The Slave Trade and The Origins of Mistrust in Africa: Citation

The Slave Trade and the Origins of Mistrust in
Africa
Citation
Nunn, Nathan, and Leonard Wantchekon. 2011. “The Slave Trade and the Origins of Mistrust in
Africa.” American Economic Review 101 (7) (December): 3221–3252. doi:10.1257/aer.101.7.3221.
http://dx.doi.org/10.1257/aer.101.7.3221.
Published Version
doi:10.1257/aer.101.7.3221
Permanent link
http://nrs.harvard.edu/urn-3:HUL.InstRepos:11986331
Terms of Use
This article was downloaded from Harvard University’s DASH repository, and is made available
under the terms and conditions applicable to Other Posted Material, as set forth at http://
nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of-use#LAA
Share Your Story

The Harvard community has made this article openly available.
Please share how this access benefits you. Submit a story .
Accessibility
American Economic Review 101 (December 2011): 3221–3252
http://www.aeaweb.org/articles.php?doi=10.1257/aer.101.7.3221
The Slave Trade and the Origins of Mistrust in Africa†
By Nathan Nunn and Leonard Wantchekon*
We show that current differences in trust levels within Africa can

be traced back to the transatlantic and Indian Ocean slave trades.
Combining contemporary individual-level survey data with histori-
cal data on slave shipments by ethnic group, we find that individuals
whose ancestors were heavily raided during the slave trade are less
trusting today. Evidence from a variety of identification strategies
suggests that the relationship is causal. Examining causal mecha-
nisms, we show that most of the impact of the slave trade is through
factors that are internal to the individual, such as cultural norms,
beliefs, and values. (JEL J15, N57, Z13)
In a recent study, Nunn (2008) examines the long-term impacts of Africa’s slave
trade. He finds that the slave trade, which occurred over a period of more than
400 years, had a significant negative effect on long-term economic development.
Although the article arguably identifies a negative causal relationship between the
slave trade and income today, the analysis is unable to establish the exact causal
mechanisms underlying this reduced-form relationship.
In this article, we examine one of the channels through which the slave trade
may affect economic development today. Combining contemporary individual-level
survey data with historical data on slave shipments by ethnic group, we ask whether
the slave trade caused a culture of mistrust to develop within Africa. Initially, slaves
were captured primarily through state organized raids and warfare, but as the trade
progressed, the environment of ubiquitous insecurity caused individuals to turn on
others—including friends and family members—and to kidnap, trick, and sell each
other into slavery (Sigismund Wilhelm Koelle 1854; P. E. H. Hair 1965; Charles
Piot 1996). We hypothesize that in this environment, a culture of mistrust may have
evolved, which may persist to this day.
* Nunn: Department of Economics, Harvard University, 1805 Cambridge Street, Cambridge, MA 02138 (e-mail:
[email protected]); Wantchekon: Department of Politics (joint with Economics), New York University, 19
West Fourth Street, New York, NY 10012 (e-mail: [email protected]). We thank referees for comments
that substantially improved the paper. We are also grateful to Daron Acemoglu, Ralph Austen, Daniel Berkowitz,
Michael Bratton, Alejandro Corvalan, William Darity, Koen Deconinck, William Easterly, James Fenske, Patrick
Francois, Avner Greif, Joseph Henrich, Karla Hoff, Joseph Inikori, Joseph Miller, Petra Moser, Elisabeth Ndour,
Ifedayo Olufemi Kuye, Torsten Persson, John Thornton, Marijke Verpoorten, Warren Whatley, and Robert
Woodberry for valuable comments. We also thank seminar participants at Boston University, Cal-Tech, Colby
College, Columbia, Dalhousie University, Dartmouth College, Georgia Tech, Harvard Business School, Harvard,
LSE, MIT, Northwestern, Simon Fraser University, Stanford, UCL, UCLA, University of Michigan, UPF-CREI,
University of Alberta, University of British Columbia, Warwick University, Yale, ASSA Meetings, EHA meetings,
NBER Political Economy Program Meeting, SED Meetings, and Sieper’s SITE conference. Sayon Deb, Eva Ng,
and Katherine Wilson provided excellent research assistance.
†
To view additional materials, visit the article page at
http://www.aeaweb.org/articles.php?doi=10.1257/aer.101.7.3221.
3221
3222 THE AMERICAN ECONOMIC REVIEW DECEMBER 2011
Our hypothesis builds on the well-established insight from cultural anthropology

that in environments where information acquisition is either costly or imperfect,
the use of heuristic decision-making strategies, or “rules-of-thumb,” can be optimal
(Robert Boyd and Peter J. Richerson 1985; Boyd and Richerson 1995). These gen-
eral rules or beliefs about the “right” action in different situations save the individual
from the costs of acquiring information. Of course, these rules-of-thumb do not
develop in a vacuum but rather evolve according to which yield the highest payoff.
Our hypothesis is that in areas heavily exposed to the slave trade, norms of mistrust
towards others were likely more beneficial than norms of trust, and therefore they
would have become more prevalent over time. If these beliefs and norms persist,
then the relationship between such norms and the history of the slave trade may still
be felt today—almost 100 years after the slave trade has ended.
To test our hypothesis, we use data from the 2005 Afrobarometer survey to exam-
ine whether individuals belonging to ethnic groups that were heavily targeted by
the slave trade in the past are less trusting of others today. We find that individuals
belonging to ethnic groups that were most exposed to the slave trades exhibit lower
levels of trust in their relatives, neighbors, coethnics, and local government today.
This finding is consistent with the historical fact that by the end of the slave trade,
it was not uncommon for individuals to be sold into slavery by neighbors, friends,
and family members.
An alternative explanation for our finding is that more slaves were supplied by
ethnic groups that initially were less trusting, and that these lower levels of trust
continue to persist today. Alternatively, there may be other historical events, such
as formal colonial rule, that are correlated with the severity of the slave trade and
subsequent levels of trust. We pursue a number of strategies to determine whether
the correlations we uncover are, in fact, causal.
Our first strategy is to control for other forms of European influence, most notably
the period of formal colonial rule that followed the slave trade. We also control for
certain precolonial characteristics of ethnic groups, including initial prosperity and
political development. We find that controlling for these observable characteristics
has little effect on the estimated effect of the slave trade on trust.
Our second strategy is to use the recent insights from Joseph G. Altonji, Todd E.
Elder, and Christopher R. Taber (2005) to calculate how much greater the influence
of unobservable factors would need to be, relative to observable factors, to com-
pletely explain away the negative relationship between the slave trade and trust. We
find that the influence of unobservable factors would have to be between three and
11 times greater than observable factors. Therefore, it is unlikely that our estimates
can be fully attributed to unobserved heterogeneity.
Our third strategy uses the distance of ethnic groups from the coast at the time of
the slave trade as an instrument for the number of slaves taken. The unique history
of sub-Saharan Africa provides a basis for the instrument’s exogeneity. Prior to the
transatlantic and Indian Ocean slave trades, Africans were not engaged in overseas
external trade. Therefore, it is unlikely that closer proximity to the ocean had an
impact on trust, other than through the slave trade. Because distance from the coast
may be correlated with access and involvement in the early trade across the Sahara
Desert, we control directly for distance to Saharan routes and the cities involved in
the Saharan trade. We also control for a number of measures of European influence,
VOL. 101 NO. 7 NUNN AND WANTCHEKON: THE ORIGINS OF MISTRUST IN AFRICA 3223
and for the historical dependence on fishing, both of which are potentially correlated
with distance from the coast. The IV regressions produce estimates that are qualita-
tively identical to the OLS estimates.
To address the concern of whether the exclusion restriction is satisfied, we perform
a number of falsification tests that examine the reduced-form relationship between
distance from the coast and trust inside and outside of Africa. Within Africa, we
find a strong positive relationship between distance from the coast and trust. This
is expected, given our IV estimates. Places farther from the coast had fewer slaves
taken, and therefore exhibit higher levels of trust today. If distance from the coast
affects trust only through the slave trade (i.e., if our exclusion restriction is satis-
fied), then there should be no relationship between distance from the coast and trust
outside of Africa, where there was no slave trade. This is exactly what we find.
Looking at samples from Asia and Europe, we estimate a statistically insignificant
relationship between distance from the coast and trust.
After establishing that the slave trade adversely affected trust, we turn to the task
of distinguishing between channels of causality. One mechanism, which is the arti-
cle’s focus, is that the slave trade altered the cultural norms of the ethnic groups
exposed to it, making them less trusting of others. However, there is also a second
potential channel. Because the slave trade resulted in a long-term deterioration of
legal and political institutions, the residents of heavily affected regions may now
be able to cheat others more easily. Individuals may be less trusting today because
those around them are less trustworthy.
We undertake three exercises to identify the relative importance of these channels.
First, we consider the determinants of respondents’ trust in their local government
and examine how the estimated effect of the slave trade changes when we control
for measures of individuals’ perceptions about the trustworthiness of their local gov-
ernment. After doing this, the estimated coefficient for slave exports decreases by
slightly less than 50 percent but remains precisely estimated, highly significant, and
very stable across specifications.
In the second exercise, rather than controlling for perceived trustworthiness, we
directly control for the effects of the slave trade on the trustworthiness of others. We
again estimate the determinants of intergroup trust, but this time we control directly
for the impact of the slave trade on the other ethnic groups living in the same loca-
tion as the respondent. Our estimates show that ethnic groups whose ancestors were
heavily enslaved in the past are less trusted today. This is consistent with the slave
trade’s adversely affecting the trustworthiness of individuals today. We find that the
estimated effect of the slave trade on internal norms of trust remains robust, even
after we control for the effect of the slave trade on the trustworthiness of others.
Our final strategy decomposes the effect of the slave trade into two channels: its
effect on factors internal to the individual, and its effect on factors external to the
individual. We do this by constructing a second measure of slave exports: the aver-
age number of slaves taken from the geographic location where the individual lives
today. This is different from our baseline measure, which is the average number
of slaves taken from an individual’s ethnic group. The logic behind including both
measures in our equation derives from the fact that when individuals relocate, their
internal beliefs move with them, even though their external environment changes.
Therefore, the two variables distinguish between the effects of the slave trade on trust
working through internal factors that are geographically mobile—such as individu-

als’ internal beliefs and values—versus through external factors that are less geo-
graphically mobile, including political, legal, institutional, and social structures. If
the slave trade primarily affects trust through internal beliefs and values, then across
individuals, what should matter is whether their ancestors were heavily affected by
the slave trade. If instead the slave trade affects trust primarily through its deteriora-
tion of institutions, social structures, or other factors external to the individual, then
what should matter is whether the individual’s external environment was heavily
affected by the slave trade. Our estimates show that both channels are important, but
that the internal channel is at least twice as large as the external channel.
These results complement recent studies documenting the importance of trust: for
economic development (Stephen Knack and Philip Keefer 1997; Marcel Fafchamps
2006; Guido Tabellini 2007; Yann Algan and Pierre Cahuc 2010); for international
trade (Avner Greif 1989; Luigi Guiso, Paola Sapienza, and Luigi Zingales 2007a);
for political institutions (Robert Putnam 2000); and for firm management practices
(Nicholas Bloom, Raffaella Sadun, and John Van Reenen 2008). Given the mount-
ing evidence of the importance of trust, our contribution here is in helping to under-
stand and explain its origins. Our evidence most directly complements those few
studies that also consider the historical determinants of differences in cultural norms
of behavior, such as Guiso, Sapienza, and Zingales’s (2007b) study that empirically
links differences in social capital within Italy to the independence of cities during
the eleventh to fourteenth centuries. Our work also complements Tabellini’s (2007)
study which shows that the levels of education and democracy in eighteenth century
Europe are important determinants of interpersonal trust today.
Our focus on the long-term historical determinants of cultural norms is not meant
to suggest that short-run determinants are unimportant. There is substantial evidence
that nonhistorical determinants of trust—for example income, education, informa-
tion flows, organization membership, and current experiences—are also important
(Raymond Fisman and Tarun Khanna 1999; Alberto Alesina and Eliana La Ferrara
2002; John Bellows and Edward Miguel 2008).
In Section I, we begin our analysis by first laying out the historical and conceptual
groundwork. We discuss the theoretical literature that seeks to understand how and
why norms evolve, as well as historical literature describing the slave trade and the
environment of insecurity that it generated. In Section II, we turn to a description
of the data, before reporting our estimates in Sections III and IV. In Section V, we
examine specific mechanisms and test whether the slave trade affects trust through
internal cultural norms or through societies’ institutional and legal structures, which
in turn affect the trustworthiness of its citizens. Section VI concludes.
I. Historical Background and Conceptual Framework
A. Historical Background
Early in the slave trade, nearly all slaves were taken in large-scale conflicts or raids,
which created an environment of extreme insecurity outside of the local community
(Mario Azevedo 1982; Andrew Hubbell 2001; Joseph E. Inikori 2000; Martin Klein
2001). Ironically, this in turn caused insecurity within communities, as individuals
began to turn on others close to them, including neighbors, friends, and even fam-
ily. Unlike most other environments of conflict and insecurity, the slave trade had
one unique feature: individuals could partially protect themselves by turning against
others within their community.1 By engaging in trickery, local kidnappings, or other
forms of small scale violence, one could exchange slaves (with Europeans, or slave
merchants) for guns and iron weapons (Abdullahi Mahadi 1992; Walter Hawthorne
1999). Slave merchants and raiders also played a role in promoting internal conflict,
often forming strategic alliances with key groups within villages and states in order
to extract slaves (Boubacar Barry 1992; Inikori 2003; Klein 2003).
In his book Planting Rice and Harvesting Slaves, Walter Hawthorne documents
the decentralized and interpersonal nature of slave capture among the Beafares of the
Guinea-Bissau region of Africa. He writes that “the Atlantic slave trade was insidi-
ous because its effects penetrated deep into the social fabric of the Upper Guinea
Coast—beyond the level of the state and to the level of the village and household …
Hence, in many areas, the slave trade pitted neighbor against neighbor” (Hawthorne
2003, pp. 106–107). Hawthorne provides a particularly telling example, taken from
André Alvarez d’Almada (1984). Households located near ports were able to profit
from the slave trade by “tricking” unsuspecting strangers and then selling them to
merchants. Almada writes that “these Beafares are so smart, that if a yokel arrives
from the interior, they pretend that they want to give him shelter, and they receive
him into their homes. After a few days have passed, they persuade him that they have
friends on the ships, and that they would like to take him and have a party. But when
they go to the ships, they sell him. In this way they trick many yokels” (Almada
1984, p. 121).2
Data on the manner of enslavement in the nineteenth century confirm the descrip-
tive evidence: by the end of the slave trade, individuals entered slavery in a variety
of ways, including by being sold into slavery by acquaintances, friends, and fam-
ily. During the 1840s, German missionary and linguist Sigismund Koelle (1854)
collected information on the manner of enslavement of 144 former slaves living in
Free Town, Sierra Leone. In his sample, the most common manner of enslavement
was kidnapping, with just over 40 percent of the slaves being taken in this manner.
Just under 25 percent of the slaves were captured during wars. Amazingly, almost
20 percent of the slaves were sold by relatives or friends. Koelle’s interviews docu-
ment numerous accounts of individuals being sold into slavery by family members,
relatives, and “supposed friends.” One of the more notable accounts is of a slave
who was sold into slavery after being “enticed on board of a Portuguese vessel” by
“a treacherous friend.” Another example is the custom of the Kabre (from Northern
Togo), developed during the nineteenth century, of selling their own kin into slav-
ery (Piot 1996). The final process of enslavement was through the judicial system.
Sixteen percent of the Koelle sample entered slavery this way. This form of enslave-
ment includes the historically common practice of accusing others of crimes such
as witchcraft, theft, adultery, or murder in order to obtain slaves. If found guilty, the
1
One environment that arguably featured a similar dynamic to the slave trade was communist East Germany dur-
ing the Cold War. Because of the strategies employed by the Stasi, an individual’s best option often was to become
an informant, turning on others, even those close to him or her (Marcus Jacob and Marcel Tyrell 2010).
2
Also see the discussion in Hawthorne (2003).
accused (and often their family) were sentenced to slavery (Paul E. Lovejoy 2000).
It even became common for the leaders of local communities to obtain slaves in this
manner (Hawthorne 1999, 2003; Klein 2001).
The fact that slaves often were taken or tricked into slavery by individuals close to
them suggests that the slave trade may have eroded trust even in the most intimate
social relationships. Furthermore, because chiefs often were slave traders, or were
forced to sell their own people into slavery, the slave trade also may have engen-
dered a mistrust of political figures, particularly local leaders. Our analysis tests for
these effects, examining whether individuals whose ancestors were most heavily
threatened during the slave trade have less trust in their family, neighbors, and their
local government council.
B. Conceptual Framework
Our notion of culture is taken from research in evolutionary anthropology that

views culture as decision-making rules-of-thumb employed in uncertain or com-
plex environments. Using theoretic models, Boyd and Richerson (1985, 2005)
show that if information acquisition is either costly or imperfect, it can be optimal
for individuals to develop heuristics or rules-of-thumb in decision making. By
relying on general beliefs about the “right” thing to do in different situations, indi-
viduals may not behave in a manner that is optimal in every instance, but they save
on the costs of obtaining the information necessary to always behave optimally.
In these models, different behavioral rules evolve through a process of natural
selection determined by the relative payoffs from different rules-of-thumb. Within
this framework, the hypothesis we test is whether the environment of insecurity
caused by the slave trade increased the returns to rules-of-thumb based on mistrust
relative to rules-of-thumb based on trust, thus causing a culture of greater mistrust
to develop.
The natural question that we face, though, is why we expect to find evidence
of increased mistrust among the descendants of those exposed to the slave trade
100 years after its end. One explanation for the persistence of mistrust can be found
in the models developed by Guiso, Sapienza, and Zingales (2007b) and Tabellini
(2008); they show how multiple equilibria in cultural outcomes can arise. The long-
run effects that we find may be the result of permanent movements to equilibria
characterized by high levels of mistrust among the lineages most affected by the
slave trade.
Another explanation for the persistence of mistrust is rooted in the existence of
complementarities between cultural norms and domestic institutions. This phenom-
enon is highlighted by the model developed in Tabellini (2008). In the model, indi-
viduals inherit norms of cooperation from their parents and make political choices
(through voting) that determine the quality of domestic institutions. Through this
mechanism, norms of cooperation will affect the equilibrium quality of domestic
institutions. When there is a negative shock to internal norms of cooperation, the
next generation will not only be less trusting, but also will choose institutions with
weaker enforcement, resulting in poor behavior and low levels of trust among future
generations. This creates a self-enforcing outcome by which low levels of trust and
weak institutions persist among future generations. When applied to Africa, the
model suggests that areas with low levels of trust have developed weaker institutions,
and the weaker institutions in turn have resulted in worse behavior and still lower
levels of trust. These societies remain trapped in an equilibrium of uncooperative
behavior, mistrust, and inefficient institutions.
It is also possible that our results arise not because African societies remain
trapped in low trust equilibria, but because the shocks caused by the slave trade—
an event lasting for over 400 years—have not yet fully dissipated. This expla-
nation is consistent with the dominant presumption that cultural change occurs
slowly (e.g., Alberto Bisin and Thierry Verdier 2000; Bisin and Verdier 2001,
2008). Rare evidence of the speed of cultural change is provided by Alesina and
Nicola Fuchs-Schündeln’s (2007) study of the effects of the division of Germany
between 1945 and 1990 on individuals’ beliefs about the benefits of redistribu-
tion and government intervention. They find that East Germans view government
intervention more favorably than West Germans and that since reunification the
beliefs of East Germans have begun to slowly converge to those of West Germans.
Although this particular shock lasted only 45 years, the authors estimate that the
differences generated by the shock will take 20–40 years to diminish to zero. In
that case, the effects of the shock will have persisted for nearly as long as the
shock itself. In our study, the negative shock lasted well over 400 years, and we
are examining its effects just over 100 years later. Therefore, it is possible that the
effects of the shock on mistrust are still being felt today, even if they are actually
temporary and ultimately will die out in the long run.
Additional evidence can also be gleaned from the experiments of Karla Hoff,
Mayuresh Kshetramade, and Ernst Fehr (2009), which show that the legacy of the
caste system within India continues to manifest itself over half a century after its
abolition, and that this finding is robust to the inclusion of controls for education
and wealth. They find that individuals from low castes are less willing to sanction
violations of a cooperation or reciprocity norm than individuals from a high-caste
background.3
A fourth explanation is also possible. The mechanism may not be the result of cul-
tural learning, as in the previous three explanations. Instead, it is (at least theoreti-
cally) possible that the persistent change in cultural norms arises because a greater
number of inherently more trusting individuals were captured and shipped from
the continent, leaving a greater proportion of inherently less trusting individuals.
Therefore, even without any cultural learning, the slave trade could still have had a
large effect on mistrust today. This explanation requires that mistrust be an inherent
or hard-wired characteristic, which would occur, for example, if trust were geneti-
cally determined.4
Our analysis is not able to distinguish between these finer transmission mech-
anisms. Instead, our aim and intended contribution is to empirically estimate the
long-term causal effects of the slave trade on trust, and to identify the extent to
which this relationship arises because of the slave trade’s effects on norms, beliefs,
3
For related evidence of the cultural legacy of the caste system in India and its potential effects on trust, see Hoff
and Priyanka Pandey (2005).
4
A number of recent studies provide suggestive evidence that genetics may be a determinant of trust, as well as
other behavioral characteristics. See, for example, David Cesarini et al. (2008).
and rules-of-thumb, all of which are internal to the individual and transmitted from
parents to children over time.
II. Data Sources and Description
A. Afrobarometer Data
The individual-level data are from the 2005 Afrobarometer surveys. These
nationally representative surveys are based on interviews conducted in the local
languages of a random sample of either 1,200 or 2,400 individuals of voting age
in each country. The surveys cover 17 sub-Saharan African countries: Benin,
Botswana, Ghana, Kenya, Lesotho, Madagascar, Malawi, Mali, Mozambique,
Namibia, Nigeria, Senegal, South Africa, Tanzania, Uganda, Zambia, and
Zimbabwe. The sample countries are concentrated in West Africa, Eastern Africa,
and Southern Africa. West Central Africa is not included (e.g., Cameroon, Gabon,
Democratic Republic of Congo, Angola), nor are countries inland of the Red Sea
(e.g., Sudan, Ethiopia, Eritrea).
From the surveys we have a potential sample of 21,822 respondents. Within
this sample, 120 of the respondents: (i) list “other” as their ethnicity; (ii) list their
country as their ethnicity; (iii) belong to an ethnic group that is not an indigenous
Africa ethnicity; or (iv) list an indigenous ethnicity that could not be matched
cleanly to the slave trade data. Removing these observations leaves us with 21,702
potential observations.
The Afrobarometer asks respondents how much they trust their relatives, neigh-
bors, and their locally elected government council. It also asks about trust of those
in the same country from other ethnic groups, and those from the same ethnic group.
The respondents choose between four possible answers: not at all, just a little, some-
what, or a lot. The distributions of responses for each question are reported in the
paper’s online Appendix. Not surprisingly, the data show that individuals exhibit
more trust in those closer to them than in those further from them. Despite this, 7
percent still report that they do not trust their relatives at all, and 18 percent report
that they trust their relatives only a little.
Since respondents’ answers to the trust questions are categorical, there are a
number of possible estimation strategies. The first is to convert the categorical
responses into a variable that assigns a number to each response. Following this
strategy, we construct a measure of trust that takes on the value of 0, 1, 2, or 3:
0 corresponds to the response “not at all”; 1 to “just a little”; 2 to “somewhat”;
and 3 to the response “a lot.” (These are the numeric values assigned to each
answer in the Afrobarometer survey.) An alternative strategy is to maintain the
categorical nature of the answers and instead estimate an ordered logit model.
As we discuss below, the estimates are qualitatively identical if we pursue this
alternative strategy.
B. Ethnicity-Level Data on Slave Exports
The estimates of the number of slaves taken from each ethnic group rely on
country-level slave export figures from Nunn (2008). They were constructed by
combining data on the total number of slaves shipped from all ports and regions of
Africa with data on the slaves’ ethnic identities. The country-level estimates cover
Africa’s four slave trades (the transatlantic, Indian Ocean, Red Sea, and trans-Saha-
ran) between 1400 and 1900.5
We disaggregate the country-level slave export figures to the ethnicity level
using the same ethnicity samples as Nunn (2008). Since only two of the four slave
trades—the transatlantic and Indian Ocean—have ethnicity data detailed enough to
construct reliable estimates of the number of slaves taken from each ethnicity, our
empirical analysis is restricted to the transatlantic and Indian Ocean slave trades.
Since the transatlantic slave trade was by far the largest of the slave trades, the omis-
sion of the Red Sea and trans-Saharan slave trades likely will have little impact.
Nunn (2008) has previously shown that the impact of the slave trades as a whole is
driven almost solely by the transatlantic slave trade. Furthermore, the countries that
were most affected by the Red Sea and trans-Saharan slave trades—namely Sudan,
Ethiopia, and Chad—are not included in the Afrobarometer sample. We also show
that our results are robust to omitting observations from the two countries in our
sample—Kenya and Mali—that shipped a significant number of slaves during either
the trans-Saharan or Red Sea slave trades.
For the transatlantic slave trade, Nunn (2008) has collected a sample of
80,656 slaves whose ethnic identity is known. The aggregate sample comprises
54 different samples that report 229 distinct ethnic designations. For the Indian
Ocean slave trade, Nunn’s aggregate sample has 21,048 slaves, covering 80 distinct
ethnic groups.
One important step in estimating the number of slaves taken from each eth-
nic group is matching the ethnic identities in the historical records to the ethnic
classification in the Afrobarometer surveys. We achieve this by first linking the
original ethnic groups to a classification that is constructed and mapped by George
Peter Murdock (1959).6 Since Murdock’s classification is similar to the one used
in the Afrobarometer surveys, it is easy to move from his classification to the
Afrobarometer data.
Figures 1A and 1B map the historical boundaries (in the late nineteenth century)
according to Murdock (1959). To provide a visual representation of the spatial dis-
tribution of each slave trade, we have grouped total slave exports between the years
1400 and 1900 into five broad categories, and we denote greater numbers of slaves
shipped with darker shades.
As Figure 1A shows, the transatlantic slave trade affected much of the African
continent. Slaves were not only taken from West Africa and West-Central Africa,
but also from Eastern Africa and Madagascar. The much smaller Indian Ocean slave
trade was confined primarily to Eastern Africa (see Figure 1B). These patterns of
enslavement are consistent with the qualitative evidence on the sources of slaves
taken during the two slave trades (e.g., Patrick Manning 1990; Lovejoy 2000).
5
Full details of the underlying data, their sources, and the construction procedure are provided in Nunn (2008).
6
The authors of the secondary sources, from which much of the data are taken, typically provide a detailed
analysis of the meanings and historical locations of the ethnic groups identified in the historical documents. This
information greatly facilitated the matching.
Panel A. Transatlantic slave trade
Atlantic slave exports

0
1 – 100,000
100,001 – 500,000
500,001 – 1,000,000
1,000,001 – 4,000,000
Panel B. Indian Ocean slave trade
Indian slave exports

0
1 – 1,000
1,001 – 50,000
50,001 – 100,000
100,001 – 1,000,000
Figure 1
Note: Maps display the total number of slaves of each ethnicity shipped during the transatlantic and Indian Ocean
slave trades.
III. Estimating Equations and Empirical Results
A. OLS Estimates
We begin by estimating the relationship between the number of slaves that were
taken from an individual’s ethnic group and the individual’s current level of trust.
Our baseline estimating equation is:
(1) trusti, e, d, c = αc + β slave exportse + X′i,e,d,c Γ + X′d,c Ω + X e′ Φ + εi,e,d,c,
where i indexes individuals, e ethnic groups, d districts, and c countries. The vari-
able trusti, e, d, c denotes one of our five measures of trust, which vary across individu-
als. αc denotes country fixed effects, which are included to capture country-specific
factors, such as government regulations, that may affect trust (e.g., Philippe Aghion
et al. 2010; Aghion, Algan, and Cahuc 2008). slave exportse is a measure of the
number of slaves taken from ethnic group e during the slave trade. (We discuss this
variable in more detail below.) Our coefficient of interest is β, the estimated relation-
ship between the slave exports of an individual’s ethnic group and the individual’s
current level of trust.
The vector X′i, e, d, c denotes a set of individual-level covariates, which include the
respondent’s age, age squared, a gender indicator variable, an indicator variable
that equals one if the respondent lives in an urban location, five fixed effects for the
respondent’s living conditions, ten fixed effects for the educational attainment of the
respondent, 18 religion fixed effects, and 25 occupation fixed effects.7 Many of the
controls are intended to proxy for individual income, which has been shown to be
correlated with trust. Although we do not have a direct measure of income, occupa-
tion, education, and living conditions are all very good proxies.8
The vector X ′d consists of two variables designed to capture the ethnic composi-
tion of the district in which the respondent lives.9 The first variable is the ethnic frac-
tionalization of the respondent’s district. Previous studies, such as William Easterly
and Ross Levine (1997), have documented a relationship between ethnic fraction-
alization and income. Perhaps through this channel, the ethnic fractionalization of
a respondent’s location may affect the respondent’s trust.10 Second, we also control
for the share of the district’s population that is of the same ethnicity as the respon-
dent.11 When respondents are part of an ethnic minority, they may be less trusting of
others; Alesina and La Ferrara (2002) find evidence of this within the United States.
Both of our measures of ethnic composition are constructed using the sample of
individuals in the Afrobarometer survey.12
7
A full description of these controls is provided in the online Appendix.
8
Occupation, as well as proxying for income, may be an important determinant of trust itself. Patrick Francois,
Thomas Fujiwara, and Tanguy van Ypersele (2010) provide evidence showing that within the United States, indi-
viduals who work in more competitive sectors have higher levels of trust.
9
A district is the level of disaggregation finer than a region/province and coarser than a village. The sample
includes 1,292 districts.
10
Ethnic fractionalization is constructed in the standard manner. See Easterly and Levine (1997) for details.
11
This measure actually varies at the district and ethnicity level. As a result, there is a slight abuse of notation in
equation (1) in our use of X ′d, c to denote the two variables.
12
For the average respondent, 48 other individuals in the survey live in the same district.
Table 1—OLS Estimates of the Determinants of Trust in Neighbors
Slave Exports/ ln (1 + ln (1 +
Dependent variable: exports Exports/ historical ln (1 + exports/ exports/
Trust of neighbors (thousands) area pop exports) area) historical pop)
(1) (2) (3) (4) (5) (6)
Estimated coefficient −0.00068 −0.019 −0.531 −0.037 −0.159 −0.743
[0.00014] [0.005] [0.147] [0.014] [0.034] [0.187]
(0.00015) (0.005) (0.147) (0.014) (0.034) (0.187)
{0.00013} {0.005} {0.165} {0.015} {0.034} {0.212}
Individual controls Yes Yes Yes Yes Yes Yes
District controls Yes Yes Yes Yes Yes Yes
Country fixed effects Yes Yes Yes Yes Yes Yes
Number of observations 20,027 20,027 17,644 20,027 20,027 17,644
Number of ethnicities 185 185 157 185 185 157
Number of districts 1,257 1,257 1,214 1,257 1,257 1,214
R2 0.16 0.16 0.15 0.15 0.16 0.15
Notes: The table reports OLS estimates. The unit of observation is an individual. Below each coefficient
three standard errors are reported. The first, reported in square brackets, is standard errors adjusted for
clustering within ethnic groups. The second, reported in parentheses, is standard errors adjusted for two-
way clustering within ethnic groups and within districts. The third, reported in curly brackets, is T. G.
Conley (1999) standard errors adjusted for two-dimensional spatial autocorrelation. The standard errors
are constructed assuming a window with weights equal to one for observations less than five degrees apart
and zero for observations further apart. The individual controls are for age, age squared, a gender indica-
tor variable, five living conditions fixed effects, ten education fixed effects, 18 religion fixed effects, 25
occupation fixed effects, and an indicator for whether the respondent lives in an urban location. The dis-
trict controls include ethnic fractionalization of each district and the share of the district’s population that
is the same ethnicity as the respondent.
The vector X ′e denotes a vector of ethnicity-level variables that are meant to cap-
ture the historical characteristics of ethnicities, as well as the differing impacts of
colonial rule on separate ethnic groups. They are important controls for our analysis,
and we discuss them as they are introduced.
Estimates of equation (1), with trust measured by individuals’ trust in their neigh-
bors, are reported in Table 1. In the first column, we use the total number of slaves
taken from an ethnic group (expressed in thousands of people) as our measure of the
intensity of the slave trade. The estimated coefficient for slave exports, β, is negative
and statistically significant. This is consistent with the hypothesis that the slave trade
adversely affected individuals’ trust of those around them. Because the distribution
of the exports is highly left skewed, with a small number of observations taking
on large values, in column 2, we report estimates using the natural log of the slave
export measure. The results are similar: we continue to find a significant negative
correlation between slave exports and trust.13
Many of the explanatory variables in equation (1) do not vary across individu-
als. Rather, they vary at either the ethnicity level (e.g., slave exportse and X ′e) or
the district level (e.g., X ′d). Given the potential for within-group correlation of the
residuals, we adjust all standard errors for potential clustering. In Table 1, we report
in square brackets standard errors adjusted for clustering of observations of the same
13
To conserve on space, we do not report the coefficient estimates of the control variables throughout the paper.
The estimates generally are in agreement with the findings from previous studies. Consistent with Alesina and La
Ferrara’s (2002) findings from a US sample, trust is increasing at a decreasing rate in age and is higher for males
than for females.
ethnicity. We also calculate standard errors and report them in parentheses, adjusted
for two-way clustering within ethnic groups and within districts.14 A third strategy
is to calculate Timothy Conley (1999) standard errors adjusted for two-dimensional
spatial dependence. These are reported in curly brackets in Table 1. These three
methods all produce standard errors that are essentially identical. For the remainder
of the article, we report standard errors adjusted for two-way clustering within eth-
nic groups and districts.
The estimates reported in column 1 use the total number of slaves as a measure
of the impact of the slave trade. One shortcoming of the measure is that it does not
account for differences in the size of ethnic groups. Column 2 reports estimates
using an alternative slave export measure that normalizes the number of slaves taken
by the area of land inhabited by the ethnic group during the nineteenth century. The
results are similar using this alternative slave export measure.
Ideally, we would prefer to use a measure of slave exports that is normalized by
the population of each ethnic group prior to the slave trade. Unfortunately, these
data are unavailable. Some historical population data are available from Murdock
(1959), but they are from the colonial period (approximately the early twentieth
century) after the end of the slave trade, and they exist for only about 85 percent of
the ethnicities in the sample. Column 3 reports estimates normalizing slave exports
using these colonial population figures. We obtain similar estimates using this alter-
native measure. Columns 4–6 report estimates using the natural log of one plus the
normalized slave export measures from columns 1–3. Again, this is done to reduce
the skewness in the slave export variables. The results remain robust to this alterna-
tive specification.
For the remainder of the analysis, we use, as our baseline measure, the natural log
of one plus slave exports normalized by land area (the specification from column 5).
This provides a measure that is normalized by the size of ethnic groups and uses a
denominator that is precisely measured and available for all ethnic groups in our sam-
ple. However, as Table 1 illustrates, the results of the article do not rest on this choice.
We now turn to the other measures of trust. Table 2 reports OLS estimates for all
five trust measures. The estimates show that the slave trade is negatively correlated
with all five measures of trust, including intragroup trust and trust of relatives. This
is consistent with the historical evidence: that the effects of the slave trade pene-
trated deep into the social fabric of societies and eventually turned friends, families,
and neighbors against each other.
Not only are the negative coefficient estimates of Table 2 statistically significant,
but they are also economically meaningful. To see this, first note that the standard
deviation of our baseline slave export variable is close to one (0.95).15 Also, the
standard deviation of each trust measure is close to one, ranging from 0.96 to 1.10.
Therefore, the reported coefficients are close to standardized “beta” coefficients,
which report the number of standard deviation changes in the dependent variable
for a one–standard deviation change in the independent variable. As we have seen,
the coefficients for slave exports (for the full sample) range from −0.10 to −0.16.
14
See Colin Cameron, Jonah Gelbach, and Douglas Miller (2006) for details on multiway clustering.
15
Summary statistics are reported in the online Appendix.
Table 2—OLS Estimates of the Determinants of the Trust of Others
Trust Trust Trust of Intra- Inter-

of of local group group
relatives neighbors council trust trust
(1) (2) (3) (4) (5)
ln (1 + exports/area) −0.133*** −0.159*** −0.111*** −0.144*** −0.097***
(0.037) (0.034) (0.021) (0.032) (0.028)
Individual controls Yes Yes Yes Yes Yes
District controls Yes Yes Yes Yes Yes
Country fixed effects Yes Yes Yes Yes Yes
Number of observations 20,062 20,027 19,733 19,952 19,765
Number of ethnicity clusters 185 185 185 185 185
Number of district clusters 1,257 1,257 1,283 1,257 1,255
R2 0.13 0.16 0.20 0.14 0.11
Notes: The table reports OLS estimates. The unit of observation is an individual. Standard
errors are adjusted for two-way clustering at the ethnicity and district levels. The individ-
ual controls are for age, age squared, a gender indicator variable, five living conditions fixed
effects, ten education fixed effects, 18 religion fixed effects, 25 occupation fixed effects, and
an indicator for whether the respondent lives in an urban location. The district controls include
ethnic fractionalization in the district and the share of the district’s population that is the same
ethnicity as the respondent.
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
* Significant at the 10 percent level.
An alternative way to assess the magnitude of the slave export coefficients is to

compare their explanatory power against other variables in the regression. To do
this, we compare the slave exports variable with all other explanatory variables in
the estimating equation (other than the country fixed effects). Using the estimates
from column 5 of Table 1, and performing a standard variance decomposition, we
find that slave exports and the other covariates together explain 5.4 percent of the
total variation of trust in neighbors. Of this 5.4 percent, 16–27 percent is explained
by slave exports.
We undertake a number of robustness and sensitivity checks which we describe
only briefly here. The details are reported in the online Appendix. Because we have
estimates for only the transatlantic and Indian Ocean slave trades, we verify that our
results are not biased by the omission of slaves exported during the trans-Saharan
and Red Sea slave trade. We also check our results for robustness to the omission
of respondents living in Kenya and Mali, the two countries in our sample that were
strongly affected by the trans-Saharan or Red Sea slave trades. Removing the two
countries results in point estimates that are nearly identical to the baseline estimates.
Finally, we check for robustness to alternative estimation methods. Using an
ordered logit model produces estimates that are qualitatively identical to our base-
line OLS estimates. Similarly, estimating versions of equation (1) where the unit
of observation is an ethnicity rather than an individual produces similar results.
The individual-level estimating equation, which is our baseline specification, has
a number of advantages. First, it allows us to explicitly control for individual-level
characteristics, which result in more precise estimates of β. Second, the finer unit
of observation is necessary to test for the causal mechanisms (see Section V). The
tests require variation across individuals and are not possible at the ethnicity level.
IV. Identifying Causal Relationships
The negative correlation between slave exports and trust that is documented in the
previous section is consistent with our hypothesis that the slave trade engendered a
culture of mistrust. However, the correlation could also be explained by omitted vari-
ables that are correlated with selection into the slave trade and with subsequent trust.
For example, if ethnic groups that were inherently less trusting were more likely to
be taken during the slave trades, and if these groups continue to be less trusting today,
then this could generate a negative relationship between the slave trade and trust.
In this section, we pursue three strategies to assess whether the correlations
documented to this point are causal. First, we control for observable characteris-
tics of ethnic groups that may be correlated with the slave trade and subsequent
trust. Second, we use selection on observable variables to assess the likelihood that
our estimates are being driven by unobserved heterogeneity across ethnic groups.
Finally, we use the historical distance from the coast of an individual’s ethnic group
as an instrument for slave exports.
A. Controlling for Observables: Initial Conditions and Colonial Rule
Within the historical context of Africa, the most important potentially omitted fac-
tor is colonial rule, which followed the slave trade and lasted from 1885 until inde-
pendence. If the parts of Africa that were most affected by the slave trade were also
the most affected by colonial rule, then not controlling for colonial rule might lead
to falsely attributing its effects to the slave trade. Therefore, we control for a num-
ber of ethnicity-level variables that are intended to capture subnational variation in
colonial rule and its determinants.16 We specifically follow Daron Acemoglu, Simon
Johnson, and James A. Robinson (2001, 2002), who put forth two primary deter-
minants of the type of institutions implemented during colonial rule: the deadliness
of the disease environment for early European settlers and precolonial prosperity.
We measure an ethnic group’s initial disease environment using the malaria ecol-
ogy of the land it inhabited. The underlying data are from the Malaria Stability
Index constructed by Anthony Kiszewski et al. (2004). The index takes into account
the prevalence and type of mosquitoes indigenous to a region, their human biting
rate, their daily survival rate, and their incubation period. It has been constructed
for 0.5-degree-by-0.5-degree grid-cells globally. Combining the malaria index and
the digitized map of historical ethnic boundaries, we construct a measure of average
malaria presence in land historically inhabited by each ethnic group.17
We also construct measures of precolonial prosperity, which Acemoglu, Johnson,
and Robinson (2002) argue affected the strategies undertaken by the colonizers. Our
ideal measure of initial prosperity would be precolonial population density, or urban-
ization rates measured at the ethnicity level. Unfortunately, no such data exist. The
earliest period for which systematic population data are available (from Murdock
16
All estimates include country fixed effects. Since colonial boundaries are nearly identical to current country
boundaries, our estimates already control for any effects of colonial rule that vary at the national level.
17
Although the malaria transmission index is taken from contemporary data, it likely provides a close approxi-
mation to historical conditions. This is because the indicators it is based on (prevalence and type of mosquitoes,
including their biting rates, within Africa) have not changed drastically over time.
1959) is from the colonial period, approximately the early twentieth century. We use
this to construct an ethnicity-level measure of colonial population density. However,
in addition to not being measured in the precolonial period, the variable suffers from
a second shortcoming: the data are missing for about 15 percent of the ethnic groups
in the sample.
Given these shortcomings, we also construct additional measures of precolonial
prosperity. The first exploits information on the locations and sizes of urban cen-
ters. Using data from Tertius Chandler (1987) on the location of African cities with
more than 20,000 inhabitants in year 1400, we construct an indicator variable that
equals one if there was a city located on the land inhabited by each ethnic group.
This provides an indicator of ethnic groups that were densely populated prior to the
slave trade.
We also use historical data from the Ethnographic Atlas to construct two addi-
tional proxies for initial levels of prosperity. The first is a set of indicator variables
that quantify the precolonial settlement patterns of ethnic groups. These variables
identify whether ethnic groups were fully nomadic (migratory), seminomadic,
semisedentary, lived in compact and impermanent settlements, in neighborhoods of
dispersed family homes, in separated hamlets forming a single community, in com-
pact and relatively permanent settlements, or in complex settlements. The categories
are listed in order of increasing economic and social development and are, almost
by definition, increasing in initial population density. The second variable that we
construct from the Ethnographic Atlas is the number of jurisdictional hierarchies
beyond the local community. This measures the sophistication of an ethnic group’s
political institutions.
Finally, we construct a number of ethnicity-level variables that directly measure
European influence during the colonial period. Using information on the location
of railway lines in the first decade of the twentieth century from Century Company
(1911), we construct an indicator variable that equals one if any part of the railway
network was built on land historically inhabited by the ethnic group. This is meant
to proxy for whether ethnic groups were historically connected to the colonial rail-
way networks. Using the same source, we construct an indicator variable that equals
one if a European explorer traveled through land historically occupied by the ethnic
group.18 Third, we construct a variable to capture European missionary contact dur-
ing the colonial period. Using information on the historical location of missions in
the early twentieth century from William R. M. Roome (1924), we calculate the
number of missions per square kilometer for each ethnic group.
Our intention is that by controlling for this extensive set of covariates, we cap-
ture any potential effects of non–slave trade European influence on long-term trust.
Estimates of equation (1) controlling for the additional controls are reported in Table
3. For each measure of trust, the estimated slave export coefficients remain negative
and highly significant.19
18
The variable captures exploration routes between 1768 and 1894.
19
Because of missing colonial population density data, the sample is reduced slightly when this variable is
included as a covariate. Estimates using the larger sample when this variable is excluded are very similar. These are
reported in the online Appendix.
Table 3—OLS Estimates of the Determinants of the Trust of Others,

with Additional Controls
Trust Trust Trust of

of of local Intragroup Intergroup
(1) (2) (3) (4) (5)
ln (1 + exports/area) −0.178*** −0.202*** −0.129*** −0.188*** −0.115***
(0.032) (0.031) (0.022) (0.033) (0.030)
Colonial population density Yes Yes Yes Yes Yes
Ethnicity-level colonial controls Yes Yes Yes Yes Yes
Number of ethnicity clusters 147 147 146 147 147
Number of district clusters 1,187 1,187 1,194 1,186 1,184
R2 0.13 0.16 0.21 0.16 0.12
Notes: The table reports OLS estimates. The unit of observation is an individual. Standard
errors are adjusted for two-way clustering at the ethnicity and district levels. The individ-
ual controls are for age, age squared, a gender indicator variable, five living conditions fixed
effects, ten education fixed effects, 18 religion fixed effects, 25 occupation fixed effects, and
an indicator for whether the respondent lives in an urban location. The district controls include
ethnic fractionalization in the district and the share of the district’s population that is the same
ethnicity as the respondent. Ethnicity-level colonial controls include the prevalence of malaria,
a 1400 urbanization indicator variable, eight fixed effects for the sophistication of precolonial
settlement, the number of jurisdictional political hierarchies beyond the local community in the
precolonial period, an indicator for integration with the colonial rail network, an indicator for
contact with precolonial European explorers, and the number of missions per square kilometer
during colonial rule. Colonial population density is the natural log of an ethnicity’s population
density during the colonial period.
B. Using Selection on Observables to Assess the Bias from Unobservables
Despite our attempts to control for observable factors, such as initial prosperity
and the impacts of colonial rule, the estimates reported in Table 3 may still be biased
by unobservable factors correlated with selection into the slave trade and subse-
quent trust. In this section, we assess the likelihood that the estimates are biased by
unobservables.
The strategy that we use exploits the insight from Altonji, Elder, and Taber
(2005) that selection on observables can be used to assess the potential bias from
unobservables. The authors provide a measure to gauge the strength of the likely
bias arising from unobservables: how much stronger selection on unobservables,
relative to selection on observables, must be to explain away the full estimated
effect.20
To see how this measure is calculated, consider two regressions: one with a
restricted set of control variables, and one with a full set of controls. Denote the
estimated coefficient for the variable of interest from the first regression βˆ (where
R
20
Altonji, Elder, and Taber (2005) consider the situation where the explanatory variable is a binary explanatory
variable. Bellows and Miguel (2009) develop the same test, but for the case where the variable of interest is continu-
ous. Full details of the test are provided in the working paper version of their study, Bellows and Miguel (2008).
Table 4—Using Selection on Observables to Assess the Bias from Unobservables

Controls in the
restricted set Controls in the full set (1) (2) (3) (4) (5)
None Full set of controls from equation (1) 4.31 4.23 3.03 4.13 3.32
None Full set of controls from equation (1), 11.54 6.98 2.65 9.22 3.80
ethnicity-level colonial controls, and
colonial population density
Age, age squared, Full set of controls from equation (1) 4.17 3.99 2.89 3.91 3.12
gender
Age, age squared, Full set of controls from equation (1), 10.93 6.52 2.57 8.44 3.59
gender ethnicity-level colonial controls, and
colonial population density
Notes: Each cell of the table reports ratios based on the coefficient for ln (1 + exports/area) from two individual-
level regressions. In one, the covariates include the “restricted set” of control variables. Call this coefficient β R. In
the other, the covariates include the “full set” of controls. Call this coefficient β F. In both regressions, the sample
sizes are the same, and country fixed effects are included. The reported ratio is calculated as: β F/(β R − β F). See
Table 3 for the description of the full set of controls from equation (1), the ethnicity-level colonial controls, and
colonial population density.
R stands for Restricted) and the estimated coefficient from the second regression
βˆ (where F stands for Full). Then, the ratio can be calculated as: βˆ /(βˆ − βˆ ).21
F F R F
The intuition behind the formula is straightforward. First, consider why the ratio is
decreasing in (βˆ − βˆ ). The smaller is the difference between βˆ and βˆ , the less
R F R F
the estimate is affected by selection on observables, and the stronger selection on

unobservables needs to be (relative to observables) to explain away the entire effect.
Next, consider the intuition behind βˆ in the numerator. The larger βˆ , the greater
F F
is the effect that needs to be explained away by selection on unobservables, and

therefore the higher is the ratio.
We consider two sets of restricted covariates: one with no controls and another
with a sparse set of individual controls that includes only age, age squared, and
the gender indicator variable. We also consider two sets of full covariates: the
baseline set of controls from equation (1), and a second adding to this the ethnic-
ity-level colonial control variables, including colonial population density. Given
our two restricted and two unrestricted sets of covariates, there are four combi-
nations of restricted and unrestricted controls that can be used to calculate the
ratios. The ratios, for each of our five measures of trust, are reported in the cells
of Table 4.
Of the 20 ratios reported in Table 4, none is less than one. The ratios range from
3.0 to 11.5, with a median ratio of 4.1. Therefore, to attribute the entire OLS esti-
mate to selection effects, selection on unobservables would have to be at least three
times greater than selection on observables and, on average, over four times greater.
In our view, these results make it less likely that the estimated effect of the slave
trade is fully driven by unobservables. In the following section, we examine this
issue further by undertaking an alternative strategy.
21
See Bellows and Miguel (2008) for the formal derivation. As well, see Altonji, Elder, and Taber (2005) for
details of the underlying assumptions.
C. IV Estimates
Our final strategy is the use of instrumental variables. This requires an instru-
ment that is correlated with the number of slaves taken from an ethnic group but
uncorrelated with any characteristics of the ethnic group that may affect the trust of
descendants. We use a measure of the distance of an individual’s ethnic group from
the coast during the slave trade. The instrument captures an ethnic group’s exposure
to the external demand for slaves, since slaves were purchased at the coast before
being shipped overseas. Further, distance from the coast is plausibly uncorrelated
with other factors that affected the trust of their descendants.
The instrument is constructed using data from Murdock (1959) on the historical
borders of ethnic groups during the nineteenth century. (The borders are shown in
Figures 1A and 1B.) We calculate the distance from the centroid of each ethnicity to
the closest point along the coast.
The history of Africa’s slave trades leaves little doubt that the instrument is rel-
evant. Various authors, including Joseph C. Miller (1996), describe the slave trade
as progressing in waves of destruction that originated from the coast. The critical
issue is whether an ethnic group’s distance from the coast in the past is uncorrelated
with factors, other than the slave trade, that may affect how trusting the ethnic group
is today—for example, initial prosperity, which may have affected an ethnic group’s
susceptibility to the slave trade, as well as its subsequent trust. Generally, we would
expect distance from the coast to be correlated with overseas trade, and thus with
initial prosperity. However, because of Africa’s particular history, this is not a con-
cern. In the regions in our sample, there was no overseas trade prior to the transatlan-
tic and Indian Ocean slave trades. This alleviates concerns that initial distance from
the coast may have had a direct effect on initial development via preexisting trade.
Despite this fact, there remain a number of other reasons why the exclusion restric-
tion may not be satisfied. First, distance from the coast may be correlated with other
forms of European contact, like colonial rule, which followed the slave trade. For this
reason, we only report IV estimates after controlling for our full set of ethnicity-level
colonial control variables. Second, locations closer to the coast were more likely to
rely on fishing as a form of subsistence. Although it is not obvious how this may
affect future trust, to be as thorough as possible we control for ethnicities’ historical
reliance on fishing. Third, for some parts of Africa, proximity to the coast implies
greater distance from the ancient trade networks across the Sahara Desert. Because
long-term trust may have been affected by a group’s involvement in this inland trade,
we also control for the average distance to the closest city in the Saharan trade, as
well as the average distance to the closest route of the Saharan trade.22
We report IV estimates for each of the five measures of trust in Tables 5 and 6.
Table 5 reports IV estimates controlling for our baseline set of control variables, the
22
An additional concern is that the distance of an individual’s ethnic group from the coast in the past may be
correlated with the individual’s distance from the coast today, which may be correlated with current income and
trust. However, throughout the analysis we control for a number of proxies for income, such as education, occupa-
tion, and living condition fixed effects. An alternative strategy is to also control for a respondent’s current distance
from the coast. This yields results that are similar to those we report here. The full estimates are reported in the
online Appendix.
Table 5—IV Estimates of the Effect of the Slave Trade on Trust

(1) (2) (3) (4) (5)
Second stage: Dependent variable is an individual’s trust
ln (1 + exports/area) −0.190*** −0.245*** −0.221*** −0.251*** −0.174**
(0.067) (0.070) (0.060) (0.088) (0.080)
Hausman test (p-value) 0.88 0.53 0.09 0.44 0.41
R2 0.13 0.16 0.20 0.15 0.12
First stage: Dependent variable is ln (1 + exports/area)
Historical distance of ethnic −0.0014*** −0.0014*** −0.0014*** −0.0014*** −0.0014***
group from coast (0.0003) (0.0003) (0.0003) (0.0003) (0.0003)
Number of clusters 147 / 1,187 147 / 1,187 146 / 1,194 147 / 1,186 147 / 1,184
F-stat of excl. instrument 26.9 26.8 27.4 27.1 27.0
2
R 0.81 0.81 0.81 0.81 0.81
Notes: The table reports IV estimates. The top panel reports the second-stage estimates, and the bottom panel
reports first-stage estimates. Standard errors are adjusted for two-way clustering at the ethnicity and district levels.
The individual controls, district controls, ethnicity-level colonial controls, and colonial population density measures
are described in Table 3. The null hypothesis of the Hausman test is that the OLS estimates are consistent.
ethnicity-level colonial controls, and colonial population density.23 The first-stage

estimates show that historical distance from the coast is negatively correlated with
slave exports. Consistent with the historical record, ethnic groups that were further
from the coast exported fewer slaves. The second-stage estimates report a negative
and highly significant effect of the slave trade on trust. Furthermore, the magnitudes
of the estimates are remarkably similar to the OLS estimates. In fact, in all specifi-
cations, the Durbin-Wu-Hausman test cannot reject the null hypothesis of the con-
sistency of the OLS estimates at the 5 percent level or lower. These results suggest
that selection into the slave trade is not strongly biasing the OLS estimates. This is
consistent with the findings in the previous section, and with the findings in Nunn
(2008), where the IV estimates of the effect of the slave trade on per capita income
across countries were similar to the OLS estimates.
Table 6 reports estimates with controls for each ethnic group’s historical reliance
on fishing and two measures of its distance from the Saharan trade. The variable
for the reliance on fishing measure is from Murdock (1967) and is measured as the
fraction of food from fish. The distance from the Saharan trade variables are: the dis-
tance to the closest city involved in the Saharan trade and the distance to the closest
route of the Saharan trade. Data on the historical locations of towns and routes are
originally from Roland Oliver (2000) and have been digitized by Matthew T. Ciolek
23
The results are similar if we do not include the colonial population density control. The results from this speci-
fication are reported in the online Appendix.
Table 6—IV Estimates of the Effect of the Slave Trade on Trust, with Additional Controls

(1) (2) (3) (4) (5)
Second stage: Dependent variable is an individual’s trust
ln (1 + exports/area) −0.172** −0.271*** −0.262*** −0.254** −0.189*
(0.076) (0.088) (0.075) (0.109) (0.103)
Hausman test (p-value) 0.98 0.42 0.05 0.53 0.44
R2 0.13 0.16 0.20 0.15 0.12
First stage: Dependent variable is ln (1 + exports/area)
Historical distance of ethnic −0.0015*** −0.0015*** −0.0015*** −0.0015*** −0.0015***
group from coast (0.0003) (0.0003) (0.0003) (0.0003) (0.0003)
Reliance on fishing Yes Yes Yes Yes Yes
Distances to Saharan city, route Yes Yes Yes Yes Yes
Number of clusters 147 / 1,187 147 / 1,187 146 / 1,194 147 / 1,186 147 / 1,184
F-stat of excl. instrument 21.7 21.6 22.2 21.8 21.6
R2 0.81 0.81 0.81 0.81 0.81
Notes: The table reports IV estimates. The top panel reports the second-stage estimates, and the bottom panel
reports first-stage estimates. Standard errors are adjusted for two-way clustering at the ethnicity and district levels.
The individual controls, district controls, ethnicity-level colonial controls, and colonial population density measures
are described in Table 3.
(2001). The estimates are similar when controlling for these additional factors; they
remain negative, significant, and virtually identical in magnitude.
Our measures capture slaves exported anytime after 1400. For West, West-Central,
and Southern African countries in our sample, overseas trade did not exist prior to
this date. However, the Indian Ocean slave trade and overseas trade in legitimate
commodities predate 1400, the first period for which we have slave data. Therefore,
it is possible that, for this region of Africa, distance from the coast directly affected
the characteristics of ethnic groups prior to the first year of our analysis. However,
IV estimates omitting ethnic groups from coastal East Africa yield similar results.24
Falsification Tests.—When we examine the reduced form, we find a strong posi-

tive relationship between the historical distance from the coast of individuals’ ances-
tors and their level of trust today.25 This correlation is consistent with the first- and
second-stage IV estimates reported in Tables 5 and 6: ethnic groups that lived closer
to the coast were more exposed to the slave trade, and today their descendants are
less trusting. Our IV strategy rests on the assumption that the slave trade is the only
channel through which historical distance from the coast affects current trust. If this
24
Space constraints prevent us from reporting the estimates here; they are reported in the online Appendix.
25
A similar relationship is found if one examines an individual’s current distance from the coast and trust.
assumption is correct, then a positive relationship between distance from the coast
and trust should not exist in parts of the world that did not experience the slave trade.
To assess the validity of the IV estimates, we undertake this falsification test:
we use two additional surveys that ask the same, or similar, trust questions as the
Afrobarometer survey and we identify the locations of individuals in the surveys. The
first sample comes from the 2003 Asiabarometer and includes individuals from the fol-
lowing ten Asian countries: Japan, South Korea, China, Malaysia, Thailand, Vietnam,
Myanmar, India, Sri Lanka, and Uzbekistan. Using data on the locations of the survey
respondents, we calculate each person’s current distance from the nearest coastline.26
The Asiabarometer asks respondents how much they trust their local government.
The question is worded “How much do you trust your local government?” Although
this differs slightly from the Afrobarometer question, which asks “How much do
you trust your locally elected government council?” both questions convey the
same general meaning. Moreover, the available answers for the two questions are
the same, further suggesting that they are comparable. We construct our dependent
variable the same way for both samples. Because income, occupation, and ethnic
fractionalization measures are unavailable from the Asiabarometer survey, these
covariates are not included in the estimating equations of either the African or Asian
samples. The covariates that are common to the two samples are also measured
slightly differently, so we report all specifications with country fixed effects only.
The first two columns of Table 7 report the reduced-form estimates of the rela-
tionship between distance from the coast and trust in the local government within
Africa. With or without the control variables, there is a strong positive relationship
between ethnic groups’ historical distance from the coast and their trust in their
local council.27 Columns 3 and 4 report the same reduced-form estimates within
Asia. Unlike the African sample, the Asian sample shows no systematic relationship
between an individual’s distance from the coast and trust. Both point estimates for
Asia are close to zero, and highly insignificant.
We also undertake a second falsification exercise using the 1990 World Values
Survey (WVS). The sample includes individuals from Chile, Norway, Sweden,
Great Britain, and Northern Ireland, the only countries in the first four rounds of
the WVS for which the geographic location of respondents is collected and a trust
question similar to one of the Afrobarometer trust questions is asked. The WVS
asks: “How much do you trust <nationality> people in general?” This is similar to
the Afrobarometer question: “How much do you trust <nationality> people from
other ethnic groups?” The possible responses for the WVS answers are slightly dif-
ferent from the Afrobarometer categories. In addition to the four answers in the
Afrobarometer survey—“not at all,” “not very much,” “a little,” and “completely”—
the WVS allows respondents the additional choice of “neither trust or distrust.” For
the WVS variable, as with the Afrobarometer measure, we assign the values 0 and 1
to the two least trustful answers, and the values of 2 and 3 to the two most trustful
26
Note that here we are using each respondent’s current distance from the coast since we do not have a measure
of his ancestor’s historical distance from the coast. Given the persistence in family locations over time, and the
strong correlation between historical and current distance within the Africa sample, we feel that the current distance
from the coast is a useful proxy for historical distance.
27
The relationship is similar if current distance from the coast is used instead of the historical distance from the
coast. See the online Appendix for full estimates.
Table 7—Reduced Form Relationship between the Distance from the Coast
and Trust within Africa and Asia
Trust of local government council

Afrobarometer sample Asiabarometer sample
(1) (2) (3) (4)
Distance from the coast 0.00039*** 0.00031*** −0.00001 0.00001
(0.00009) (0.00008) (0.00010) (0.00009)
Country fixed effects Yes Yes Yes Yes
Individual controls No Yes No Yes
Number of observations 19,913 19,913 5,409 5,409
Number of clusters 185 185 62 62
R2 0.16 0.18 0.19 0.22
Notes: The table reports OLS estimates. The unit of observation is an individual. The depen-
dent variable in the Asiabarometer sample is the respondent’s answer to the question: “How
much do you trust your local government?” The categories for the answers are the same in
the Asiabarometer as in the Afrobarometer. Standard errors are clustered at the ethnicity level
in the Afrobarometer regressions and at the location (city) level in the Asiabarometer and the
WVS samples. The individual controls are for age, age squared, a gender indicator, education
fixed effects, and religion fixed effects.
answers. For the additional intermediate category, “neither trust or distrust,” we

assign a value of 1.5. This coding ensures that both variables have the same range,
from 0 to 3.28
The estimation results are reported in Table 8. Columns 1 and 2 report the reduced-
form relationship within Africa, with and without the set of controls. (Because the
WVS does not include measures of education, living conditions, or religion, these
are not included in the regressions.) Again, we find a strong positive relationship
between an individual’s distance from the coast and trust. Columns 3 and 4 report
the same estimates using the WVS sample. For this non-African sample, we do not
find evidence of a positive relationship between distance from the coast and trust.
Both coefficients are statistically insignificant.
One potential concern is that the differences in the relationship between distance
from the coast and trust within and outside of Africa are driven solely by differ-
ences in the two surveys, such as their sample sizes or the precision of the data.
However, as shown in column 5, when we look only at respondents from African
countries within the WVS—i.e., Nigeria—we estimate a strong positive relationship
between individuals’ distance from the coast and trust. This suggests that the differ-
ent reduced-form relationships that we find are not the result of differences in the
underlying surveys.
Robustness to Violations of Perfect Exogeneity.—Although our falsification tests

do provide evidence for the validity of our instrument, we recognize that the require-
ment of perfect exogeneity is a knife requirement that, strictly speaking, is unlikely
to hold exactly. To gain a sense of the robustness of our IV estimates, we relax the
28
The results are not sensitive to this assumption. They are qualitatively identical if we instead use a trust vari-
able that takes on the values 0, 1, 2, 3, and 4.
Table 8—Reduced Form Relationship between the Distance from the Coast
and Trust within and Outside of Africa
Intergroup trust
Afrobarometer sample WVS non-Africa sample WVS Nigeria
(1) (2) (3) (4) (5)
Distance from the coast 0.00039*** 0.00037*** −0.00020 −0.00019 0.00054***
(0.00013) (0.00012) (0.00014) (0.00012) (0.00010)
Country fixed effects Yes Yes Yes Yes n/a
Individual controls No Yes No Yes Yes
Number of observations 19,970 19,970 10,308 10,308 974
Number of clusters 185 185 107 107 16
R2 0.09 0.10 0.09 0.11 0.06
Notes : The table reports OLS estimates. The unit of observation is an individual. The dependent variable in the
WVS sample is the respondent’s answer to the question: “How much do you trust <nationality> people in gen-
eral?” The categories for the respondent’s answers are: “not at all,’’ “not very much,’’ “neither trust nor distrust,’’
“a little,’’ and “completely.” The responses take on the values 0, 1, 1.5, 2, and 3. Standard errors are clustered at the
ethnicity level in the Afrobarometer regressions and at the location (city) level in the Asiabarometer and the WVS
samples. The individual controls are for age, age squared, a gender indicator, an indicator for living in an urban
location, and occupation fixed effects.
assumption of perfect exogeneity and examine the bounds we are able to place on
the true effect of the slave trade on trust as we deviate from perfect exogeneity.
Consider a generalization of the standard IV equations that allows the instrument
to also enter linearly in the second-stage regression with a coefficient γ. In other
words, we allow distance from the coast to affect trust directly. Conley, Christian
Hansen, and Peter E. Rossi (2008) show how one can obtain consistent estimates
of the effect of interest (in our case, the slave trade on trust β ) if γ is known.
Furthermore, the estimates of the relationship between distance from the coast and
trust in countries where there was no slave trade provide consistent estimates of γ.29
Applying Conley, Hansen, and Rossi (2008), the first finding is that in our setting
when γ < 0, the bounds on the strength of β are actually further from zero (i.e.,
a stronger effect) relative to the IV estimate of β. In other words, if areas further
from the coast have lower trust, then the IV coefficient provides an underestimate of
the true effect of the slave trade on trust. This is reassuring, since three of our four
falsification exercises report negative point estimates for the correlation between
distance from the coast and trust, γ.
Applying Conley, Hansen, and Rossi (2008), we can show that the positive
estimate of γ reported in column 4 is not above the value of γ necessary to lose
confidence in the finding of a negative impact of the slave trade on trust. For the
90 percent confidence interval for β to include zero, γ must be larger than 56 × 10−6.
This is over eight times greater than the estimate of 7 × 10−6 from column 4 of
29
To see this, note that in general the reduced-form relationship between distance from the coast and trust cap-
tures both β and γ. But if we are certain that β is zero (as is the case in the parts of the world where there was no
slave trade), then it only captures γ.
Table 4. Therefore, even allowing for plausible amounts of imperfect exogeneity, we

are still able to confirm a negative effect of the slave trade on trust.30
V. Testing for Channels of Causality: Effects of the Slave Trade

on Internal Norms versus External Factors
Up to this point, we have asked whether the slave trade caused the descendants of
those exposed to it to become less trusting. The evidence we presented is consistent
with our hypothesis that the evolution of behavioral norms was influenced during
the 400-year period of the slave trade. Those exposed to the trade became less trust-
ing, and their descendants remain less trusting today. However, a second explanation
is also possible. The slave trade may be correlated with lower trust today because it
resulted in a deterioration of preexisting states, institutions, and legal structures. If
these institutional effects persist, then people today may have lower levels of trust
because poor institutions permit poor behavior, which engenders mistrust.
In this section, we perform three empirical tests to distinguish between the two
channels. The first focuses on individuals’ trust in their local government council.
We have already shown that individuals with heavily threatened ancestors have less
trust today in their local government. This relationship could be due in part to the
adverse effects of the slave trade on local institutions. Individuals may mistrust their
local government council not because they have developed internal norms of mis-
trust, but rather because the council is not trustworthy. We account for this by con-
trolling directly for the perceived trustworthiness of the local government council.
In the survey, respondents were asked whether they approve or disapprove of the
way their locally elected government councilor performed his/her job over the past
12 months. Respondents chose from the following responses: strongly disapprove,
disapprove, approve, or strongly approve. Respondents also were asked two addi-
tional questions: (i) how many of their locally elected councilors were corrupt, and
(ii) whether their local council members listen to their concerns. For the corruption
question, the respondents were given the option of answering that none, some, most,
or all of the councilors are corrupt. For the question about whether councilors listen,
the respondents were given the option of answering: never, only sometimes, often,
or always.
In the data, we find that individuals with ancestors who were more heavily affected
by the slave trade are more likely to disapprove of their local government council,
to report that more of their councilors are corrupt, and to feel that councilors do not
listen. Therefore, it is possible that the relationship between the slave trade and mis-
trust in the local council arises because the slave trade adversely affected the actual
trustworthiness of the local government council.
In column 1 of Table 9 we check for this possibility by estimating equation (1)
while controlling for the three measures of the perceived quality of the local coun-
cil.31 We include three sets of fixed effects constructed from the responses to each
30
These results are from an IV regression with trust in the local council as the dependent variable, and control-
ling for the baseline controls and the ethnicity-level colonial controls.
31
Throughout this section, colonial population density is included in the regression. As we show in the online
Appendix, the results are qualitatively identical if this control is omitted.
Table 9—Identifying Channels of Causality
Intergroup trust
Within Within Within
Trust of local council town district province
(1) (2) (3) (4) (5)
Ethnicity-based slave export measure −0.072*** −0.070*** −0.102*** −0.120*** −0.098***
(baseline measure) (0.019) (0.019) (0.028) (0.027) (0.029)
Average slave export measure among other −0.037 −0.063** −0.091***
ethnicities in the same location (0.029) (0.030) (0.035)
Council trustworthiness fixed effects Yes Yes No No No
Five public goods fixed effects No Yes No No No
Baseline controls Yes Yes Yes Yes Yes
Number of clusters 146/1,172 145/1,130 147/725 147/737 147/1,127
R2 0.37 0.37 0.12 0.12 0.12
Notes: The table reports OLS estimates. The unit of observation is an individual. Standard errors are adjusted for
two-way clustering at the ethnicity-based ethnicity level and at the location-based ethnicity level. “Average slave
export measure among other ethnicities in the same location” is the average slave export measure of respondents in
the Afrobarometer survey living in the same village, district, or region as the respondent. The “Five public goods
fixed effects” are for the existence of the following public goods in the respondent’s town/village: school, health
clinic, sewage, piped water, and electricity. See Table 3 for a description of the baseline controls, the ethnnicity-level
colonial controls, and the colonial population density variables.
question. Even with the inclusion of these additional controls, the estimated relation-
ship between slave exports and trust remains negative and highly significant. The
estimates of β from a regression without the quality of local council fixed effects,
but using the same sample of observations, is −0.141 (the standard error is 0.024 ).
Therefore, controlling for the quality of the local council decreases the magnitude of
the estimated coefficient by just under 50 percent. This result suggests that over half
of the estimated relationship between slave exports and trust cannot be explained by
a deterioration in the trustworthiness of the local council.
It is possible that including further controls for the trustworthiness of the local
council would result in point estimates that are close to zero. To cast doubt on this
possibility, we include additional controls. Column 2 of Table 9 shows that includ-
ing objective proxies for the quality of the local government—measured by the
existence of public goods—has little effect on the coefficient. The variables are con-
structed from Afrobarometer survey questions that ask whether electricity, piped
water, sewage, health clinics, and schools are available in the respondent’s village.
Using this information, we control for five indicator variables that equal one if the
respondent has access to each of the five public goods.
In a second exercise, we further distinguish between the effects of the slave trade
through a change in the internal norms of trust versus a change in the trustworthiness
of others. We focus on intergroup trust and reestimate equation (1) while controlling
for how much the slave trade affected others from different ethnic groups living in
the same area. For each observation, we calculate the average slave export inten-
sity of those belonging to different ethnic groups living in the respondent’s town,
district, or region. The measure is intended to capture any effects of the slave trade
on the trustworthiness of other ethnic groups living near the individual.
The estimates are reported in columns 3–5 of Table 9. Column 3 reports esti-
mates of equation (1), controlling for the average interethnic slave export intensity
of others in the respondent’s town. If there are no individuals from other ethnic
groups living in the same town in the Afrobarometer sample, then the variable
takes on a missing value. As a result, controlling for the measure results in a
smaller sample of 9,673 observations. Columns 4 and 5 report estimates using the
district and region as the geographic area when constructing the interethnic slave
trade variable. As the geographic region is broadened, there are fewer missing
observations, because it is more likely that other ethnicities in the sample live in
the same location as the respondent.
The estimated effect of the slave trade on intergroup trust is robust to controlling
for the effect of the slave trade on trustworthiness. Moreover, if we compare the
point estimates to estimates using the same specification and sample, but not includ-
ing the interethnic slave exports control (which are −0.104, −0.126 and −0.107,
respectively), we find that the point estimates are barely affected by the inclusion of
the control. These findings suggest that essentially all of the estimated effect of the
slave trade on intergroup trust is not explained by the effect of the slave trade on the
trustworthiness of others.32
Our final strategy is to estimate directly how much of the slave trade’s effect
on trust works through an individual’s external environment—such as the rule of
law and the trustworthiness of others—versus through individuals’ internal norms
of mistrust. We do this by constructing a second slave-export variable. Unlike our
baseline measure of the number of slaves taken from an individual’s ethnic group, it
measures the number of slaves taken from the geographic area in which the individ-
ual is currently living. We first identify the current location of each respondent and
then determine which ethnic group historically inhabited that location. The location-
based slave-export variable takes on the value of the slave exports measure for the
ethnic group that historically lived in the location. Therefore, the second variable
measures the slave trade’s impact on an individual’s geographic location, rather than
on the individual’s ancestors.
The two slave export measures identify the internal and external channels by
exploiting the fact that when individuals relocate, their cultural beliefs, norms, and
values move with them, but their external environment is left behind. Therefore, if
the slave trade primarily affects trust through internal factors, then mistrust should
be most strongly correlated with the extent to which individuals’ ancestors were
affected by the slave trade. If the slave trade affects trust primarily through external
factors, like the deterioration of domestic institutions, which lead to a decline in
the trustworthiness of others, then mistrust should be most strongly correlated with
the slave trade’s impact on the environment in which the individual lives today.
By including the ethnicity-based and location-based slave export variables in our
32
This is not to say that the slave trade did not affect trustworthiness. The estimates provide evidence for this.
The coefficients for the interethnic slave exports variable are negative in all three specifications, and significant in
two of the three. This is consistent with the slave trade’s negatively affecting the trustworthiness of individuals,
which causes them to be trusted less by others today.
Table 10—Identifying Channels of Causality
Trust of Trust of Trust of Intragroup Intergroup

relatives neighbors local council trust trust
(1) (2) (3) (4) (5)
Ethnicity-based slave export measure −0.155*** −0.182*** −0.100*** −0.169*** −0.090***
(baseline measure) (0.029) (0.029) (0.023) (0.033) (0.030)
Location-based slave export measure −0.045*** −0.045*** −0.045** −0.043** −0.047**
(0.014) (0.016) (0.018) (0.018) (0.020)
Baseline controls Yes Yes Yes Yes Yes
Number of clusters 146/269 146/269 145/272 146/269 146/269
R2 0.13 0.16 0.20 0.16 0.12
Notes: The table reports OLS estimates. The unit of observation is an individual. Standard errors are adjusted for
two-way clustering at the ethnicity-based ethnicity level and at the location-based ethnicity level. “Ethnicity-based
slave export measure” is our baseline measure of slave exports used throughout the article; it is the log of the num-
ber of slaves taken from an individual’s ethnic group (normalized by land area). “Location-based slave export mea-
sure” is our alternative measure of slave exports, which is the log of the number of slaves taken from the location
where an individual is currently living (normalized by land area). See Table 3 for a description of the baseline con-
trols, the ethnicity-level colonial controls, and the colonial population density variables.
estimating equation, we are able to distinguish between the effects of the slave trade
through the two channels.33
If an individual currently lives where his ancestors lived, then the two slave export
measures will be the same.34 When we include both variables in the estimating equa-
tion, the “movers” in the sample (i.e., those living in a location different from their
ancestors) are the source of identification. Therefore, the estimates are an average
effect among the movers only, and they may not apply to the population more gener-
ally. Because movers constitute 45 percent of the population, knowing the average
effect among this group is still informative.35
Estimates of equation (1) with both slave-export variables included are reported in
Table 10. The estimates for the baseline ethnicity-based slave export measure remain
robust to the inclusion of the location-based slave-export variable. The coefficients
remain negative and highly significant, and their magnitudes decrease by only about
10–15 percent when the location-based slave-exports variable is included. (The esti-
mates of β without the location-based control are −0.187, −0.204, −0.136, −0.190,
−0.116, respectively.)
33
The logic of the test is the same as that used in previous studies that examine migrants to test whether cultural
differences can explain differences in on-the-job shirking (Andrea Ichino and Giovanni Maggi 2000), financial
decisions (Guiso, Sapienza, and Zingales 2004), living arrangements (Paola Giuliano 2007), or female labor force
participation and fertility (Raquel Fernández and Alessandra Fogli 2007).
34
Not surprisingly, we find that the geography- and ethnicity-based measures of slave exports are highly cor-
related (the correlation coefficient is 0.74). For 55 percent of the respondents in the sample, both variables take on
the same value.
35
Relative to nonmovers, movers are more likely to live in urban locations that are more ethnically fragmented
and with fewer coethnics. This is consistent with the migration patterns observed within African countries, where
individuals, in search of better employment opportunities, move from ethnically homogenous rural villages to
larger, more ethnically diverse urban centers. Full details of these differences are reported in the online Appendix.
The location-based measure of slave exports always enters with a negative and
significant coefficient, suggesting that the slave trade affects trust through geo-
graphically fixed factors, like domestic institutions. Comparing the magnitudes
of the coefficients for the two variables, we find that the ethnicity-based slave
export coefficient is always at least twice the magnitude of the location-based
slave export coefficient. This suggests that, although the slave trade adversely
affected trust through factors both internal and external to the individual, the inter-
nal channel was more important.
Overall, the results from our three tests suggest that much of the slave trade’s
effect on trust, identified in Sections III and IV, arises from a change in the internal
norms and beliefs of the descendants of those affected by the slave trade. Our first
test suggests that over 50 percent of the relationship between the slave trade and
trust in the local council can be explained by internal norms. Our second and third
tests suggest that internal norms explain 85–100 percent of the total effect of the
slave trade on interpersonal trust.
VI. Conclusions
This article adds to a new and growing literature in economics that seeks to better
understand the role that culture, norms, and beliefs play in individual decision mak-
ing. Generally, the empirical literature has focused on either showing that culture
exists or on identifying the economic impacts of cultural differences. The next natu-
ral step is to try to understand the origins of cultural differences, which this study
does by looking back into history.
We have shown that within Africa low levels of trust can be traced back to the leg-
acy of the slave trade. Individuals’ trust in their relatives, neighbors, coethnics, and
local government is lower if their ancestors were heavily affected by the slave trade.
To determine whether this relationship is causal, we pursued a number of differ-
ent strategies. First, we controlled for initial ethnicity characteristics and for the
potential impact of colonial rule. Second, using recently developed techniques from
Altonji, Elder, and Taber (2005), we showed that on average selection based on
unobservables would have to be four times greater than selection on observables in
order for the negative effect of the slave trade on trust to be completely spurious.
Finally, we reported IV estimates that use the historical distance from the coast of an
individual’s ethnic group as an instrument for slave exports. The IV estimates also
show a negative effect of the slave trade on trust.
Motivated by the possibility that the instrument does not satisfy the exclusion
restriction, we then performed a number of falsification exercises. Within Africa, we
observe a robust positive reduced-form relationship between distance from the coast
and trust. However, in samples outside of Africa, we find no reduced-form relation-
ship. These correlations are consistent with distance from the coast affecting trust
only through the slave trade (i.e., that the exclusion restriction is satisfied).
We then turned to specific mechanisms and examined two explanations for the
relationship between the slave trade and trust. The first is that over the 400 years of
insecurity generated by the slave trade, general beliefs or “rules-of-thumb” based on
mistrust evolved. These beliefs were then transmitted from parents to children over
time, and they continue to manifest themselves today, more than 100 years after the
end of the slave trade. The second explanation is that the slave trade resulted in a
deterioration of legal and political institutions. Because these weakened institutions
continue to persist today, individuals are not constrained to act in a trustworthy man-
ner, and this lack of trustworthiness results in lower trust.
We performed three tests to determine the relative importance of the two chan-
nels. We find evidence for both mechanisms. The evidence suggests that the slave
trade had an adverse effect on the external environment, which continues to affect
trustworthiness to this day. We also find evidence that the slave trade altered the trust
of modern Africans through internal factors, such as norms, beliefs, and values. Our
tests suggest that the internal channel accounts for at least half of the reduced-form
effect of the slave trade on trust.
Overall, the findings provide evidence for the importance of internal norms and
beliefs in transmitting the impacts of a historical shock, in this case the slave trade.
One reason that history matters today is through the evolution of cultural norms.
REFERENCES
Acemoglu, Daron, Simon Johnson, and James A. Robinson. 2001. “The Colonial Origins of Compar-
ative Development: An Empirical Investigation.” American Economic Review, 91(5): 1369–401.
Acemoglu, Daron, Simon Johnson, and James A. Robinson. 2002. “Reversal of Fortune: Geography
and Institutions in the Making of the Modern World Income Distribution.” Quarterly Journal of
Economics, 117(4): 1231–94.
Aghion, Philippe, Yann Algan, and Pierre Cahuc. 2008. “Can Policy Interact with Culture? Minimum
Wage and the Quality of Labor Relations.” Unpublished.
Aghion, Philippe, Yann Algan, Pierre Cahuc, and Andrei Shleifer. 2010. “Regulation and Distrust.”
Quarterly Journal of Economics, 125(3): 1015–49.
Alesina, Alberto, and Eliana La Ferrara. 2002. “Who Trusts Others?” Journal of Public Economics,
85(2): 207–34.
Alesina, Alberto, and Nicola Fuchs-Schündeln. 2007. “Good-Bye Lenin (or Not?): The Effect of Com-
munism on People’s Preferences.” American Economic Review, 97(4): 1507–28.
Algan, Yann, and Pierre Cahuc. 2010. “Inherited Trust and Growth.” American Economic Review,
100(5): 2060–92.
d’ Almada, André Alvares. 1984. Tratado Breve Dos Rios de Guiné Do Cabo-Verde. Liverpool: Uni-
versity of Liverpool.
Altonji, Joseph G., Todd E. Elder, and Christopher R. Taber. 2005. “Selection on Observed and Unob-
served Variables: Assessing the Effectiveness of Catholic Schools.” Journal of Political Economy,
113(1): 151–84.
Azevedo, Mario. 1982. “Power and Slavery in Central Africa: Chad (1890–1925).” Journal of Negro
History, 67(3): 198–211.
Barry, Boubacar. 1992. “Senegambia from the Sixteenth to the Eighteenth Century: Evolution of the
Wolof, Sereer and ‘Tukuloor.’” In General History of Africa, Volume 5: Africa from the Sixteenth
to the Eighteenth Century, ed. B. A. Ogot, 262–99. Berkeley, CA: University of California Press.
Bellows, John, and Edward Miguel. 2008. “War and Collective Action in Sierra Leone.” Unpublished.
Bellows, John, and Edward Miguel. 2009. “War and Local Collective Action in Sierra Leone.” Journal
of Public Economics, 93(11–12): 1144–57.
Bisin, Alberto, and Thierry Verdier. 2000. “‘Beyond the Melting Pot’: Cultural Transmission, Mar-
riage, and the Evolution of Ethnic and Religious Traits.” Quarterly Journal of Economics, 115(3):
955–88.
Bisin, Alberto, and Thierry Verdier. 2001. “The Economics of Cultural Transmission and the Dynam-
ics of Preferences.” Journal of Economic Theory, 97(2): 298–319.
Bisin, Alberto, and Thierry Verdier. 2008. “Cultural Transmission.” In The New Palgrave Dictionary of
Economics, ed. Steven N. Durlauf and Lawrence E. Blume. New York: Palgrave McMillan.
Bloom, Nicholas, Raffaella Sadun, and John Van Reenen. 2008. “Measuring and Explaining Decentral-
ization across Firms and Countries.” Unpublished.
Boyd, Robert, and Peter J. Richerson. 1985. Culture and the Evolutionary Process. Chicago: Univer-
sity of Chicago Press.
Boyd, Robert, and Peter J. Richerson. 1995. “Why Does Culture Increase Human Adaptability?”
Ethology and Sociobiology, 16(2): 125–43.
Boyd, Robert, and Peter J. Richerson. 2005. The Origin and Evolution of Cultures. New York: Oxford
University Press.
Cameron, A. Colin, Jonah B. Gelbach, and Douglas L. Miller. 2006. “Robust Inference with Multi-Way
Clustering.” National Bureau of Economic Research Technical Working Paper 0327.
Century Company. 1911. “The Century Atlas: Africa [map].” Buffalo, NY: Matthews-Northrup (470
English statute miles to 1 inch).
Cesarini, David, Christopher T. Dawes, James H. Fowler, Magnus Johannesson, Paul Lichtenstein,
and Björn Wallace. 2008. “Heritability of Cooperative Behavior in the Trust Game.” Proceedings of
the National Academy of Sciences, 105(10): 3721–26.
Chandler, Tertius. 1987. Four Thousand Years of Urban Growth: An Historical Census. Lewistown,
NY: Edwin Mellen Press.
Ciolek, Matthew T. 2001. “Georeferenced Data Set (Series 1 - Routes): NW African Trade Route
500–1900 CE.” Asia Pacific Research Online OWTRAD Dromographic Digital Data Archives
(ODDDA). Old World Trade Routes (OWTRAD) Project. http://www.ciolek.com/owtrad.html
(accessed September 1, 2009).
Conley, T. G. 1999. “GMM Estimation with Cross Sectional Dependence.” Journal of Econometrics,
92(1): 1–45.
Conley, Timothy, Christian Hansen, and Peter E. Rossi. 2008. “Plausibly Exogenous.” Unpublished.
Easterly, William, and Ross Levine. 1997. “Africa’s Growth Tragedy: Policies and Ethnic Divisions.”
Quarterly Journal of Economics, 112(4): 1203–50.
Fafchamps, Marcel. 2006. “Development and Social Capital.” Journal of Development Studies, 42(7):
1180–98.
Fernández, Raquel, and Alessandra Fogli. 2007. “Culture: An Empirical Investigation of Beliefs,
Work, and Fertility.” Unpublished.
Fisman, Raymond, and Tarun Khanna. 1999. “Is Trust a Historical Residue? Information Flows and
Trust Levels.” Journal of Economic Behavior and Organization, 38(1): 79–92.
Francois, Patrick, Thomas Fujiwara, and Tanguy van Ypersele. 2010. “Competition Builds Trust.”
Unpublished.
Giuliano, Paola. 2007. “Living Arrangements in Western Europe: Does Cultural Origin Matter?” Jour-
nal of the European Economic Association, 5(5): 927–52.
Greif, Avner. 1989. “Reputation and Coalitions in Medieval Trade: Evidence on the Maghribi Traders.”
Journal of Economic History, 49(4): 857–82.
Guiso, Luigi, Paola Sapienza, and Luigi Zingales. 2004. “The Role of Social Capital in Financial
Development.” American Economic Review, 94(3): 526–56.
Guiso, Luigi, Paola Sapienza, and Luigi Zingales. 2007a. “Cultural Biases in Economic Exchange.”
Unpublished.
Guiso, Luigi, Paola Sapienza, and Luigi Zingales. 2007b. “Social Capital as Good Culture.” National
Bureau of Economic Research Working Paper 13712.
Hair, P. E. H. 1965. “The Enslavement of Koelle’s Informants.” Journal of African History, 6(2): 193–203.
Hawthorne, Walter. 1999. “The Production of Slaves Where There Was No State: The Guinea-Bissau
Region, 1450–1815.” Slavery & Abolition, 20(2): 97–124.
Hawthorne, Walter. 2003. Planting Rice and Harvesting Slaves: Transformations along the Guinea-
Bissau Coast, 1400–1900. Portsmouth, NH: Heinemann.
Hoff, Karla, Mayuresh Kshetramade, and Ernst Fehr. 2009. “Caste and Punishment: The Legacy of
Caste Culture in Norm Enforcement.” World Bank Policy Research Working Paper 5040.
Hoff, Karla, and Priyanka Pandey. 2005. “Opportunity Is Not Everything: How Belief Systems and
Mistrust Shape Responses to Economic Incentives.” Economics of Transition, 13(3): 445–72.
Hubbell, Andrew. 2001. “A View of the Slave Trade from the Margin: Souroudougou in the Late Nine-
teenth-Century Slave Trade of the Niger Bend.” Journal of African History, 42(1): 25–47.
Ichino, Andrea, and Giovanni Maggi. 2000. “Work Environment and Individual Background: Explain-
ing Regional Shirking Differentials in a Large Italian Firm.” Quarterly Journal of Economics,
115(3): 1057–90.
Inikori, Joseph E. 2000. “Africa and the Trans-Atlantic Slave Trade.” In Africa, Volume I: African His-
tory before 1885, ed. Toyin Falola, 389–412. Durham, NC: Carolina Academic Press.
Inikori, Joseph E. 2003. “The Struggle against the Transatlantic Slave Trade: The Role of the State.”
In Fighting the Slave Trade: West African Strategies, ed. Sylviane A. Diouf, 170–98. Athens, OH:
Ohio University Press.
Jacob, Marcus, and Marcel Tyrell. 2010. “The Legacy of Surveillance: An Explanation for Social Capital
Erosion and the Persistence of Economic Disparity between East and West Germany.” Unpublished.
Kiszewski, Anthony, Andrew Mellinger, Andrew Spielman, Pia Malaney, Sonia Ehrlich Sachs, and Jef-
frey Sachs. 2004. “A Global Index Representing the Stability of Malaria Transmission.” American
Journal of Tropical Medicine and Hygiene, 70(5): 486–98.
Klein, Martin A. 2001. “The Slave Trade and Decentralized Societies.” Journal of African History,
42(1): 49–65.
Klein, Martin. 2003. “Defensive Strategies: Wasulu, Masina, and the Slave Trade.” In Fighting the Slave
Trade: West African Strategies, ed. Sylviane A. Diouf, 62–78. Athens, OH: Ohio University Press.
Knack, Stephen, and Philip Keefer. 1997. “Does Social Capital Have an Economic Payoff? A Cross-
Country Investigation.” Quarterly Journal of Economics, 112(4): 1251–88.
Koelle, Sigismund Wilhelm. 1854. Polyglotta Africana: Or a Comparative Vocabulary of Nearly Three
Hundred Words and Phrases, in More than One Hundred Distinct African Languages. London:
Church Missionary House.
Lovejoy, Paul E. 2000. Transformations in Slavery: A History of Slavery in Africa. 2nd ed. New York:
Cambridge University Press.
Mahadi, Abdullahi. 1992. “The Aftermath of the Jihad in the Central Sudan as a Major Factor in the
Volume of the Trans-Saharan Slave Trade in the Nineteenth Century.” In The Uncommon Market:
Essays in the Economic History of the Atlantic Slave Trade, ed. Elizabeth Savage, 111–28. London:
Fran Cass.
Manning, Patrick. 1990. Slavery and African Life: Occidental, Oriental and African Slave Trades.
New York: Cambridge University Press.
Miller, Joseph C. 1996. Way of Death: Merchant Capitalism and the Angolan Slave Trade, 1730–1830.
Madison, WI: University of Wisconsin Press.
Murdock, George Peter. 1959. Africa: Its Peoples and Their Culture History. New York: McGraw-Hill.
Murdock, George Peter. 1967. Ethnographic Atlas. Pittsburgh: University of Pittsburgh Press.
Nunn, Nathan. 2008. “The Long-Term Effects of Africa’s Slave Trades.” Quarterly Journal of Eco-
nomics, 123(1): 139–76.
Nunn, Nathan, and Leonard Wanchekon. 2011. “The Slave Trade and the Origins of Mistrust in
Africa: Dataset.” American Economic Review. http://www.aeaweb.org/articles.php?doi=10.1257/
aer.101.7.3221.
Oliver, Roland. 2000. The African Experience: From Olduvai Gorge to the 21st Century. Boulder, CO:
Westview Press.
Piot, Charles. 1996. “Of Slaves and the Gift: Kabre Sale of Kin during the Era of the Slave Trade.”
Journal of African History, 37(1): 31–49.
Putnam, Robert. 2000. Bowling Alone: The Collapse and Revival of American Community. New York:
Simon & Schuster.
Roome, William R. M. 1924. “Ethnographic Survey of Africa: Showing the Tribes and Languages;
Also the Stations of Missionary Societies [map].” (1: 5,977,382).
Tabellini, Guido. 2007. “Culture and Institutions: Economic Development in the Regions of Europe.”
Unpublished.
Tabellini, Guido. 2008. “The Scope of Cooperation: Values and Incentives.” Quarterly Journal of Eco-
nomics, 123(3): 905–50.

The Slave Trade and The Origins of Mistrust in Africa: Citation

Uploaded by

Copyright:

Available Formats

The Slave Trade and The Origins of Mistrust in Africa: Citation

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Slave Trade and The Origins of Mistrust in Africa: Citation

Uploaded by

Copyright:

Available Formats

The Slave Trade and the Origins of Mistrust in

Share Your Story

The Slave Trade and the Origins of Mistrust in Africa†

By Nathan Nunn and Leonard Wantchekon*

We show that current differences in trust levels within Africa can

Our hypothesis builds on the well-established insight from cultural anthropology

working through internal factors that are geographically mobile—such as individu-

I. Historical Background and Conceptual Framework

Our notion of culture is taken from research in evolutionary anthropology that

II. Data Sources and Description

B. Ethnicity-Level Data on Slave Exports

Panel A. Transatlantic slave trade

Atlantic slave exports

Panel B. Indian Ocean slave trade

Indian slave exports

III. Estimating Equations and Empirical Results

(1) trusti, e, d, c = αc + β slave exportse + X′i,e,d,c Γ + X′d,c Ω + X e′ Φ + εi,e,d,c,

Table 1—OLS Estimates of the Determinants of Trust in Neighbors

Table 2—OLS Estimates of the Determinants of the Trust of Others

Trust Trust Trust of Intra- Inter-

An alternative way to assess the magnitude of the slave export coefficients is to

IV. Identifying Causal Relationships

A. Controlling for Observables: Initial Conditions and Colonial Rule

Table 3—OLS Estimates of the Determinants of the Trust of Others,

Trust Trust Trust of

B. Using Selection on Observables to Assess the Bias from Unobservables

Table 4—Using Selection on Observables to Assess the Bias from Unobservables

Trust Trust Trust of

the estimate is affected by selection on observables, and the stronger selection on

is the effect that needs to be explained away by selection on unobservables, and

Table 5—IV Estimates of the Effect of the Slave Trade on Trust

Trust Trust Trust of

ethnicity-level colonial controls, and colonial population density.23 The first-stage

Trust Trust Trust of

Falsification Tests.—When we examine the reduced form, we find a strong posi-

Trust of local government council

answers. For the additional intermediate category, “neither trust or distrust,” we

Robustness to Violations of Perfect Exogeneity.—Although our falsification tests

Table 4. Therefore, even allowing for plausible amounts of imperfect exogeneity, we

V. Testing for Channels of Causality: Effects of the Slave Trade

Table 9—Identifying Channels of Causality

Table 10—Identifying Channels of Causality

Trust of Trust of Trust of Intragroup Intergroup

You might also like