indigenous sleavey in africa's history

7/29/2019 Indigenous Sleavey in Africa's History

1/38

1

Indigenous Slavery in Africas History:

Conditions and Consequences

Dirk Bezemer*, Jutta Bolt* and Robert Lensink**.

ABSTRACT

This paper is the first study to conduct an econometric analysis of indigenous slavery in

Africa. We distinguish indigenous slavery from export slavery and survey the literature in

order to identify the factors that shaped its prevalence and its impact on Africas long-term

development. We present data collected from colonial records and utilize these in a statistical

analysis. The results show that indigenous slavery was more common in Equatorial and West

Africa (specifically the Belgian colonies) and in societies with more developed states. Our

analysis also shows that indigenous slavery is robustly and negatively associated with long-

term income development. We find evidence that this effect runs via human capital formation,

especially health status.

We thank Robert Inklaar and participants in a 10 September 2008 seminar at the Utrecht School of

Economics, where an earlier version of this paper was presented, for their helpful discussions and

suggestions. Any errors are ours.

Keywords: Africa; Slavery; Pre-Colonial Societies; Long-term Economic Growth; HumanCapital

* Department of International Economics and Business, Faculty of Economics, University of

Groningen, Groningen, the Netherlands. **Department of Finance, Faculty of Economics, University

of Groningen, Groningen, the Netherlands & Development Economics Group, Wageningen University,

Wageningen, The Netherlands. Bezemer is correponding author ([email protected])
mailto:[email protected]:[email protected]:[email protected]:[email protected]


2/38

2

Indigenous Slavery in Africas History:

Conditions and Consequences

I. Introduction

This paper contributes to an expanding literature that has recently sought to identify the

impact of colonial-era conditions on long-term development in ex-colonies (Grier, 1999;

Englebert, 2000a; Acemoglu et al, 2001; Bertocchi and Canova 2002; Lange 2004; Fielding

and Torres 2008). Specifically for Africa, Englebert (2000b) attempted to link pre-colonial

institutions to the quality of post-colonial states and to long-run economic development in

tropical Africa. Gennaioli and Rainer (2007) investigated how the structure of pre-colonial

African societies affected long-term development. Bolt and Bezemer (2008) studied how

colonial-era human capital formation affects long-term growth in Sub-Saharan Africa. A

prominent argument in the African context is, of course, that its long-term development is

seriously hampered by slave trades, which took place alongside colonialization. Many papers

have studied export slavery - introduced by Arab settlers and European colonizers, and

greatly extended by the latter - and its impact on long-term development (Manning 1990;

Bairoch 1993; Nunn 2007) and some authors have assessed the impact of New World slavery

in econometric analysis (Engerman and Sokoloff, 1997, 2002; Mitchener and McLean, 2003).

But until Nunns (2008) recent path breaking study, Africas slavery had yet to be examined

in a systematic empirical fashion. Nunn (2008) estimated the numbers of slaves exported from

Africas main ports during 1400-1800 and combines this with other data to gauge the long-

term effect of export slavery on development. He finds this relation to be robustly negative,

also when controlling for geographical conditions, natural endowments and nationality of

colonizers. He also offers some evidence that export slavery may have induced ethnic

fractionalization, which in turn impeded economic growth in later times, to the present day.

The present paper extends the investigation to Africas indigenous slavery, by which we

denote customs of slavery that were innate in African societies before, during and after

colonization. Indigenous slavery is contrasted to export slavery (which was introduced and

developed in Africa by Arab and European invaders and colonizers) although indigenous and

export slavery also existed in symbiosis (Nunn, 2008; Vansina, 1989; Hilton, 1985). Wecollected data from colonial yearbooks and other records to investigate under what conditions


3/38

3

indigenous slavery arose and existed, what its relation to export slavery was, what its impact

on long-run income development is, and what the plausible channels were through which

indigenous slavery influenced long-term development. In doing so, we build on a long-

standing qualitative and descriptive literature on indigenous slavery, which is currently further

being developed by, for instance, Ayittey (2006) and Perbi (2004) 1

In the next section we define and explore African indigenous slavery. We review its linkages

with export slavery and our state of knowledge regarding its impact on economic

development. In section III we conduct a number of statistical analyses on the historical

conditions and the consequences of indigenous slavery in Africa. Previewing our results, we

find that indigenous slavery was more common in countries closer to the Equator, in West

Africa and specifically in the West-Central African Belgian colonies (present-day Rwanda,

Burundi and the Congo Democratic Republic). There is also weak evidence that it was more

prevalent among societies with more developed states and in those that had written records.

We find no clear impact of Islam or of export slavery on the prevalence of indigenous slavery,

despite frequent debates on these issues in the literature. In Section III we conduct an

econometric analysis of the consequences of indigenous slavery. We find a clearly negative

impact on present-day income levels, also if we control for geographical conditions,

nationality of colonizers, export slavery and the possible endogeneity of indigenous slavery to

income development. And in section IV we find robust evidence for the relevance of human

capital as a channel through which indigenous slavery would have influenced long-term

income development. Section V concludes with a stock taking of the results, a critical

discussion of our findings in the context of the literature, and some suggestions for future

research.

.

1The academic study of indigenous slavery in Africa is a niche research area that rapidly developed in the 1960s

and 1970s, predominantly in universities in the UK, US and in France - francophone West Africa was the hub of

Africas slave trade, both continental and for export - as well as at some African institutes (such as the Institute

for African Studies of the University of Ghana at Legon). Scholars could draw on colonial archives, eyewitness

accounts (often of colonial officials), oral traditions, and interviews: there were then still living witnesses of

indigenous slavery, which continued well into the 20th century. But the subject remains relatively under-

researched; Ewald noted in 1992 of indigenous slavery (in Islamic Africa) that it has not engaged European

and American scholars as intensely as slavery in the Atlantic world. This remains true today of Africas

indigenous slavery systems generally.


4/38

4

II. Defining and Exploring Indigenous Slavery

Slavery in Africas history can be categorized into indigenous slavery and export slavery. The

distinguishing feature that we adopt is whether or not slaves were traded beyond the

continent 2

In contrast Hopkins (1973) building on Nieboer (1900), suggested that indigenous slavery

was a response to scarcity of labour especially in West Africa, where under conditions of

simple agricultural technologies, the costs of acquiring and maintaining slaves were less than

the cost of hiring labour (Hopkins (1973:25). This economic rationalism approach assumesthe treatment of slaves as a commodity, since the slave was chattel. It also takes a dimmer

. Thus, African indigenous slavery includes both slavery and the slave trade in

Africa, which often occurred across long distances within Africa. The distinction is important

because - as we elaborate below - the conditions of indigenous slavery, its economic role, and

slaves status within African society were all very different from the conditions characteristic

of varieties of export slavery especially so in the case of the last form of export slavery that

Africa experienced, the trans-Atlantic slave trade.

Indigenous slavery has been studied by scholars in two broad frameworks: functionalism and

(economic) rationalism. In the first school, Miers and Kopytoff (1977) argued that most forms

of indigenous slavery in Africa cannot be understood simply as a commodification of people,

or even as slavery as the legal institutionalization of persons as property as Tuden and

Plotnicov (1970:12) defined it. This would be a more appropriate interpretation of ancient

Roman or modern New World slavery. Instead, Miers and Kopytoff (1977) view indigenous

slavery in Africa as part of a continuum of social relationships within a kinship system, of

which slavery was the most marginal. Terms like (private) property and (individual)

freedom, they stressed, are unhelpful to understanding traditional African society. We

elaborate on this below. With this framework comes a relative benign view on slaves social

status and living conditions.

2 A note on definitions is in order. Nunn (2008:139,159) uses the term Africas slave trades or external slavetrades to denote exclusively export slavery, using shipping records to gauge the extent of slave trade. Nunnjuxtaposes the slave trades (both Islamic and Atlantic) to domestic slavery (2008:159). It should be noted,however, that slave trade was rife also within the African continent (i.e. domestically), and many traded slaveswere never shipped and never left the continent. Nunns (2008) study is thus about export slavery as definedhere. Perbi (2001) writes of internal and external slavery, where external slavery involves slave trade within

and beyond Africa. Our export slavery is the trans-African part of Perbis (2001) external slavery. Finally, wedistinguish our usage from some authors who have denoted by indigenous slavery African involvement in theexport slave trades.


5/38

5

view of slaves circumstances. Klein (1978:601) suggests (based on fieldwork in Senegal)

that this approach might be most appropriate in high-density slavery systems.

Historical records suggest that indigenous slavery was an ingrained feature of most African

societies since they have been studied3. Many have speculated that indigenous slavery may

have been an important precursor for (and facilitator of) export slavery, and that the existence

of indigenous slave trade opened up ... societies to the temptation of the Atlantic trade

(Klein, 1978:605)4

However that may be, the indigenous slave trade from its earliest observations was part of

pan-African trade in goods including salt, copper and dates from the Sahara, and millet,

sorghum, wheat, livestock, gum, shea butter, ivory and gold from West Africa. There were

two major slave routes, one between North and West Africa, and one linking East, Central

. Perbi (2004), describing indigenous slavery in Ghana, takes the view that

it predated the Atlantic slave trade, coexisted with it from the sixteenth to the nineteenth

centuries and survived it through the early twentieth century. Perbi (2004) locates its origins

in the neolithic and iron ages, and its institutionalization with the formation and expansion of

precolonial states (Miescher, 2004:157). Others argue that in fact indigenous slavery may not

have predated New World export slavery. On the contrary, as the Guyanese historian Rodney

(1996) suggested, the Atlantic slave trade may have stimulated slavery within Africa. Nunn

(2008:159) explains that [w]hether the parts of Africa that were untouched by the Islamic

trades had chattel slavery has been the subject of an old debate (e.g., Fage, 1962; Rodney,

1970). Since this debate, evidence has been brought forth suggesting that domestic slavery

may not have existed prior to the trans-Atlantic slave trade. Nunn (2008) proceeds to

examine linguistic evidence that there were no words for slaves prior to the New Worldexport trades (see Vansina, 1989; Hilton, 1985).

3 There is early pre-colonial evidence of slavery among the Berbers of Morocco and Algeria, the Tuaregs of theSahara, the Ethiopeans, Egyptians and Somalis of northeast Africa, the Buhganda states and Nyamwezi andChagga peoples of inland East Africa, the Mrima and Omani Arabs of coastal East Africa, the Wolof of Senegaland the Gambia, the Mende and Temne of Sierra Leone, the via of Liberia, the Duala of Cameroon, the Bakongond Luande of Congo, the Lozi of Zambia and, as Perbi (2001:2) notes, virtually al the states and societies inGuinea, Ghana, Ivory Coast, Dahomey, Mali, Nigeria, etc.

4This is not to suggest that export slavery started with the Atlantic slave trades. African slaves were acquired bythe ancients Egyptians, the Greeks and the Romans and by mediaeval Europe, Arabia, the Ottoman Empire, andAsia. From the 1435 capture by the Ottomans of Constantinople which halted the flow of white slaves from theBlack Sea regions and Balkans, mediaeval Europe turned completely to Africa for its slave labour (Perbi, 2001:3;

Mc Kay et al, 1992). In modern times, export slavery was towards the Oriental, Islamic and, especially, Atlanticworlds during the 15th to 19th centuries (Perbi, 2001:4).


6/38

6

and Southern Africa. Important West African slave markets are recorded from as early as the

year 1000. Ghana obtained slaves - mainly in returns for its abundant gold resources from

the 1st to the 16th

The geographic dimension of indigenous slavery is illustrated in Table 1. This is based on

data taken from an Ethnographic Atlas compiled by Murdoch (1967), in turn based on data

previously published in the journal Ethnology. For sub-Saharan Africa, Murdoch included

292 societies, with a wealth of ethnographic data. One of these is whether a society

historically had the institution of (indigenous) slavery. The actual time periods to which this

data refer is dependent on the earliest period for which Murdoch could find reliable data,

which is from 1850 onward but not later than 1950, and varying over societies. Following

Bolt and Smits (2008), we combine this data with population data from the Atlas Narodov

Mira (1964) to assign indigenous societies to present-day countries. We so calculate the share

of the population within modern-day national borders that historically had the institution of

indigenous slavery

century. Bono Manso and Begho in Ghana were important slave markets

from AD 1000 to around 1750 (Perbi, 2001:4); others were Ouagadougou in Burkina Faso

and Bonduku and Buna in Ivory Coast.

5

5 For instance, let us assume that only the Mende and Temne peoples - who lived in lands now part of Sierra

Leone - historically had the institution of indigenous slavery, and no other ethnicities in present-day Sierra Leonedid. If the Mende make up 17 % of the population of present-day Sierra Leone and the Temne 8 %, then ourmeasure for indigenous slavery in Sierra Leone is 25 %.

. Thus, it is not a measure for actual slaves scaled by population (a

quantity which is unknown), but for population fractions historically practicing the institution

of indigenous slavery. An important distinction with measures of export slavery (such as

Nunns) is that our measure is about the country that practised indigenous slavery, while

Nunns measure relates to countries that losttheir people to countries that practised slavery.


7/38

7

Table 1: Population Fraction In Todays Borders That Historically Had Indigenous Slavery

Lesotho 0 Djibouti 0.76Madagascar 0 Zimbabwe 0.77Swaziland 0 Malawi 0.81South Africa 0.02 Burkina Faso 0.84Eq. Guinea 0.08 Central African Rep. 0.86Kenya 0.10 Gambia, The 0.88Botswana 0.13 Congo, Dem. Rep. 0.89Gabon 0.15 Ethiopia 0.90Mozambique 0.22 Uganda 0.90Sudan 0.27 Guinea 0.92Guinea-Bissau 0.34 Togo 0.92Chad 0.36 Senegal 0.93Cape Verde 0.38 Nigeria 0.94Cameroon 0.40 Sierra Leone 0.94

Liberia 0.45 Zambia 0.94Angola 0.53 Rwanda 0.98Congo, Rep. 0.66 Somalia 0.98Tanzania 0.67 Burundi 0.99Benin 0.70 Ghana 0.99Cote d'IVoire 0.70 Mali 0.99Namibia 0.72 Mauritania 0.99

Niger 0.99Source: Authors compilation based on Atlas Narodov Mira (1964) and Murdoch (1967)\

Table 1 ranks the countries in our sample by increasing values of this indigenous slaveryvariable. It shows that indigenous slavery was prevalent among the peoples of most of todays

African countries. On average, it was an indigenous institution in 63 % of the population; and

in 20 of the 43 countries in our sample, it occurred among over 80 % of the population. It is

also noteworthy that nearly all these very high values (and none of the 18 lowest-ranking

countries) are West or West-Central African countries north of the Equator. This is where

both the institution of indigenous slavery and the slave markets were concentrated.

Indigenous slaves in Africa were obtained by means of raiding and kidnapping, warfare, as

tribute paid by conquered nations or by pawning (where debtors unable to repay their dues

went in to slavery, or sent dependents into slavery). For instance, almost all the states

conquered by the famous Asante empire (in what is now Ghana) from 1700 to the end of the

19th century paid annual tributes in slaves and goods to the Asante capital of Kumasi the

state of Gonja 1000, Slaga 600, and Akwapim 1000 slaves (Perbi, 2001:5). Slavery lay at the

core of Ghanas precolonial states, whose economy was almost totally dependent on slave

labour (Perbi, 2004:111). Slaves in pre-colonial Africa were used for a variety of purposes -to work in agriculture, trade and industry; in the administration and military; for domestic


8/38

8

chores and in harems; for social prestige; for procreation; and occasionally for ritual sacrifice.

This stands in stark contrast to the vast numbers of slaves shipped out of Africa in the trans-

Atlantic trades, which were used almost exclusively for hard plantation labour.

There was an intimate connection between slavery for export and for indigenous use. Ewald

(1992:466) notes that the same networks taking millions of slaves out of Africa also

transported others within the contintent. We use our indigenous slavery variable introduced

in Table 1 to explore the relation between indigenous and export slavery, which is measured

by the total number of slaves taken from each country during various slave trades between

1400 and 1900, scaled by populations. This is a variable constructed by Nunn (2008), to

which we refer for further detail on the underlying data. In Figure 1 we plot country-level

observations of the prevalence of indigenous slavery against the logarithm of slave exports

per population. The correlation between these two measures in our sample is substantial, as a

positively sloped trend line demonstrates, but far from complete: the bivariate correlation

coefficient is only 30 % (n = 43). While we must bear in mind that the two measures are not

identical definitions of slavery prevalence and therefore will never be perfectly correlated, it

is interesting to observe that a low prevalence of indigenous slavery occurred in both high-

level and low-level export slavery environments. But a high prevalence of indigenous slavery

seems more clearly associated to intense export slavery environments, as the concentration inthe upper-right corner of Figure 2 shows. This is in line with scholarly work suggesting that

high levels of slave exports was an important driving force for the development of indigenous

slavery (Nunn, 2008; Vansina, 1989; Hilton, 1985).


9/38

9

Figure 1: Scatter Plot Of The Prevalence Of Indigenous Slavery and Slave Exports

y = 2.925x + 8.2405

R2 = 0.0923

0

2

4

6

8

10

12

14

16

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

indigenous slavery

ln(exports/population)

Sources:Atlas Narodov Mira (1964), Murdoch (1967), Nunn (2008)

African modes of indigenous slavery were very different from slavery as known in North

America, the West Indies, or Europe (including ancient Greece and Rome). Perbi (2001:12)

records how 19th century European observers expressed surprise at the humane treatment of

indigenous African slaves. The typical situation was that slaves were integrated, in various

ways, into the extended family of their owners by adoption or marriage 6

6For instance, slaves in Tuareg society were often regarded as fictive children and used kinship term to address

members of the owners family (Miers and Kopytoff, 1977: 391-403). Perbi (2001:8-9) details the different

practices to assimilate slaves into owner familes among the Makara and Bambara in Niger, the Wolof and Serer

in Senegal and the Gambia, the Bajongo, Baluba and Lunda in Central Africa and the Sena of Mozambique.

Only a few peoples are known to have not integrated slaves habitually in existing kinship structures (Perbi

mentions the Batawana of Botswana; the Ila in Zambia; the Yao in East Africa; the Duala of Cameroon; and the

Shebro of Sierra Leone).

. They usually had the

right to be fed and clothed, to marry in legal ceremonies, to earn an independent income, to

work a plot of land for own consumption, to hold and inherit property, and to have legal

protection. Slave owners had no absolute power over their slaves and were not allowed to kill


10/38

10

them at their discretion - only the king or chief could impose a death penalty on both free and

enslaved persons. In cases of maltreatment, owners could be punished in accordance with

local legal custom (e.g. fined, as in the Asante state). There are only five records of revolts of

indigenous slaves in Africa (Perbi, 2004). Among most of the African states and societies

there were also avenues for upward social and political mobility for slaves (Miers and

Koptykoff, 1977:134-170). For instance, in the Hausa-Fulani Emirates, slaves could be

appointed village heads. Slaves among the Mende of Sierra Leone could obtain the political

positions of chiefs. Asante slaves were granted occupation of stools, the traditional symbols

of authority (Perbi, 2001:11). And indigenous slaves, unlike export slaves, did not reproduce

as slaves but often married free persons and had children who were free people: children

born into slavery could expect to grow up as members of their masters lineage (Lovejoy,

1983:127).

All this is not to say that slaves and free persons were equals in Africas indigenous societies.

People entered into slavery by an act of violence which stripped them of their social identity

in their own kinship system, to be only marginally integrated into an alien kinship system.

Precisely the danger that a slave might escape in the area where (s)he had social linkages was

an important reason why slaves were transported and traded across societies (Klein,

1978:601). Slaves, while often not chattel, were exploited one way or another, moreintensively so in the high-density slave systems. The archives of Senegal contain the

account of the Governor-General in Bamako (now capital of Mali) who during a 1904 survey

of slavery in Africa was the only administrator who actually went out and talked to slaves. He

encountered consistent complaints about malnourishment among the Soninke slaves

(recounted in Klein, 1978:607). There was economic rationalization resulting in harsh slave

systems among the Soninke, Juula, Hausa and Swahili slaves comparable, Klein suggests, to

New World slavery conditions. These slaves mostly did not live with their owner families but

in separate compounds, in conditions were their membership in the owners kinship system

was a mere formality.

Furthermore, slaves had no familial rights, did not control their children and could not make

bequest decisions (Klein, 1978:602). They could be used as sacrifice or to pay off debts; were

expected to work harder and dress simply; could not freely mix with free men and women;

needed the owners permission to embark on any enterprise; and received only the simplest

burials (Perbi, 200:11). Klein (1978:602) recounts that after slavery was abolished in French


11/38

11

Africa, hundreds of thousands of slaves left their masters and went home. Those who stayed

often did so because they had no other place to go or because the costs of building up a new

life were too large; or they formed separate communities locally. These ex-slaves often

remained in relations of dependence and exploitation, as in the western Sudan were released

slaves farming for themselves generally paid their master the amount of grain necessary to

feed a person for one year. It is also noteworthy that most observations on indigenous slavery

were made in the period of colonial rule, which, as Klein (1978:601) stresses, deprived ruling

[indigenous] elites of their capacity to coerce. Thus historically, observed indigenous slavery

may have been the milder form.

Indigenous slavery lasted longer than export slavery. In 1807 Britain passed a law abolishing

the Atlantic slave trade, but the African colonies had laws against indigenous slavery passed

by their colonizers much later and at different dates: in 1874 in the Gold Coast Colony (the

southernmost part of Ghana), but not until 1908 in the Asante and Northern Territories of

Ghana (Perbi, 2001:12). And even then, indigenous slavery often continued in practice. The

abolition brought about a labour shortage that induced an increase in pawning, to which

former slaves and their descendants - often in the most vulnerable strata of society were the

natural victims (Perbi, 2004). Laws against slave trading were more strictly enforced thanlegislation on slavery, which was often a dead letter Servile labour remained important in

many areas well into the interwar period; and in a few economic backwaters it persisted even

longer, generally with the knowing complicity of colonial regimes (Klein, 1978:599,608).

Against this background we conduct an econometric analysis of the conditions under which

indigenous slavery arose and existed. The review of the literature above suggest that

indigenous slavery was more likely to be encountered in Central-West Africa (home to the

most important slave markets). We therefore include in our empirical analysis below the

degrees of longitude and the distance to the Equator (latitude) into the regression equation. As

colonizers adopted different attitudes towards indigenous slavery, the nationality of the

colonizer is plausibly relevant: we also include five binary variables for colonizer nationality

(capturing the British, French, Belgian, German and Portugese colonizing powers 7

7 The reference group comprises countries that were never colonies (Ethiopia and Liberia), Equatorial Guinea,Lybia and Namibia (a former German colony)

). Cooper

(1981) and others describe how religion generally and Islam specifically was a distinctive


12/38

12

force in the development of indigenous slavery. Thus the prevalence of indigenous slavery

may well have been higher in those areas where Islam was the dominant religion. We include

the Islamic population share to capture this. The presence of export slavery may have

stimulated the extent of indigenous slavery, as Rodney (1966), Ewald (1992) and Nunn (2008)

suggest and we include a control variable for the number of exported slaves per land area 8.

More hierarchical societies with stronger states may have induced indigenous slavery in their

lands in various ways (e.g. by asking tribute in slaves as the Asante did) and we include a

variable on state development, defined as the share of non-European population that belongs

to indigenously 'centralised' ethnic group 9 . These considerations 10 lead to a model

specification of the form

Slaveryi = C + ijXij + eij

Where Slavery

with i = 1,2,,43 and j = 1,2,,11

i is the prevalence of indigenous slavery in country i, C is a constant, ij is the

coefficient reflecting the impact of condition Xj in country i on Slavery and e ij is a white-

noise error term. We refer to the Appendix for full details on data sources and definitions. We

checked that independents are not seriously multicollinear with others 11

8 Using the numbers of exported slaves per population gives qualitatively similar results in all regressions below.9 We here follow the definition by Gennaioli and Rainer (20007), who provide more detail on data and sourcesfor this variable.10 There are other variables that we considered but did not include in the analysis. Given the variation of forms ofindigenous slavery over societal modes (Cooper, 1979; 1981), we experimented with variables on socialheterogeneity, on community organization, on the presence of written records, on population density and onstate-community relations (Gennaioli and Rainer, 2007; Bolt and Smits, 2008). But only the variable presence

of written records contains enough observations, and exploratory regressions indicated no robust correlationswith indigenous slavery for other variables. Nor did their inclusion improve the explained variation inindigenous slavery. This is actually an understandable finding taking into account that indigenous slavery waspervasive across very different societies. It is clear from the literature that the type of indigenous slavery did varyover societies; but, according to our analysis, not the generic institution of indigenous slavery itself. This study isnot about defining an the nature of African indigenous slavery, which would probably be in vain: efforts to findan essence of slavery in Africa as a whole leads away from identifying more valid distinctions between slavesystems (Cooper, 1981:274; also Cooper, 1979). Another consideration we pursued is that colonizers tended todiscourage indigenous slavery, and plausibly more so in the coastal areas where they had most influence. Weexperimented with including a variable capturing the share of the population living within a hundred kilometersfrom the coast but left it out for similar reasons: it reduced the number of observations and did not improve theexplained variation in indigenous slavery.11 Our cut off point is a Variance Inflation Factor exceeding 5. To compute this, we regress each independent on

all others and compute the R2. The Variance Inflation Factor is then 1/(1-R2), If more then 80 % of the variationof a dependent is explained by variations in other dependents (i.e. if the Variance Inflation Factor exceeds 5),then we consider this variable multicollinear. This was not the case for any of the 11 variables.

. Even so, we have 11

variables on the 43 African countries in our sample for which we have indigenous slavery

observations. We address the resulting degrees of freedom problem by estimating sequentially

six models were we each we take out groups of variables from the full model in equation (1).


13/38

13

This procedure also serves as a test on the robustness of coefficients to varying the model

specification 12. In order to account for the censored nature of our variable for indigenous

slavery (which takes on values between 0 and 100) we estimate the more efficient Tobit

specification instead of OLS 13

We find that indigenous slavery was more common in countries closer to the Equator, in West

Africa

. In table III.1 we present estimation results.

14

12Another method of model variable selection is Hendrys general-to-specific stepwise regression, sequentiallyexcluding variables with coefficient estimates of significance levels with, for instance, p > 20 %. Weexperimented with this but found that the outcome is too sensitive to the initial set, yielding too arbitrary finalvariable selection results.13 Tobit specifications give qualitatively similar but statistically more significant results compared to OLS

estimations of the same equations.14 Degrees of longitude start from zero at Greenwich (UK) and take increasingly negative values in morewesterly countries.

and specifically in the West-Central African Belgian colonies (present-day Rwanda,

Burundi, and the Congo Democratic Republic). It was more prevalent among societies with

more developed states and in those that had written records (although these are not robust

results across all model specifications). We find no clear impact of Islam or of export slavery

on the prevalence of indigenous slavery.

In conclusion of this section, we identified indigenous slavery as conceptually separate from

the export slave trade in that it is slavery and the slave trade within Africa. While indigenous

slavery in Africa did not involve the vast numbers of people traded in the Atlantic slave trades,

it was more pervasive across Africas traditional states and societies than export slavery. It is

also more recent and plausibly more deeply engrained in the fabric of society. It was mostly -

but by no means always - more benign towards slaves. But it dislocated and disenfranchised

large numbers of Africans, was a major motivation in inter-African wars, and intertwinedcommerce with warfare. While often less cruel than the Atlantic slave trades, it was arguably

a strong and pervasive impediment to the development of political stability and human capital

in traditional African states and societies before and during the colonial era, and well after

export slavery had been ended. Its pervasiveness and longevity suggest that its impact on

Africas development may have been strong, possible enduring to the present day. We now

turn to investigate this issue.


14/38

14

Table III.1: Conditions of Indigenous Slavery: Tobit regressions

Dependent variable: slavery

model (1.1) (1.2) (1.3) (1.4) (1.5) (1.6)

colony1 0.045 -0.016 -0.026 0.028 0.051

(0.094) (0.097) (0.097) (0.097) (0.093)

colony3 -0.162 -0.245 -0.348* -0.198 -0.236

(0.156) (0.184) (0.177) (0.154) (0.166)

colony4 0.332* 0.333* 0.487** 0.493*** 0.389**

(0.178) (0.179) (0.185) (0.177) (0.185)

colony5 -0.514* -0.479* -0.216 -0.456 -0.444

(0.257) (0.282) (0.293) (0.292) (0.276)

longitude -0.008*** -0.007** -0.006* -0.004 -0.009***

(0.003) (0.003) (0.003) (0.002) (0.003)

abs_latitude -0.019*** -0.020** -0.020** -0.011* -0.015*

(0.007) (0.008) (0.008) (0.006) (0.008)

State_dev 0.406** 0.417** 0.560*** 0.095 0.344*

(0.177) (0.183) (0.177) (0.156) (0.178)

writtenrec~s 0.005* 0.005* 0.004 0.002 0.006***

(0.002) (0.002) (0.003) (0.002) (0.002)

Islam 0.002 0.002 0.002 0.003 0.004***

(0.002) (0.002) (0.002) (0.002) (0.001)

ln_expor_~a 0.004 0.014 0.039** 0.014 0.016

(0.015) (0.016) (0.015) (0.014) (0.016)

_cons 0.651*** 0.609*** 0.524*** 0.313** 0.619*** 0.600***

(0.107) (0.147) (0.138) (0.131) (0.137) (0.141)

LR chi2 31.928 30.354 25.254 23.777 28.643 31.901

N 41 41 41 41 43 41

Note: ***, **, and * denote statistical significance at probability levels below 1 %, 5 % and 10 %, respectIVely.

Sources: see Appendix

III. Indigenous Slavery and Long-Term Income Development

Nunn (2008) showed statistically that export slavery had a negative impact on Africas long-term development, as measured by the GDP per capita (income) levels in the year 2000. In

this section we investigate whether this long-run effect is also observable for indigenous

slavery.

To begin with, we plot the percentage of the population within todays borders of an African

country that had the institution of indigenous slavery, against the logarithm of its per capita

income in 2000, for 43 countries (Figure 2). The negative relation (with bivariate correlationcoefficient of 55%) is already clear from visual inspection, and this is confirmed by


15/38

15

computation of a trend line with negative coefficient of -1.12. We find that nearly a third

(R2=30%) of the variation in sub-Saharan Africas current income levels is statistically

associated with variation in indigenous slavery.

Figure 2: Indigenous Slavery Correlates Negatively to 2000 Income levels in 41 African

Countries

y = -1.1191x + 7.6668

R2

= 0.2961

5

5.5

6

6.5

7

7.5

8

8.5

9

9.5

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

indigenous slavery prevalence

ln(GDP

per

capita2000)

Sources: Atlas Narodov Mira (1964), Murdoch (1967), Maddison (2000)

Our baseline OLS equations to test this relation more rigorously follow the specifications in

Nunn (2008):

ln(income2000) i = C + ijXij + eij

Where ln(income2000)

with i = 1,2,,43 and j = 1,2,,11

i is the natural logarithm of average per capita GDP in the year 2000 in

country i, C is a constant, ij is the coefficient reflecting the impact of condition Xj in country

i on year-2000 per capita income levels and eij is a white-noise error term. We refer to the

Appendix for full details on data sources and definitions. The set of Xj variables always

includes the indigenous slavery variable, and in one model also Nunns (2008) export slavery


16/38

16

variable 15. This specification allows us to examine if indigenous slavery has a long-term

growth effect apart from the effect of export slavery already identified by Nunn (2008) 16

We start with a simple univariate model (2.1). as explored in Figure 2. In model (2.2) we add

geographical position captured by longitude and latitude; in model (2.3) additionally a

number of variables capturing climate, location, legal origin and religion. Of these variables,

only longitude and coastline appear to contribute to explaining income variations, and in

model (2.4) we estimate an equation with only these control variables. We then add point

resources (oil, gold and diamonds) in model (2.5), of which only oil endowments appear to

have a significant coefficient. We include oil endowments in model (2.6), along with export

slavery and control variables for colonial origin. This is our preferred model.

.

15 Compared to the Nunn (2008) specification, this model does not include a dummy for North Africa, as wehave no North African countries in the sample.16 We also note that, in contrast to the regressions in section II above, we are not now concerned with the qualityof the model explaining current income levels. Rather, we aim to probe the effect of our variable of interest,indigenous slavery. Therefore we need not worry about the large number of variables or about colinearity of

control variables, both of which are also features of models reported in Nunn (2008, table III) who includes up to19 control variables in his model explaining current income levels by export slavery, in a sample of 52 Africancountries.


17/38

17

Table III.2: Indigenous Slavery and 2000 Income levels: OLS regressions

Dependent variable: log of GDP per capita in 2000

(2.1) (2.2) (2.3) (2.4) (2.5) (2.6)

Slavery -1.119*** -1.112*** -0.880* -1.147*** -1.010*** -0.514**(0.300) (0.358) (0.441) (0.303) (0.339) (0.227)

abs_latitude 0.011 0.012

(0.015) (0.017)

longitude -0.006 -0.006 -0.004 -0.003 -0.006

(0.004) (0.006) (0.004) (0.005) (0.004)

rain_min -0.002

(0.010)

humid_max 0.004

(0.013)

low_temp -0.006

(0.024)

ln_coastli~a 0.043 0.021 0.018(0.042) (0.031) (0.032)

island_dum -0.289

(0.529)

Islam -0.003

(0.004)

legor_fr -0.149

(0.247)

ln_avg_oil~p 0.044 0.072***

(0.031) (0.021)

ln_avg_all~p 0.028

(0.064)ln_avg_gol~p 0.002

(0.017)

ln_export_~a -0.098***

(0.024)

colony0 -0.998***

(0.276)

colony1 -0.821***

(0.177)

colony2 -0.966***

(0.180)

colony3 -1.049***

(0.307)colony4 -1.744***

(0.329)

colony5 -0.275

(0.230)

_cons 7.667*** 7.616*** 7.496*** 7.769*** 8.137*** 9.241***

(0.242) (0.419) (0.867) (0.262) (0.340) (0.245)

R2 0.296 0.330 0.372 0.323 0.385 0.724

N 43 43 43 43 43 43

Note: ***, **, and * denote statistical significance at probability levels below 1 %, 5 % and 10 %, respectively.

Robust standard errors are reported in parentheses. We use Maddisons income data, as in Nunn (2008). Using

other sources such as WDI (2008) gives qualitatively identical results. See the Appendix for data definitions and

sources.


18/38

18

We find that colonial origin variables have statistically significant coefficient estimates 17.

The baseline countries for colonial origin are Libya (a former Italian colony) and the former

German colony Namibia, which is significantly richer than the sample average (its per capita

GDP level in 2000 was US$ 3,641 compared to US$ 1,119 for the sample average).

Compared to this reference set, all formerly British (colony1), French (colony2), and Spanish

(colony3) colonies as well as Liberia and Ethiopia which were never colonised (colony0) are

significantly poorer, with negative coefficients of about the same magnitude, around one. The

former Belgian colonies Rwanda, Burundi, and the Congo Democratic Republic (colony4) are

poorer still, with a much larger negative coefficient. Also, export slavery is clearly and

negatively correlated to income in 2000, as in Nunn (2008). Most importantly, in this and in

the other five specifications indigenous slavery is also a significant and negative correlate of

variations in current per capita income levels. While it is worthy of note that its coefficient

halves when including export slavery in the model, and the total explained variation nearly

doubles, still indigenous slavery has a negative long-term growth effect which is independent

of the income-depressing effect of export slavery18

While this is prima facie evidence that indigenous slavery depressed economic development

in the long run, this is only a baseline set of estimations which requires further exploration.

There may be a selection problem, where countries destined by climate or location to remain

relatively poor also selected into the practice of indigenous slavery. One plausible selection

mechanism would run via technology. Hopkins (1973:25) attributed indigenous slavery

institutions to scarcity of labour especially in West Africa, where under conditions of simple

agricultural technologies (which restrain income growth), the costs of acquiring and

maintaining slaves were less than the cost of hiring labour. Similarly, Fielding and Torres

(2008) explain that particular combinations of endowments and climates - prevalent

especially in West and Central Africa - stimulated the development of plantation and mining

economies with their attendant extractive institutions (Acemoglu et al, 2001) inhibiting

long-term development. This is another possible selection mechanism. In such and similar

.

17 We include four colony dummies and left out the colony variable for Spansih Equitoreal Guinea (colony5).Including all five led to estimated covariance matrix of moment conditions not of full rank, which underminesreliability of the standard errors and model tests.18

We explored other specifications within the set of variables in table 2, and found that indigenous slavery isrobust also in this larger model set (and so is, incidentally, the Belgian colony variable). Results are available onrequest.


19/38

19

scenarios, we observe a negative correlation of indigenous slavery with todays income levels,

but this is no conclusive evidence of causation. Conducting a simple OLS regression would

then lead to biased estimates.

In order to assess how serious this selection problem is, we want to know if countries with

endogenous slavery were already poor at the time of observation of indigenous slavery in our

data set. Historically, as shown by (among others) Acemoglu et al (2002), Africas population

density is a good indicator for prosperity. We explored this relation for the 27 countries for

which we have data on both historical population density (in the 15th

But we cannot be certain, of course, that our OLS estimates do not suffer from endogeneity

problems in some other way. In order to address this possible problem, we apply an

instrumental-variable approach to estimating the slavery-income relation

century) and indigenous

slavery. With a correlation coefficient of +35%, it appears that if anything, there is a positive

association between the two. This is also reported by Nunn (2008) on export slavery who

finds it were the richer, not the poorer countries that selected into the slave trades.

19 . We use

instruments that are correlated with indigenous slavery, but not with the error terms in the

equation explaining present-day income. We base our specification choice on model (2.6)

above, where we now instrument indigenous slavery. We take the results in table III.1

(models 1.1 to 1.6) as a guide to the selection of instruments for indigenous slavery. We there

found that of the truly exogenous variables, longitude, latitude and colony dummies

(especially colony4) were associated with the prevalence of indigenous slavery20

. Also, we

take account of the fact that export slavery itself may be an endogenous variable. As in Nunn

(2008), we instrument it with a countrys shortest sailing distance to African coasts where

important export slave trade ports were located: the Red Sea coast, the Atlantic Ocean coast,

the Mediterranean coast and the Indian Ocean coast.

19 The instrumental-variable approach This also controls for the fact that our indigenous slavery variable is aconstructed variable, which may render OLS estimates biased. It should be noted, however, that IV regressionshave bad small sample properties. Therefore, we present the IV regressions as just an additional robustnesscheck of the OLS regression results in table III.2 We also note that we use a 2sls strategy with an OLS regressionin the first stage. We do not use a Tobit regression technique in the first stage (as we did in table III.1) since thisrequires that in the second stage, the impact of indigenous slavery is identical across countries. A two-stage leastsquares method is less efficient, but its estimates are consistent also if the impact of indigenous slavery is not

identical across countries (which is plausible).20 In addition, state development was also associated with the prevalence of indigenous slavery, but this variableis plausibly endogenous and therefore not suited as an instrument.


20/38

20

Estimations results are reported in table III.3a and III.3b below, where we report first and

second stages, respectively, of IV regressions of four models (3.1) to (3.4). In each of the four

models we report, the first-stage equation has indigenous slavery as the dependent variable. In

model (3.1) we instrument both indigenous and export slavery, so that there are two estimated

equations in the first stage, one with indigenous slavery as the dependent variable and one

with export slavery as the dependent variable. In model (3.2) we omit instrumented export

slavery. In model (3.3) we employ another instrumentation of indigenous slavery, adding the

geographical variables capturing the presence of point resources (oil and diamonds) and

measuring a countrys coastline. Finally in model (3.4.) we extend the second stage by

including all four colony dummy variables. Note that models (3.3) and (3.4) have identical

first stage specifications.

In all four models, the coefficient for instrumented indigenous slavery takes a negative value

(between -1.23 and -2.71) which is highly significant statistically. We conclude from this that

indigenous slavery was a robust long-term influence on African development, even taking

account of any endogeneity problems and controlling for the presence of export slavery. We

probe the validity of our instrumentation choices in two ways, reported in table III.3c. We ask

whether we have chosen the right instruments (Hansen J test) and whether instrumentation is

warranted at all (endogeneity test) 21

21 The Hansen J test is a test of overidentifying restrictions. The joint null hypothesis is that the instruments arevalid instruments, i.e., exclude instruments are uncorrelated with the error term. Hence a rejection of the nullhypothesis casts doubt on the validity of the instruments. In the endogeneity test, the test statistic is defined asthe difference of two Sargan-Hansen statistics: one for the equation with the smaller set of instruments where thesuspect regressor(s) are treated as endogenous, and one for the equation with the larger set of instruments, where

the suspect regressors are treated as exogenous. We consider the null hypothesis that the specified endogenousregressors can actually be treated as exogenous. If this tests fails to reject the null hypothesis, this supports theinstrumentation strategy

. Since the values for all Hansen J statistics are

statistically insignificant in the Table, we cannot reject the null hypothesis that the

instruments are valid instruments for any of the four models. This strengthens confidence in

the choice of instruments. And since the endogeneity test statistic is significant in for all

models at the 10 % cut-off point,. instrumentation seems warranted.


21/38

21

Table III.3a: IV regression of indigenous slavery on 2000 income levels: First stages

model (3.1) (3.1) (3.2) (3.3) & (3.4)

dependent Export slavery Indigenousslavery IndigenousSlavery IndigenousslaveryIndependents

Atlantic -1.258** -0.126*

(0.615) (0.070)Indian -1.230* -0.163

(0.664) (0.112)Saharan -1.501 0.024

(1.816) (0.252)red_sea -1.320 -0.462*

(1.606) (0.271)

longitude -0.137 -0.045** -0.006** -0.006**(0.116) (0.022) (0.003) (0.003)

latitude 0.006 0.007 -0.015** -0.015**

(0.080) (0.010) (0.007) (0.007)colony1 1.424 0.161 0.085 0.048

(1.992) (0.170) (0.172) (0.189)colony2 0.132 0.139 0.076 0.037

(1.963) (0.182) (0.175) (0.193)colony3 0.782 -0.051 -0.210 -0.213

(2.359) (0.214) (0.222) (0.224)colony4 -1.865 0.378 0.420* 0.350

(2.696) (0.227) (0.235) (0.275)Oil 0.241 -0.022 -0.032** -0.031**

(0.162) (0.018) (0.014) (0.015)Diamonds -0.082 0.004 -0.016 -0.017

(0.290) (0.031) (0.020) (0.020)coastline -0.010

(0.021)(constant) 33.755** 4.468** 0.508** 0.545**

(12.579) (1.910) (0.195) (0.210)

R2 0.40 0.53 0.40 0.40

N 52 43 43 43

Notes: ***, **, and * denote statistical significance at probability levels below 1 %, 5 % and 10 %, respectively.

Robust standard errors are reported in italicsparentheses. We use Maddisons income data, as in Nunn (2008).

Using other sources such as WDI (2008) gives qualitatively identical results. See the Appendix for data

definitions and sources.


22/38

22

Table III.3b: IV regression of indigenous slavery on 2000 income levels: second stages

Dependent variable: log of per capita incomes in 2000

Model (3.1) (3.2) (3.3) (3.4)

Independents

Export slavery -0.127**

(0.050)

Indigenous slavery -1.229*** -1.739*** -1.737*** -2.714**

(0.417) (0.582) (0.572) (1.178)

longitude -0.012** -0.007 -0.007 -0.016*

(0.006) (0.006) (0.006) (0.009)

Oil 0.066*** 0.033 0.033 0.018

(0.024) (0.029) (0.029) (0.035)Diamonds -0.025 0.011 0.011 -0.034

(0.038) (0.050) (0.050) (0.046)

colony1 -0.314

(0.504)

colony2 -0.456

(0.511)

colony3 -1.310**

(0.604)

colony4 -0.126

(0.803)

constant 8.790*** 8.454*** 8.453*** 9.283***

(0.357) (0.381) (0.373) (0.733)

Regression statistics

Hansen J test 6.470 6.210 6.215 0.048

(0.49) (0.18) (0.29) (0.83)

Endogeneity test 6.764** 3.447* 3.307* 5.773**

(0.03) (0.06) (0.07) (0.016)

R2 0.495 0.279 0.279 -0.017

N 43 43 43 43

Notes: ***, **, and * denote statistical significance at probability levels below 1 %, 5 % and 10 %, respectively.

Robust standard errors are reported in parentheses below the coefficients. Chi-square statistics are included in

parenthesis below Hanne J and endogeneity test statistics. We use Maddisons income data, as in Nunn (2008).

Using other sources such as WDI (2008) gives qualitatively identical results. See the Appendix for data

definitions and sources.


23/38

23

Statistical issues of significance and instrumentation aside, we may also ask what the

substantial impact of indigenous slavery on long-term development of African income levels

was. How much income development did it cost the continent? A back-of-the-envelope

calculation (detailed in appendix A) suggests that a one standard deviation increase in the

measure for indigenous slavery is equivalent to between 39 and 84 years of income growth, at

the average 1960-200 growth rate and depending on which mode coefficients are used. This

impact is in the same order of magnitude as the effect of export slavery. While the numbers

are indicative rather than precise, the exercise demonstrates that the estimated coefficients

represent substantial development effects.

IV. Indigenous Slavery and Long-Term Development: the Human Capital Channel

In view of these findings, the natural question to ask is how indigenous slavery influenced

long-term development. For export slavery, the prime candidate for a possible transmission

channel is political development. Export slavery was a major motivation in inter-African wars,

intertwined commerce with warfare and decimated populations unevenly, so hindering the

development of political stability and economic confederations in traditional African states

and societies. Indeed, Nunn (2008) persuasively argues that export slavery had long term

pernicious effects in fostering ethic tension and strife, a central factor in Africas continuing

political instability and economic stagnation. But in contrast to export slavery, it is much less

clear that indigenous slavery would have led to higher current levels of ethnic

fractionalization or political instability. Unlike export slavery, it was more evenly shared over

societies, did not lead to locally concentrated, disruptive population losses, did not upset the

political order, and underpinned rather than undermined the economic system.

A more plausible channel is a human capital channel. The term human capital was defined

by its originator Theodore Schultz (1961:1) as skills and knowledge which are a form of

capital and are in substantial part a product of deliberate investment. He also posited the

link with economic growth, observing that increases in economic output have been large in

comparison with increases in land, man-hours and physical reproducible capital. Investment

in human capital is probably the major explanation for this difference (1961:1). Health status

is one dimension of human capital, another one being education. Both have been shown bye.g. Schultz (1999) (another Schultz) to be relevant to Sub-Saharan economic development.


24/38

24

Our focus is on the health channel: indigenous slavery dislocated and disenfranchised large

numbers of Africans and - while often less cruel than the Atlantic slave trades it clearly

involved lower standards of living for slaves than for the free population, especially so in the

high-density slavery systems of the West African plantation and mining economies (Klein,

1978). Fielding and Torres (2008:1084), drawing on Galeano (1973), explain that in these

settings, the intensive use of forced labour exacerbated the effects on health of a tropical

climate that was already associated with a high mortality rate.

We would argue that the pervasiveness of indigenous slavery as well as its long duration

suggest an enduring effect across generations. It may seem far-fetched to ascribe the current

health status of African populations partly to their having had the indigenous slavery

institution up to a century ago. But actually there is ample precedent for this suggestion.

Population segments that were systematically maltreated relative to the majority population

(by earning lower incomes and having lower social status) have identifiably worse health

states and lower incomes even several generation later. Theoretically, Loury (1981) shows

that the allocation of training resources among the young people of any generation depends

upon the distribution of earnings among their parents, which in turn negatively depends on

parents health status (see also Lunberg and Satz, 1998). Also empirically, mortality rates, and

self-rated general health status are well documented for US blacks, and have been robustly

linked to their persistently lower income levels relative to the white population (Gaskell at al

2005; Markinodes and Miranda 1997; Headon et al 2003). Lower income begets worse health.

Smith and Kington (1997) use 1994 data to find that an extra US$ 1,000 of social security

income is associated with a one-half point reduction in a persons functional limitation score.

Lower health status, in turn, begets lower income and wealth levels: those with poor health

tend to leave lower inheritances to their children (Smith, 1999). These mutually enforcing

effects may persist across generations. Smith and Kington (1997) report that the health status

of past, concurrent and future generations of relatives influence ones health outcomes also in

old age. Sacerdote (2005) finds that it took two generations for the educational and

occupational gaps between descendants of free blacks and former slaves in the US to close22

22 Note that this does NOT imply that the effects of slavery on US blacks had disappeared two generations after

the abolition of slavery; but rather that by 1920 all blacks were affected equally by the remaining legacy ofslavery (Sacerdote 2005:217). That adverse legacy in terms of health, education and income disparities betweenblack and white communities remains significant to this day (e.g. Gaskell et all, 2005).

.

Foster et al (2000) found that Afro-American third-generation children remained at higher


25/38

25

risk of low birth weight and pre-term delivery than were white children (Foster et al,

2000:213). Such intergenerational health linkages can be either hereditary as e.g. in the

Barker (1990) study, which finds that adult cardiovascular disease is closely related to

neonatal mortality of siblings. Or poor health can be transmitted across generations by

socioeconomic conditions. Children in poor health are more like to have parents with low

educational attainment, linked in turn to lower incomes and lower health status (Flores, 1999).

Either way, health disadvantages tend to persist over a lifetime, to be passed on to the next

generation, to be related to educational and occupational disadvantages, and to negatively

affect income generation capacity.

In present-day African populations many people have ancestors who were indigenous slaves

living in worse conditions than the free population. Therefore these effects may still be

significant - just like US blacks are a population segment that has been historically

underprovided in terms of income, health care and social status, with measurable effects on

their health and income to this day. After all, indigenous slavery in Africa was more

widespread and is more recent than slavery of blacks in the US. Our suggestion, then, is to

pose a channel from indigenous slavery to worse long-term health outcomes, which implies a

lower quality of human capital leading to the negative effect on long-term economic

development that was observed in the preceding section.23

23 We focus on health, not on the educational aspect of human capital which is likely to be less persistent.

Sacerdote (2005) identifies intermarriage as an important factor driving the convergence in educational and

occupational differences between descendants from slaves and other blacks in the US. Since intermarriage with

free people was quite common among indigenous slaves and their children, we do not expect strong persistence

in educational and occupational disadvantages. In contrast, the health channel of transmission may be much

stronger. While the social mobility of indigenous slaves and their descendants was sometimes large, in many

contexts the differences with free people and their descendants with respect to place of residence, living

conditions, and occupational and social standing persisted.

It is well established that health

outcomes do indeed have a robust impact on economic growth in general and for reasons of

space and focus, we will not separately show that also this relation exists also in our sample.

Suffice it to note that we find strong correlation of current health outcomes (measured by the

logarithm of life expectancy at birth, averaged over the 1990s) with per capita incomes in

2000, with correlation coefficient +0.63. In turn, our indigenous slavery measure is

negatively correlated with average 1990s life expectancy with -0.31 correlation coefficients.


26/38

26

This health effect is the possible channel from indigenous slavery to lower 2000 per capita

income levels, and we therefore probe it more rigorously. We find that it is robust also when

testing it in a regression framework and adding exogenous control variables, as table IV.1

shows. In model (4.1) we introduce colony dummies as controls. In model (4.2) we add

latitude, longitude and climate variables. In model (4.3) we omit colony dummies and climate,

and add coastline and island status. In model (4.4) we add to this specification colony

dummies again. Finally in models (4.5) we include legal origin and point resources, and also

estimate these effects in the presence of colony dummies in model (4.6). In all specifications,

our measure for historical indigenous slavery is negatively and significantly correlated to life

expectancy in the 1990s 24. The models with the highest explanatory power also yield the

largest coefficients for the indigenous slavery effect on present-day life expectancy. The

magnitude of the coefficient varies between -0.11 and -0.21. The substantive significance is

that a one standard deviation (0.39) increase in the population share practicing indigenous

slavery in the 19th

24

Incidentally, this finding is also robust to including the export slavery measure - which indeed theoreticallyshould have little effect on life expectancy in our sample. It primarily affected the health of people removed fromthe populations that we study.

century is robustly associated with between 3.8 and 7.3 per cent lower life

expectancy. With a sample average of 52.7 years of life expectancy, this works out at between

2.0 and 3.9 years lower life expectancy on average in our sample. But we note that we are not

concerned only with the shorter life span, but also the implied lower lifetime health status that

this measure proxies. Human capital theory and evidence shows that this will have pervasive

and significant impacts on, among others, labour productivity and the long-term growth rate

of income.


27/38

27

Table IV.1: Historical Indigenous Slavery Causes Current Lower Life Expectancy

Dependent: logarithm of life expectancy at birth, average 1990-1999model (4.1) (4.2) (4.3) (4.4) (4.5) (4.6)

slavery -0.144** -0.212*** -0.107* -0.152*** -0.129** -0.145**

(0.057) (0.069) (0.058) (0.047) (0.062) (0.060)

colony0 -0.317** -0.201 -0.332*** -0.374**

(0.142) (0.178) (0.106) (0.170)

colony1 -0.201 -0.132 -0.134 -0.215*

(0.120) (0.139) (0.090) (0.126)

colony2 -0.150 -0.120 -0.098 -0.276

(0.119) (0.146) (0.092) (0.212)

colony3 -0.245* -0.267 -0.359*** -0.377

(0.131) (0.159) (0.100) (0.223)colony4 -0.356** -0.261 -0.216* -0.472**

(0.134) (0.166) (0.107) (0.219)

colony5 -0.337* -0.303 -0.337** -0.495*

(0.168) (0.223) (0.131) (0.256)

abs_latitude 0.003 0.003 0.002

(0.004) (0.002) (0.002)

longitude -0.002* -0.001 -0.001

(0.001) (0.001) (0.001)

rain_min -0.001

(0.002)

low_temp 0.006

(0.006)humid_max -0.003

(0.003)

ln_coastli~a 0.011 0.018***

(0.007) (0.006)

island_dum 0.190 0.352***

(0.130) (0.103)

legor_fr -0.007 0.108

(0.043) (0.171)

legor_uk 0.000 0.000

. .

ln_avg_gol~p -0.001 -0.002

(0.004) (0.004)

ln_avg_oil~p 0.001 0.002

(0.006) (0.006)

_cons 4.162*** 4.303*** 3.925*** 4.123*** 3.957*** 4.185***

(0.123) (0.173) (0.060) (0.110) (0.065) (0.135)

R2 0.392 0.506 0.361 0.724 0.129 0.407

N 43 43 43 43 43 43

Note: : ***, **, and * denote statistical significance at probability levels below 1 %, 5 % and 10 %, respectively.

Standard errors are included in parenthesis.


28/38

28

V. Summary, Discussion and Conclusions

In this paper we conducted the first systematic quantitative assessment of the long-term

impact of Africas indigenous slavery on its economic development. We document that

indigenous slavery was a pervasive institution across traditional African societies, and

construct a measure that indicates its prevalence especially in West-Central Africa north of

the Equator. Indigenous slavery lasted well into the 20 th century, much longer than export

slavery. Our review of the literature suggests that indigenous slaves mostly lived in better

health and social conditions than did slaves exported to New World and other destinations.

But it also highlight the large scale violence, social identity loss and disenfranchisement that

indigenous slaves experienced. All this suggests that it was arguably a strong and pervasive

impediment to the development of, among others, human capital in traditional African states

and societies before and during the colonial era, to the present day. This may have inhibited

economic development.

In order to research this suggestion, we use historical data in a sample of 43 countries to

estimate the conditions under which indigenous slavery existed, and its consequences for

todays health and income levels. Our first set of estimations confirms that indigenous slavery

was concentrated in West-Central Africa. They also cast doubt on the frequently suggested

link with Islam or with export slavery. The second statistical analysis shows that indigenous

slavery has been clearly harmful to long-term economic development. The effect is

statistically significant and also substantial in economic terms. We estimate that the

continuing impact of Africas indigenous slavery is about equal to of its total post-war income

growth. It is in the same order of magnitude as the long-term impact of export slavery. Our

third analysis concerns the transmission channel from indigenous slavery to current income

levels. The literature suggest that intergenerational effects of slavery and the discrimination it

entails especially on health are significant, in turn leading to lower income levels. We find

confirmation of this also in our sample, where the measure for historical indigenous slavery is

negatively and significantly correlated with life expectancy in the 1990s. These results are

robust to including various sets of control variables.

Discussion future research.


29/38

29

Variable

Appendix A: Descriptive Statistics

Obs Mean Std. Dev. Min Max

indigenous slavery 43 0.627674 0.349849 0 0.99

ln(income2000) 52 7.133545 0.825317 5.384495 9.273503

export slavery 52 3.259615 3.894839 -2.30259 8.818254

colony0 52 0.038462 0.194184 0 1

colony1 52 0.346154 0.480385 0 1

colony2 52 0.403846 0.495455 0 1

colony3 52 0.096154 0.297678 0 1

colony4 52 0.057692 0.235436 0 1

colony5 52 0.019231 0.138675 0 1

colony6 52 0.019231 0.138675 0 1

colony7 52 0.019231 0.138675 0 1

abs_latitude 52 13.55 9.858598 0.2 36

Longitude 52 16.69876 20.21442 -24.0443 57.79387

rain_min 52 8.865385 16.05509 0 69

humid_max 52 71.67308 11.94714 35 95

low_temp 52 8.75 7.488226 -9 19

ln_coastli~a 52 -0.23779 3.235199 -4.60517 6.983902

island_dum 52 0.096154 0.297678 0 1

Islam 52 35.31923 39.08479 0 100

ln_avg_gol~p 52 -7.48399 5.664152 -13.8155 3.084304

ln_avg_oil~p 52 -6.71477 4.030915 -9.21034 3.235896

ln_avg_all~p 52 -5.4901 2.396281 -6.90776 2.186849

atlantic_d~m 52 7.380799 3.280256 3.646842 16.39266

indian_dis~m 52 6.934245 4.238689 0.03191 16.77543

saharan_di~m 52 3.511205 1.567817 0.309734 6.637325

red_sea_di~m 52 3.445517 1.466612 0.06439 6.465437

state_dev 47 0.580383 0.329134 0 1

Writtenrec~s 43 10.28837 24.35616 0 97.5


30/38

30

We report in table III.4 coefficients from all OLS and IV models that we estimated,

but for brevity here discuss only the IV coefficients. An increase of one standard deviation in

the measure for indigenous slavery would result in between 43% and 95% lower income in

the IV models, depending on the Table III.3b model used. (In comparison, an increase of one

standard deviation in the measure for export slavery would result in 46 % lower income

Appendix B: How much Income Development did Indigenous Slavery Cost Africa?

What was the substantial impact ofindigenous slavery on long-term development of African

income levels was - how much income development did it cost the continent? To explore this

issue we ask how a one standard deviation change in the slavery variable would affect per

capita incomes in the year 2000. This then reflects the impact on long-term development of a

sample-specific, typical change in indigenous or export slavery.

Since the dependent variable is in logarithmic terms but the independent is not, the

coefficient is a semi-elasticity: it tells us the percentage change in 2000 per capita incomes

resulting from a one unit change in the slavery variable. To see this, note that if we estimate

the equation ln(y) = C + b*S + e (wherey, S and e are 2000 per capita incomes, slavery and

the error term, respectively), then coefficient b equals the first derivative of ln(y) with respect

to S, denoted d(lny)/dS (using symbol dfor infinitesimal changes). Because d(lny)/d(y) = 1/y,

it follows that d(lny) = d(y)/y, which is the relative change in y, expressed as a fraction.

Hence coefficient b is the ratio ofd(y)/y over dS, or the relative change in y (expressed as a

fraction) over a one unit change in S. Thus, if we consider a change in S from its sample

average 0.63 by 0.01 unit to 0.64, this would in model 3.1 (were b = -1.26) lead to a growth

decline of 0.01 * -1.26 = 0.0126, or -1.26 percentage points. An increase in S by one

standard deviation (0.35) results in a change in the 2000 income level equal to (0.35*-1.26),

which is a 43 percentage points decline (as in line one of table III.4).

25

25

.) To

put this into context, total average growth in per capita income achieved during 1960-2000 in

the same sample was 0.93 % annually. The growth loss due to a one standard variation change

in the measure for indigenous slavery is thus equivalent to between 39 and 84 years of

contemporary growth ( since ln(1.43) / ln(1.0093) = 39 and ln(2.17) / ln(1.0093) = 84),

depending on which coefficient from the Table III.3a models we use. But note that the

magnitude of their two coefficients cannot simply be compared, as one is a population

fraction and the other is (the log of) enslaved and traded persons per land area. Also the


31/38

31

number of people enslaved and exported during 1400-1800 per country population, given in

Nunn (2008), would not be comparable in value terms to the fraction of the population living

in societies that historically had the institution of indigenous slavery. All people included in

the first measure were slaves, but not all people included in the last measure were. And the

first is snapshot view in the mid-19th

century, while the second is a cumulative measure over

four centuries. We thus compare two measures, taking note of the underlying differences in

definition.

This figure also suggests that in the most conservative (model 3.1) estimate, the

income loss due to Africas indigenous slavery was equal to 71 years of contemporary

(average 1960-2000) income growth. Comparing the absence of indigenous slavery with

actual indigenous slavery implies comparing S=0 to its sample mean of 0.63. With the model

(3.1) coefficient of -1.23, this results in a change in the 2000 income level equal to (0.63*-

1.23), which is a 77 percentage points decline, equal to 71 years of contemporary growth.

But we note that this is only indicative, as we use point estimates with a margin of

uncertainty, and also as we extrapolate the coefficient originally estimated around the sample

mean. In any case, this exercise demonstrates that the estimated coefficients represent very

substantial development effects.

Table III.4: Magnitude of slavery effects on long-term income development

(a) (b) (a*b)

Model coefficient stnd.dev. income change

indigenous slavery 2.1 -1.119 0.3498 -0.392.2 -1.112 0.3498 -0.392.3 -0.88 0.3498 -0.312.4 -1.147 0.3498 -0.402.5 -1.01 0.3498 -0.352.6 -0.514 0.3498 -0.18

3.1 -1.258 0.3498 -0.433.2 -2.141 0.3498 -0.613.3 -1.800 0.3498 -0.613.4 -2.412 0.3498 -0.95

export slavery 3.1 -0.095 3.5982 -0.46

Sources: authors calculations


32/38

32

Appendix C: Definitions and Sources

Variable Source Definition


33/38

33

log of GDP pc in 2000 Maddison (2003) Real per capita GDP in 2000

colony1 Nunn (2008) British Colony

colony3 Nunn (2008) Portuguese Colony

colony4 Nunn (2008) Belgian Colony

colony5 Nunn (2008) Spain (Equatorial Guinea)

colony6 Nunn (2008) German (Namibia)

longitude Nunn (2008) The longitude of each countries centroid, measured indegrees. The centroid of each country is calculated usingthe Centroid Utility in ArcGIS. For the countries where thecentroid is located outside the land borders of the country(Cape Verde, Gambia, Mauritius, Seychelles andSomalia), a point within the country closest to the centroidis used. The location on the coast that is closest to eachcountry's centroid is identified using the Proximity Utility in

ArcGISabs_latitude Nunn (2008) The absolute value of latitude of each country's centroid

measured in degrees

state_dev Genaiolli en Rainer (2007) For each country the share of non-European populationthat belongs to indigenously 'centralised' ethnic groups

writtenrec~s Bolt en Smits (2008) For each country the share of non-European populationthat belongs to an indigenous group that had writtenrecords

islam The percent Islamic variable is the percent of a countryspopulation that is Islamic.

ln(export/area) Nunn (2008)Total number of slaves taken from each country duringvarious slave trades between 1400 and 1900, normalisedby country size (measured by land area -millions ofsquared kilometres)

rain_min Nunn (2008) The average total rainfall, in the driest month of the year,measured in millimetres.

humid_max Nunn (2008)The average of the maximum afternoon humidity,measured in percent, during the hottest month of the year

ln_coastli~a Nunn (2008) Countries' total coastline per land area measured inthousands of kilometres

island_dum Nunn (2008)

ln_avg_gol~p Nunn (2008)Natural log of the average annual gold production perthousand inhabitants from 1970 to 2000 measured in

kilograms

ln_avg_oil~p Nunn (2008)Natural log of the average annual crude petroleumproduction per thousand inhabitants from 1970 to 2000measured in thousands of tonnes

ln_avg_all~p Nunn (2008)Natural log of the average annual gemstones andindustrial diamond production per thousand inhabitantsfrom 1970 to 2000 measured in thousands of carats


34/38

34

Acemoglu, D, S Johnson, and J Robinson, The Colonial Origins of Comparative

Development: An Empirical Investigation,American Economic Review XCI (2001),

13691401

References

Ayittey, G (2006). Indigenous African Institutions

Bairoch, P (1993)

. New York: Transnational

Publishers, Inc.

Economics and World History: Myths and Paradoxes

Barker, D (199) The Fetal and Infant Origins of Adult Diseases: The Womb May be

More Important than the Home.British Medical Journal 301(676):1111

Bertocchi, Graziella, and Fabio Canova, Did Colonization Matter for Growth? An

Empirical Exploration into the Historical Causes of Africas Underdevelopment,

European Economic Review, XLVI (2002), 18511871.

. University of

Chicago Press, Chicago

Bolt, J and D Bezemer (2008) Understanding Long-Run African Growth: Colonial

Institutions or Colonial Education?Journal of Development Studies, forthcoming

Bolt, J and JP Smits (2008) Can The Nature Of Pre-Colonial Institutions Explain

Governance Quality In Africa?Mimeo, Groningen University, the Netherlands

Cooper, F (1979) The problem of Slavery in African Studies. Journal of African

History, XX: 103-125

Cooper, F (1981) Islam and Cultural Hegemony. In: Lovejoy, P. (ed.) The Ideology of

Slavery in Africa. London: Sage


35/38

35

Curtin, Philip D. (1992) The Slavery Hypothesis for Hypertension among African

Americans: The Historical Evidence.American Journal of Public Health, 82(12):

1681-1686

Dewey, J. (1916) Democracy and Education

Engerman, S.L. and K.L. Sokoloff (2000). History Lessons: Institutions, Factor

Endowments, and Paths of Development in the New World, Journal of Economic

Perspectives, 14 (3): 217-232.

. The Macmillan Company.

Djankov , S., Glaeser, E.L., La Porta, R., Lopez-de-Silanes, F., Shleifer, A. (2003)The New Comparative Economics,Journal of Comparative Economics, 31, 595-619.

Engerman, S.L. and K.L. Sokoloff (2002). Factor endowments, inequality, and paths

of development among new world economics,NBER Working Paper Series, paper no.

9259 (Cambridge-MA: NBER).

Englebert, P (2002a) Pre-Colonial Institutions, Post-Colonial States, and Economic

Development in Tropical Africa, Political Research Quarterly, LIV, 736.

Englebert, P (2002b) Solving the Mystery of the Africa Dummy, World

Development, XXVIV, 18211835.

Ewald, J (1992) Slavery in Africa and the Slave Trades from Africa: Review Article

American Historical Review, 97(2): 465-85

Fage, J (1962) An Introduction to the History of West Africa, 3r

Flores, G (1999) The Impacts of Ethnicity, Family Income, and Prenatal Education on

edition. Cambridge:

Cambridge University Press.

Fielding, D and S Torres (2008) Cows and Conquistadors: A Contribution on the

Colonial Origins of comparative Development.Journal of Development Studies 44:8,

1081:1099.


36/38

36

Childrens Health and Use of Health Services. American Journal of Public Health

89(7):1066-1072

Foster, H, L Wu, M. Bracken K. Semeny and J Thomas (2000) Intergenerational

Effects of High Socioeconomic Status on Low Birth Weight and Preterm Birth in

African Americans.Journal of the National Medical Association 92(5):213-221

Gaskin, Darrell J.; Headen Jr., Alvin E.; White-Means, Shelley I.. (2005) Racial

Disparities In Health And Wealth: The Effects Of Slavery And Past Discrimination.

Review of Black Political Economy, 32(3/4): 95-110

Gennaioli, N. and I. Rainer (2007). The Modern Impact of Precolonial Centralization

in Africa,Journal of Economic Growth 12: 185-234.

Grier, Robin M., Colonial Legacies and Economic Growth, Public Choice, XCVIV,

317335

Klein, M. (1978) The Study of Slavery in Africa.Journal of African History, XIX(4):

599-609

Lange, Matthew K., British Colonial Legacies and Political Development, World

Development, XXXII (2004), 905922.

Lovejoy, P (1981) (ed) .) The Ideology of Slavery in Africa

Lovejoy, P (1983)

. London: Sage

Transformations in Slavery

Manning, Patrick, Contours of Slavery and Social Change in Africa,American

Historical Review, LXXXVIV (1983), 835857.

. African Studies Series no. 36.

Cambridge: Cambridge University Press.

McKay, Hill and Bucklar (1992) A History of World Societies. 3 r edition


37/38

37

Markides, K and Manual Miranda (eds.) (1997) Minorities, Aging and Health

Miers, S. and I Kopytoff (eds.) (1977)

.

Thousand Oaks, CA: SAGE Publications

Slavery in Africa. Historical and

Anthropological Perspectives. University of Wisconsin Press, Madison

Miescher, S (2004) Book review of Perbi, A. (2004) A History of the Indigenous

Slavery in Ghana: From the 15th to the 19th

Mitchener, Kris James, and IanW. McLean, The Productivity of U.S. States Since

1880,Journal of Economic Growth, VIV (2003), 73114.

Century.Africa Today 2004:157-159.

Nunn, N. (2007) Historical Legacies: A Model Linking Africa's Past to Its Current

Underdevelopment.Journal Of Development Economics 83(1): 157 -175

Nunn, N. (2008) The Long-Term Effects Of Africa's Slave Trades. Quarterly Journal

of Economics 123(1):139-176

Perbi, A (2001) Slavery and the Slave Trade in Pre-Colonial Africa. Mimeo, Paper

presnetd at the University of Illinois, 5 April 2001

Perbi, A (2004) A History of the Indigenous Slavery in Ghana: From the 15 t to the

19th

Century. Accra: Sub-Saharan Publishers.

Rodney , W. (1966) African Slavery and Other Forms of Social Oppression on the

Upper Guinea Coast in the Context of the Atlantic Slave Trade.Journal of African

History, VII: 431-443

Sacerdote, B (2005) Slavery and the Intergenerational Transmission of Human Capital

The Review of Economics and Statistics 87(2): 217-234

Schultz, T.P. (1999) Health and Schooling Investments in Africa. Journal of

Economic Perspectives (13:3), pp 67-88.


38/38

Schultz, T.W. (1961) Investment in Human Capital, The American Economic

Review 51 (1), 1-17.

Smith, J and R Kington (1997) Demographic and Economic Correlates of Health in

Old Age.Demography 34(1):159-170

Smith (1999) Healthy Bodies and Thick Wallets: The Dual Relation Between Health

and Economic Status.Journal of Economic Perspectives 13(2): 145-166

Tuden, A and L Plotnicov (1970). Social Stratification in Africa

. New York.

indigenous sleavey in africa's history

Documents