Professional Documents
Culture Documents
Stratification in Africa
By Warren Whatley and Rob Gillezeau
The last decade has been a golden age This work does not capture the mechanism
for the study of African economic history. through which this underdevelopment may
As noted by Antony G. Hopkins (2009) have occurred and while Nathan Nunn and
economists have produced a new economic Leonard Wantchekon (forthcoming) make
history of Africa in the course of the past an effort to explain the process through the
decade, with two primary narratives. The development of mistrust it seems unlikely
first narrative is that Africa has suffered that this is the only causal mechanism. The
a reversal of fortune during the past 500 strong positive relationship between ethnic
years. The second is that ethnic frag- fractionalization and slave exports found in
mentation, which has deep historical roots, this paper suggests that increased ethnic
is a distinct cause of African economic fractionalization may have been a promi-
backwardness (page 155). In this arti- nent factor in this underdevelopment. This
cle, we argue that these narratives are in- would be consistent with work by Robert
terconnected. In particular, we argue that Levine and William Easterly (1997), Paul
Africas slave trade, which helped drive its Collier (1998), Robert Bates (2008) and
reversal of fortune, increased the degree others, but rather than see ethnic fraction-
of ethnic fragmentation in Africa today. In alization as an exogenous source of social
a prior paper, Warren Whatley and Rob conflict this paper presents evidence that it
Gillezeau (2011) show that under plausible is also an endogenous outcome of the so-
conditions the slave trade may have con- cial conflict associated with slaving. Fi-
strained the geographic scope of authority nally, this paper cautions that controlling
and increased the salience of ethnic identity. for ethnic fractionalization will result in un-
In this paper we empirically test this rela- derestimates of the impact of slavery on de-
tionship. Using both OLS and instrumen- velopment.
tal variables analysis, we find an economi-
cally and statistically significant positive re- I. Ethnic Identity, the Slave Trade, and
lationship between various measures of eth- Development
nic fragmentation in the present and slave
exports from the western coast of Africa in A number of important studies have fo-
the past. cused on ethnic stratification and its ex-
These results are significant when consid- ogenous impact on economic performance
ered in the context of other recent work in Africa. The best known is a study
in African economic history. In partic- by Levine and Easterly (1997) which ar-
ular, the results aid in the interpreta- gues that roughly 25 percent of the differ-
tion of Nathan Nunns analysis of the ence in the growth experiences of African
slave trade and GDP (2008) in which he and Asian economies can be attributed
finds that the transatlantic slave trade re- to the greater ethnic diversity in Africa.
sulted in the long-term, systematic under- While it is unclear precisely how ethnicity
development of many African economies. influences economic performance, the au-
thors present some evidence on a negative
Whatley: Department of Economics, University of
relationship between ethnic diversity and
Michigan, 611 Tappan Street, Ann Arbor, MI, 48104,
wwhatley@umich.edu. Gillezeau: Department of Eco-
under-investment in schooling, weak finan-
nomics, University of Michigan, 611 Tappan Street, Ann cial institutions, poor infrastructure and
Arbor, MI, 48104, gillezr@umich.edu. black-market premia.
1
2 PAPERS AND PROCEEDINGS MAY 2011
Collier (1998) cautions that the relation- malleable, politically manipulable and situ-
ship between ethnicity and economic per- ational. Posner (2006), for example, devel-
formance is more complex and contextual ops and tests a model explicitly designed
than this. While arguing that ethnic diver- to identify the conditions under which indi-
sity can become a drag on growth, Collier vidual Zambians choose to organize around
adds the proviso that the negative effects one particular identity rather than another.
are largely confined to economies with lim- Individuals are viewed as having a portfo-
ited individual rights. In fact, ethnic diver- lio of identities from which they can choose,
sity can be a plus. While democratic in- and it is postulated that individuals choose
stitutions can effectively mitigate the nega- the one that has the best chances of putting
tive effects of ethnic diversity, highly diverse them in the winning coalition. The impor-
countries are less likely (not more likely) to tant political choice in post-colonial Zam-
break out into ethnic conflict, presumably bia is between ethnic identity and language
because of the higher cost of inter-ethnic co- identity, and Posner is successful in re-
operation. Bates (2008) contextualizes the vealing the conditions under which people
impact of ethnicity in a similar way. He ar- choose one or the other. Still, in this formu-
gues that the predatory nature of the post- lation ethnic identity as distinct from lan-
colonial state in many African countries cre- guage identity retains a high degree of ex-
ated political and military challenges to its ogeniety. The choice is between ethnic and
authority. When the challenges intensified, language identity, not between competing
ethnic stratification also intensified to the ethnic identities.
point where things fell apart. Ethnic identity becomes more endoge-
The literature on ethnic conflict tends to nous and malleable when one leaves the
assume that the oppositional character of realm of rational choice and takes a his-
ethnic identity, with its insider-outsider dis- torical view. Posner (2006, pages 21-88),
tinctions, is a source of conflict that im- for example, spends two chapters tracing
pedes growth. A useful alternative view is the historical origins of Zambias ethnic and
offered by Joan Esteban and Debraj Ray language groupings. The conventional wis-
(2008). In situations where political be- dom here emphasizes the role played by the
havior can be modeled as prize grabbing institutions of colonial rule, not the conflict
mass mobilization, there is a built-in bias and violence of the slave trade. Quoting
towards ethnic rather than class mobiliza- Posner,
tion because ethnic groups include the rich,
In tracing the origins of con-
who have the resources, and the poor, who
temporary Zambian ethnic iden-
provide the labor needed to mount a mass
tity to the institutions of colo-
movement. Conflict will tend to occur along
nial rule, I am following an ex-
ethnic lines, not because ethnic identity
tremely well-trodden path. In
is inherently conflictual but because it is
fact, the notion that the colonial
easier to mount an ethnically-based mass
state created or heightened the
movement.
importance of ethnic identities in
In all of these examples ethnicity is postcolonial Africa is so accepted
treated as exogenous and given to the situ- these days that to argue other-
ation. In fact, Collier expresses an uneasi- wise would probably be controver-
ness about the negative connotations being sial (2006, page 23).
attached to ethnic diversity in Africa be-
cause ... there is nothing a country can le- Yet otherwise is precisely what we want
gitimately do about its ethnic composition to argue. The conventional view roots the
(1998, pg. 387-88). But there is a large and salience of ethnic identity in Africa in what
growing literature which attempts to endo- Kathryn Firmin-Sellers calls the logic of
genize ethnic identity, to varying degrees. indirect rule (1996). Colonial administra-
This literature tends to emphasize the fact tions, befuddled by the variety of local eth-
that people have multiple identities that are nic political economies they encountered,
VOL. 1 NO. 1 THE SLAVE TRADE AND ETHNIC STRATIFICATION 3
soil and terrain slope constraints, popula- The circular buffer zones overlap along
tion density in 1960,2 elevation,3 local agri- the coast, but they have the virtue of en-
cultural suitability as measured by climate, compassing much interior territory. To en-
forest coverage, and desert coverage.4 sure the robustness of the results, we take
We perform our regression analysis with steps to reduce overlap. We draw buffers
3 different circular buffer sizes: 125 kilome- around every other coastal point (resulting
tres, 250 kilometres, and 500 kilometres in in 100 buffer zones) and every fourth (pro-
radius. In our analysis using the 125 kilo- ducing 50 buffer zones). We also perform
metre buffer our environmental variables our analysis on a set of buffer zones that
are based on their mean value in that re- are just tangent to each other, with no over-
gion, the number of ethnicities is the total lap. This produces 67, 40 and 19 zones re-
number found within that buffer, and slave spectively for the 125km, 250km and 500km
exports are the total exported from slaving buffers. Finally, we also perform our anal-
ports falling within that buffer. In addition ysis at the country level.
to the buffer method, we perform our anal-
ysis assigning each ethnicity to the nearest III. Results
of the coastal points. Using each of these
methods, we perform the following OLS re- In Table 1 we present results from several
gression: of the OLS and 2SLS regression specifica-
tions. The complete set of OLS and first
and second stage 2SLS results for both of
Ei = + 1 Si + Xi + vi our ethnicity measures are available in the
online Appendix.
Where Ei is the number of ethnicities as- Columns 2 and 3 contain a subset of our
signed to coastal point i, is the intercept, OLS results, using both the Peoples At-
Si is the number of slave exports assigned las and Murdock measures of ethnicity to
to coastal point i, Xi is a vector of environ- generate our outcome variables. It is clear
mental covariates assigned to coastal point that there is a robust, positive relationship
i, and vi is an error term. between slave exports and the number of
There is almost certainly some degree of ethnicities as defined by the Peoples At-
reverse causation in the above specifica- las. Among the controls, agricultural suit-
tions. If slaving was taboo within ones own ability and population are positively related
ethnic group it would have been necessary to the number of ethnicities while eleva-
for other ethnicities to be present nearby in tion, forest, and desert cover are all neg-
order to capture slaves. In order to present atively related to the number of ethnicities.
a causal estimate of the impact of slaving The positive relationship between slave ex-
on the development of ethnicity, we make ports and ethnicity tends to persist using
use of the instruments developed by Nunn the Murdock (1959) measure, although it is
(2008) which in this analysis are the dis- a weaker relationship than with the Peoples
tances between the coastal points and the Atlas. In general, the results are stronger
nearest slave destination in the Americas or the larger the buffer zone and the greater
North Africa.5 the number of observations.
In columns 4 and 5, we display the coef-
2 UNESCO. UNEP Sioux Falls Clearninghouse ficient on slave exports from our IV regres-
http://na.unep.net/datasets/datalist.php (accessed Oc-
tober 1, 2010) sions. As in the OLS regressions, there ex-
3 USGS. USGS Geographic Data Download ists a strong positive relationship between
http://edc2.usgs.gov/geodata/index.php. (accessed slave exports and the number of ethnicities
October 1, 2010)
4 IIASA in each region. The treatment effects are
Global Agro-Ecological As-
sessment for Agriculture in the 21st significantly larger in specifications using
Century: Methodology and Results.
http://www.iiasa.ac.at/Research/LUC/GAEZ/index.htm Haiti, Kingston, Dominica, Guyana, Salvador, Rio de
(accessed October 1, 2010) Janeiro, and the North African destinations are Algiers,
5 The American destinations are Virginia, Havana, Tunis, Tripoli, Bengahzi, Cairo.
VOL. 1 NO. 1 THE SLAVE TRADE AND ETHNIC STRATIFICATION 5
the Peoples Atlas and only slightly larger local increase (over a much larger area) of
in specifications using Murdocks mapping. 43.6 to 110.95 ethnicities. Since, ethnicities
The results from the first stage of the re- overlap across buffer zones this overstates
gressions are available in the online Ap- the treatment effect. This should be viewed
pendix. The instrument is powerful and the as an upper bound on the treatment effect.
strategy does not appear to suffer from a For reference, the Peoples Atlas contains
weak instruments problem. roughly 3700 ethnicities for all of Africa.
The results are robust for all buffer sizes It is difficult to pin down a precise treat-
and numbers of observations for the Peoples ment effect, but it is clear that the effect is
Atlas. The Murdock results are generally economically significant even at the lower
robust, although weaker for smaller buffer bound.
zones. The results are robust to remov-
IV. Implications
ing most observations from North Africa
and South Africa. In columns 6 and 7, We have argued that the slave trade in-
we present results using the Peoples At- creased the number of ethnic groupings in
las where buffer zones are not allowed to present-day Africa. We do not claim to un-
overlap. As such, we have 67, 40, and 19 derstand the mechanism, although we be-
observations respectively. The downside to lieve that the slave trade likely constrained
limiting the level of overlap is that we are the geographic scope of political authority
unable to take advantage of much of the and heightened the incentive to distinguish
inland variation in ethnicity. These spec- insider from outsider. Our regressions iden-
ifications provide results that are positive tify a positive and statistically significant
and are statistically significant in half of the relationship between the number of slaves
specifications. The results are also robust leaving the west coast of Africa and the lim-
to performing the analysis at the country- ited geographic scope of twentieth century
level. ethnic groupings. This relationship is ro-
To get an idea of the size of the treat- bust to changes in the scheme for drawing
ment effect, we multiply the coefficients by ethnic boundaries, to the choice of obser-
mean slave exports. The small buffer zone vational unit, and to the inclusion of a va-
estimates, using the Peoples Atlas, indicate riety of variables thought to influence the
that the slave trade resulted in an average geographic scope of ethnic groupings. Our
increase of 0.9 to 2.3 local ethnicities in each IV estimates produce support for the view
of the 200 coastal regions. This should be that causality runs from slaving to ethnic
viewed as a lower bound on the treatment diversity. We believe this finding has broad
effect. The larger buffer zone estimates, us- implications for research in the economic
ing the Peoples Atlas, suggest an average history of Africa.
6 PAPERS AND PROCEEDINGS MAY 2011