Precolonial Centralization, Koranic Education and School Participation in Nigeria

2.1 Islam and Islamic Education in Nigeria¹

Islam first reached the Savannah region of western Africa around the 8th century A. D, some 1000 years before Christian missionaries first set their foot on the shores of Nigeria. The spread of Islam in this region has brought about extensive commercial relationships with North Africa (Fafunwa, 1991). With the established religious and trade relationship between the savannah region of West Africa and North Africa, new cultural experiences, which included the development of new intellectual and literacy skills that later made the region the hub of intellectual activities for many centuries, began to emerge.

Through the efforts of Islamic cleric and scholar Hamed Muhammed Mani, Islam was introduced to the ruler of Kanem Bornu between 1085 and 1097 (Fafunwa, 1991). From Kanem-Bornu, traders and scholars took Islam to Hausa Land in the early 14th century during the reign of Ali Yaji. From 1452 to 1463, books on Islamic theology and jurisprudence were introduced to Hausaland by Fulani scholars (Fafunwa, 1991). During the 1463-99 reign of Muhammed Rumfa in Kano, Islam became firmly established, and many Islamic learning centers were established. Islamic scholars from Timbuktu also taught and preached Islam in Kano. Islam later spread to Katsina in the 15th century, with Katsina receiving pilgrims from Mecca and Timbuktu scholars, with latter groups from Timbuktu often visiting books on Islamic divinity and etymology.

From Timbuktu, Islam was introduced to Yorubaland towards the close to the 19th century (Johnson, 1966), although before 1627, this area was already known to Muslims, when Ahmed Baba identified Yorubaland as a location where paganism was prevalent and Islam had no root (Fafunwa, 1991). Learned Islamic scholars formally brought Islam to Yorubaland through lorin in 1830. Some of these scholars made contact with Ibadan, about the time the town was established, and later with Ijebu-Ode and Abeokuta. Ilorin became a center of Islamic learning, with many basic and advanced schools established to teach Islamic religion. Ilorin often received famous Islamic scholars from Katsina, Kano, and the south-west. With the passage of time, Ibadan became a renowned center of excellence in Islamic education, with the construction of many mosques and Islamic schools to encourage the spread of Islam.

3.1 School Enrolment Data

Our dependent variables, the fraction of UPE and UBE cohorts reported to have enrolled in schools, were taken from the district level summary of Nigeria’s 2006 census, which contains relevant information. We restricted ourselves to cohorts who, by virtue of age, should have participated in the universal primary or basic education program. For the UPE cohorts, this included those reported to be in the 35-39 and 40-44 age brackets. Similarly, those in the 6-9 and 10-14 age bracket for the UBE cohorts were included in this study. The choice of these cohorts is primarily due to the fact that post-1970, the number of schools more than doubled (Archibong, 2019). The number of schools remained unchanged at about 35,000 until 1989, and again increased to about 69,979 (Archibong, 2019) by 2010, a period of 11 years after the re-introduction of UPE as UBE, with basic education becoming free and compulsory for the first time. The slash in the cost of education, coming from the direct reduction in tuition fees and access to schooling facilities, was required to increase educational attainment.

3.2 Explanatory Variables

Murdock’s (1967) Ethnographic Atlas is the source of state centralization and other relevant historical data. We matched Murdock’s ethnographic data using the information available in Archibong (2019), who identified which of the historical variables could be matched to the 774 districts or LGAs. The Atlas provides descriptions of the spatial distribution of ethnic groups across Africa, and for much of Africa, the data capture the pre-colonial attributes of these ethnicities, although they were sampled in the 20th century (Murdock, 1967; Michalopoulos, 2012; Obikili, 2016). Murdock constructed historical variables and the political complexity of each ethnic group using both primary and secondary sources, including other 59 historical variables besides the political complexity of state centralization. Murdock’s 243 ethnic locations out of 1265 fall within the African continent and 117 are currently within Nigeria. At the most, over three ethnic groups would overlap a given district today; they tend to share the same level of state centralization or political complexity value.

Following Gennaioli and Rainer (2007), Michalopoulos (2012), Alesina et al. (2013), and Archibong (2019), we adopted Murdock’s (1967) information on “Jurisdictional Hierarchy beyond the Local Community Level”. This measure ranges from zero to four, starting from the absence of hierarchical authority to four levels or layers of authority. This measure describes the complexity of political systems that allow pre-colonial states to deliver on the provision of social services and collect taxes.

For our benchmark model, we choose additional historical variables depicting various degrees of political and economic development. First, the Ethnographic Atlas describes the complexity of human settlement, namely nomadic or fully migratory, semi-nomadic, semi-sedentary, compact but impermanent settlement, neighborhood of dispersed family homesteads, separated hamlets forming a single community, compact and relatively permanent settlements, and complex settlements. The values range from 1 to 8 in order of complexity. Second, ethnographic variable 40 depicts the use of large animals by ethnic groups inhabiting the 774 districts in Nigeria. The seven categories are the absence or near absence of large domestic animals (1); pigs are the only large domestic animals (2); sheep and goats without large domestic animals (3); equine animals such as horses and donkeys (4); deer (5); camels, alpacas, or llamas (6); and bovine animals and cattle, Mithun, water buffalo, and yaks (7). The variable for the use of large animals was constructed by assigning 1 to locations with ethnic groups that fell within categories 2 to 7 and 0 otherwise.

Third, we used an indicator variable constructed from the Murdock Ethnographic Atlas v39, indicating whether groups inhabiting any district in our sample were using animal plow at the time of observation or whether plow existed before contact. Fourth, the historical variable indicates the intensity of the agricultural practices. This was obtained from v28 of the Murdock Ethnographic Atlas, with the following categories of intensity: no agriculture, casual agriculture, extensive or shifting agriculture, horticulture, intensive agriculture, and intensive irrigated agriculture. It ranges from 1 to 6, with higher values indicating a greater intensity.

The final historical variables in our baseline specification were the extent to which districts had groups that mainly hunt and herd large animals as a form of subsistence. The hunting variable taken from v2 and the herding variable are from v4 of the Murdock Ethnographic Atlas. The level of dependence for hunting and animal husbandry is graded as 0-5 %, 6-15%, 16-25%, 26-35%, 36-45%, 46-55%, 56-65%, 66-75%, 76-85% and 86-100%.

More Historical Variables

Because the present-day level of school enrolment could have been influenced by contact (or lack thereof ) with Europeans, we introduced several historical variables that reflect this experience and could serve as confounding factors if not accounted for.

We included a variable measuring the number of Christian mission stations (Protestant and Catholic) per square kilometer as of 1924 for each of the 774 districts. Alternatively, we introduced a variable measuring the number of mission schools in 1925. Data originally provided by Century Company (1911) and used by Nunn and Wantchekon (2011)² to construct a dummy variable assigning 1 to every district which had a European explorer passing through it, otherwise zero. Using the century map again, we constructed an indicator variable for every district through which the colonial rail passed.

District Geographic Controls

In addition to the variables listed in our benchmark specifications, we introduce several geographic controls. First, we used the district terrain ruggedness index obtained by Nunn and Puga (2012). This geographic variable measures the extent to which land can be worked upon, or the extent to which it facilitates or impedes access to certain localities, including schools and other public facilities.

The second was the extent of temperature experienced in a district. It was constructed by including temperatures at the time of birth for the UPE and UBE cohorts in our specification. The data were obtained from GeoAid (Goodman et al., 2019). This is because drought, prompted by high temperatures and water shortages, can impede several activities (Maccini and Yang, 2009), including participation in formal education. Third, we constructed district rainfall variables for UPE and UBE cohorts in the same way as we did for the temperature variable. Fourth, we included the presence or absence of gemstones (21 types) in each of the 774 districts. The presence of gold and crude oil was also included in the extended econometric specifications. Fifth, various dimensions of conflict were added to our specification because conflict can affect enrolment (for UBE because of the data used).

Sixth, the distance variables that can affect school enrolment are included in our augmented specifications. This includes the mean distance to the coast, water bodies, roads, country borders, major cities, lootable gold locations, gemstone deposit areas, drug cultivation, and onshore crude oil exploration sites³. The summary statistics for the dependent variables and the selected baseline variables are presented in Table 1.

5.1 Robustness Tests: Sample Selection and Influential Observations

Although the estimates of the three key explanatory variables are statistically significant at 1 % with baseline control variables, there are concerns that specification issues due to influential observations from these explanatory variables might be at play. Therefore, we ran regressions that accounted for the presence of extreme values. Table 3 presents the results. First, we eliminate districts with a centralization index greater than three (four are at the highest level). Second, we removed districts with a centralization index of less than one (zero is the lowest) and those with a centralization index greater than three. Third, we exclude districts with no investment in Koranic education. Fourth, we also removed districts with more than two-thirds of the 1914-1946 district cohorts with Koranic education. The results for both the UPE and UBE cohorts are reported in Table 3. Columns 1-8 show that most estimates are economically and statistically significant for the three key variables.

We used more standard methods to account for the influential observations. First, we used Student-T residuals that eliminated observations with plus or minus 2.0, 2.5, and 3.0. Second, we used Cook-Distance (based on d=4/771) to remove influential observations. Appendix Table 1 reports the results of the estimates. All estimates are economically significant and are statistically significant at 1%. Furthermore, we used the DFIT and DFBETA⁴ methods to check for outlying observations. Appendix Table 2 (columns 1-4) indicate that the presence of influential observations has no perceptible effects on our dependent variables. Based on RREG and QREG econometric methods⁵, We ran regressions that corrected for influential observations based on the RREG and QREG econometric methods. Again, Appendix Table 2 (columns 5-8) show that estimates for the three variables barely changed, remaining economically and statistically significant.

Finally, we used dummy variables for the three core variables. The Koranic education dummy was constructed by assigning 1 to districts with equal or higher mean values (lga_kor46 ≥ 0.3), otherwise zero. For the state centralization index dummy, we assigned 1 to a district higher than one level of political authority (v33 ≥ 2). The interaction variable was constructed using the product of the Koranic and State Centralization dummies. The results were all significant (results not shown, but available).

5.2 Additional Explanatory Variables⁶

Several covariates were included in our econometric specification to mitigate omitted variable bias. The included variables accounted for European influence in Appendix Table 3 (columns 1-4), conflict in Panels A and B columns 5-9 Appendix Table 3, rainfall and temperature (Appendix Table 4, Panels A and B), natural resources ( Appendix Table 4, Panels A and B column 3), Chinese foreign aid (Appendix Table 4.0, Panels A & B Column 4), distance to centers of economic activity (Appendix Table 4 Panels A and B column 5), d iseases ( (Appendix Table 4, Panels A and B column 6), with these additional explanatory variables, running regressions separately for male and female folks (Appendix Table 4, Panels A and B columns 7-8) and UBE cohorts in the 10-14 (Column 9) age category did not change results in any materially significant way.

5.3 Additional Robustness Checks

So far, the regression results from our baseline model and robustness checks have shown that state centralization has negative effects on school enrolment in districts where Koranic educational investment has been well above the average. The results still hold when many explanatory variables that could impact school enrolment are included in our econometric model. As we are not sure that our model is correctly specified, it is possible that one or two explanatory variables omitted from the econometric specification are strong enough to undermine the observed empirical relationships. In other words, our reported regression estimates may be biased by unobservables that are strongly correlated with our key explanatory variables, as well as school enrolment. In this concluding section (reported in Table 4), we attempt to perform a few more tests to mitigate the potential effects of unobservables on the observed results.

First, we want to ensure that the permanent historical and geographical differences between northern and southern Nigeria are not confounding the reported results. To facilitate comparison, we show the beta coefficients in brackets while showing the normal standard error in parentheses. Therefore, we introduced an indicator variable to the north. Columns 1 and 2 report our regression results. All three key explanatory variables are statistically and economically significant.

Second, we ran regressions for the northern sample to ensure that the vast differences in the northern and southern regions were not driving results. In this case, we compare areas in the north with differential exposure to both historical state complexity and Koranic education investment. To a large extent, this should mitigate the role that unobservables might play in influencing the results. Columns (3) and (4) show the regression estimates. Again, all the key variables remain significant. This shows that the difference between north and south cannot account for our reported results. Restricting regressions to a pair of LGAs within 200 km of each other yielded similar results to the previous regressions⁷ (Columns 5 and 6). The empirical estimates reported in columns 7-10 showed where that Koranic education has the greatest effect on education outcomes. This is done by splitting LGAs into two groups, one group having to 3-4 layers of governmental authority beyond the local level (high state centralization LGAs) and the other having to 1-2 layers of authority (low-state centralization LGAs). For the two cohorts (6-9 and 34-39), estimates of the effects of Koranic education on school participation are negative and statistically significant. However, the effects in high-state centralization LGAs are nearly four times the effect on low-state centralization LGAs (Columns 7-10).

Finally, we implemented the Instrumental Variable (IV) method, which exploits the fact that Koranic education was ongoing in areas where Christian missionaries were restricted from reaching by the colonial government, and not necessarily because these areas were unreachable (Okoye and Pongou (2024). Therefore, we used the Emirate dummy variable as an instrument for Koranic education, assigning 1 to LGAs considered restricted areas for missionaries⁸. Table 5 presents the regression results. IV regressions are reported for the 6-9 and 34-39 cohorts for all samples, as well as high- and low-state centralization LGAs (Column 1-6). Koranic education had negative and significant effects on school enrolment for both cohorts and the whole sample, but significantly more so for high-state centralization areas. IV regressions for the northern sample implemented for high- and low-state centralization LGAs produce similar results. Koranic education tends to have greater negative effects on school enrolment in high-state centralization LGAs than in low-state centralization LGAs (Columns 7-10). These results should only be regarded as suggestive of causal effects because we do not have definitive evidence to make any serious claims about causal inference.