2.1 The missionary roots of modern medical provision in sub-Saharan Africa

The history of modern medicine in sub-Saharan Africa is closely linked to the development of missionary activity. Missionaries were among the first to provide modern health care until the middle of the 20th century [Vaughan (Reference Vaughan1991, Ch. 3)]. Missionary work was often conducted by doctors or nurses. David Livingstone—probably the best-known figure in the history of African missionaries—was a doctor himself. Although Livingstone's project was the promotion of trade with local populations as means to improve development and end the slave trade, he emphasized the crucial role of health and education because “neither civilization nor Christianity can be promoted alone. In fact, they are inseparable.” “Civilization,” roughly understood as the advancement of economic prosperity and Western values of which Western medicine was an essential part, was thus considered a central element of the evangelical mission from an early period.

From the end of the 19th century to the early 20th century, there was a rapid increase in medical work among missionaries. The Edinburgh Medical Missionary Association was founded in 1841. In 1891, the Church Mission Society established a new “Medical Committee” that was specifically in charge of managing the medical work of the society. Medical missionaries had a central role in the provision of health care through the construction of hospitals, dispensaries, the provision of midwifery services, the treatment of leprosy, vaccination campaigns, and the training of lay health workers. Although the colonial methods employed, such as the segregation of lepers in asylums, are now questioned, the case remains that an important part of the provision of health care was initiated by missionaries. Moreover, secular investments in health partly relied on the Christian missionary networks. For a more detailed history, refer to Vaughan (Reference Vaughan1991, Ch. 5).

The early Christian provision of health care thus persisted after colonization and has been particularly influential in the design of contemporary health care in low-income countries [Idler (Reference Idler2014)]. In 1968, for instance, the World Council of Churches established the Christian Medical Commission (CMC), institutionalizing the long-lasting tradition of Christian medical and healing work around the world.Footnote ¹³ In 2001, the World Christian Encyclopedia inventoried 37 different Christian societies working on the provision of health care in sub-Saharan Africa (unfortunately, more detailed statistics are not reported). The Christian Medical Fellowship (CMF), a modern medical Christian society, also illustrates the extent of the Christian influence in contemporary medical provision. It inventories on its website 64 Christian partner hospitals,Footnote ¹⁴ many of them with satellite institutions and mobile clinics. In countries such as Malawi or Rwanda, the CMF's partner institutions alone account for up to 9% of the country's total hospital beds.

2.2 The Christian roots of modern health provision

Missionaries' central objective was to convert the local population to the Christian faith, and to some extent, missionary medicine was considered by some as instrumental toward this goal. The notion of healing the body and soul is indeed deeply rooted in the Christian dogma. For instance, the Gospel of Matthew 10:1 reads: “Jesus called his twelve disciples to him and gave them authority to drive out impure spirits and to heal every disease and sickness.” Medical work—a form of “healing authority”—was “part of a programme of social and moral engineering through which Africa would be saved” [Hardiman (Reference Hardiman2006)]. The following excerpt from Vaughan (Reference Vaughan1991, p. 72), narrating an encounter between a missionary doctor and a traditional healer, is also illustrative of how medical missionary work was sometimes conceived as embedded in the evangelization work, rather than as an independent scientific activity.

Therefore, when studying the legacy of missionary activity on health outcomes, we need to bear in mind that missionary medicine was a part of a larger conversion enterprise. The history of the early colonial and missionary campaigns targeting STDs illustrates the tension between the Christian and the scientific response to these diseases. In Uganda, there was a large increase in the prevalence of syphilis from the mid-19th century, which led to the first public health campaigns launched in sub-Saharan Africa by the colonial powers [Setel (Reference Setel1999, p. 99)].Footnote ¹⁵ The priority of the campaigns was not the treatment of patients or the diffusion of scientifically accurate information, but the fight against what colonial authorities considered to be the innate sinfulness of the African society [Vaughan (Reference Vaughan1991)]. This moral combat was in turn supposed to eradicate the “disease of immorality,” the term used by missionary doctor Albert Cook to refer to syphilis in Uganda [Vaughan (Reference Vaughan1991, p. 138)]. The interventions by colonial powers, missionaries, and local authorities to fight the crisis took different forms, but they always implied the necessity to change sexual behaviors by incentivizing monogamous marriages at a younger age. Missionaries campaigned for the recognition of customary law marriages in 1918, capping the levels of required bride wealth, banning divorce for reasons other than adultery, and imposing restriction on women's mobility [Doyle (Reference Doyle2013)].

However, the prioritizing of morality over medical responses cannot be generalized to all missionaries. A revealing example can be found in the archives of the Society for the Propagation of the Gospel (SPG).Footnote ¹⁶ Around 1907, as a response to the SPG's inquiry on the needs of their missionaries in the field, Reverend Lattimer Fuller, a missionary doctor in South Africa, replied with a letter that reads rather like a public health report than a moralizing pamphlet. Without references to a moralizing enterprise, he simply emphasizes the need for dispensaries (“cottage hospitals”) and funds to train personnel, to contain a growing epidemic of syphilisFootnote ¹⁷:

The role of medical missions in the development of Western medicine in sub-Saharan Africa, beyond its instrumental value for conversion, is also emphasized in the historical work on Protestant and Catholic medical missions. For instance, Good (Reference Good1991) writes: “The idea that medical care for Africans should be subordinate to the goal of spiritual salvation—especially strong among Protestant missions—gradually tempered to accommodate the practical view that the treatment of disease, relief of suffering, and training of Africans as medical auxiliaries constituted worthy ideals of Christian service in their own right.”

Medical missionaries, in fact, were extremely active in the diagnosis and treatment of many conditions, in particular STDs. Exploring the specific case of the Uganda mission-hospital, Doyle et al. (Reference Doyle, zu Selhausen and Weisdorf2018) present the continuous efforts to diagnose and treat syphilis and other related STDs both for Christians and non-Christians, throughout the colonial period and after independence, when the missionary hospital was nationalized. The missionary doctors' efforts in the containment of STDs are likely to have mattered, as STD-related genital ulcers have been shown to be catalyzers of the early spread of the HIV epidemic [de Sousa et al. (Reference de Sousa, Müller, Lemey and Vandamme2010)].

Today, medical missionaries continue to have an active role in the HIV-prevention campaigns, even when the messages diffused conflict with Christian dogmas. For instance, in 2009, the United Church of Christ HIV Ministry Network released a statement recommending that houses of worship and faith-based education institutions distribute condoms and provide comprehensive sexual education. Other HIV-relief programs also run by faith-based organizations are less clear on the importance of distributing condoms.Footnote ¹⁸ In this paper, we capture this heterogeneity by comparing the effect of missions that specialized in health provision to those that did not. We show indeed those only regions around missions that invested in health exhibit lower HIV prevalence today.

2.3 Missions, Christianity, and HIV

The history of African Christianity, and the way it shaped the HIV epidemic from its very roots, goes beyond the historical and persistent provision of public health. In this section, we show that there is both historical and contemporary evidence suggesting that early exposure to Christianity could have affected HIV prevalence. This evidence provides grounds for our investigation into the long-term consequences of missionary presence on HIV.

Recent research at the intersection between history and biology has shown that the origins of HIV in Africa can be traced back to the first half of the 20th century. The evolution of HIV from a successful cross-species contagion of SIV (Simian Immunodeficiency Virus, the “version” of HIV that contaminates monkeys) into a large-scale human epidemic is tightly linked to colonial history, and the way it brought up a large-scale rapid cultural and urban transformation. The 1920s are the estimated timing of the group-0 cross-species transmission between apes and humans [Hemelaar (Reference Hemelaar2012)]. The disease likely diffused after one or several incidents of cross-species contamination, around the Congo-river basin in the 1930s and 1940s. A single diffusion cause is impossible to identify, but it is believed that the contagion was made possible as a result of colonial transformations, which changed local demographic structures, consolidated a commercial sex work industry, and facilitated the spread of diseases through poorly organized health campaigns that heavily relied on injectable medicines [Pépin (Reference Pépin2011), Lowes and Montero (Reference Lowes and Montero2017b)].

Large scale public infrastructure projects, such as the Congo-Océan railway project (1921–1932), and rapid urbanization, consolidated population centers with skewed sex ratios. In these population centers, the commercial sex work sector grew rapidly. Sex workers were highly prone to STDs leading to genital ulcers, which in turn highly increase the contagion risk of HIV [de Sousa et al. (Reference de Sousa, Müller, Lemey and Vandamme2010)]. These centers were also frequently exposed to public health campaigns where frequent use of unsterilized needles facilitated iatrogenic diffusion.

By diffusing a religion that promoted changes in sexual behaviors and family structures, missionaries may have affected HIV since its very origins. Pépin (Reference Pépin2011, p. 98), for instance, gives a glimpse on how missionaries, who actively discouraged polygamy and tried promoting monogamous marriages, may have contributed to the spread of the disease:

It is not necessary to go as far as spanning a causality chain from Christianity to the sex industry to understand that conversion to Christianity may have affected HIV prevalence. For instance, missionaries devoted substantial efforts into diminishing polygyny, a marriage structure that today is associated with lower HIV prevalence, because it creates “small isolates of concurrent partnerships in which the virus is trapped” [Reniers and Watkins (Reference Reniers and Watkins2010)]. This network is less likely to spread HIV than one where both men and women have multiple and changing partners.Footnote ¹⁹

Today, the focus of campaigns against HIV is often on teaching the ABC approach: “Abstain, Be Faithful, Condomize.” But while religious groups have been open to teaching this message, there is qualitative evidence that the “Condomize” section is often ignored. Many churches—in particular Pentecostal churches—indeed consider the use of condoms to be a sin [Mash and Mash (Reference Mash and Mash2013)]. Religious groups are also heavily involved in the design of sex-education curricula delivered in many public schools. According to Duflo et al. (Reference Duflo, Dupas and Kremer2015), in Kenya as well as in many other sub-Saharan African countries, the resulting curriculum teaches the biology of AIDS and HIV transmission, how to care for people living with AIDS, and prevention. The prevention component stresses abstinence before marriage, followed by marital faithfulness as the most effective ways to prevent STDs. The official textbook does not mention condoms or contraception.Footnote ²⁰

Yet, it has been shown in the literature that HIV curricula stressing abstinence before marriage do not reduce risky sexual behaviors, whereas those that do explain the use of condoms are effective in the reduction of STDs [Dupas (Reference Dupas2011), Duflo et al. (Reference Duflo, Dupas and Kremer2015), Dupas et al. (Reference Dupas, Huillery and Seban2018)].

3.1 Historical data

Different sources were used to construct the dataset of missions and historical controls. We describe them in turns in this section.

3.1.1 Protestant missionary investments

Regarding Protestant missions, we originally compiled the data of missionary investments and their locations in a previous paper [Cagé and Rueda (Reference Cagé and Rueda2016)]. We constructed the mission-level data and geocoded the maps of sub-Saharan African regions from the Geography and Atlas of Christian Missions [Beach (Reference Beach1903)]. The maps locate all the Protestant mission stations in 1903 (an example of these maps is provided in Figure 2b).

Figure 2. Plates representing all (Catholic and Protestant) missionary stations. Note: The figure reproduces a plate locating Catholic missionary stations (top) and a plate locating Protestant missionary stations (bottom).

As opposed to other available geographic datasets of Protestant missions [Akyeampong et al. (Reference Akyeampong, Bates, Nunn and Robinson2014, Ch. 16)], the one generated for this paper contains detailed information for each mission settlement. In the Geography and Atlas of Christian Missions, each mission station is uniquely identified in a statistical appendix, which provides information on the mission's size (number of students, of missionaries, etc.) and a detailed record of its activities and investments. For example, we know whether each mission had a printing press, a school, or a health facility. The exhaustive list of variables and a reproduction of one page of the statistical index are provided in the online Appendix to Cagé and Rueda (Reference Cagé and Rueda2016). Our sample of sub-Saharan African missions includes a total of 723 Protestant missions out of which 99 had a dispensary, and 61 had a hospital or a declared doctor.

3.1.3 Determinants of missionary location and investments

Figure 3 reports the historical location of both the Catholic and Protestant missions as well as their health investments. To the extent of our knowledge, we are the very first to produce such a dataset. Overall, we provide information for a total of 1,259 missions.

Figure 3. Catholic and Protestant mission stations with and without health investments. Note: This map is a digitized and geocoded version of two geography atlas. For Protestant missions, we used digitized and geocoded plates 14 to 18 of Dennis, Beach, and Fahs (1903), from Cagé and Rueda (Reference Cagé and Rueda2016). Health investments include dispensaries and hospitals. For Catholic missions, the map is a digitized and geocoded version of the 1929 Atlas Hierarchicus. The geocoding was conducted by the authors. Health investments include hospitals.

We rely on the aforementioned atlases to construct our data because they are, to the best of our knowledge, the only systematic records of missions and their investments for the entire continent. However, recent research focused on Ghana has shown that this sample is more likely to capture the earlier and better endowed European missions than the later missions led by Africans, which settled in less favored regions, further from the coast, and more suitable to malaria [Jedwab et al. (Reference Jedwab, zu Selhausen and Moradi2018)]. Although it is difficult to extrapolate from the unique case of Ghana, we acknowledge that our approach captures the effect of the European missionary expansion. Nonetheless, we want to highlight that our within-mission approach restricts the analysis to regions that are close to these European settlements and thus locally comparable. In light of Jedwab et al. (Reference Jedwab, zu Selhausen and Moradi2018), a question that arises is whether our effects are mostly channeled through the later expanding network of local missions. This could be the subject of further research that would suppose the collection of more extensive data, in particular with a larger geographical reach. It is, however, beyond the scope of this paper.

According to this existing literature, European missionaries were disproportionately more likely to locate in geographically favored areas [Johnson (Reference Johnson1967), Nunn (Reference Nunn2010), Akyeampong et al. (Reference Akyeampong, Bates, Nunn and Robinson2014), Cagé and Rueda (Reference Cagé and Rueda2016), Jedwab et al. (Reference Jedwab, zu Selhausen and Moradi2018)]. In the online Appendix summary statistics Table C.3, we perform a t-test on the equality of means for geographic and historical characteristics of towns located near (less than 100 km) and far (more than 100 km) from the historical mission settlements in our sample. The results confirm that missionary locations are more geographically favored regions with higher precolonial and subsequent population densities.

Due to potential selection in missionary location, our specifications always control for all the geographic and historical characteristics that may have influenced location choice, such as the malaria ecology, proximity to the coast, suitability for rainfed crops, suitability to Tse-Tse fly, and proximity to historical rail lines and explorer routes. More importantly, following Cagé and Rueda (Reference Cagé and Rueda2016) our results on the effects of health investments are carried “within missions,” that is to say that we compare within our sample, missions that invested in health with missions that did not.Footnote ²²

Moreover, as we compare missions depending on whether they invested in health, a second concern would be that of endogenous selection into health investments. Table 1 compares the geographic and historical characteristics of missions that did invest in health with those that did not. Protestants and Catholics are analyzed separately. Although the difference for most geographic indicators is not statistically significant, we observe that missions that invested in health facilities were, on average, located in regions that were slightly cooler, closer to the coast (not statistically significant), with lower elevation. The most important difference is pre-colonial population density, which is (at least) twice is high around missions with health facilities. The patterns observed suggest that missions with health investments were set up in more favorable areas, with larger population densities. We also know that urbanization is one of the most significant determinants of HIV prevalence [Fox (Reference Fox2010), Oster (Reference Oster2010), Pépin (Reference Pépin2011)]. In most sub-Saharan African countries, the ratio of HIV prevalence is more than twice as high in cities than in rural areas [UNAIDS (2015, Figure 5)]. In a country like Côte d'Ivoire, 23% of the national population lives in Abidjan, but the city accounts for 30% to 50% of the country's population infected with the disease [UNAIDS (2015)]. Therefore, if geographically confounding factors were driving our results, we would expect these locations to have larger, and not smaller HIV risk.

Table 1. Determinants of the location of the health investments

Note: *p < 0.10, **p < 0.05, ***p < 0.01. The table compares the characteristics of the places where missions with and without a health investment did locate. Panel A presents the results for Protestant missions, and Panel B for Catholic missions. In both panels, column (1) presents the results for missions without a hospital. Column (2) presents the results for missions with a hospital. In column (3), we perform a t-test on the equality of means (robust standard errors are in parentheses). Variables are described in the online Appendix.

Finally, since pre-colonial characteristics also shaped colonial and post-colonial development outcomes [Michalopoulos and Papaioannou (Reference Michalopoulos and Papaioannou2013)], Table 2 compares pre-colonial marital structures of the missions with and without a health investment. The rationale for this comparison is that these structures shape sexual behaviors that can ultimately determine HIV patterns. We use data from the Ethnographic Atlas, and do not observe statistically significant differences in pre-colonial marriage patterns and sexual practices linked to the location of the health investments.

Table 2. Determinants of the location of the health investments: precolonial marital structures

Note: *p < 0.10, **p < 0.05, ***p < 0.01. The table compares the characteristics of the places where missions with and without a health investment did locate. Panel A presents the results for Protestant missions, and Panel B for Catholic missions. In both panels, column (1) presents the results for missions without a hospital. Column (2) presents the results for missions with a hospital. In column (3), we perform a t-test on the equality of means (robust standard errors are in parentheses). The variables are from the Ethnographic Atlas.

3.2 Contemporary data

Our analysis relies on the individual-level data from the DHS distributed by USAID. The DHS are standardized nationally representative surveys in developing countries. Women and men answer questions related to their anthropometrics, living conditions, health, sexual reproduction behaviors, and attitudes and beliefs regarding different health outcomes.

Since the 2000s, the DHS also collect biomarker data on STDs. Biomarkers are objective physical or biologic measures of health conditions. Using field-friendly technologies, the DHS surveyors can collect biomarker data on HIV.Footnote ²³ Hence, our measure of HIV infection is based on actual HIV contamination information, not on self-reported surveys where the mean rates of HIV infection are likely to be severe underestimates.Footnote ²⁴ The DHS data also provide information on anemia for women; just like for HIV, biomarker testing is used for anemia, making them comparable. We use anemia in section 5 to show that our result are specific to STDs (i.e., not driven by poorer health conditions in general).

We use the DHS data for sub-Saharan African countries where we have at least one missionary hospital reported, and where both GPS information and biomarker data on HIV are provided. Figure 4a shows the locations of these clusters. There are a total of 344,212 individuals, distributed in 8,498 clusters distributed among 17 countries: Burkina Faso, Burundi, the Democratic Republic of Congo (DRC), Cameroon, Ethiopia, Gabon, Guinea, Kenya, Liberia, Lesotho, Malawi, Rwanda, Senegal, Swaziland, Uganda, Zambia, and Zimbabwe. The information is collected in several rounds, from 2003 to 2014. The surveys are not systematically carried out for all countries every year, so each country appears 1–3 times in the sample.Footnote ²⁵

Figure 4. DHS locations. Note: The figure shows the location of DHS clusters in sub-Saharan Africa that report both HIV biomarker data and GPS information.

The DHS now also releases geospatial covariates for all the clusters for which it provides GPS coordinates. We rely on these data to obtain geocovariates that are standard in the literature. For each location, we observe land characteristics (drought episodes, aridity, rainfall, temperature, malaria ecology, etc.), average night-time luminosity, and proximity to water sources, among others. Importantly, this source gives information on the average travel time to a major urban center. This variable captures urbanization and also connectedness to the infrastructure network, which is of particular interest to us, as Oster (Reference Oster2010) shows that road density is an important determinant of HIV prevalence.

4.1 Specification

To examine the correlation between proximity to missions (with and without health investments) and HIV/AIDS prevalence, we estimate the following equation:

(1)$$y_{irct} = {\bf Dis}_r^{\prime} \beta + {\bf X}^{\prime}_{irt} \gamma _1 + {\bf W}^{\prime}_r \gamma _2 + \zeta _t + \theta _c + \varepsilon _{irct}$$

where y _irct is a binary variable equal to one if the respondent i is tested positive for HIV, in the DHS cluster r, in country c, and year t, and to zero otherwise. In the remainder of the paper, the DHS clusters are referred to as “towns.” ${\bf Dis}_r$ is a vector of distances. It includes the explanatory variables of interest: the logarithm of the distance between town r and the closest historical mission settlement, and the logarithm of the distance between town r and the closest mission with a health investment (a hospital or a dispensary). Online Appendix Table C.1 presents summary statistics on these distances.

The vector ${\bf X}_{irt}$ contains individual level controls that are not “bad controls” in the sense of Angrist and Pischke (Reference Angrist and Pischke2009): sex, age, age square, and marriage status. Our results are robust for introducing a larger set of controls (accounting for education level status or wealth indicators, see section 4.2). ${\bf W}_r$ contains geographical and historical-level controls for town r in country c. The geographical controls are distance to the capital city; distance to the coast; the average malaria ecology of the land; the slope of the terrain; the elevation; temperature per month, rainfall per month, aridity, and the length of the growing season in days (these are all from the DHS geocovariates set). We also include the Tse-Tse fly suitability index [Alsan (Reference Alsan2015)]. The historical controls are the distances to cities in 1,400 and 1,800; and distance to colonial railways and to initial explorer routes [Nunn (Reference Nunn2008)]. Because they are potential “bad controls,” our baseline specification does not include urbanization proxies, in particular, the average travel time to an urban center that captures both urbanization and connectedness into the transport infrastructure. However, our results are robust to including these controls, and also to restricting the analysis to the most connected areas (see section 5).

Finally, ζ and θ are year and country fixed effects. Unless otherwise stated, all the regressions' standard errors are clustered at the town level, and all the regressions restrict the analysis to towns that lie within a 100 km buffer from a historical missionary settlement (depending on the specification, this is any settlement, a Protestant settlement, or a Catholic settlement).

5.1 Consent to test

Our estimates rely on the availability of HIV test results. One potential issue is that regions with higher mistrust in medicine will have fewer individuals consenting to HIV tests, thus lowering the HIV-positive results. Given the persistence of mistrust in medicine in regions where colonial medical campaigns were established [Lowes and Montero (Reference Lowes and Montero2017b)], lack of consent in regions close to missions could pose a potential threat to the estimation. Table C.12 in the online Appendix shows that proximity to a missionary station with a health investment is neither a strong nor a statistically significant predictor of consent to HIV testing. If anything, we observe that proximity to a hospital increases the consent to test, but this result collapses and loses statistical significance when decomposing the analysis separately for Protestant and Catholic missions.

5.2 Precolonial marriage patterns

Traditional marriage patterns matter in the understanding of the HIV epidemic. In particular, the literature has shown that polygamy is associated with a decrease in HIV prevalence, as it tends to “shrink the sex networks” compared to sequential monogamous relationships or non-faithful monogamous relationships (the last are common in our sample) [Reniers and Watkins (Reference Reniers and Watkins2010)]. It would be problematic if we found that there were different pre-colonial marriage patterns in regions close to the missions of interest. First, we can remark that most of the sample in the Ethnographic Atlas reports some form of polygamy [Murdock (Reference Murdock1967)]. Second, Table 2 shows that the Ethnographic Atlas' groups closest to each mission report no statistically significant differences in precolonial sexual and marriage patterns between missions that invested in health and those that did not.

5.3 Education

The long-lasting effects of missionaries on education have been studied thoroughly [Woodberry (Reference Woodberry2004), Acemoglu et al. (Reference Acemoglu, Gallego and Robinson2014), McCleary (Reference McCleary2015), Valencia-Caicedo (Reference Valencia-Caicedo2018)]. If individuals living in regions close to missions with health investments tend to be more educated, then our results could be driven by human capital rather than by differences in sexual behaviors. Although the existing literature has found no systematic relation between education (in general) and STDs at the micro and short-term levels [Duflo et al. (Reference Duflo, Dupas and Kremer2015)], we cannot rule out the role of education in the long run. Table C.13 in the online Appendix reports the results of estimating equation (1) when the dependent variable is the number of years of education. As expected, proximity to a mission is correlated with more education. Importantly, we find no statistically significant relationship between proximity to missions that invested in health and the number of years of schooling. It is thus unlikely that all of our effect would be driven by human capital accumulation.

5.4 Non-sexually transmitted diseases and access to health care

This section explores the possibility that our results could be unspecific to STDs, and capturing other factors that would correlate with systematically worse health around regions where missions settled, in particular Protestant ones.

The DHS do not provide test results for other STDs, but do report other health-related outcomes, mostly for women. The DHS female dataset indeed includes results for anemia tests and a few anthropometric measures. Tables C.16 and C.17 in the online Appendix report the results of estimating equation (1) when the dependent variable is a binary variable equal to one if the respondent is either anemic or has stunted growth, and to zero otherwise.

Regarding proximity to missions, the results show that, unlike for HIV, proximity to a mission does not statistically significantly correlate with these health outcomes. If anything, we observe improved outcomes (less positive results of anemia or stunted growth), but the relationship is not significant. Therefore, it is unlikely that our results capture confounders associated with worse health outcomes around missions. Also, we do not find evidence for conflicting effects between proximity to missions and other health outcomes.

Proximity to a health investment is also associated with improved outcomes in particular regarding the likelihood that maternal care is delivered by medically trained specialists (see Table C.14 in the online Appendix). These outcomes go in line with better access to health facilities, a channel that is explored in further detail in the next section. Regarding other diseases, such as anemia or anthropometric measures, the results usually do not meet statistical significance. A possibility is that we do not observe enough variance within regions close to missions on these outcomes because there are no conflicting effects of missionary influence on non-STDs. Similarly, it is possible that better access of health infrastructure alone is not sufficient to improve all health outcomes. Our results would then have to be considered in combination with the safer sexual behaviors observed around missions with health facilities (see section 6). This result would be consistent with the results of Calvi and Mantovanelli (Reference Calvi and Mantovanelli2018) in the case of India.

5.5 Central equatorial Africa and the rest

The recent biological history of HIV has emphasized the likely role of health institutions in the early spread of the epidemic in the equatorial central African region, especially around the Congo river. The widespread use of non-sterilized needles in mass health campaigns to treat sleeping sickness, leprosy, and other diseases during colonial times is likely to have been a key vector of transmission for the disease in its early phase. This transmission is what catalyzed the formation of a large-scale epidemic [Pépin (Reference Pépin2011), Lachenal (Reference Lachenal2014), Lowes and Montero (Reference Lowes and Montero2017a)]. For this reason, proximity to a health mission, where many of these campaigns were also organized, might not have had a positive effect in Central African countries. We examine whether this is the case by considering separately the DRC and Gabon, the only countries in our sample where this early history of the epidemic might apply, and the rest of the sample.

Table C.18 in the online Appendix presents the results. Columns (1) to (3) exclude the DRC and Gabon and columns (4) to (6) restrict the analysis to the DRC and Gabon. The previously described results does not hold for Central Africa, with no statistically significant effects. Columns (4) to (6) show the same analysis, but for non-Central African countries. First, the results for the sample excluding the DRC and Gabon are similar—both in terms of statistical significance and of magnitude—to the ones presented in Table 3. Second, there is no statistically significant effect of proximity to a health station for DHS locations in the DRC and Gabon. Our results are thus consistent with the biomedical and economic history literature that emphasize the ambiguous effect of colonial medical campaigns on health and trust and medicine around the Congo river.

5.6 Controls

We show that our results do not depend on the set of controls included. Online Appendix Table C.21 shows that our results are robust to a specification that only includes year and country fixed effects. In this specification, the effect of distance to a health investment remains positive, statistically significant, and of the same order of magnitude as in our baseline set of results. There is a change in the effect of distance to missions, that become larger in absolute terms, negative and statistically significant for all missions, as opposed to just for Protestant.

5.7 Urbanization

In sub-Saharan Africa, HIV is a disease that is mostly present in urban and well-connected areas where large numbers of people transit [Pépin (Reference Pépin2011), Oster (Reference Oster2012)]. Given the large evidence of the long-term legacy of missions on long-term development, there is a possibility that our effects, in particular the long-term worsening of HIV around Protestant missions, are entirely driven by modernization.

First, we address this issue regarding missions that invested in health. The new geocovariates in the DHS (available since 2017) report the traveling connecting time to an urban area for all the geocoded DHS clusters. This variable captures both urbanization and connectedness into the transport infrastructure network, which are the key drivers of the relationship between HIV and modernization. Table C.19 in the online Appendix shows that distance to a health investment is associated with an increase in travel time.Footnote ²⁶ Therefore, if modernization was the unique channel at play, we would expect proximity to a health investment to be associated with larger, and not lower HIV prevalence.

Table C.20 in the online Appendix reports the result of estimating our baseline equation, but in a sample restricted to the 10% most urbanized and connected clusters in the DHS. It shows that our results hold when applying this restriction. In this sample, proximity to a mission, both Catholic and Protestant, is associated with increased chances of HIV; whereas proximity to a health investment is associated with lower chances. Therefore, it seems plausible to think that urbanization and modernization are not completely driving the result.

Modernization is a channel that we acknowledge in the interpretation of our results, in particular the long-term worsening risk of HIV close to Protestant missions. The relationship between HIV and modernization has been highlighted in the literature [Fox (Reference Fox2010), Pépin (Reference Pépin2011), among others]. Our results can then be seen as further quantitative evidence for this colonial origin of the disease, thus adding nuance to the view that the missionary legacy was solely positive for development. However, within urbanized areas, we still observe are long-term effects of missionary presence (see table C.20 in the appendix). The goal of section 6 is to explore further channels of persistence, in particular relating to conversion to Christianity, sexual behaviors, and the persistence of health infrastructure.

6.1 The persistent effect of missionary health care

In section 4.2, we have showed that within regions close to missions, proximity to a mission that invested in health care is associated with lower HIV prevalence, and it is unlikely that this relationship is fully driven by unobservables. In this section, we wish to understand the channels through which such correlation can operate.

6.1.1 Infrastructure in the medium term

To understand whether the positive relationship between distance to missionary health stations and HIV prevalence can be driven by persistence in health infrastructure, we collect information on the location of British colonial hospitals up to the mid-1920s from the British African Blue Books. This information is particularly helpful to understand whether our results for the proximity with missionary health stations are driven by persistence in health infrastructure, such as missionary hospitals. The focus here is on Protestant health stations given that the information we have on hospitals in the 1920s is restricted to former British colonies.

The persistence of healthcare infrastructures may be an important explanatory channel given that HIV prevention policies are implemented in these infrastructures. Moreover, other health policies that can decrease the HIV risk, e.g., the provision of treatment for other, treatable, STDs, in particular STDs generating genital ulcers, take place in these hospitals. Protestant and colonial health centers devoted substantial resources for the treatment and the prevention of syphilis [de Sousa et al. (Reference de Sousa, Müller, Lemey and Vandamme2010), Doyle et al. (Reference Doyle, zu Selhausen and Weisdorf2018)].Footnote ²⁷

First, to capture the persistence of missionary health stations in the medium term, we want to gage whether the health facilities reported are also entered as health facilities in the colonial statistics. This could arrive for the larger hospitals, if they started receiving colonial funds or even became colonial public services [Good (Reference Good1991), Doyle et al. (Reference Doyle, zu Selhausen and Weisdorf2018)]. The first row in Table 5 gives the average number of colonial health facilities reported in the Blue Books inside a 10 km buffer of mission stations. The numbers are larger, albeit no statistically significantly so, around missions with health facilities.

Table 5. Presence of health services around missions with and without health facilities

Note: This table presents the average number of colonial hospitals (“Hospitals in the Blue Books”) and health services today (“Facilities in SPA”) inside a 10 km buffer of missions with no health facilities [Column (1)] and missions with health facilities [Column (2)]. Samples differ. Only countries in the British Empire are considered for the analysis with the Blue Books. Data availability constraint the analysis for the SPA to Kenya, Malawi, Rwanda, and Sénégal.

Second, Table 6 records the results from the same analysis ran in Table 3, but replacing the distance to missionary health stations with the distance to any health center (missionary or colonial). Because the data on the location of colonial hospitals comes from the British Blue Books, the sample is restricted to former British colonies. Columns (1) and (2) show that the results are similar to those in our baseline regression. The results for Catholic health stations are not significant, which can be due to the sample restriction.

Table 6. Persistence in health care: effect of missions and colonial health investments

Note: *p < 0.10, **p < 0.05, ***p < 0.01. The table reports OLS estimates. The unit of observation is the individual. The dependent variable is a binary variable equal to one if the respondent is HIV positive, and zero otherwise. Standard errors in parentheses are clustered at the closest mission level. All regressions restrict the analysis to regions at least 100 km close to missions. Controls are described in the text. The regressions are restricted to former British colonies.

6.1.3 Sexual behaviors

To capture the extent of risky sexual behaviors and general knowledge about prevention, we present results on six different outcomes that capture two important dimensions of HIV transmission: (i) paying for sex from sex workers (for men) and (ii) monogamy and number of sexual patterns. Figure 9 reports the results in graphical form to facilitate comparisons between outcomes.Footnote ²⁸ Proximity to a mission with a health station is correlated with a number of safer sexual behaviors. Individuals living in regions close to health investments declare lower use of sex workers' services, over a lifetime or in the year of the questionnaire. Sex workers have been shown to be a crucial vector of transmission of HIV [Pépin (Reference Pépin2011)].Footnote ²⁹ Similarly, we show that proximity to a mission decreases the likelihood of being in a monogamous relationship, but also decreases the number of sexual partners other than spouse or the number of sexual partners over the lifetime. This is consistent with a pattern of more isolated sexual networks in polygamous structures that are associated with lower HIV prevalence (as explained in section 2). We find no evidence suggesting that individuals around historical health facilities exhibit more usage or knowledge about condoms (see online Appendix Figure B.5). Therefore, it seems that proximity to historical health facilities is associated with safer sexual norms that lead to less HIV, rather than with the practice of safer sex.

Figure 9. Missionary investments and sexual behaviors. Note: The figures report OLS estimates. The point represents the estimated effect and the lines represent the 90% confidence interval. The unit of observation is an individual. The dependent variables vary for each graph and are reported in the legend. Standard errors are clustered at the town level. Controls are described in the text. All regressions restrict the analysis to regions at least 100 km close to missions.

These results can stem from the fact that sex prevention campaigns are run from the health facilities that have persisted (as shown in the previous section), for instance campaigns informing individuals about the risk of contracting STDs from sex workers. It could also stem from the fact that missions that specialized in health were less active in conversion and disrupted less traditional marriage patterns, in particular polygamy [Good (Reference Good1991)]. We do indeed observe that within regions close to missions, individuals are less likely to be Christian around missions that invested in health (see online Appendix Table C.28).

6.2 The persistent effect of missionary presence

The results presented in Table 3 suggest that missionary presence is consistently associated with higher HIV prevalence. This effect is mostly relevant for Protestant missionary presence. Overall, the formation of urban centers around missions and the higher prevalence of HIV in these urban centers is one channel explaining these results. The higher HIV prevalence is, in this sense, a side effect of the “modernization” that occurred in locations where missionaries settled, in particular Protestant missions.

In this section, we argue that conversion may be another important channel. In analyzing the effects of missionary proximity per se, we move away from the strict within-mission analysis that has characterized our previous research [Cagé and Rueda (Reference Cagé and Rueda2016)] and this paper's findings regarding health stations. Although the results should thus be interpreted with caution, we consider important to document conversion—a cultural channel—as an additional factor explaining the persistent effect of missionary presence on HIV.

Missions had a long-run cultural imprint. They indeed started a large-scale conversion process, and Africa exhibits today among the largest growth in Christian population. Table C.28 in the online Appendix corroborates the results by Nunn (Reference Nunn2010), who shows that Christian conversion persisted around missions.

To begin with, we compare the sexual behaviors of the Christian (Catholic and Protestant) and of the non-Christian citizens surveyed in the DHS. Table 7 presents the results. We observe that despite being more educated on average than non-Christians, Christians have riskier sexual behaviors. They have more sexual partners over their lifetime and are more likely to use the services of sex workers. Furthermore, they are also less likely to be abstinent before marriage. Despite being more likely to know that condoms lower the chances of transmitting HIV, they are less likely to know where to find them. These results also hold when restricting our sample of analysis to individuals living in urban areas (online Appendix Table C.29).

Table 7. Descriptive statistics: HIV prevalence and sexual behaviors, Christians vs. non-Christians (t-test)

Note: The table reports descriptive statistics. The unit of observation is an individual. Column (1) reports the means for non-Christians. Column (2) reports the means for Christians. Column (3) performs a t-test on the equality of means.

Focusing on conversion also helps understanding why the worsening of HIV around missions is mostly driven by Protestant missions. According to the historical literature, Catholic missionaries were less successful in their conversion enterprise. The introductions of the 1913 and the 1929 Atlas Hierarchicus state the Vatican's worries about lagging behind Protestant missions in the Empire [Streit (Reference Streit1913, Reference Streit1929)]. Catholic missions were far less successful in consolidating a local clergy, the very effective conversion army of Protestants. This endeavor was indeed particularly challenging given Catholicism's exigency of priest celibacy, and its reliance on Latin for clerical education liturgy before Vatican II in 1959 [Hastings (Reference Hastings1996), Frankema (Reference Frankema2012), Jedwab et al. (Reference Jedwab, zu Selhausen and Moradi2018)]. Moreover, Catholic missions concentrated in former French and Belgium colonies. Their religious activities were viewed with more hostility by colonial administrators in the French anticlerical political environment before World War One [Woodberry (Reference Woodberry2004), Foster (Reference Foster2013)].Footnote ³⁰

Moreover, a comparison within Christians, between Catholics and Protestants, shows that Catholics have safer sexual behaviors. These results are to be interpreted with caution as religious affiliation is not reported in a standardized way across surveys, and categories separating Catholics and Protestants are not always mutually exclusive. Table 8 presents the results. As before, Protestants are significantly more educated and there is virtually no difference regarding urban residence. Catholics exhibit certain riskier behaviors, like a larger age gap inside the household, or a larger number of sex partners. Protestants are nonetheless more likely to use the services of sex workers over their lifetime, which is a very strong determinant of HIV transmission, and less likely to know that condoms lower the chances of HIV contamination. Although it is statistically significant, the difference is quantitatively very small.

Table 8. Descriptive statistics: HIV prevalence and sexual behaviors, Catholics vs. Protestants

Note: The table reports descriptive statistics. The unit of observation is the individual. Column (1) reports the statistics for Protestants. Column (2) reports similar statistics for Catholics. Column (3) performs a t-test on the equality of means.

The statistics in Table 8 can strike as a surprise; after all, the most well-known opposition to anti-HIV policies favoring access to condoms was the Vatican's official publication “Family Values versus Safe Sex,” of which the first chapter is entitled “The Catholic Church's Criticism of the Condom in AIDS-Prevention Programmes” [López-Trujillo (Reference López-Trujillo2003)]. However, the Vatican's positions do not necessarily reflect the diversity of practices in the field. Furthermore, our data do not allow us to control for the intensity of religiosity, nor to decompose accurately the effects between different denominations of Protestants that can substantially differ in prevalence and sexual behaviors [Trinitapoli and Weinreb (Reference Trinitapoli and Weinreb2012)]. Further research could aim at understanding this particular point.

To test the plausibility of conversion as one explanatory channel, we run the same regressions as those in Table 3, and add an interaction term between distance to the mission and religion, as well as a religion binary variable. We pool all Christian denominations together into one single binary variable because, as explained before, the DHS religious categories are not standardized across waves. The religious variable allows, however, to distinguish precisely between Muslims, Christians, and Traditional religions. The results are reported in Table 9.

Table 9. Individual level OLS: distance to mission, religion interactions, and HIV status

Note: *p < 0.10, **p < 0.05, ***p < 0.01. The table reports OLS estimates. The unit of observation is an individual. The dependent variable is a binary variable equal to one if the respondent is HIV positive, and zero otherwise. Standard errors in parentheses are clustered at the town level. Controls are described in the text. All regressions restrict the analysis to regions at least 100 km close to Catholic missions [columns (1) and (3)] or Protestant missions [columns (2) and (4)].

Columns (1) and (3) show that distance to a Catholic mission does not seem to affect the probability of being tested positive for HIV for either Christians or Muslims. Similarly, the negative effect of distance to a Protestant mission is stronger for Protestants [columns (2) and (4)], and not statistically significant for Muslims.

Taken together, these results combined with the historical evidence suggest that conversion is a channel of transmission that contributes to explaining the long-lasting effects missions on HIV prevalence. Given data limitations and potential confounders the magnitudes of these findings have to be interpreted with caution. Despite these caveats, the possible negative effect of missionary activity on development through cultural transformations in the colonial context is a relevant finding in the growing economic literature on the, mostly positive, persistent effects of missionary activity.