WEST AFRICA: EUROPEAN CONTACT AND ECONOMIC DEVELOPMENT BEFORE 1900

The economic history of West Africa helps to frame the investigation of the impact of European trade, slavery, and early colonialism on human capital development in Western Sudan. An aspect that we will assess later is the heterogeneity of regional developments within this region. Indeed, some of the existing quantitative accounts of pre-colonial and colonial West Africa provide a gloomy view on the role played by Europeans, whereas other contributions point out more positive aspects.

Inikori (Reference Inikori2007) claims that the rise of the Atlantic economy in the eighteenth century prompted the growth of a regional divide within Western Africa. Prior to European contact, West Africa had experienced the rise of a well-integrated regional market. After the establishment of European outposts, however, the coastal regions of West Africa became more connected to the Atlantic economy, whereas the interior was more isolated and experienced growing conflicts and slave raids, following a growing demand for slaves from the Americas (see Richardson Reference Richardson1989 on the evolution of Atlantic slave trades between 1700 and 1810). Klas Rönnbäck (Reference Rönnbäck2015) critically discussed the hypothesis that slavery was an important determinant of the pattern of regional development within the area. He finds that slavery had only a small influence on total agricultural production in the Gold Coast, mostly to feed slaves and crews, although this small impact did tend to be concentrated in the coastal regions. Since we explore the effect of peanut farming on numeracy later, we expect to find a heterogeneous impact as well.

Scholars have explored a number of positive developments linked to European influence and trade. For example, Gareth Austin (Reference Austin and Baten2016) has pointed out the possibility that West African farmers might have benefitted from the new trading opportunities provided by European trade networks. Expanding peanut production after the 1840s in the Gambia, Sine-Salum, the Rufisque and Petite-Cote region, in Cayor and along the lower Senegal was done mainly by free labor.Footnote ¹ It also stimulated the development of a host of activities, such as transportation, marketing, and food provision (Atmore Reference Atmore, Roland and Sanderson1985; Brooks Reference Brooks1975; Searing Reference Searing1993, pp. 187–8). However, this latter point is still a debated issue in the context of the rise of peanut production in West Africa. Evidence from early colonial Kano in Nigeria shows that African initiatives were central to the rise of peanut as a cash crop, but also that slavery was heavily used to increase production (Salau Reference Salau2010). Joye L. Bowman (Reference Bowman1987) confirmed the importance of slavery for the rise of groundnut production in Portuguese Guinea, which was done on relatively large estates by African producers. She also stressed that migrant labor was important; yet, these workers could not accumulate property: “more often than not […] the workers had very little of their peanut harvest left after clearing their debt [with the landlord]” (Bowman Reference Bowman1987, p. 97).

Evidence on the economic and social impact of the peanut trade in Senegambia is more controversial with respect to other West African regions. Bernard Moitt (Reference Moitt1989) argued that slavery played an important role. However, his evidence is based on a clear regional pattern. Slaves were primarily imported from Western Mali (French Sudan) through a route following the Senegal River, and they ended up in the peanut basin of Cayor and Baol. Furthermore, Moitt made a clear connection between Wolof societies and families, who inhabited mostly the northwest of Senegambia, and the use of slaves. Technological constraints and factor endowments did not allow an increase in productivity, so that extensive growth based on more land and more labor was the only chance to reap the benefits of increasing foreign demand.

Factor endowments and the organization of society itself differed in other regions of Senegal, such as Sine and Salum. Babacar Fall (Reference Fall2010) provides qualitative evidence suggesting that the impact of peanut production on society and the economy of the Senegambia was positive, especially in the coastal regions of today's Senegal, such as Salum. First, he notes that production was often organized at the family level, using labor from the household, which was possible given high population density in some rural regions of today's Senegal (high population densities in some regions are confirmed by Rousseau Reference Rousseau1929, pp. 400–3).

Fall also argues that the growth of peanut exports increased the peasants' living standard. According to him, up to 1930, peasants of some rural areas in Senegal earned more than laborers in Dakar (idem, p. 59). Furthermore, Fall (Reference Fall2010) notes that even though labor migrations occurred in Senegambia before 1900, there was a clear regional pattern. People migrated temporarily from the north and northeast of Senegal (Jolof and Cayor mainly) towards the south, such as Kaolack in the Salum region (idem, p. 62).

The main hypothesis that slavery, early colonialism, and European trade brought about a growing regional divide—especially as peanut production benefitted only specific groups and regions—can be assessed quantitatively by studying the development of human capital in the area of Senegal, Gambia, and Western Mali. The French outposts of St. Louis and Gorée, along the coast of today's Senegal, as shown in Figure 1, grew quickly as important centers for slavery and trade during the eighteenth century (Becker et al. Reference Becker, Martin and Schmitz1983). Their growing integration in the Atlantic economy fits Inikori's hypothesis of economic divergence between the coast and the interior of West Africa: Gorée became particularly important in the eighteenth century, as the island was used to gather slaves who would soon be deported to the Americas (Knight-Baylac Reference Knight-Baylac1970, Reference Knight-Baylac1977). Consistent with the evidence offered by James F. Searing (Reference Searing1993), the data we discuss later shows that captives in the two French outposts often performed semi-skilled occupations such as carpenters, sailors, and textile workers among men and laundrywomen, spinners, and cleaners among women. This pattern of specialization was brought about by the presence of very few Europeans on the Atlantic coast, who needed rather skilled labor to be used within the growing market economy that was brought about by the Atlantic slave trade.

The long-term contact with Europeans made society along the West coast of Senegambia quite diverse and mixed. In 1794, the freeborn inhabitants of Gorée and St. Louis were formally granted French citizenship, extended in 1848 to Rufisque and Dakar (Idowu Reference Idowu1969). These towns, the so-called Four Communes, had mixed populations, which can be divided in three main categories: Europeans (mostly French), black Africans from Western Sudan and descendants of African women and European traders (called métis). The latter would later become a distinct class of the Four Communes, known as the habitants. They enjoyed French citizenship, inherited their European fathers' names and to a large extent they also inherited their fathers' European culture due to their involvement in business, attendance of schools, and practice of Catholicism (Searing Reference Searing1993). Therefore, we expect to observe high values of numeracy in these regions.

Other aspects of colonialism, such as military expansion and the violent annexation of new territory, clearly had a more negative impact on Africans. During the mid-nineteenth century, the Four Communes served as bases in France's expansion in Western Sudan led by governors Faidherbe and Brier de L'Isle. The French first annexed part of the region of Cayor and established a few posts along the Senegal River by defeating the Walo. During the 1880s and early 1890s, other regions south and southeast of Gorée and St. Louis became part of the French colony. By 1891, the last pockets of independence were eliminated in Jolof, Futa Toro, and Cayor (Robinson Reference Robinson2000, p. 63). Salum was also formally under French rule, although France's presence “was minimal. […] In 1890, France was officially represented in the cercle of Sine-Salum by one lone Administrator. But French military power was in the wings, always available, though seldom visible” (Klein Reference Klein1968, p. 150; Barry Reference Barry1998, who describes France's involvement in Sine-Salum as “half-hearted”). The Dakar-Saint-Louis railway, completed in 1885, allowed further control over the new territories. Casamance was the last region to be annexed by the French during the early twentieth century. We expect to observe a negative relationship between military conflicts and numeracy in our analysis later.

Besides European trade and early contact, local endowments shaped the features of West African societies and, as a result, the way that they reacted to external opportunities and constraints. In particular, the difference between the Wolof society and that of the Serer stands out (Barry Reference Barry1998, pp. 16–17). The Wolof inhabited the area of today's metropolitan Dakar and the regions of Cayor, Baol, Walo, Futa Toro, and Jolof. Highly mobile, they were characterized by a hierarchical society. Slaves were used to cope with the poor quality of soils in the dry savanna environment, as well as the limited availability of labor to dig wells (Searing Reference Searing1993, p. 3). By contrast, the Serer lived in the heavily forested and well-watered region of Sine and, especially, Salum. They were able to use land intensively and more than once a year, due to their heavy reliance on manure produced by cattle that were held in soon-to-be-cultivated fields.

South of the Gambia, the Joola lived in the western area of the Casamance. Because of their location along the coast, a large share of their population was enslaved and deported to the Americas. Finally, the Mandingos were heirs to the great Mali Empire and inhabited a large area that stretched between Mali and southeast Senegal, including portions of both Portuguese and French Guinea to the south. They had a long history of owning and maintaining slaves and were characterized as a society based on castes with a comparatively inefficient agricultural technology (Curtin Reference Curtin1975, p. 29).

The different societies and polities of Senegambia experienced several changes after the first contact with Europeans. First, the rise of the slave trade in the eighteenth century brought about growing instability in the kingdoms of the former Jolof Empire, where the response to the growing demand for slaves initiated a violent age of warlords, called ceddo (Getz Reference Getz2004, pp. 12–13). During the nineteenth century, the ceddo aristocracy fought the rise of the marabouts (Islamic religious leaders) following waves of Muslim migration from the region north of the Senegal River, which put even more pressure on the cohesiveness of pre-colonial kingdoms. The hierarchical structure of the Wolof likely favored the rise of the ceddo and of internal conflicts, whereas more stability and cohesiveness in Sine and Salum were favored by the equal and uniform society of the Serer. As Boubacar Barry (Reference Barry1998, pp. 87–88) notes, “[…] since the history of SiinFootnote ² was shaped by the narrowness of its boundaries, the context of generalized violence forced it to grow even more isolationist and to develop an agricultural system of shifting cultivation in a bid to maintain internal cohesion.” According to Barry, the society of Salum was slightly more mixed than that of Sine given its large territory, which helped Islam to gain strength during the late-nineteenth century. Yet, this shift was not accompanied by violent campaigns by the French to maintain control of their territory, as in Cayor and Futa Toro. Instead, in Sine and Salum, the concurrent development of Islam and peanut cultivation, carried out by African peasants on small landholdings, helped many of them to avoid the pillaging of ceddo rulers.

While differences among West African societies were hardly static, we explore whether some of these pre-colonial societal features and factor endowments might have influenced the response to incentives brought about by Europeans. The study of human capital trends in West Africa before 1900 has long remained an uncharted territory, and our estimates provide a way to assess some of the aforementioned hypotheses on the role played by European trade, slavery, and factor endowments on the long-term development of human capital in West Africa.

DATA SOURCES AND METHODOLOGY

The French colonial administration in Senegal was quite active in collecting information about the population of Gorée, and a number of census lists and electoral registers survived in the National Archives of Senegal (Dakar). While the censuses of Gorée and St. Louis are known to scholars and have been studied before (Becker et al. Reference Becker, Martin and Schmitz1983), to our knowledge the underlying micro data have not been used to study human capital development. In particular, we use individual-level records organized by birth cohort to estimate long-term trends in numeracy in three different periods: 1770–1799, 1800–1849, and 1850–1899.Footnote ³ Although the sources do not allow us to draw a well-defined line between the métis (Franco-African Creoles) and African groups, the history of Senegambia suggests that religion can be used as a proxy for differentiating between them in broad terms, or at least to differentiate between Christian Europeans (a tiny minority) and métis people on the one hand and Muslim West Africans on the other.Footnote ⁴

We drew data for analysis mainly from census lists, which cover the early nineteenth to early twentieth centuries. We observed a large portion of the population recorded. We took all the individuals available through the sources. Different “census years” sometimes refer to different stages in preparation of the same census, so that repetition is very unlikely; yet, we also avoided repetition by checking the names of the people included in the dataset.Footnote ⁵ Table 1 provides more details on the sources used and the number of observations available for each year. In addition, we collected evidence from electoral registers that started in 1851, which include information on the population of the Four Communes.

Table 1 SOURCES OF INDIVIDUAL DATA SORTED BY YEAR AND TYPE (RECORDED IN GORÉE AND DAKAR)

Source: National Archives of Senegal, Dakar: Electoral registers were found in the 3G262 and 3G265 folders, and were collected in 1851, 1887, and 1913; census lists were found in folders from 3G2120 to 3G2140 and in folders 22G29 and 22G34, and include enumerations collected between 1832 and 1921.

We use the terms “human capital” and “numeracy” as synonyms in the following discussion even though the concept of human capital may include other components. Although economic historians are well aware of the importance of human capital for economic growth and development (Allen Reference Allen2011; Baten and van Zanden Reference Baten and Luiten van Zanden2008; Temple and Voth Reference Temple and Voth1998), they must often rely on imperfect measures of educational quality because information on numerical skills and genuine ability is unavailable. The most common proxy for educational outcomes from a historical perspective is literacy. Yet, literacy rates do not always represent the quality of education, and several problems with this index have been highlighted by the literature (A'Hearn, Baten, and Crayen Reference A'Hearn, Baten and Crayen2009). Taking these matters into consideration, researchers have tried to develop new measures of human capital, such as the production of books, which proxies for advanced literacy (Baten and van Zanden Reference Baten and Luiten van Zanden2008).

A recent line of inquiry has developed an approach to measuring basic quantitative skills, reaching back to the medieval period and earlier. Brian A'Hearn, Joerg Baten, and Dorothee Crayen have found that the basic ability to count and complete simple calculations is negatively correlated with age heaping, that is, the tendency to report one's own age as rounded numbers such as 0 and 5. The extent of heaping for a given population can be summarized by the Whipple Index (Shryock and Siegel Reference Shryock and Siegel1976), which is the ratio of the observed frequency of ages ending in 0 or 5 to the frequency predicted by assuming a uniform distribution of terminal digits (one-fifth of total individuals), as described in Equation 1. The ABCC index, a linear transformation of the Whipple Index suggested by A'Hearn, Baten, and Crayen (Reference A'Hearn, Baten and Crayen2009), represents the share of the population able to report their own age exactly and ranges between 0 and 100, as described in Equation 2.

(1)

By exploring the extent of age heaping, numeracy can be inferred from a large variety of sources: census lists and household surveys, marriage registers, court records, and death registers are only some of the sources that can be used to reconstruct age heaping through historical samples.Footnote ⁷

We base our work on a two-step methodology. First, we estimate basic human capital for a number of macro-regions and half-century periods within Western Sudan by documenting age-heaping. Second, we explore the determinants of regional human capital formation in this panel by drawing on the literature discussed in the previous section.

The first step works as follows. Senegal is divided into seven macro-regions, based on characteristics of West African societies and kingdoms and the nature and intensity of interaction with Europeans. Figure 1 provides more detail. The region of “direct influence” is defined as the coastal area stretching from Rufisque on the petite côte up to St. Louis on the mouth of the Senegal River, including Dakar and the small island of Gorée. These Four Communes, as we have explained earlier, were characterized by long-term contact (1650–1900) between Europeans and West Africans due to commercial ties. There, European and local populations coexisted and often mixed. Next, the region of “colonial influence” is defined as the area that was inhabited mainly by Wolof people, including Walo, Cayor, Baol, Jolof, and Futa Toro. This area was characterized by major slave deportations as well as early colonial military campaigns, wars, and occupation. Third and fourth, the regions of Sine and Salum, south of Baol and north of the Gambia River on the west coast, were mostly inhabited by Serer groups and were only marginally touched by the military campaigns of the late nineteenth century. Fifth, Casamance, south of the Gambia and southeast Senegal, was annexed to Senegal later than any other regions and inhabited mostly by Jola and Mandingos (see Appendix Table 1 for an overview on how we recoded regions within Senegal). In addition to Senegal, we are able to obtain numeracy levels of people born in the Gambia and in the region of Kayes, in today's Mali, east of the upper Senegal River.

To calculate regional numeracy from micro-level data, we assign each individual in the sample to these areas according to their place of birth. Besides information on the latter, our database includes data on each individual's age, occupation, gender, captivity, and religion. We classify individuals according to their type of occupation following the taxonomy suggested by Alan Armstrong (Reference Armstrong1974), which is, in the case of our African evidence, virtually identical to what one would obtain by using HISCLASS (Leeuwen and Maas Reference Leeuwen, van and Maas2011).Footnote ⁸

We estimate regional numeracy trends by following Joerg Baten and Johan Fourie (Reference Baten and Fourie2015). Specifically, we estimate a regression in which the dependent variable equals 1 if the person has an age that does not end in 0 or 5, and the key independent variables are interaction terms that include religion (to reflect whether the background of the concerned individual was Creole or mostly African), the region of birth, and the half-century of birth to capture regional ABCC trends over the long run.Footnote ⁹ The interaction terms will capture the premium on numeracy (expressed relative to the constant of the model) for each background-region-period chosen; the control variables that we include in the regressions will help to limit the influence of potential sources of selectivity. In other words, this first-step-regression approach is preferable to simple averages for obtaining numeracy values. For example, if the sample referring to a specific region and period was characterized by a high share of slaves, while this was not the case for the underlying population, the value of numeracy obtained through simple averages for that particular region and time would be biased.

FIRST-STEP RESULTS: HUMAN CAPITAL TRENDS IN WESTERN SUDAN, 1770–1900

The estimates of regional numeracy trends are based on the regressions reported in Table 2.Footnote ¹⁰ To obtain regional ABCC figures, the religion-region-time coefficients are added to the value of the reference category (the constant refers to an unskilled, free Christian born between 1850 and 1899 in the region of The Four Communes, male, aged 33 to 72 and recorded via censuses). These regional estimates, in turn, will be our dependent variable when we explore the factors associated with the development of numeracy across areas of Western Sudan in the next section.

Table 2 ESTIMATING REGIONAL TRENDS IN NUMERACY

P-values in parentheses.

∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1

Note: The dependent variable “numerate” is one if an age not ending in 0 or 5 is reported. Marginal effects are reported, expressed in ABCC percentages. The constant refers to an unskilled, free Christian born between 1850 and 1899 in the region of direct influence (The Four Communes and their surroundings), male, aged 33 to 72 and recorded via census. Survey-year dummies and “gender unknown” dummies are included. The results when the religious group and gender could not be identified (NA) are not reported in the table. Only 8,669 observations from the original dataset remain when the age interval 23–72 is concerned.

Source: See text.

The regional trends reconstructed from the interaction terms (model in column 3) are reported in Figure 2 for Christians and Figure 3 for Muslims.Footnote ¹¹ The human capital of Christians in Gorée and the Four Communes grew remarkably over the period considered, especially after 1850. The same trend is discernible for Christians whose region of birth could not be determined and, to a lesser extent, the furthest region of Casamance.

Notes: The ABCC for the region of Casamance and East is based on a small sample of 43 observations during the period 1850–99. “Direct influence” are the four cities at the coast and their surroundings, in which contact with Europeans developed earliest. Sources: See text.

Figure 2 THE NUMERACY OF (MOSTLY CREOLE) CHRISTIANS BY REGION OF SENEGAL, C. 1750–1900

The regional trends for Muslims show a quite different, more heterogeneous pattern of human capital development, characterized by large and growing regional inequalities. On average, our figures are slightly more pessimistic than the 60 percent literacy suggested by Sylviane A. Diouf (Reference Diouf2013). Previous estimates of 60 percent literacy would have implied a somewhat higher numeracy level of approximately 80–90 percent (A'Hearn, Baten, and Crayen Reference A'Hearn, Baten and Crayen2009). However, we find that the numeracy of Muslims in large areas of Western Sudan only approached 60 percent by the mid-nineteenth century. The trend in the Four Communes reflects the human capital of unskilled Christians, although a gap of approximately 10 percentage points in numeracy compared to Muslims persists. The trend in the region of colonial influence, including Cayor, Baol, Jolof, and Futa Toro, shows that the human capital of Africans from Western Sudan declined to 52 percent during the second half of the nineteenth century (or stagnated, as this is probably not statistically significant). The result is similar to the trend for Muslims whose region of birth was unknown. The numeracy of Muslims in the Casamance region was still quite low (61 percent) during the second half of the nineteenth century. However, the human capital of Muslims in Sine-Salum was fairly high (c. 60 percent) during the early nineteenth century and grew considerably after 1850.

Table 3 compares our estimates to those obtained from other countries in sub-Saharan Africa, North Africa, and the Middle East. Mali (Kayes) ranks among the countries with the lowest numeracy, whereas the Four Communes rank among the highest, with the area of colonial influence and Casamance ranking somewhere in between. After discussing potential selectivity concerns in the next section, we outline and explore a number of hypotheses regarding the factors that might have affected this regional pattern of human capital accumulation.

Table 3 COMPARATIVE LEVELS OF NUMERACY IN REGIONS OF SENEGAL AND SUB-SAHARAN AND MIDDLE-EAST COUNTRIES

Source: Our own elaboration from Crayen and Baten (Reference Crayen and Baten2010).

Potential Sources of Selectivity

This sub-section discusses a number of factors that could have led to positive or negative selection of the recorded individuals into our historical samples with respect to their underlying population. This issue needs to be addressed when inferring human capital trends from historical samples, which are typically not entirely random.

First, regarding the electoral registers, because the French Colonial Government in Senegal aimed to provide voting rights to the indigenous and métis adult males in the second half of the nineteenth century, positive or negative bias from legal restrictions should not be an issue in principle.Footnote ¹² In practice, however, access to electoral registers was likely rather limited for the poorer and less educated strata of the population. Therefore, we would expect an upward bias in numeracy for individuals recorded via this source. We attempt to control for this in the regressions with a dummy variable for observations obtained from electoral registers (22 percent of our sample). The bias may or may not be constant, though, and so we perform a robustness test by simply excluding all individuals recorded via this type of source. We find that the results are very similar to the baseline specification that includes all observations (see column 4 of Table 2).

Notes: French Sudan (Mali), Salum, and Casamance and East are based on small samples, as noted in the legend. “Direct influence”: see notes to Figure 2. Sources: See text.

Figure 3 THE NUMERACY OF MUSLIMS BY REGION OF SENEGAL, C. 1750–1900

It is also possible that skilled individuals were more likely to migrate to the coastal region of the Four Communes and, as a result, more likely to be included in the Gorée census data. To assess this possibility, we compare the level of numeracy measured in coastal versus interior regions of Senegambia constructed from Gorée's sources with similar measures obtained from records of Senegambian slaves deported to Havana in the early nineteenth century. All slaves arriving to Havana, the trade capital of Latin America in much of the period, were recorded (Eltis Reference Eltis2007). If relative regional values are similar across these two different samples, one can argue that migrant selectivity was probably not strong because selection into slavery was, to a large extent, randomly distributed across Western Sudan. Indeed, considering evidence on the heights of slaves, Eltis has argued strongly that the bias between African slaves brought to the Americas and the populations from which they came was small or even negligible (Eltis Reference Eltis1982, pp. 455–56; see also Austin, Baten, and Van Leeuwen Reference Austin, Baten and Van Leeuwen2012; Baten and Blum Reference Baten and Blum2014).Footnote ¹³ Figure 4 shows that the numeracy of African slaves who were brought to Havana closely reflects the level of human capital of West African Muslims that we estimate from the Gorée census surveys. We repeat the same selectivity assessment comparing our samples with slaves deported to other world regions, and none of these comparisons suggests strong selectivity, as shown in Table 4.Footnote ¹⁴

Notes: Most of the slaves recorded in the Havana dataset were born during the period 1780–1810, while the Muslim Africans we observe were all born between 1850 and 1900; hence, there is a gap of (on average) 60 to 70 years between the samples. However, note that our estimates of regional trends in numeracy suggest that human capital stagnated in various areas of Western Sudan. Sources: The “Havana_Dataset” refers to a dataset on the shipment of African slaves to the port of Havana during the early nineteenth century obtained from the Transatlantic Slave Trade Database. Two main categories of individuals are considered: “coast” refers to slaves shipped from the port of Bissau (Portuguese Guinea) and “interior” represents slaves belonging to Mandingo tribes, who inhabited an area that stretched from contemporary southwestern Mali to southeastern Senegal. The “Goree_Dataset” refers to our dataset on Western Sudan. Here, “coast” refers to Muslim individuals born in the coastal cities called the Four Communes, while “interior” refers to a sample of Muslims born in French Sudan, today's Kayes, which is located in Mali.

Figure 4 AN ASSESSMENT OF POTENTIAL MIGRANT SELECTIVITY THROUGH ESTIMATES OF AFRICANS' NUMERACY (PERCENT) REPORTED IN TWO DIFFERENT SOURCES

Table 4 NUMERACY IN WESTERN MALI ACCORDING TO DIFFERENT SOURCES

Notes: We obtained a numeracy value from the Gorée census list of 35 and compared it in the paper with the numeracy of Mandingo people, who inhabited a region that largely overlapped with that of Kayes, in Western Mali. We obtain estimates of numeracy for the Mandingoes by using lists of slaves deported to Havana, and we find a very similar value to that of Gorée (31). Moreover, we can compare the latter to other slave lists in order to assess the robustness of the estimates. For example, Mandingo slaves deported to Brazil also had a numeracy of 31 and Mandingo slaves deported to the Southern region of the US had a numeracy of 32.5. Although the data from Gorée concern people born several decades later than slaves deported to the Americas, note that the estimated trends for numeracy in the interior regions of Africa highlight stagnation over time (see Figure 3 in the paper). The numeracy calculated from Gorée census lists refers to Muslim Africans born in Western Mali (French Sudan) between 1850 and 1900. The numeracy of slaves deported to Havana, US and Brazil refers to Mandingo people who inhabited the same region, born in 1750–1810 (Havana dataset) and 1770–1780 (Brazil data). The number of observations of each sample is shown is brackets.

Sources: Data from Gorée were collected from the National Archives of Senegal in Dakar. Data on slaves deported to Havana were collected through the Trans-Atlantic Slave Trade Database, see Eltis (Reference Eltis2007). The U.S. dataset is based on slave records from Louisiana, collected by Midlo-Hall (Midlo-Hall 2000). Data on slaves deported to Brazil are based on the work of Hawthorne (Reference Hawthorne2010). Both the United States and Brazil data are available online through the project “Slave biographies: Atlantic database network.” See http://slavebiographies.org/project/.

If there were migrant selectivity, we conjecture that it would be more likely upward than downward, given migration costs. The patterns discussed later focus on the low level and stagnation of numeracy in the interior regions. In the presence of positive migrant selectivity from the interior, the true differences would be even stronger than observed in Gorée census surveys. Thus, while we cannot dismiss migrant selectivity completely, the available evidence suggests that potential bias is not likely to be substantial enough to undermine the basic results.

Finally, we discuss another potential source of bias in the data for slaves (27 percent of our sample). In particular, we are concerned that some of the slaves' age statements might have been declared by their owners. This may positively bias estimates of numeracy because age was declared by Christian owners who might have had the opportunity to examine previously stated ages in other lists and calculate an unrounded age (e.g., if the slaves had been in their households for three years). We control for this in Table 2 by including interaction terms between the dummies that capture religion and those that identify slaves. Indeed, slavery is associated with a positive premium for Muslims, which suggests that slaves did not always declare their ages themselves and that their masters sometimes did so. However, a part of this premium might be due to relatively high numeracy of slaves in Gorée and Dakar. The results remain practically unchanged when we exclude slaves from the sample as an additional robustness check (column 5 of Table 2).

Last but not least, it is possible that Muslim Africans might have avoided cooperation with census takers, not least because of differences in language. However, we find that the numeracy of Muslims in Western Sudan was not extremely low, as we might have expected if all ages were estimated by census personnel due to non-cooperation. For example, it was higher than the numeracy of eighteenth century Muslims and non-Muslims in South Africa, for which there is evidence that each individual was asked for their age (Baten and Fourie Reference Baten and Fourie2015). Because of the very mixed character of Gorée's society, we believe that language barriers could be easily overcome. Moreover, there is some consistency if we estimate numeracy for the group whose information on religion was unknown to the authorities and compare it to the Muslim Africans—levels and trends were quite similar. This evidence suggests that while non-cooperative behavior might have existed, it is not likely to represent a major obstacle to estimating numeracy.Footnote ¹⁵

SECOND-STEP RESULTS: CORRELATES OF NUMERACY ACROSS REGIONS OF WESTERN SUDAN

This section presents the second step in our analysis: we exploit the variation of numeracy across macro-regions of Western Sudan and over time (values obtained through step 1) to explore the correlation between human capital and the factors outlined in the literature review. We focus on four main explanatory variables: (1) distance from Gorée, (2) the presence of cash crops and its interaction with societal features, (3) slavery, and (4) military activities—including additional controls.

First, as discussed earlier, Inikori (Reference Inikori2007) suggests that the rise of the Atlantic slave trade brought about a disintegration of Senegambia's regional market before the early nineteenth century. Coastal areas became more integrated with the Atlantic trade network, while interior suppliers ceased to specialize, reverted to self-subsistence, and abandoned previous trade networks while pillaging increased. We use the distance from Gorée to assess whether exposure to the Atlantic economy and the disintegration of Senegambia's regional market was associated with human capital development. If so, we expect distance to be negatively correlated with numeracy.

Second, we also explore whether the contact with Europeans was related to the regional growth of numeracy after the twilight of the Atlantic slave trade, when the demand for peanuts grew exponentially in the second half of the nineteenth century. As noted earlier, features of West African societies interacted with the rise of peanut as a cash crop. The more stable regions of Sine and Salum, experienced the most remarkable increase in numeracy during the second half of the nineteenth century. Peanut production was likely to benefit a large share of the Serer society for many distinct but related reasons. The Serer used land more intensively than their neighbors did, and exhibited denser patterns of settlement (Curtin Reference Curtin1975, p. 27), which in turn depended on horizontal ties and resistance to encroachment that allowed the development of more stable economic units among the Serer peasantry. This qualitative evidence is consistent with that discussed by Fall (Reference Fall2010), who argues that the rise of peanut production had a positive impact in regions characterized by high population density and domestic peanut production on relatively small farms.

We study how the interaction between the incentives from producing peanuts as a cash crop and societal features affected human capital formation. We code a dummy which is equal to one if the majority of a macro-region's territory was in the so-called peanut basin. Given that the impact of cash crops may have been heterogeneous across macro-regions of Western Sudan, in one specification we also interact this with a dummy equal to one for regions characterized by higher population densities, more equal and stable societies, where slavery was less prominent in the production of groundnuts (namely, Sine and Salum).

A third hypothesis—partly connected to Inikori's idea—is that the intensity of slavery within African societies might have hampered the development of human capital, even beyond slave deportations, since slavery was linked to slave raids and pillaging. We capture this mechanism by including a measure of the share of slaves in each regional sample, the coefficient of which we expect to be negative. Although it is hard to come up with a figure for the intensity of slavery in Senegambia during the nineteenth century, estimates provided by Moitt (Reference Moitt1989, pp. 29–30) suggest that between 25 and 40 percent of the population of Cayor was enslaved in 1904. Our share for 1800–1850 (52 percent) is broadly consistent with these figures, as it concerns data from the pre-abolition period.

Fourth and finally, French colonial military campaigns were likely related to the numeracy investment of West African parents. We code a dummy variable that takes value one if there was a military campaign during a certain half-century period and in a certain region of Senegal, Mali, or the Gambia, taking the information from the Historical Atlas of Africa (Ajayi and Crowder 1985). We expect its coefficient to be negative, too.

We use weighted least squares (weights being the number of individuals in each region as a share of the total observations in our sample) to regress regional ABCC on a number of covariates that could be related to the numeracy differences observed in our sample. Summary statistics are reported in Table 5 (data are available through the Open ICPSR repository, see Cappelli and Baten Reference Gabriele and Baten2017). Besides the aspects discussed thus far, we include further control variables. A dummy is used to identify areas that were (negatively) affected by jihad and marabout wars. This has been coded by drawing on A Political Chronology of Africa (Lea and Rowe Reference Lea and Rowe2001). According to this source, only one major conflict affected the region of colonial influence in 1750–1800, when the Fulani Kingdom rose to dominance in Futa Toro. However, several jihad wars fueled by marabouts affected the same region between 1850 and 1900 (idem, pp. 358–59).Footnote ¹⁶ Finally, two dummy variables are used to identify regional numeracy calculated from Muslim individuals and from individuals whose religion was unknown, as compared to regional figures based on Christian inhabitants. We expect especially the first one to be negatively correlated with numeracy.

Table 5 SUMMARY STATISTICS FOR WEIGHTED LEAST SQUARES REGRESSION

Sources: See text.

We estimate one specification with pooled data and one with time-fixed effects (i.e., half-centuries of birth as previously defined). In a third specification we include the interaction between areas where peanut production was prominent and regions characterized by better endowments and a limited use of slavery for its production. In a last model, we test whether the distance from Trans-Saharan trade networks (measured as the distance from Timbuktu) played any role in explaining the pattern of regional numeracy that we find.Footnote ¹⁷ Although the present analysis should be regarded as exploratory given the small number of observations, we can assess some of the interpretations offered in the previous section.Footnote ¹⁸ The results should also be taken as correlates rather than definitive evidence of causal relationships.

The econometric results are reported in Table 6. The distance from Gorée confirms the existence of a regional NW-SE gradient in the levels of human capital in Western Sudan. This is consistent with the argument that the incentives brought about by the rise of trade and other interactions with Europeans fostered development in numeracy in coastal regions. According to our fullest models, moving 100 km further away from Gorée is associated to a reduction of numeracy equal to about 5 percentage points.

Table 6 CORRELATES OF NUMERACY ACROSS REGIONS OF WESTERN SUDAN, 1770–1890

Robust standard errors in parentheses.

∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1

Notes: The distance from Gorée refers is calculated by referring to: (1) Dakar for the area of direct French rule, i.e., the Four Communes; (2) Banjul (formerly Bathurst) for British Gambia; (3) Ziguinchor for Casamance; (4) Fatick for Sine; (5) Kaolack for Salum; (6) Kayes for French Sudan (Mali); finally, given the extension of the area of colonial influence, we calculate an unweighted average of the distance between Gorée and Diourbel (Baol) and Gorée and Mboul (Cayor). We also control for the share of females in the sample. We exclude region fixed effects because they are likely to capture the entire source of variation in the data, given that our independent variables are mostly time-invariant and coded as dummies, and that we can extract only 25 observations for regional ABCC values. The share of young adults (23–32) is included to capture a potential remaining numeracy premium associated with youth.

^a In column 3, we include the dummy that identifies low-slavery areas, characterized by higher population densities and more stable and equal societies than inner regions, and its interaction with areas characterized by peanut production.

Sources: See text.

When we take into account the passage of time by using half-century dummies, the coefficient of slavery is scaled down, to the point that it becomes small and statistically indistinguishable from zero. This suggests that the correlation between numeracy and slavery is largely the result of a common time trend.

We find that the cultivation of groundnut did not itself guarantee faster human capital accumulation. Only in areas characterized by better endowments and thus high population densities, by more stable and equal societies, and by limited reliance on slavery (like Sine and Salum) did peanut production bring about higher levels of human capital. The coefficient of the interaction term in column 3, equal to 20.5 additional numeracy percentage points, is of substantial magnitude and statistically significant. The presence of a rather equal and cohesive society, characterized by small-scale production and a widespread economic culture (Austin Reference Austin and Baten2016) given the proximity to the Gambia River, interacted with the opportunities provided by peanut exports. This allowed the peasants to accumulate resources to fight back ceddo regimes and promote human capital accumulation in the area, which was also largely untouched by French early colonialism.

The other results are mostly as expected. Having been influenced by French military campaigns is always strongly and negatively correlated with numeracy (but only in one model the result appears to be statistically significant). Consistent with previous estimates, Muslim religion is associated with lower numeracy. The half-century dummies underline a positive trend in numeracy across regions of Western Sudan in the second specification. Finally, we estimate a positive (but not statistically significant) coefficient for the distance from Timbuktu (column 4). While this contrasts with Ghislaine Lydon's hypothesis on the importance of Islam and Trans-Saharan trade for the rise of a commercial culture and literacy in the region (Lydon Reference Lydon2009, p. 353), the religious wars initiated in Western Sudan by Islamic leaders might explain why being more distant from Timbuktu is associated to higher numeracy. This is consistent with previous findings by Valeria Prayon (Reference Prayon2013), who identified a negative cross-country relationship between Islam and numeracy.