Slavery, Education, and Inequality*
Graziella Bertocchi**
Arcangelo Dimico***
April 2014
ABSTRACT
We investigate the effect of slavery on the current level of income inequality across US counties. We
find that a larger proportion of slaves over population in 1860 persistently increases inequality, and in
particular inequality across races. We also show that a crucial channel of transmission from slavery to
racial inequality is human capital accumulation, i.e., current inequality is primarily influenced by
slavery through the unequal educational attainment of blacks and whites. Finally, we provide
suggestive evidence that the underlying links run through the political exclusion of former slaves and
the resulting negative influence on the local provision of education.
JEL Codes: E02, D02, H52, J15, O11.
Key words: Slavery, inequality, education, institutions.
* We would like to thank two anonymous referees, Daron Acemoglu, Costas Azariadis, Roland Benabou, Elena
Esposito, Raquel Fernandez, Chiara Gigliarano, Oded Galor, Nippe Lagerlöf, Nathan Nunn, Elias Papaioannou,
Dietrich Vollrath, as well as participants at the NBER Summer Institute Workshop on Income Distribution and
Macroeconomics, the IV Workshop of the Social Choice Research Group, the PRIN Workshop on Political
Institutions and Demographic Dynamics, the Rimini Conference in Economics and Finance, the IX Brucchi
Luchino Labour Economics Workshop, the CEPR ESSIM, the IEA, and seminars at the Universities of
Nottingham, Glasgow, Göteborg, IMT Lucca, National University of Singapore, and Sheffield, for comments
and suggestions. Generous financial support from the Italian University Ministry and Fondazione Cassa
Risparmio di Modena is gratefully acknowledged.
** Affiliations: University of Modena and Reggio Emilia, CEPR, CHILD e IZA - Address: Dipartimento di
Economia Marco Biagi, Viale Berengario, 51, I-41121 Modena, Italy - Phone: 39 059 2056856 E-mail: [email protected] - Web: http://www.economia.unimore.it/bertocchi_graziella.
*** Affiliation: Queen's University Belfast - Address: Management School, Riddel Hall 185 Sranmillis Rd, BT9
5EE, Belfast, UK - Phone: +44(0)2890974167 - E-mail: [email protected].
1
1. Introduction
Recent advances in growth theory have debated the long-run determinants of current comparative
economic performances. In this paper we address the same issue by investigating the legacy of slavery
in the US. Our goal is to assess if, nearly a century and a half after its abolition, this peculiar institution
still exerts an influence on the American economy and society. We are especially interested on its
potential effect on current inequality. To concentrate on a single country facilitates the empirical
investigation on several grounds, since it reduces the risk of omitted variable bias that typically plagues
cross-country investigations. At the same time, because of their size and history, the US still presents
sufficient variations along both the geographic and the institutional dimensions to make such
investigation worthwhile.
Slavery was introduced in North America as early as in the sixteenth century and its diffusion escalated
throughout the next centuries. Overall, the Middle Passage brought an estimated 645,000 slaves, mostly
from Africa, to the territories that today represent the US. Initially most of the slaves were forcibly
settled in the coastal Southern colonies, where they were employed primarily in agriculture. Later,
between the American Revolution and the Civil War, with the Second Middle Passage around a million
slaves were relocated toward the inland regions where the plantation economy was developing (Berlin,
2003). By the 1860 Census the US slave population had grown to four million, to represent about 13%
of the entire population, distributed within 15 slave states, mostly in the South. In the same year, almost
90% of the blacks living in the US were slaves. The American Civil War led to the abolition of slavery
in 1865.
We measure the historical legacy of slavery with the share of slaves over population in 1860 and we
find that the current level of income inequality is indeed shaped by it, since the distribution of per
capita income is more unequal today in counties associated in the past with a larger proportion of
slaves in the population. Moreover, we show that past slavery affects the racial component of
inequality, while it is not associated with higher inequality within each race.
To investigate the mechanism through which slavery determines current racial inequality, we compare
two alternative channels: human capital transmission and racial discrimination. To be noticed is that, at
least in principle, these two channels are not mutually exclusive since, as suggested by Sokoloff and
2
Engerman (2000) and Acemoglu et al. (2008), institutional and economic development paths may be
interlinked and jointly determined by various factors. For instance, the institution of public schooling,
which has been a major vehicle of capital accumulation, may have been more rapid and more effective
in the same counties which were less exposed to de facto and de jure discrimination.
In evaluating the human capital transmission mechanism, we test the hypothesis that the long-term
influence of slavery may run through its negative impact on the educational attainment of blacks
relative to those of whites. According to this hypothesis, which is closely associated with Smith (1984)
and the literature on race and human capital, the counties more affected by slavery should be associated
with worse educational attainment for the black population. For a cross section of counties, our
empirical investigation does support this hypothesis.
The second channel of transmission we test is motivated by those racial discrimination theories which
have emphasized racial differences in the value of skills. Racial discrimination can manifest itself on
the schooling dimension, through worse quantity and quality of the education publicly provided, or
directly on the labor market, by denying blacks access to certain jobs (see Smith, 1984). We find that
the evidence supporting the relevance of this second potential channel of transmission is much weaker
than that collected for the first channel.
Our empirical findings are confirmed after robustness checks involving other potential confounding
factors such as factor endowments, IV estimation, and quasi-experimental evidence.
We complete our investigation with a series of reduced forms illustrated by scatter plots which further
explore the underlying mechanisms linking slavery to local politics, and in turn to education policies
and the resulting racial gap in education. The resulting evidence, together with our previous results,
supports the following theory. Slave counties have been persistently characterized by restricted political
participation achieved both through de jure and de facto methods. The political exclusion of former
slaves resulted in a limited supply of education since, being education locally administered and
financed, political exclusion implied undertaxation and reduced funding.
Our empirical implementation of the concept of slavery involves the fraction of slaves over population
in 1860, which reflects the intensity of the diffusion of this specific form of work organization together
3
with its influence on local institutions. The long-term legacy of slavery can also run through
institutional differences, such as national or state-level institutions aimed at protecting property rights
in human labor and the associated balance of political power, which may be influential even the
presence of small fractions of slaves even though should be eliminated by state fixed effects. Despite
controlling for de-jure institutional differences using state-dummies, we do find the same effect on
racial inequality even when we control for the slave share measured in 1800, at a time when small
fractions of slaves were still present in some Northern states. We interpret these results as supportive of
our empirical strategy.
The rest of the paper is organized as follows. In Section 2 we discuss the relevant literature. In Section
3 we examine the effect of slavery on inequality. In Section 4 we explore the channels through which
this effect materializes. Section 5 is devoted to a quasi-experiment and Section 6 to the reduced-form
evidence. In Section 7 we derive our conclusions.
2. Related Literature
The historiography of slavery in North America is huge. Economic historians have focused on the
profitability and the efficiency of slavery. In their provocative and controversial empirical work on the
antebellum South, Fogel and Engerman (1974) suggested that slavery was both productive and
economically efficient, a conclusion which was criticized, among others, by David et al. (1976) and
Ransom and Sutch (2001). Lagerlöf (2009) and Acemoglu and Wolitzky (2011) model the economics
of labor coercion from a related perspective.
Since our goal is to establish whether slavery can be viewed as a deep determinant of long-run
development, we contribute to the research line, initiated by Hall and Jones (1999), which has searched
for fundamental, rather than proximate, growth factors. Other authors have already focused on the
long-term legacy of slavery, in a number of dimensions. While Nunn (2008a) has examined the
implications of slave trades in Africa, Engerman and Sokoloff (2005a) and Nunn (2008b) have looked
at the impact of slavery in the receiving countries. In particular, on the basis of historical evidence, the
former formulate the hypothesis that factor endowments, through large-plantation slavery and other
inequality-perpetuating institutions, may have hampered subsequent economic growth. The latter
estimates the influence of slavery on the current performances of the US economy at the county level,
4
to find that slave use is negatively correlated with subsequent economic development, but that this
relationship is not driven by large-scale plantation slavery, i.e., a more precise measure of factor
endowments. He also finds a positive impact of slavery on 1860 land inequality, which is in turn
correlated with current income inequality, but no impact of 1860 land inequality on current income,
which suggests that inequality may not be the channel of influence running from factor endowments to
the current level of development. Mitchener and McLean (2003) find that the legacy of slavery has a
strong and persistent effect on productivity levels, measured as income per worker, across US states in
the 1880-1980 period. Lagerlöf (2005) explores the link between geography and slavery and also
documents a negative relationship between slavery and current income at the county level for a sample
of former slave states. A common conclusion for this stream of the literature is a negative relationship
between past slavery and current income per capita across US states and counties, even though
Bertocchi and Dimico (2010) find that this relationship is not robust after controlling for structural
differences among US states and regions. More generally, economic historians have documented how,
despite a decline in the immediate aftermath of the Civil War,1 from the mid‐1870s per capita income in
the South started to converge to that of the rest of the country, with an acceleration after WW2. The
abolition of the Jim Crow Laws in 1964-5 may have contributed to the process by improving blacks'
schooling, skills, occupations, employment rate, and therefore possibly productivity (see Margo, 1990).
Indeed Donohue and Heckman (1991) find that governmental anti-discrimination efforts, starting with
the 1964 Civil Rights Act, had a substantial impact on the economic status of blacks.2
A separate research line, which is more relevant to our goal, has focused on the impact of race on
inequality. This work has documented that, since emancipation and especially since 1940, the average
income of black Americans has increased greatly, both in absolute and relative terms. The determinants
1
See for example Fogel and Engerman (1974) and Ransom and Sutch (2001). Irwin (1994) shows that
the decline in labor productivity in the South was positively correlated with the extent of slavery before
the Civil War. Evidence of a catch-up between North and South is also presented by Caselli and
Coleman (2001).
2
Acemoglu and Robinson (2008) suggest that, at least initially, the impact of the abolition of slavery on
the Southern economy was limited by the fact that, after the Civil War, the landed elites managed to
maintain economic institutions based on low-skilled, repressed labor through the exercise of de facto
political power.
5
of the relative improvement of the economic status of blacks after WWII, however, have been the
subject of debate. Both the civil rights movement, with its impact on the labor market through
affirmative action laws, and long-term changes in human capital have been advanced as possible
explanations of the observed trend (Heckman, 1990 and Margo, 1990). The main contributions to the
line of research on race and human capital are Smith (1984), Smith and Welch (1989), followed by
Margo (1990) and Collins and Margo (2006).3 The evidence collected by these authors documents the
evolution of racial differences both in the quality and the quantity of education. After the Civil War,
African-Americans had essentially no exposure to formal schooling, as a legacy of the extremely high
rates of illiteracy that existed under slavery. The first generations of former slaves were able to
complete far fewer years of schooling, on average, than whites and they had access to racially
segregated public schools, mostly in the South, where they received a qualitatively inferior education,
even if compared to that received by Southern whites.4 Initially the combination of low educational
attainment and inferior educational quality determined the persistence of large wage and income gaps.
Subsequently, however, the racial schooling gap declined, as successive generations of black children
received more and better schooling, with an eventual impact on earnings. Overall, despite the initial
conditions and the persistence of discrimination, the reported evidence on the evolution of educational
differences, in a wide number of dimensions (such as literacy rates, years of educational attainment,
spending per pupil, and returns to literacy), overwhelmingly points to long-term convergence.5
3. Slavery and Inequality
According to the Engerman and Sokoloff hypothesis, the initial presence of specific factor endowments
explains the development of agricultural production techniques based on slave labor, which in turn
resulted in extreme economic inequality and in a set of political (Engerman and Sokoloff, 2005b),
3
See also Goldin and Margo (1992), Goldin (1998), and Goldin and Katz (1999).
4
Naidu (2010) estimates the effect of the nineteenth century disenfranchisement laws for blacks in the
South and finds that they are associated with a fall in black educational inputs and thus with lowquality Southern schooling. Bertocchi and Dimico (2012b) find that black disenfranchisement
negatively affects black education over a sample of Mississippi counties.
5
A related stream of the literature has measured the long-term influence of family background. See for
example Cameron and Heckman (2001) and Sacerdote (2005).
6
redistributive (Sokoloff and Zolt, 2007), and educational (Mariscal and Sokoloff, 2000) institutions that
reflected this inequality. The link between factor endowments and inequality is also empirically
documented. Galor et al. (2009) find evidence that land inequality - which can proxy for factor
endowments - adversely affected the emergence of human capital promoting institutions, as measured
by educational expenditure across US states in the 1900-1940 period. Vollrath (2013) documents a
negative effect of inequality on property tax revenues in 1890. Ramcharan (2010) tests the relationship
between land inequality and redistribution and uncovers significant effect of land inequality on
redistributive policies in the 1890-1930 period. Over a cross section of slave counties, Lagerlöf (2005)
finds that counties which in 1850 had a larger slave population display higher racial inequality today.
However, the link between slavery and current economic and racial inequality still remains unclear. The
channel through which this link may have worked is also poorly understood.
Given this state of the art, we investigate the long-run effects of slavery on inequality using a sample of
US counties (at 1860 boundaries) distributed across 42 states, including 15 slave states. The share of
slaves over population in 1860 varies considerably, both across and within states.6 It is highest in South
Carolina and Louisiana (at about 55 percent, respectively) and lowest in Delaware and Missouri (at
about 2 and 8 percent, respectively).
In order to measure inequality, we use a Theil decomposition to decompose the level of income
inequality at a county level into two components: inequality across races (between inequality, or racial
inequality) and inequality within races (within inequality).7 Theoretically one should expect a positive
effect of slavery on overall inequality, running through inequality across races, since blacks are likely
to be confined at the bottom of the income distribution. On the other hand the effect of slavery on
within inequality is a priori less clear-cut. While it is likely that there is no effect on income distribution
among blacks (for the same reason as above), the effect of slavery on inequality among whites is likely
to be more complex. One can expect a positive effect if slavery has contributed to establish a small
white elite (former slaveholders) who has preserved its wealth and power over time. On the other hand
the effect is likely to be insignificant if political and economic structural changes have washed out such
6
See Table A1 in the Table Appendix and the Data Appendix for data sources.
7
The decomposition is obtained using the Stata plug in program INEQDECO by Jenkins (1999).
7
privileges. As a result the overall effect is potentially ambiguous.8
Following Acemoglu et al. (2002) and Nunn (2008b) we also control for population density in 1860 as
a proxy for initial prosperity. To be noticed, however, is that our goal is to investigate whether slavery
can be viewed as a fundamental, deep determinant of inequality over the long run. Thus, we employ the
most parsimonious possible specification, where proximate factors, such as migration, etc. are omitted
(see Hall and Jones, 1999).9
In Table 1 we test the relationship between slavery and the current level of overall, between and within
income inequality. All specifications include a full set of state dummies. In Model 1 we find that
slavery has a positive and significant effect on overall inequality. The effect is economically significant,
with an increase in inequality of 0.045 per a one percent increase in the share of slaves. In Model 2 we
look at inequality between races: the coefficient on slavery is almost unchanged and significant at a 1
percent level. Finally in the Model 3 we look at the effect of slavery on inequality within races and we
find it is not significant.10
8
Note that the sample represents existing counties in 1860. In these counties whites and blacks together
still represent more than 95% of the total population. As a result inequality between races largely
captures the difference between whites and blacks.
9
The cross-county correlation between the fraction of slaves in 1860 and the fraction of blacks in 2000
is 0.80. Therefore, despite the massive relocation of blacks from the rural South to the urban North,
race and past slavery still appear closely related. In the 1970s there was actually a reversal of black
migration flows due to employment and integration opportunities in the South (Ashenfelter and
Heckman, 1976).
10
We also experiment with a quadratic form on the share of slaves in 1860. Its effect on racial
inequality, which is our main focus, is only marginally significant and still positive. Therefore we retain
a linear form in the following models.
8
Table 1: Slavery and Inequality Estimation Method: OLS
Dependent Variables
Model 1
Overall Ineq.
0.0452***
(5.96)
0.00597***
(5.50)
0.251***
(22.07)
Model 2
Racial Ineq.
0.0492***
(10.67)
0.00308***
(18.23)
0.00623***
(4.70)
Model 3
Within Ineq.
-0.00393
(-0.73)
0.00289***
(3.10)
0.245***
(19.28)
Yes
Yes
Yes
R-squared
1,980
0.427
1,980
0.334
1,980
0.353
Sample
All Counties
All Counties
All Counties
Slaves/Population 1860
Population Density 1860
Constant
State Dummies
Observations
*** p<0.01, ** p<0.05, * p<0.1. Robust t statistics in parentheses.
The long-term influence of slavery can also run through institutional differences such as de jure
institutions aimed at protecting property rights in human.11 The influence of these factors may be
powerful even in the presence of small fractions of slaves even though should be eliminated by state
fixed effects. In turn the intensity of slavery, as a form of work organization shaping social and
institutional relations, is more likely to reflect the effect of de facto institutions which, according to
Key (1964), should be more binding in counties with a larger share of slaves.12 To gain further insight
into the relevance of our empirical implementation of the concept of slavery, in the specifications
shown in Table 1 we substitute the slave share in 1860 with the slave share in 1800, i.e., measured at a
time when several Northern states still had small fractions of slaves. The results from this specification
(which we do not report for brevity) still show a significant effect of the slave share on racial
inequality, while the effect on within inequality becomes significantly negative and makes the effect on
overall inequality insignificant.13
11
Einhorn (2002) discusses the links between US property law and slave property law.
12
Wright (2006) draws a similar distinction between slavery as a form of work organization and
slavery as a set of property rights.
13
To verify whether the effect of the slave share on inequality vanishes after accounting for a proxy of
current racism, we also add a control for the share of black population in 2000 (as in Lagerlöf, 2005).
Despite the fact that the new variable exhibits a positive coefficient on racial inequality, the effect of
9
Engerman and Sokoloff (1997, 2005a) have influentially argued that the diffusion of agricultural crops
best suited for the employment of slave labor is determined by factor endowments, such as soils,
climate, etc., best suited to large-scale plantations. The resulting unequal structure of society has in turn
contributed to the evolution of a set of legal, political, and educational institutions meant to preserve
the privileges of the elites, which have exerted a persistent impact on economic outcomes long after
the abolition of slavery, by determining paths of development characterized by marked inequalities. If
this theory is verified, the influence of the legal institution of slavery would come from its association
with factor endowments.14 In our context, the latter could be proxied by land inequality, which should
in turn reflect the diffusion of those crops that were typical of large-scale plantations and thus of the
use of slave labor. To control for this potential historical confound, we construct an index of land
inequality similar to the one employed by Lagerlöf (2005), Nunn (2008b), and Galor et al. (2009).15 It
is reasonable to expect that, within counties with a prevalence of large-scale plantations and therefore
large land inequality, income inequality in mid-nineteenth century was higher, because of the presence
of a small white elite and a large share of poor blacks. This in turn implied, in those days, a larger
degree of inequality between blacks and whites (Nunn, 2008b). This initial racial inequality may have
persisted until the present day and contributed to the higher overall economic inequality, as suggested
by our previous results.
In Table 2 we insert controls for factor endowments. In Models 1-3 we add the index of land inequality
in 1860 to the specifications of Table 1. In Model 1 we find a marginal effect of land inequality on the
slavery in confirmed, suggesting that the legacy of slavery may persist through other channels other
than racism. See Morgan (1975) for a discussion on the development of slavery and the emergence of
racism in colonial Virginia.
14
The direct link between endowments and slavery, where the former are measured by temperature,
elevation, and precipitation, has been examined for the US by Lagerlöf (2005).
15
Consistently with the literature the index reflects the size, rather than the value, of land holdings and
is calculated using information about the size of each farm (1860 Census). Sizes of farms fall in the
following ranges: (1) 9 acres or less, (2) 10 to 19 acres, (3) 20 to 49 acres, (4) 50 to 99 acres, (5) 100 to
499 acres, (6) 500 to 999 acres, and (7) 1,000 acres or more. We assume that farms are uniformly
distributed within each category and for the category of 1,000 acres or more we use 1,000 acres.
10
overall level of inequality in 2000. However, this hardly impacts on the coefficient of the slave share
variable. In addition, as Model 3 shows, the marginal effect of land inequality on overall inequality is
likely to be related to the effect on inequality within races, since there is no effect of land inequality on
inequality between races (Model 2). Therefore, although land inequality and slavery are statistically
associated their effects on the current overall inequality run through different channels.16 Similar results
(in regressions which we do not report for brevity) are obtained using alternative measures of
inequality in 1860 which are constructed using data on household wealth and the value of real estate
holdings from IPUMS. Both controls exert a significant effect on current racial inequality but they
hardly affect the coefficient on the share of slaves in 1860.
Table 2: Slavery and Inequality – Controlling for Factor Endowments Estimation Method: OLS
Model 1
Model 2
Model 3
Model 4
Model 5
Model 6
Dependent Variables
Overall Ineq.
Racial Ineq.
Within Ineq.
Overall Ineq.
Racial Ineq.
Within Ineq.
Slaves/Population 1860
0.0468***
0.0494***
-0.0026
0.0403***
0.0477***
-0.00746
(6.11)
(10.70)
(-0.48)
(5.37)
(10.50)
(-1.37)
0.00588***
0.00308***
0.00280***
0.00553***
0.00553***
0.00297***
(5.63)
(18.30)
(3.15)
(5.45)
(19.77)
(2.91)
0.0214*
0.00170
0.0197**
0.00884
0.00884
-0.00250
(1.79)
(0.24)
(2.23)
(0.74)
(-0.35)
(1.28)
0.00206*
0.000564
0.00150*
(1.79)
(1.28)
(1.70)
0.000794
0.000450
0.000344
(1.42)
(1.49)
(0.78)
-0.00574***
-0.00139***
-0.00435***
(-8.08)
(-4.49)
(-7.27)
Population Density 1860
Land Inequality 1860
Soil PH
Longitude
Latitude
Constant
0.243***
0.00263
0.240***
0.608***
0.608***
0.117***
(24.61)
(0.88)
(25.62)
(8.31)
(3.38)
(7.98)
State Dummies
Yes
Yes
Yes
Yes
Yes
Yes
Observations
1,875
1,875
1,875
1,836
1,836
1,836
R-squared
0.418
0.327
0.343
0.440
0.332
0.369
Sample
All Counties
All Counties
All Counties
All Counties
All Counties
All Counties
*** p<0.01, ** p<0.05, * p<0.1. Robust t statistics in parentheses.
16
According to Wright (2006) this distinction can be the result of different property rights emerging
from slave and free labor production systems.
11
In Models 4-6 of Table 2 we also control for latitude and longitude, which largely explain climatic
differences within the US, and for a proxy of soil neutrality (pH) collected from the Harmonized World
Soil Database (FAO GeoNetwork). Climate and soil quality should largely determine the crops
produced in an area providing a deeper control for crops suitability. The impact of the share of slaves is
hardly affected despite a significant effect of latitude on all measures of inequality and a marginal
effect of soil neutrality on overall and within inequality. Therefore, the effect of slavery persists even
after controlling for the diffusion of large-scale plantations as determined by factor endowments.
Identical results obtain in a variant excluding land inequality (which we omit for brevity).
Our results so far indicate a strong and persistent effect of slavery on current inequality and in
particular on current racial inequality. The relationship between slavery, land inequality, and racial
inequality may also be biased by omitted variables and measurement error. As a final robustness check
we test the effect of slavery in 1860 on the latter using an IV model in order to provide some source of
exogenous variation, first to the share of slaves in 1860 and then also to land inequality in 1860, using
distinct instruments.
Mann (2011) argues that the reason why slavery developed in the US is because of the resistance of
African slaves to tropical diseases, and in particular to malaria. Using data on the force and stability of
malaria transmission and on malaria endemicity, Esposito (2013) documents a strong cross-sectional
correlation between geographic suitability to malaria and the diffusion of slavery across US counties.
Based on these findings, we use a similar instrument for the share of slaves in 1860 based on a proxy of
environmental suitability for transmission of malaria falciparum. In order to construct our proxy we use
disaggregated GIS data at a 1km level from the Malaria Atlas Project, which then is aggregated at a
county level (1860 boundaries). The resulting index of suitability is a relative measure of the effect of
environmental conditions on vectorial capacity and reproductive number which is proxied using a
biological model incorporating the impact of climate on vector lifespan and the duration of
P.
falciparum sporogeny (Gething et al., 2011). Given that the index is constructed using exogenous
variations in climate the environmental hazard is likely to remain constant before and after the
introduction of slavery. In addition, unlike Africa, in the US malaria has been eradicated since the
1940s, so that the malaria ecology is unlikely to have a direct impact on current economic outcomes.
12
In principle, as argued by Wright (2006), land inequality could be endogenous to slavery, as slave
owners could easily acquire the best lands. Moreover, the relationship between slavery, land inequality,
and racial inequality may also be biased by omitted variables and measurement errors. Therefore, to
gain further insights we use FAO GAEZ data to construct indexes of soil suitability to cotton and
tobacco - which normally are associated with scale economies - to construct instruments for land
inequality in 1860. Cotton and tobacco have also been associated with the use of slavery. Therefore, in
Table 3 we report the pairwise correlation between the presumed endogenous variables and their
proposed instruments. Our measure of environmental suitability for malaria falciparum is strongly
correlated with the share of slaves in 1860 (0.57) and much less correlated with land inequality in 1860
(0.27). The index of soil suitability to tobacco is correlated strongly with land inequality and only
weakly correlated with the share of slaves. The cross-correlation between malaria’s and tobacco’s
suitability is also rather low (0.29). The facts that each of these two candidate instruments is highly
correlated with one of the endogenous variable only and that the cross-correlation between instruments
is quite low are important in order to disentangle the orthogonal part of the instrument correlated with
endogenous variables (Shea, 1997). Cotton suitability on the other hand is quite significantly correlated
with both slavery and land inequality, which confirms the evidence that land inequality and slavery
were more frequent along the Cotton Belt.
Table 3: Slavery and Factor Endowments – Pairwise Correlations
Slaves
Land Ineq.
Malaria
Cotton
Slaves/Population 1860
1.0000
Land Inequality 1860
0.1297
1.0000
Climate Suitability to Malaria F.
0.5744
0.2758
1.0000
Soil Suitability to Cotton
0.4160
0.3842
0.6910
1.0000
Soil Suitability to Tobacco
0.1294
0.4591
0.2956
0.0994
Tobacco
1.0000
In Table 4 we re-estimate Model 2 in Table 2 using an IV estimator where the dependent variable is
racial inequality. In Model 1 we only use an instrument for the share of slaves in 1860 and the
coefficient on slavery is now much larger and still highly significant. In Model 2 we also instrument for
land inequality using soil suitability to tobacco and the coefficient on the share of slaves in 1860 is
almost unchanged. In Model 3 we also enter the index of soil suitability to cotton as an additional
instrument and the null that the model is not over-identified (Hansen Test) cannot be rejected. Beside
13
diagnostic tests, the suitability of the instruments is also confirmed by the fact that the coefficients on
the share of slaves and land inequality are extremely stable across models, which is unlikely to happen
if the instruments are endogenous (Wooldridge, 2002) given that it is difficult to have a bias which is
systematically the same across specifications.17
Table 4: Slavery and Racial Inequality – IV Estimates EstimationMethod: IV
Dependent variable
Model 1
Model 2
Racial Inequality
Model 3
0.0915***
0.0967***
0.0934***
(10.67)
0.00304***
(20.26)
0.00817
(1.05)
-0.00120
(-0.35)
(9.58)
0.00315***
(14.96)
-0.0176
(-1.06)
0.01000
(1.38)
(5.58)
0.00314***
(5.79)
-0.0164
(-1.00)
0.00949
(1.25)
Variables Instrumented
Slaves
Slaves and Land Ineq
Slaves and Land Ineq
Instruments
Malaria
Suitability
Malaria Suitab.+
Tobacco Suitab.
Malaria Suitab.+
Tobacco Suitab. +
Cotton Suitab.
88.453
88.842
16.38
67.149
33.525
7.03
0.0000
Yes
0.0000
Yes
75.977
25.336
13.43
0.5696
0.0190
Yes
1,833
0.269
All Counties
1,833
0.247
All Counties
1,830
0.258
All Counties
Slaves/Population 1860
Population Density 1860
Land Inequality 1860
Constant
Anderson Canon. LR Stat.
Cragg Donald F-Stat
Stock and Yogo Critic. Val
Hansen J-stat
Endogeneity
State Dummies
Observations
R-Squared
Sample
*** p<0.01, ** p<0.05, * p<0.1. Robust t statistics in parentheses.
17
As a falsification exercise we also regress malaria suitability against our best proxy of agricultural
conditions (soil neutrality) and we find an insignificant effect of soil pH on malaria, consistent with the
idea that the slave share is independent of particular agricultural conditions.
14
Table A2 in the Appendix reports first stage statistics. We find that the effect of the index of malaria
suitability on the share of slaves in 1860 is highly significant, while the index of tobacco suitability
does not exert any effect on slavery. The opposite pattern of significance occurs for land inequality.
Malaria suitability does not exert any effect on land inequality while tobacco suitability is highly
significant. Therefore each instrument has an independent effect on the two endogenous variables
providing enough orthogonality between predicted values, as confirmed by the Cragg-Donald Fstatistics which is much larger than the critical values in Stock and Yogo (2005). As a result, the
instruments are both relevant.18
4. Channels of Transmission
So far we have shown that there is a robust and persistent relationship between slavery and current
inequality, particularly on its racial component. In this section we try to understand which is the
channel through which this influence manifests itself. While the legacy of slavery for contemporary
economic outcomes may have run through multiple channels, ranging from social capital19 and wealth
inequality to law enforcement and criminalization, here we focus on two specific mechanisms: racial
discrimination and human capital transmission.
According to the racial discrimination mechanism, one of the legacies of slavery was a gap in the
earnings of blacks which can be attributed to discrimination.20 To test this hypothesis we proceed as
follows. We start by creating a measure of racial discrimination. To this end, we estimate returns on
education at the county level for blacks and whites through a model akin to a macro-Mincerian
18
IV estimates of Model 5 of Table 2, which involves a longer list of controls for factor endowments,
yield identical results. Therefore we refer to Model 2 of Table 2 as a benchmark for the rest of the
paper.
19
Chay and Munshi (2012) study network formation among blacks after abolition and how it affected
their voting and migration behavior.
20
Sundstrom (2007) estimates the white-black earnings gap of men in the South in 1940, adjusting for
individual differences in schooling and experience, and finds evidence of discrimination, which
depressed black workers’ wages and also prevented them from acquiring skills.
15
equation.21 The estimates are reported in Table A3 in the Table Appendix.22 Beside educational
attainment, we also control for experience, as proxied by the employment rate and median age for each
group, for the proportion of whites and blacks in the labor force, to capture clusters or network effects,
and for fixed geographical effects (namely, for counties within North Eastern and South Atlantic
states.23 In Table A4 we summarize the descriptive statistics resulting from our estimates. As expected
income per capita tends to increase with the level of education. On average, for educated whites income
per capita is 71.3 percent higher than for whites who did not complete high school, while for educated
blacks income per capita is only 36.5 percent higher. We use predicted returns to construct a proxy of
discrimination between blacks and whites which is equal to the ratio of average returns for blacks to
average returns for whites. A ratio below one denotes the existence of a possible racial discrimination.
According to the human capital transmission channel, the legacy of slavery runs through educational
inequality. This happens since blacks, the vast majority of whom descend from slaves with no
education (Smith, 1989 and Margo, 1990), have accumulated a gap in terms of education which results
in inequality between blacks and whites, and in turn in overall inequality. As a preliminary test of this
hypothesis, we construct a measure of racial inequality for education,24 based on information on the
years of schooling for blacks and whites.25
Table A5 in the Table Appendix reports descriptive statistics for the two proxies we have constructed in
21
For a discussion of macro-Mincerian equations see Krueger and Lindhal (2001) and references
therein.
22
In order to estimate returns for whites and blacks at the county level we compute separate estimates
for the two groups. Data are from the 2000 US Census.
23
The geographic dummies are considered since there is evidence that the vicinity to the coast, in
particular to the Atlantic Ocean, has a direct effect on current economic outcomes (see Lagerlöf, 2005).
See also Rappaport and Sachs (2003) for an analysis of the coastal concentration of economic activity
in the US.
24
Data on educational attainment by race are from the 2000 US Census. For each race we compute the
average years of schooling and then we derive a Gini index.
25
See Birdsall and Londono (1997) and Thomas et al. (2002) for cross-country analyses of education
Gini indices.
16
order to test the two potential channels. We present these statistics for the entire sample of counties and
also for the sub-sample of counties belonging to former slave states. At the mean, the ratio of the
expected returns on education we estimated for blacks and whites is 0.51, across all counties. When
confined to former slave states, the blacks to whites ratio of returns on education is even smaller,
suggesting the presence of more discrimination down in the South. The distribution of education is also
more unequal within slave states.
Table 5: Slavery and Inequality – Channels of Transmission
OLS Estimates
Dependent Variable
Slaves/Population 1860
Population Density 1860
Land Inequality 1860
Constant
IV Estimates
Model 1
Model 2
Model 3
Model 4
Racial
Educational
Inequality
Blacks/Whites
Returns on Edu
Racial
Educational
Inequality
Blacks/Whites
Returns on
Edu
0.0346***
(18.60)
0.00204***
(6.49)
0.00283
-0.208***
(-6.51)
-0.0054***
(-3.28)
0.0889
0.0852***
(11.64)
0.00206***
(6.54)
-0.00724
(0.82)
0.0133***
(3.71)
(1.54)
0.425***
(7.55)
(-0.92)
0.00643*
(1.83)
Variables Instrumented
Instruments
Anderson Canon. LR Stat.
Cragg Donald F-Stat
Stock and Yogo Critic. Val
Hansen J-stat
Endogeneity
State Dummies
Yes
Yes
Observations
1,849
1,803
R-Squared
0.594
0.144
Sample
All Counties
All Counties
*** p<0.01, ** p<0.05, * p<0.1. Robust t statistics in
parentheses.
-0.267
(-1.60)
-0.00366***
(-2.97)
-0.267
(-1.49)
0.680***
(8.17)
Slaves and Land Ineq
Malaria Suitab.+ Tobacco Suitab. +
Cotton Suitab.
75.41
25.143
13.43
0.2054
0.000
Yes
1,805
0.323
All Counties
67.696
22.521
13.43
0.9249
0.0436
Yes
1,759
0.128
All Counties
17
In Table 5 we compare our two hypotheses. In Model 1 we regress our measures of racial educational
inequality on the previous set of controls and we find a significant effect of the share of slaves, which
increases racial inequality in education by almost 0.005 per a standard deviation in the share of slaves.
In Model 2 we replace the dependent variable with our measures of racial discrimination in the labor
market and we also find that the share of slaves has a significant effect on the black/white ratio of
returns on education, by increasing the gap by almost 0.097 per a standard deviation. To test the
robustness of these results in Models 3 and 4 we also use an IV estimator using the same set of
instruments as in Model 4 in Table 426 and we find that the effect of slavery on racial educational
inequality is reinforced, while the coefficient on the returns loses significance.
The evidence we collected so far thus leads us to conclude that current income inequality between races
is primarily influenced by slavery through the impact exerted by the latter on the unequal educational
attainment between races.27 In other words, it is through human capital transmission that slavery
determines the cross-county distribution of inequality in the US today.
Our results share a number of features with the literature on race and inequality. For example Smith and
Welch (1986) find that black relative wages rose from 18 to 35 percent in the 1960-80 period and that
education is the overwhelmingly dominant factor in explaining this relative advance, while Card and
Krueger (1992) attribute to schooling between 15 and 20 percent of the overall growth in black-white
relative earnings between 1960 and 1980. At the same time, Neal (2006) collects evidence suggesting
that, even though the black-white educational gaps diminished over most of the twentieth century, this
convergence process is still far from complete. Moreover, over a panel dataset covering the 1940-2000
period at the state level, Bertocchi and Dimico (2012a) find that the racial educational gap significantly
depends on its 1940 level, which is in turn larger in former slave states: in other words, states with a
larger initial gap still have larger racial educational inequality.28 These results are consistent with the
present finding that the inequality of education between races is the main determinant of income
26
First stages are presented in Table A2 in the Table Appendix.
27
In the context of nineteenth century Puerto Rico, Bobonis and Morrow (2010) find that the abolition
of labor coercion reduced laborers’ incentives to accumulate human capital.
28
The correlation between the racial gap in education in 1940 and the fraction slaves/population in
1860 is 0.90 at the high-school level and 0.78 at the bachelor-degree level.
18
inequality because of its impact on racial income inequality.
5. A Quasi Experiment
In this section we exploit the exogenous variation in slavery determined by the instruments used for our
IV specification to carry out an additional test. According to Engerman and Sokoloff (1997, 2005a) and
Mann (2011) slavery largely depends on factor endowments and malaria hazard. Therefore we can
exploit this exogenous variation to predict the probability of slavery by using the “discontinuity” in the
use of slavery along the a sort of “extended Mason-Dixon line” dividing between slave and non-slave
states: in other words, we can match counties which did and did not make use of slavery despite having
the same probability of developing a slave society. The discontinuity in the use of slavery can therefore
be interpreted as the result of state and local politics only. As a result, if we control for this channel then
we should obtain an unbiased estimate of the effect of slavery on inequality.
In more details, if we look at counties along the border between slave and non-slave states (as
illustrated in Figure A1 in the Appendix), then counties within slave and non-slave states should be
very similar in terms of factor endowments and because of that they should have the same probability
of developing slavery. However, only those counties within slave states actually ended up introducing
this form of labor coercion, mainly because of differences in state politics. Therefore if within the same
sample of counties with a similar probability of developing slavery we manage to control for state
politics and state policy variables then we should be able to estimate the counterfactual.
From an empirical point of view first we need to match counties which have a similar probability of
developing slavery and then we can estimate the effect of the latter on long-term outcomes. This can be
done using a propensity score model which is then followed by an OLS estimator. The propensity score
model will assign the probability of developing slavery on the basis of factor endowments and will
match counties which have a similar probability of developing slavery by choosing the region within a
common support. Then we can look at counties within this region in order to understand what would
have happened to counties which have not been treated with slavery if they had developed this form of
labor coercion. This is done by exploiting the variation in slavery between matched counties as in a sort
19
of county-pair identification model.29 The fact that we use a propensity score to match counties will
permit us to have a better matching since counties are chosen on the basis of specific criteria related to
slavery and they are not assumed to be a priori identical. As a consequence we will be able to control
for any county-specific factor which could be related to slavery (except for those related to state
politics and policies) and at the same time we will be able to control for spatial fixed effects in the same
way a county-pair identification model does. As usual, clustering at the state-border level allows to
control for the correlation across counties sharing the same borders.
The basic assumption behind our estimation procedure is conditional independence, which in our case
means that there is a set of observed characteristics Z (related to factor endowments and malaria
environment) such that the outcome is independent on the probability of slavery conditional on a set of
pre-treatment independent variables X. In other words the selection into treatment depends on
observables up to a random factor, so that our estimation procedure can be considered as a weaker
version of a pure randomization in which the treatment is independent of pre-treated independent
variables. Of course there are costs associated with the use of such an estimator. The main one relates
to the limited sample size, which in our case is confined to the set of counties along the border between
slave and non-slave states. The smaller sample size affects the variation in our main explanatory
variable (slavery) given that the use of slavery was largely restricted to the deep South. On the other
side, this simple quasi-experiment will help us to understand what would have happened to counties
which used slavery if this form of labor coercion had been banned in the first place.
Table A6 and A7 in the Table Appendix show the propensity score model and the distribution of the
propensity score in the treatment and matched comparison groups across blocks (Rosenbaum and
Rubin, 1983). The difference in mean between counties exposed and not exposed to slavery is not
significantly different across blocks, which confirms that the characteristics of the treatment and
comparison groups are not significantly different from each other within blocks. This test is particularly
important for our analysis given that it ensures that matched counties with and without slavery are
exactly the same, so that there are no unobserved characteristics which can affect the estimates once we
control for state politics and policies by using a set of state dummy variables.
29
A county-pair identification strategy has been used among others by Dube (2010) and Naidu (2009),
to test the effect of the minimum wage and of disenfranchisement schemes, respectively.
20
Table 6 provides estimates restricted to the region of common support. In Model 1 we still find a
significant effect of slavery on income inequality across races. In Model 2 we test the effect of the
share of slaves on racial educational inequality and we can also confirm the significant effect of the
former on the latter. In Model 3 we replace the dependent variable with the gap in the returns to
education for blacks and whites and, consistently with the IV estimates in Section 4, we do not find a
statistical significant effect of slavery, which implies that racial educational inequality is the main
channels through which slavery affects the inequality between races.
To conclude, our robustness checks confirm the results found in the previous section suggesting that the
persistence in the racial gap in education is a crucial channel of transmission of from slavery to current
inequality within the US.
Table 6: Quasi Experiment (Matching Model) Estimation Method: OLS
Model 1
Model 2
Dependent Variables
Racial
Inequality
Racial Educational
Inequality
Model 3
Blacks/Whites
Returns on Edu.
0.0286***
(4.81)
0.0539***
(4.69)
0.0262*
(2.00)
0.0337***
(5.10)
0.0321***
(6.87)
0.0225**
(2.54)
-0.0747
(-0.63)
-0.288**
(-2.24)
0.271
(0.62)
Constant
-0.00122
(-0.21)
-0.00239
(-0.58)
0.385**
(2.55)
State Dummies
Observations
R-Squared
Yes
Yes
Yes
145
0.497
Common Support
143
0.547
Common Support
133
0.106
Common Support
Slaves/Population 1860
Population Density 1860
Land Inequality 1860
Sample
*** p<0.01, ** p<0.05, * p<0.1. Robust t statistics in parentheses.
21
6. Why Education?
As outlined in the previous sections one of the main reasons why the effect of slavery on inequality
runs through the inequality between races is related to the huge impact that education has on wages
(see also Goldin and Katz, 1999). The natural question to ask now is why slavery may have affected
education inequality. One of the main reasons is related to local politics and to the close link between
them and educational policies. In other words, educational policies as shaped by local politics represent
a channel through which the gap in education between whites and blacks has persisted. Using a panel
of US states over the 1940-2000 period Bertocchi and Dimico (2012a) show that a larger educational
gap between whites and blacks in 1940 has caused a long-lasting gap in education which still persists
today. This gap is the result of the post-Reconstruction politics and “separate but equal” educational
policies. Using data on voting registration for blacks and whites within the state of Mississippi
Bertocchi and Dimico (2012b) show that in 1917 the student-teacher ratio for blacks is much higher in
counties with lower black registration. This higher ratio is the result of state-funding diversion from
black to white pupils. Since local officers in former slave states could use funding for black pupils in
order to finance education for white pupils, they applied a lower property tax. The property tax is the
principal source of revenue for localities/counties and it also represents the main source of funding for
education, with nearly half of all property-tax revenues used for public elementary and secondary
education (Kenyon, 2007). Even after the abolition of “separate but equal” counties which used to have
a lower property tax still kept a lower tax rate, with a negative effect on public school funding and
therefore education for blacks.
Plots in Figure 1 show the relationship between the slave share in 1860 and local politics, in terms of
presidential voting turnout and Democratic share.30 The two plots at the top show the relationship for
the year 1900, when almost all the Southern states - with the exception of Texas, Alabama, and
Virginia - had introduced forms of disenfranchisement for blacks. The plot on the left shows that
turnout in counties with a massive share of slaves in 1860 is much smaller than in counties with a lower
share of slaves (almost -0.7 percentage points in turnout per a 1 percent increase in the share of slaves)
while the plot on the right shows that slavery is also closely associated with a larger share for the
Democratic party which, at the time, represented the interests of the white elites. The two plots at the
30
Data on presidential elections are from Clubb et al. (2006).
22
bottom at the figure show that this relationship still persists until 1960, i.e., until the realignment of
party politics and the approval of the Voting Rights Act. Of course one can argue that presidential
elections data are not appropriate to evaluate local politics, but in the US presidential turnout has
always been the highest if compared to other elections, so that in the worst case scenario results should
be downward biased.31 The connection between disenfranchisement of blacks and local politics is
confirmed by a more formal analysis in Bertocchi and Dimico (2012b), Naidu (2011) and Acharya et
al. (2013), while Acemoglu et al. (2013) offer a survey of the literature on democracy, redistribution
and public good provision including education.
0
0
20
20
40
40
60
60
80
80
100
100
Figure 1: Slavery, Presidential Turnout, and Democratic Share
0
.2
.4
.6
.8
1
0
slave_1860
1900 Presidential Turnout (%)
.2
.4
.6
.8
1
slave_1860
Fitted values
1900 Democratic Vote Share (%)
Fitted values
0
0
20
20
40
40
60
60
80
80
100
100
0
.2
.4
.6
slave share 1860
1960 Presidential Turnout (%)
.8
1
0
.2
.4
.6
.8
1
slave_1860
Fitted values
1960 Democratic Vote Share (%)
Fitted values
While the above documented political balance of power had broader implications, the next link
between slavery and education inequality is to be found in the local nature of education provision and
31
We use data on turnout for presidential elections because historical data on local elections are not
available.
23
funding. As mentioned above, the property tax is the main source of funding for localities/counties and
thus for public education support. Because of the “separate but equal” educational policies applied in
Southern states until the 1960s, local officers could divert state funding for blacks to finance education
for whites. As a result they could impose a lower property tax and spend less in education. The
relationship between slavery and per capita property taxes is shown in plots in Figure 2 for 1962 (on
the left) and 1992 (on the right). The plots show that a larger share of slaves is associated with a
smaller per capita tax in 1962 and the relationship still hold is in 1992, which implies a sort of
persistence in the education policies implemented before and after the abolition of black
disenfranchisement.
0
0
2
.2
4
6
.4
8
.6
10
Figure 2: Slavery and Property Taxes 0
.2
.4
.6
.8
slave_1860
1962 Per Capita Property Tax
Fitted values
1
0
.2
.4
.6
.8
1
slave_1860
1992 Per Capita Property Tax
Fitted values
Finally in Figure 3 on the left we show the relationship between slavery and education expenditure per
capita in 2000 (from the National Center for Education Statistics). Education expenditure in 2000 is
still negatively affected by slavery. On the right we plot a measure of the average education attainment
of the black population over age 25 in 2000 (the source is Census), which is inversely related to past
slavery despite the above-mentioned evidence of convergence between blacks and whites in terms of
education.
24
6.5
1
2
7
3
7.5
4
8
5
Figure 3: Slavery and Education Expenditure
0
.2
.4
.6
.8
slave_1860
Log Total Education Expenditure p.c. (2000)
Fitted values
1
0
.2
.4
.6
.8
1
slave_1860
2000 Weighted Average Black Education blk 25+
Fitted values
While the evidence we have reported in this section of course does not imply causality, it does provide
a clear illustration of the strong links between the legacy of slavery, local politics, fiscal policy choices
and current education outcomes.
7. Conclusion
In this paper we have shown that the legacy of slavery still plays a major role in the US economy and
society, since the use of slave labor persistently affects current inequality, and particularly its racial
component. In other words, those US counties that in the past exhibited a higher slave share over
population turn out to be still more unequal in the present day. We also show that human capital
transmission is a crucial channel through which slavery manifests its legacy.
The declared goal of recent reform programs for the American schooling system has been the removal
of the persistent racial and ethnic educational gaps that are still pervasive in the US society. This is the
case both for No Child Left Behind, the federal education program enacted in 2002 under President
Bush, and for Race to the Top, the current program created by President Obama. This goal is consistent
with the evidence we provide, showing how deeply educational inequality is rooted in the history of the
country.
25
REFERENCES
Acemoglu, D., Bautista, M. A., Querubín, P. and Robinson, J. A., 2008, Economic and Political
Inequality in Development: The Case of Cundinamarca, Colombia, in Helpman, E. (ed.), Institutions
and Economic Performance, Harvard University Press, Cambridge, 181-245.
Acemoglu, D., Johnson, S. and Robinson, J. A., 2002, Reversal of Fortune: Geography and
Institutions in the Making of the Modern World Income Distribution, Quarterly Journal of Economics
117, 1231-1294.
Acemoglu, D., Johnson, S., Robinson, J. A. and Yared, P., 2008, Income and Democracy, American
Economic Review 98, 808-842.
Acemoglu, D., Naidu, S., Restrepo, P. and Robinson, J. A., 2013, Democracy, Redistribution and
Inequality, NBER Working Paper No. 19746.
Acemoglu, D. and Robinson, J. A., 2008, Persistence of Power, Elites, and Institutions, American
Economic Review 98, 267-293.
Acemoglu, D. and Wolitzky, A., 2011, The Economics of Labor Coercion, Econometrica 79, 555600.
Acharya, A., Blackwell, M. and Sen, M., 2013, The Political Legacy of American Slavery, mimeo,
University of Rochester.
Ashenfelter, O. and Heckman, J. J., 1976, Measuring the Effect of an Anti-discrimination Program,
in Ashenfelter, O. and Blum, J. (eds.), Estimating the Labor Market Effects of Social Programs,
Princeton University Press, Princeton, 46-89.
Berlin, I., 2003, Generations of Captivity: A History of African American Slaves, Harvard University
Press, Cambridge.
Bertocchi, G. and Dimico, A., 2012a, The Racial Gap in Education and the Legacy of Slavery, Journal
of Comparative Economics 40, 581-595.
Bertocchi, G. and Dimico, A., 2012b, De Jure and de Facto Determinants of Power: Evidence from
Mississippi, CEPR Discussion Paper No. 9064.
Bertocchi, G. and Dimico, A., 2010, Slavery, Education, and Inequality, CEPR Discussion Paper No.
8073.
Birdsall, B. and Londono, J. J., 1997, Asset Inequality Matters: An Assessment of the World Bank's
Approach to Poverty Reduction, American Economic Review Papers and Proceedings 87, 32-37.
Bobonis, G. J. and Morrow, P. M., 2010, Labor Coercion and the Accumulation of Human Capital,
26
mimeo, University of Toronto.
Cameron, S. V. and Heckman, J. J., 2001, The Dynamics of Educational Attainment for Blacks,
Hispanics, and Whites, Journal of Political Economy 109, 455-499.
Card, D. and Krueger, A. B., 1992, School Quality and Black-White Relative Earnings: A Direct
Assessment, Quarterly Journal of Economics 107, 151-200.
Caselli, F. and Coleman, W. J., 2001, The U.S. Structural Transformation and Regional Convergence:
A Reinterpretation, Journal of Political Economy 109, 584-616.
Chay, K. and Munshi, K.,
2012, Black Networks After Emancipation: Evidence from
Reconstruction and the Great Migration, mimeo, Brown University.
Clubb, J. M., Flanigan, W. H. and Zingale, N. H., 2006, Electoral Data for Counties in the United
States: Presidential and Congressional Races, 1840-1972, Inter-university Consortium for Political and
Social Research.
Collins, W. J. and Margo, R. A., 2006, Historical Perspectives on Racial Differences in Schooling in
the United States, in Hanushek, E. and Welch, F. (eds.), Handbook of the Economics of Education,
Elsevier, New York, 107-154.
David, P. A. and Stampp, K. M. (eds.), 1976, Reckoning With Slavery: Critical Essays in the
Quantitative History of American Negro Slavery, Oxford University Press, New York.
Donohue, J. J., III and Heckman, J., 1991, Continuous Versus Episodic Change: The Impact of Civil
Rights Policy on the Economic Status of Blacks, Journal of Economic Literature 29, 1603-1643.
Dube, A., Lester, T. W. and Reich, M., 2010, Minimum Wage Effects Across State Borders:
Estimates Using Contiguous Counties, Review of Economics and Statistics 92, 945-964.
Einhorn, R. L., 2002, Patrick Henry's Case Against the Constitution: The Structural Problem With
Slavery, Journal of the Early Republic 22, 549-573.
Engerman, S. L. and Sokoloff, K. L., 2005a, Institutional and Non-Institutionals Explanations of
Economic Differences, in Menard, C. and Shirley, M. M. (eds.), Handbook of New Institutional
Economics, Springer, Dordrecht, 639-665.
Engerman, S. L. and Sokoloff, K. L., 2005b, The Evolution of Suffrage Institutions in the Americas,
Journal of Economic History 65, 891-921.
Engerman, S. L. and Sokoloff, K. L., 1997, Factor Endowments, Institutions, and Differential Paths
of Growth Among New World Economies: A View from Economic Historians of the United States, in
Harber, S. (ed.), How Latin America Fell Behind, Stanford University Press, Stanford, 260-304.
27
Esposito, E., 2013, Side Effects of Immunities: The African Slave Trade, mimeo, University of
Bologna.
Fogel, R. W. and Engerman, S. L., 1974, Time on the Cross: The Economics of American Negro
Slavery, Little Brown, Boston.
Galor, O., Moav, O. and Vollrath, D., 2009, Inequality in Landownership, the Emergence of HumanCapital Promoting Institutions, and the Great Divergence, Review of Economic Studies 76, 143-179.
Gething, P. W., Van Boeckel, T. P., Smith, D. L., Guerra, C. A., Patil, A. P., Snow, R. W. and Hay, S.
I., 2011, Modelling the Global Constraints of Temperature on Transmission of Plasmodium Falciparum
and P. Vivax, Parasites & Vectors 4, 92-102.
Goldin, C. and Margo, R. A., 1992, The Great Compression: The US Wage Structure at Mid-Century,
Quarterly Journal of Economics 107, 1-34.
Goldin, C., 1998, America’s Graduation from High School: The Evolution and Spread of Secondary
Schooling in the Twentieth Century, Journal of Economic History 58, 345-374.
Goldin, C. and Katz, L. F., 1999, The Shaping of Higher Education: The Formative Years in the
Unites States, 1890 to 1940, Journal of Economic Perspectives 13, 37-62.
Hall, R. E. and Jones, C. I., 1999, Why Do Some Countries Produce So Much More Output Per
Worker Than Others? Quarterly Journal of Economics 114, 83-116.
Heckman, J. J., 1990, The Central Role of the South in Accounting for Economic Progress of Black
Americans, American Economic Review 80, 242-246.
Irwin, J. R., 1994, Explaining the Decline in Southern per Capita Output after Emancipation,
Explorations in Economic History 31, 336-356.
Jenkins, S. P., 1999, INEQDEC0: Stata Module to Calculate Inequality Indices with Decomposition
by Subgroup, Statistical Software Components S366007, Boston College.
Kenyon, D. A., 2007, The Property Tax-School Funding Dilemma (Policy Focus Report), The
Lincoln Institute of Land Policy, Cambridge.
Key, V. O., 1964, Politics, Parties and Pressure Groups, 5th Edition, Crowell, New York.
Krueger, A. B. and Lindahl, M., 2001, Education for Growth: Why and For Whom, Journal of
Economic Literature 39, 1101-1136.
Lagerlöf, N., 2009, Slavery and Other Property Rights, Review of Economic Studies 76, 319-342.
Lagerlöf, N., 2005, Geography, Institutions, and Growth: The United States as a Microcosm, mimeo,
York University.
Mann, C. C., 2011, 1493: How the Ecological Collision of Europe and the Americas Gave Rise to
28
the Modern World, Granta Books, London.
Margo, R. A., 1990, Race and Schooling in the South, 1880-1950: An Economic History, University
of Chicago Press, Chicago.
Mariscal, E. and Sokoloff, K. L., 2000, Schooling, Suffrage, and the Persistence of Inequality in the
Americas, 1800–1945, in Haber, S. (ed.), Political Institutions and Economic Growth in Latin America:
Essays in Policy, History, and Political Economy, Hoover Institution Press, Stanford, 159-218.
Mitchener, K. J. and McLean, I. W., 2003, The Productivity of U.S. States Since 1880, Journal of
Economic Growth 8, 73-114.
Morgan, E., 1075, American Slavery, American Freedom: The Ordeal of Colonial Virginia, W. W.
Norton & Company, New York & London.
Naidu, S., 2010, Suffrage, Schooling, and Sorting in the Post-Bellum U.S. South, mimeo, Harvard
University.
Neal, D., 2006, Why Has Black-White Skill Convergence Stopped? in Hanushek, E. and Welch, F.
(eds.), Handbook of Economics of Education, Vol. 1, Ch. 9, North-Holland, Amsterdam, 511-576.
Nunn, N., 2008a, The Long-Term Effects of Africa's Slave Trades, Quarterly Journal of Economics
123, 139-176.
Nunn, N., 2008b, Slavery, Inequality, and Economic Development in the Americas: An Examination
of the Engerman-Sokoloff Hypothesis, in Helpman, E. (ed.), Institutions and Economic Performance,
Harvard University Press, Cambridge, 48-180.
Putterman, L. and Weil, D. N., 2010, Post-1500 Population Flows and the Long-Run Determinants of
Economic Growth and Inequality, Quarterly Journal of Economics 125, 1627-1682.
Ramcharan, R., 2010, Inequality and Redistribution: Evidence from U.S. Counties and States, 18901930, Review of Economics and Statistics 92, 729-744.
Ransom, R. L. and Sutch, R., 2001, One Kind of Freedom: The Economic Consequences of
Emancipation, 2nd Edition, Cambridge University Press, New York.
Rappaport, J. and Sachs, J. D., 2003, The United States as a Coastal Nation, Journal of Economic
Growth 8, 5-46.
Rosenbaum, P. R. and Rubin, D. B., 1983, The Central Role of the Propensity Score in Observational
Studies for Causal Effects, Biometrika 70, 41-55.
Sacerdote, B., 2005, Slavery and the Intergenerational Transmission of Human Capital, Review of
Economics and Statistics 87, 217-234.
29
Sachs, J. D., and Warner, A. M., 2001, The Curse of Natural Resources, European Economic
Review 45, 827-838.
Shea, J., 1997, Instrument Relevance in Multivariate Linear Models: A Simple Measure, Review of
Economics and Statistics 79, 348-352.
Smith, J. P., 1984, Race and Human Capital, American Economic Review 74, 685-98.
Smith, J. P. and Welch, F. R., 1989, Black Economic Progress After Myrdal, Journal of Economic
Literature 27, 519-564.
Sokoloff, K. L. and Engerman, S. L., 2000, Institutions, Factor Endowments, and Paths of
Development in the New World, Journal of Economic Perspectives 14, 217-232.
Sokoloff, K. L. and Zolt, E. M., 2007, Inequality and the Evolution of Institutions of Taxation:
Evidence from the Economic History of the Americas, in Edwards, S., Esquivel, G. and Marquez, G.
(eds.), The Decline of Latin American Economies: Growth, Institutions, and Crises, University of
Chicago Press, Chicago, 83-136.
Sundstrom, W. A., 2007, The Geography of Wage Discrimination in the Pre-Civil Rights South,
Journal of Economic History 67, 410-444.
Thomas, V., Wang, Y. and Fan, X., 2002, A New Dataset on Inequality in Education: Gini and Theil
Indices of Schooling for 140 Countries, 1960-2000, mimeo, World Bank.
Vollrath, D., 2013, School Funding and Inequality in the Rural United States, 1890, Explorations in
Economic History 50, 267-284.
Wooldridge, J. M., 2002, Introductory Econometrics, 2nd Edition, South-Western, Cincinnati.
Wright, G., 2006, Slavery and American Economic Development, Louisiana State University Press,
Baton Rouge.
30
Table Appendix
Table A1: Slave Shares, 1860
States
Mean
Std. Dev.
Nr of Counties
Slave State
ALABAMA
ARKANSAS
CALIFORNIA
COLORADO
CONNECTICUT
DELAWARE
FLORIDA
GEORGIA
IOWA
ILLINOIS
INDIANA
KANSAS
KENTUCKY
LOUISIANA
MASSACHUSETTS
MARYLAND
MAINE
MICHIGAN
MINNESOTA
MISSOURI
MISSISSIPPI
NORTH CAROLINA
NEBRASKA
NEW HAMPSHIRE
NEW JERSEY
NEW MEXICO
NEVADA
NEW YORK
OHIO
OREGON
PENNSYLVANIA
RHODE ISLAND
SOUTH CAROLINA
TENNESSEE
TEXAS
UTAH
VIRGINIA
VERMONT
WASHINGTON
WISCONSIN
0.391
0.218
0
0
0
0.019
0.334
0.395
0
0
0
0.00002
0.173
0.548
0
0.205
0
0
0
0.083
0.516
0.310
0.0004
0
0.00003
0
0
0
0
0
0
0
0.551
0.190
0.229
0.0003
0.383
0
0
0
0.219
0.188
0
0
0
0.022
0.163
0.208
0
0
0
0.0001
0.127
0.189
0
0.170
0
0
0
0.083
0.209
0.162
0.002
0
0.00009
0
0
0
0
0
0
0
0.157
0.142
0.179
0.001
0.200
0
0
0
52
55
43
1
8
3
36
129
97
102
92
35
109
47
14
22
16
61
59
113
60
86
28
10
21
8
2
60
88
18
65
5
30
84
131
13
93
14
16
58
1
1
0
0
0
1
1
1
0
0
0
0
1
1
0
1
0
0
0
1
1
1
0
0
0
0
0
0
0
0
0
0
1
1
1
0
1
0
0
0
Total
0.156
0.216
1984
1050 Counties
Note: In New Jersey a few colored apprentices for life remained after an act to abolish slavery was passed on April 18,
1846. In the 1860 Census, these apprentices are classified as slaves. In Kansas, Nebraska and Utah slavery was still
permitted though they were not slave states.
31
Table A2: First Stage Regressions
Table 4, Model 3
Table 4, Model 2
Table 4, Model 1
Dependent Variable
Malaria Suitability
and
Slave/Population
1860
Slave/Population
1860
Land
Inequality
1860
Table 5, Models 3 and 4
Slave/Population Land
1860
Inequality
1860
0.0690***
(8.77)
0.0701***
(8.50)
-0.00395
(-0.83)
0.00214
(0.44)
0.248***
(9.39)
0.0704***
(7.93)
-0.00378
(-0.81)
0.00113
(0.21)
0.236***
(3.66)
0.000856
(0.30)
0.00178
(1.47)
-1.475***
(-2.67)
Tobacco Suitability
Cotton Suitability
-0.00410***
(-3.12)
-0.635***
(-8.43)
-0.00412***
(-3.13)
-0.612***
(-8.17)
0.00184
(1.50)
-1.709***
(-8.07)
-0.000190
(-0.22)
-0.00409***
(-3.14)
-0.573***
(-7.78)
Yes
1,833
Yes
1,833
Yes
1,833
Yes
1,833
Yes
1,833
0.674
0.674
0.339
*** p<0.01, ** p<0.05, * p<0.1. Robust t statistics in parentheses.
0.674
0.341
Population Density
Constant
State Dummies
Observations
R-squared
32
Table A3: Returns on Education, by Race, 2000
Dependent Variables:
Income (2000) Whites Only
Income (2000) Blacks Only
Estimation Method: OLS
Model 1
Model 2
High-School Diploma
0.683***
0.260***
(4.84)
(2.82)
0.596***
0.532***
(2.95)
(3.63)
1.646***
0.789***
(11.47)
(6.24)
1.890***
1.409***
(8.54)
(7.54)
1.601***
0.368***
(5.62)
(2.68)
0.0330***
0.0778***
(6.31)
(6.78)
0.0134***
0.0110
(4.42)
(1.49)
0.0109***
0.0267***
(6.13)
(11.03)
-0.0165
0.0235
(-0.52)
(0.47)
0.0410
0.0276
(1.55)
(0.59)
0.0388
-0.0720**
(1.36)
(-2.62)
6.680***
7.000***
(24.26)
(51.20)
Sample
All Counties
All Counties
Observations
3030
2714
R-squared
0.77
0.27
Some Years of College (no Bachelor)
Bachelor Degree
Post-Graduate Education (Master or PhD)
Employment Rate
Whites in Labor Force
Blacks in Labor Force
Median Age
North East Dummy
South Atlantic Dummy
South Central Dummy
Constant
*** p<0.01, ** p<0.05, * p<0.1. Robust t statistics in parentheses.
33
Table A4: Predicted Returns on Education, by Race, 2000
Variable
Obs
Mean
Std. Dev.
Min
Max
Estimated Returns Whites
3074
0.713
0.139
0.341
1.499
Estimated Returns Blacks
2799
0.365
0.170
0.016
1.415
Table A5: Channels of Transmissions: Descriptive Statistics
All Counties
Variable
Obs
Mean
Std. Dev.
Min
Max
Returns Blacks/Returns Whites in 2000
2799
0.511
0.224
0.0284
2.531
Racial Educational Inequality in 2000
3016
0.012
0.013
0.00013
0.141
Slave States Only
Variable
Obs
Mean
Std. Dev.
Min
Max
Returns Blacks/Returns Whites in 2000
1358
0.452
0.163
0.071
2.015
Racial Educational Inequality in 2000
1366
0.016
0.014
0.00062
0.073
Table A6: Test of Similarity in the Propensity Score
Block 1
Block 2
Block 3
Block 4
Block 5
Mean
Mean
Mean
Mean
Mean
Mean (0)
Mean (1)
Average Mean
0.169
0.144
0.160
0.324
0.312
0.319
0.496
0.509
0.502
0.670
0.706
0.695
0.858
0.907
0.904
Difference (M(0) – M(1))
-0.026
-0.011
-0.013
0.036
-0.049
Mean(0) – Mean(1) = 0
Null not Rej.
Null not Rej.
Null not Rej.
Null not Rej.
Null not Rej.
34
Table A7: P
Propensity Score Moddel
D
Dependent V
Variable: Sl
Slavery (binaary)
Soil PH
4.791
(00.92)
2.575***
(22.80)
1.723
(00.86)
Malaria Suitaability
Tobacco Suittability
Cotton Suitabbility
0.136
(00.50)
-880.51*
(-1.68)
Constant
Observations
1553
Figuree A1: Border Sample
35
Data Appendix
The following datasets have been used:
a) The Historical, Demographic, Economic, and Social Data: The United States, 1790-2000,
downloaded
from
the
Inter-University
Consortium
for
Political
and
Social
Studies
(http://www.icpsr.umich.edu/), is used for the following variables: Total slaves number in 1860, Total
population in 1860, and Size of farms in 1860.
b) The United States Decennial Census 2000, downloaded through the American Fact Finder
(http://factfinder.census.gov/home/saff/main.html?_lang=en), is used for the 2000 levels of:
Households income, Shares of population for each race, Total Population, Educational attainment,
Median age, Employment rate, and Civilian labor force participation.
c) Geographical dummy variables are constructed using the US Census regional classification.
d) Data on soil neutrality (pH) are collected from Harmonized World Soil Database (FAO
GeoNetwork), downloadable at http://www.fao.org/geonetwork/srv/en/metadata.show?id=37140.
e) Data on the index of environmental suitability for malaria falciparum is taken from the Malaria Atlas
Project, downloadable at http://www.map.ox.ac.uk/.
f) Data on soil suitability to cotton and tobacco are from FAO GAEZ, downloadable at
http://www.fao.org/nr/gaez/en/.
g) Data on presidential elections are from Clubb et al. (2006)
h) Data on education expenditure are from the National Centre for Education Statistics
(http://nces.ed.gov/) while data on property taxes are from the County and City Book (relevant years)
downloaded
from
the
Inter-University
Consortium
for
Political
and
Social
Studies
(http://www.icpsr.umich.edu/).
36