Long-Term Effects of Access to Health Care: Medical Missions in Colonial India

Transcription

1 Long-Term Effects of Access to Health Care: Medical Missions in Colonial India Rossella Cali 1 and Federico G. Mantoanelli 2 1 Department of Economics, Boston College 2 Analysis Group This Version: April, 2015 First Draft: August, 2014) Abstract We study the long-term effect of access to health care on indiiduals health status by inestigating the relationship between the proximity to a Protestant medical mission in colonial India and health outcomes today. We use indiiduals anthropometric indicators to measure health status and geocoding tools to calculate the minimum distance between the location of indiiduals today and Protestant health facilities founded in the nineteenth century. We exploit ariation in actiities of missionary societies and use an instrumental ariable approach to show that proximity to a Protestant medical mission has a causal effect on current indiiduals health status. We inestigate some potential transmission channels and we find that the long-run effect of access to health care is not drien by persistence of infrastructure, but by improements in indiiduals health potential and, possibly, by changes in hygiene and self-care habits. Keywords: India, health, body mass index, historical persistence, Protestant missions. JEL codes: I15, N35, O12, O15, Z12, Z13. l: rossella.cali@bc.edu. federico.mantoanelli@analysisgroup.com. The iews presented in this paper are those of the author and do not reflect those of Analysis Group or its clients. We are grateful to Samson Ala, S Anukriti, Andrew Beauchamp, Donald Cox, Scott Fulford, Arthur Lewbel, Dilip Mookherjee, Mathis Wagner, Francisco Pino, and Robert Woodberry for their aluable comments. We thank all participants at the Boston College Applied Micro Seminar, the 2015 ASREC Conference and the Boston Uniersity Deelopment Reading Group for their suggestions and insight. All errors are ours.

2 1 Introduction According to the most recent World Health Organization WHO) estimates, India is ranked 150 th in the world in terms of life expectancy and one out of three Indian adults is underweight. 1 These country-leel figures hide a considerable regional ariation in health conditions: while South Indian states are doing relatiely well in terms of health performance, the same can t be said for some of the most populous states, such as Uttar Pradesh and Bihar, 2 whose health indicators are worse than those of many Sub-Saharan African countries Pal and Ghosh 2007)). We inestigate one potential source of the spatial heterogeneity in health performance by analyzing the long-term effect of access to health care facilities on current indiidual health outcomes. To this end, we exploit historical information about the diffusion, location and actiities of Protestant missionaries in colonial India in the nineteenth century. 3 Although most missionary societies initially ignored the importance of medicine for the dissemination of Christianity, from the 1860s onwards the deficiencies of the orthodox missionary methods became clear. The last decades of the nineteenth century witnessed an unprecedented growth in the medical mission enterprise and the number of doctors and medical personnel sent to India by Protestant societies increased dramatically, together with the number of missionary medical facilities and the quality of medical serices proided Richter 1908), Fitzgerald 2001)). We find empirical eidence of the existence of a long-term effect of access to health care facilities on current indiidual health status. Important policy implications may be drawn from our analysis. On one hand, understanding historical determinants of regional ariation in health conditions may proide useful guidance and insight in addressing issues related to health inequality in modern India. On the other hand, in light of the existence of long-run effects, the expansion of healthcare access becomes an een more critical priority, as it proides benefits both for current and future generations. 4 We use a noel dataset that combines contemporary indiidual-leel data from the India World Health Surey 2003) with historical data on Protestant medical missions settlements. Consistent with preious literature Schultz 2002), Schultz 2003)), we measure health status 1 Figures are een higher when only female adults are considered. 2 According to the 2011 Census of India, Uttar Pradesh and Bihar account for one fourth of the total Indian population. 3 Our decision of focusing on Protestant medical missions is mainly data drien. To the best of our knowledge, there is no publicly aailable document that systematically records medical actiities of Catholic missions. Moreoer, India was the preferred destination of the Protestant missionary enture due to its British element. Protestant were especially committed to operate among non-elites and socially disadantaged groups, while Catholic missionaries, at least until the Second Vatican Council in 1965, strategically decided to focus their efforts on the elite groups of society as they belieed that, by doing so, they would acquire a position of prominence in the local communities Mantoanelli 2014)). 4 In order to improe healthcare, the Indian goernment has increased its share of expenditure on total healthcare from 22% in 2005 to 33% in 2012, according to World Bank data. Some of the most recent national leel programs include National Rural Health Mission, National Urban Health Mission, National Health Insurance Programme Rashtriya Swasthya Bima Yojana) and Pradhan Mantri Swasthya Suraksha Yojana, a program aiming at correcting the imbalances in the aailability of affordable healthcare facilities in different parts of the country. 1

3 with indiiduals anthropometric indicators. We use geocoding tools to calculate the minimum distance between the surey s respondents and the location of a Protestant health facility founded in the nineteenth century. We find that the proximity to a Protestant medical mission has a positie and significant relationship with current indiiduals Body Mass Index BMI). Moreoer, we find that the larger was the closest Protestant medical facility, the stronger is the effect on indiiduals BMI today. We erify that this is not the result of other non-medical missionary actiities by showing that the icinity to Protestant missions without health facilities does not contribute to explain current health outcomes. We use additional historical data on Protestant schools to rule out that our results are drien by proximity to educational institutions. To gie a causal interpretation to our findings, we undertake a series of identification strategies. We control for an extensie set of geographic and indiidual specific characteristics and we include district specific fixed effects to account for district-leel heterogeneity. In addition, we address two types of selection that may affect our findings. On one hand, historical and geographic factors may hae drien missionaries decision of settling down in specific locations. On the other hand, Protestant missions may hae endogenously selected themseles into building a health facility and into proiding health care. In the spirit of Cagé and Rueda 2013), we tackle the first problem by inestigating whether results are robust to restricting our sample to indiiduals liing within arying distances from a Protestant mission, and the latter by instrumenting the proximity to the closest Protestant medical facility. We exploit the fact that Protestant missionary organizations differ in their inclination in undertaking medical actiities. We use these differences to compute the share of medical missions for each society in all regions of the world outside of India. We then construct our instrument as the sum of the shares of medical missions outside of India for all medical missions historically located less than K kilometers away from a surey respondent and we show that our findings are robust to the instrumental ariable approach. We analyze three potential transmission channels: persistence of infrastructure, improements in health potential health status of preious generations, nutrition and wealth), and improements in health habits hygiene and self-care awareness). First, consistent with preious literature Chaudhury and Hammer 2004), Banerjee et al. 2004)), we show that proximity to current health facilities does not play a role. Second, we find that the long-term effect of historical medical missions is partially, but not entirely, drien by improements in health potential. Finally, we proide preliminary eidence of the fact that proximity to a health care facility in the past may hae affected indiidual habits regarding hygiene, self-care and health awareness, which hae been bequeathed oer time and hae influenced health outcomes of later generations. 2

4 This paper adds to the growing empirical literature that looks at historical institutions as important determinants of current outcomes and inestigates path persistence in deeloping countries Acemoglu et al. 2001), Glaeser and Shleifer 2002), Nunn 2009), Alesina et al. 2013)). More specifically, we contribute to preious literature that recognizes the importance of religion and the actiities of religious organizations as possible fundamental sources of economic and social outcomes Barro and McCleary 2003), Barro and McCleary 2005)). Preious works hae identified Protestantism as an important determinant of adancements in human capital by focusing on its contribution to literacy and education Weber 1930), Becker and Woessmann 2009), Gallego and Woodberry 2009), Bai and Kung 2014), Cagé and Rueda 2013), Mantoanelli 2014)). In this paper, we take a first step in analyzing the historical persistence of health in deeloping countries. Health can influence not only indiiduals early surial and life expectancy, but also cognitie performance, schooling decisions, and productiity as adults. Understanding accumulation of health capital in deeloping countries means understanding one of the foremost determinants of economic deelopment Schultz 2009)). Recent works study how religious affiliation may help explaining differences in health outcomes Almond and Mazumder 2011), Brainerd and Menon 2013)). We differentiate from that strand of literature as we focus on the long-term effect of religious institutions. To our knowledge, this paper represents the first attempt to inestigate how the exposure to Protestant medical missions may hae determined adancements in current indiidual health outcomes. The rest of the paper is organized as follows. Section 2 gies a brief historical description of the Protestant medical enterprise in colonial India. In section 3, we describe our noel dataset, combining historical, geographic and contemporary data, and discuss the use of anthropometric indicators as a measure of health. Section 4 describes our empirical strategy. Main results and robustness checks are presented in section 5. In section 6, we inestigate potential transmission channels. Finally, section 7 concludes. 2 Medical Missions in Colonial India The period between represents an era of great missionary expansion led by the Protestants of Britain and the United States of America. The intensity of the missionary eangelization was such that that period is known as the great century of missions. India was the preferred destination of the Protestant missionary enture due to its British element. In the mid-nineteenth century more than a quarter of all the Protestant missionaries were stationed in India. On the ee of the first World War 5,200 Protestant missionaries of whom 2,500 British and 1,800 America) 3

5 were posted in the Indian sub-continent. Only China, at that time, rialed India in terms of Protestant presence. 5 At first missionary societies showed ery little interest in medicine. Initially Protestant organizations did not consider the establishment of hospitals and the proision of medical care as crucial for the scope of their missions. The missionary moement was coninced that the biblical command of eangelizing eery nation was limited only to preaching and teaching the Gospel message. 6 Most missionary societies completely dismissed the role that medicine may hae played as a way to disseminate Christianity. Estimates suggest that by 1840 there were no more than 40 Protestant medical missionaries at work worldwide. Things started to change from the 1860s onwards in correspondence with the growing awareness of the deficiencies of the orthodox missionary methods. Anxiety oer the lack of tangible results proided a powerful trigger for a change in official mission policies and practices. The belief was that the missionary sphere could rightfully include a range of actiities beyond those of preaching and teaching: I feel coninced that medical missions are amongst the means best calculated [...] to gie a fresh impetus to the cause of the Gospel, to help lift the chariot out of the rut in which sometimes it seems for a moment set fast and its progress retarded Re. Thomas V. French, Bishop of Lahore, quoted by Bishop McDougal in a speech at the Oxford Missionary Conference of 1877) Medical missions were finally recognized as aluable auxiliaries to the work of propagating Christianity, The work of the doctor is to open the door that the eangelist may enter in Medical Missionary Journal [1874], 9, 58-59) In contrast to official colonial medicine, mission medicine sought to place itself within non- European social and institutional milieus and reached out to embrace all classes of the natie society, the high-caste and the low-caste, the men and women Hardiman 2008)). In light of the recognition of the importance of medicine as an effectie weapon in the eangelization process, the last decades of the nineteenth century witnessed an unprecedented growth in the medical mission enterprise. The number of doctors and medical agents sent to India by the Protestant societies rose from 28 in 1870 to 335 in This was accompanied by a rapid increase in the number of mission hospitals and dispensaries that, by 1912, treated approximately three million patients annually. The quality of medical missionary medicine increased as well in 5 For more details refer to Richter 1908). 6 The constitution of the Baptist Missionary Society, for example, stated that the propagation of Christianity consisted of the preaching of the Gospel, the translation of the Holy Scriptures and the establishment of Schools. 4

6 the closing years of the nineteenth century. At the Bombay Missionary Conference of a resolution was passed stating that medical missionaries in India should inariably possess a medical degree or diploma sufficient to qualify as a license to practice in the West. Despite initial fears, it was surgery that determined acceptance of Western medicine among the Indian people and their greater willingness to use the resources of medical missions. Missionary reports indicate that oer time patients became increasingly willing to enter a mission hospital for periods of in-patient care and were generally applying for medical assistance at an earlier stage in their illness Fitzgerald 2001)). It should also be noticed that the majority of medical agents to India were women, reflecting the priority gien to women s medical mission work as a way of reaching India s female population Fitzgerald 2001)). 7 Moreoer, in addition to foreign staff, medical missions trained and employed Indians working as assistants around 700 in 1912), creating another ital contribution to the local communities. 3 Data To examine the relationship between the medical actiities of Protestant missionaries and current indiidual health status, we use an original data set that combines both historical and contemporary information. 3.1 Historical Data Information about the location of Protestant missions in colonial India comes from the Statistical Atlas of Christian Missions the Atlas, hereinafter), published in 1910 in correspondence with the World Missionary Conference held in Edinburgh Scotland). We geocoded the maps of India contained in the Atlas so that we were able to identify the exact location of Protestant missions as of Figure A1 in the Appendix shows an example of these maps. To identify missionary medical infrastructures we rely on the Centennial Surey of Foreign Missions Dennis, ed., 1902) which proides a worldwide list of all the Protestant missions with medical facilities and a record of their basic characteristics. For example, we know the designation of the facility hospital or dispensary), the foundation date, the name of the supporting Protestant society, the number of patients and surgical cases. A reproduction of one page of the Surey is proided in Figure A2 in the Appendix. We digitize and geocode this information for the Indian 7 This was an important factor in the success of the missionary moement as women could reach that part of the female population especially in high ranked castes) socially secluded in the Zenanas, which was the part of the house dedicated to the female members of a family. 5

7 subcontinent. We use the same data source to identify Protestant missions with educational institutions. Our final sample includes a total of 1,069 Protestant missions. Of these, 183 are equipped with either a hospital or dispensary. From the middle of the nineteenth century, the Goernment of India started a great transportation infrastructure project aimed at deeloping a railroad network in the Indian sub-continent. Between 1853 and 1930, a total of 67,000 kilometers of railroads were built, penetrating inland regions and leading to significant changes in the technology of trading in India. It is likely that railroads play a significant role in determining the missionaries location decisions. Areas with a well-deeloped railways network were more likely to be targeted by missionaries as they were more easily accessible. To take that into account we digitized and geocoded historical maps contained in the Constable s Hand Atlas of India. Figure A3 in the Appendix reports a section of one of those maps showing the status of the Indian railways network as of Finally, in our empirical analysis we also include information about the location of historical Catholic missions. We rely on maps contained in the Atlas Hierarchicus, an official document of the Vatican published in 1913 and showing the worldwide geographic distribution of Catholic missions as of Figure A4 in the Appendix shows one of these maps. 3.2 Contemporary Data Current indiidual data are from the 2003 India - World Health Surey WHS), produced by the World Health Organization WHO). Since 2003 the WHO has been deeloping and implementing a surey program to compile comprehensie information on the health of populations of 70 deeloped and deeloping countries and on the outcomes associated with inestments in the health sector. The 2003 India WHS contains information at indiidual and household leel of subjects residing in seen states: Assam, Karnataka, Madhya Pradesh, Maharashtra, Rajasthan, Uttar Pradesh and West Bengal. While the surey is centered around health-specific issues, it also proides demographic information of the respondents and data about household composition and educational leel. Crucially for the purpose of our research, the surey discloses the geographic coordinates of where the respondents lie. This information allows us to combine the contemporary indiidual data with the location of historical Protestant hospitals. Figure 1 shows the locations of surey respondents and of the Protestant health facilities actie in

8 3.3 Geographic Data We control for geographic characteristics at WHS location leel. We use the Global Gridded Population Database from CIESIN to compute the logarithm of the population in an area contained within a 5 kilometers radius from the surey s respondents location. 8 We also calculate the aerage eleation within the 5 km buffer using the SRTM 90m Digital eleation Data compiled by the CGIAR Consortium for Spatial Information. 9 Finally, to account for the importance of access to water sources especially in rural enironments) we also include a ariable measuring the number of water streams within the 5 km buffer around the WHS locations. Data are from the Digital Chart of the World. 10 [FIGURE 1 HERE] 3.4 Descriptie Statistics Table 1 contains some descriptie statistics of the ariables used in our empirical analysis. Columns 1 to 3 present figures for the entire surey sample, while columns 4 and 5 show figures for the baseline estimation sample, limited to indiiduals of age 20 to 60 and BMI 15 to 30. There are no significant differences in means and standard deiations between the surey sample and our preferred estimation sample. [TABLE 1 HERE] The location of a Protestant mission or health facility dispensary or hospital) is likely not to be random. A number of geographic and historical factors may hae drien the settlements decisions of missionaries in colonial India, such as access to clean water, population density, altitude, accessibility by railways and the proximity to a Catholic mission. Preious literature widely discusses the potential determinants of mission locations Johnson 1967), Nunn 2010), Cagé and Rueda 2013)), showing that Protestant missionaries choose to locate in geographically faorable and more accessible regions. In Table 2, we inestigate whether and how geographic and historical controls ary as we restrict the sample to areas closer to a Protestant mission or to a Protestant medical mission. The descriptie statistics suggest that the location of both missions and hospitals may be affected by historical and geographic factors and is, therefore, not random. Columns ) 5-6) report 8 Center for International Earth Science Information Network geolib/gis/dcw.html 7

9 means and standard deiations of historical and geographic controls when the sample is restricted to obserations below the 25 th 50 th ) 75 th ) percentile of minimum distance from a mission Panel A) or from a medical mission Panel B). 11 Population in 5 kilometers radius increases as we restrict the sample to nearby areas, suggesting that missionaries prefer to settle in more densely populated rather than to build missions in more remote locations. Moreoer, medical missions are on aerage located in areas that hae better access to colonial railways and water sources compared to missions with no health care facility. In addition, historical mission settlements seem to hae on aerage higher altitude, while medical mission are located at lower altitude, probably due to a higher need of accessibility. Finally, the distance from a Catholic mission increases as further away areas are included in the sample, suggesting, on one hand, that the geographic and historical factors driing location decisions of Protestant missionaries are similar to those driing location decisions of Catholic missionaries, and, on the other hand, that the presence of a Catholic mission does not act as a deterrent to the settlement of Protestant missionaries. [TABLE 2 HERE] 3.5 Measuring Health Measurement of health has recently eoled to rely on anthropometric indicators of physical deelopment. Preious literature has used both height and weight for height to measure indiidual health. 12 In this paper we use weight for height or Body Mass Index, hereinafter BMI, 13 as a proxy for current indiidual health status Waaler 1984), Deaton 2008), Schultz 2009)). 14 That the relationship between health status and BMI is nonlinear is a well known fact, both at the indiidual and at the aggregate leel. Waaler 1984) clearly illustrates this non-linearity 11 Since medical missions are a subset of Protestant missions, the distance percentiles are lower when considering the distance from a mission than when considering the distance from a medical mission. The mission distance percentiles based on all sample statistics) are 14 kilometers 25 th ), 29 kilometers 50 th ) and 49 kilometers 75 th ); the hospital distance percentiles based on all sample statistics) are 28 kilometers 25 th ), 75 kilometers 50 th ) and 121 kilometers 75 th ). 12 Height as an adult includes the long-run effect of fetal and childhood nutritional limitations and disease enironment. It is referred to as stunting when height is two standard deiations below the aerage in a reasonably well-fed reference population. In contrast, weight for height BMI) responds to the short-run nutritional balance among food and disease and is referred to as wasting when BMI alues are two standard deiations below aerage Fogel 1994), Schultz 2002), Schultz 2003), Weil 2005), Steckel 2008)). 13 The BMI is defined as follows: BMI = weightkg) height 2 m) = weightlb) height 2 in) The WHO proides an international classification of adult underweight, oerweight and obesity according to BMI and fixes 18.5 as the cut-off alue between underweight and normal-range, 25 as the cut-off alue between normal-range and oerweight and 30 as the cut-off alue between oerweight and obese. Eidence, howeer, suggests that Asian populations hae different associations between BMI, percentage of body fat, and health risks. A WHO expert consultation in 2004 Shiwaku et al. 2004)) addresses the recent debate about interpretation of recommended body-mass index BMI) cut-off points for determining oerweight and obesity in Asian populations, and deelops population-specific cut-off points for BMI. The current BMI cut-off alues for Asian Indians are 18.5, 23, 25 and 30 for the thresholds between underweight and normal-range, between normal-range and oerweight, between oerweight and pre-obese and between pre-obese and obese, respectiely. 8

10 by showing a U-shaped relationship between relatie risk of mortality and BMI using Norwegian data from the 1970s. Especially in low income countries, howeer, increases in the caloric intake shifts the lower tail of the BMI distribution to the right. Fogel 1994, 2004) interprets this shift as an accumulation of the population s health human capital stock, which tends to be associated with both declines in mortality and increases in labor productiity. [FIGURE 2 HERE] Figure 2 shows the results of a non-parametric estimation of the relationship between health status and BMI for respondents of age between 20 and 60. We exclude children, teenagers and elderly in order to aoid changes in indiiduals BMI due to biological growth and aging and not directly to their health status. The grey areas identify the 95% confidence interals. Panel a) suggests the presence of a non-monotonic concae relationship between respondents self-reported health status 15 and BMI, while panel b) indicates a conex relationship between the probability of seeking health care in the preious month and BMI. 16 In our empirical analysis we focus on indiidual with BMI between 15, an indicator for staration, and 30, the obesity threshold. We therefore concentrate only on the monotonic part of the BMI-health relationship. [FIGURE 3 HERE] Figure 3 shows the distribution of the logarithm of BMI in the sample, together with the kernel density 17 solid line) and a Normal distribution dashed line). Consistent with preious literature Burmaster and Crouch 1997), Hjelmborg et al. 2008)), BMI in our surey sample is log normally distributed. Unless otherwise stated, in the rest of the paper we refer to the scale of natural logarithm of BMI Empirical Strategy Our baseline equation is as follows BMI id = α d β Hospital distance id X idγ 1 W dγ 2 ɛ id 1) 15 Surey respondents are asked to rate their health status, from ery bad here equal to zero) to ery good here equal to 4). 16 Kernel regression estimates: the bandwidth is optimally chosen and equal to 1.11 in panel a) and 1.86 in panel b); the Epanechniko kernel function is used in both panels. 17 Bandwidth: 0.06; Epanechniko kernel function. 18 Findings are robust to a change of the specification to a linear-log setting. 9

11 where BMI id is the logarithm of the BMI of indiidual i, liing in illage in district d. The surey proides anthropometric information for one single indiidual per household, so the indiidual and household dimensions coincide in our dataset. Hospital distance id is the logarithm of the minimum aerial distance between the Protestant medical mission operating in 1902 and the location of surey respondent i. 19 Our parameter of interest is β, measuring the percentage change in BMI following a 1% change in the distance from a location of a Protestant medical mission. We control for a large set of coariates influencing both the BMI of indiiduals today and the location of a Protestant hospital at the end of the nineteenth century. The ector X id contains indiidual and household leel characteristics such as gender, marital status, highest completed leel of education, ethnic group, age and household size. In our preferred specification, W d includes illage specific geographic controls, such as population density population liing within a 5 kilometers radius), number of riers or lakes in a 5 kilometers radius and altitude. We include district leel fixed effects, α d, to account for unobsered heterogeneity across districts. In order to identify the causal effect of the proximity to a historical Protestant medical facility and current indiiduals health outcomes, we address two types of selection. On one hand, as discussed in section 3.4, historical and geographic characteristics may hae drien the missionaries decision of settling down in specific locations. On the other hand, Protestant missions may hae endogenously selected themseles into building a health facility and into proiding health care. We deal with the first potential selection issue in two ways. First, we expand W d to include the number of British colonial railways within a 5 kilometers radius and the minimum distance from a Catholic mission. These ariables, while potentially endogenous, may capture factors that also determine the location decisions of Protestant missionaries. Second, in order to correct for possible systematic differences between regions with and without Protestant hospitals, we restrict our sample of analysis to indiiduals liing with a certain distance from a Protestant mission. 20 The underlying idea is that limiting the analysis to more concentrated areas should minimize the risk of our findings being drien by within district unobsered characteristics. To tackle the additional problem of endogenous selection of Protestant mission into building a health facility, we follow Cagé and Rueda 2013) and deelop an instrumental ariable approach. The fact that Protestant missionaries built medical missions in historically and geographically faored areas may bias the OLS estimates of β upwards, while potential downward bias of OLS 19 We consider the logarithm of the distance, instead of the ariable in leels, as we beliee that percentage changes in distance from a Protestant medical mission matters more than leel changes, i.e. we beliee that an increase in distance from 10 km to 50 km does not hae the same effect of an increase in distance from 100 km to 150 km. 20 We consider different radii 14, 29 and 49 kilometers) which represent the the 25 th, 50 th and 75 th percentiles of the distance distribution. 10

12 estimates may arise due to the fact that Protestant missionaries built medical missions in areas where medical care was more needed and to measurement error. In the nineteenth century, missionary societies differed in their propensity to undertake medical actiities. Using information contained in the Centennial Surey of Foreign Missions, we compute the share of medical missions outside India for each missionary society affiliated with the Protestant health facility m: Society hospital share m = Hm M m 2) where H m and M m are the society s number of medical missions and total missions outside India, respectiely. 21 We then construct our instrument as the sum of these shares for all medical missions historically located less than K kilometers away from indiidual i liing in illage : Hospital share K id = k K The first stage of the IV approach is defined as Society hospital share mk) id 3) Hospital distance id = α d λhospital share K id X idδ 1 W dδ 2 ɛ id 4) where Hospital distance id is the logarithm of the minimum distance between the Protestant medical mission and the location of surey respondent i and Hospital share K id is our instrument as defined in 3). X id is a ector of indiidual and household leel characteristics, W d includes illage specific geographic controls and α d are district fixed effects. Identification comes from the assumptions that Hospital share K id is uncorrelated with the error term in our outcome equation 1), while adequately correlated with Hospital distance id. The larger Society hospital share K id, the more likely it is for a mission associated with this society to be equipped with a hospital. The larger Hospital share K id, the more likely it is for an indiidual to be closer to a medical mission. Moreoer, we beliee that each society s share of medical missions outside India is uncorrelated with any within district unobsered heterogeneity that may hae led missionaries to select themseles into building a health facility in India in the nineteenth century. 21 Table A1 in the Appendix contains information about the missionary societies we include in our analysis. 21 societies hae positie alues of Society hospital share m. The table includes only societies affiliated with medical missions historically located less than 50 kilometers away from indiidual i liing in illage. 11

13 5 Results 5.1 OLS Estimation Table 3 presents OLS estimates. As discussed in section 3.5, the estimation sample include indiiduals aged 20 to 60 with BMI between 15 and 30. Standard errors are corrected for heteroskedasticity and clustered at the primary sampling unit leel. Column 1 shows that the unconditional elasticity of BMI with respect to the distance from a medical mission is equal to , suggesting that a 10% increase in the distance from a Protestant medical mission is associated with a decrease in BMI by nearly 0.12 percentage points. Alternatiely, for the aerage indiidual, liing about 40 kilometers closer to the location of a medical mission is associated with an increase of her BMI by about 0.1. Estimates in columns 2, 3 and 4 indicate that the negatie correlation between proximity to a medical mission and health outcome today is robust to the inclusion of district fixed effects, indiidual and geographic controls. District fixed effects account for any unobsered heterogeneity across districts, while geographic controls aim at controlling for illage leel obsered heterogeneity within districts. While geographic controls do not seem to play a role in determining indiiduals health status, some of the indiidual leel ariables are significantly associated with BMI. As expected, better educated indiiduals hae higher leels of BMI. Moreoer, the logarithm of BMI increases with age, but at a decreasing rate. The magnitude and significance of the coefficient of interest are practically unchanged across all specifications. [TABLE 3 HERE] As discussed in section 3.4, areas near historical Protestant missions with or without medical facilities) are on aerage more densely populated and hae better access to colonial railways and water sources. Moreoer, missions without health facilities are located, on aerage, at higher altitudes, while medical missions are located at lower altitudes. The non-random location of missionary stations may introduce some bias in the OLS estimates presented in table 3. In particular, if the proximity to a Protestant medical mission captured within district unobsered heterogeneity positiely negatiely) correlated with indiiduals health outcome today, the OLS estimates would be upward downward) biased. In table 4, we inestigate whether the potential selection bias is indeed driing our findings. Columns 1 to 3 present the results of the OLS estimation of our baseline specification 1) oer restricted samples. In column 1, the estimation sample is limited to areas within 49 kilometers 12

14 from a Protestant mission 75 th percentile), while in columns 2 and 3 is limited to areas within the 29 kilometers and 14 kilometers from a mission 50 th and 25 th percentiles), respectiely. Such restrictions represent an attempt to correct for possible systematic differences between regions with and without Protestant missions. Our coefficient of interest, β, representing the elasticity of current BMI with respect to proximity to Protestant medical mission, remains statistically significant at the 5% leel as the sample size decreases, suggesting that our results are not drien by endogenous location selection. Considering indiiduals liing within the 25 th percentile of the minimum distance from a Protestant medical mission, we find that a 10% increase in proximity increases BMI by 0.149%. If only indiiduals liing within the median distance are included, doubling the distance from a medical mission decreases the BMI by about 1.2%, while if only the upper quartile is excluded from the sample, we find that a 100% increase in the distance decreases BMI by 1.37%. We then augment our set of controls to include historical ariables, such as the number of British colonial railways within a 5 kilometer radius and the minimum distance from a Catholic mission. Common factors drie the location decisions of Protestant and Catholic missionaries and British colonizers. If these factors were the driing force behind our results, then we should find a significant correlation between our dependent ariable and these additional historical controls. Column 4 shows this is not the case, proiding further support to our claim that, once controlling for district fixed effects, endogenous location of Protestant missions does not drie our results. [TABLE 4 HERE] 5.2 IV Estimation To address the potential problem of endogenous selection of Protestant missions into building a medical facility, we adopt an instrumental ariable approach. On one hand, if Protestant missionaries built medical missions in historically and geographically faorable areas, then the estimate of β in equation 1) would be upward biased. On the other hand, if Protestant missionaries built medical missions in areas where medical care was more needed, then our estimate of β would be downward biased. Moreoer, an instrumental ariable approach tackles the attenuation bias due to measurement error. Table 5 shows the IV estimates The instrument is defined as follows: Hospital share 50 id = Society hospital share mk) id 5) k 50 We build our instrument using a 50 kilometer radius. In section 5.3 we show that our results are robust to changes in the radius. 13

15 [TABLE 5 HERE] Columns 1 and 2 report the results of the first stage, without and with indiidual and geographic controls. Due to their potential endogeneity, historical controls, i.e. the number of colonial railways within 5 kilometers and the minimum distance from a Catholic missions are not included. The logarithm of the distance from a Protestant medical mission is negatiely correlated with the instrument Hospital share 50 id. In both specifications, the coefficient of interest in equation 4), λ, is significant at the 1% leel. When indiidual and geographic controls are included, a 1% increase in Hospital share 50 id is associated with an increase in the logarithm of the distance by about 1%. The first stage F statistics 23 is largely aboe 10, indicating that Hospital share 50 id is a non-weak instrument for our endogenous regressor Hospital distance id. Columns 3 and 4 show the results of the second stage, without and with the inclusion indiidual and geographic controls. The statistically significant negatie relationship between distance from a medical mission and indiiduals BMI presented in section 5.1 is robust to the instrumental ariable approach. According to both specifications, proximity to a Protestant medical missions increases BMI. Een though the leel of significance of β decreases to 5%, the magnitude of the estimated coefficient triples if compared to the OLS estimates in table 3. Thus, OLS estimates are downward biased, suggesting that the potential upward bias related to missionaries endogenous location selection is more than offset by the downward bias related to potential measurement error and to the fact that medical missions may hae been located in areas where medical care was more needed. When indiidual and geographic controls are included, doubling Hospital distance decreases BMI by 3.13%. These results corroborate the existence of a positie long-term effect of proximity to a historical medical mission on indiiduals health status, as proxied by BMI. 5.3 Robustness Checks We perform a series of robustness checks to test the sensitiity of our results Self-Reported Health Status In our main analysis, we follow the literature and rely on BMI as a proxy of indiiduals health status. In an alternatie specification, we use a self-reported ordinal ariable to measure health. Table A5 in the Appendix shows the estimation results, both using OLS column 1) and an 23 We report here the Kleibergen-Paap F statistics, which is robust to non-iid errors. 14

16 ordered logit model column 2). 24 Our preious findings are robust to the use of the self-reported measure of health. The sign of the estimated coefficient in column 2 suggests that the latent ariable increases with the proximity to a Protestant medical mission. 25 Figure A5 displays the marginal effects at the mean) on the probability of an indiidual picking any integer alue between 0 and 4 when describing their current health status. As expected, the marginal effects are positie for worse health rankings and negatie for better health rankings. This suggests that an increase in distance from a historical medical facility is associated with a decrease in the probability that an indiidual describes his/her health as good or ery good Estimation sample In section 3.5 we discuss the use of BMI as a measure of health and our decision to restrict the estimation sample to indiiduals of age 20 to 60 and BMI between 15 and 30. We expect drastic ariations in the estimation sample to affect our findings. In particular, gien the non-monotonic relationship between indiiduals BMI and health status, we expect our results not to hold when we include indiiduals in the far right tail of the BMI distribution, i.e. indiiduals who would fall into the obese categories, as defined by the WHO. Table A6 in the Appendix shows the results oer different estimations samples. Clearly, the sample size decreases drastically once we include only people with BMI larger than 30 in the estimation sample and the relationship between proximity to a Protestant medical mission and indiiduals BMI today becomes not statistically significant. Our results are robust to extending the estimation sample to indiiduals with BMI between 15 and 35. This finding seems to be drien by indiiduals with BMI between 15 and 20, since the relationship becomes not significant when only indiiduals with BMI between 20 and 35 are considered Instrument We test whether our findings are robust to the way we construct the instrumental ariable. First, we find our results from the IV estimation to be robust to ariations in the radius used to construct 24 In a ordered logit model, ordered outcomes are modeled as sequentially arising as a latent ariable y crosses higher and higher thresholds. Estimates of these thresholds are reported in table A5. For the ordered logit model the error term is assumed to be logistically distributed. The alternatie specification is as follows. Health ranking id = α d β Hospital distance id X idγ 1 W dγ 2 ɛ id 6) where Health ranking id is a self-reported description of current health status and is an ordinal ariable from 0, ery bad, to 4, ery good ) determined by the response to the surey question how would you rate your health today?. Hospital distance id and the indiidual and geographic controls are defined as in 1). 25 The marginal effect on the indiidual probability of choosing alternatie j, with j = 0, 1, 2, 3, 4, when Hospital distance id changes is gien by P rhealth ranking id = j) Hospital distance id 15

17 the instrument as in equation 5). In section 5.2 we presented the estimation results obtained using a radius, K, equal to 50 kilometers. Tables A2 and A3 in the Appendix show the estimation results when K is equal to 30 kilometers and 10 kilometers respectiely. Our preious findings are confirmed when we use these different radii to build our instrument Hospital share K id. 26 Moreoer, we show that our findings do not change if we construct our instrumental ariable as the sum of shares, defined as in equation 2), for all missions, both medical and non-medical. To this end, we use data from the World Atlas Of Christian Missions 1911) that includes information about medical and non-medical missions actie in 1911 and their affiliated missionary societies. 27 Table A4 in the Appendix shows the estimation results for K equal to 50 kilometers. Nonmedical missions were often affiliated with societies with relatiely low Society hospital share m. The correlation coefficient between the two ersions of the instrument is 0.89 and statistically significant at the 1% leel. Our preious findings are confirmed when both medical and nonmedical missions are considered. 6 Transmission Channels In this section, we inestigate in more details the findings presented in sections 5.1 and 5.2 and discuss potential channels underlying the long-term effect of Protestant medical missions. 6.1 Mission and School Proximity As discussed in section 2, Protestant societies initially showed limited interest in medicine and did not consider the establishment of hospitals and the proision of medical care as crucial for the scope of their missions. At the same time, missionaries inested in seeral educational, cultural and philanthropic actiities and established schools, seminaries, printing presses and orphanages. Preious literature stressed the role of Protestantism in increasing human capital by focusing on its contribution to the adancement of literacy and education Cagé and Rueda 2013), Mantoanelli 2014)). In this section, we assess whether our preious results are indeed drien by icinity to a medical mission and not by proximity to a Protestant school or to a generic Protestant mission. In order to determine whether the long-term effect on indiiduals health is indeed drien 26 An increase in the releant radius, howeer, decreases both the first stage F statistics and the magnitude of the IV estimates for the parameter of interest, β. When a radius equal to 30 kilometers is used to construct the instrument and we control for indiidual and geographic obserable characteristics and district leel unobsered heterogeneity, a 10% increase in the proximity to a Protestant medical mission increase indiiduals BMI today by 0.293%; if a radius equal to 10 kilometers is used, a 10% increase in the proximity to a Protestant medical mission increase indiiduals BMI today by 0.223%. 27 We classify a mission to be a medical mission if the presence of a doctor is recorded. 16

18 by the proximity to a medical mission and not to a generic mission, we add to our baseline specification the distance to the closest Protestant mission, independently from it being equipped with a medical facility or not. 28 Column 1 in table 6 shows indeed that proximity to a generic Protestant mission does not play a role in determining current indiidual BMI. Our coefficient of interest β remains negatie and statistically significant. To assess whether the long-term effect is transmitted ia an educational, rather than medical, channel, we control for the minimum distance to different types of education institutions, i.e. boarding schools, high schools and uniersities. Figure 4 shows the spatial distribution of Protestant boarding schools, high schools and uniersities operating in colonial India in Columns 2 to 4 in table 6 show that this is not the case. Een when these different educational institutions are added to our analysis, we still find a negatie and significant relationship between hospital distance and current BMI. 29 [FIGURE 4 HERE] [TABLE 6 HERE] 6.2 Medical Mission Size and Medical Clusters We analyze the effect of the size of the closest Protestant medical facility on indiiduals BMI today and test our ex ante hypothesis that larger hospitals hae a more sizable long-run affect on indiidual health status. From the Centennial Surey of Foreign Mission we retriee data about the total number of patients, treatments and surgeries of each medical mission in We use this information as a proxy for the size of the Protestant medical facility. 30 We also inestigate whether liing closer to a historical cluster of Protestant medical missions has an independent effect on indiidual current BMI. We proxy medical missions clustering with the number of medical missions located within a 50 kilometer radius from the surey respondent. Table 7 shows the OLS estimates for alternatie specifications of our baseline model. column 1, the baseline specification is extended to include the size of the closest Protestant health facility in logarithm). In column 2, the number of medical missions within a 50 km buffer is included in the baseline model. Column 3 reports OLS estimates of a more complete model 28 As shown in table 1, the aerage minimum distance from a Protestant mission is less than half the aerage minimum distance from a Protestant medical mission, indicating that only a small fraction of historical missions had at least one hospital or one dispensary. 29 The statistical significance of our main regressor falls to 10% if we extend our baseline specification to simultaneously include the minimum distances to a generic mission, a boarding school, a high school and a uniersity. The coefficient on the medical mission is, howeer, the only significant one, suggesting that a the reduction in the statistical significance of our estimates may be mainly drien by a multicollinearity issue. 30 When more than one hospital or dispensary are present in the same illage, the total number of patients, treatment and surgeries is computed at the illage leel. 17 In

19 featuring the distance from the closest Protestant medical mission and its size together with our measure of medical clustering. As we expected, the larger the closest historical health facility, the larger the long-run affect on indiidual health status. Both the proximity to a Protestant medical mission and the size are statistically significant at the 1% leel. Eerything else equal, a 10% decrease in the distance from a medical mission is associated with an increased in indiidual current BMI by 0.123%. Analogously, a 10% increase in the total number of patients, treatments and surgeries of the closest medical mission is associated with an increased in current BMI today by 0.043%. The magnitude of the coefficients may be biased downwards due to measurement error, related to misreporting of hospital actiities and geocoding discrepancies, and to the fact that the number of patients, treatment and surgeries may be higher in less healthy and deeloped areas. Estimates in both columns 2 and 3 suggest that, once controlling for proximity to a medical mission, the long-term effect on indiiduals health is not drien by the presence of a clusters of medical missions. [TABLE 7 HERE] 6.3 Long-Term Transmission We classify the channels driing the long-term effect of access to health care on indiiduals health status in three main categories: channels related to infrastructure, to improements in health potential and in health culture, i.e. hygiene and health awareness. To illustrate the three transmission channels, we model indiidual health as follows H = HP, I, D) 7) where P is health potential, I are health inputs and D are exogenous enironmental conditions. These exogenous conditions are assumed to depend on local obserable characteristics, i.e. D = DW ). We define A SR as current access to health care and A LR as long-term access to health care, i.e. access to health care of preious generations. Since access to health care and health status of one generation can leae their mark on the health potential of the following cohorts, both ia an intergenerational transmission of health Bhalotra and Rawlings 2011), Coneus and Spiess 2012)) and wealth, we hae indiidual health potential depend on A LR, i.e. P = PA LR ). We can define indiidual health inputs to be a function of current access to health care, health culture C, such as hygiene and health awareness and health promoting practices, and socioeconomic indiidual characteristics, i.e. I = IA SR, C, X). Moreoer, we hae health culture 18

20 depend on current and past access to health care: C = CA SR, A LR ). On one side, proximity to health facilities today may directly affect health culture and stimulate the diffusion of healthy practices. On the other side, health and hygiene awareness may hae been bequeathed by the family or the social enironment and therefore depend on the access to health care of preious generations. An indiidual s health production function is therefore a function of long-term access to health care, access to health care today, socio-economic obserable indiidual characteristics and local enironmental conditions. Assuming a simple Cobb-Douglas production function and conditioning on obserable characteristics at the indiidual leel and at the illage leel, indiidual s health is produced as follows H = HA LR, A SR ) = κa LR ) β 1 A SR ) β 2 8) Since we cannot measure changes in health culture, while we can obsere current access to health care and measure health potential, we attribute any residual long-term effect of access to health care on indiiduals current health status to changes in hygiene and self-care awareness, as bequeathed by the family or the social enironment Persistence of Infrastructure In preious sections, we omitted current access to health care from our baseline specification to aoid identification issues. In this section, we inestigate the independent effects of long-term and short-term access to health care by estimating the health production function in equation 8). We measure indiidual s health with BMI logarithm), long-term access to health care with the minimum distance logarithm) from a Protestant medical mission and access to health care today with the minimum distance logarithm) from a illage with at least one hospital, as from the Village Directory of the 2001 Census of India 31 BMI id = α d β 1 Hospital distance LR id β 2 Hospital distance SR id X idγ 1 W dγ 2 ɛ id 9) Our parameters of interest are β 1 and β 2, measuring the elasticities of indiidual health with respect to long-term access to health care and to access to health care today, respectiely. If the only releant channel was the persistence of health infrastructure oer time, we would expect the 31 We successfully geocoded 4,617 illages in Assam, Karnataka, Madhya Pradesh, Maharashtra, Rajasthan, Uttar Pradesh and West Bengal with at least one allopathic hospital 569 in Assam, 213 in Karnataka, 427 in Madhya Pradesh, 947 in Maharashtra, 46 in Rajasthan, 2,252 in Uttar Pradesh and 165 in West Bengal) and computed the aerial minimum distance between WHS respondents and the allopathic hospital. For the sake of simplicity, we exclude Ayuredic, Unani and homeopathic hospitals, dispensaries, maternity and child welfare centers, maternity homes, primary health centers, family welfare centers, T. B. clinics, nursing homes, priate and subsidized medical practitioners and community health workers. 19

21 coefficient on Hospital distance LR id to become smaller in magnitude and, potentially, insignificant once including Hospital distance SR id in the model. 32 Fixed effects, α d, control for district leel unobsered heterogeneity. 33 Column 1 of Table 8 contains our estimation results and proides preliminary eidence of the fact that the long-run effect of access to health care facilities on indiiduals health is not drien by persistence of infrastructure. Hospital distance SR id is not associated with indiidual BMI and the magnitude and significance of the coefficient on Hospital distance LR id do not change when current access to health care is included in the baseline specification. This suggests that there may be other mechanisms health potential and culture) playing a releant role Health Potential and Alternatie Channels The cross-sectional nature of the WHS does not allow us to match indiiduals BMI today with health outcomes of preious cohorts. Height as an adult, howeer, includes the long-run effect of fetal and childhood) nutrition and exposure to diseases, together with the genetic height potential. Via intergenerational transmission of health, access to health care and health status of indiiduals in one generation could therefore affect the stature of indiiduals in the next one. In this section, we use height a proxy of long-run health capital 34 to inestigate whether the long-term effect of proximity to a Protestant medical mission on indiiduals BMI is drien by improements in health potential. In addition, we assess whether this effect is merely drien by increases in weight due to better nutrition. We also control for total household expenditure, as a proxy for income. We use principal component analysis to create an index household wealth, which we include to control for intergenerational transmission of wealth and additional health inputs. 35 Column 2 to 5 contain the OLS estimates of an alternatie ersion of our baseline specification in equation 1) featuring our measures of health potential as additional controls. Een controlling for indiidual height, the coefficient on the distance from a Protestant medical mission remains statistically significant and negatie, suggesting that the intergenerational transmission of health may not fully explain the long-term effect on current BMI. Column 3 presents the OLS estimates of an alternatie specification aiming at disentangling the effect of access to better nutrition in 32 In an alternatie specification, we proxy the icinity of a health care center today with the self-reported number of minutes it took an indiidual to reach the closest health care proider while seeking care for him/herself or his/her children in the preious 12 months. This information is unfortunately aailable only for indiiduals who needed health assistance in the preious year 3,540 obserations), which could introduce additional downward bias in the OLS estimates. Results are shown in table A7 in the Appendix. Results confirm the findings presented in this section. 33 In other words, we assume κ to be additiely separable in district leel and indiidual idiosyncratic components, i.e. κ id = α d ɛ id. 34 See section We create a wealth index using the first principal component among six different measures of household assets and wealth: number of rooms in the dwelling, ownership of a fridge, a t, a radio and a scooter. 20

22 terms of quantity and quality of food) from the icinity to a Protestant medical mission. Despite the magnitude of the coefficient of the distance from a Protestant hospital slightly decreases when we control for quantity and quality of nutrition, the long-term effect of proximity to a historical health facility on BMI today remains statistically significant at the 5% leel. This indicates that proximity to a medical mission has an effect on BMI today that is independent from a better access to food and short-run changes in weight. In columns 4 and 5 we reject that current income and wealth are the only driers of our finding: een controlling for household total expenditure and assets, proximity to a Protestant medical mission is significantly associated with higher indiidual current BMI. [TABLE 8 HERE] The estimation results in column 6 proide conincing eidence in faor of our hypothesis that the cultural channel may in fact play a role in the long-run transmission of the effect of access to health care. Een controlling for indiidual current access to health care, income, wealth, nutrition and height, socio-demographic indiidual characteristics, illage leel geographic characteristics and across districts unobsered heterogeneity, a 100% increase in the distance from a Protestant medical mission is associated with a 1.2% lower indiidual BMI today. Preious results suggests that we can consider this as a lower bound for the actual magnitude of the effect. Proximity to a health care facility, doctors and nurses may affect eeryday habits regarding hygiene, self-care and health awareness, which are bequeathed oer time and may influence health outcomes of subsequent generations. 7 Conclusion We analyze one potential source of the regional ariation of health performance in India by inestigating the existence of a long-term effect of access to health care facilities on current indiidual health outcomes. We combine historical information about the location and actiities of Protestant medical missions in colonial India with contemporary indiidual leel data to identify the long-term causal effect of access to medical facilities on current indiidual health status. We show that proximity to the location of a Protestant medical mission has a positie and significant effect on current indiiduals Body Mass Index BMI). We erify that our findings are not drien by other non-medical missionary actiities and show that the icinity to Protestant missions without health facilities does not contribute to explain current health outcomes. The inestigation of potential transmission channels shows that persistence of infrastructure and current access to 21

23 health care do not play a role, while improements in health potential and, possibly, changes in health habits are the main driers of the long-term effect. This paper represents a first attempt to analyze the persistence of health in deeloping countries and to inestigate how exposure to Protestant medical missions may hae determined adancements in current indiidual health. In light of the existence of long-run effects, the expansion of healthcare access becomes an een more critical priority, as it beneficially affects both current and future generations. Further work should focus on explicitly connect the long-term effect of access to health care, current health status, productiity and economic outcomes. 22

24 Bibliography Acemoglu, D., S. Johnson, and J. A. Robinson 2001): The Colonial Origins of Comparatie Deelopment: An Empirical Inestigation, American Economic Reiew, 91, [3] Alesina, A., P. Giuliano, and N. Nunn 2013): On the Origins of Gender Roles: Women and the Plough, The Quarterly Journal of Economics, 128, [3] Almond, D. and B. Mazumder 2011): Health Capital and the Prenatal Enironment: The Effect of Ramadan Obserance during Pregnancy, American Economic Journal: Applied Economics, 3, [3] Bai, Y. and J. K. S. Kung 2014): Diffusing Knowledge While Spreading God s Message: Protestantism and Economic Prosperity in China, , Forthcoming. Journal of the European Economic Association. [3] Banerjee, A., A. Deaton, and E. Duflo 2004): Health care deliery in rural Rajasthan, Tech. rep. [2] Barro, R. J. and R. McCleary 2003): Religion and Economic Growth, NBER Working Papers 9682, National Bureau of Economic Research, Inc. [3] Barro, R. J. and R. M. McCleary 2005): Which Countries Hae State Religions? The Quarterly Journal of Economics, 120, [3] Becker, S. O. and L. Woessmann 2009): Was Weber Wrong? A Human Capital Theory of Protestant Economic History, The Quarterly Journal of Economics, 124, [3] Bhalotra, S. and S. B. Rawlings 2011): Intergenerational persistence in health in deeloping countries: The penalty of gender inequality? Journal of Public Economics, 95, , new Directions in the Economics of Welfare: Special Issue Celebrating Nobel Laureate Amartya Sen s 75th Birthday. [18] Brainerd, E. and N. Menon 2013): Religion and Health in Early Childhood: Eidence from the Indian Subcontinent, Working Papers 65, Brandeis Uniersity, Department of Economics and International Business School. [3] Burmaster, D. E. and E. A. Crouch 1997): Lognormal distributions for body weight as 23

25 a function of age for males and females in the United States, , Risk Analysis, 17, [9] Cagé, J. and V. Rueda 2013): The long Term Effects of the Printing Press in Sub Saharan Africa, PSE Working Papers halshs , HAL. [2], [3], [7], [10], [16] Chaudhury, N. and J. S. Hammer 2004): Ghost Doctors: Absenteeism in Rural Bangladeshi Health Facilities, The World Bank Economic Reiew, 18, [2] Coneus, K. and C. K. Spiess 2012): The intergenerational transmission of health in early childhoodñeidence from the German Socio-Economic Panel Study, Economics and Human Biology, 10, [18] Deaton, A. 2008): Height, Health, and Inequality: The Distribution of Adult Heights in India, American Economic Reiew, 98, [8] Fitzgerald, R. 2001): Clinical Christianity : the emergence of medical work as a missionary strategy in colonial India, in Health, Medicine and the Empire: New Perspecties on Colonial India, ed. by B. Pati and M. Harrison, London: Sangam Books. [1], [5] Fogel, R. W. 1994): Economic Growth, Population Theory, and Physiology: The Bearing of Long-Term Processes on the Making of Economic Policy, American Economic Reiew, 84, [8], [9] 2004): The Escape from Hunger and Premature Death, 1700Ð2100, no in Cambridge Books, Cambridge Uniersity Press. [9] Gallego, F. A. and R. Woodberry 2009): Christian Missionaries and Education in Former African Colonies: How Competition Mattered, Working Papers ClioLab 2, EH Clio Lab. Instituto de Economía. Pontificia Uniersidad Católica de Chile. [3] Glaeser, E. L. and A. Shleifer 2002): Legal Origins, The Quarterly Journal of Economics, 117, [3] Hardiman, D. 2008): Missionaries and their medicine: a Christian modernity for tribal India, Manchester Uniersity Press. [4] Hjelmborg, J.., C. Fagnani, K. Silentoinen, M. McGue, M. Korkeila, K. Christensen, A. Rissanen, and J. Kaprio 2008): Genetic Influences on Growth Traits of BMI: A Longitudinal Study of Adult Twins, Obesity, 16, [9] 24

26 Johnson, H. B. 1967): The Location of Christian Missions in Africa, Geographical Reiew, 57, pp [7] Mantoanelli, F. G. 2014): The Protestant Legacy: Missions and Literacy in India, Unpublished manuscript. [1], [3], [16] Nunn, N. 2009): The Importance of History for Economic Deelopment, Annual Reiew of Economics, 1, [3] 2010): Religious Conersion in Colonial Africa, American Economic Reiew, 100, [7] Pal, P. and J. Ghosh 2007): Inequality in India: A surey of recent trends, Working Papers 45, United Nations, Department of Economics and Social Affairs. [1] Richter, J. 1908): A History of Missions in India, F. H. Reell. [1], [4] Schultz, T. P. 2002): Wage Gains Associated with Height as a Form of Health Human Capital, American Economic Reiew, 92, [1], [8] 2003): Wage rentals for reproducible human capital: eidence from Ghana and the Iory Coast, Economics & Human Biology, 1, [1], [8] 2009): Population and Health Policies, Working Papers 974, Economic Growth Center, Yale Uniersity. [3], [8] Shiwaku, K., E. Anuurad, B. Enkhmaa, K. Kitajima, and Y. Yamane 2004): Appropriate Body-Mass Index For Asian Populations and Its Implications For Policy and Interention Strategies, Lancet. [8] Steckel, R. H. 2008): Heights and Human Welfare: Recent Deelopments and New Directions, NBER Working Papers 14536, National Bureau of Economic Research, Inc. [8] Waaler, H. T. 1984): Height. Weight and Mortality The Norwegian Experience, Acta Medica Scandinaica, 215, [8] Weber, M. 1930): The Protestant Ethic and the Spirit of Capitalism, , London: G. Allen and Unwin. [3] Weil, D. N. 2005): Accounting for the Effect of Health on Economic Growth, Tech. rep. [8] 25

27 8 WHS respondents Protestant Medical Missions Districts States ,000 Kilometers Figure 1: Protestant Medical Missions and WHS Respondents 26

28 Self-reported health BMI a) Self-reported health status Health care needed in preious 30 days BMI b) Health care needed in the past month Figure 2: Non-parametric regression of health status on BMI 27

29 Density BMI log) Figure 3: Distribution of the logarithm of BMI in the sample 28

30 ) 8 ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) Protestant Medical Missions Protestant Uniersities Protestant High School Protestant Boarding Schools Districts States Figure 4: Protestant Medical Missions and Schools 29 1,000 Kilometers

31 Table 1: Descriptie Statistics All sample Estimation sample Obs. Mean St. De. Obs. Mean St. De. 30 Body Mass Index BMI) Female) Married) Major ethnic group) No formal schooling) Less than primary school) Primary school completed) Secondary school completed) High school completed) College completed) Post graduate degree completed) Age Household size No. riers/water sources in 5 km radius Population in 5 km radius Aerage altitude in 5 km radius No. colonial railways in 5 km radius Distance from Protestant medical mission km) Distance from Protestant mission km) Distance from Catholic mission km) Distance from Protestant mission with uniersity km) Distance from Protestant mission with boarding school km) Distance from Protestant mission with high school km) Distance from allopathic hospital from Census 2001 km) Only indiiduals of age 20 to 60 and BMI 15 to 30 are included in the estimation sample.

32 Table 2: Geographic and Historical Controls Panel A: Distance from Protestant mission Percentile 25% 50% 75% Mean St. De Mean St. De Mean St. De 1) 2) 3) 4) 5) 6) 31 No. riers in 5km radius Population in 5km radius Ag. altitude in 5km radius No. colonial railways in 5km radius Distance from Catholic mission km) Panel B: Distance from Protestant hospital No. riers in 5km radius Population in 5km radius Ag. altitude in 5km radius No. colonial railways in 5km radius Distance from Catholic mission km) Mission distance percentiles based on all sample statistics. 14 kilometers: 25 th ; 29 kilometers: 50 th ; 49 kilometers: 75 th. Hospital distance percentiles based on all sample statistics. 28 kilometers: 25 th ; 75 kilometers: 50 th ; 121 kilometers: 75 th.

33 Table 3: OLS estimates Dependent ariable BMI log) 1) 2) 3) 4) Distance from medical mission log) ) ) ) ) 1Female) ) ) 1Married) ) ) 1Less than primary school) ) ) 1Primary school completed) ) ) 1Secondary school completed) ) ) 1High school completed) ) ) 1College completed) ) ) 1Post graduate degree completed) ) ) 1Major ethnic group dummy) ) ) Age ) ) Age ) ) Household size ) ) Household size ) ) Population in 5 km radius log) ) Aerage altitude in 5 km radius log) ) No. riers/water sources in 5 km radius ) Constant ) ) ) ) N District FE No Yes Yes Yes Indiidual controls No No Yes Yes Geographic controls No No No Yes Robust standard errors in parentheses. p < 0.10, p < 0.05, p < Standard errors clustered at the PSU leel. Only indiiduals of age 20 to 60 and BMI 15 to 30 are included. 32

34 Table 4: Geographic and Historical Selection Dependent ariable BMI log) Distance percentile 75% 50% 25% 100% 1) 2) 3) 4) Distance from medical mission log) ) ) ) ) No. colonial rails in 5km radius 1891) ) Distance from Catholic miss. log) ) Constant ) ) 0.104) ) N District FE Yes Yes Yes Yes Indiidual controls Yes Yes Yes Yes Geographic controls Yes Yes Yes Yes Robust standard errors in parentheses. p < 0.10, p < 0.05, p < Standard errors clustered at the PSU leel. Only indiiduals of age 20 to 60 and BMI 15 to 30 are included. Mission distance percentiles based on all sample statistics. 14 kilometers: 25 th ; 29 kilometers: 50 th ; 49 kilometers: 75 th. OLS estimates. Table 5: IV estimates Instrument: Hospital share, 50 km) First Stage OLS) Second Stage IV) Dependent ariable Distance log) BMI log) 1) 2) 3) 4) Hospital share 50 km only medical) ) 0.215) Distance from medical mission log) ) ) Constant ) 0.595) ) ) Kleibergen-Paap rk Wald F statistic N District FE Yes Yes Yes Yes Indiidual controls No Yes No Yes Geographic controls No Yes No Yes Robust standard errors in parentheses. p < 0.10, p < 0.05, p < Standard errors clustered at the PSU leel. Only indiiduals of age 20 to 60 and BMI 15 to 30 are included. 33

35 Table 6: Protestant Missions and Schools Dependent ariable BMI log) 1) 2) 3) 4) 5) Distance from medical mission log) ) ) ) ) ) Distance from Protestant mission log) ) Distance from Protestant mission with boarding school log) ) ) 34 Distance from Protestant mission with high school log) ) ) Distance from Protestant mission with uniersity log) ) ) Constant ) ) ) ) ) N District FE Yes Yes Yes Yes Yes Indiidual controls Yes Yes Yes Yes Yes Geographic controls Yes Yes Yes Yes Yes Robust standard errors in parentheses. p < 0.10, p < 0.05, p < Standard errors clustered at the PSU leel. Only indiiduals of age 20 to 60 and BMI 15 to 30 are included. OLS estimates

36 Table 7: Medical Mission Size and Medical Clusters Dependent ariable BMI log) 1) 2) 3) Distance from medical mission log) ) ) ) Hospital size log) ) ) No. medical missions in 50 km radius ) ) Constant ) ) ) District FE Yes Yes Yes Indiidual controls Yes Yes Yes Geographic controls Yes Yes Yes Robust standard errors in parentheses. p < 0.10, p < 0.05, p < Standard errors clustered at the PSU leel. Only indiiduals of age 20 to 60 and BMI 15 to 30 are included. OLS estimates 35

37 Table 8: Long-run Transmission Channels Dependent ariable BMI log) 1) 2) 3) 4) 5) 6) Distance from medical mission log) ) ) ) ) ) ) Distance from allopathic hospital log) ) ) Height log) ) ) Fruit & egetables serings per day) ) ) 36 Expenditure on food log) ) ) Total expenditure log) ) ) Wealth index ) ) Constant ) 0.131) ) ) ) 0.134) N District FE Yes Yes Yes Yes Yes Yes Indiidual controls Yes Yes Yes Yes Yes Yes Geographic controls Yes Yes Yes Yes Yes Yes OLS estimates. Robust standard errors in parentheses. p < 0.10, p < 0.05, p < Standard errors clustered at the PSU leel. Only indiiduals of age 20 to 60 and BMI 15 to 30 are included. Expenditure on food is the amount of rupees spent in the preious 4 weeks on things as rice, meat, fruits, egetables, and cooking oils. It includes the alue of any food that was produced and consumed by the household, and excludes alcohol, tobacco and restaurant meals. Total expenditure is the total amount spent by the household in the past 4 weeks. Wealth index is the first principal component between 6 assets.

38 A Appendix [FIGURE A1 HERE] [FIGURE A2 HERE] [FIGURE A3 HERE] [FIGURE A4 HERE] [FIGURE A5 HERE] [TABLE A5 HERE] [TABLE A2 HERE] [TABLE A3 HERE] [TABLE A6 HERE] [TABLE A7 HERE] 37

39 Figure A1: Protestant missions locations as of 1910 Statistical Atlas of Christian Missions) Figure A2: Page from the Centennial Surey Surey of Foreign Missions 38

40 Figure A3: Status of Indian railways network as of 1891 Constable s Hand Atlas of India) 39

41 Figure A4: Catholic missions locations as of 1911 Atlas Hierarchicus) Marginal Effect of Distance Self-reported health 95% CI Marginal Effect Figure A5: Ordered Logit Marginal Effects MEM) 40