436) (2.539) (2.366) (2.265) (2.479)
Baltimore, Md 0.454 0.331 0.576 0.722 0.321 0.319 0.232 0.340 0.461 0.175
Baton Rouge, La 0.009 0.012 0.007 0.001 0.017 0.012 0.021 0.009 0.004 0.017
Charleston, SC 0.077 0.105 0.049 0.022 0.097 0.121 0.138 0.117 0.069 0.182
Frederick, Md 0.030 0.032 0.028 0.025 0.033 0.027 0.050 0.021 0.019 0.024
Louisville, Ky 0.046 0.055 0.037 0.037 0.038 0.039 0.049 0.037 0.042 0.030
Mobile, Al 0.025 0.032 0.018 0.004 0.044 0.024 0.038 0.020 0.008 0.037
Nashville, Ky 0.019 0.027 0.012 0.011 0.013 0.024 0.036 0.021 0.016 0.027
New Orleans, La 0.203 0.238 0.169 0.063 0.355 0.223 0.238 0.220 0.119 0.357
Petersburg, Va 0.085 0.101 0.069 0.089 0.034 0.132 0.121 0.135 0.186 0.064
Richmond, Va 0.051 0.068 0.035 0.027 0.049 0.080 0.078 0.081 0.076 0.087
Observations 11,592 5,774 2,121 3,697 5,818 4,269 842 1,451 1,976 3,427
Means, with standard deviations (except for dichotomous variables) in parentheses.
As discussed in Section 3, we have imputed wealth values at the ward level.
All same-gender subpopulation pairs of variables have means significantly different at p0.01 (p0.001 in almost all
cases), except men’s age and the no-occupation indicator for white versus black men. Due to undersampling of
whites, the city indicators can only be compared for black and mulatto subsamples; those other than Fredericksburg,
Louisville, and Richmond consistently differ across African-American classification with p0.001.
Sources: U.S. Bureau of the Census, Eighth Census (1860), population census manuscripts.
-34-
Table 2: Determinants of ln (household wealth) in the urban south: ordinary least squares regression
Dependent variable Male-headed households Female-headed households
is ln(wealth) (1) (2) (3) (4) (1) (2) (3) (4)
AfAm† -1.731*** -1.553*** -1.257*** -0.759*** -1.481*** -1.335*** -1.187*** -1.092***
(0.247) (0.250) (0.186) (0.169) (0.240) (0.233) (0.193) (0.199)
Black -2.052*** -1.842*** -1.456*** -0.949*** -1.908*** -1.738*** -1.452*** -1.344***
(0.271) (0.278) (0.220) (0.202) (0.247) (0.239) (0.199) (0.213)
Mulatto -1.215*** -1.091*** -0.945*** -0.462*** -0.958*** -0.855*** -0.875*** -0.814***
(0.213) (0.215) (0.152) (0.131) (0.219) (0.218) (0.198) (0.200)
Age 0.159*** 0.133*** 0.136*** 0.124*** 0.090** 0.081** 0.078** 0.077**
(0.023) (0.020) (0.019) (0.019) (0.027) (0.027) (0.026) (0.027)
Age squared -0.001*** -0.001*** -0.001*** -0.001*** -0.001** -0.001* -0.001* -0.001*
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Immigrant -0.714*** -0.618** -0.559*** -0.228^ -0.239 -0.18 -0.503* -0.480^
(0.181) (0.184) (0.138) (0.119) (0.318) (0.321) (0.245) (0.250)
Migrant 0.313 0.374^ 0.172 0.091 0.282 0.346 -0.059 -0.088
(0.199) (0.193) (0.142) (0.123) (0.236) (0.232) (0.249) (0.262)
Literate 0.991*** 0.838** 0.830*** 0.501* 0.584*** 0.344^ 0.322 0.278
(0.240) (0.253) (0.223) (0.215) (0.149) (0.201) (0.224) (0.232)
Adults in hhold 0.105*** 0.135*** 0.123*** 0.104* 0.088* 0.086*
(0.028) (0.024) (0.023) (0.040) (0.036) (0.036)
Literate males -0.011 -0.047 -0.059 -0.09 -0.094 -0.105
(0.048) (0.042) (0.039) (0.094) (0.078) (0.077)
Literate females 0.162* 0.197*** 0.115* 0.205* 0.251* 0.254**
(0.066) (0.052) (0.053) (0.102) (0.099) (0.091)
SEI 0.038*** 0.007
(0.002) (0.012)
No Occupation 0.15 0.306
(0.182) (0.185)
City controls No No Yes Yes No No Yes Yes
Constant 0.980^ 0.958^ 0.184 -0.2 2.627*** 2.290** 1.863* 1.653*
(0.564) (0.526) (0.432) (0.405) (0.734) (0.765) (0.733) (0.812)
Adj. R-squared 0.163 0.185 0.226 0.310 0.156 0.180 0.228 0.231
N 11592 11592 11592 11592 4269 4269 4269 4269
prob F>0 for Mulatto==Black 0.000 0.000 0.001 0.001 0.000 0.000 0.000 0.000
†The first row (italicized) reports only the African-American indicator variable estimated in equations that also
included the variables in the associated columns but did not include the Black and Mulatto variables.
Other notes: Robust standard errors (allowing for correlation within wards, Stata®’s cluster option) are in
parentheses. Probability weighting was used to adjust for undersampling of white-headed households. Significance
is indicated by ^ p0.1, * p0.05, ** p0.01, *** p0.001.
-35-
Table 3: Complexion-based wealth decompositions’ treatment (T) and characteristics (C) effects
Male-headed Advantaged group (H): Mulatto White White White
Disadvantaged (L): Black Mulatto Black African-American
mean log wealth: ln(wH), ln(wL) 4.846, 3.876 5.888, 4.846 5.888, 3.876 5.888, 4.230
geometric mean wealth ($): wH, wL 127.2, 48.2 360.7, 127.2 360.7, 48.2 360.7, 68.7
raw differential: R = ln(wH) - ln(wL) 0.970 1.042 2.012 1.658
raw ratio (%): wL/wH = e-R 38% 35% 13% 19%
Model From Table 2: (3) (4) (3) (4) (3) (4) (3) (4)
treatment: T = R – C, (as % of R) treatment: T = R – C, (as % of R)
Advantaged group, C = bH(xH - xL) 0.444 0.402 0.888 0.431 1.354 0.900 1.184 0.729
(46%) (41%) (85%) (41%) (67%) (45%) (71%) (44%)
Disadvantaged group, C = bL(xH - xL) 0.391 0.344 0.686 0.404 0.664 0.308 0.701 0.378
(40%) (35%) (66%) (39%) (33%) (15%) (42%) (23%)
Population-weighted, C = b*(xH - xL) 0.411 0.365 0.882 0.431 1.321 0.873 1.149 0.704
(42%) (38%) (85%) (41%) (66%) (43%) (69%) (42%)
Frequency of advantaged group 0.365 0.972 0.953 0.928
Assumed return to characteristics in
absence of discrimination
WH/WL, estimated wealth due to
characteristics ratio = eC = eR-T 1.75 1.83 1.17 1.84 2.00 3.13 1.66 2.60
D = eT - 1 — see notes 51% 44% 142% 54% 275% 139% 215% 102%
Female-headed Advantaged (H): Mulatto White White White
Disadvantaged (L): Black Mulatto Black African-American
mean log wealth: ln(wH), ln(wL) 4.439, 3.450 5.538, 4.439 5.538, 3.450 5.538, 3.869
geometric mean wealth ($): wH, wL 84.7, 31.5 254.2, 84.7 254.2, 31.5 254.2, 47.9
raw differential: R = ln(wH) - ln(wL) 0.989 1.099 2.088 1.669
raw ratio (%): wL/wH = e-R 37% 33% 12% 19%
Model from Table 2: (3) (4) (3) (4) (3) (4) (3) (4)
treatment: T = R – C, (as % of R) treatment: T = R – C, (as % of R)
Advantaged group, C = bH(xH - xL) 0.422 0.380 0.761 0.716 1.325 1.245 1.086 1.021
(43%) (38%) (69%) (65%) (63%) (60%) (65%) (61%)
Disadvantaged group, C = bL(xH - xL) 0.421 0.387 0.587 0.475 1.122 1.043 0.833 0.729
(43%) (39%) (53%) (43%) (54%) (50%) (50%) (44%)
Population-weighted, C = b*(xH - xL) 0.421 0.384 0.739 0.686 1.291 1.212 1.022 0.947
(43%) (39%) (67%) (62%) (62%) (58%) (61%) (57%)
Frequency of advantaged group 0.423 0.874 0.836 0.746
Assumed return to characteristics in
absence of discrimination
WH/WL, estimated wealth due to
characteristics ratio = eC = eR-T 1.76 1.83 1.43 1.51 2.22 2.40 1.91 2.06
D = eT - 1 — see notes 52% 47% 109% 99% 264% 236% 178% 158%
Notes: Calculated from regressions specified as those reported in columns 3 and 4 of Table 2, but run separately for
each color and gender subpopulation. Oaxaca’s (1973) adaptation of Becker’s (1971) discrimination coefficient, D
= ((wH/wL)-(wH/wL)) / (wH/wL) = eT – 1, is the increase in wealth due to unmeasured differences, using the boldface
(population-weighted) counterfactual estimate of the treatment residual, C = b*(xH – xL). |
Ntfs_encryption
"Cyniquian" Level Poster
Username: Ntfs_encryption
Post Number: 315
Registered: 10-2005
Rating: N/A
Votes: 0 (Vote!) | | Posted on Tuesday, July 04, 2006 - 04:17 pm: |
|
Uhhhh....I guess there is a point here? |
Prettybabygirl
"Cyniquian" Level Poster
Username: Prettybabygirl
Post Number: 356
Registered: 04-2005
Rating: N/A
Votes: 0 (Vote!) | | Posted on Tuesday, July 04, 2006 - 05:07 pm: |
|
Is there a point to the documentation of racism Ntfs?
|
Tonya
"Cyniquian" Level Poster
Username: Tonya
Post Number: 2881
Registered: 07-2005
Rating: N/A
Votes: 0 (Vote!) | | Posted on Tuesday, July 04, 2006 - 06:39 pm: |
|
This should be a LITTLE clearer, sorry y'all.
. . . . . .
ABSTRACT
Black is not always black. Subtle distinctions in skin tone translate into significant differences in outcomes. Data on more than 15,000 households interviewed during the 1860 federal census exhibit sharp differences in wealth holdings between white, mulatto, and black households in the urban South. We document these differences, investigate the relationships between wealth and the recorded household characteristics, and decompose the wealth gaps into treatment and characteristic effects. In addition to higher wealth holdings of white households as compared to free African-Americans in general, there are distinct differences between both the characteristics of and wealth of free mulatto and black households, whether male- or female-headed. While black-headed households' mean predicted log wealth was only 20% of white-headed households', mulatto-headed households' was nearly 50% that of whites'. The difference between light- and dark-complexion is highly significant in semi-log wealth regressions. In the decomposition of this wealth differential, treatment effects play a large role in explaining the wealth gap between all subpopulation pairs.
Colorism and African-American Wealth: Evidence from the Nineteenth-Century South
1. Introduction
The first documented case of a white man being punished for engaging in miscegenation
occurred in Virginia in 1630, just 10 years after the arrival of the first Africans into the colony (Mumford 1999). Interracial sex is not new and neither are the questions of the relationship between identity and race. Throughout the nineteenth and twentieth centuries, both whites and blacks were uncomfortable with miscegenation and mixed-race people(Bynum 1998). Although the number of interracial unions in the United States is currently increasing, the desire of mixedrace individuals to seek a racial identity separate from their constituent racial ancestries is not new. For many of African descent, black is not black—both in terms of how they view themselves and how others view them. There are meaningful subtleties of shade, differences that have socially and politically meaningful distinctions today—just as they have in the past.1 Despite the complex history of miscegenation, researchers have been complacent about the content of federally defined racial categories (Snipp 2003). With a few recent exceptions (Bodenhorn and Ruebeck 2003; Darity, Mason, and Stewart Forthcoming), economists have used the blunt instrument of binary racial classifications to categorize individuals as either black or white. Researchers consider all individuals within a category to be subject to similar treatment. Relatively little attention, if any, is paid to the concordance between social constructions of race and official categorizations. What may appear to be self-evident—whether an individual is black
_______________
1 Modern studies of the social consequences of skin shade differences include Russell, Wilson and Hall (1992), Root (1996), Daniel (2002), Tizard and Phoenix (2002), Herman (2002), and Doyle (undated).
_____________________
-2-
or white—actually masks a complex reality and likely does not capture the subtleties of racial interactions. If different skin tones are treated unequally, the continued practice of scholarly racial categorization based on anachronistic notions developed in the Jim Crow South may not lead to useful or relevant insights into modern phenomena. As Trina Jones (2000, p. 1499) notes, to understand “the nuanced ways in which discrimination operates and differentially impacts similarly but not identically situated people, we must examine and attempt to understand color. Such an understanding begins with history.” Embracing Jones’ call for historical understanding, our paper turns to history to document the economic consequences of colorism, thereby providing fresh insights into an increasingly modern concern. Using data on more than 15,000 households interviewed during the 1860 federal census, we find sharp differences in wealth holdings between white, mulatto and black households in the urban South.2 This study advances the literature in two ways. First, we consider the operation and effects of colorism closer to its historical origins. Historians attribute the origins of lightcomplexion preferences to slaveholders. By investigating data on complexion-based differences in attainment of free African-Americans in the pre-Civil War South, our study provides a more objective basis for such beliefs and provides a baseline against which to consider modern outcomes. Second, recent economic and sociological research has stressed the importance of household wealth, rather than income, for the determination of socioeconomic status (Conley 1999; Shapiro 2004). Accumulations of household wealth have been shown to insulate middle-
__________________
2 Teo (2003) notes how the use of terms such as mulatto and colored, as well as the conception of mixed-blood, are terms heavily laden with racial, often racist, connotations. This is not the place to develop a new terminology. We will use the term mulatto in the contemporary context to describe mixed-race and, generally, light-complected African Americans. We recognize that some readers dislike the word, and ask for their indulgence. Some have suggested the terms biracial or multiracial be used instead, but we resist because the modern use of these terms is politically charged and does not capture the meaning of mulatto as it was used in the mid-nineteenth century. We will also use the terms light-complected and dark-complected to recognize the likely factor that determined the “mulatto” and “black” identifications and to refer to modern studies of colorism.
________________________
-3-
and working-class families from falling into poverty when they face even short-term labor market disruption or other temporary crises (Oliver and Shapiro 1995). Previous studies of the economic consequences of modern colorism have focused on income (Goldsmith, Hamilton, and Darity (2004); Darity, Dietrich and Hamilton (Forthcoming)), which incompletely captures the consequences of differential treatment according to skin shade. With its twin focus on wealth holdings and complexion, this paper demonstrates that the economic consequences of colorism were (and likely are) far reaching. Section 2 summarizes the history and modern experience with colorism in the African American community, including just a handful of recent economic studies. We then proceed through an empirical analysis in three steps. Section 3 describes the data and takes a first look at the three sub-populations’ differences evident in the 1860 census manuscripts. In Section 4 we present several regressions in which we show that bivariate racial classification does not capture the subtleties of race even after controlling for a wide variety of individual and household influences. By separating African-Americans into blacks and mulattoes, we find meaningful differences in outcomes between households with a black or a mulatto head. Section 5 presents decompositions of the complexion wealth gap to identify how the wealth gaps are influenced by differences in wealth-generating characteristics between the groups and by differences in treatment between the groups not explained by the observed characteristics other than skin color. The decompositions show that this treatment effect explains a substantial fraction of the observed complexion wealth gaps among all sub-populations. Section 6 summarizes and offers some concluding remarks.
2. Colorism and socioeconomic outcomes
-4-
History provides compelling evidence of a consistent pattern of preferences shown, by both
whites and blacks, toward light-complected African Americans. The economic advantages of a light complexion date back to slavery. Reuter (1918) and Frazier (1957), among others, argue that a complexion advantage appeared early in the slavery era when slaveholders disproportionately selected light-complected blacks to work as house servants and field foremen. Indeed, some slaveholders contended that “no man would buy a mulatto for field work” (Johnson 1996, p. 111), while others reported that light-complected slaves resisted field work by arguing that their physical constitutions could not stand up to the demands of hard labor (Kemble 1863, p. 193). Although few slaveholders balked at setting light-complected slaves to field work, they preferentially selected light-complected slaves for craft training and apprenticeship. According to Frazier (1957) a fair complexion improved a slave’s life chances by significantly reducing his or her toil and drudgery, by improving his or her access to food and shelter, and by exposure to the culture, as well as the manners, dress, and linguistic conventions of whites. Light-complected, mixed-race slaves were also more likely to be educated and manumitted. Moreover, despite legal prohibitions in many states against slaveholders providing for slaves in their wills, masters did occasionally provide for their mixed-race offspring at their passing, sometimes at the expense of white offspring (Horowitz 1973). In the antebellum South, free light-complected blacks were more likely than darkcomplected blacks to be literate. Horton and Horton (1997, p. 154), for example, report general white opposition to black school attendance except for those of “very light complexion.” With geater access to education, mulattoes labored at more remunerative occupations than blacks. Neither mulattoes nor blacks labored in high prestige occupations but, among African Americans, mulattoes dominated positions at the top of the employment ladder—merchants,
-5-
shopkeepers, and skilled craftsmen. Those with fair complexions were also overrepresented among skilled workers and professionals. Among rural free African Americans, Bodenhorn (2003) finds that mulattoes were more likely than blacks to work as tenants or to own their own farms. Gatewood (2000) shows that social preferences for light-skinned blacks persisted through the post-Civil War period, but the increasing severity of Jim Crow and the movement toward the one-drop rule3 worked to politically and socially align light and dark blacks, groups that were often at odds prior to the Civil War. Nevertheless, color-based differences persisted and by 1920, colorism and complexion-based stratification in the African American community gained the attention of sociologists, notably E. B. Reuter (1918, 1928). A number of studies conducted through the 1930s and 1940s, including Myrdal’s (1944) magnum opus, found support for Reuter’s contention that light skin tones and perceptible traces of non-African heritage were associated with material advantages for African Americans (Hill 2000). A rhetoric of complexion egalitarianism, however, seemingly reigned through the Civil Rights era and into the 1970s, which led to fewer studies of colorism during the 1960s and 1970s (Freeman et al. 1966 is a notable exception). Since the early 1990s, however, sociologists have become interested in complexion-based differences in non-white communities. Modern evidence documents the continuing, perhaps growing, significance of colorism.
___________________
3 In 1787 Virginia defined anyone with one or more black grandparents as mulatto, a definition that was reiterated in 1792 and 1819. In 1833 the Virginia legislature conceded that the one fourth definition created a class of people that were neither black nor mulatto nor white. An act defined such people with less than one-fourth black ancestry to petition the local court for certificates of freedom that would establish their legal “whiteness.” It was not until 1866 that the term mulatto was eliminated from the civil code and replaced with the word colored. In 1910 the statutory limit was changed so that anyone with one-sixteenth black ancestry was considered colored. Finally, in 1930 the law was changed so that anyone with any ascertainable amount of black ancestry was colored (Wadlington 1966, pp. 1196, 1201). Mississippi’s state constitution in 1911 classified individuals with one-fourth black ancestry as legally colored (Anonymous 1911). By 1948, the constitution had been amended so that individuals with one-eighth black ancestry were colored (Bynum 1998). Legal definitions in other southern states followed similar trajectories.
_____________________
-6-
Using data from the National Survey of Black Americans conducted in 1980, Hughes and Hertel (1990), Keith and Herring (1991), and Seltzer and Smith (1991) find that dark-skinned blacks acquire less education, work in less prestigious occupations, and earn lower incomes than lighter complected blacks, even after controlling for other factors believed to be correlated with achievement. Using longitudinal analysis, Hill (2000) finds that biracial men enjoyed modestly advantaged backgrounds compared to blacks, but multivariate analysis reveals that family background variables account for only a small portion of complexion-based differences in outcomes. Color is the decisive factor. Light-complected men are more than twice as likely as dark men to find high-prestige employment. Research has also documented skin-tone differences among Latinos, which suggests that colorism characterizes a host of minority communities (Murguia and Telles 1996; Telles and Murguia 1990; Mason 2004; Darity and Boza 2003). Gullickson’s (2003) interpretation that the significance of skin tone differentials are declining stands in sharp contrast to an emergent literature that finds complexion differences still loom large in the life experiences of mixed-race youth (Russell, Wilson and Hall (1992), Root (1996), Daniel (2002), Tizard and Phoenix (2002), Herman (2002), and Doyle (undated)). Despite the attention that colorism has garnered among sociologists, economists have not accounted for intraracial differences in economic outcomes that may result from colorism. One recent exception that uses nationally representative samples of African Americans from 1979 and 1992 finds that the wages of light-complected blacks exceed those of dark-complected blacks by about eight to ten percent (Goldsmith, Hamilton, and Darity 2004). By comparison, their regression decompositions suggest color-based wage differentials (treatment effects) between whites and light-complected blacks of about five percent. Thus, colorism accounts for about twice the income gap between light- and dark-complected blacks as between whites and light
- 7-
complected blacks. Thus, subtle distinctions in skin tone seemingly translate into significant outcome differences, including educational attainment, occupational achievement and earnings, among others. We now turn to a description of the data source that allows us to contribute to this emergent and important discussion.
3. Data
This study uses a large, regionally representative sample of free African Americans residing in the urban South, constructed from information in the original population manuscripts of the 1860 census. In discussing the outcomes of free African Americans we are, by definition, excluding the experience of slaves, who represented the majority of blacks in the antebellum South and held little or no wealth.4 The cities that are included in the sample are listed along with their relative frequencies in the lower part of Table 1’s summary statistics. Williamson (1980) notes that the border states of the Old South formed a sort of mulatto belt, which contained a high proportion of mulattoes relative to blacks. The Lower South cities of Charleston, Mobile, and New Orleans also had substantial populations of gens de couleur libre. The size or proportions of their respective mulatto and black populations did not drive the choice of cities for this study. Rather, cities were selected to provide a representative sample of all southern, urban free African Americans in 1860. Every household headed by an African American (black or mulatto) from these cities was recorded while collecting data, along with an equal number of randomly selected white households. Some non-white households were later dropped because the census failed to report information on one or more of the variables of interest or because the values reported were
_________________________
4 Kenzer (1997) provides a discussion of wealth accumulation among free blacks, slaves and former slaves for North Carolina.
__________________________________
-8-
clearly erroneous. The resulting sample includes information on 15,861 households. Of these, 5,774 households were headed by white males, 2,121 were headed by mulatto males, and 3,697 by black males. The sample of 4,269 female-headed households includes 842 headed by whites, 1,451 headed by mulattoes, and 1,976 headed by blacks. The 1860 census is a particularly valuable source for the study of the economic consequences of colorism because census marshals were instructed to classify individuals as white, black or mulatto. Specifically, marshals were directed to leave the “Color” column blank if the individual was white, to insert the letter “B” if the person was black without admixture, and to insert the letter “M” if the person was a mulatto or of mixed heritage. It is unclear whether marshals inquired into a person’s racial heritage or classified him or her based on physical appearance, but Gross (1998) contends that local residents, including census takers, used locallyaccepted notions of race and admixture to draw conclusions about one’s racial heritage. Most southerners believed the differences between blacks and mulattoes were obvious and involved complexion and physiology, as well as the complexion and physiology of those with whom one associated.5 There is no reason to believe that a census taker’s assignations of color would have deviated from accepted community standards of racial classification. A second concern is whether a person classified as a mulatto in Baltimore would have been similarly classified in New Orleans. Even if there were inter-city differences in classification standard, to the extent
___________________
5 Of tangential concern to us is the possibility (in fact, the likelihood) that the identity of one’s associates also played a part in the census taker’s determination of color. This connection between an individual’s choices and his or her identity is important but outside the scope of this paper. It is an active area of research (see Bodenhorn and Ruebeck (2003) for references), but here we seek to document the effects of colorism, with less concern for questions of its source and its links to individuals’ identity.
________________________
-9-
that they existed at all, they do not vitiate the results presented here. What ultimately determined whether an individual received differential treatment was whether he or she met the local physiological or heritage standards, because that individual interacted with members of the local community. Inclusion of city dummy variables in the regressions reported below also controls, in part, for local differences in the relationship between treatment and racial categorizations. In addition to the valuable information on complexion, federal census takers from the mid-nineteenth century collected and recorded information on a number of demographic and economic variables. Household structure is easily inferred from the manuscripts because the head of each household is listed first, identified by both given name and surname. Only first names are recorded for spouses and children. Members of the extended family were listed next, followed by unrelated members of the household (typically servants or apprentices), both identified by given name and surname. The ages and genders of all household members were recorded, along with their state or nation of birth, and whether the individuals were literate or had attended school in the past year. The census marshals also recorded economic data, including the occupations of household heads. In a few instances, marshals recorded the occupations of the wives of male-headed households, but no use is made of this information because it was inconsistently reported. Finally, census marshals were asked to provide estimates of the dollar value of each household’s real and personal property. The value of household real estate was to be obtained by asking the householder for his or her valuation of the real property owned by the household ignoring any lien or encumbrance on the property. Thus, the reported figure is gross, not net, real estate wealth and does not include the value of rental property. Marshals were also instructed to obtain the householder’s valuation of all other property (personal estate), including financial
-10-
assets, slaves, livestock, jewelry, fixtures and furniture. The instructions recognized that an accurate valuation may not be possible, but marshals were encouraged to obtain as “near and prompt” an estimate as they could. The Census Bureau anticipated the reluctance of many householders to divulge information about their wealth and instructed marshals to cajole and reassure respondents that the information was confidential and would not be used for private gain or any other public (i.e., tax) purposes. It is clear that some marshals were better at cajoling and reassuring householders than others. It was not uncommon for marshals to return a blank (nonresponse) when reporting real and personal estate in the manuscripts. An empty cell in the real estate column is generally taken to represent zero wealth, reflecting that the family rented rather than owned its current habitation, but historians have long debated the meaning of nonresponses for personal wealth, and the debate is far from resolved. Some contend that marshals left the cell blank rather than recording zeroes. Others contend that marshals failed to report modest or hard-to-value property holdings, so that nonresponses likely represent small but nonzero personal wealth. Conley and Galenson (1998) and Bodenhorn (2003) review the debate and the data and conclude that marshals had idiosyncratic, nonzero censoring points for personal wealth below which they returned a blank. Just as there is no consensus concerning the interpretation of unreported personal wealth for a household, there is no consensus among economic historians concerning a preferred empirical technique. Conley and Galenson, and Bodenhorn, estimate quantile or median regressions; others estimate Tobit models. Given the common assertion that nonresponses are typically indicative of small, but nonzero personal property values, we attribute to each nonresponse one-half the minimum value recorded by any marshal in each city ward, and then correct the standard errors for ward-level clustering. We conducted a number of robustness
-11-
checks (available by request): we estimated models that excluded households with unreported wealth; we excluded wards if nonresponses exceeded 10, or 25, or 45 percent of sampled households; we replaced the natural log of wealth as the dependent variable with the inverse hyperbolic sine, among other procedures. In every instance, the basic results reported below hold. We prefer our procedure because estimating the models using Ordinary Least Squares allows us to report the full results of standard Oaxaca-Blinder regression decompositions. Thus, although the wealth data drawn from the 1860 U.S. census are imperfect, they are rich enough to generate meaningful comparisons of wealth accumulation among white, black, and mulatto households. In addition to age and complexion, we employ a number of variables expected to explain differences in household wealth. We label household heads Migrants if they migrated across state lines.6 We also label as Immigrants any head of household born outside the United States. Migrants and immigrants are often thought to be self-selected, highly motivated individuals, suggesting that, all else equal, movers will accumulate more wealth than nonmovers (Borjas 1994). On the other hand, movers may have financed their migration by consuming previously
accumulated wealth, which may imply lower current wealth. There is no reason to expect, a
priori, one effect to dominate. Summary statistics reported in Table 1 show that while a little over 10 percent of all male heads of households migrated across state lines, another 25 percent immigrated from outside the United States. Women were as likely as their male counterparts to have migrated across state lines, but a much smaller proportion of female heads of household were immigrants. Given the restrictions on black in-migration in most southern states, it is not surprising that blacks and mulattoes were less likely than whites to migrate across state lines
_____________________
6 Given the proximity of Baltimore and Washington, D.C., individuals who moved from Washington to Baltimore were not considered interstate migrants.
________________________
-12-
(Guild, 1969). Despite such restrictions, about 7 percent of African-American male household heads and nearly 9 percent of African-American female heads had migrated across state lines. Further, 1.5 percent of African-American male heads of households had emigrated from foreign, mostly Caribbean, countries, as had 2.7 percent of African-American female heads of households. Another set of variables is included to control for the household’s human capital. We include a dummy variable equal to one if the head of the household was literate, a variable equal to the number of literate males 20 years of age or older (including the head of the household), and a variable equal to the number of literate females 20 years of age or older in the household. Table 1 shows that about 82 percent of male heads of household and 74 percent of female heads were literate. Both male- and female-headed households on average had more than one literate female (1.109 and 1.280, respectively), but female-headed households generally had far less than one literate adult male while the mean male-headed household had more than one. Given the absence of publicly funded education for African Americans in the antebellum South, it is not surprising that African American households had fewer literate adults than white households. Indeed, the level of literacy among African Americans is notable given the outright animosity among many whites toward the education of blacks and mulattoes (Woodson 1919). Two additional variables are constructed to capture possible education or training effects. The first (SEI) is a Duncan-style socioeconomic index (Reiss 1961), intended to indicate occupational prestige. It is based on wages and educational levels associated with several hundred occupations in the 1950s. For occupations recorded in the 1860 census with the same title (e.g., barber, blacksmith, carpenter, bricklayer, etc.) the occupational prestige index value was taken directly from the 1950s index. For some occupations recorded in the 1860 census,
-13-
assigning a corresponding Duncan occupational prestige score required some ingenuity. Duncan‘s index has no entry for carriage drivers (ignoring only those few still found in Central Park and other tourist locations), but fortunately Duncan did list a prestige score for the modern counterpart of the carriage driver, namely the taxi driver. Similarly, modern bus drivers are roughly the equivalent of stagecoach drivers. Because Duncan’s classification includes several hundred occupations—from bootblacks (SEI = 8) and charwoman (SEI = 10) to lawyers (93) and dentists (96)—and nearly every major industrial category, it is possible to assign a reasonable occupational prestige value to every occupation listed in the 1860 census. 7 When a specific occupation listed in the 1860 census had no reasonable corresponding Duncan value, that occupation was given Duncan’s general prestige score for job class (foreman, operative, or laborer) and industry (textiles, metals, machinery, services, etc.). Finally, just as marshals failed to report wealth holdings for some households, they also failed to report occupations for every household head. These observations are also not deleted from the sample. We attribute to each of these households the average occupational prestige value by the gender of the household head (15 for males, 14 for females), and we include a dummy variable to indicate these observations. The dummy variable should capture any fundamental differences between reporting and nonreporting households, but we can see from Table 1 that there may be much less concern for this issue in the men (6% of the male sample) than females (46% of the female sample, with representation decreasing significantly from white to mulatto to black women). It is likely that the labor supply decisions of women and men could be very different, as they often can be today, and as a result the determinants of wealth accumulation may have been quite different across gender as well. This leads us, as is the usual practice, to present separate results for male- and female-headed households. Historians and labor economists have noted
________________________
7 A detailed appendix of occupational classifications is available from the authors on request.
___________________________
-14-
differential economic and occupational opportunities for women, and have discussed the negative consequences of single parenthood (Moehling 2004). We also report our full model regressions both with and without the occupational variables because occupations may be endogenous: to the extent that family wealth persists across generations, greater wealth in previous generations may imply both greater wealth and a more prestigious occupation for the current head of household. Because we cannot account for intergenerational transfers and because we lack any reasonable instruments to perform two-stage least squares, we ask the reader to interpret those regressions including the occupational variables with care. Before moving on to the regressions, we call attention to the substantial differences in means across same-gender subpopulations (comparing among white, mulatto, and black as well as comparing white to African-Americans as a whole). Nearly all these differences were highly statistically significant (see the comments below Table 1 for exceptions). Note especially the significant differences in characteristics between blacks and mulattoes. In every category mulatto-headed households were better endowed than their black counterparts, but mulattoes were still not as well endowed as heads of white households.
4. The determinants of household wealth Our empirical strategy in this section is to estimate household-level wealth regressions that include several explanatory economic and demographic characteristics as well as the complexion of the household head, Black and Mulatto (omitting White). The parameter vectors b and g in the equation
(1) ln(wj ) = b 0 + b 1 X j + b 2 Yj + b 3 Z j + g 1 Black j + g2Mulattoj + e j
-15-
are estimated using ordinary least squares regression. The dependent variable ln(wj ) is the natural logarithm of the jth household’s wealth (following the usual semi-log regression form due to the skewed nature of the distribution of wealth in levels), Xj is a vector of individual characteristics of the household head (age and its square, immigrant and interstate migrant status, literacy, and occupational prestige score), Yj is a vector of household characteristics (the number of literate males, the number of literate females, and the size of the household), and Zj is a vector of city dummy variables (excluding Baltimore) to capture any systematic regional differences in the ability of households to accumulate wealth. e j is the error term associated with the jth household. We also estimated the equation
(2) ln(wj ) = b 0 + b 1 X j + b 2 Yj + b 3 Z j + gAfAmj + e j ,
where both subpopulations with African-American heritage are grouped into the same category, AfAm. We only present the value of this shift variable’s coefficient in our results because the other parameters did not differ substantively from those estimated using (1). Table 2 reports the regressions. Probability weights are applied to account for the sampling method described in the Section 3.8 The columns labeled (3) and (4) are our preferred specifications because they offer the largest number of exogenous controls. A mentioned above, we include regression (4) to account for occupational differences but recognize that occupations may be endogenous to wealth. Before turning to the issue of colorism, we summarize and interpret the other individual and household-level correlates and the overall performance of the regressions.
__________________________
8 These weights take into account the undersampling of white-headed households. We took the assertion that the census included all free households at face value and used the probability weights to adjust for the relative probability of sampling from each household subpopulation.
______________________________
-16-
Household wealth for both male- and female-headed households increased with age at a
decreasing rate. The maximum for men occurs at about 65 years, an age which little more than five percent of our sample exceeded, although a handful of ages in the 100’s were recorded for both genders. Households headed by male immigrants had about half (e-0.56 = 0.57) of the wealth of households headed by native-born men, indicating that the benefits of initiative were outweighed by the costs of moving (at least for the current generation), although the results in column (4) indicate that immigrants’ advantage may be explained by their occupations. The results in columns (1)-(3) are consistent with the economic studies of immigration that show that firstgeneration immigrants are slow to assimilate often because immigrants tend to congregate in ethnic enclaves (Borjas 1994; Borjas 2000). We have no evidence on the ethnic or racial compositions of the neighborhoods in which southern urban African Americans resided, but relying on crude data at the level of census wards Curry (1981) found that blacks were considerably less likely to be crowded into proto-ghettoes in southern than in northern cities in the antebellum era. It may also be the case that these free blacks’ forebears had not passed significant wealth on to them. Female-headed household wealth did not have a significant relationship with immigrant status. Unlike international immigration, internal migration has no statistically or economically meaningful effect on household wealth once the city controls are added. Household size was significantly related to household wealth. The addition of one person to a male-headed household was associated with more than a 10 percent increase in male-headed household wealth, while the increase was somewhat smaller in a female-headed household. The results are consistent with the hypothesis that larger extended households, which were relatively
-17-
common in 1860, brought together larger agglomerations of wealth than smaller nuclear households. Literacy, as expected, had an economically meaningful association with household wealth. Households in which the male head was literate had more than double (e0.83 = 2.3) the wealth of households with an illiterate male head (about 50% more if we control for occupational status). There was little relationship between wealth and female heads’ literacy. Interestingly, the addition of a literate female was associated with an increase in male-headed household wealth of about 20 percent, and an increase in female-headed household wealth by about 25 percent, but the addition of a literate adult male did not add significantly to household wealth. We do not have an explanation for this phenomenon. Turning now to the issue of colorism, we consider the estimated coefficients on the Black and Mulatto variables, as well as that on the AfAm variable in the incompletely reported companion regressions. (See the first row of coefficients in Table 2 for the latter, and the second and third rows for the former pair of variables.) It is immediately apparent that standard specifications of a single race identifier do not fully capture the subtleties of race and racial wealth differences. Simply comparing the means of the white, mulatto, and black subpopulations implies that a household headed by an African-American male had just 30 to 50 percent (e-1.26 to e-0.76)—or less—of the wealth of the average household headed by a white male. But when we separate African-American households into those with mulatto and black heads of household, there are significant and meaningful differences between them. Households headed by a black male had on average less than 40 percent (e-0.95)—probably less than 25 percent (e- 1.46)—of the wealth of households headed by white males, while the mean household with male mulatto head had at least 40 percent – and perhaps as much as 63 percent (e-0.46) -- of white male
-18-
wealth. Even more distinct bounds can be placed on female-headed households’ wealth differentials by racial ancestry. The fact that the separate male and female regressions generate such similar results increases our confidence that we are identifying a real phenomenon. Further evidence is provided in the following section by estimates of separate wealth equations for each subpopulation where we decompose the differences into characteristics and treatment effects. We recognize that, if colorism was as powerful a social force as we contend, ancestry and skin shade may have influenced wealth accumulation because lighter-complected individuals received preferential treatment at younger ages that created occupational opportunities not available to those with darker complexions. Dark individuals may have been discriminated against through the apprenticeship system or through other types of occupational training. Columns (4) in Table 2 address this concern by including the Occupational Prestige (SEI) and No Occupation variables. As we recognize above, the coefficients on the colorism variables are smaller than in columns (3), but the colorism effect does not disappear. It remains a powerful force even after controlling for occupation. In every specification, we reject the null hypothesis of Black and Mulatto coefficient equality at confidence level p0.001. The shortcoming of the analysis in Table 2 is that it imposes equality restrictions on the non-race (non-colorism) coefficients across the subpopulations. Thus the analysis above ignores the possibility that, in controlling for the household’s characteristics, we have missed the differing return to characteristics across subpopulations. In the next section, we report the results of standard Oaxaca-Blinder regression decompositions to investigate the extent to which the colorism effect was attributable to differences in characteristics across groups versus arising from the treatment afforded those of different complexions.
-19-
Before moving on to the decompositions, we should also note that the regressions’
adjusted-R2 values indicate that there is much variation in household wealth remaining to be
explained. Because the decomposition techniques we use below rely on an analysis of residuals, the results must still be considered only imperfect insights into the treatment effects that follow from colorism.
5. Did light-complected households receive preferential treatment? Evidence from regression decompositions Wealth differences between any two groups may be due to differences in the average features of that group’s members (characteristic effects), or to systematic differences in how each group is rewarded for those features by the marketplace (treatment effects). The premise of this article is that colorism influences economic outcomes, generating greater household wealth in lightcomplected households for a given set of observable characteristics. In order to better understand the contributions of these two effects of colorism, we use the following methodology. For each gender, separate coefficients are estimated for the white, mulatto, and black subpopulations. Because there are some concerns that the occupational prestige variable may be endogenous, regressions and decompositions are reported with and without the occupational variables, as in columns 3 and 4 of Table 2. We omit the regression results themselves to conserve space and because they do not meaningfully differ from those in Table 2. We then use these regressions to further investigate the differences between subpopulations’ wealth using the decomposition method of Oaxaca (1973) and Blinder (1973), in addition to Cotton’s (1988) modified form.
-20-
The goal of the decomposition is to consider the counterfactual implied by the assertion that different groups have different wealth levels, due perhaps to colorism: what wealth would
the members of each group expect to have in the absence of their identification as a member of their group? Although we will construct a decomposition for all four pairs of sub-populations, consider for concreteness the pair of subpopulations whose wealth differences may illustrate colorism: mulattoes and blacks. Label the associated pair of regressions lnwMulatto = aMulatto + XMulatto bMulatto + e and lnwBlack = aBlack + XBlack bBlack+ e. To decompose the advantages accrued by those with lighter skin, we consider the difference in predicted log wealth evaluated at the explanatory variables’ means (indicated by “ ” notation) and parameter estimates (indicated by “ ˆ ” notation)
(3) lnwMulatto - lnwBlack = ˆ a Mulatto - ˆ a Black + X Black ( ˆ b Mulatto - ˆ b Black ) [ ]+ (X Mulatto - X Black ) ˆ b Mulatto [ ] ,
where the first bracketed term is the unexplained, or treatment, effect and is presumed to be due, at least in part, to colorism. The second term is the characteristic effect, the effect on mean log wealth due to differences between the two groups. With the structure of (3) we are taking the perspective that in the absence of colorism the advantaged group’s wealth structure would prevail. The opposite assumption is to presume that all groups would receive the disadvantaged group’s wealth structure in the absence of colorism, or
(4) lnwMulatto - lnwBlack = ˆ a Mulatto - ˆ a Black + X Mulatto ( ˆ b Mulatto - ˆ b Black ) [ ]+ (X Mulatto - X Black ) ˆ b Black [ ] .
These two decompositions provide a range within which we expect the actual effect of colorism to lie. Many alternative decompositions have been proposed to predict a particular value inside this range. Because we feel, along with other authors, that the counterfactual wealth is likely to lie closer to the larger group’s prevailing wealth, we calculate Cotton’s (1988) decomposition as well. That is,
-21-
(5) lnwMulatto - lnwBlack = ˆ a Mulatto - a*+X Mulatto ( ˆ b Mulatto - b*) [ ]+ a*- ˆ a Black + X Black ( b*- ˆ b Black ) [ ] + (X Mulatto - X Black ) b* [ ]
The vector of parameters a*, b* ( ) generalizes the previous two cases: a*, b* ( )= aLight , bLight ( ) in (3), while a*, b* ( )= aDark, bDark ( ) in (4). The parameter vector that we actually use for a*, b* ( ) is a weighted sum of the previous parameters, where the weights are the relative frequencies of the two populations9. This decomposition has the added feature that it apportions the treatment effect into components that indicate the gain of the advantaged subpopulation and the loss of the disadvantaged population (respectively the first and second bracketed terms above), but we do not report these separately because our focus is instead on the total treatment effect, the addition of those first two bracketed terms. It is also worth noting that, because the weights are the subpopulations’ relative frequencies, when comparing the treatment of people of color (blacks, mulattoes, or African-Americans as a whole) to the white subpopulation this decomposition weights the advantaged population (the majority whites) more heavily, but when considering the existence of colorism it is the disadvantaged population (the majority blacks) that receives the higher weight.
The two panels of Table 3 first provide estimates of the subpopulation pairs’ relative
predicted wealth and the resulting raw wealth differentials, as well as the subpopulations’
predicted geometric mean wealth (the exponential of the arithmetic log mean). Note that our
discussion of the raw wealth ratios compared across subgroups from the previous section’s
results with pooled regression is similar to the results presented here from separate regressions. Households headed by black men hold on average 13% of white wealth, while mulatto-headed
__________________
9 Due to our sampling method, the relative frequencies are actually calculated as the sum of each subpopulation’s probability weights.
_____________________________
-22-
households’ average wealth is 35% of white-headed households. The results for female-headed households are quite similar to those for male-headed households. Next in Table 3’s two panels, the raw differential is decomposed into two parts using the method described above and estimation of the model from either column 3 (no occupation variables) or column 4 (with occupation variables) of Table 2. The raw differential is decomposed into both the treatment and characteristic effects, but to save space Table 3 presents only the treatment effect. This effect is the (imperfect) estimate of discrimination or colorism. As well, the characteristic effect (to which we return later in the table) need not be listed explicitly because it is simply the remainder of the raw differential. Our discussion focuses on the third set of treatment effect estimates, those typeset in boldface and listing the characteristic effect as C = b* (XH - - XL ), using Cotton’s (1988) decomposition. For example, on the right side of Table 3 where the results of pairing the white and African-American subpopulations are presented, we see that for male-headed households the treatment effect is estimated at 69% of the raw differential with model (3) and 42% with model (4). Thus the characteristic effect is 31% and 58%, respectively, for the two models’ application to this pair of subpopulations. These percentages are the focus of this portion of the table: how much of the raw wealth differential is explained by treatment effects?
For the moment, compare the boldface treatment effects to those presented in the rows
above them for each gender. We see—as should have been expected—that Cotton’s estimate is always between those of the traditional Oaxaca and Blinder decompositions. Its counterfactual assumption is a weighting of the two groups’ returns to characteristics, while the Blinder-Oaxaca decompositions take two possible extreme counterfactual assumptions: that either the advantaged
-23-
or the disadvantaged group’s returns to characteristics would obtain in the absence of differential treatment across groups. Our first evidence of colorism in Table 3 is simply to note the large estimate of the treatment effect in the first column’s decomposition of the Mulatto and Black subpopulation pair of regressions. This treatment effect is about 40 percent of the raw wealth differential, with little difference dependent on including occupational status. Considering all sub-population pairs, the differences between models 3 and 4 illustrate another result we should expect: controlling for occupation dampens the treatment effect; more of the raw wealth differential can be explained by including this characteristic of the household head to in the regressions. Duncan’s SEI is one way in which all subpopulations are different from each other, it has a positive relationship with wealth, and the more advantaged subpopulations have higher SEI. On the other hand the larger white-mulatto treatment effect (85%) as compared to the white-black treatment effect (66%), without controlling for occupation (model 3), can be explained by mulattoes’ characteristics having greater similarity to whites’ characteristics than blacks’ characteristics have to whites’ characteristics. Even with a higher return to mulattoes’ characteristics as compared to blacks’, more of the raw wealth differential is left unexplained by any difference in mulattoes’ characteristics as compared to whites precisely because their
characteristics are less different. The resulting smaller characteristics effect leads to a larger treatment effect. Once occupation is included (model 4), the difference in treatment effects is small (41% and 43%, for males). The treatment effect for the mulatto-black pair of
subpopulations shows little change whether we include occupation or not (38% versus 42%).
The final portion of each of Table 3’s panels then uses the boldface estimate of the
treatment effect (Cotton’s method) to estimate two more comparisons of the results for the
-24-
subpopulation pairs. First is the estimated return to characteristics, the wealth ratio. It is the ratio of the advantaged group’s wealth to the disadvantaged group’s wealth. The values are all greater than 1, reflecting the inherent advantage in characteristics that the lighter-skinned group has, regardless of differential treatment. Loury (1998) focuses on this as an issue that remains even after dealing with discriminatory practices: the fact that a subpopulation has a disadvantage in characteristics should still be of concern even if there is no treatment effect due to discrimination. First note that including another explanatory variable (in particular, occupation) in the regression will always increase the return to characteristics. These comparisons confirm our earlier observation that whites had a much greater advantage in characteristics over blacks (a factor of 2 to 3) than they did over mulattoes (always less than 2). Likewise, the advantage of mulattoes in characteristics as compared to blacks (remarkably similar for males and females) is about a factor of 1.8. The final comparison in each panel is Oaxaca’s (1973) adaptation of Becker’s (1971) discrimination coefficient, the relative benefit to the advantaged group from discrimination, assuming that the entire treatment residual can be attributed to discrimination (as measured by Cotton’s residual). It is measured as the percentage gain in the wealth ratio as compared to what it would be without differential treatment. Similar patterns are seen in thesecomparisons as those above. The gains from treatment are large, and the differences within African-Americans are also large. So we see that, although differences in underlying characteristics were a driving force behind the mulatto-black wealth gap, treatment disadvantages were substantial. The magnitude of the treatment disadvantage estimates supports the colorism hypothesis, and reflecting on the characteristic effect further strengthens that hypothesis. Further research needs to uncover the extent to which pre-market discrimination led to differences in observable wealth-generating
-25-
characteristics. It is well known that light-complected slaves were more likely to receive skill training (Margo 1992), for example. It may have been that mixed-race free African Americans, too, received more or better education and job skills than monoracial blacks. Uncovering whether and why this conjecture was true represents a potentially fruitful line of further inquiry.
6. Concluding Remarks Does black always signify Black? In matters of race economists have assumed it does. Nearly every economic investigation into the consequences of race uncritically and without pause separates Whites and Blacks into distinct and mutually exclusive categories. For the purposes of empirical work done during the past 50 years such a dichotomous racial classification scheme was probably appropriate, an intellectual and academic recognition of a broad social acceptance of Jim Crow’s “one-drop” rule. In the future, however, dichotomous racial categories may not capture the increasing subtleties of race as race mixing spreads and becomes increasingly socially acceptable. Empirical economics cannot continue to model race as a binary variable and expect the resulting coefficient to meaningfully capture the subtleties of race. Our paper shows that mulattoes were treated differently than blacks in the mid-nineteenth century. As a percentage of white households’ log wealth, the predicted log wealth of blackheaded households was 20 percentage points lower than that of mulatto-headed households. Regression decompositions reveal that treatment effects played a large role in explaining the white-mulatto wealth gap, even after including a measure of occupation; treatment effects played a comparable role for other subpopulation pairs as well. Our results are consistent with those from a sociological literature dating to Reuter (1917) and an emergent economics literature that finds that colorism remains a powerful social phenomenon in minority communities. An
-26-
important line of further inquiry will be to understand how color-based identities are established and reinforced in historical and modern populations. We intend to contribute to that developing literature as well. |
Misty
"Cyniquian" Level Poster
Username: Misty
Post Number: 144
Registered: 02-2006
Rating: 
Votes: 1 (Vote!) | | Posted on Tuesday, July 04, 2006 - 08:11 pm: |
|
LOL...tonya i don't lknow how much more specific you could have been in the first post....there are some people who just dont WANT to get what you're saying. As it's been reported time and time again, there's a huge difference in the way that lighter skinned blacks and darker blacks are raised, and treated. the difference in income says it all. Naturally kids raised in families with higher income are treated differently and raised differently rather they're black or white. When you factor that in with skin tone this becomes even more obvious.
The post simply brings home what you've been saying all along....those with lighter skin recieve better treatment and are seen in a more favorable light by society...DUUUUHHH...LOL |
Ntfs_encryption
"Cyniquian" Level Poster
Username: Ntfs_encryption
Post Number: 318
Registered: 10-2005
Rating: N/A
Votes: 0 (Vote!) | | Posted on Wednesday, July 05, 2006 - 05:23 am: |
|
"Is there a point to the documentation of racism Ntfs?"
Of course! If it is well documented and researched,cleary written (minus bloated and un-necessary abstractions) and interesting to read.
|
Tonya
"Cyniquian" Level Poster
Username: Tonya
Post Number: 2885
Registered: 07-2005
Rating: N/A
Votes: 0 (Vote!) | | Posted on Wednesday, July 05, 2006 - 12:05 pm: |
|
Ntfs_encryption, how could it have been better researched and documented? What should they have done? Point out the changes you would have made? Be specific about how you would have done them. And explain why you expect an actual raw, uncut, scientific study result to be entertaining. |
Serenasailor
"Cyniquian" Level Poster
Username: Serenasailor
Post Number: 521
Registered: 01-2006
Rating: N/A
Votes: 0 (Vote!) | | Posted on Wednesday, July 05, 2006 - 05:12 pm: |
|
The truth was that there was a strict mullatto elite in those days. The mullatto's hated darker skinned blacks and only intermarried with one another. They set up there own communities.
The slave master breed a new breed of slaves. These slave were like Shemika said in one of here post "mentally and physically" indistinguishable from there slave masters. Just like the offspring of these self-hating BM and WW. You are breeding a new form of semi-whites who hate Black ppl. |
Nels
"Cyniquian" Level Poster
Username: Nels
Post Number: 435
Registered: 07-2005
Rating: N/A
Votes: 0 (Vote!) | | Posted on Wednesday, July 05, 2006 - 11:28 pm: |
|
Serenasailor --
"You are breeding a new form of semi-whites who hate Black ppl."
No, they don't hate them. However, if they were to read your diatribes, they might change their minds. |
Ntfs_encryption
"Cyniquian" Level Poster
Username: Ntfs_encryption
Post Number: 325
Registered: 10-2005
Rating: N/A
Votes: 0 (Vote!) | | Posted on Friday, July 07, 2006 - 05:25 am: |
|
"No, they don't hate them. However, if they were to read your diatribes, they might change their minds."
Thank you. You are so right. |
|
Email This Page