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 longterm 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; Human Capital

* 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 (d.j.bezemer@rug.nl)

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 longterm 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). We collected data from colonial yearbooks and other records to investigate under what conditions 2

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 longstanding 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.

The 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 underresearched; 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.

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 . 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. 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 assumes the treatment of slaves as a commodity, since the slave was chattel. It also takes a dimmer

A note on definitions is in order. Nunn (2008:139,159) uses the term Africas slave trades or external slave trades to denote exclusively export slavery, using shipping records to gauge the extent of slave trade. Nunn juxtaposes 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 slaves were never shipped and never left the continent. Nunns (2008) study is thus about export slavery as defined here. 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, we distinguish our usage from some authors who have denoted by indigenous slavery African involvement in the export slave trades.

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 studied 3. 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. 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 World export trades (see Vansina, 1989; Hilton, 1985). 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

There is early pre-colonial evidence of slavery among the Berbers of Morocco and Algeria, the Tuaregs of the Sahara, the Ethiopeans, Egyptians and Somalis of northeast Africa, the Buhganda states and Nyamwezi and Chagga peoples of inland East Africa, the Mrima and Omani Arabs of coastal East Africa, the Wolof of Senegal and the Gambia, the Mende and Temne of Sierra Leone, the via of Liberia, the Duala of Cameroon, the Bakongo nd Luande of Congo, the Lozi of Zambia and, as Perbi (2001:2) notes, virtually al the states and societies in Guinea, Ghana, Ivory Coast, Dahomey, Mali, Nigeria, etc. This is not to suggest that export slavery started with the Atlantic slave trades. African slaves were acquired by the ancients Egyptians, the Greeks and the Romans and by mediaeval Europe, Arabia, the Ottoman Empire, and Asia. From the 1435 capture by the Ottomans of Constantinople which halted the flow of white slaves from the Black 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, Atlantic worlds during the 15th to 19th centuries (Perbi, 2001:4).

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 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.

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 5 . 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 lost their people to countries that practised slavery.

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 Leone did. If the Mende make up 17 % of the population of present-day Sierra Leone and the Temne 8 %, then our measure for indigenous slavery in Sierra Leone is 25 %.

Table 1: Population Fraction In Todays Borders That Historically Had Indigenous Slavery Lesotho Madagascar Swaziland South Africa Eq. Guinea Kenya Botswana Gabon Mozambique Sudan Guinea-Bissau Chad Cape Verde Cameroon Liberia Angola Congo, Rep. Tanzania Benin Cote d'IVoire Namibia 0 0 0 0.02 0.08 0.10 0.13 0.15 0.22 0.27 0.34 0.36 0.38 0.40 0.45 0.53 0.66 0.67 0.70 0.70 0.72 Djibouti Zimbabwe Malawi Burkina Faso Central African Rep. Gambia, The Congo, Dem. Rep. Ethiopia Uganda Guinea Togo Senegal Nigeria Sierra Leone Zambia Rwanda Somalia Burundi Ghana Mali Mauritania Niger 0.76 0.77 0.81 0.84 0.86 0.88 0.89 0.90 0.90 0.92 0.92 0.93 0.94 0.94 0.94 0.98 0.98 0.99 0.99 0.99 0.99 0.99

Source: 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 slavery variable. 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 7

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 transAtlantic 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 highlevel 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 in the 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).

Figure 1: Scatter Plot Of The Prevalence Of Indigenous Slavery and Slave Exports
16 14

ln(exports/population)

12 10 8 6 4 2 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 indigenous slavery

y = 2.925x + 8.2405 R2 = 0.0923

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. 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

For 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).

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, more intensively 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

10

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 than legislation 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). Cooper (1981) and others describe how religion generally and Islam specifically was a distinctive

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

11

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 specification of the form
10

lead to a model

Slaveryi = C + ij X ij + e ij

with

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

Where Slaveryi is the prevalence of indigenous slavery in country i, C is a constant, ij is the coefficient reflecting the impact of condition X j in country i on Slavery and e ij is a whitenoise 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. 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).
8 9

Using the numbers of exported slaves per population gives qualitatively similar results in all regressions below. We here follow the definition by Gennaioli and Rainer (20007), who provide more detail on data and sources for this variable. 10 There are other variables that we considered but did not include in the analysis. Given the variation of forms of indigenous slavery over societal modes (Cooper, 1979; 1981), we experimented with variables on social heterogeneity, on community organization, on the presence of written records, on population density and on state-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 correlations with indigenous slavery for other variables. Nor did their inclusion improve the explained variation in indigenous slavery. This is actually an understandable finding taking into account that indigenous slavery was pervasive across very different societies. It is clear from the literature that the type of indigenous slavery did vary over societies; but, according to our analysis, not the generic institution of indigenous slavery itself. This study is not about defining an the nature of African indigenous slavery, which would probably be in vain: efforts to find an essence of slavery in Africa as a whole leads away from identifying more valid distinctions between slave systems (Cooper, 1981:274; also Cooper, 1979). Another consideration we pursued is that colonizers tended to discourage indigenous slavery, and plausibly more so in the coastal areas where they had most influence. We experimented with including a variable capturing the share of the population living within a hundred kilometers from the coast but left it out for similar reasons: it reduced the number of observations and did not improve the explained 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 variation of 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.

12

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. In table III.1 we present estimation results.

We find that indigenous slavery was more common in countries closer to the Equator, in West Africa 14 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 intertwined commerce 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.

Another method of model variable selection is Hendrys general-to-specific stepwise regression, sequentially excluding variables with coefficient estimates of significance levels with, for instance, p > 20 %. We experimented with this but found that the outcome is too sensitive to the initial set, yielding too arbitrary final variable 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 more westerly countries.

12

13

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

Dependent variable: slavery model colony1 colony3 colony4 colony5 longitude abs_latitude State_dev writtenrec~s Islam ln_expor_~a _cons (1.1) 0.045 (0.094) -0.162 (0.156) 0.332* (0.178) -0.514* (0.257) -0.008*** (0.003) -0.019*** (0.007) 0.406** (0.177) 0.005* (0.002) 0.002 (0.002) 0.004 (0.015) 0.651*** (0.107) (1.2) (1.3) -0.016 (0.097) -0.245 (0.184) 0.333* (0.179) -0.479* (0.282) -0.007** (0.003) -0.020** (0.008) 0.417** (0.183) 0.005* (0.002) 0.002 (0.002) 0.014 (0.016) 0.609*** (0.147) (1.4) -0.026 (0.097) -0.348* (0.177) 0.487** (0.185) -0.216 (0.293) (1.5) 0.028 (0.097) -0.198 (0.154) 0.493*** (0.177) -0.456 (0.292) -0.004 (0.002) -0.011* (0.006) (1.6) 0.051 (0.093) -0.236 (0.166) 0.389** (0.185) -0.444 (0.276) -0.009*** (0.003) -0.015* (0.008) 0.344* (0.178) 0.006*** (0.002)

-0.006* (0.003) -0.020** (0.008) 0.560*** (0.177) 0.004 (0.003) 0.002 (0.002) 0.039** (0.015) 0.524*** (0.138)

0.095 (0.156) 0.002 (0.002) 0.003 (0.002) 0.014 (0.014) 0.313** (0.131)

0.004*** (0.001) 0.016 (0.016) 0.619*** (0.137)

0.600*** (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 longterm 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 correlation coefficient of 55%) is already clear from visual inspection, and this is confirmed by 14

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
9.5 9

ln(GDP per capita 2000)

8.5 8 7.5 7 6.5 6 5.5 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 indigenous slavery prevalence

y = -1.1191x + 7.6668 R2 = 0.2961

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 + ij X ij + e ij

with

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

Where ln(income2000) 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 X j in country i on year-2000 per capita income levels and e ij is a white-noise error term. We refer to the Appendix for full details on data sources and definitions. The set of X j variables always includes the indigenous slavery variable, and in one model also Nunns (2008) export slavery

15

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.

Compared to the Nunn (2008) specification, this model does not include a dummy for North Africa, as we have 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 quality of 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 to 19 control variables in his model explaining current income levels by export slavery, in a sample of 52 African countries.

15

16

Table III.2: Indigenous Slavery and 2000 Income levels: OLS regressions
Dependent variable: log of GDP per capita in 2000 (2.1) Slavery abs_latitude longitude rain_min humid_max low_temp ln_coastli~a island_dum Islam legor_fr ln_avg_oil~p ln_avg_all~p ln_avg_gol~p ln_export_~a -1.119*** (0.300) (2.2) -1.112*** (0.358) 0.011 (0.015) -0.006 (0.004) (2.3) -0.880* (0.441) 0.012 (0.017) -0.006 (0.006) -0.002 (0.010) 0.004 (0.013) -0.006 (0.024) 0.043 (0.042) -0.289 (0.529) -0.003 (0.004) -0.149 (0.247) (2.4) -1.147*** (0.303) (2.5) -1.010*** (0.339) (2.6) -0.514** (0.227)

-0.004 (0.004)

-0.003 (0.005)

-0.006 (0.004)

0.021 (0.031)

0.018 (0.032)

0.044 (0.031) 0.028 (0.064) 0.002 (0.017)

0.072*** (0.021)

-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.

17

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 slavery 18.

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

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 undermines reliability 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 is robust also in this larger model set (and so is, incidentally, the Belgian colony variable). Results are available on request.

17

18

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 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.

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
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 slavery 20. 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.

The instrumental-variable approach This also controls for the fact that our indigenous slavery variable is a constructed variable, which may render OLS estimates biased. It should be noted, however, that IV regressions have bad small sample properties. Therefore, we present the IV regressions as just an additional robustness check of the OLS regression results in table III.2 We also note that we use a 2sls strategy with an OLS regression in the first stage. We do not use a Tobit regression technique in the first stage (as we did in table III.1) since this requires that in the second stage, the impact of indigenous slavery is identical across countries. A two-stage least squares 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 variable is plausibly endogenous and therefore not suited as an instrument.

19

19

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

. 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.

The Hansen J test is a test of overidentifying restrictions. The joint null hypothesis is that the instruments are valid instruments, i.e., exclude instruments are uncorrelated with the error term. Hence a rejection of the null hypothesis casts doubt on the validity of the instruments. In the endogeneity test, the test statistic is defined as the difference of two Sargan-Hansen statistics: one for the equation with the smaller set of instruments where the suspect 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 endogenous regressors can actually be treated as exogenous. If this tests fails to reject the null hypothesis, this supports the instrumentation strategy

21

20

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

model dependent
Independents Atlantic Indian Saharan red_sea longitude latitude colony1 colony2 colony3 colony4 Oil Diamonds coastline (constant)

(3.1) Export slavery

(3.1) Indigenous slavery

(3.2) Indigenous Slavery

(3.3) & (3.4) Indigenous slavery

-1.258** (0.615) -1.230* (0.664) -1.501 (1.816) -1.320 (1.606) -0.137 (0.116) 0.006 (0.080) 1.424 (1.992) 0.132 (1.963) 0.782 (2.359) -1.865 (2.696) 0.241 (0.162) -0.082 (0.290)

-0.126* (0.070) -0.163 (0.112) 0.024 (0.252) -0.462* (0.271) -0.045** (0.022) 0.007 (0.010) 0.161 (0.170) 0.139 (0.182) -0.051 (0.214) 0.378 (0.227) -0.022 (0.018) 0.004 (0.031)

-0.006** (0.003) -0.015** (0.007) 0.085 (0.172) 0.076 (0.175) -0.210 (0.222) 0.420* (0.235) -0.032** (0.014) -0.016 (0.020)

33.755** (12.579)

4.468** (1.910)

0.508** (0.195)

-0.006** (0.003) -0.015** (0.007) 0.048 (0.189) 0.037 (0.193) -0.213 (0.224) 0.350 (0.275) -0.031** (0.015) -0.017 (0.020) -0.010 (0.021) 0.545** (0.210)

R2 N

0.40 52

0.53 43

0.40 43

0.40 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.

21

Table III.3b: IV regression of indigenous slavery on 2000 income levels: second stages
Dependent variable: log of per capita incomes in 2000 Model Independents Export slavery Indigenous slavery longitude Oil Diamonds colony1 colony2 colony3 colony4 constant 8.790*** (0.357) 8.454*** (0.381) 8.453*** (0.373) -0.127** (0.050) -1.229*** (0.417) -0.012** (0.006) 0.066*** (0.024) -0.025 (0.038)

(3.1)

(3.2)

(3.3)

(3.4)

-1.739*** (0.582) -0.007 (0.006) 0.033 (0.029) 0.011 (0.050)

-1.737*** (0.572) -0.007 (0.006) 0.033 (0.029) 0.011 (0.050)

-2.714** (1.178) -0.016* (0.009) 0.018 (0.035) -0.034 (0.046) -0.314 (0.504) -0.456 (0.511) -1.310** (0.604) -0.126 (0.803) 9.283*** (0.733)

Regression statistics Hansen J test 6.470 (0.49) 6.764** (0.03) 0.495 43 6.210 (0.18) 3.447* (0.06) 0.279 43 6.215 (0.29) 3.307* (0.07) 0.279 43 0.048 (0.83) 5.773** (0.016) -0.017 43

Endogeneity test

R2 N

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.

22

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 by e.g. Schultz (1999) (another Schultz) to be relevant to Sub-Saharan economic development. 23

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 close 22. Foster et al (2000) found that Afro-American third-generation children remained at higher

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 of slavery (Sacerdote 2005:217). That adverse legacy in terms of health, education and income disparities between black and white communities remains significant to this day (e.g. Gaskell et all, 2005).

22

24

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 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.

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.

25

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 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.

24

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

26

Table IV.1: Historical Indigenous Slavery Causes Current Lower Life Expectancy
Dependent: logarithm of life expectancy at birth, average 1990-1999

model slavery colony0 colony1 colony2 colony3 colony4 colony5 abs_latitude longitude rain_min low_temp humid_max ln_coastli~a island_dum legor_fr legor_uk ln_avg_gol~p ln_avg_oil~p _cons

(4.1) -0.144** (0.057) -0.317** (0.142) -0.201 (0.120) -0.150 (0.119) -0.245* (0.131) -0.356** (0.134) -0.337* (0.168)

(4.2) -0.212*** (0.069) -0.201 (0.178) -0.132 (0.139) -0.120 (0.146) -0.267 (0.159) -0.261 (0.166) -0.303 (0.223) 0.003 (0.004) -0.002* (0.001) -0.001 (0.002) 0.006 (0.006) -0.003 (0.003)

(4.3) -0.107* (0.058)

(4.4) -0.152*** (0.047) -0.332*** (0.106) -0.134 (0.090) -0.098 (0.092) -0.359*** (0.100) -0.216* (0.107) -0.337** (0.131) 0.002 (0.002) -0.001 (0.001)

(4.5) -0.129** (0.062)

(4.6) -0.145** (0.060) -0.374** (0.170) -0.215* (0.126) -0.276 (0.212) -0.377 (0.223) -0.472** (0.219) -0.495* (0.256)

0.003 (0.002) -0.001 (0.001)

0.011 (0.007) 0.190 (0.130)

0.018*** (0.006) 0.352*** (0.103) -0.007 (0.043) 0.000 . -0.001 (0.004) 0.001 (0.006) 3.957*** (0.065) 0.108 (0.171) 0.000 . -0.002 (0.004) 0.002 (0.006) 4.185*** (0.135)

4.162*** (0.123)

4.303*** (0.173)

3.925*** (0.060)

4.123*** (0.110)

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.

27

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 20th 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.

28

Appendix A: Descriptive Statistics

Variable Obs Mean Std. Dev. Min Max

indigenous slavery ln(income2000) export slavery colony0 colony1 colony2 colony3 colony4 colony5 colony6 colony7 abs_latitude Longitude rain_min humid_max low_temp ln_coastli~a island_dum Islam ln_avg_gol~p ln_avg_oil~p ln_avg_all~p atlantic_d~m indian_dis~m saharan_di~m red_sea_di~m state_dev Writtenrec~s

43 52 52 52 52 52 52 52 52 52 52 52 52 52 52 52 52 52 52 52 52 52 52 52 52 52 47 43

0.627674 7.133545 3.259615 0.038462 0.346154 0.403846 0.096154 0.057692 0.019231 0.019231 0.019231 13.55 16.69876 8.865385 71.67308 8.75 -0.23779 0.096154 35.31923 -7.48399 -6.71477 -5.4901 7.380799 6.934245 3.511205 3.445517 0.580383 10.28837

0.349849 0.825317 3.894839 0.194184 0.480385 0.495455 0.297678 0.235436 0.138675 0.138675 0.138675 9.858598 20.21442 16.05509 11.94714 7.488226 3.235199 0.297678 39.08479 5.664152 4.030915 2.396281 3.280256 4.238689 1.567817 1.466612 0.329134 24.35616

0 5.384495 -2.30259 0 0 0 0 0 0 0 0 0.2 -24.0443 0 35 -9 -4.60517 0 0 -13.8155 -9.21034 -6.90776 3.646842 0.03191 0.309734 0.06439 0 0

0.99 9.273503 8.818254 1 1 1 1 1 1 1 1 36 57.79387 69 95 19 6.983902 1 100 3.084304 3.235896 2.186849 16.39266 16.77543 6.637325 6.465437 1 97.5

29

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

What was the substantial impact of indigenous 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 (where y, 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 d for 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 of d(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). 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 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

25

30

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 Model indigenous slavery 2.1 2.2 2.3 2.4 2.5 2.6 3.1 3.2 3.3 3.4 export slavery Sources: authors calculations 3.1 (a) coefficient -1.119 -1.112 -0.88 -1.147 -1.01 -0.514 -1.258 -2.141 -1.800 -2.412 -0.095 (b) stnd.dev. 0.3498 0.3498 0.3498 0.3498 0.3498 0.3498 0.3498 0.3498 0.3498 0.3498 3.5982 (a*b) income change -0.39 -0.39 -0.31 -0.40 -0.35 -0.18 -0.43 -0.61 -0.61 -0.95 -0.46

31

Appendix C: Definitions and Sources

Variable Source Definition

32

log of GDP pc in 2000 colony1 colony3 colony4 colony5 colony6 longitude

Maddison (2003) Nunn (2008) Nunn (2008) Nunn (2008) Nunn (2008) Nunn (2008) Nunn (2008)

Real per capita GDP in 2000 British Colony Portuguese Colony Belgian Colony Spain (Equatorial Guinea) German (Namibia) The longitude of each countries centroid, measured in degrees. The centroid of each country is calculated using the Centroid Utility in ArcGIS. For the countries where the centroid is located outside the land borders of the country (Cape Verde, Gambia, Mauritius, Seychelles and Somalia), a point within the country closest to the centroid is used. The location on the coast that is closest to each country's centroid is identified using the Proximity Utility in ArcGIS The absolute value of latitude of each country's centroid measured in degrees For each country the share of non-European population that belongs to indigenously 'centralised' ethnic groups For each country the share of non-European population that belongs to an indigenous group that had written records The percent Islamic variable is the percent of a countrys population that is Islamic.

abs_latitude state_dev writtenrec~s

Nunn (2008) Genaiolli en Rainer (2007) Bolt en Smits (2008)

islam ln(export/area) Nunn (2008)

Total number of slaves taken from each country during various slave trades between 1400 and 1900, normalised by country size (measured by land area -millions of squared kilometres) The average total rainfall, in the driest month of the year, measured in millimetres. The average of the maximum afternoon humidity, measured in percent, during the hottest month of the year

rain_min humid_max

Nunn (2008) Nunn (2008)

ln_coastli~a island_dum ln_avg_gol~p

Nunn (2008) Nunn (2008) Nunn (2008)

Countries' total coastline per land area measured in thousands of kilometres

Natural log of the average annual gold production per thousand inhabitants from 1970 to 2000 measured in kilograms Natural log of the average annual crude petroleum production per thousand inhabitants from 1970 to 2000 measured in thousands of tonnes Natural log of the average annual gemstones and industrial diamond production per thousand inhabitants from 1970 to 2000 measured in thousands of carats

ln_avg_oil~p

Nunn (2008)

ln_avg_all~p

Nunn (2008)

33

References

Acemoglu, D, S Johnson, and J Robinson, The Colonial Origins of Comparative Development: An Empirical Investigation, American Economic Review XCI (2001), 13691401

Ayittey, G (2006). Indigenous African Institutions. New York: Transnational Publishers, Inc.

Bairoch, P (1993) Economics and World History: Myths and Paradoxes. University of Chicago Press, Chicago

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.

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

34

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. 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, 595619. 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.

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, 3rd 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.

Flores, G (1999) The Impacts of Ethnicity, Family Income, and Prenatal Education on 35

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. London: Sage

Lovejoy, P (1983) Transformations in Slavery. African Studies Series no. 36. Cambridge: Cambridge University Press.

Manning, Patrick, Contours of Slavery and Social Change in Africa, American Historical Review, LXXXVIV (1983), 835857. McKay, Hill and Bucklar (1992) A History of World Societies. 3rd edition

36

Markides, K and Manual Miranda (eds.) (1997) Minorities, Aging and Health. Thousand Oaks, CA: SAGE Publications

Miers, S. and I Kopytoff (eds.) (1977) 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 Century. Africa Today 2004:157-159.

Mitchener, Kris James, and IanW. McLean, The Productivity of U.S. States Since 1880, Journal of Economic Growth, VIV (2003), 73114.

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 15th 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. 37

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.

38