Abstract

By focusing on the unique setting of one Protestant mission (Livingstonia Mission) dating back to the late 19th century in Malawi, this study investigates the long-lasting missionary influence on female marital practices, based on individual-level data provided by the Malawi Demographic and Health Survey (2000, 2004 and 2010). Exploiting geographical distance to the influential mission station as a measure of exposure to the missionary influence, together with an abundance of historical, geographic and climate controls, this study finds that the mission encouraged females to postpone their first marriage, while discouraging their engagement in polygynous relationships. In addition, due to the missionary influence, females were also motivated to convert to Christianity as well as to attain academic skills. These findings suggest that Christian attitudes and values, along with the missionary educational investment, may play a role in explaining the missionary influence on marital practices.

Introduction

Christian missionaries should be given considerable attention by economists because their activities may have resulted in long-lasting effects on economic development by altering cultural values and/or the structure of institutions (Alesina and Giuliano, 2015). For example, Nunn (2010) demonstrated that European missionary activities have ingrained religious values among African indigenous people and, according to the famous Weberian hypothesis, the Protestant work ethic may act as a facilitator of a capitalist economy (e.g., Becker and Woessmann, 2009; Arruñada, 2010; Cantoni, 2015).

More recently, a growing body of empirical research has also explored the investment by missionaries in local public goods and the resulting welfare consequences on education, social capital and urbanisation. Examples of missionary investment include development of modern educational systems (e.g., Gallego and Woodberry, 2010; McCleary, 2013; Nunn, 2014; Valencia-Caicedo, 2014; Wantchekon et al., 2016); foundation of schools and hospitals and the subsequent diffusion of ‘useful’ knowledge (Bai and Kung, 2015); promotion of democracy (e.g., Woodberry and Shah, 2004; Woodberry, 2004, 2012) and introduction of the printing press (Cagè and Rueda, 2016).

In contrast, there is a marked paucity of empirical studies exploring the missionary influence on marriage practices, with only a few exceptions (e.g., Fenske, 2015). This research gap exists even though examination of the family formation process is a highly important factor to consider when examining economic development. While both rigorous empirical and theoretical research remain scarce, several notable studies have recently demonstrated the considerable welfare impacts of the practices of early marriage and polygyny (e.g., Tertilt, 2005, 2006; Schoellman and Tertilt, 2006; Field and Ambrus, 2008; Bove and Valeggia, 2009; Edlund and Lagerlöf, 2012).1

It can be hypothesised that marriage practices were influenced by missionary activities. First, religious teachings may instill marriage-related cultural norms in the indigenous population, such as the avoidance of polygyny. Second, economic development resulting from the missionary investment in public goods may affect local marriage practices by altering the perceived value of women in the marriage and labour markets. Based on this premise, the current study explores the case of a single missionary venture that took place in Malawi in the late 19th century and continued into the 20th century (Livingstonia Mission). The aim is to evaluate the long-term impacts on female marriage choices (here, early marriage and polygyny) using repeated cross-sectional data drawn from the Malawi Demographic and Health Surveys (MDHS) conducted in 2000, 2004 and 2010.

The Livingstonia Mission of the Free Church of Scotland was founded in 1875. It has long been recognised as one of the most important missions aimed at introducing Christianity into Malawi and throughout Central Africa. Although the picture should not be over-simplified, historical research conducted in Malawi suggests that Christianity expanded from the northern areas, where the mission's influential station, Livingstonia, was established. Therefore, in the present study, a community's distance to Livingstonia serves as a measure of the missionary influence. On the other hand, it is also true that Christianity was less appealing to the Yao, an ethnic group that was largely proselytised into Islam, as their ivory and slave trade with the Arabs existed before the arrival of the Christian mission. Consequently, this study primarily analyses the missionary influence on non-Yao ethnic groups. Nonetheless, data about the Yao people are also exploited to facilitate the interpretation of the mechanisms responsible for the identified marriage effects on the non-Yao population.

As will be shown in this study, a one-standard-deviation increase in the distance to Livingstonia from the mean value results in a 24–27% increase in the probability of non-Yao females under 18 years of age entering into marriage. This one-standard-deviation increase in distance also results in a 7–8% increase in the probability of non-Yao females of all ages engaging in polygynous relationships. In addition to these main findings, the current study also demonstrates that non-Yao females residing further from Livingstonia are less likely to attain academic skills (as measured by formal schooling and reading aptitude) and less likely to convert to Christianity than their non-Yao counterparts living in closer proximity to Livingstonia. These findings imply that Christian values and missionaries’ educational investment are potential factors explaining the marriage impacts.

To elucidate which of these two forces exerted greater influence on local marriage practices, this study also examines the influences of a community's distance to Livingstonia on the adoption of Christianity, academic skills and marriage practices of ‘Yao’ females, a group that is largely Muslim. These results are subsequently compared with those for non-Yao females, who are largely Christian. This comparison may allow identification of the channels for impacting marriage practices of the non-Yao population. This assertion can be made because it seems that educational investment affects all local people to a similar degree, whereas the religious teachings mainly influence the behaviour of adherents. Based on this strategy, it is argued that the marriage impacts on non-Yao females may have been driven by the attitudes and values promoted by Christianity as well as the promotion of education by the Livingstonia mission.

The Livingstonia mission was established ‘before’ European institutions and other early missions had the opportunity to exert significant commercial, political and religious influence on the local population. Therefore, by focusing on this pioneering mission, it is arguably possible to separate the effects attributed to missionary activities from those ascribed to these other institutional forces, while encouraging the causal interpretation of an intent-to-treat (ITT) effect of this frontier mission. The in-depth within-country nature of the analysis also helps identify impacts resulting from missionary activities, which are separate from the influence of other environmental/institutional factors. This identification is usually difficult in macro-level studies. In particular, the wealth of information on this important mission provided by the relevant historical studies may also facilitate persuasive discussion of several empirical issues.

Despite this promising setting for causal identification, it must be acknowledged that the distance to Livingstonia may also be correlated with other socio-economic factors affecting marriage decisions. In this work, three primary approaches are taken to address this concern and facilitate the causal inference. First, an attempt is made to control for pre-determined local conditions that might have affected both the advancement of missionary penetration and colonial operations. Hence, a great number of geographic and climate conditions affecting the surveyed communities (e.g., climatology, landscape typology, soil and terrain, and crop season parameters) are used as controls in the estimations. In addition, the regressors also include historical information on the travel routes of European explorers, railway lines that were in operation during the first decade of the 20th century, and the extent of slave export in the 19th century. As these pieces of information cannot be discerned from the MDHS, the relevant data were derived from other data sets of the Third Integrated Household Survey (IHS) 2010–2011 and Nunn and Wantchekon (2011). The second approach is to conduct several falsification tests with the aim of verifying that the estimated missionary effects are not attributable to other environmental factors. Third, following Oster (forthcoming)'s method, this study also assesses the importance of unobservables (relative to the observed controls) required to explain the identified missionary effects.

The results yielded by the present investigation provide two noteworthy contributions to the extant economic literature. First, the economic impacts of Christian missions are explored by focusing on their influence on local marriage practices. In this sense, the findings support Fenske (2015), who linked the missionary influence to the lower prevalence of polygynous marriages in several African countries. However, differences between these studies still exist. Specifically, the current study explores the missionary influence on both early marriage and polygyny via a detailed analysis of one particular mission. In addition, the previously mentioned mission-related empirical literature to date has largely highlighted missionaries’ investment in local public goods and the subsequent impacts on economic development. In addition to this economic perspective, the role of religious values is also discussed in the present study.

Second, this study improves empirical understanding of the factors determining polygyny and early marriage. Thus far, a handful of empirical studies have explored the determinants of polygyny from several perspectives. These perspectives include women's marginal contribution to agricultural production (Jacoby, 1995); the slave trade (Dalton and Leung, 2014); colonial and missionary education (Fenske, 2015) and the desire to acquire many (possibly male) offspring (Grossbard-Shechtman, 1986; Milazzo, 2014).2 Empirical research aimed at understanding the factors responsible for the practice of early marriage (possibly resulting in/from early pregnancy and fertility) is also scarce, with the exception of several studies focusing on the role of HIV/AIDS (Ueyama and Yamauchi, 2009); female labour market opportunities (Jensen, 2012) and education subsidies (Duflo et al., 2015).

The findings yielded by this research may also attract the interest of policymakers and practitioners who are concerned with the effects of religious institutions on the development process. For example, Commission for Africa (2005, pp. 31, 127–129) highlights the potential of such institutions, which provide both material resources and spiritual services, to exert positive influence on the projected development of the African continent.3

This paper is organised into seven sections. Section 2 provides historical background on the Livingstonia Mission. A data overview is given in Section 3, followed by the empirical strategy presented in Section 4. The main findings are reported in Section 5. Section 6 discusses the channels through which the missionaries exerted influence on marital practices, while concluding remarks are provided in Section 7.

Livingstonia Mission

This section provides historical background on the settlement pattern of the Livingstonia Mission, which facilitates the identification strategy explained in Section 4. The historical information is largely sourced from McCracken (1977), Msiska (1995), Pike (1968), Shillington (2005) and Vail and White (1991). Detailed historic accounts can also be found in those literature sources and elsewhere (e.g., Pike, 1965; Kalinga, 1985; Thompson, 1995; Bone, 2000; McCracken, 2012).

As described in more detail in Appendix A.1, David Livingstone (1813–73), one of the most renowned explorers to make a transcontinental journey across Africa during the mid-19th century, laid the groundwork for the mission, which was named ‘Livingstonia’ in his honour. Sponsored by the British government, the Scottish missionary headed the ‘Zambezi Expedition’ between 1858 and 1863, which aimed to catalogue the natural resources of the Zambezi River area, as well as to identify the trade routes necessary for transporting raw materials from the African interior to coastal trading points so that they could eventually be sold in the British market.

In this expedition, Livingstone concluded that a deep-water route from the Shire River to Lake Malawi by steamer would be the only practical means of linking the interior with the coast. Moreover, he deemed the Shire Highlands, a plateau in southern Malawi, a suitable area for white settlement as well as for the creation of a cash-crop economy. However, his statement quickly encountered harsh criticism. Consequently, the British government decided to withdraw the Zambezi expedition, which had lasted 6 years, and many observers at the time commented that the expedition was a failure with none of its purposes fulfilled.

After a decade in which the Free Church of Scotland and other societies ignored Livingstone's proposals, James Stewart (1831–1905), a head of the Lovedale Institution in South Africa, drafted and sent a memorandum on ‘Livingstonia, Central Africa’ to Scotland in 1874.4 He presented the key details of Livingstone's suggestions to the General Assembly of the Free Church. He suggested that the Shire at the southern end of Lake Malawi could be reached from the coast by waterway via the mouth of the Zambezi, with Lovedale operating as the operational base for a new mission. In response to his presentation, the Free Church authorities decided to found the Livingstonia Mission.5

In 1875, the Livingstonia Mission established its central station at Cape Maclear (see Figure 1), a hilly promontory at the south end of Lake Malawi that served as a suitable port for the mission steamer. Based on a residential mission policy that required Africans to be housed and trained in mission sites isolated from the ‘temptations’ of their own society, the mission attracted a variety of people to the site (e.g., freed slaves returning to their homeland, a local chief's son sent to acquire Western education and refugees defiant of the authority of local rulers). However, the settlement expansion encountered several issues, such as a shortage of sanitation facilities, insufficient food for the settlers and the failure to regulate settlers’ behaviours in many spheres of social life (e.g., violence, theft, Sunday meetings, beer drinking and polygamy). Moreover, it soon became evident that the mission station was nearly useless because, along with its great distance from the nearest villages, it was situated on the edge of barren and tsetse-infested plains unfriendly to animal life and lacking the fertile land needed for cotton production. Facing these unfavourable environmental conditions, the pioneering party decided to move the central station to Bandawe (see Figure 1), halfway up the west coast of Lake Malawi, in 1881. Consequently, the years spent by the missionaries at Cape Maclear were seen as a period of adjustment to the realities of the African situation. The missionaries’ departure also immediately weakened their influence at Cape Maclear.

Figure 1:

Early Mission Stations.

Notes: (1) The location of the early mission stations was sourced from Nunn (2010). (2) The map of Malawi is sourced from DIVA-GIS (http://www.diva-gis.org/datadown). (3) In this figure, Bandawe, Blantyre, Cape Maclear and Livingstonia refer to ‘Kasangazi’, ‘Blantyre’, ‘Mkope Hill’ and ‘Hondowe’ named in Nunn's (2010) data.

Figure 1:

Early Mission Stations.

Notes: (1) The location of the early mission stations was sourced from Nunn (2010). (2) The map of Malawi is sourced from DIVA-GIS (http://www.diva-gis.org/datadown). (3) In this figure, Bandawe, Blantyre, Cape Maclear and Livingstonia refer to ‘Kasangazi’, ‘Blantyre’, ‘Mkope Hill’ and ‘Hondowe’ named in Nunn's (2010) data.

The relocation to Bandawe was a milestone in the history of the Livingstonia Mission for several reasons. First, it showed the mission's intent to shift the entire axis of its activities to Northern Region.6 Second, the mission abandoned its previous residential policy and decided to work directly with existing villages, retaining only a small residential component at the mission. The exploratory and evangelistic visits to the neighbouring villages made under the new strategy helped the missionaries extend their Christian and educational influence outside the settlement considerably. Third, in the early stages of the Livingstonia Mission, in the absence of any local authority, the missionaries often exercised civil powers to impose discipline on the settlers as well as to respond to crimes such as thefts occurring near the settlement. Similarly, at the original location, the mission was involved in local disputes, acting as an authoritative third party at the request of indigenous headmen. At Bandawe, however, the missionaries attempted to avoid any involvement in local politics.

Despite the landmark nature of the move to Bandawe, however, both the missionaries and the Foreign Missions Committee of the Free Church at home regarded the location as a provisional outstation until a better site was found. Several issues contributed to their lack of enthusiasm. First, the low-lying site on the lake shore seemed malaria-prone due to its close proximity to swamps and marshes. In addition, its susceptibility to attacks and lack of protection from waves made the site inadequate for a mission steamship harbour. While the missionary activities at Bandawe experienced unparalleled success for a mission in East or Central Africa in this period, the mission eventually decided in 1894 to relocate the central station to Khondowe, further north, which later developed into the small town presently known as Livingstonia (see Figure 1). The new site was located on the highlands between Lake Malawi and Nyika Plateau and was not prone to malarial mosquitoes.

Under the directorship of Scottish missionary Robert Laws (1851–1934) at Livingstonia, the mission contributed greatly to providing educational facilities and services at both the primary and post-primary levels.7 The educational expansion was followed by the widespread adoption of evangelical Christianity. Significant improvement in evangelical strategies (e.g., using mission-educated natives as evangelists) enabled Christianity to spread as a genuinely popular movement from the mid-1890 s. From the long-term venture started 20 years earlier at Cape Maclear, the mission finally established a solid base for its activities at Livingstonia. Thus, a great movement towards Christianity can be said to have begun in northern Malawi.

Data

The information utilised in the present study is primarily sourced from the repeated cross-sectional data yielded by the MDHS (2000, 2004 and 2010) conducted by the National Statistical Office (NSO) from July to November 2000, from October 2004 to January 2005 and from June to November 2010. The MDHS is a nationally representative household survey providing information on population, health and nutrition in Malawi, including data on marriage, fertility, family planning, reproductive health, child health and HIV/AIDS status. Given these specific areas of interest, women of reproductive age are the primary target of this survey. The 2010 MDHS completed interviews with 23,020 females aged 15–49 residing in 24,825 households located in 849 enumeration areas (communities).8 For the previous survey in 2000 (resp., 2004), 11,698 (13,220) resident females in 13,664 (14,213) households situated in 521 (559) communities were interviewed.9 The MDHS households were chosen by stratified random sampling with stratification based on study domains and urban/rural considerations.10 Although the MDHS has been conducted multiple times, the data yielded did not include a panel element in terms of either the clusters or households. As explained in Section 4.2 in more detail, the analyses conducted in this study also exploit community-level geographic and climate variables sourced from the IHS 2010–11 and ethnicity-level historical controls obtained from Nunn and Wantchekon (2011) in order to complement the limited information available from the MDHS.

The Malawi population consists of several ethnic groups. As described in Appendix A.2, the Yao ethnic group, uniquely among these groups, largely converted to Islam in the late 19th century as a result of their ancestors’ strong alliances with the Arabs, which predated the arrival of the Livingstonia Mission. Consequently, they were less amenable to missionary activities than other ethnic groups were. Therefore, the empirical analysis in this study primarily focuses on data about non-Yao respondents.

The distribution of religious affiliation among all female respondents is provided in Table 1. As can be seen, approximately 85.9% of the females in the sample practiced some form of Christianity, whereas about 13.0% practiced Islam and 1.0% indicated other/no religion.11 Apart from the category designated as ‘Other Christian’, Presbyterianism (the Church of Central Africa, Presbyterian; CCAP) and Catholicism are the two major types of Christianity in Malawi, with CCAP being directly descended from the Livingstonia Mission.12 In addition, as indicated above, the Yao are predominantly Muslim. In contrast, only about 4% of the non-Yao population is Muslim (the absolute number of individuals practicing Islam included in this non-Yao sample is nonetheless large). The bivariate correlation between the indicator for practicing Christianity and the non-Yao dummy is high, with a coefficient of 0.68.

Table 1:

Religious Affiliation

 Total Yao Non-Yao 
(1) Christianity 
 Anglican 0.02 0.01 0.02 
 Catholic 0.21 0.05 0.23 
 Seventh Day Advent/Baptist 0.06 0.01 0.07 
 The CCAP 0.17 0.04 0.18 
 Other Christian 0.37 0.09 0.42 
(2) Islam 0.13 0.76 0.03 
(3) Other or no religion 0.01 0.00 0.01 
No. of respondents 47,920 6,171 41,749 
 Total Yao Non-Yao 
(1) Christianity 
 Anglican 0.02 0.01 0.02 
 Catholic 0.21 0.05 0.23 
 Seventh Day Advent/Baptist 0.06 0.01 0.07 
 The CCAP 0.17 0.04 0.18 
 Other Christian 0.37 0.09 0.42 
(2) Islam 0.13 0.76 0.03 
(3) Other or no religion 0.01 0.00 0.01 
No. of respondents 47,920 6,171 41,749 

Notes: The number is the proportion relative to the total number of respondents in each category. This is the unweighted proportion. In order to calculate the true proportion of the entire population from the sample data, appropriate sample weights need to be used.

The summary statistics related to non-Yao respondents only are reported in Table 2. These summary statistics are reported for selected variables, along with a check of the equality of the mean between two groups classified according to their residential community's distance to Livingstonia (latitude, 10°36′S; longitude, 34°06′E), which is expected to reflect the missionary influence. The sample was divided according to the median distance from the mission station. Using the GPS-based coordinates provided by the MDHS, the distance was calculated as the great-circle distance (GCD) between the MDHS communities and Livingstonia.

Table 2:

Descriptive Statistics (non-Yao)

 Distance to Livingstonia 
Below median Above median 
Mean Std. No. of obs. Mean Std. No. of obs. 
(A) Individual controls 
 Age at first marriage (years)1 17.59*** 3.02 16,678 17.29 3.31 16,925 
 Polygyny (dummy)2 0.19*** 0.39 14,563 0.10 0.31 13,932 
 Christian (dummy) 0.95** 0.20 20,621 0.94 0.21 20,616 
 Education (years) 5.39*** 3.56 20,627 4.63 3.72 20,621 
 Unable to read at all (dummy) 0.32*** 0.46 20,573 0.36 0.48 20,579 
 Age (years) 27.82* 9.30 20,628 27.99 9.29 20,621 
 Birth order 3.67 2.37 20,568 3.67 2.44 20,562 
 No. of alive siblings at age 10 4.07*** 1.99 20,568 4.02 2.16 20,562 
 No. of late siblings at age 10 0.83** 1.41 20,568 0.79 1.35 20,562 
(B) Selected community-level controls 
 Distance to Livingstonia (100 km) 2.63*** 1.31 20,628 5.95 0.70 20,621 
 Distance to the nearest mission station (100 km) 0.26*** 0.14 20,628 0.18 0.12 20,621 
 Descent rule       
  Patrilineal descent (dummy) 0.49*** 0.49 20,628 0.19 0.39 20,621 
  Matrilineal descent (dummy) 0.38*** 0.48 20,628 0.75 0.42 20,621 
  Dual descent (dummy) 0.12*** 0.32 20,628 0.04 0.20 20,621 
(C) Selected geographic and climate
controls 
 Longitude 33.80*** 0.39 20,628 35.08 0.39 20,621 
 Latitude −12.54*** 1.43 20,628 −15.60 0.61 20,621 
 Annual mean temperature (×10°C), 1960–1990 213.43*** 20.24 20,628 222.71 18.32 20,621 
 Std. of temperature (×100), 1960–1990 2,238.15*** 257.35 20,628 2,455.73 201.08 20,621 
 Mean precipitation (mm), 1960–1990 1,039.01*** 246.32 20,628 1,177.12 276.82 20,621 
 Coef. of var. of precipitation, 1960–1990 109.92*** 11.07 20,628 95.94 11.88 20,621 
 Elevation (m) 992.71*** 329.16 20,628 718.16 319.70 20,621 
 Slope (percent) 4.36*** 3.78 20,628 5.71 4.80 20,621 
(D) Ethnicity-level historical controls 
 European explorers (dummy) 0.97*** 0.15 19,335 0.57 0.49 19,416 
 Railway networks (dummy) 0.06*** 0.23 19,335 0.07 0.26 19,416 
 Slave exports normalised by land area 0.27*** 0.41 19,335 0.66 1.36 19,416 
 Distance to Livingstonia 
Below median Above median 
Mean Std. No. of obs. Mean Std. No. of obs. 
(A) Individual controls 
 Age at first marriage (years)1 17.59*** 3.02 16,678 17.29 3.31 16,925 
 Polygyny (dummy)2 0.19*** 0.39 14,563 0.10 0.31 13,932 
 Christian (dummy) 0.95** 0.20 20,621 0.94 0.21 20,616 
 Education (years) 5.39*** 3.56 20,627 4.63 3.72 20,621 
 Unable to read at all (dummy) 0.32*** 0.46 20,573 0.36 0.48 20,579 
 Age (years) 27.82* 9.30 20,628 27.99 9.29 20,621 
 Birth order 3.67 2.37 20,568 3.67 2.44 20,562 
 No. of alive siblings at age 10 4.07*** 1.99 20,568 4.02 2.16 20,562 
 No. of late siblings at age 10 0.83** 1.41 20,568 0.79 1.35 20,562 
(B) Selected community-level controls 
 Distance to Livingstonia (100 km) 2.63*** 1.31 20,628 5.95 0.70 20,621 
 Distance to the nearest mission station (100 km) 0.26*** 0.14 20,628 0.18 0.12 20,621 
 Descent rule       
  Patrilineal descent (dummy) 0.49*** 0.49 20,628 0.19 0.39 20,621 
  Matrilineal descent (dummy) 0.38*** 0.48 20,628 0.75 0.42 20,621 
  Dual descent (dummy) 0.12*** 0.32 20,628 0.04 0.20 20,621 
(C) Selected geographic and climate
controls 
 Longitude 33.80*** 0.39 20,628 35.08 0.39 20,621 
 Latitude −12.54*** 1.43 20,628 −15.60 0.61 20,621 
 Annual mean temperature (×10°C), 1960–1990 213.43*** 20.24 20,628 222.71 18.32 20,621 
 Std. of temperature (×100), 1960–1990 2,238.15*** 257.35 20,628 2,455.73 201.08 20,621 
 Mean precipitation (mm), 1960–1990 1,039.01*** 246.32 20,628 1,177.12 276.82 20,621 
 Coef. of var. of precipitation, 1960–1990 109.92*** 11.07 20,628 95.94 11.88 20,621 
 Elevation (m) 992.71*** 329.16 20,628 718.16 319.70 20,621 
 Slope (percent) 4.36*** 3.78 20,628 5.71 4.80 20,621 
(D) Ethnicity-level historical controls 
 European explorers (dummy) 0.97*** 0.15 19,335 0.57 0.49 19,416 
 Railway networks (dummy) 0.06*** 0.23 19,335 0.07 0.26 19,416 
 Slave exports normalised by land area 0.27*** 0.41 19,335 0.66 1.36 19,416 

Notes: (1) The equality of means between those residing in close to Livingstonia and those away from Livingstonia are examined by T-tests. ***Significance at 1%, **significance at 5% and *significance at 10%. (2) The information is relevant only to:

1Females who have ever been married by the time of the DHS.

2Females who were in marital relationships at the time of DHS.

These two groups were found to be significantly different in terms of many characteristics at the individual (panel A), community (panels B and C) and ethnic levels (panel D). For example, the results presented in panel A show that individuals residing in the vicinity of Livingstonia were more likely to be Christian and postpone their first marriage, while somewhat unexpectedly engaging more frequently in polygyny, relative to those living further from the mission station. In addition, compared with persons residing far from Livingstonia, the residents in communities closer to the station attained higher levels of education.

Along with these observations, several further findings are potentially compatible with historical records on the advancement of the mission. For instance, in its pioneering years, the Livingstonia Mission moved and established a settlement in Northern Region partly to avoid the unfavourable climate conditions in the south (e.g., at Cape Maclear). This resulted in the subsequent establishment of the mission's main work centres in the northern highlands, as explained in Section 2. Consistent with this account, the area surrounding Livingstonia is characterised by high altitude and lower temperatures and precipitation.

Empirical strategy

It is expected that the Livingstonia Mission affects present marriage practices primarily through two mechanisms. First, missionary activities potentially altered people's values and beliefs concerning marital practices, which then may have been transmitted from parents to children.13 It is certainly possible that this change in values and beliefs discouraged descendants of those in contact with the mission from entering into polygyny and early marriage. Indeed, abandoning polygyny and the consumption of beer were seen as strict prerequisites for becoming a Christian in the early periods of missionary activity (McCracken, 1977, pp. 195, 253).

Second, by investing in the creation of legal, institutional and economic foundations for a modern market economy (e.g., by provision of education), the mission may have altered women's values in both the marriage and labour markets during the last century, which now affects present female marital practices. For example, as the economy advances, greater value is placed on children's education in modern labour markets. As a result, males may prefer to marry educated females, as they are perceived as being more capable of rearing educated children. Under this mechanism, female education (resp., early marriage) may be encouraged (discouraged) because education increases the perceived value of women in marriage markets. As shown in Gould et al. (2008), this change in male preferences reduces the incidence of polygyny while encouraging skill-based assortative monogamous mating.

In addition, in primitive societies in which search frictions in the process of finding a spouse are prevalent, female marriage can be interpreted as a decision by the bridal parents (or the bride) to accept an offer of marriage. Under this framework, parents accept the first proposal that provides them with utility higher than their reservation payoff (Ermisch, 2003). Thus, if female academic skills are sought after and valued in the labour market, it is possible that parents would decide to invest in their daughter's education as well as to delay her marriage. This decision would be based on the greater gains derived from increased female job opportunity, as an educated daughter can supplement family income before entering into a marital relationship (while also increasing the reservation utility).14

A primary goal of this study is to identify the ‘total’ effects of these two forces. However, the underlying factors are also discussed in Section 6.

Specification

One of the most influential stations of the Livingstonia Mission was Livingstonia, a town located in the northern area of Malawi. Thus, this study measures the influence of the mission by using a community's distance to Livingstonia. As described in Section 3, it is expected that missionary activities had a limited influence on the Yao due to their strong socio-economic ties with Arab Muslims, which existed before the advent of the Christian mission. Therefore, the analyses conducted in this study primarily focus on data from non-Yao respondents. Consequently, for a non-Yao female i living in an MDHS community j, the effects of the Livingstonia Mission are estimated based on the following empirical model:  

(1)
yij=α1+α2dj+α3xij+εij,
where yij denotes marriage-related outcomes; dj is the GCD between community j and Livingstonia; vector xij contains several determinants of the outcomes specific to this female, her original household and her community, in addition to district fixed effects (30 groups) and survey-round fixed effects; and εij represents the stochastic error.15 In all subsequent estimations, the standard errors are corrected to allow for intra-community correlation. In addition, when yij is a binary outcome, a probit model is primarily exploited for the estimations and the marginal effects (ME) are reported.

Evaluating xij prior to the respondent (or her parents) engaging in marriage decisions is preferable. Hence, to capture the levels of wealth at a household's disposal at that point, in addition to her birth order and other standard controls (e.g., age), xij includes the number of siblings that had died, as well as those still living when she was 10 years old, based on the survey responses. The number of deceased siblings is included in xij, based on the assumption that the mortality information is positively correlated with the poverty status of the surveyed woman's original household. Conditional on the mortality data, the number of existing siblings may reflect a household's financial capacity to raise children.

Notably, descent principles in Malawi are closely linked to ethnic identity, resulting in marked differences in marriage-related customs, such as residential place (e.g., matrilocal and patrilocal) and bride wealth payment. For example, it is possible that male members of patrilineal ethnic groups are inclined to marry later than those belonging to matrilineal ones, because the former must accumulate sufficient wealth to pay the bride price. In addition, the descent system may also affect the frequency of marital dissolution by providing the wife with greater bargaining power in households belonging to matrilineal communities relative to patrilineal communities.

To explicitly separate the missionary influence from the effects of these cultural differences, xij also includes an indicator, equal to one for residents in a matrilineal community, as well as fixed effects for the languages typically spoken at home in a community (possible languages are Chewa, Lambya, Lomwe, Ngoni, Nkhonde, Nyakyusa, Nyanja, Sena, Senga, Sukwa, Tonga, Tumbuka, Yao and other). By matching IHS communities with the closest MDHS community in proximity (see Section 4.2.1 for the IHS and identification process), the IHS can be used as the data source for such community-level information that cannot be discerned directly from the MDHS data. Similarly, this study also controls for individual ethnicity fixed effects in some specifications based on the categories provided by the MDHS (Chewa, Lambya, Lomwe, Mang'anja, Ndali, Ngoni, Nkhonde, Nyanja, Sena, Tonga, Tumbuka, Yao and other).

While the Livingstonia Mission is one of the most important missions that introduced Christianity into Malawi, other missions also existed in this country and exerted various degrees of influence on the population (e.g., the Zambezi Industrial Mission headed by Joseph Booth). To control for the influence of other missions, an MDHS community's distance to the nearest early mission station was also included as a regressor. The distance is calculated based on the positional information of the early mission stations shown in Figure 1, sourced from Nunn (2010) (it should be noted, however, that exclusion of this control had a negligible effect on the implications obtained from the regression analysis).

By exploiting the distance to Livingstonia as a measure of the missionary influence on the ‘present non-Yao’ respondents, the empirical strategy implicitly assumes that the spatial mobility of the population has been completely limited at the ethnic level. However, the mission's involvement in political disputes between indigenous leaders, which was sometimes observed in the early periods of the missionary penetration, could have potentially altered the spatial distribution of ethnic groups to a certain degree.16 However, while ethnic conflicts and the associated population movement were common until the turn of the 20th century, such massive ethnic-level migration has not been supported by findings provided by relevant historical research (e.g., Pike, 1968; McCracken, 2012). The spatial distribution of linguistic groups based on the IHS data (see Figure 2) is also very similar to that of the ethnic groups demonstrated in Pike and Rimmington (1965, Fig. 32, p. 139).17 Thus, the assumption made here actually allows for the spatial mobility of the ethnic groups which, while potentially existing, likely would not have been strong enough to make the distance variable invalid. Moreover, even if such (unlikely) population displacement took place, the estimated value of α2 can still be interpreted as the ITT effect of the Livingstonia Mission.

Figure 2:

Spatial Distribution of the Most Common Language Spoken at Home in a community (IHS).

Notes: (1) Figure in () is the number of communities. (2) The background map is sourced from DIVA-GIS (http://www.diva-gis.org/datadown).

Figure 2:

Spatial Distribution of the Most Common Language Spoken at Home in a community (IHS).

Notes: (1) Figure in () is the number of communities. (2) The background map is sourced from DIVA-GIS (http://www.diva-gis.org/datadown).

Controlling for pre-determined conditions

A community's distance to Livingstonia may be correlated with its distances to other locations of relevance to the missionaries as well as to installations of the British Government maintaining the colonial state. If such distances had an independent influence on present-day marriage practices, the estimate would be biased. To alleviate this concern, this study attempts to control for pre-determined local conditions that characterised the entry and expansion of the missionary venture as well as for colonial administration.

Geographic and climate controls

The settlement pattern of the missionaries was influenced by several factors. As indicated by Johnson (1967) and Nunn (2010, 2014), these generally included health-related factors such as the availability of clean water and malaria-preventing geographic and climatic conditions (e.g., low temperature and high altitude); economic considerations such as access to trade routes to and from Europe (which is likely affected by railway networks in colonial periods), and the availability of fertile land needed for the creation of a cash-crop economy; and factors related to the mission's benevolent attitude towards slave trade cessation. Some of these points are discussed in Section 2.

To alleviate the concern that these factors would prevent causal inference of the missionary effects, an attempt was made to control for a large number of geographic and climate conditions that would have been encountered by the missionaries. However, pre-missionary a set of data suitable for empirical analyses was not available. Thus, given the assumption that these conditions have not noticeably changed over the last century, more recent information was utilised. In the subsequent analysis, information provided by a survey in the IHS 2010–2011 was employed, as a suitable equivalent was absent from the MDHS data set.

With technical assistance offered by the World Bank Living Standards Measurement Study-Integrated Surveys on Agriculture (LSMS-ISA) team, the NSO in Malawi implemented the IHS in the period from March 2010 to March 2011. Employing stratification based on geography, respondents belonging to 12,271 households in 768 enumeration areas (communities) were randomly contacted through the IHS, which provided information on various aspects of the welfare and socio-economic status of the population.18 The IHS data also contained abundant information on geography and climate pertinent to the surveyed communities, such as climatology, landscape typology, soil and terrain, and crop season parameters (see Appendix B for details).

As a part of the MDHS and IHS projects, the GPS coordinates of the surveyed communities were published after applying a random offset within a specified range to the positions (see Appendix C.1 for details). The GPS latitude and longitude position was utilised in the present study when calculating the GCD between the MDHS and IHS communities. Figure A1 depicts the locations of the sample communities in both the MDHS and IHS (for ease of visual identification, only the 2010 MDHS communities were charted with the IHS ones in the figure). Because both the MDHS and IHS communities were spatially dispersed across the country, an IHS community located relatively close to each community surveyed in the MDHS could be identified with ease (see Appendix C.2 for the details of the identification process). In fact, the nearest IHS community matched for approximately 95% (resp., 99%) of the MDHS communities were situated within 10 (15) km. Consequently, the geographic and climate information of the MDHS communities required for the analyses were based on the data pertaining to the nearest corresponding IHS communities. Therefore, these geographic and climate controls are measured at the community level. By performing several exercises, the goodness of fit of the community-level characteristics sourced from the IHS to individual characteristics of the MDHS females was also verified (see Appendix C.3).

Historical controls

The primary objective of utilising detailed information on geography and climate in a community in these analyses is controlling for the missionaries’ considerations of health-related factors and land productivity in selecting their settlement. While this information (e.g., elevation, slope and terrain roughness) may also be associated with the administration of trade routes from/to the coast and the intensity of slave trade, it may still be effective to control for these additional factors more directly. Thus, to reinforce the primary controls for geography and climate, the empirical analysis performed in this study also exploited additional covariates measuring European influence during colonial periods, as well as the severity of slavery during the 19th century.

All pertinent data were sourced from Nunn and Wantchekon (2011), which contained (i) an indicator that takes the value of one if a European explorer travelled through land historically inhabited by an ethnic group; (ii) a dummy variable equal to one if any part of a railway line in the first decade of the 20th century, as drawn from Century Company (1911), passed through land historically occupied by an ethnic group and (iii) the total number of slaves taken from an ethnic group, normalised by the area of land inhabited by the ethnic group during the 19th century (log of one plus the normalised slave export measure). Unlike the aforementioned geographic and climatic controls measured at the community level, these items were evaluated at the ethnic level.19 Thus, the information was appended to the MDHS data, using the names of ethnic groups sourced from the two independent data sets. Consequently, only a few ethnic groups in the MDHS that were not identified in Nunn and Wantchekon (2011)'s data were excluded from the subsequent regression analysis.20 These omitted groups represent approximately 5% of all females in the sample.

Estimation results

Early marriage

Table 3 presents the estimation results for Equation (1) concerning the practice of early marriage. For each outcome reported in this table (and Tables 4 and 6), the community-level geographic and climatic controls (climatology, landscape typology, soil and terrain, crop season parameters) are included in the first column, together with individual controls (age, birth order, number of living and deceased siblings at age 10); an indicator for matrilineal communities; a community's GPS-based coordinates; fixed effects of languages commonly spoken in a community; district fixed effects and survey-round fixed effects. The estimation in the second column additionally included ethnic-level historical controls. The historical controls were replaced by ethnicity fixed effects in the third column. Instead of a probit model, a linear probability model (LPM) was estimated in column (g) in Table 3 (and columns (d) and (h) in Table 4 and columns (d) and (k) in Table 6), whereby all the regressors, excluding the historical controls, were exploited together with the ethnicity fixed effects.

Table 3:

Missionary Influence on Early Marriage (non-Yao)

Dependent variables Age at first marriage One if married Years to first marriage 
Sample All ever married Aged below 18 Aged below 18 
OLS OLS OLS Probit
(ME) 
Probit
(ME) 
Probit
(ME) 
LPM Hazard
ratio 
Hazard
ratio 
Hazard
ratio 
(a) (b) (c) (d) (e) (f) (g) (h) (i) (j) 
 Distance to Livingstonia (100 km) −0.433**
(0.184) 
−0.325
(0.239) 
−0.324*
(0.193) 
0.144***
(0.051) 
0.128**
(0.053) 
0.134***
(0.052) 
0.172**
(0.069) 
3.007***
(1.275) 
2.739**
(1.319) 
2.786**
(1.204) 
 Distance to the nearest mission station (100 km) −0.559***
(0.197) 
−0.554***
(0.204) 
−0.527***
(0.196) 
0.083*
(0.044) 
0.078*
(0.043) 
0.085*
(0.044) 
0.095*
(0.050) 
1.763
(0.659) 
1.678
(0.642) 
1.781
(0.666) 
 Shoenfeld res. (p-val.)        0.860 0.230 0.936 
Oster (forthcoming)'s δ           
  Rmax=1.3R˜ −1.755 −2.937 −3.526    −0.524    
  Rmax = 1.0 −0.028 −0.035 −0.037    −0.049    
R2 0.037 0.037 0.038 0.146 0.146 0.148 0.110    
 No. of obs. 33,505 31,379 33,505 5,508 5,179 5,508 5,508 5,508 5,183 5,508 
Panel A: With a control of the distance to Blantyre (100 km) 
 Distance to Livingstonia (100 km) −0.386**
(0.189) 
−0.257
(0.245) 
−0.260
(0.198) 
0.133**
(0.053) 
0.124**
(0.055) 
0.123**
(0.053) 
0.158**
(0.071) 
2.526**
(1.115) 
2.453*
(1.245) 
2.373*
(1.070) 
 Distance to the nearest mission station (100 km) −0.539***
(0.199) 
−0.533***
(0.205) 
−0.501**
(0.198) 
0.078*
(0.044) 
0.077*
(0.043) 
0.081*
(0.044) 
0.089*
(0.051) 
1.632
(0.614) 
1.621
(0.623) 
1.661
(0.626) 
 Distance to Blantyre (100 km) −0.142
(0.147) 
−0.174
(0.150) 
−0.191
(0.144) 
0.032
(0.029) 
0.009
(0.029) 
0.032
(0.029) 
0.043
(0.037) 
1.546*
(0.382) 
1.263
(0.328) 
1.491
(0.370) 
 No. of obs. 33,505 31,379 33,505 5,508 5,179 5,508 5,508 5,508 5,183 5,508 
Panel B: With no control of other missions 
 Distance to −0.457** −0.354 −0.352* 0.148*** 0.131** 0.139*** 0.177** 3.110*** 2.789** 2.893** 
  Livingstonia (100 km) (0.185) (0.241) (0.195) (0.052) (0.053) (0.052) (0.070) (1.334) (1.347) (1.265) 
 No. of obs. 33,505 31,379 33,505 5,508 5,179 5,508 5,508 5,508 5,183 5,508 
Panel C: Only for Lomwe, Ngoni, Nyanja and Sena 
 Distance to −2.265 −2.032 −2.191 0.819 0.649 0.697 0.754    
  Livingstonia (100 km) (1.742) (1.769) (1.741) (0.537) (0.528) (0.530) (0.609)    
 No. of obs. 13,504 13,255 13,504 2,079 2,051 2,079 2,113    
Panel D: Non-Yao excluding Lomwe, Ngoni, Nyanja and Sena 
 Distance to −0.280 −0.079 −0.151 0.170*** 0.155*** 0.153*** 0.194*** 3.474** 3.552** 3.126** 
  Livingstonia (100 km) (0.193) (0.261) (0.203) (0.055) (0.056) (0.055) (0.072) (1.687) (1.898) (1.542) 
 No. of obs. 20,001 18,124 20,001 3,381 3,086 3,381 3,395 3,395 3,100 3,395 
Panel E: Only for Lomwe, Nyanja and Sena 
 Distance to −3.650 −4.568 −4.192 2.112 1.787 1.791 2.343    
  Livingstonia (100 km) (3.006) (3.086) (3.016) (1.742) (1.834) (3.318) (2.105)    
 No. of obs. 9,034 8,785 9,034 1,320 1,292 1,320 1,345    
Panel F: With time-varying ‘age’ 
 Distance to        2.994** 2.724** 2.771** 
  Livingstonia (100 km)        (1.270) (1.313) (1.197) 
 No. of obs.        5,508 5,183 5,508 
 Individual controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Matrilineal com. dummy Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 GPS coordinate Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Ethnicity FE No No Yes No No Yes Yes No No Yes 
 Community-language FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Geography and climate Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Historical controls No Yes No No Yes No No No Yes No 
 District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Round FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Dependent variables Age at first marriage One if married Years to first marriage 
Sample All ever married Aged below 18 Aged below 18 
OLS OLS OLS Probit
(ME) 
Probit
(ME) 
Probit
(ME) 
LPM Hazard
ratio 
Hazard
ratio 
Hazard
ratio 
(a) (b) (c) (d) (e) (f) (g) (h) (i) (j) 
 Distance to Livingstonia (100 km) −0.433**
(0.184) 
−0.325
(0.239) 
−0.324*
(0.193) 
0.144***
(0.051) 
0.128**
(0.053) 
0.134***
(0.052) 
0.172**
(0.069) 
3.007***
(1.275) 
2.739**
(1.319) 
2.786**
(1.204) 
 Distance to the nearest mission station (100 km) −0.559***
(0.197) 
−0.554***
(0.204) 
−0.527***
(0.196) 
0.083*
(0.044) 
0.078*
(0.043) 
0.085*
(0.044) 
0.095*
(0.050) 
1.763
(0.659) 
1.678
(0.642) 
1.781
(0.666) 
 Shoenfeld res. (p-val.)        0.860 0.230 0.936 
Oster (forthcoming)'s δ           
  Rmax=1.3R˜ −1.755 −2.937 −3.526    −0.524    
  Rmax = 1.0 −0.028 −0.035 −0.037    −0.049    
R2 0.037 0.037 0.038 0.146 0.146 0.148 0.110    
 No. of obs. 33,505 31,379 33,505 5,508 5,179 5,508 5,508 5,508 5,183 5,508 
Panel A: With a control of the distance to Blantyre (100 km) 
 Distance to Livingstonia (100 km) −0.386**
(0.189) 
−0.257
(0.245) 
−0.260
(0.198) 
0.133**
(0.053) 
0.124**
(0.055) 
0.123**
(0.053) 
0.158**
(0.071) 
2.526**
(1.115) 
2.453*
(1.245) 
2.373*
(1.070) 
 Distance to the nearest mission station (100 km) −0.539***
(0.199) 
−0.533***
(0.205) 
−0.501**
(0.198) 
0.078*
(0.044) 
0.077*
(0.043) 
0.081*
(0.044) 
0.089*
(0.051) 
1.632
(0.614) 
1.621
(0.623) 
1.661
(0.626) 
 Distance to Blantyre (100 km) −0.142
(0.147) 
−0.174
(0.150) 
−0.191
(0.144) 
0.032
(0.029) 
0.009
(0.029) 
0.032
(0.029) 
0.043
(0.037) 
1.546*
(0.382) 
1.263
(0.328) 
1.491
(0.370) 
 No. of obs. 33,505 31,379 33,505 5,508 5,179 5,508 5,508 5,508 5,183 5,508 
Panel B: With no control of other missions 
 Distance to −0.457** −0.354 −0.352* 0.148*** 0.131** 0.139*** 0.177** 3.110*** 2.789** 2.893** 
  Livingstonia (100 km) (0.185) (0.241) (0.195) (0.052) (0.053) (0.052) (0.070) (1.334) (1.347) (1.265) 
 No. of obs. 33,505 31,379 33,505 5,508 5,179 5,508 5,508 5,508 5,183 5,508 
Panel C: Only for Lomwe, Ngoni, Nyanja and Sena 
 Distance to −2.265 −2.032 −2.191 0.819 0.649 0.697 0.754    
  Livingstonia (100 km) (1.742) (1.769) (1.741) (0.537) (0.528) (0.530) (0.609)    
 No. of obs. 13,504 13,255 13,504 2,079 2,051 2,079 2,113    
Panel D: Non-Yao excluding Lomwe, Ngoni, Nyanja and Sena 
 Distance to −0.280 −0.079 −0.151 0.170*** 0.155*** 0.153*** 0.194*** 3.474** 3.552** 3.126** 
  Livingstonia (100 km) (0.193) (0.261) (0.203) (0.055) (0.056) (0.055) (0.072) (1.687) (1.898) (1.542) 
 No. of obs. 20,001 18,124 20,001 3,381 3,086 3,381 3,395 3,395 3,100 3,395 
Panel E: Only for Lomwe, Nyanja and Sena 
 Distance to −3.650 −4.568 −4.192 2.112 1.787 1.791 2.343    
  Livingstonia (100 km) (3.006) (3.086) (3.016) (1.742) (1.834) (3.318) (2.105)    
 No. of obs. 9,034 8,785 9,034 1,320 1,292 1,320 1,345    
Panel F: With time-varying ‘age’ 
 Distance to        2.994** 2.724** 2.771** 
  Livingstonia (100 km)        (1.270) (1.313) (1.197) 
 No. of obs.        5,508 5,183 5,508 
 Individual controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Matrilineal com. dummy Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 GPS coordinate Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Ethnicity FE No No Yes No No Yes Yes No No Yes 
 Community-language FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Geography and climate Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Historical controls No Yes No No Yes No No No Yes No 
 District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Round FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 

Notes: (1) Figures () are standard errors. ***Significance at 1%, **significance at 5% and *significance at 10%. (2) Standard errors are robust to heteroskedasticity and clustered residuals within each community. (3) The individual controls include age (years), birth order, no. of alive siblings at age 10 and no. of late siblings at age 10. (4) The ethnicity is classified into 12 groups, i.e., Chewa, Lambya, Lomwe, Mang'anja, Ndali, Ngoni, Nkhonde, Nyanja, Sena, Tonga, Tumbuka and other. (5) The community language is classified into 14 groups, i.e., Chewa, Lambya, Lomwe, Ngoni, Nkhonde, Nyakyusa, Nyanja, Sena, Senga, Sukwa, Tonga, Tumbuka, Yao and other. (6) The geographic and climate controls contain community-level information on climatology, landscape typology, soil and terrain, crop season parameters. See Appendix B for the details. (7) The ethnicity-level historical controls include (i) a dummy variable, equal to one if a European explorer travelled through land historically inhabited by an ethnic group; (ii) a dummy variable, equal to one if any part of railway lines in the first decade of the 20th century drawn from Century Company (1911) passed through land historically inhabited by an ethnic group and (iii) the total number of slaves taken from an ethnic group that was normalised by the area of land inhabited by the ethnic group during the 19th century (log of one plus the normalised slave export measure).

Table 4:

Missionary Influence on Polygyny (non-Yao)

Dependent variables One if polygyny One if polygyny (zero if unmarried) 
Sample All married All 
Probit
(ME) 
Probit
(ME) 
Probit
(ME) 
LPM Probit
(ME) 
Probit
(ME) 
Probit
(ME) 
LPM 
(a) (b) (c) (d) (e) (f) (g) (h) 
 Distance to 0.062*** 0.051* 0.064*** 0.079** 0.044*** 0.036* 0.041*** 0.060** 
  Livingstonia (100 km) (0.023) (0.028) (0.023) (0.031) (0.015) (0.018) (0.015) (0.023) 
 Distance to the nearest 0.114*** 0.111*** 0.114*** 0.123*** 0.089*** 0.087*** 0.090*** 0.106*** 
  mission station (100 km) (0.024) (0.024) (0.024) (0.027) (0.016) (0.018) (0.016) (0.020) 
Oster (forthcoming)'s δ         
  Rmax=1.3R˜    −0.684    −0.899 
  Rmax = 1.0    −0.019    −0.024 
R2 0.071 0.070 0.072 0.061 0.092 0.092 0.093 0.061 
 No. of obs. 28,411 26,623 28,411 28,411 41,130 38,635 41,130 41,130 
Panel A: With a control of the distance to Blantyre (100 km) 
 Distance to 0.043* 0.028 0.044* 0.057* 0.031** 0.020 0.028* 0.042* 
  Livingstonia (100 km) (0.023) (0.029) (0.024) (0.032) (0.015) (0.019) (0.015) (0.024) 
 Distance to the nearest 0.104*** 0.102*** 0.103*** 0.113*** 0.082*** 0.080*** 0.082*** 0.098*** 
  mission station (100 km) (0.024) (0.025) (0.024) (0.027) (0.017) (0.017) (0.017) (0.020) 
 Distance to 0.064*** 0.069*** 0.067*** 0.069*** 0.046*** 0.048*** 0.048*** 0.053*** 
  Blantyre (100 km) (0.019) (0.019) (0.018) (0.019) (0.013) (0.013) (0.012) (0.014) 
 No. of obs. 28,411 26,623 28,411 28,411 41,130 38,635 41,130 41,130 
Panel B: With no control of other missions 
 Distance to 0.064*** 0.054* 0.067*** 0.085*** 0.046*** 0.038** 0.044*** 0.065*** 
  Livingstonia (100 km) (0.023) (0.029) (0.023) (0.031) (0.015) (0.019) (0.015) (0.023) 
 No. of obs. 28,411 26,623 28,411 28,411 41,130 38,635 41,130 41,130 
Panel C: Only for Lomwe, Ngoni, Nyanja and Sena 
 Distance to 0.277* 0.297* 0.281* 0.368 0.214** 0.228** 0.217** 0.325*** 
  Livingstonia (100 km) (0.161) (0.163) (0.161) (0.227) (0.102) (0.103) (0.102) (0.115) 
 No. of obs. 11,084 10,870 11,084 11,093 16,426 16,132 16,426 16,438 
Panel D: Non-Yao excluding Lomwe, Ngoni, Nyanja and Sena 
 Distance to 0.059** 0.050 0.061** 0.070** 0.042** 0.032 0.038*** 0.051** 
  Livingstonia (100 km) (0.027) (0.035) (0.028) (0.033) (0.018) (0.023) (0.019) (0.024) 
 No. of obs. 17,318 15,730 17,318 17,318 24,692 22,475 24,692 24,692 
Panel E: Only for Lomwe, Nyanja and Sena 
 Distance to 0.521 0.624 0.524 0.699* 0.405 0.482* 0.398 0.661** 
  Livingstonia (100 km) (0.329) (0.354) (0.331) (0.370) (0.230) (0.233) (0.233) (0.310) 
 No. of obs. 7,251 7,041 7,251 7,383 10,586 10,298 10,586 10,782 
 Individual controls Yes Yes Yes Yes Yes Yes Yes Yes 
 Matrilineal com. dummy Yes Yes Yes Yes Yes Yes Yes Yes 
 GPS coordinate Yes Yes Yes Yes Yes Yes Yes Yes 
 Ethnicity FE No No Yes Yes No No Yes Yes 
 Community-language FE Yes Yes Yes Yes Yes Yes Yes Yes 
 Geography and climate Yes Yes Yes Yes Yes Yes Yes Yes 
 Historical controls No Yes No No No Yes No No 
 District FE Yes Yes Yes Yes Yes Yes Yes Yes 
 Round FE Yes Yes Yes Yes Yes Yes Yes Yes 
Dependent variables One if polygyny One if polygyny (zero if unmarried) 
Sample All married All 
Probit
(ME) 
Probit
(ME) 
Probit
(ME) 
LPM Probit
(ME) 
Probit
(ME) 
Probit
(ME) 
LPM 
(a) (b) (c) (d) (e) (f) (g) (h) 
 Distance to 0.062*** 0.051* 0.064*** 0.079** 0.044*** 0.036* 0.041*** 0.060** 
  Livingstonia (100 km) (0.023) (0.028) (0.023) (0.031) (0.015) (0.018) (0.015) (0.023) 
 Distance to the nearest 0.114*** 0.111*** 0.114*** 0.123*** 0.089*** 0.087*** 0.090*** 0.106*** 
  mission station (100 km) (0.024) (0.024) (0.024) (0.027) (0.016) (0.018) (0.016) (0.020) 
Oster (forthcoming)'s δ         
  Rmax=1.3R˜    −0.684    −0.899 
  Rmax = 1.0    −0.019    −0.024 
R2 0.071 0.070 0.072 0.061 0.092 0.092 0.093 0.061 
 No. of obs. 28,411 26,623 28,411 28,411 41,130 38,635 41,130 41,130 
Panel A: With a control of the distance to Blantyre (100 km) 
 Distance to 0.043* 0.028 0.044* 0.057* 0.031** 0.020 0.028* 0.042* 
  Livingstonia (100 km) (0.023) (0.029) (0.024) (0.032) (0.015) (0.019) (0.015) (0.024) 
 Distance to the nearest 0.104*** 0.102*** 0.103*** 0.113*** 0.082*** 0.080*** 0.082*** 0.098*** 
  mission station (100 km) (0.024) (0.025) (0.024) (0.027) (0.017) (0.017) (0.017) (0.020) 
 Distance to 0.064*** 0.069*** 0.067*** 0.069*** 0.046*** 0.048*** 0.048*** 0.053*** 
  Blantyre (100 km) (0.019) (0.019) (0.018) (0.019) (0.013) (0.013) (0.012) (0.014) 
 No. of obs. 28,411 26,623 28,411 28,411 41,130 38,635 41,130 41,130 
Panel B: With no control of other missions 
 Distance to 0.064*** 0.054* 0.067*** 0.085*** 0.046*** 0.038** 0.044*** 0.065*** 
  Livingstonia (100 km) (0.023) (0.029) (0.023) (0.031) (0.015) (0.019) (0.015) (0.023) 
 No. of obs. 28,411 26,623 28,411 28,411 41,130 38,635 41,130 41,130 
Panel C: Only for Lomwe, Ngoni, Nyanja and Sena 
 Distance to 0.277* 0.297* 0.281* 0.368 0.214** 0.228** 0.217** 0.325*** 
  Livingstonia (100 km) (0.161) (0.163) (0.161) (0.227) (0.102) (0.103) (0.102) (0.115) 
 No. of obs. 11,084 10,870 11,084 11,093 16,426 16,132 16,426 16,438 
Panel D: Non-Yao excluding Lomwe, Ngoni, Nyanja and Sena 
 Distance to 0.059** 0.050 0.061** 0.070** 0.042** 0.032 0.038*** 0.051** 
  Livingstonia (100 km) (0.027) (0.035) (0.028) (0.033) (0.018) (0.023) (0.019) (0.024) 
 No. of obs. 17,318 15,730 17,318 17,318 24,692 22,475 24,692 24,692 
Panel E: Only for Lomwe, Nyanja and Sena 
 Distance to 0.521 0.624 0.524 0.699* 0.405 0.482* 0.398 0.661** 
  Livingstonia (100 km) (0.329) (0.354) (0.331) (0.370) (0.230) (0.233) (0.233) (0.310) 
 No. of obs. 7,251 7,041 7,251 7,383 10,586 10,298 10,586 10,782 
 Individual controls Yes Yes Yes Yes Yes Yes Yes Yes 
 Matrilineal com. dummy Yes Yes Yes Yes Yes Yes Yes Yes 
 GPS coordinate Yes Yes Yes Yes Yes Yes Yes Yes 
 Ethnicity FE No No Yes Yes No No Yes Yes 
 Community-language FE Yes Yes Yes Yes Yes Yes Yes Yes 
 Geography and climate Yes Yes Yes Yes Yes Yes Yes Yes 
 Historical controls No Yes No No No Yes No No 
 District FE Yes Yes Yes Yes Yes Yes Yes Yes 
 Round FE Yes Yes Yes Yes Yes Yes Yes Yes 

Notes: (1) Figures () are standard errors. ***Significance at 1%, **significance at 5% and *significance at 10%. (2) Standard errors are robust to heteroskedasticity and clustered residuals within each community. (3) The individual controls include age (years), birth order, no. of alive siblings at age 10 and no. of late siblings at age 10. (4) The ethnicity is classified into 12 groups, i.e., Chewa, Lambya, Lomwe, Mang'anja, Ndali, Ngoni, Nkhonde, Nyanja, Sena, Tonga, Tumbuka and other. (5) The community language is classified into 14 groups, i.e., Chewa, Lambya, Lomwe, Ngoni, Nkhonde, Nyakyusa, Nyanja, Sena, Senga, Sukwa, Tonga, Tumbuka, Yao and other. (6) The geographic and climate controls contain community-level information on climatology, landscape typology, soil and terrain, crop season parameters. See Appendix B for the details. (7) The ethnicity-level historical controls include (i) a dummy variable, equal to one if a European explorer travelled through land historically inhabited by an ethnic group; (ii) a dummy variable, equal to one if any part of railway lines in the first decade of the 20th century drawn from Century Company (1911) passed through land historically inhabited by an ethnic group and (iii) the total number of slaves taken from an ethnic group that was normalised by the area of land inhabited by the ethnic group during the 19th century (log of one plus the normalised slave export measure).

With reasonably strong significance, the estimated missionary effects on early marriage revealed a relatively stable pattern across the columns in Table 3. The estimation results for age (in years) at first marriage presented in columns (a)–(c) indicate that, compared with non-Yao females living in closer proximity to Livingstonia, females residing further from the mission station are more likely to marry early. The age at marriage is recorded only for females who have been married at the time or prior to the DHS. To alleviate potential selection concerns arising from using only data pertaining to those females, the probability of being married was also estimated based on data pertaining to both single and married female respondents under the age of 18 years, and the results are presented in columns (d)–(g) in Table 3. Due to their young age, the likelihood of these females being widowed or divorced was minimal; therefore, the married/unmarried distinction is simplified to a married/single dichotomy. As the results show, the Livingstonia Mission has contributed to reducing the likelihood of females entering into marriages at a very young age.

To assess the robustness of this finding, the number of years from birth to first marriage was also analysed using a Cox proportional hazard model for both single and married female respondents younger than 18 years of age. The estimated hazard ratio is reported in columns (h)–(j) in Table 3, whereby a ratio greater (resp., smaller) than one indicates that the variable encourages (discourages) early marriage. Due to the proportional hazard assumption, the ratio should be interpreted as the hazard ratio at any point in time between two individuals that only vary by one unit of covariates. Consistent with the results presented in the previous columns, these findings support the premise that the influence of the Livingstonia Mission discouraged females from marrying early. The Schoenfeld residual (p-values) reported at the bottom of Table 3 also failed to reject the assumption of proportional hazard. While all covariates were assumed to be time invariant in the benchmark survival analysis, in the estimations performed in panel (F) at the bottom of the table, respondent age was allowed to explicitly vary over time. The reported hazard ratio relevant to the distance to Livingstonia did not alter the implications yielded by the benchmark analysis.

While the upper bound of respondents’ age used in the estimations presented in columns (d)–(j) in Table 3 was set at 18, which is the current minimum legal age of marriage in Malawi, variation in the upper bound allows analysis of different samples of data pertaining to females included in this study.21 The estimated missionary influence is graphically reported in Figure 3 with 95% confidence intervals, with the estimate for age m shown on the horizontal axis from the regression using data on females aged 15 to m–1 years.22 As this figure shows, the missionary effects discouraging early marriage are robust even when using other cut-off values. Moreover, it appears that such effects gradually diminish as the upper bound imposed on female age increases, which is particularly evident in the estimation results of the hazard model (right-hand panel). Given that almost all women in Malawi enter into a (first) marital relationship by their mid-20s, this finding suggests that the Livingstonia Mission indeed prevented ‘early’ marriage, rather than discouraging marriage ‘generally’.

Figure 3:

Missionary Influence on Early Marriage (non-Yao): Different Cut-off of Age.

Notes: (1) This figure reports α2 in Equation (1) with 95% confidence intervals by changing the exploited sample by the respondents’ age. (2) Age m in the horizontal axis means that the estimation uses data pertaining to female respondents aged 15 to m–1 years. (3) The regressors of the probit model and of the hazard model are the same as those used in columns (f) and (j) in Table 3, respectively.

Figure 3:

Missionary Influence on Early Marriage (non-Yao): Different Cut-off of Age.

Notes: (1) This figure reports α2 in Equation (1) with 95% confidence intervals by changing the exploited sample by the respondents’ age. (2) Age m in the horizontal axis means that the estimation uses data pertaining to female respondents aged 15 to m–1 years. (3) The regressors of the probit model and of the hazard model are the same as those used in columns (f) and (j) in Table 3, respectively.

Polygyny

In the analysis presented in columns (a)–(d) in Table 4, an indicator for polygyny, equal to one if the marriage type was polygyny and zero otherwise, was estimated. The results suggest that the Livingstonia Mission discouraged females from engaging in polygyny. Similar to the estimations of age at marriage provided in Table 3, the marriage type can be observed only for currently married females.23 As the data pertaining to divorced or widowed females were not used in these regressions, bias could potentially be introduced into the estimates if those females had unobserved personal traits correlated with incidence of polygyny. To mitigate this potential selection concern, as well as to estimate the missionary influence on the likelihood of entering into marital relationships and remaining in polygyny, the polygyny indicator was again estimated for both single and married respondents, with the results reported in columns (e)–(h). For currently unmarried females, the indicator was assumed to take the value of zero. These additional exercises also provided support for the view that the exposure to the Livingstonia Mission reduced the likelihood of females engaging in polygynous relationships.

Based on Yatchew (1997, 1998)'s difference-based semi-parametric estimation of a partial lineal, Figure A2 (upper panels) also reports the estimated non-parametric function of the missionary influence. Clearly, a community's distance to Livingstonia positively correlates with the likelihood of non-Yao females engaging in early marriage and polygyny.24

As the identified missionary influence on marriage practices suggests, a distance one standard deviation above the mean to Livingstonia (187.81 km) increases the probability of non-Yao females below the age of 18 entering into marriages and of non-Yao females of all ages engaging in polygyny by approximately 24–27% and 7–8%, respectively. It appears that these impacts are remarkably large, given the fact that the mean proportions of corresponding females in such early marital and polygynous relationships are 13% and 10%, respectively.

Based on the estimated effects of a community's distance to the nearest early mission station reported in Tables 3 and 4, other Christian missions also discouraged female engagement in early marriage and polygynous relationships. While the statistical significance of the estimates relevant to early marriage is more pronounced in the estimations of age at first marriage (at the 1% level) and the probability of entering into marriage before the age of 18 years (at the 10% level), the corresponding estimates achieve statistical significance at the 12% level, even in the analyses of years to the first marriage performed in columns (h) and (j) in Table 3.

In addition, along with the Livingstonia Mission, the Blantyre Mission of the Church of Scotland was also an influential pioneering mission in Malawi (see also Appendix A.3 for more information and Figure 1 for the position of Blantyre). Given its potential significance, an MDHS community's distance to Blantyre was also included in the analyses (latitude, 15°47′S; longitude, 35°02′E) as an additional control, and the estimates are reported in panel (A) at the bottom of Tables 3 and 4. As the results suggest, this pioneering mission has also contributed to the avoidance of early marriage (as inferred from the sign of the estimates) and polygyny.25 Compared with this mission, the non-state Presbyterian nature of the Livingstonia and its rigorous teachings might have had stronger influence on the likelihood of females entering into marriage at a very young age.

Moreover, in the estimations performed in panel (B) in Tables 3 and 4, all the variables related to the influence of other missions (i.e., a community's distances to the nearest early mission station and to Blantyre) were excluded from the regressors while retaining distance to Livingstonia. As the results show, the previously identified influence of the Livingstonia Mission is robust to the presence or absence of all controls related to other missions.

Robustness checks

Even though numerous controls were exploited in the previous estimations, it is still important to conduct more stringent tests in order to verify that the estimated effects are not attributed to other socio-economic factors that correlate with a community's distance to Livingstonia. Several discussions on this issue are presented below.

Influence of the ‘South’

Given the thin strip of land occupied by Malawi and the location of Livingstonia, in many cases, communities located far from Livingstonia are located in southern areas. This spatial distribution raises a concern that the distance to Livingstonia may indeed be a proxy for the distance to or location in the ‘South’. In particular, the Yao enjoyed prosperity and strong political power before the arrival of the mission because of the wealth they had accumulated from exporting ivory and slaves to the Arabs. Consequently, the impacts of distance to Livingstonia reported in Tables 3 and 4 may reflect effects related to the prosperity enjoyed by the Yao in the past. To mitigate this concern, in this study, the previous estimations were also conducted for a subset of the population belonging to the Lomwe, Ngoni, Nyanja and Sena ethnic groups only. As seen from Figure 2, these groups primarily reside in the more southerly regions of the country compared with the Yao. Consequently, for any two communities A and B within this selected sample, it is highly likely that community A is located at greater distance from Livingstonia than community B, which is also positioned further from the Yao residential area than the community B. If the estimation results based on the data drawn from these selected groups yield similar implications to those obtained from the analysis presented in Tables 3 and 4, the distance variable is indeed likely to capture the influence of the ‘North’ (i.e., the mission).

The results reported in panel (C) at the bottom of Tables 3 and 4 support this original interpretation.26 Moreover, the impacts of the distance to Livingstonia were also estimated for the remaining non-Yao ethnic groups, with the results presented in panel (D). The implications of these results are very similar to those yielded by the main estimation results explained in Sections 5.1 and 5.2. Since part of the Ngoni were historically located in the north (Pike and Rimmington, 1965, Fig. 32, p. 139) and were influenced by Christian missionary activities (Thompson, 1995), the same exercises were repeated in panel (E), using data only on group members belonging to the Lomwe, Nyanja and Sena only. The estimated effects of the distance to Livingstonia still suggest a marked influence of the ‘North’, although the impacts appear to be somewhat imprecisely estimated, as seen from the increases in the estimates’ magnitude and the associated standard errors. The loss of precision in the estimates reported in panel (E) (as well as in panel (C)) is likely due not only to the reduction in the sample size, but also less variation in the distance within these ethnic groups, which are primarily located in the southerly regions relative to that noted within the whole sample. Finally, it should also be recalled that in the analyses presented in panel (A) of Tables 3 and 4, the influence of the Livingstonia Mission was robust when controlling for the distance to Blantyre, which is located in the Southern Province.

Pre-missionary economic prosperity

In this section, an attempt was also made to examine the relationship between the distance to Livingstonia and the marital outcomes in the periods ‘before’ the missionary was founded. The premise for this analysis is that the absence of a significant correlation between these variables in pre-missionary periods may imply that the findings presented in Tables 3 and 4 can indeed be attributed to the missionary impacts.

While no information on marriage practices in the late 19th century is available, it is likely that such behaviour was correlated with economic prosperity, which may in turn be measured by population density (Acemoglu et al., 2002). Therefore, in this study, the relationship between the pre-missionary population density and the distance was explored.

While it is relatively difficult to obtain population data for the historical periods predating the arrival of the mission, this study benefitted from two independent data sets. The first set of historical population data was sourced from the History Database of the Global Environment (HYDE) 3.1. Within the field of economics, Fenske (2013) has recently used this database to explain pre-colonial land tenure and slavery in Africa. The HYDE provides estimates of historical population from 10,000 BC to 2005 AD with a spatial resolution of 5-min longitude/latitude in raster format (Goldewijk et al., 2010). The present investigation employed the data on population density in 1900 plotted in Figure 4. In the regression analysis, the 5 × 5-min cell serves as the unit of observation and, accordingly, the distance to Livingstonia was calculated as that between the central point of each cell and Livingstonia.

Figure 4:

Historical Population Density in 1900 (Inhabitants/km2).

Notes: (1) The data on population density are sourced from History Database of the Global Environment (HYDE) version 3.1. (2) The map of Malawi is sourced from DIVA-GIS (http://www.diva-gis.org/datadown).

Figure 4:

Historical Population Density in 1900 (Inhabitants/km2).

Notes: (1) The data on population density are sourced from History Database of the Global Environment (HYDE) version 3.1. (2) The map of Malawi is sourced from DIVA-GIS (http://www.diva-gis.org/datadown).

Since historical population estimates are unavoidably imprecise, in this study, census data were also employed as an alternative to the HYDE. However, it appeared that the earliest census data related to this empirical analysis can only be sourced from the post-colonial periods, more precisely, the ‘Malawi Population Census 1966’. To compensate for the disadvantages of having to use data on the post-colonial population density, the association of the distance to Livingstonia with the density of the population aged above 60 years was specifically examined. This exercise is based on the presumption that the density of the elderly cohort in the 1960s can serve as a proxy for the true density of the total population in the late 19th or early 20th century. However, this assumption may not be valid because the longevity of the cohort prior to the middle 20th century may be correlated with the process of economic development in the areas where the elderly resided. Acknowledging this potential selection bias, nevertheless, the population density of the elderly cohort identified in the 1966 census was also exploited. This analysis used the distance between a district's capital city or major town and Livingstonia.

Table 5 presents the estimation results relating the pre-missionary population density to the distance to Livingstonia. The estimation results shown in columns (1a) and (2a) are derived from the population data sourced from the HYDE, whereas the results in all the other columns are based on the 1966 census data. In columns (1a) and (2a), the unit of observation is a 5 × 5-min cell. On the other hand, a district (belonging to 23 groups) is employed as the unit of observation in columns (1b), (1c), (2b) and (2c). Finally, the results presented in columns (1d), (1e), (2d) and (2e) are based on the district–age–gender cohort (a total of 92 groups). The district–age–gender cohort consists of 23 districts × two age cohorts (age 60–64 and greater than 65 years) × two groups by gender.

Table 5:

Population Density in the Early 20th Century (OLS)

Dependent variable Log of population density 
Data sources of population HYDE Malawi Population Census 1966, Final report 
Population in 1900 Population in 1966 
60 years or above Age cohorts aged 60 years or above 
(1a) (1b) (1c) (1d) (1e) 
Distance to Livingstonia (100 km) 0.240 0.330 0.196 0.329 0.329 
(0.196) (0.230) (0.277) (0.202) (0.204) 
Male proportion   6.300  −0.014 
  (8.421)  (0.034) 
Average age   −0.143   
  (0.182)   
65 years or above (dummy)     0.880*** 
    (0.063) 
Longitude 0.012 0.452 0.668 0.550** 0.550** 
(0.288) (0.282) (0.415) (0.215) (0.217) 
Latitude −0.171 −0.079 −0.181 −0.061 −0.061 
(0.158) (0.186) (0.237) (0.159) (0.161) 
R2 0.378 0.629 0.665 0.466 0.715 
No. of obs. 1,115 23 23 92 92 
Dependent variable Log of population density 
Data sources of population HYDE Malawi Population Census 1966, Final report 
Population in 1900 Population in 1966 
60 years or above Age cohorts aged 60 years or above 
(1a) (1b) (1c) (1d) (1e) 
Distance to Livingstonia (100 km) 0.240 0.330 0.196 0.329 0.329 
(0.196) (0.230) (0.277) (0.202) (0.204) 
Male proportion   6.300  −0.014 
  (8.421)  (0.034) 
Average age   −0.143   
  (0.182)   
65 years or above (dummy)     0.880*** 
    (0.063) 
Longitude 0.012 0.452 0.668 0.550** 0.550** 
(0.288) (0.282) (0.415) (0.215) (0.217) 
Latitude −0.171 −0.079 −0.181 −0.061 −0.061 
(0.158) (0.186) (0.237) (0.159) (0.161) 
R2 0.378 0.629 0.665 0.466 0.715 
No. of obs. 1,115 23 23 92 92 
 (2a) (2b) (2c) (2d) (2e) 
Distance to Livingstonia (100 km) × non-Yao (proportion) −0.369 −0.821 −0.064 −0.797 −0.797 
(0.436) (0.930) (1.197) (0.764) (0.773) 
Distance to Livingstonia (100 km) 0.577 1.256 0.444 1.194 1.194 
(0.475) (0.972) (1.205) (0.797) (0.806) 
Non-Yao (proportion) 1.824 3.198 −0.924 3.319 3.319 
(1.961) (4.058) (5.882) (3.346) (3.387) 
Male proportion   11.847  −0.014 
  (15.624)  (0.034) 
Average age   −0.108   
  (0.159)   
65 years or above (dummy)     0.880*** 
    (0.064) 
Longitude 0.021 0.406 0.581* 0.528** 0.528** 
(0.275) (0.252) (0.337) (0.192) (0.194) 
Latitude −0.179 −0.110 −0.234 −0.082 −0.082 
(0.157) (0.195) (0.240) (0.159) (0.161) 
R2 0.383 0.661 0.702 0.483 0.732 
No. of obs. 1,115 23 23 92 92 
Unit of obs. Cell District District Cohort Cohort 
Regional FE Yes Yes Yes Yes Yes 
 (2a) (2b) (2c) (2d) (2e) 
Distance to Livingstonia (100 km) × non-Yao (proportion) −0.369 −0.821 −0.064 −0.797 −0.797 
(0.436) (0.930) (1.197) (0.764) (0.773) 
Distance to Livingstonia (100 km) 0.577 1.256 0.444 1.194 1.194 
(0.475) (0.972) (1.205) (0.797) (0.806) 
Non-Yao (proportion) 1.824 3.198 −0.924 3.319 3.319 
(1.961) (4.058) (5.882) (3.346) (3.387) 
Male proportion   11.847  −0.014 
  (15.624)  (0.034) 
Average age   −0.108   
  (0.159)   
65 years or above (dummy)     0.880*** 
    (0.064) 
Longitude 0.021 0.406 0.581* 0.528** 0.528** 
(0.275) (0.252) (0.337) (0.192) (0.194) 
Latitude −0.179 −0.110 −0.234 −0.082 −0.082 
(0.157) (0.195) (0.240) (0.159) (0.161) 
R2 0.383 0.661 0.702 0.483 0.732 
No. of obs. 1,115 23 23 92 92 
Unit of obs. Cell District District Cohort Cohort 
Regional FE Yes Yes Yes Yes Yes 

Notes: (1) The unit of observations is a 5 × 5-min cell demonstrated in Figure 4 in columns (1a) and (2a); a district in columns (1b), (1c), (2b) and (2c); and a district–age–gender cohort in columns (1d), (1e), (2d) and (2e). (2) Figures () are standard errors. ***Significance at 1%, **significance at 5% and *significance at 10%. (3) Standard errors are robust to heteroskedasticity and clustered residuals within each district. (4) The non-Yao proportion is the proportion of the non-Yao population relative to a district's overall population in 1966. See also Section 5.3.2 for the details. (5) The latitude and longitude are the coordinates of (a centroid of) each cell in columns (1a) and (1b), and those of a district's capital city (major town) in all the other columns. (6) The distance is measured as that between Livingstonia and a centroid of each cell in columns (1a) and (2a), and that between Livingstonia and a district's capital city (major town) in all the other columns. (7) In columns (1b), (1c), (2b) and (2c) (columns (1d), (1e), (2d) and (2e)), the population density is a district's (a cohort's) population (African, European, Asian and other) divided by land area (square mile). In columns (1a) and (2a), the density is evaluated per square kilometre. (8) In columns (1c) and (2c), a district's average age is included as controls. It is calculated by weighting based on the population in each age category of below 5 years, assumed to be 1-year old in the calculation; 5–9 years, assumed to be 5-year old; 10–14 years, assumed to be 10-year old; 15–19 years, assumed to be 15-year old; 20–24 years, assumed to be 20-year old; 25–29 years, assumed to be 25-year old; 30–34 years, assumed to be 30-year old; 35–39 years, assumed to be 35-year old: 40–44 years, assumed to be 40-year old; 45–49 years, assumed to be 45-year old; 50–54 years, assumed to be 50-year old; 55–59 years, assumed to be 55-year old; 60–64 years, assumed to be 60-year old; and 65 years or above, assumed to be 65- year old. (9) In columns (1e) and (2e), an indicator for a cohort aged 65 years or above is included as a control for age. (10) The analysis uses three regional fixed effects (i.e., north, central and south).

First, as the results reported in columns (1a)–(1e) demonstrate, the distance to Livingstonia had no significant relationship with the pre-missionary population density.27 In the context of the current study, however, the main objective of these exercises is verifying that the population density is not correlated with the distance within the non-Yao population. Hence, the estimations presented in columns (2a)–(2e) attempted to examine the interaction of distance with the proportion of the non-Yao population relative to the total population of an observational unit (i.e., 5 × 5-min cell, district or cohort) and explore the statistical relationship between the interaction term and the pre-missionary population density. However, the HYDE does not contain population information for specific ethnic groups. As a result, the result in column (2a) is generated by an analysis exploiting the non-Yao proportion of a district in 1966 in which each cell in Figure 4 is located. On the other hand, in the analyses reported in columns (2b)–(2e), the exact non-Yao proportion of each district in 1966 was utilised. These results imply that the interaction term between distance and the non-Yao proportion had no significant association with the pre-missionary population density. While this examination relies on the assumption that a statistically systematic relationship between population density and marital practices existed prior to the arrival of the missionaries, these findings still support the view that the significant impacts of the distance reported in Tables 3 and 4 can be attributed to the missionary influence.28

Abandoned mission station

As described in Section 2, the Livingstonia Mission initially established its central station at Cape Maclear, which was eventually abandoned due to its unfavourable environmental factors. Drawing upon the work of Valencia-Caicedo (2014), the aim of the analyses presented in this section is to demonstrate the lack of effect of the missionary activities at Cape Maclear on present marital practices, as measured by a community's distance to this station. Although the absence of Cape Maclear influence does not necessarily imply that a community's distance to Livingstonia has no correlation with other unobserved determinants of marriage outcomes, it is still interesting to conduct this exercise.

The relevant estimation results on Cape Maclear are reported in Table A1. These results do not offer strong support to the view that the missionary services provided at Cape Maclear discouraged early marriage and polygyny, while encouraging religious conversion or educational attainment of local people.29 On the other hand, the Livingstonia's influence is robust to the inclusion of the distance to Cape Maclear as one of the regressors. As argued by Valencia-Caicedo (2014), these findings may suggest that the missionaries’ activities, rather than their original settlement locations, had long-term influence on the economic development of the local community.

Assessment of bias attributed to unobservables

By exploiting an approach elaborated by Oster (forthcoming) based on Altonji et al. (2005), the importance of omitted variables that are required to explain the missionary effects and share covariance properties with the observed controls was assessed. Oster (forthcoming) developed a strategy for calculating the importance of such unobservables, denoted as δ (i.e., the coefficient of proportionality on selection assumptions). A value of δ > 1 indicates that the unobservables are more important than the observables. On the other hand, a negative δ value suggests that including the unobserved controls in regressions increases the magnitude of the estimated effect, rather than absorbing the effect size.

In adopting this approach, it is necessary to employ the value of R2 obtained from a hypothetical regression of the outcome on the treatment, observed and unobserved controls, denoted as Rmax. In the present study, two values of Rmax were utilised. Referring to the value of R2 arising from a regression on the treatment and observed controls as R˜, Oster (forthcoming) heuristically suggested Rmax=1.3R˜. Here, for each value of R2 corresponding to the OLS estimation results reported in Tables 3, 4 and 6, 1.3R˜ was first adopted as the Rmax estimate, while Rmax = 1 served as the alternative, least conservative value. The δ values corresponding to the OLS estimation results were reported in Tables 3, 4 and 6. As all values were negative, it is likely that the previous marriage effects were attenuated if any bias existed.

Table 6:

Missionary Influence on Christianity and Education (non-Yao)

Dependent variables One if Christian Education (years) One if unable to read 
Sample All All All 
Probit (ME) Probit (ME) Probit (ME) LPM OLS OLS OLS Probit (ME) Probit (ME) Probit (ME) LPM 
(a) (b) (c) (d) (e) (f) (g) (h) (i) (j) (k) 
 Distance to −0.044*** −0.060*** −0.036** −0.082*** −1.546*** −1.318*** −0.993*** 0.207*** 0.177*** 0.138*** 0.123*** 
  Livingstonia (100 km) (0.015) (0.017) (0.015) (0.021) (0.330) (0.411) (0.333) (0.040) (0.047) (0.040) (0.036) 
 Distance to the nearest 0.010 0.010 0.010 −0.015 −2.673*** −2.614*** −2.649*** 0.259*** 0.244*** 0.257*** 0.235*** 
  mission station (100 km) (0.013) (0.013) (0.012) (0.024) (0.351) (0.358) (0.343) (0.039) (0.039) (0.039) (0.036) 
Oster (forthcoming)'s δ            
  Rmax=1.3R˜    −1.416 −0.879 −1.180 −2.079    −1.187 
  Rmax = 1.0    −0.117 −0.183 −0.233 −0.351    −0.113 
R2 0.209 0.219 0.220 0.127 0.246 0.249 0.258 0.103 0.105 0.108 0.131 
 No. of obs. 41,118 38,551 40,949 41,118 41,129 38,634 41,129 41,033 38,557 41,033 41,033 
Panel A: With a control of the distance to Blantyre (100 km) 
 Distance to −0.048*** −0.065*** −0.041*** −0.087*** −1.268*** −0.998** −0.734** 0.172*** 0.138*** 0.106*** 0.093** 
  Livingstonia (100 km) (0.015) (0.018) (0.015) (0.022) (0.342) (0.428) (0.342) (0.041) (0.049) (0.041) (0.037) 
 Distance to the nearest 0.007 0.007 0.006 −0.017 −2.552*** −2.509*** −2.537*** 0.243*** 0.232*** 0.244*** 0.222*** 
  mission station (100 km) (0.013) (0.013) (0.012) (0.023) (0.346) (0.354) (0.339) (0.038) (0.034) (0.038) (0.035) 
 Distance to 0.010 0.011 0.013 0.014 −0.857*** −0.829*** −0.785*** 0.108*** 0.100*** 0.096*** 0.091*** 
  Blantyre (100 km) (0.008) (0.008) (0.008) (0.013) (0.243) (0.248) (0.244) (0.028) (0.029) (0.028) (0.027) 
 No. of obs. 41,118 38,551 40,949 41,118 41,129 38,634 41,129 41,033 38,557 41,033 41,033 
Panel B: With no control of other missions 
 Distance to −0.043*** −0.059*** −0.035** −0.083*** −1.663*** −1.458*** −1.134*** 0.215*** 0.186*** 0.149*** 0.136*** 
  Livingstonia (100 km) (0.015) (0.017) (0.015) (0.021) (0.339) (0.421) (0.341) (0.042) (0.048) (0.042) (0.038) 
 No. of obs. 41,118 38,551 40,949 41,118 41,129 38,634 41,129 41,033 38,557 41,033 41,033 
Panel C: Only for Lomwe, Ngoni, Nyanja and Sena 
 Distance to −0.296*** −0.298*** −0.270** −0.344*** −4.705*** −4.004** −4.183** 1.108*** 0.966*** 1.024*** 0.634*** 
  Livingstonia (100 km) (0.124) (0.683) (0.120) (0.113) (1.804) (1.757) (1.760) (0.305) (0.306) (0.302) (0.205) 
 No. of obs. 15,990 15,714 15,990 16,437 16,438 16,160 16,438 16,393 16,116 16,393 16,399 
Panel D: Non-Yao excluding Lomwe, Ngoni, Nyanja and Sena 
 Distance to −0.038*** −0.054*** −0.034** −0.100*** −1.539*** −1.455*** −0.924*** 0.208*** 0.205*** 0.135*** 0.127*** 
  Livingstonia (100 km) (0.013) (0.015) (0.013) (0.024) (0.339) (0.431) (0.341) (0.041) (0.050) (0.041) (0.037) 
 No. of obs. 24,681 22,124 24,512 24,681 24,691 22,474 24,691 24,634 22,435 24,634 24,634 
Panel E: Only for Lomwe, Nyanja and Sena 
 Distance to −0.314* −0.286 −0.237 −0.877** −11.480*** −10.199*** −9.994*** 1.862** 1.368* 1.516* 1.209** 
  Livingstonia (100 km) (0.159) (0.162) (0.157) (0.423) (3.482) (3.468) (3.433) (0.835) (0.781) (0.799) (0.487) 
 No. of obs. 10,676 10,403 10,676 10,781 10,782 10,504 10,782 10,746 10,468 10,746 10,757 
 Individual controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Matrilineal com. dummy Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 GPS coordinate Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Ethnicity FE No No Yes Yes No No Yes No No Yes Yes 
 Com.-language FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Geography and climate Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Historical controls No Yes No No No Yes No No Yes No No 
 District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Round FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Dependent variables One if Christian Education (years) One if unable to read 
Sample All All All 
Probit (ME) Probit (ME) Probit (ME) LPM OLS OLS OLS Probit (ME) Probit (ME) Probit (ME) LPM 
(a) (b) (c) (d) (e) (f) (g) (h) (i) (j) (k) 
 Distance to −0.044*** −0.060*** −0.036** −0.082*** −1.546*** −1.318*** −0.993*** 0.207*** 0.177*** 0.138*** 0.123*** 
  Livingstonia (100 km) (0.015) (0.017) (0.015) (0.021) (0.330) (0.411) (0.333) (0.040) (0.047) (0.040) (0.036) 
 Distance to the nearest 0.010 0.010 0.010 −0.015 −2.673*** −2.614*** −2.649*** 0.259*** 0.244*** 0.257*** 0.235*** 
  mission station (100 km) (0.013) (0.013) (0.012) (0.024) (0.351) (0.358) (0.343) (0.039) (0.039) (0.039) (0.036) 
Oster (forthcoming)'s δ            
  Rmax=1.3R˜    −1.416 −0.879 −1.180 −2.079    −1.187 
  Rmax = 1.0    −0.117 −0.183 −0.233 −0.351    −0.113 
R2 0.209 0.219 0.220 0.127 0.246 0.249 0.258 0.103 0.105 0.108 0.131 
 No. of obs. 41,118 38,551 40,949 41,118 41,129 38,634 41,129 41,033 38,557 41,033 41,033 
Panel A: With a control of the distance to Blantyre (100 km) 
 Distance to −0.048*** −0.065*** −0.041*** −0.087*** −1.268*** −0.998** −0.734** 0.172*** 0.138*** 0.106*** 0.093** 
  Livingstonia (100 km) (0.015) (0.018) (0.015) (0.022) (0.342) (0.428) (0.342) (0.041) (0.049) (0.041) (0.037) 
 Distance to the nearest 0.007 0.007 0.006 −0.017 −2.552*** −2.509*** −2.537*** 0.243*** 0.232*** 0.244*** 0.222*** 
  mission station (100 km) (0.013) (0.013) (0.012) (0.023) (0.346) (0.354) (0.339) (0.038) (0.034) (0.038) (0.035) 
 Distance to 0.010 0.011 0.013 0.014 −0.857*** −0.829*** −0.785*** 0.108*** 0.100*** 0.096*** 0.091*** 
  Blantyre (100 km) (0.008) (0.008) (0.008) (0.013) (0.243) (0.248) (0.244) (0.028) (0.029) (0.028) (0.027) 
 No. of obs. 41,118 38,551 40,949 41,118 41,129 38,634 41,129 41,033 38,557 41,033 41,033 
Panel B: With no control of other missions 
 Distance to −0.043*** −0.059*** −0.035** −0.083*** −1.663*** −1.458*** −1.134*** 0.215*** 0.186*** 0.149*** 0.136*** 
  Livingstonia (100 km) (0.015) (0.017) (0.015) (0.021) (0.339) (0.421) (0.341) (0.042) (0.048) (0.042) (0.038) 
 No. of obs. 41,118 38,551 40,949 41,118 41,129 38,634 41,129 41,033 38,557 41,033 41,033 
Panel C: Only for Lomwe, Ngoni, Nyanja and Sena 
 Distance to −0.296*** −0.298*** −0.270** −0.344*** −4.705*** −4.004** −4.183** 1.108*** 0.966*** 1.024*** 0.634*** 
  Livingstonia (100 km) (0.124) (0.683) (0.120) (0.113) (1.804) (1.757) (1.760) (0.305) (0.306) (0.302) (0.205) 
 No. of obs. 15,990 15,714 15,990 16,437 16,438 16,160 16,438 16,393 16,116 16,393 16,399 
Panel D: Non-Yao excluding Lomwe, Ngoni, Nyanja and Sena 
 Distance to −0.038*** −0.054*** −0.034** −0.100*** −1.539*** −1.455*** −0.924*** 0.208*** 0.205*** 0.135*** 0.127*** 
  Livingstonia (100 km) (0.013) (0.015) (0.013) (0.024) (0.339) (0.431) (0.341) (0.041) (0.050) (0.041) (0.037) 
 No. of obs. 24,681 22,124 24,512 24,681 24,691 22,474 24,691 24,634 22,435 24,634 24,634 
Panel E: Only for Lomwe, Nyanja and Sena 
 Distance to −0.314* −0.286 −0.237 −0.877** −11.480*** −10.199*** −9.994*** 1.862** 1.368* 1.516* 1.209** 
  Livingstonia (100 km) (0.159) (0.162) (0.157) (0.423) (3.482) (3.468) (3.433) (0.835) (0.781) (0.799) (0.487) 
 No. of obs. 10,676 10,403 10,676 10,781 10,782 10,504 10,782 10,746 10,468 10,746 10,757 
 Individual controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Matrilineal com. dummy Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 GPS coordinate Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Ethnicity FE No No Yes Yes No No Yes No No Yes Yes 
 Com.-language FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Geography and climate Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Historical controls No Yes No No No Yes No No Yes No No 
 District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Round FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 

Notes: (1) Figures () are standard errors. ***Significance at 1%, **significance at 5% and *significance at 10%. (2) Standard errors are robust to heteroskedasticity and clustered residuals within each community (3) The individual controls include age (years), birth order, no. of alive siblings at age 10 and no. of late siblings at age 10. (4) The ethnicity is classified into 12 groups, i.e., Chewa, Lambya, Lomwe, Mang'anja, Ndali, Ngoni, Nkhonde, Nyanja, Sena, Tonga, Tumbuka and other. (5) The community language is classified into 14 groups, i.e., Chewa, Lambya, Lomwe, Ngoni, Nkhonde, Nyakyusa, Nyanja, Sena, Senga, Sukwa, Tonga, Tumbuka, Yao and other. (6) The geographic and climate controls contain community-level information on climatology, landscape typology, soil and terrain, crop season parameters. See Appendix B for the details. (7) The ethnicity-level historical controls include (i) a dummy variable, equal to one if a European explorer travelled through land historically inhabited by an ethnic group; (ii) a dummy variable, equal to one if any part of railway lines in the first decade of the 20th century drawn from Century Company (1911) passed through land historically inhabited by an ethnic group and (iii) the total number of slaves taken from an ethnic group that was normalised by the area of land inhabited by the ethnic group during the 19th century (log of one plus the normalised slave export measure).

Discussion on mission influence channels

Thus far, the total missionary effects on female marriage practices, which are assumed to result from the dissemination of Christian values and the missionary investment in local public goods (e.g., education), have been identified. In this section, the potential channels for the missionary effects are discussed, as the relevant public policy implications differ depending on the underling mechanisms.

Christianity and education

Before proceeding with the discussion, it is worth exploring non-Yao respondents’ religious conversion and educational attainment, as prompted by the Livingstonia Mission, based on specification (1).30 The relevant estimation results are reported in Table 6. This exercise was performed because the conversion to Christianity primarily facilitates acquisition of Christian values by the local inhabitants, whereas the Livingstonia Mission exerted additional influence by greatly investing in promotion of education. As outlined in the detailed historical account of the educational influence of the Livingstonia Mission provided in Appendix A.3, the educational advancement facilitated by this mission triggered the subsequent economic, political and social transformations in Malawi.

Given that communities located further from Livingstonia are assumed to have less exposure to the missionary venture, their members are less likely to be Christian. This relationship is confirmed by the results reported in columns (a)–(d) in Table 6.

Next, the results reported in columns (e)–(k) highlight the educational impacts of the Livingstonia Mission. They reveal the presence of a significant, negative correlation between distance to Livingstonia and educational attainment both in terms of formal schooling (i.e., years of the highest level of education completed) and reading skills.31 This finding suggests that, due to limited exposure to the missionary services, non-Yao females living further away from Livingstonia are less likely to attain academic skills than their non-Yao counterparts residing in closer proximity to Livingstonia. The educational effects are consistent with the evidence provided in the previous relevant historical and economic research (e.g., Gallego and Woodberry, 2010; Nunn, 2014).

Moreover, the estimation results reported in Tables 3, 4 and 6 suggest that an MDHS community's distance to the nearest mission station had a significant influence on both the marriage practices of interest and educational attainment among the non-Yao population, but not on the likelihood of their conversion to Christianity. Thus, it is possible that other missions’ effects on local marriage practices may primarily be driven by their educational promotion, which may also be true of the Livingstonia Mission.

Comparison with influence on the Yao

However, the findings reported in Table 7 may mitigate the view of ‘only education matters’. In this table, the exercises performed in Tables 3, 4 and 6 were repeated for the Yao ethnic group and the estimation results are presented.32

Table 7:

Missionary Influence on the Yao Ethnic Group

Dependent variables One if Christian One if unable to read Education (years) Age at first marriage One if married Years to first marriage 
Sample All All All All ever married Aged below 18 Aged below 18 
Probit (ME) LPM Probit (ME) LPM OLS OLS Probit
(ME) 
LPM Hazard
ratio 
(a) (b) (c) (d) (e) (f) (g) (h) (i) 
 Distance to 0.520 0.517 1.286*** 0.952*** −1.340 0.041 −0.267 0.724 0.271 
  Livingstonia (100 km) (0.392) (0.318) (0.454) (0.283) (2.161) (1.805) (1.310) (0.815) (1.858) 
 Distance to the nearest −0.030 −0.030 0.353*** 0.302*** −2.941*** −0.046 0.090 −0.041 0.806 
  mission station (100 km) (0.093) (0.083) (0.097) (0.080) (0.645) (0.559) (0.154) (0.148) (0.741) 
 Shoenfeld res. (p-val.)         0.745 
R2 0.238 0.265 0.169 0.212 0.309 0.039 0.142 0.146  
 No. of obs. 6,079 6,104 6,083 6,089 6,104 5,166 682 769 769 
Dependent variables One if polygyny One if polygyny (zero if unmarried)      
Sample All married All      
Probit
(ME) 
LPM Probit
(ME) 
LPM  
(j) (k) (l) (m)      
 Distance to 0.147 0.135 0.103 0.134      
  Livingstonia (100 km) (0.533) (0.241) (0.364) (0.223)      
 Distance to the nearest 0.173** 0.182** 0.127*** 0.151***      
  mission station (100 km) (0.074) (0.072) (0.051) (0.053)      
R2 0.106 0.102 0.122 0.095      
 No. of obs. 4,248 4,312 6,001 6,104      
 Individual controls Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Matrilineal com. dummy Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 GPS coordinate Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Ethnicity FE No No No No No No No No No 
 Com.-language FE Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Geography and climate Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Historical controls No No No No No No No No No 
 District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Round FE Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Dependent variables One if Christian One if unable to read Education (years) Age at first marriage One if married Years to first marriage 
Sample All All All All ever married Aged below 18 Aged below 18 
Probit (ME) LPM Probit (ME) LPM OLS OLS Probit
(ME) 
LPM Hazard
ratio 
(a) (b) (c) (d) (e) (f) (g) (h) (i) 
 Distance to 0.520 0.517 1.286*** 0.952*** −1.340 0.041 −0.267 0.724 0.271 
  Livingstonia (100 km) (0.392) (0.318) (0.454) (0.283) (2.161) (1.805) (1.310) (0.815) (1.858) 
 Distance to the nearest −0.030 −0.030 0.353*** 0.302*** −2.941*** −0.046 0.090 −0.041 0.806 
  mission station (100 km) (0.093) (0.083) (0.097) (0.080) (0.645) (0.559) (0.154) (0.148) (0.741) 
 Shoenfeld res. (p-val.)         0.745 
R2 0.238 0.265 0.169 0.212 0.309 0.039 0.142 0.146  
 No. of obs. 6,079 6,104 6,083 6,089 6,104 5,166 682 769 769 
Dependent variables One if polygyny One if polygyny (zero if unmarried)      
Sample All married All      
Probit
(ME) 
LPM Probit
(ME) 
LPM  
(j) (k) (l) (m)      
 Distance to 0.147 0.135 0.103 0.134      
  Livingstonia (100 km) (0.533) (0.241) (0.364) (0.223)      
 Distance to the nearest 0.173** 0.182** 0.127*** 0.151***      
  mission station (100 km) (0.074) (0.072) (0.051) (0.053)      
R2 0.106 0.102 0.122 0.095      
 No. of obs. 4,248 4,312 6,001 6,104      
 Individual controls Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Matrilineal com. dummy Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 GPS coordinate Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Ethnicity FE No No No No No No No No No 
 Com.-language FE Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Geography and climate Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Historical controls No No No No No No No No No 
 District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Round FE Yes Yes Yes Yes Yes Yes Yes Yes Yes 

(1) Figures () are standard errors. ***Significance at 1%, **significance at 5% and *significance at 10%. (2) Standard errors are robust to heteroskedasticity and clustered residuals within each community. (3) The individual controls include age (years), birth order, no. of alive siblings at age 10 and no. of late siblings at age 10. (4) The community-language is classified into 14 groups, i.e., Chewa, Lambya, Lomwe, Ngoni, Nkhonde, Nyakyusa, Nyanja, Sena, Senga, Sukwa, Tonga, Tumbuka, Yao and other. (4) The geographic and climate controls contain community-level information on climatology, landscape typology, soil and terrain, crop season parameters. See Appendix B for the details.

These estimations were conducted because comparing the missionary influence on the Yao with the influence on non-Yao respondents may help identify whether Christian values or the missionary investment in education exerted greater influence on the latter group's engagement in early marriage and polygyny. While Christianity may have failed to instill its cultural norms and beliefs among the Yao that were under considerable Islamic influence, all ethnic groups potentially benefited from the missionary investment in education and the resulting economic development to a similar extent. Given this assertion, if the analyses of Yao respondents’ data provide similar findings to those obtained when the data of non-Yao groups is analysed, the results may imply that missionary investment had a more pronounced effect in triggering the marriage effects on the non-Yao population. Conversely, if the results are different for the Yao and non-Yao groups, this may mean that Christianity played a greater role in the subsequent marriage effects on the latter group.

Table 7 reports two interesting findings. First, the results presented in columns (c)–(e) indicate that a community's distance to Livingstonia was significantly and negatively correlated with Yao's academic skills, as measured by years of formal schooling and reading proficiency. This effect is similar to the influence identified for non-Yao respondents and reported in Table 6. Thus, this suggests that the influence of educational promotion at the Livingstonia Mission is captured by distance to some extent. On the other hand, distance had no significant association with Yao's Christian identity and marital practices, as indicated by the results reported in the remaining columns in Table 7. This is in sharp contrast with the estimation results for non-Yao respondents.

Admittedly, the variation in the community distance to Livingstonia is less pronounced for the Yao ethnic group relative to the non-Yao population, because the Yao primarily reside at the southern end of Lake Malawi (see Figure 2). Nevertheless, these findings may still suggest that the missionary investment, at least in education, cannot alone explain the marriage effects on the non-Yao population, while underscoring the role of Christian values as one of the facilitators of the impacts.

In fact, in a short questionnaire-based survey, the author conducted in rural Malawi specifically for the purpose of this study, approximately 90% of the respondents, whether Christian or Muslim, acknowledged that, unlike Islam, Christianity prohibits polygyny.33 While a few respondents shared the view that Christian teachings explicitly discouraged early marriage, it is also possible that Christian teachings could prevent this practice indirectly. This conclusion is reached because religious authorities often preach about the negative health consequences of early childbirth and sexual intercourse, which may possibly result in/from early marriage. In short, when these findings are considered alongside those presented in Section 6.1, it may be surmised that a combination of Christian values and the missionary investment in education plays a role in explaining the marriage effects of the Livingstonia Mission.

More generally, the Christian missionary investment in public goods and its influence on marriage practices is highlighted in the extant literature. Examining several African countries, Fenske (2015) provided evidence suggesting the powerful influence of colonial and missionary education on present polygynous practices. Consistent with this view, other early missions reduced the incidence of polygyny among the Yao while facilitating their educational attainment. This link is supported by the coefficients pertaining to an MDHS community's distance to the nearest mission station provided in columns (j)–(m) and columns (c)–(e) in Table 7, respectively. However, given that the distance is uncorrelated with the probability of Yao females entering into marriage at a very young age, educational promotion alone cannot explain how these early missions discouraged non-Yao females from practicing early marriage, as revealed in Table 3. Thus, the pathways responsible for the impacts of these early missions on the practice of early marriage potentially differ from the mechanisms in which these missions discouraged polygyny.

Related to this point, as the results in Tables 3, 4 and 6 show, the Livingstonia Mission influenced the likelihood of non-Yao females converting to Christianity, engaging in early marriage and polygyny, and attaining academic skills. Other missions also affected these outcomes, albeit without facilitating religious conversion. Consequently, the difference between the Livingstonia Mission and other missions may underscore the historical significance of the Livingstonia Mission, which introduced Christianity into Malawi. This may also suggest that missionary activities were heterogeneous in those days. In fact, in contrast with the Blantyre Mission, which acted largely within a small network of the neighbouring villages based on its traditional residential policy (McCracken 1977, p. 69), the Livingstonia Mission took an exploratory approach. Its members actively visited the surrounding villages, as described in Section 2, for example. Consequently, it may also be possible that the underlying forces driving the marriage effects differ from one mission to another. However, further empirical research is required to precisely identify the channels involved.34

Conclusion

By focusing on one Protestant mission (Livingstonia Mission) dating back to the late 19th century in Malawi, this study investigated the impacts of Christian missionaries on female marriage practices.

To measure the influence of the mission, analyses exploited a community's distance to the influential station, Livingstonia, and estimated the impacts of the distance on non-Yao ethnic groups. Two assumptions (based on historical accounts) encouraged this decision. First, in Malawi, Christianity expanded from the northern areas. Second, the Yao largely converted to Islam because of their alliance with the Arabs based on ivory and slave trades that predated the arrival of the Christian mission.

Using the distance to Livingstonia as a proxy for the missionary influence, along with numerous historical, geographic and climate controls, this study revealed that the Livingstonia Mission encouraged non-Yao females to postpone their first marriage, while discouraging their engagement in polygynous relationships. While it is difficult to completely exclude potential endogeneity concerns, several exercises performed as a part of this work also failed to establish that the estimated impacts were entirely attributed to the estimation bias. In addition to these main findings, non-Yao females residing closer to Livingstonia are also more likely to be Christian as well as attain academic skills than the other non-Yao females located further away from the mission.

However, it is still unclear whether Christian values or the missionary investment in local public goods such as education exerted greater influence on the identified marriage effects. By comparing the welfare consequences of missionaries’ activities between the Christian and non-Christian populations (e.g., non-Yao and Yao ethnic groups in the present context), it may be possible to separate these two forces. This view is based on the premise that the missionary investment affects all local people equally, whereas the religious doctrine primarily influences the adherents. The discussion and use of this testing approach is another contribution of the present paper. Application of this strategy revealed that the identified effects have likely stemmed from marriage-related norms and knowledge mediated by Christianity, as well as from the educational investment and promotion facilitated by the Livingstonia Mission. Nevertheless, empirical findings supporting this view are only suggestive. Therefore, further research would be useful to validate the significance of these and possibly other factors.

Through the in-depth analysis of a single but well-known mission in Malawi (and Central Africa) that was founded before European institutions and other early missions became influential, this study aimed to increase the internal validity of the findings. On the one hand, due to the pioneering nature of the Livingstonia Mission, local inhabitants may not have felt inclined to completely submit to the views and practices of the mission, particularly in the early periods of its activities. On the other hand, for the same reason, this mission may be perceived as offering greater marginal returns arising from religious conversion of the local population and its missionary investment in public goods compared with other subsequent missions. The external validity of the findings may depend upon establishing which of these two conflicting forces had a greater influence.

1
Field and Ambrus (2008) revealed that delaying marriage significantly improved female schooling in Bangladesh. In addition, Tertilt (2005) quantitatively demonstrated that legal prohibition of polygyny resulted in a marked increase in savings and a decrease in fertility in sub-Saharan Africa, which is a location historically characterised by a high incidence of polygyny. Similar implications have been identified from transferring the right to choose a husband from fathers to daughters (Tertilt, 2006). In the same vein, Edlund and Lagerlöf (2012) argued, from a theoretical perspective, that monogamy boosted human capital investment by encouraging young married males to dedicate more of their time to educating their children, rather than pursuing leisure as they did during bachelorhood.
2
Based on Becker (1981)’s theory of marriage, polygynous marriage emerges even when the number of males and females is equal. This phenomenon arises because ‘superior’ males endowed with resources complementary to women's marginal contribution to the marital output (e.g., land-rich males) can oust ‘inferior’ males (e.g., resource-poor peasants) from the marriage market. Accordingly, several sources of male inequality, such as income and the number of sisters (Bergstrom, 1994), as well as differences in technological efficiency of human capital creation between young and old generations (Edlund and Lagerlöf, 2012), have been analysed in available theoretical studies.
3
It should also be noted that, during the last decade, significant political effort has been expended to ensure that religious forces and institutions are given a prominent place on the development agenda (Haar and Ellis, 2006). For instance, UNDP (2004) views religion as a source of cultural diversity and sees this cultural liberty as a vital aspect of human development. The UK Department for International Development also financed a series of multi-million-pound comparative research projects conducted by the Religions and Development Research Programme Consortium between 2005 and 2010 (see http://r4d.dfid.gov.uk/Project/3896/ and http://www.religionsanddevelopment.org/index.php?section=10#mod_58 for the details).
4
Lovedale was a mission station and educational institute established in Cape Province, South Africa.
5
Until the early 20th century, the mission primarily relied on a small group of philanthropic industrialists for financial support, most of whom were operationally based in Glasgow, rather than relying on Free Church official funds. The Glasgow businessmen discerned the economic potential of Lake Malawi.
6
Such a change in direction partly stemmed from a proposal made by James Stevenson, one of the directors of the African Lakes Company (ALC; a trading body formed by the Glasgow industrialists mentioned in footnote 5). Stevenson proposed building a road between Lake Malawi and Lake Tanganyika, which would enable the ALC to distribute commercial products to a wide inland area while allowing them to work in close cooperation with the missionaries.
7
In its long history, some debate existed among the authorities of the Livingstonia Mission about the levels and types of educational services to provide to the local population. In contrast with Laws, who prioritised the promotion of post-primary education, for example, Donald Fraser (1870–1933) highlighted the importance of providing practical training adjusted to local conditions.
8
In the present study, the terms ‘enumeration areas’ and ‘communities’ are used interchangeably.
9
In the survey, all females aged 15–49 residing in the selected households and all males aged 15–54 in one-third (one-fourth in 2000) of the selected households were eligible for the interviews.
10
Similar sampling protocols were adopted for all the surveys. For example, in the 2010 MDHS, households were selected in two stages. First, by separating the 27 study domains (districts) into urban and rural areas, the nation was partitioned into 54 sampling units consisting of the 9,144 enumeration areas established in the 2008 Malawi Population and Housing Census (PHC). The selection of 849 clusters (158 urban and 691 rural) from these enumeration areas was made in the first stage. In the second stage, 20 and 35 households were selected from each urban and rural cluster, respectively, resulting in the target sample size of 27,345 households at the national level. See Malawi DHS Final Report (2000, 2004 and 2010) for a detailed description of the sampling framework.
11
The proportions given in Table 1 are unweighted. Although calculating the true proportion of the entire population from the sample data requires appropriate sample weights, it is expected that the weighting would not affect the overall picture significantly. For example, based on the 2010 estimate provided by Pew Forum on Religion and Public Life (http://features.pewforum.org/global-christianity/total-population-percentage.php), approximately 82.7% of the total population in Malawi was Christian, which closely corresponds to the unweighted proportion observed in the MDHS data.
12
In 1911, the Livingstonia and Blantyre Synods agreed to join together to form the CCAP.
13
It is expected that family plays a pivotal role in characterising or transmitting cultural values from parents to children (e.g., Dohmen et al., 2011; Farré and Vella, 2013), although several recently conducted empirical studies failed to yield evidence supporting the intergenerational transmission of cultural values (e.g., Cipriani et al., 2013).
14
Or, more simply, a daughter's earning potential increases the opportunity cost of her early marriage. Moreover, greater life expectancy may further encourage (resp., discourage) female schooling (early marriage), because the benefits of schooling (and, once again, reservation utility) increase due to extended longevity (e.g., Soares, 2006).
15
While the results are not reported due to space restrictions, Equation (1) was estimated separately using single-round data and the results yielded implications similar to those provided by the estimations based on the data from the three DHS rounds of 2000, 2004 and 2010.
16
For example, see the relationship of the mission with the lakeside Tonga and the northern Ngoni in the early periods of the Livingstonia Mission (McCracken 1977, pp. 73–99).
17
However, the author could not find the reference period of Pike and Rimmington (1965)'s map.
18
For details on the sampling design, see Third Integrated Household Survey (IHS3) 20102011 Basic Information Document, March 2012 at http://siteresources.worldbank.org/INTLSMS/Resources/3358986-1233781970982/5800988-1271185595871/IHS3.BID.FINAL.pdf.
19
As a part of this investigation, an MDHS community's distances to the nearest point of the railway lines and explorer routes were also calculated, based on the maps provided by Nunn and Wantchekon (2011). Including these community-level historical controls in regressors did not alter the implications of the estimation results reported in this paper. The estimation results controlling for these distances are available upon request.
20
The following ethnic groups were identified in both the MDHS and Nunn and Wantchekon (2011) data sets: the Chewa, Lomwe, Ngoni, Lambya, Manga'nja, Nkhonde, Sena, Tonga, Tumbuka and Yao.
21
Note that not all respondents who married before the age of 18 years entered into illegal marital relationships because the sample in the respective round of the MDHS includes females who are currently between the ages of 15 and 49 years and some respondents entered into marriage prior to the enforcement of this law.
22
The estimates reported in Table 3 correspond to m = 18.
23
Married females include both those currently in a marital union (92%) and those living with a partner (8%).
24
A command developed in Lokshin (2006) was used in these estimations.
25
Note that, while the statistical significance is weak in general, the effect is nonetheless important, as seen from the 10% and 11% levels in columns (h) and (j) in panel (A) of Table 3, for example.
26
However, it was not possible to obtain estimation results for the hazard model due to a convergence failure.
27
It is possible that population density in the late 19th century affected the mission's decision to establish the station such as Livingstonia. The exercise in Table 5 is performed solely as a means to examine the correlation between the distance to Livingstonia and the population density (conditional on some covariates), rather than to report causal impacts of the distance.
28
While it is difficult to expect particular relationships between the pre-missionary population size and location of Christian missionaries to be consistent across Africa (Johnson 1967, p. 171), the absence of a statistically significant relationship between the pre-missionary population density and the distance to Livingstonia may still be surprising. However, the analyses discussed in this section revealed that the correlation was absent when controlling for a cell's or district's positional information (i.e., latitude, longitude). Although the results are not reported here for brevity, similar exercises to those in Table 5 were conducted without using positional information as regressors, which revealed a statistically significant association between population density and the distance to Livingstonia. Some of these results indicate that Livingstonia was located in a less densely populated area in the late 19th or early 20th century. This finding is plausible, given that the Yao, who primarily settled in the southern region of Malawi, enjoyed economic prosperity before the arrival of Christian missionaries, whereas Livingstonia is located in the north.
29
As inferred from the results reported in columns (j)–(m) in Table A1, the coefficients related to the distance to Cape Maclear reveal that the influence of this location appears to have discouraged polygyny. However, the distance to Cape Maclear was not correlated with either the practices of early marriage or the attainment of academic skills. The estimation results reported in columns (a) and (b) also indicate that the estimated influence on religious conversion is counter-intuitive. Hence, the estimates on the distance to Cape Maclear do not yield a meaningful interpretation of its influence.
30
Nunn (2014) explored the missionary influence on educational promotion in sub-Saharan Africa. Compared with Catholic missionaries, Nunn (2014) found that Protestant missionaries placed a greater emphasis on female schooling, and thus had greater long-term positive effects on the educational attainment of females relative to males.
31
See also Figure A2 (lower panels) for the estimated non-parametric function of the missionary influence on Christianity and academic skills based on the difference-based semi-parametric estimation of a partial lineal model by Yatchew (1997, 1998). In contrast with the remaining outcomes reported in this figure, the relationship between Christianity and the distance to Livingstonia is somewhat non-monotonic. However, non-Yao females living more than 650 km away from Livingstonia constitute only approximately 10% of the entire non-Yao sample. Within the 650 km radius from Livingstonia, the distance to the mission station is clearly negatively correlated with Christian identity. In addition, as only a small proportion of non-Yao females were identified as non-Christian in the exploited data set (see Table 1), this data issue might have rendered the semi-parametric estimation of religions conversion rather sensitive to the specification. While the estimation results are not reported, exploiting the same specifications and samples as those in columns (a) and (b) in Table 6 and estimating the non-parametric function of the missionary influence revealed a monotonically negative relationship between the distance to Livingstonia and Christianity.
32
In this analysis exploiting data on Yao respondents only, neither ethnic-level historical controls nor ethnicity fixed effects were included in the regressors. This treatment was chosen to avoid a multicollinearity problem.
33
To examine people's perceptions of marital and inheritance practices, as well as the relationship between these practices and religious beliefs, the author conducted a short questionnaire-based survey in three districts (Machinga, Mulanje and Zomba) in southern Malawi in 2013. After obtaining the village list from the respective district councils, the author randomly selected at least one village from each district, resulting in five villages surveyed. In each village, between two and five residents took part in individual interviews lasting approximately 30–60 min. To ensure confidentiality and increase data reliability, during the interviews, the respondent was alone with the author and the research assistant (responsible for translation to and from Chewa). Since the interviewed respondents were not randomly selected due to limited resources (i.e., they were identified through convenience sampling), it is difficult to generalise the findings based on their responses. In total, the survey sample comprised 8 male and 12 female adult respondents belonging to 4 ethnic groups (the Lomwe, Ngoni, Nyanja and Yao), of whom 11 were Muslim and 9 were Christian.
34
Utilising data pertaining to both Yao and non-Yao respondents, the following difference-in-differences (DID) model was also estimated based on the ordinary-least squares (OLS) technique:
yij = β1 + β2djeij + β3eij + β4xij + vj + εij,
where eij is a dummy variable for non-Yao ethnic groups. Here, all time-invariant unobserved factors that are specific to each community and affect the outcomes are measured by the community-level fixed effects, vj. Overall, the estimation results reported in Table A2 indicate that the implications of the estimated β2 (i.e., impacts on non-Yao ethnic groups) were similar to those yielded by the findings reported in Tables 3, 4 and 6.
35
To Livingstone, it seemed that the region had already laid the foundations of a successful cash-crop economy, as cotton of good quality was cultivated in many villages in the region.
36
In contrast with Livingstone's view described in footnote 35, when Stewart arrived at the highlands, he found that no cotton was produced in the region, and that slow and primitive methods of spinning were exploitive. Nevertheless, even after the recall of the expedition ordered by the British government, Livingstone still insisted on the importance and practicability of introducing small colonies into the region, contending that cotton production was not in full swing at the moment of Stewart's visit.
37
Concurrently, the Universities’ Mission to Central Africa (UMCA), whose establishment was inspired by Livingstone's speeches at Cambridge and Oxford in 1857, built the station at Magomero between modern Zomba and Blantyre in 1861 and made the decision to move its work centre to Zanzibar in 1862. UMCA (1857–1965) was a missionary society established by members of the Anglican Church from the universities of Cambridge, Dublin, Durham and Oxford.
38
In the literature, many other reasons are also proposed for the spread of Islam in Malawi. For instance, Islam was simply considered fashionable among the Yao in the late 19th century (Pike 1968, p. 69). In addition, people were sometimes eager to become Muslims because they viewed conversion as a means of increasing their incomes (Msiska 1995, p. 61). This is because once they became Muslim teachers, they could typically collect fairly substantial fees from their disciples. Those disciples also had to serve their teachers until they left as full-fledged Muslims. Moreover, the colonial administration also indirectly contributed to the spread of Islam because government officials sometimes preferred Muslims and helped them build mosques, and for the period from 1888 to 1889, the Nyasaland Government banned Christian missionaries from working in Muslim areas, for example (Msiska 1995, pp. 63–64).
39
Note that this statement is not intended to imply that other missions had no influence on educational promotion (Kalinga, 2012). However, Livingstonia's preeminent status was partly the result of limited financial support that many other independent churches received from international bodies (McCracken 1977, p. 284). As noted in footnote 5, the Livingstonia Mission greatly benefitted from such support, in particular from financial contributions from philanthropic Glasgow businessmen.

References

Acemoglu
D.
,
Johnson
S.
,
Robinson
J. A.
(
2002
) ‘
Reversal of Fortune: Geography and Institutions in the Making of the Modern World Income Distribution
’,
Quarterly Journal of Economics
 ,
117
(
4
):
1231
1294
.
Alesina
A.
,
Giuliano
P.
(
2015
) ‘
Culture and Institutions
’,
Journal of Economic Literature
 ,
53
(
4
):
898
944
.
Altonji
J. G.
,
Elder
T. E.
,
Taber
C. R.
(
2005
) ‘
Selection on Observed and Unobserved Variables: Assessing the Effectiveness of Catholic Schools
’,
Journal of Political Economy
 ,
113
(
1
):
151
184
.
Arruñada
B.
(
2010
) ‘
Protestants and Catholics: Similar Work Ethic, Different Social Ethic
’,
Economic Journal
 ,
120
(
547
):
890
918
.
Bai
Y.
,
Kung
J. K.
(
2015
) ‘
Diffusing Knowledge While Spreading God's Message: Protestantism and Economic Prosperity in China, 1840–1920
’,
Journal of the European Economic Association
 ,
13
(
4
):
669
698
.
Becker
G. S.
(
1981
)
A Treatise on the Family
 .
Cambridge, MA
:
Harvard University Press
, Enlarged Edition.
Becker
S. O.
,
Woessmann
L.
(
2009
) ‘
Was Weber Wrong? A Human Capital Theory of Protestant Economic History
’,
Quarterly Journal of Economics
 ,
124
(
2
):
531
596
.
Bergstrom
T. C.
(
1994
)
On the Economics of Polygyny
 .
University of Michigan, mimeo
.
Bone
,
D. S.
2000
. Malawi's Muslims: Historical Perspectives. Christian Literature Association in Malawi (CLAIM), Blantyre, Malawi.
Bove
R.
,
Valeggia
C.
(
2009
) ‘
Polygyny and Women's Health in sub-Saharan Africa
’,
Social Science & Medicine
 ,
68
(
1
):
21
29
.
Cagè
J.
,
Rueda
V.
(
2016
) ‘
The Long-term Effects of the Printing Press in sub-Saharan Africa
’,
American Economic Journal: Applied Economics
 ,
8
(
3
):
69
99
.
Cantoni
D.
(
2015
) ‘
The Economic Effects of the Protestant Eeformation: Testing the Weber Hypothesis in the German Lands
’,
Journal of the European Economic Association
 ,
13
(
4
):
561
598
.
Century Company.
1911
. The Century Atlas: Africa [Map].
Cipriani
M.
,
Giuliano
P.
,
Jeanne
O.
(
2013
) ‘
Like Mother Like Son? Experimental Evidence on the Transmission of Values from Parents to Children
’,
Journal of Economic Behavior & Organization
 ,
90
:
100
111
.
Commission for Africa.
2005
. Our Common Interest: Report of the Commission for Africa. http://www.commissionforafrica.info/2005-report.
Dalton
J. T.
,
Leung
T. C.
(
2014
) ‘
Why is Polygyny More Prevalent in Western Africa? An African Slave Trade Perspective
’,
Economic Development and Cultural Change
 ,
62
(
4
):
599
632
.
Dohmen
T.
,
Falk
A.
,
Huffman
D.
(
2011
) ‘
The Intergenerational Transmission of Risk and Trust Attitudes
’,
Review of Economic Studies
 ,
79
(
2
):
645
677
.
Duflo
E.
,
Dupas
P.
,
Kremer
M.
(
2015
) ‘
Education, HIV, and Early Fertility: Experimental Evidence from Kenya
’,
American Economic Review
 ,
105
(
9
):
2757
2797
.
Edlund
,
L.
,
Lagerlöf
,
N.-P.
2012
. Polygyny and its Discontents: Parental Age and Human Capital Accumulation. Discussion Paper No. 1213-11, Department of Economics, Columbia University.
Ermisch
J. F.
(
2003
)
An Economic Analysis of the Family
 .
Princeton, New Jersey
:
Princeton University Press
.
Farré
L.
,
Vella
F.
(
2013
) ‘
The Intergenerational Transmission of Gender Role Attitudes and its Implications for Female Labor Force Participation
’,
Economica
 ,
80
(
318
):
219
247
.
Fenske
J.
(
2013
) ‘
Does Land Abundance Explain African Institutions?
’,
Economic Journal
 ,
123
(
573
):
1363
1390
.
Fenske
J.
(
2015
) ‘
African Polygamy: Past and Present
’,
Journal of Development Economics
 ,
117
:
58
73
.
Field
E.
,
Ambrus
A.
(
2008
) ‘
Early Marriage, Age of Menarche, and Female Schooling Attainment in Bangladesh
’,
Journal of Political Economy
 ,
116
(
5
):
881
930
.
Gallego
F. A.
,
Woodberry
R.
(
2010
) ‘
Christian Missionaries and Education in Former African Colonies: How Competition Mattered
’,
Journal of African Economies
 ,
19
(
3
):
294
329
.
Goldewijk
K. K.
,
Beusen
A.
,
Janssen
P.
(
2010
) ‘
Long-term Dynamic Modeling of Global Population and Build-up Area in a Spatially Explicit Way: HYDE 3.1
’,
The Holocene
 ,
20
(
4
):
565
573
.
Gould
E. D.
,
Moav
O.
,
Simhon
A.
(
2008
) ‘
The Mystery of Monogamy
’,
American Economic Review
 ,
98
(
1
):
333
357
.
Grossbard-Shechtman
A.
(
1986
) ‘
Economic Behavior, Marriage and Fertility: Two Lessons from Polygyny
’,
Journal of Economic Behavior & Organization
 ,
7
(
4
):
415
424
.
Haar
G. T.
,
Ellis
S.
(
2006
) ‘
The Role of Religion in Development: Towards a New Relationship between the European Union and Africa
’,
European Journal of Development Research
 ,
18
(
3
):
351
367
.
Jacoby
H. G.
(
1995
) ‘
The Economics of Polygyny in sub-Saharan Africa: Female Productivity and the Demand for Wives in Côte d'ivoire
’,
Journal of Political Economy
 ,
103
(
5
):
938
971
.
Jensen
R.
(
2012
) ‘
Do Labor Market Opportunities Affect Young Women's Work and Family Decisions? Experimental Evidence from India
’,
Quarterly Journal of Economics
 ,
127
(
2
):
753
792
.
Johnson
H. B.
(
1967
) ‘
The Location of Christian Missions in Africa
’,
Geographical Review
 ,
57
(
2
):
168
202
.
Kalinga
O. J.
(
1985
) ‘
Colonial Rule, Missionaries and Ethnicity in the North Nyasa District, 1891-1938
’,
African Studies Review
 ,
28
(
1
):
57
72
.
Kalinga
O. J. M.
(
2012
)
Historical Dictionary of Malawi
 ,
4th edn
.
Lanham, Maryland
:
Scarecrow Press, Inc
.
Lokshin
M.
(
2006
) ‘
Difference-based Semiparametric Estimation of Partial Linear Models
’,
The Stata Journal
 ,
6
(
3
):
377
383
.
McCleary
,
R. M.
2013
. Protestantism and Human Capital in Guatemala and the Republic of Korea. ADB Economics Working Paper Series 332.
McCracken
J.
(
1977
)
Politics and Christianity in Malawi 1875–1940: The Impact of the Livingstonia Mission in the Northen Province
 .
Cambridge, London
:
Cambridge University Press
.
McCracken
J.
(
2012
)
A History of Malawi 1859–1966
 .
New York
:
Boydell & Brewer Inc
. 668 Mt Hope Avenue Rochester, NY 14620-2731.
Milazzo
,
A.
2014
. Son Preference, Fertility and Family Structure: Evidence from Reproductive Behavior among Nigerian Women. Policy Research Working Paper 6869, Human Development and Public Service Team, Development Research Group, World Bank.
Msiska
A. W.
(
1995
) ‘
The Spread of Islam in Malawi and its Impact on Yao Rites of Passage, 1870-1960
’,
The Society of Malawi Journal
 ,
48
(
1
):
49
86
.
Nunn
N.
(
2010
) ‘
Religious Conversion in Colonial Africa
’,
American Economic Review: Papers & Proceedings
 ,
100
(
2
):
147
152
.
Nunn
N.
(
2014
) ‘Gender and Missionary Influence in Colonial Africa’, in
Akyeampong
E.
,
Bates
R.
,
Nunn
N.
,
Robinson
J. A.
(eds)
Africa's Development in Historical Perspective
 .
New York
:
Cambridge University Press
.
Nunn
N.
,
Wantchekon
L.
(
2011
) ‘
The Slave Trade and the Origins of Mistrust in Africa
’,
American Economic Review
 ,
101
(
7
):
3221
3252
.
Oster
E.
(forthcoming) ‘
Unobservable Selection and Coefficient Stability: Theory and Validation
’,
Journal of Business and Economic Statistics
 . http://amstat.tandfonline.com/doi/full/10.1080/07350015.2016.1227711?scroll=top&needAccess=true#.WImJfNG7rPT.
Pike
J. G.
(
1965
) ‘
A Pre-colonial History of Malawi
’,
Nyasaland Journal
 ,
18
(
1
):
22
54
.
Pike
J. G.
(
1968
)
A Political and Economic History
 .
London, W. I
:
Pall Mall Press Ltd
.
Pike
J. G.
,
Rimmington
G. T.
(
1965
)
A Geographical Study
 .
Malawi
:
Oxford University Press
.
Schoellman
T.
,
Tertilt
M.
(
2006
) ‘
Marriage Laws and Growth in sub-Saharan Africa
’,
American Economic Review
 ,
96
(
2
):
295
298
.
Shillington
K.
(
2005
)
Encyclopedia of African History
 ,
Vol. 2
.
New York
:
Fitzroy Dearborn
.
Soares
R. R.
(
2006
) ‘
The Effect of Longevity on Schooling and Fertility: Evidence from Brazilian Demographic and Health Survey
’,
Journal of Population Economics
 ,
19
(
1
):
71
97
.
Tertilt
M.
(
2005
) ‘
Polygyny, Fertility, and Savings
’,
Journal of Political Economy
 ,
113
(
6
):
1341
1371
.
Tertilt
M.
(
2006
) ‘
Polygyny, Women's Rights, and Development
’,
Journal of the European Economic Association
 ,
4
(
2–3
):
523
530
.
Thompson
T. J.
(
1995
)
Christianity in Northern Malawi: Donald Fraser's Missionary Methods and Ngoni Culture
 .
Leiden, New York, Köln
:
E.J. Brill
.
Ueyama
M.
,
Yamauchi
F.
(
2009
) ‘
Marriage Behavior Response to Prime-age Adult Mortality: Evidence from Malawi
’,
Demography
 ,
46
(
1
):
43
63
.
UNDP.
2004
. Human Development Report 2004: Cultural Liberty in Today's Diverse World. New York, USA.
Vail
L.
,
White
L.
(
1991
) ‘Tribalism in the Political History of Malawi’, in
Vail
L.
(ed.)
Chapter 5 in The Creation of Tribalism in Southern Africa
 .
Berkeley and Los Angeles
:
University of California Press
.
Valencia-Caicedo
,
F.
2014
. The Mission: Human Capital Transmission, Economic Persistence and Culture in South AmericaMimeo.
Wantchekon
L.
,
Klašnja
M.
,
Novta
N.
(
2016
) ‘
Education and Human Capital Externalities: Evidence from Benin
’,
Quarterly Journal of Economics
 ,
130
(
2
):
703
757
.
Woodberry
,
R. D.
2004
. The shadow of empire: Christian Missions, colonial policy, and democracy in post colonial societies. PhD Thesis, Department of Sociology, University of North Carolina.
Woodberry
R. D.
(
2012
) ‘
The Missionary Roots of Liberal Democracy
’,
American Political Science Review
 ,
106
(
2
):
244
274
.
Woodberry
R. D.
,
Shah
T. S.
(
2004
) ‘
Christianity and Democracy: The Pioneering Protestants
’,
Journal of Democracy
 ,
15
(
2
):
47
61
.
Yatchew
A.
(
1997
) ‘
An Elementary Estimator of the Partial Linear Model
’,
Economics Letters
 ,
57
:
135
143
.
Yatchew
A.
(
1998
) ‘
Nonparametric Regression Techniques in Economics
’,
Journal of Economic Literature
 ,
36
(
2
):
669
721
.

Appendix

A. Historical background

A.1 Zambezi expedition

David Livingstone (1813–73), one of the most renowned explorers to make a transcontinental journey across Africa during the middle years of the 19th century, laid the groundwork for the Livingstonia Mission, which was named ‘Livingstonia’ in his honour. Sponsored by the British government, the Scottish missionary headed the ‘Zambezi Expedition’ between 1858 and 1863, which aimed to catalogue the natural resources of the Zambezi River area, as well as to identify trade routes necessary for transporting raw materials from the African interior to coastal trading points and eventually selling these materials in the British market. The opening of the African continent to the world economy and the promotion of local commercial activities were believed to have contributed to the decline in the African slave trade by creating the ‘legitimate’ trade of products (e.g., cotton and ivory), so that Africans did not have to sell their own people to obtain the guns, gun-powder and cloth that they desired. The expedition was also greatly motivated by Livingstone's zeal for ending the slave trade and bringing Christianity and civilisation to the Africans. He also urged the cultivation of cotton (and other crops) in the unexplored territory to make the missionary activities self-supporting and to bypass the slave-owning American states from which most of Britain's raw cotton was sourced.

In this expedition, he reached the conclusions that the only practicable means of linking the interior with the coast was to establish a deep-water route from the Shire River to Lake Malawi by steamer. He identified the Shire Highlands, a plateau in southern Malawi, as being a suitable area for white settlement as well as for the creation of a cash-crop economy.35 However, his statements shortly encountered harsh criticism from James Stewart (1831–1905). As a devout adherent of Livingstone's model of an ‘industrial mission’, Stewart travelled to the Shire Highlands in 1861 to establish an agricultural and Christian settlement. However, in contrast with the indications provided by Livingstone, Stewart, in his journey, eventually concluded that no commercial benefits could be obtained from settlement in the region and discovered that the Zambezi–Shire route was shallow and difficult to navigate by steamship.36,37 Shortly, the British government decided to withdraw the Zambezi expedition, which had lasted 6 years, and many at the time commented that it was a failure with none of its purposes fulfilled.

A.2 The spread of Islam among the Yao

The Yao are a major ethnic group that primarily resides at the southern end of Lake Malawi. They originally inhabited northern Mozambique, and after an attack launched by the Makua people around 1830, they migrated from their traditional home to present-day Malawi and Tanzania, which shaped their current population distribution (see Figure 2 for the recent spatial distribution of linguistic groups). The Yao are predominantly Muslim and indeed, Table 1, using the pooled data set of the 2000, 2004 and 2010 MDHS, reports that 76% of the interviewed Yao females professed the Islamic faith.

Historically, the Yao were under considerable Islamic Influence because of their alliance with Arabs involved in the caravan trade through which the east coast of Africa was linked to markets in the African interior. For example, it was observed that by the middle years of the 18th century, a Yao caravan came to Kilwa, a great Arab port, to trade with the Arabs (Pike 1968, pp. 58–59). The Yao–Arab relationship was that of a senior and a junior business partner, through which the Arabs learned of the interior of Africa from the Yao, who in turn traded beads, cloth, guns and gun-powder for ivory, tobacco and slaves.

While the Yao had maintained a relationship with Arab traders since at least the early 18th century, it was not until the 1870–90s that the rapid expansion of Islam among the Yao became apparent (Pike 1968, p. 69; Msiska 1995, p. 52). It was believed that several factors contributed to the mass conversion of the Yao. First, powerful Yao chiefs (e.g., Makanjira and Mponda) adopted Islam to strengthen their economic ties with their Arab trading partners, and using their commercial prowess, to command their subjects’ loyalty. The chiefs’ conversion was typically followed by that of their subjects. Second, after the arrival of the Christian mission, Islam provided a more acceptable solution to Yao's cultural requirements than Christianity for several reasons. First, the Islamic faith did not interfere with Yao traditional customs and social institutions such as polygamy and partial circumcision. Second, in Yao society, slave labour was a fundamental feature and the chiefs needed slaves not only for selling on the export market but also for domestic physical labour (e.g., farming, building, making baskets and sewing garments). Thus, it was not surprising that Christianity, with its missionaries’ attempts to stop the slave trade, lost the battle to bring the Yao into its religious domain. Another reason for the Islamic conversion may be Yao's ongoing clashes with the Ngoni people, another powerful group that had migrated from the Natal region of present-day South Africa. Threatened by Ngoni raids on their territory, adopting Islam was an attempt by the Yao chiefs to form a tactical alliance with the Arab traders who supplied them with flintlocks and Enfield rifles.38

A.3 Educational influence of the Livingstonia Mission

While the Livingstonia Mission is one of the most important missions that introduced Christianity into Malawi, this does not mean that no other missions existed in this country (for instance, Zambezi Industrial Mission headed by Joseph Booth is another example). However, the educational contribution made by the Livingstonia Mission makes its position unique because other missions tended to have less capability for the promotion of education (Shillington 2005, p. 907).39

The Blantyre Mission of the Church of Scotland was a notable exception. It not only actively encouraged educational advancement but, together with the Livingstonia Mission, represents one of the two pioneering missions in Malawi (see Figure 1 for the location of Blantyre). Compared with the Livingstonia Mission, however, it appears that the Christian and educational influence of the Blantyre Mission was less widespread. This limited influence was due to the Blantyre Mission acting largely within a small network of neighbouring villages, owing to its traditional residential policy (McCracken 1977, p. 69). Moreover, while the Livingstonia Mission leads to several independent religious movements (e.g., Watch Tower movement by Elliot Kenan Kamwana and Seventh Day Baptist movement by Charles Domingo), it was still by far the most dominant religious and educational entity in northern Malawi, as well as one of the best organised missionary societies in Central Africa during the early period of its activities (McCracken 1977, pp. 220–221).

While its influence should not be over-simplified, the educational advancement facilitated by the Livingstonia Mission triggered the subsequent economic, political and social transformation in this country. For example, the Overtoun Institution founded at Livingstonia served as a training centre for post-primary education, and thus a source of skilled labour (e.g., clerks, typists, telegraphists and mechanics), not only for the European-controlled economy of Northern Region, but other parts of South and Central Africa as well (e.g., Tanganyika and Northern Rhodesia). Indeed, in the absence of significant commercial opportunities in Northern Region compared with European agricultural estates in the Southern Province, the Livingstonia elites often migrated to seek paid jobs in other areas. This played a crucial role in creating the migrant labour system, which was a central feature of Malawi's colonial economy. Remittances sent by those migrants to their families that remained in their place of origin were also an important source of revenue for the colonial government that imposed a hut tax upon the citizens.

The mission is also often seen as a seedbed of Malawian nationalism, starting with the formation of the Nyasaland African Congress (NAC) in 1944, the first political association established during the colonial period whose membership extended to all regions of Malawi. While a variety of welfare organisations (e.g., North Nyasa Native Association) concerned with the social consequences of colonial administration were founded as a predecessor of the NAC in the late 1910s and 1920s, the mission supplied the educated elites to these native associations. The origin of many educational and economic policies pursued by the independent Malawi government in the 1960s can also be traced back to the demands of these associations.

Another significant contribution of the Livingstonia Mission is the creation of new forms of ethnic identity and social differentiation (McCracken 1977, pp. 147, 294; Vail and White 1991). For instance, while the Tumbuka ethnic group residing in northern areas (see Figure 2) served the Ngoni soldiers as slaves and serfs before the arrival of the mission, by receiving the missionary education, they successfully created a strong ethnic ideology. Indeed, this group is a core of the educated elites in present-day Malawi. On the other hand, the Yao (see Figure 2), an economically powerful ethnic group during the pre-missionary periods, isolated themselves from the mission's educational contacts and, as a result, currently constitute one of the most underdeveloped populations in this country.

B. Geographic and climate controls

This section describes community-level geographic and climatic controls (as well as the original sources), which are all publicly available in the IHS data set. The variables description refers to ‘Geovariables.Description.pdf’ (http://microdata.worldbank.org/index.php/catalog/1003).

B.1 Climatology

The original data on climatology are sourced from ‘WorldClim—Global Climate Data’, University of California, Berkeley.

Mean temperature: average temperature (multiplied by 10°C) based on monthly climate data between 1960 and 1990.

Temperature seasonality: standard deviation of temperature (multiplied by 100) based on monthly climate data between 1960 and 1990.

Mean precipitation: average annual precipitation (mm) based on monthly climate data between 1960 and 1990.

Precipitation seasonality: coefficient of variation of annual precipitation (mm) based on monthly climate data between 1960 and 1990.

B.2 Landscape typology

Agricultural land: percentage under agriculture within approximately 1 km buffer in 2009 based on ‘GlobCover Version 2.3’, sourced from the European Space Agency (ESA) and Université Catholique de Louvain.

Agro-ecological zones: categorical variables for agro-ecological zones in 2009, sourced from HarvestChoice and International Food Policy Research Institute (IFPRI). These zones are (a) tropic-warm/semiarid (reference group); (b) tropic-warm/subhumid; (c) tropic-cool/semiarid and (d) tropic-cool/subhumid.

B.3 Soil and terrain

Elevation: elevation (m) based on the Shuttle Radar Topography Mission (SRTM) 90 m data sourced from the National Aeronautics and Space Administration (NASA).

Slope: slope (percent) based on the SRTM 90 m data sourced from the U.S Geological Survey (USGS).

Topographic wetness index: potential wetness index based on the modified SRTM 90 m data sourced from the Africa Soil Information Service (AfSIS). This index is calculated as ln(Atanb), where A is flow accumulation or effective drainage areas and b is slope gradient.

Terrain roughness: categorical variables for terrain roughness based on the SRTM 90 m data sourced from the LSMS-ISA. Terrain types include (a) plains (reference group); (b) mid-altitude plains; (c) high-altitude plains; (d) lowlands; (e) rugged lowlands; (f) platform (very low plateaus); (g) low plateaus; (h) mid-altitude plateaus; (i) hills; (j) low mountains and (k) mid-altitude mountains.

Nutrient availability: categorical variables for nutrient availability based on ‘Harmonised World Soil Database’ sourced from the Food and Agriculture Organization of the United Nations (FAO); classified as (a) no, slight or moderate constraint (reference group); (b) severe or very severe constraint; (c) mainly non-soil and (d) water.

Nutrient retention capacity: categorical variables for nutrient retention capacity based on the ‘Harmonised World Soil Database’ sourced from the Food and Agriculture Organization of the United Nations (FAO); classified as (a) no, slight or moderate constraint (reference group); (b) severe or very severe constraint and (c) water.

Rooting conditions: categorical variables for rooting conditions based on the ‘Harmonised World Soil Database’ sourced from the Food and Agriculture Organization of the United Nations (FAO); classified as (a) no, slight or moderate constraint (reference group); (b) severe or very severe constraint; (c) mainly non-soil and (d) water.

Oxygen availability to roots: categorical variables for oxygen availability to roots based on the ‘Harmonised World Soil Database’ sourced from the Food and Agriculture Organization of the United Nations (FAO); classified as (a) no, slight or moderate constraint (reference group); (b) severe or very severe constraint and (c) water.

Excess salts: categorical variables for excess salts based on the ‘Harmonised World Soil Database’ sourced from the Food and Agriculture Organization of the United Nations (FAO); classified as (a) no, slight or moderate constraint (reference group); (b) severe or very severe constraint and (c) water.

Toxicity: categorical variables for toxicity based on the ‘Harmonised World Soil Database’ sourced from the Food and Agriculture Organization of the United Nations (FAO); classified as (a) no, slight or moderate constraint (reference group); (b) severe or very severe constraint and (c) water.

Field-management constraint: categorical variables for field-management constraint based on the ‘Harmonised World Soil Database’ sourced from the Food and Agriculture Organization of the United Nations (FAO); classified as (a) no, slight or moderate constraint (reference group); (b) severe or very severe constraint; (c) mainly non-soil and (d) water.

B.4 Crop season parameters

Greenness changes: mean total change in greenness (averaged by district), the integral of the daily Enhanced Vegetation Index (EVI) values within a primary growing season between 2001 and 2010, sourced from ‘Land Cover Dynamics—MODIS’, Boston University.

C. Merging IHS community-level information with MDHS data

C.1 Community-level positions

The MDHS collected the coordinates of households groupings known as clusters (communities). To maintain the confidentiality of the surveyed respondents, the GPS latitude/longitude position was publicised after displacing the coordinates by applying a random offset within a specified range to the position. After this adjustment was made, urban clusters contained 0–2 km of positional error. On the other hand, rural clusters contained 0–5 km of error with a further 1% of them offset by 0–10 km. Nevertheless, this displacement still made the surveyed clusters fall within an original surveyed area of the country's second administrative level (district). For details, see http://www.measuredhs.com/What-We-Do/GPS-Data-Collection.cfm.

In the IHS, GPS-based household location data were collected. To ensure respondent confidentiality, this information was disseminated as a community-level value after manipulating the household-level GPS coordinates. This manipulation included first computing the average of household-level coordinates in a community (enumeration area) and then following the MDHS methodology of applying a random offset to the average coordinate value. For urban areas, a range of 0–2 km was applied as the random offset, in contrast with a range of 0–5 km for the offset used in rural areas. An additional 0–10 km offset was also used for 1% of the rural clusters effectively to increase the publicly known range of positional displacement (for all rural points) to the level of 10 km with minimal noise. Similar to the MDHS, this displacement was made in such a manner to keep a community's representative location in its original district. For the details, see Third Integrated Household Survey (IHS3) 2010–2011 Basic Information Document, March 2012 (http://siteresources.worldbank.org/INTLSMS/Resources/3358986-1233781970982/5800988-1271185595871/IHS3.BID.FINAL.pdf).

C.2 Finding the nearest IHS community

By using the community-level GPS coordinates provided by both the MDHS and IHS, this study selected, from 768 communities surveyed in the IHS, the geographically closest one to each community surveyed in the MDHS, which contained more than 1,900 communities over all three surveys from 2000, 2004 and 2010.

When calculating the distance between the MDHS and IHS communities, this study used the GCD, the shortest distance between any two points on the surface of a sphere measured along a path on the surface of the sphere. More specifically, in this paper, the GCD between two points i and j was computed as follows:  

(C.1)
Radius(6378.7km)×arccos[sin(latitudei57.2958)×sin(latitudej57.2958)+cos(latitudei57.2958)×cos(latitudej57.2958)×cos(longitudej57.2958longitudei57.2958)].

Because both the MDHS and IHS communities were spatially spread over the country (see Figure A.1, the sample communities in both the surveys shown as individual points. For ease of visual identification, only the 2010 MDHS communities were compared with the IHS communities in the figure), it was not difficult to identify the IHS community with the closest proximity (i.e., with relatively short distance) to all the MDHS communities. As a matter of fact, approximately 95% (99%) of the MDHS communities could be matched to its nearest IHS community within 10 (15) km, with the MDHS community having a maximum distance of approximately 67 km to the nearest IHS community.

C.3 Goodness of fit of the IHS data to the MDHS data

To observe the goodness of fit of the characteristics of the nearest IHS communities to the MDHS data, three informal checks were performed. In column (a) in Table A.3, this study first regressed an indicator, equal to one if the MDHS sample females were Christian and zero otherwise, on a dummy variable, which takes one if Christianity was the most common religion practiced in the nearest IHS communities. If the community-level characteristics of the IHS data fit well to the MDHS data, a significantly positive relationship is likely to arise, which is indeed observed. Similar exercises associated with Islam and other or no religion were also conducted in columns (b) and (c) in the table, providing further support for the fitness.

With an emphasis on ethnicity, the second exercise exploited a similar idea as the first check. For example, in column (d) in Table A.3, a dummy taking the value of one if the MDHS sample females were identified as the Chewa ethnic group was related to an indicator variable, taking the value of one if the most common language spoken at home in the nearest IHS communities was Chewa and zero otherwise. Again, given the good fit of the IHS community characteristics to the MHDS data, it is quite likely that these two variables have a significantly positive association, which was indeed confirmed in the result presented in that column. The analysis in the remaining columns in the table reports the estimation results implemented for the other ethnic groups. As a whole, it appears that the results provided good support for the goodness of fit between the two data sets.

Finally, in columns (f)–(j) in Table A.4, an indicator, which takes one if the MDHS sample females in wedded relationship were not born in their current residential location and zero otherwise, was regressed on a dummy variable, equal to one if the nearest IHS communities traced their descent through their mothers and zero otherwise. This exercise was conducted to check if married females were less likely to be migrants to the present residential location if the nearest IHS communities traditionally adopted a matrilineal descent system. This is because the default norm of marriage-related relocation in a matrilineal society is matrilocal. In matrilocality, females stay in their natal villages, to which their husbands relocate. This concept is in contrast with patrilocality (females leaving their natal homes and marrying into their husband's villages), which is associated with a patrilineal descent system. As the norm may not be strictly enforced in urban areas, the analysis examined the interaction between the dummy for the nearest IHS communities characterised by the matrilineal descent rule and the distance (km) to the nearest town having population over 20,000. In the analysis, the distance took a categorical form in which communities were separated into six groups in column (f), five groups in (g), four groups in (h), three groups in (i) and two groups in (j). In all these columns, the reference group consists of communities situated the farthest from the nearest population centre.

As expected, the results show that compared with their counterparts in patrilineal communities, married females residing in matrilineal communities were less likely to be migrants to their current residential location. In addition, within matrilineal communities, married females were more likely to stay in their natal homes if their current residences were situated at greater distances from the nearest population centre. The analysis in columns (k)–(o) limits its attention to data pertaining to females living in MDHS communities located less than 10 km from the nearest IHS communities. The estimation results also revealed a similar pattern to those obtained from the analysis using the full sample. Consequently, the relationship between the migrant probability and matriliny may further support the goodness of fit of the community-level characteristics of the nearest IHS communities to the individual characteristics of the MDHS females.

Moreover, using only the IHS data (i.e., the female observations of the IHS) and a similar set of controls and specifications to those exploited in columns (f)–(o) in Table A.4, the migrant probability was again related to a matrilineal descent rule in columns (a)–(e) in that table. The estimation results about the relationship between the migrant probability and matriliny revealed a remarkably similar pattern to those obtained from the analysis in columns (f)–(o) using the MDHS data matched with the IHS. This finding may also provide support for the goodness of fit between these two data sets.

Figure A1:

Spatial Distribution of Sampled Communities: MDHS 2010 and IHS 2010–11.Note: The background map is sourced from DIVA-GIS (http://www.diva-gis.org/datadown).

Figure A1:

Spatial Distribution of Sampled Communities: MDHS 2010 and IHS 2010–11.Note: The background map is sourced from DIVA-GIS (http://www.diva-gis.org/datadown).

Figure A2:

Estimated Non-parametric Function of the Missionary Influence (non-Yao).Notes: (1) This figure is based on (Yatchew, 1997; Yatchew, 1998)'s difference-based semi-parametric estimation of a partial lineal model. (2) The values in the vertical axis are outcomes minus the estimated parametric part. (3) With an exception of the non-parametric part of the distance to Livingstonia, the specifications and samples exploited for the estimations correspond to those in column (f) in Table 3 (early marriage); columns (c) and (g) in Table 4 (polygyny) and columns (c), (f) and (j) in Table 6 (Christianity and academic skills).

Figure A2:

Estimated Non-parametric Function of the Missionary Influence (non-Yao).Notes: (1) This figure is based on (Yatchew, 1997; Yatchew, 1998)'s difference-based semi-parametric estimation of a partial lineal model. (2) The values in the vertical axis are outcomes minus the estimated parametric part. (3) With an exception of the non-parametric part of the distance to Livingstonia, the specifications and samples exploited for the estimations correspond to those in column (f) in Table 3 (early marriage); columns (c) and (g) in Table 4 (polygyny) and columns (c), (f) and (j) in Table 6 (Christianity and academic skills).

Table A1:

Influence of an Abandoned Mission Station (non-Yao)

Dependent variables One if Christian One if unable to read Education (years) Age at first marriage One if married Years to first marriage 
Sample All All All All ever married Aged below 18 Aged below 18 
Probit (ME) LPM Probit (ME) LPM OLS OLS Probit (ME) LPM Hazard
ratio 
(a) (b) (c) (d) (e) (f) (g) (h) (i) 
 Distance to −0.063*** −0.120*** 0.137*** 0.123*** −0.948*** −0.217 0.134** 0.171** 2.690** 
  Livingstonia (100 km) (0.015) (0.024) (0.043) (0.039) (0.358) (0.205) (0.055) (0.073) (1.288) 
 Distance to the nearest −0.000 −0.025 0.257*** 0.234*** −2.637*** −0.499** 0.085* 0.095* 1.767 
  mission station (100 km) (0.012) (0.024) (0.039) (0.036) (0.343) (0.199) (0.044) (0.051) (0.667) 
 Distance to 0.046*** 0.071*** 0.000 0.001 −0.083 −0.200 −0.000 0.002 1.057 
  Cape Maclear (100 km) (0.009) (0.016) (0.029) (0.027) (0.245) (0.145) (0.031) (0.037) (0.293) 
 Shoenfeld res. (p-val.)         0.575 
R2 0.224 0.129 0.108 0.131 0.258 0.038 0.148 0.110  
 No. of obs. 40,949 41,118 41,033 41,033 41,129 33,505 5,508 5,508 5,508 
Dependent variables One if polygyny One if polygyny (zero
if unmarried) 
     
Sample All married All      
Probit
(ME) 
LPM Probit
(ME) 
LPM      
(j) (k) (l) (m)      
 Distance to 0.048* 0.057* 0.029* 0.040      
  Livingstonia (100 km) (0.025) (0.034) (0.016) (0.025)      
 Distance to the nearest 0.109*** 0.117*** 0.086*** 0.101***      
  mission station (100 km) (0.029) (0.027) (0.016) (0.020)      
 Distance to 0.036** 0.043** 0.028** 0.037**      
  Cape Maclear (100 km) (0.018) (0.020) (0.012) (0.015)      
R2 0.072 0.061 0.093 0.061      
 No. of obs. 28,411 28,411 41,130 41,130      
 Individual controls Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Matrilineal com. dummy Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 GPS coordinate Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Ethnicity FE Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Com.-language FE Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Geography and climate Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Historical controls No No No No No No No No No 
 District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Round FE Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Dependent variables One if Christian One if unable to read Education (years) Age at first marriage One if married Years to first marriage 
Sample All All All All ever married Aged below 18 Aged below 18 
Probit (ME) LPM Probit (ME) LPM OLS OLS Probit (ME) LPM Hazard
ratio 
(a) (b) (c) (d) (e) (f) (g) (h) (i) 
 Distance to −0.063*** −0.120*** 0.137*** 0.123*** −0.948*** −0.217 0.134** 0.171** 2.690** 
  Livingstonia (100 km) (0.015) (0.024) (0.043) (0.039) (0.358) (0.205) (0.055) (0.073) (1.288) 
 Distance to the nearest −0.000 −0.025 0.257*** 0.234*** −2.637*** −0.499** 0.085* 0.095* 1.767 
  mission station (100 km) (0.012) (0.024) (0.039) (0.036) (0.343) (0.199) (0.044) (0.051) (0.667) 
 Distance to 0.046*** 0.071*** 0.000 0.001 −0.083 −0.200 −0.000 0.002 1.057 
  Cape Maclear (100 km) (0.009) (0.016) (0.029) (0.027) (0.245) (0.145) (0.031) (0.037) (0.293) 
 Shoenfeld res. (p-val.)         0.575 
R2 0.224 0.129 0.108 0.131 0.258 0.038 0.148 0.110  
 No. of obs. 40,949 41,118 41,033 41,033 41,129 33,505 5,508 5,508 5,508 
Dependent variables One if polygyny One if polygyny (zero
if unmarried) 
     
Sample All married All      
Probit
(ME) 
LPM Probit
(ME) 
LPM      
(j) (k) (l) (m)      
 Distance to 0.048* 0.057* 0.029* 0.040      
  Livingstonia (100 km) (0.025) (0.034) (0.016) (0.025)      
 Distance to the nearest 0.109*** 0.117*** 0.086*** 0.101***      
  mission station (100 km) (0.029) (0.027) (0.016) (0.020)      
 Distance to 0.036** 0.043** 0.028** 0.037**      
  Cape Maclear (100 km) (0.018) (0.020) (0.012) (0.015)      
R2 0.072 0.061 0.093 0.061      
 No. of obs. 28,411 28,411 41,130 41,130      
 Individual controls Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Matrilineal com. dummy Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 GPS coordinate Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Ethnicity FE Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Com.-language FE Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Geography and climate Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Historical controls No No No No No No No No No 
 District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Round FE Yes Yes Yes Yes Yes Yes Yes Yes Yes 

Notes: (1) Figures () are standard errors. ***Significance at 1%, **significance at 5% and *significance at 10%. (2) Standard errors are robust to heteroskedasticity and clustered residuals within each community. (3) The individual controls include age (years), birth order, no. of alive siblings at age 10 and no. of late siblings at age 10. (4) The ethnicity is classified into 12 groups, i.e., Chewa, Lambya, Lomwe, Mang'anja, Ndali, Ngoni, Nkhonde, Nyanja, Sena, Tonga, Tumbuka and other. (5) The community language is classified into 14 groups, i.e., Chewa, Lambya, Lomwe, Ngoni, Nkhonde, Nyakyusa, Nyanja, Sena, Senga, Sukwa, Tonga, Tumbuka, Yao and other. (6) The geographic and climate controls contain community-level information on climatology, landscape typology, soil and terrain, crop season parameters. See Appendix B for the details.

Table A2:

Missionary Influence (DID, OLS)

Dependent variables One if Christian  
(a) (b) (c) (d)     
Distance to Livingstonia (100 km) × non-Yao −0.090*** −0.091*** −0.093*** −0.070***     
(0.011) (0.011) (0.011) (0.011)     
Distance to Livingstonia (100 km) 0.002 −0.016 0.006      
(0.022) (0.024) (0.022)      
Non-Yao (dummy) 1.034*** 1.009***  0.822***     
(0.056) (0.056)  (0.058)     
R2 0.551 0.559 0.552 0.635     
No. of obs. 47,222 44,728 47,222 44,728     
Dependent variables Education (years) One if unable to read 
(e) (f) (g) (h) (i) (j) (k) (l) 
Distance to Livingstonia (100 km) × non-Yao −0.304*** −0.222*** −0.213*** −0.169** 0.037*** 0.029*** 0.028*** 0.018** 
(0.072) (0.074) (0.074) (0.070) (0.009) (0.009) (0.009) (0.009) 
Distance to Livingstonia (100 km) −1.263*** −1.156*** −0.867***  0.153*** 0.139*** 0.112***  
(0.323) (0.400) (0.327)  (0.036) (0.042) (0.035)  
Non-Yao (dummy) 2.334*** 1.596***  1.188*** −0.290*** −0.202***  −0.135*** 
(0.386) (0.398)  (0.373) (0.045) (0.046)  (0.047) 
R2 0.258 0.260 0.267 0.396 0.141 0.143 0.145 0.231 
No. of obs. 47,233 44,738 47,233 44,738 47,122 44,646 47,122 44,646 
Dependent variables Age at first marriage (all ever married) One if married (aged below 18) 
(m) (n) (o) (p) (q) (r) (s) (t) 
Distance to Livingstonia (100 km) × non-Yao −0.153** −0.073 −0.085 −0.107 0.012 0.012 0.018 −0.003 
(0.060) (0.062) (0.062) (0.068) (0.013) (0.013) (0.013) (0.020) 
Distance to Livingstonia (100 km) −0.234 −0.187 −0.194  0.161** 0.128* 0.144**  
(0.191) (0.243) (0.199)  (0.068) (0.070) (0.068)  
Non-Yao (dummy) 1.037*** 0.704**  0.882** −0.079 −0.073  0.024 
(0.316) (0.332)  (0.367) (0.065) (0.070)  (0.103) 
R2 0.036 0.036 0.037 0.099 0.103 0.101 0.105 0.389 
No. of obs. 38,671 36,545 38,671 36,545 6,277 5,952 6,277 5,952 
Dependent variables One if polygyny (all married) One if polygyny (all, zero if unmarried) 
(u) (v) (w) (x) (y) (z) (aa) (ab) 
Distance to Livingstonia (100 km) × non-Yao 0.023*** 0.024*** 0.028*** 0.020** 0.021*** 0.022*** 0.024*** 0.017*** 
(0.008) (0.008) (0.008) (0.008) (0.006) (0.006) (0.006) (0.006) 
Distance to Livingstonia (100 km) 0.052* 0.041 0.050  0.041* 0.032 0.035  
(0.031) (0.039) (0.031)  (0.024) (0.029) (0.024)  
Non-Yao (dummy) −0.179*** −0.180***  −0.153*** −0.150*** −0.149***  −0.119*** 
(0.041) (0.043)  (0.047) (0.031) (0.032)  (0.034) 
R2 0.063 0.062 0.064 0.141 0.064 0.063 0.064 0.122 
No. of obs. 32,723 30,935 32,723 30,935 47,234 44,739 47,234 44,739 
Individual controls Yes Yes Yes Yes Yes Yes Yes Yes 
Dis. to the nearest mission station Yes Yes Yes No Yes Yes Yes No 
Matrilineal com. dummy Yes Yes Yes No Yes Yes Yes No 
GPS coordinate Yes Yes Yes No Yes Yes Yes No 
Ethnicity FE No No Yes No No No Yes No 
Com.-language FE Yes Yes Yes No Yes Yes Yes No 
Geography and climate Yes Yes Yes No Yes Yes Yes No 
Historical controls No Yes No Yes No Yes No Yes 
District FE Yes Yes Yes No Yes Yes Yes No 
Community FE No No No Yes No No No Yes 
Round FE Yes Yes Yes Yes Yes Yes Yes Yes 
Dependent variables One if Christian  
(a) (b) (c) (d)     
Distance to Livingstonia (100 km) × non-Yao −0.090*** −0.091*** −0.093*** −0.070***     
(0.011) (0.011) (0.011) (0.011)     
Distance to Livingstonia (100 km) 0.002 −0.016 0.006      
(0.022) (0.024) (0.022)      
Non-Yao (dummy) 1.034*** 1.009***  0.822***     
(0.056) (0.056)  (0.058)     
R2 0.551 0.559 0.552 0.635     
No. of obs. 47,222 44,728 47,222 44,728     
Dependent variables Education (years) One if unable to read 
(e) (f) (g) (h) (i) (j) (k) (l) 
Distance to Livingstonia (100 km) × non-Yao −0.304*** −0.222*** −0.213*** −0.169** 0.037*** 0.029*** 0.028*** 0.018** 
(0.072) (0.074) (0.074) (0.070) (0.009) (0.009) (0.009) (0.009) 
Distance to Livingstonia (100 km) −1.263*** −1.156*** −0.867***  0.153*** 0.139*** 0.112***  
(0.323) (0.400) (0.327)  (0.036) (0.042) (0.035)  
Non-Yao (dummy) 2.334*** 1.596***  1.188*** −0.290*** −0.202***  −0.135*** 
(0.386) (0.398)  (0.373) (0.045) (0.046)  (0.047) 
R2 0.258 0.260 0.267 0.396 0.141 0.143 0.145 0.231 
No. of obs. 47,233 44,738 47,233 44,738 47,122 44,646 47,122 44,646 
Dependent variables Age at first marriage (all ever married) One if married (aged below 18) 
(m) (n) (o) (p) (q) (r) (s) (t) 
Distance to Livingstonia (100 km) × non-Yao −0.153** −0.073 −0.085 −0.107 0.012 0.012 0.018 −0.003 
(0.060) (0.062) (0.062) (0.068) (0.013) (0.013) (0.013) (0.020) 
Distance to Livingstonia (100 km) −0.234 −0.187 −0.194  0.161** 0.128* 0.144**  
(0.191) (0.243) (0.199)  (0.068) (0.070) (0.068)  
Non-Yao (dummy) 1.037*** 0.704**  0.882** −0.079 −0.073  0.024 
(0.316) (0.332)  (0.367) (0.065) (0.070)  (0.103) 
R2 0.036 0.036 0.037 0.099 0.103 0.101 0.105 0.389 
No. of obs. 38,671 36,545 38,671 36,545 6,277 5,952 6,277 5,952 
Dependent variables One if polygyny (all married) One if polygyny (all, zero if unmarried) 
(u) (v) (w) (x) (y) (z) (aa) (ab) 
Distance to Livingstonia (100 km) × non-Yao 0.023*** 0.024*** 0.028*** 0.020** 0.021*** 0.022*** 0.024*** 0.017*** 
(0.008) (0.008) (0.008) (0.008) (0.006) (0.006) (0.006) (0.006) 
Distance to Livingstonia (100 km) 0.052* 0.041 0.050  0.041* 0.032 0.035  
(0.031) (0.039) (0.031)  (0.024) (0.029) (0.024)  
Non-Yao (dummy) −0.179*** −0.180***  −0.153*** −0.150*** −0.149***  −0.119*** 
(0.041) (0.043)  (0.047) (0.031) (0.032)  (0.034) 
R2 0.063 0.062 0.064 0.141 0.064 0.063 0.064 0.122 
No. of obs. 32,723 30,935 32,723 30,935 47,234 44,739 47,234 44,739 
Individual controls Yes Yes Yes Yes Yes Yes Yes Yes 
Dis. to the nearest mission station Yes Yes Yes No Yes Yes Yes No 
Matrilineal com. dummy Yes Yes Yes No Yes Yes Yes No 
GPS coordinate Yes Yes Yes No Yes Yes Yes No 
Ethnicity FE No No Yes No No No Yes No 
Com.-language FE Yes Yes Yes No Yes Yes Yes No 
Geography and climate Yes Yes Yes No Yes Yes Yes No 
Historical controls No Yes No Yes No Yes No Yes 
District FE Yes Yes Yes No Yes Yes Yes No 
Community FE No No No Yes No No No Yes 
Round FE Yes Yes Yes Yes Yes Yes Yes Yes 

Notes: (1) Figures () are standard errors. ***Significance at 1%, **significance at 5% and *significance at 10%. (2) Standard errors are robust to heteroskedasticity and clustered residuals within each community. (3) The individual controls include age (years), birth order, no. of alive siblings at age 10 and no. of late siblings at age 10. (4) The ethnicity is classified into 13 groups, i.e., Chewa, Lambya, Lomwe, Mang'anja, Ndali, Ngoni, Nkhonde, Nyanja, Sena, Tonga, Tumbuka, Yao and other. (5) The community language is classified into 14 groups, i.e., Chewa, Lambya, Lomwe, Ngoni, Nkhonde, Nyakyusa, Nyanja, Sena, Senga, Sukwa, Tonga, Tumbuka, Yao and other. (6) The geographic and climate controls contain community-level information on climatology, landscape typology, soil and terrain, crop season parameters. See Appendix B for the details. (7) The ethnicity-level historical controls include (i) a dummy variable, equal to one if a European explorer travelled through land historically inhabited by an ethnic group; (ii) a dummy variable, equal to one if any part of railway lines in the first decade of the 20th century drawn from Century Company (1911) passed through land historically inhabited by an ethnic group and (iii) the total number of slaves taken from an ethnic group that was normalised by the area of land inhabited by the ethnic group during the 19th century (log of one plus the normalised slave export measure).

Table A3:

Goodness of the Fit of the IHS Community Characteristics to the MDHS Data: Religion and Ethnicity (OLS)

Dependent variables based on the MDHS One if 
Practice Christianity Practice Islam Practice other or no religion Chewa Lambya Lomwe Ngoni Nkhonde Nyanja Sena Tonga Tumbuka Yao Other ethnicity 
(a) (b) (c) (d) (e) (f) (g) (h) (i) (j) (k) (l) (m) (n) 
Independent variables based on the IHS 
The most common religion practiced in a community 
 Christianity 0.121***              
(0.019)              
 Islam  0.137***             
 (0.023)             
 Traditional   0.009*            
  (0.005)            
The most common language spoken at home in a community 
 Chewa    0.074***           
   (0.020)           
 Lambya     0.108*          
    (0.063)          
 Lomwe      0.059         
     (0.041)         
 Ngoni       0.041        
      (0.049)        
 Nkhonde        0.301***       
       (0.064)       
 Nyanja         0.005      
        (0.004)      
 Sena          0.205***     
         (0.050)     
 Tonga           0.343***    
          (0.097)    
 Tumbuka            0.442***   
           (0.042)   
 Yao             0.180***  
            (0.034)  
 Other              0.228*** 
             (0.055) 
District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Year FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
R2 0.323 0.349 0.014 0.505 0.113 0.404 0.340 0.351 0.035 0.508 0.477 0.518 0.328 0.196 
No. of obs. 47,371 47,371 47,371 47,369 47,369 47,369 47,369 47,369 47,369 47,369 47,369 47,369 47,369 47,369 
Dependent variables based on the MDHS One if 
Practice Christianity Practice Islam Practice other or no religion Chewa Lambya Lomwe Ngoni Nkhonde Nyanja Sena Tonga Tumbuka Yao Other ethnicity 
(a) (b) (c) (d) (e) (f) (g) (h) (i) (j) (k) (l) (m) (n) 
Independent variables based on the IHS 
The most common religion practiced in a community 
 Christianity 0.121***              
(0.019)              
 Islam  0.137***             
 (0.023)             
 Traditional   0.009*            
  (0.005)            
The most common language spoken at home in a community 
 Chewa    0.074***           
   (0.020)           
 Lambya     0.108*          
    (0.063)          
 Lomwe      0.059         
     (0.041)         
 Ngoni       0.041        
      (0.049)        
 Nkhonde        0.301***       
       (0.064)       
 Nyanja         0.005      
        (0.004)      
 Sena          0.205***     
         (0.050)     
 Tonga           0.343***    
          (0.097)    
 Tumbuka            0.442***   
           (0.042)   
 Yao             0.180***  
            (0.034)  
 Other              0.228*** 
             (0.055) 
District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Year FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
R2 0.323 0.349 0.014 0.505 0.113 0.404 0.340 0.351 0.035 0.508 0.477 0.518 0.328 0.196 
No. of obs. 47,371 47,371 47,371 47,369 47,369 47,369 47,369 47,369 47,369 47,369 47,369 47,369 47,369 47,369 

Notes: (1) Figures () are standard errors. ***Significance at 1%, **significance at 5% and *significance at 10%. (2) Standard errors are robust to heteroskedasticity and clustered residuals within each community.

Table A4:

Goodness of the Fit of the IHS Community Characteristics to the MDHS Data: Descent Rules and Migrant Probability of Married Females (OLS)

Dependent variable One if migrant 
Data sources IHS MDHS matched with IHS 
Sample Females who resided in all the MDHS communities Females who resided in the MDHS communities with distance to the nearest IHS communities <10 km 
(a) (b) (c) (d) (e) (f) (g) (h) (i) (j) (k) (l) (m) (n) (o) 
Matrilineal descent community × A community's distance (km) to the nearest town having population >20,000 
 1st quantile 0.138 0.113 0.056 0.033 0.056 0.132*** 0.109*** 0.120*** 0.062*** 0.079*** 0.123*** 0.106*** 0.113*** 0.055** 0.076*** 
(0.098) (0.081) (0.066) (0.045) (0.043) (0.040) (0.037) (0.033) (0.023) (0.022) (0.042) (0.038) (0.035) (0.024) (0.023) 
 2nd quantile 0.162** 0.141** 0.062 0.065*  0.120*** 0.132*** 0.080** 0.042**  0.106** 0.127*** 0.081** 0.041**  
(0.074) (0.066) (0.057) (0.036)  (0.040) (0.036) (0.033) (0.017)  (0.042) (0.038) (0.034) (0.018)  
 3rd quantile 0.264*** 0.198*** 0.005   0.092** 0.064* 0.011   0.087** 0.064* 0.013   
(0.073) (0.070) (0.053)   (0.038) (0.035) (0.028)   (0.039) (0.036) (0.030)   
 4th quantile 0.204*** 0.103*    0.043 0.010    0.036 0.011    
(0.074) (0.053)    (0.033) (0.030)    (0.035) (0.031)    
 5th quantile 0.179***     0.035     0.025     
(0.058)     (0.033)     (0.035)     
A community's distance (km) to the nearest town having population >20,000 
 1st quantile −0.172* −0.105 −0.123* −0.029 −0.092** −0.048 −0.028 −0.018 0.019 −0.046** −0.045 −0.032 −0.011 0.031 −0.040** 
(0.096) (0.084) (0.070) (0.048) (0.042) (0.032) (0.030) (0.028) (0.019) (0.018) (0.034) (0.032) (0.030) (0.020) (0.019) 
 2nd quantile −0.144* −0.128* −0.155*** −0.067**  −0.067** −0.070** −0.053* −0.029*  −0.054 −0.067** −0.054* −0.022  
(0.074) (0.068) (0.057) (0.033)  (0.033) (0.030) (0.028) (0.015)  (0.035) (0.032) (0.029) (0.015)  
 3rd quantile −0.228*** −0.177*** −0.068   −0.059* −0.022 0.005   −0.052 −0.030 −0.001   
(0.066) (0.059) (0.044)   (0.031) (0.029) (0.024)   (0.033) (0.031) (0.025)   
 4th quantile −0.116** −0.140***    −0.018 0.005    −0.017 −0.003    
(0.059) (0.043)    (0.027) (0.025)    (0.028) (0.026)    
 5th quantile −0.133***     −0.027     −0.021     
(0.048)     (0.027)     (0.028)     
Matrilineal descent −0.126** −0.084* 0.002 0.000 0.005 −0.062** −0.055** −0.044* −0.028 −0.031* −0.056* −0.056* −0.045* −0.027 −0.032* 
(0.055) (0.049) (0.047) (0.039) (0.039) (0.029) (0.028) (0.025) (0.020) (0.018) (0.030) (0.029) (0.026) (0.020) (0.019) 
Dual descent 0.023 0.011 0.018 0.015 0.017 0.065*** 0.061*** 0.062*** 0.067*** 0.068*** 0.070*** 0.066*** 0.067*** 0.072*** 0.071*** 
(0.047) (0.045) (0.045) (0.045) (0.045) (0.022) (0.022) (0.022) (0.022) (0.022) (0.023) (0.022) (0.022) (0.022) (0.022) 
Age Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Education Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Religion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Ethnicity FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Household size Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Longitude/latitude Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Community-level controls 
 Geography and climate Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Other Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Round FE      Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
R2 0.313 0.313 0.312 0.312 0.311 0.182 0.182 0.183 0.182 0.182 0.183 0.183 0.184 0.183 0.182 
No. of obs. 6,702 6,702 6,702 6,702 6,702 32,761 32,761 32,761 32,761 32,761 30,911 30,911 30,911 30,911 30,911 
Dependent variable One if migrant 
Data sources IHS MDHS matched with IHS 
Sample Females who resided in all the MDHS communities Females who resided in the MDHS communities with distance to the nearest IHS communities <10 km 
(a) (b) (c) (d) (e) (f) (g) (h) (i) (j) (k) (l) (m) (n) (o) 
Matrilineal descent community × A community's distance (km) to the nearest town having population >20,000 
 1st quantile 0.138 0.113 0.056 0.033 0.056 0.132*** 0.109*** 0.120*** 0.062*** 0.079*** 0.123*** 0.106*** 0.113*** 0.055** 0.076*** 
(0.098) (0.081) (0.066) (0.045) (0.043) (0.040) (0.037) (0.033) (0.023) (0.022) (0.042) (0.038) (0.035) (0.024) (0.023) 
 2nd quantile 0.162** 0.141** 0.062 0.065*  0.120*** 0.132*** 0.080** 0.042**  0.106** 0.127*** 0.081** 0.041**  
(0.074) (0.066) (0.057) (0.036)  (0.040) (0.036) (0.033) (0.017)  (0.042) (0.038) (0.034) (0.018)  
 3rd quantile 0.264*** 0.198*** 0.005   0.092** 0.064* 0.011   0.087** 0.064* 0.013   
(0.073) (0.070) (0.053)   (0.038) (0.035) (0.028)   (0.039) (0.036) (0.030)   
 4th quantile 0.204*** 0.103*    0.043 0.010    0.036 0.011    
(0.074) (0.053)    (0.033) (0.030)    (0.035) (0.031)    
 5th quantile 0.179***     0.035     0.025     
(0.058)     (0.033)     (0.035)     
A community's distance (km) to the nearest town having population >20,000 
 1st quantile −0.172* −0.105 −0.123* −0.029 −0.092** −0.048 −0.028 −0.018 0.019 −0.046** −0.045 −0.032 −0.011 0.031 −0.040** 
(0.096) (0.084) (0.070) (0.048) (0.042) (0.032) (0.030) (0.028) (0.019) (0.018) (0.034) (0.032) (0.030) (0.020) (0.019) 
 2nd quantile −0.144* −0.128* −0.155*** −0.067**  −0.067** −0.070** −0.053* −0.029*  −0.054 −0.067** −0.054* −0.022  
(0.074) (0.068) (0.057) (0.033)  (0.033) (0.030) (0.028) (0.015)  (0.035) (0.032) (0.029) (0.015)  
 3rd quantile −0.228*** −0.177*** −0.068   −0.059* −0.022 0.005   −0.052 −0.030 −0.001   
(0.066) (0.059) (0.044)   (0.031) (0.029) (0.024)   (0.033) (0.031) (0.025)   
 4th quantile −0.116** −0.140***    −0.018 0.005    −0.017 −0.003    
(0.059) (0.043)    (0.027) (0.025)    (0.028) (0.026)    
 5th quantile −0.133***     −0.027     −0.021     
(0.048)     (0.027)     (0.028)     
Matrilineal descent −0.126** −0.084* 0.002 0.000 0.005 −0.062** −0.055** −0.044* −0.028 −0.031* −0.056* −0.056* −0.045* −0.027 −0.032* 
(0.055) (0.049) (0.047) (0.039) (0.039) (0.029) (0.028) (0.025) (0.020) (0.018) (0.030) (0.029) (0.026) (0.020) (0.019) 
Dual descent 0.023 0.011 0.018 0.015 0.017 0.065*** 0.061*** 0.062*** 0.067*** 0.068*** 0.070*** 0.066*** 0.067*** 0.072*** 0.071*** 
(0.047) (0.045) (0.045) (0.045) (0.045) (0.022) (0.022) (0.022) (0.022) (0.022) (0.023) (0.022) (0.022) (0.022) (0.022) 
Age Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Education Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Religion FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Ethnicity FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Household size Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Longitude/latitude Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Community-level controls 
 Geography and climate Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
 Other Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Round FE      Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
R2 0.313 0.313 0.312 0.312 0.311 0.182 0.182 0.183 0.182 0.182 0.183 0.183 0.184 0.183 0.182 
No. of obs. 6,702 6,702 6,702 6,702 6,702 32,761 32,761 32,761 32,761 32,761 30,911 30,911 30,911 30,911 30,911 

Notes: (1) Figures () are standard errors. ***Significance at 1%, **significance at 5% and *significance at 10%. (2) Standard errors are robust to heteroskedasticity and clustered residuals within each community. (3) The geographic and climate controls contain a community-level information on climatology, landscape typology, soil and terrain, crop season parameters and GPS-based coordinates. See Appendix B for the details. (4) The ‘Other’ community controls are sourced from the IHS and contain characteristics identified at the point of the survey. They include (i) a community's population; (ii) a dummy for a major urban centre; (iii) dummies for the most common religion practiced in a community, i.e., Christian, Muslim or Traditional (reference group) and (iv) dummies for the most common language spoken at home in a community, i.e., Chewa, Lambya, Lomwe, Ngoni, Nkhonde, Nyakyusa, Nyanja, Sena, Senga, Sukwa, Tonga, Tumbuka Yao and other (reference group). (5) In columns (a)–(e), the completed level of education is measured by categorical variables of none (reference group), Primary School Leaving Certificate (PSLC), Junior Certificate Examination (JCE), Malawi School Certificate of Education (MSCE), non-university diploma, university diploma/degree and post-graduate degree. On the other hand, the estimations in columns (f)–(o) use a continuous measure of the completed level of education (years). (6) In the IHS, an individual's ethnicity was not directly indicated by the survey responses. Thus, alternatively, the estimations in columns (a)–(e), the ethnicity is proxied by typical languages that a household head spoke at home.

Author notes

An earlier version of this paper was circulated under the title ‘Religion and Polygamy: Evidence from the Livingstonia Mission in Malawi’. I thank an editor, three anonymous referees, Marc Bellemare, James Fenske, Casper Worm Hansen, Assi Kimou, Tomohiro Machikita, Momoe Makino, Dorothy Nampota, Gil Shapira, Tsutomu Takane, Shinichi Takeuchi, Selma Telalagić and participants at the ASREC Conference 2015 (Boston), the CSAE Conference 2014 (Oxford) and the NEUDC Conference 2013 (Harvard) and seminars/workshops at GRIPS, Hitotsubashi and IDE-JETRO for insightful comments and suggestions. Financial support from the IDE-JETRO for my field trip to Machinga, Mulanje and Zomba is gratefully acknowledged. My great thanks in the field trip go to McDonald Chitekwe, James Mkandawire (Invest in Knowledge Initiative) and rural respondents in the survey. The findings, interpretations and conclusions expressed in this paper are entirely those of the author and do not represent the views of the IDE-JETRO. All errors are my own.