The Long-Term Effects of Christian Activities in China

Yuyu Chen Hui Wang Guanghua School of Management Guanghua School of Management Peking University Peking University [email protected] [email protected]

Se Yan Guanghua School of Management Peking University [email protected]

September 2013

Does culture, and in particular religion, exert an independent causal effect on long-term economic performance, or is it merely a reflection of the latter? We explore this issue by studying the experience of Christianity in China during the late 19th and early 20th century. Despite of the arduous process, the fast spread of Christianity, especially among the less developed and isolated regions of China, open a new window for the local people to see the outside world. Such cultural interactions generate persistent and profound impacts on regional development. Using county-level datasets on Christian activities in 1920 and socioeconomic indicators in 2000, we find a robust positive relationship between the intensity of Christian activities and today’s performances. To better understand if the relationship is causal, we review historical documents on the location pattern of missionaries’ selection and use instrumental variable (IV) approach. Our IV estimates confirm and enhance the OLS results. Together the empirical evidence suggests that increased Christian activities had a positive effect on long-term socioeconomic development. We further investigate the transmission channels between history and today, and find that Christian activities affect today’s performances through improvement in education, health care, as well as other unobserved mechanisms.

Keywords: Christianity, Economic Growth, Education, Foreign Direct Investment, China

JEL Classification Numbers: I20, N15, N35, O11, O43, Z12

Preliminary; For Seminars and Conferences only; No Dissemination or Citation Please “The missionaries …have been among the pioneers of civilization.” --- President William McKinley, 19001

1. Introduction

Does culture, and in particular religion, exert an independent causal effect on politics and the economy, or is it merely a reflection of the latter? This question is the subject of a long-standing debate in social science, with Karl Marx and Max Weber among its most famous proponents. The former famously opined that while the economy did influence culture, the reverse was not true. The latter, on the other hand, rejected that view and insisted that causality runs both ways. In particular, in “The Protestant ethic and the Spirit of Capitalism”, Weber claims that Reformed Protestantism, by nurturing stronger preferences for hard work and thriftiness has led to greater economic prosperity. Empirical test of the relationship between religious beliefs, with their impacts on moral reasoning and the related behavioral incentives, and economic growth is intimidating. Indeed, religions, formal institutions, and economic performance are often entangled with each other in one society. As a result, it is very challenging to examine the causal effects among these socioeconomic variables. In recent years, scholars begin to use unique historical examples to examine the causal effects of religion on long-run economic developments and provide empirical tests for the theories of Marx and Weber. Using county-level data from late 19th century Prussia, Becker and Woessman (2009) exploit the initial concentric dispersion of the Reformation and use distance to Wittenberg as an instrument for Protestantism. They find that Protestantism led to higher level of economic prosperity. But the channel of this causal effect was not work ethic or religious ideology, but rather education and literacy thanks to reading the Bible. Taking German Lands of the Holy Roman Empire as an example, Cantoni (2009) provides another test for Weber’s hypothesis. Using German population figures in a dataset comprising 272 cities in the period of 1300 to 1900, his OLS and IV estimations find no effects of Protestantism on economic growth. Other studies also obtain mixed results on the role of religion on long-run economic performances (e.g. Glaeser and Glendon, 1998; La Porta et al. 1998; Lipset and Lenz, 2000; Putnam, 2000; Stulz and Williamson, 2001; Sacerdote and Glaseser, 2001). In this paper, we use the lens of history to better understand this fundamental question by exploiting a quasi-natural experiment in China. In particular, we want to find out how the spread of Christianity in China in the late 19th and early 20th century has affected the long-term economic performance today. The Qing government forbade Christian activities until being defeated by Britain in two opium wars and signing the Treaty of Tianjin and the Treaty of Peking in 1858 and 1860 respectively.

1 The president address for the Ecumenical Missionary Conference in the Carnegie Music Hall, New York City, April 21, 1900. The widespread of Christian activities after 1860 caused serious conflicts with the Chinese central and local governments, the elite class and ordinary people. Such conflicts culminated in the Boxer Rebellion in 1900, when hundreds of Christian missionaries and thousands of Christian converts were killed. This rebellion led to a war between the Qing government and the Eight Powers. The Qing government was defeated again and was forced to sign the humiliating Boxer Protocol in 1901. Since then, Christian activities in China have been well protected. There are two reasons why China’s experience is important for testing the role of religion on long-run economic performance. First, China provides a perfect example to study within-country variations in religious activities and economic performances. Many existing studies have already compared the outcomes of religious activities in different countries. These cross-country studies are illuminating. However, formal and informal institutions as well as natural endowments could vary a lot across different countries. Consequently, the internal mechanisms of religion and economic growth in these studies are sometimes murky. Despite of this, due to data availability, few studies has explored within-country variations.2 We contribute to the literature by using China to conduct such a study. China is a large country with relatively homogeneous culture and consistent political system but quite heterogeneous economic performances. This setting provides variations large enough to conduct a clean empirical test in order to examine the role of religion on long-run economic performances. Second, China was dominated by Confucianism for thousands of years. Christianity was illegal and marginal in most periods of the ancient Chinese society. It became legalized and wide-spread in China since the late 19th century as a result of foreign pressures. Since then, foreign missionaries came to spread the God’s gospel in different places of China, and their location choices were influenced by various factors and historical randomness. Therefore, the spread of Christianity in China was to a large extent a quasi-natural experiment. This feature allows us to examine the long-lasting effects of a sudden and exogenous historical event on the variations of economic performances eighty years later. On this regard, our research contributes to the large body of literature of the effects of historical events on current economic developments, such as Acemoglu, Johnson and Robinson (2002; 2005) and Nunn (2009). Our paper provides the first empirical evidence on Christian activities on long-run economic development in China. We construct a dataset mapping the historical Christian activities and current economic indicators at the county level and find that the diverse socioeconomic performances across different counties in 2000 are positively correlated with the degree of Christian activeness at the beginning of the 20th century. However, such correlation might be subject to endogeneity issue due to unobservable county level heterogeneities. It is possible that those more economically and socially developed counties that managed to attract more Christian activities in the past tend to continue to perform better in the present. In this case, we

2 Becker and Woessman (2009) and Cantoni (2009) are exceptions. might obtain a positive correlation between past Christian activities and present economic outcomes, even though the former do not have any causal effect on the latter. We pursue a number of strategies to better understand the reason behind the relationship between Christian activities and current socioeconomic development. First, we review the evidence from historians on the nature of selection into Christian activities. Historical archival evidence shows that it was actually less developed areas of China that tended to be the hotbed for more intensive Christian activities. We then discuss the logics behind this seemingly paradoxical relationship in detail. Second, we use instruments to estimate the causal effect of Christian activities on subsequent economic development. The instruments are the frequency of floods and droughts between 1900 and 1920. Like the OLS coefficients, the IV coefficients are positive and significant, suggesting that increased intensity of Christian activities leads to better subsequent socioeconomic performances. We further explore the precise channels of causality underlying the relationship between Christian activities and socioeconomic development. Using historical evidence as a guide, we manage to confirm that higher intensity of Christian activities results in better education and health conditions today. However, there might still be other unobserved channels between Christian activities and today’s socioeconomic development that are not explained human capital stories. In the next section, we provide some historical background on the spread of Christianity during the late 19th and early 20th century. Section III describes the data we use to study the long-term effects of the spread of Christianity. Section IV provides OLS estimations to assess the relationship between the spread of Christianity and today’s economic performances and then use an instrumental variable approach to address the issue of causality. Section V and VI explore the improvement in education and health care by Christian missionaries respectively. Section VII studies the channels that improved human capital persisted to today. Section VIII studies other channels not captured by education and health care that missionaries have exerted influence in. The final section discusses the implications of our findings and concludes.

2. Historical Background

Christianity came to China as early as in the 17th century, but was never a major religion in this Confucianism-dominated country. It was recorded that missionaries came to China since 635 AD. As Europe –China’s bilateral connections became more active, Roman Catholic Church expanded rapidly in China. It was estimated that there were more than a hundred thousand Christian converts in China in the mid seventeenth century. However, Pope Clement XI forbade Chinese Christians to engage in Confucius- related activities in 1704, which outraged Emperor Kangxi. As a result, he completely banned Christian activities in 1720. This policy was reinforced by the successive emperors. Since the early 19th century, the trade between Europe and China reached a historically high record, and Christian missionaries attempted to penetrate China again. But as Christian activities were still banned by the Qing court, Christian presence was still negligible up to the 1840s. China was forced to open up to the western powers when China was defeated by Britain in 1842. As a result, Christian missionaries were allowed to live in China in 1846. Both the Roman Catholics and the Protestants began to revive gradually in China. Defeated again in 1860, the Qing government signed the Convention of Peking which granted freedom of religion in China and allowed the missionaries to own lands and build churches. Since then Christian missionary activities became legal, and a large number of missionaries came to China from Europe and America, spreading across many provinces in China. By 1900, there were more than 80 thousand Protestant converts in China (Wang, 1991). The rapid expansion of Christianity led to a great number of conflicts and oppositions in China. There are three reasons of these conflicts and oppositions. First of all, the monotheism of Christianity was not accepted by the polytheism and ancestor worship in the Chinese culture. Second, the egalitarian tradition of Christianity clashed with the entrenched hierarchical tradition in the Chinese society. More generally, western customs accompanying the expansion of Christianity were drastically different from Chinese customs. Mistrust between Chinese people and foreign missionaries accumulated and led to frequent conflicts. Such conflicts culminated in the Boxer Rebellion in 1900. In this tragic xenophobic conflict, more than 20 thousand Christians were killed. This rebellion led to a war between the Qing government and the Eight Powers. China was defeated again and was forced to sign the Boxer Protocol in 1901 in which the government guaranteed safety of the missionaries. Since then, Christian activities in China received good protection. The first two decades of the 20th century witnessed the biggest expansion of Christianity in China. 1,500 foreign Protestant missionaries arrived in China in 1900. The number climbed to 3,445 in 1905 and 8,000 in 1927. More than half of these newly arrived missionaries were from the US. There were about 80 thousand Protestant converts in 1900, and 130 thousand in 1904. In 1922 this number soared to 402,539. By the time of 1920, there were more than 120 Protestant denominations in China, and church activities penetrated nearly 70 percent of the counties (Wang, 1991). The unprecedented expansion of Christianity in China in the early twentieth century was not purely a consequence of improved security condition. The changed missionary strategy after 1900 was a major cause of the surge of Christian activities in China. Learning the lesson from the tragedies in the Boxer Rebellion, many Christian organizations came to realize that the hostility toward Christianity was largely caused by Chinese people’s ignorance of western civilization and mistrust of foreigners. As a result, the missionaries began to take measures to build trust between foreigners and Chinese, and disseminate western science and technology to all walks of life in Chinese society. An important way of building mutual trust was disaster relief. Timothy Lee, a famous missionary in China in the late nineteenth century, once said that disaster relief was “an ideal way to reduce prejudices and prepare the ways for the Chinese to accept Christianity” (Gu, 2010). China is a country frequently inflicted by natural disasters. In the long history of fighting with disasters, China has developed a set of quite sophisticated disaster relief mechanisms. However, as the central empire got weakened in the late nineteenth century, the government’s disaster relief mechanisms gradually collapsed. Noticing the absence of government support, missionaries began to raise fund and provide food and other resources to people in drought-stricken areas. Between 1876 and 1879, a catastrophic drought hit five provinces, leading to a famine in which about one million people died of hunger. Over 100 missionaries went to those areas and helped those in need (Gu, 2004, p289). On Jan 26, 1878, they even set up a disaster-relief funding committee to coordinate all the Christian relief activities. This was the first Christian relief organization in China. Since this severe drought, Christian missionaries began to actively engage in almost every natural disaster in China. In 1888, a major flood and the subsequent cholera struck millions of people in northern China, and missionaries helped hundreds of thousands of people in the flood. In the Yellow River flood in 1899, missionaries hired thousands of flood-hit workers to build river banks, bridges and roads. Christian disaster relief became even more active in the early twentieth century. In 1920, northern China was struck by another severe drought. In order to coordinate international disaster relief efforts, missionaries set up a nation-wide organization, China International Famine Relief Commission and raised over 30 million US dollars for the famine. In the efforts of saving life in the disaster-hit areas, missionaries realized that tremendously lagged-behind science, technology and infrastructure in China were the key impediments for disaster relief. In order to better relieve and prevent disasters, missionaries actively disseminated knowledge of modern science and technology in China. Christian missionaries gradually found out that the underdeveloped education system was the fundamental source of all the problems they found in disaster reliefs. As a result, they called for fundamental education reform in China, trying to persuade Chinese government to give up the outdated Confucius education system and focus on natural science. Moreover, many missionaries put the idea of education reform in practice, and established a number of elementary schools, middle schools, and universities across China. Many of these Christian institutions soon became quite prestigious, and some survived the political turmoil in the ensuing years and were still active even today. Due to lack of modern medical knowledge and functioning public health system, cholera and plague usually killed numerous people when disasters took place. For this reason, missionaries made great efforts in establishing hospitals and promoting medical knowledge in those areas. All these practices by the missionaries gradually changed people’s attitude on science and openness, and generated profound and long-lasting effects in the Chinese society. . 3. Data To examine the impacts of Christian activities on long-run economic performances, we assemble three county-level data sets: Christian activities and socioeconomic conditions in 1920; socioeconomic conditions in 2000; and geophysical conditions of all the counties in the data. In order to match historical data with contemporary data, we have to handle the issue of changing county borders with GIS method. We explain the details in this section. 3.1 Historical Data Our data of Christian activities in China in 1920 are obtained from a published statistical report: The Christian Occupation of China: A General Survey of the Numerical Strength and Geographical Distribution of the Christian Forces in China ("COC" afterwards). The report was compiled by the China Continuation Committee, the central organization of Christian churches in China. One main task of this organization was to coordinate and promote more effective evangelization in China. For that purpose, beginning from 1918, the Committee spent three years in conducting a county-level survey on the status quo of Christian activities in China and published the survey results in “COC”. “COC” reports various measures on Christian activities such as numbers of missionaries, converts, churches, and Christina schools and hospitals in Chinese counties. It also collects and publishes other important socioeconomic information of surveyed counties, such as political orientation, population, education condition. The Committee made special efforts to ensure the quality of survey data in order to deliver a reliable and objective description of the situation of Christian influence in Chinese counties. Therefore, COC is considered to be a good source to study socioeconomic conditions in Chinese counties in the early twentieth century. [TABLE 1 ABOUT HERE] In this paper we measure the intensity of Christian activities in each county by three indicators: the numbers of Christian converts, churches and vicars for every one thousand people. The summary statistics are reported in Table 1. On average there were 0.76 converts, 0.017 churches, and 0.025 vicars per 1,000 people in each Chinese county in 1920. We also show the density of Christian converts in Figure 1. Each grid represents a county in China. Darker grids are the counties with higher densities of Christian converts. Although many coastal counties had higher density of Christian converts, the figure suggests that there were also active Christian presence in quite a number of inland counties. Therefore, our data suggest that the different intensity levels of Christian activities were not simply driven by the distances of these counties to the coastline. [FIGURE 1 ABOUT HERE] 3.2 Contemporary Data We assemble contemporary socioeconomic indicators of Chinese counties to study the long-term impacts of Christian activities. The original sources of these data are various surveys conducted by the National Bureau of Statistics of China. The county-level GDP and educational expenditure data are obtained from the Statistical Materials of Public Finance of Cities and Counties. Demographic variables (e.g., population and mortality rate) and variables of education outcome (e.g., years of schooling and literacy rate) are drawn from the Fifth National Population Census. The main dependent variable in our paper is GDP per capita in 2000. The mean of GDP per capita is 5,445 yuan (about $778). Figure 2 reports GDP per capita of all the counties in our data set, and each grid represents each county. The darker the grids are, the higher GDP per capita the counties have. Similar to Figure 1, it is evident that not all rich counties are coastal ones. Quite a number of inland counties also have high GDP per capita. [FIGURE 2 ABOUT HERE] 3.3 Climate Data Our historical climate data are drawn from the Distribution Gallery of Droughts and Floods in the Past Five Hundred Years of China. This gallery was compiled by the Institute of Synoptic Meteorology and Climatology under the administration of China Meteorology Administration (“CMA” hereafter). The original sources of historical climate data are county gazetteers and official archives. Combining historical and contemporary climate data, this gallery allows us to build a data set of droughts and floods for 120 stations from 1470 to 2000. For each station in each year, the gallery uses a 5-scale Drought-Flood Index (“DF-index” hereafter) to categorize local climate: 1 for serious flood, 2 for flood, 3 for normal condition, 4 for drought, 5 for serious drought.3 This data set has been widely used in various studies, and its consistency and reliability have been carefully examined and confirmed by many meteorologists (Yao, 1982; Ronberg and Wang, 1987). In our paper, we count the incidence rate of DF-index=1 or DF-index=2 for a station over our sample periods and use it as a measure of flood frequency.4 Similarly, we construct drought frequency by counting the incidence rate of DF-index=4 or DF-index=5.5 Our climate data are collected from only 120 stations. We use a conventional method to convert station-level drought and flood frequencies into county-level variables: the inverse-distance-weighted method (“IDW” method hereafter). This method assumes that a county’s climate is an average outcome of the climates of all nearby stations, weighted by the distances from this county to nearby stations. We report the details of processing the climate data in the appendix.

3 See Zhang and Crowley (1989) for detailed description of this categorization method. 4 In different specifications we test our results for three periods: 1800-1840, 1900-1920, and 1978-2000. 5 Using incidence rate of DF-index=1 for flood and incidence rate of DW-index=5 fordrought does not alter the main results. 3.4 Geophysical Data In our paper we include county-level geophysical data as control variables in our regression specifications. These data are drawn from the Surface Meteorological Database constructed by the CMA. The database contains annual precipitation and temperature for 754 meteorological stations between 1990 and 2000. We use the same IDW method mentioned in the previous subsection to convert station-level variables to county-level ones. 3.5 County Matching A big challenge in our study is that the territories of many counties changed because of regime changes over this long historical period. We solve the problem by comparing the GIS data for both 1920 and 2000 and converting county-level variables in 1920 to those in 2000 using overlapping area as weights. Historical GIS data are obtained from the China Historical GIS Project by Harvard University and Fudan University.6 County GIS data in 2000 (including longitudes and latitudes) are obtained from the ACASIAN Data Center at Griffith University in Brisbane, Australia. Average altitudes of these counties are obtained from the SRTM 90m Digital Elevation Data (Jarvis et al., 2008). As a result, we build a data set combining historical Christian intensity data, historical and contemporary climate data, contemporary socioeconomic data and geophysical data for 1,743 counties in China proper.

4. Empirical Results 4.1 Baseline Correlations We first study the simple correlation between county-level Christian intensities in 1920 and economic performances in 2000. Figure 3 shows the correlation between log numbers of converts per 1,000 people in 1920 and log per capita GDP in 2000 for all these counties. It is evident that these two variables have a strong positive correlation. [FIGURE 3 ABOUT HERE] We further examine this relationship with OLS regressions controlling for other county-level characteristics. The baseline specification is given by:

' ln(per capita GDP 2000)i = α + β‧ln(Christian activityi)+Xi ‧γ+ εi ,(1) where ln(Christian activityi) is the log Christian converts per 1,000 people in 1920; Xi is a vector of control variables including county locations, climates, and other geophysical characteristics; and εi is the error term. Table 2 reports the results. Column (1) contains only the variables of interest. This result reflects the relationship shown in Figure 4. In Column (2) we include a set of regional fixed effects.7 While the magnitude of the marginal effect is reduced, the

6 The historical county-level GIS data can be downloaded at http://www.fas.harvard.edu/~chgis/. 7 China proper is divided into 7 regions: North China includes Beijing, Tianjin, Hebei, Shanxi; East China correlation between Christian activities and GDP per capita remains positive and significant at 1% level. In Column (3), we control for a set of county geophysical variables, including longitudes, latitudes and altitudes, and distances to provincial capitals. We further control county climates by adding in precipitation and temperature in column (4), and contemporary frequencies of extreme weather (1978-2004) in column (5). The coefficients on Christian activities remain significantly positive. [TABLE 2 ABOUT HERE] 4.2. Establishing Causality: Instrument Variables Results The positive correlation between Christian activities in 1920 and contemporary economic performances in 2000 reported in Table 2 does not guarantee a causal relationship. One major concern for causality is that selection of counties for preachment in 1920 was not random. It is possible that missionaries preferred to preach in richer counties in 1920 and these counties are still richer today. In this subsection we study the location choices of Christian preachment before 1920 from different perspectives. In fact, historical evidence suggests that missionaries were more likely to go to less developed regions. Therefore, the OLS regressions shown in Table 2 underestimate the long-term effects of Christian activities on economic performances. A direct way to test if the missionaries were more active in the less developed counties is to examine the economic indicators of those counties in 1920. However, direct measures of historical economic performances are unavailable. Instead, we use two proxies for historical economic conditions: population density and per capita land tax revenue. Population density is a commonly used, although quite crude, proxy for historical economic prosperity.8 Per capita land tax revenue can also largely reflect economic conditions in Chinese counties. Figure 4 plots these two measures against the number of churches and vicars per 1,000 people for Chinese counties in 1920.9 All figures show significant negative correlations, suggesting that Christians were more active in less developed regions. [FIGURE 4 ABOUT HERE] In order to rigorously examine the causality between historical Christian activities and contemporary economic performances, we use instrumental variable technique to formally address the potential selection problems. As described in Section II, since the beginning of the twentieth century, Christian organizations decided to use disaster relief as a main method to gain trust from Chinese people and includes Shanghai, Jiangsu, Zhejiang, Shandong, Anhui; Northeast China includes Liaoning, Jinlin, Heilongjiang; Middle China includes Hubei, Hunan, Henan, Jiangxi; Southern China includes Guangdong, Guangxi, Hainan, Fujian. Southwest includes Sichuan, Chongqing, Guizhou, Yunnan. Northwest includes Shanxi, Gansu, Xinjiang, Qinghai, Ningxia. 8 See Acemoglu, Johnson and Robinson (2002) for example. 9 We use the number of missionaries and the number of churches in each county because we want to measure the location choices of Christian missionaries before 1920. As a robustness check, we also use the number of converts and find similar results. better spread the God’s message. In light of this fact, we construct county-level incidence rate of droughts and floods and use them as instruments for intensity of Christian activities. [TABLE 3 ABOUT HERE] We report the 2SLS estimates in Table 3. In Column (1) of Table 3 we report the results without any control variables. In column (2) we control regional fixed effects. In column (3) we add the geophysical control variables. We further include contemporary climates in column (4) and contemporary natural disasters in column (5). Second-stage estimation results are reported in Panel A. Coefficients on ln(convert) remain positive and significant across different columns, ranging from 0.13 to 0.17. These results are significant not only statistically but also economically. The results in column (5) show that one standard deviation change in ln(convert) causes 0.43 standard deviation change in log GDP per capita in 2000. For a county with the mean of income at 5,419 yuan, one more convert per 1,000 people causes an increase of 1140.8 yuan (about 175 US dollars) in per capita GDP. In addition, the magnitudes of the 2SLS coefficients are significantly higher than those of OLS. This confirms our conjecture of underestimated impacts of historical Christian activities because Christians were more active in less developed counties. The first-stage results are reported in Panel B. The coefficients on both flood and drought frequencies are significantly positive, suggesting more Christian activities in the counties with more disasters. The F-statistics of the instrument variables is higher than the critical value suggested by Stock and Yogo (2005), which rules out weak instrument concern. The validity of our instruments relies on the assumption that extreme weathers one hundred years ago did not directly affect current economic performances. Despite that all of our specifications pass over-identification test at conventional statistical levels, one might still worry about persistency of extreme weathers over time. Counties with more floods or droughts in history might also suffer from more floods and droughts today, which in turn adversely affect current economic performances. In fact, this channel is unlikely. Different from one hundred years ago, China’s economic growth today relies more on manufactures, thus insensitive to weather shocks. In addition, with advanced communication, transportation and weather forecasting technologies, the impacts of extreme weather on today’s economy are much smaller than those in history. To confirm that, in Table A1 we regress land tax per capita in 1820 and GDP per capita in 2000 on drought and flood frequencies in 1800-1820 and 1978-2000, respectively. Consistent with our expectations, in the specifications with full controls, the coefficients of disaster frequencies are significant in the regression with land tax in 1820 (Column 2 of Table A1), but insignificant in the regression with GDP per capita in 2000 (Column 4 of Table A1). We further formally test the validity of our instruments in our econometric specifications. Column (5) in Table 3 tackles the validity concern by directly controlling contemporary disasters. Compared with other specifications, this regression delivers a much higher P-value in the Sargan over-identification test. This result strongly supports the exclusive restriction of our instruments. Meanwhile, the coefficient on ln(convert) is barely affected. 4.3. Robustness Checks In Table 4 we conduct robustness checks for our estimations. Column (1) of Table 4 addresses the concerns of potential outliers by omitting potentially influential observations.10 Compared with our preferred specification in table 3, the coefficient on convert density is unaffected. [TABLE 4 ABOUT HERE] Coastal counties are easier to access than hinterlands in the past. Furthermore, coastal regions usually enjoy more preferred policies in present. Therefore, one might wonder if the long-term impacts of Christian activities are merely driven by coast-hinterland disparity. In column (2) we examine this issue by taking out all counties in coastal provinces. The effects of Christian activities on long-run economic performances become even stronger among the hinterland sample. A small number of counties in our full sample are in fact big cities. For both geographical accessibility and economic policy reasons they could be positively correlated with both Christian activities and present economic performance. As a robustness check, in column (3) we drop the urban counties and focus only on the rural ones. Our estimation results on the long-term effects of Christianity are barely changed. Instead of using instruments to tackle the selection issue, in column (4) of Table 4, we also report the results by using historical land tax as a proxy for the initial economic conditions and include it directly into the regression as a robustness check. The effects of Christian activities become slightly bigger than those in the main specification (column 5 in Table 3), which is reasonable due to the nature of selection bias discussed in Subsection 4.2. We also use another proxy, population density in 1920, and find similar results. In this paper we focus on the effects of Protestants instead of Catholics on economic development. Their practice strategies and influences in China were quite different. In absence of detailed survey for Catholics activities in that period, we collect information of Catholic activities from the COC and use it as a control variable in our robustness check. In particular, we count the number of Catholics mission stations for each prefecture. The result is reported in column (5) of Table 4. Coefficient on the Christian converts decreases by 20% but remains positively significant. In the main specification we use the number of converts instead of churches or

10 “Potentially influential observations” are identified using Cook’s distance, which measures how much each of the estimated coefficients change when each observation is deleted. If an observation has a Cook’s distance greater than 4/N, where N=1743 is our sample size, it is excluded from regression in Column (1) of Table 4A. vicars to measure the intensity of Christian activities in each county. We do so because the number of converts captures the outcome of Christian preachment, while the numbers of churches and vicars captures the inputs. As robustness checks, in columns (6) and (7) of Table 4 we report the 2SLS results using numbers of churches and vicars, and find that our main results are not affected. We also find that the coefficients of churches and vicars are bigger than that of converts. This is because one additional church or vicar should yield much larger effects than one additional convert.

5. The Effects of Christian Activities on Education

We now discuss the channels through which historical Christian activities affected China’s economic performance today. We find that Christian missionaries were pioneers of disseminating western science and technology in China, and the main channel of doing that was promoting western education system. The Christians’ efforts in promoting education generated profound and long-lasting effects in the economic development of China in the subsequent years. China completely rebuilt the education system in the early twentieth century. One of the most critical catalysts of this reform was Christian promotion in modern education. Since the late nineteenth century, missionaries began to build Christian schools in China, and profoundly impacted the old Confucius education system. Exactly as John K. Fairbank noted, “in the end the Christian influence was probably strongest in education” (Fairbank, 1974, p13). Christian missionaries began to realize the fundamental flaw of traditional Confucius education soon after they came to China in the early nineteenth century.11 In the process of disaster relief, missionaries found that many Chinese people, including some elites, knew very little about western science and technology. Lacking scientific knowledge prevented effective disaster reliefs in many places. Missionaries believed that traditional Confucius education was the root of ignorance of modern science and technology, and building western education system became an imperative task. For instance, a well-known missionary, Timothy Lee, wrote a proposal to the governor of Shanxi Province when he was conducting disaster relief in 1884, emphasizing the importance of reforming China’s education system. He said, “Education is the first priority for China. As the western countries keep developing their education day-to-day, China will lose his chance to overtake them in 10 years. In conclusion, education is the most significant and urgent thing for China” (Lee, 1889). Christian missionaries determined to build new school system in China for the purpose of not only educating Chinese, but also converting Chinese. They believed that western education was crucial to dissipate mistrust and misunderstanding of

11 The first one to criticize the traditional Chinese education system is E. C. Bridgman, the first American missionary who came to China in 1830 (Barnett and Fairbank, 1985, p. 100). Chinese people toward Christianity. Alvin Pierson Parker, chairman of the Educational Association of China, explicitly expressed this opinion in 1896, “as a Christian educators’ association, we should play a dominating role in China’s education reform and satisfy the interests of Christianity” (Educational Association of China, 1896). Driven by the motives of educating and converting Chinese, Christian missionaries made tremendous efforts in building western schools in China since the late nineteenth century. Between the 1870s and 1920s, the number of Christian schools rose sharply from 347 in 1877 to 7,382 in 1922, almost tripling every twenty years (Gregg, 1946, p.16-17). Among all these 7,382 Christian schools in 1922, 6,599 (86%) were elementary schools, 291 (7%) were middle schools, 16 (0.2%) were colleges, and 75 (1%) were vocational schools (China Educational Commission, 1922, p. 416).12 Primary schools are the foundation of modern mass education. To empirically test the effects of Christian activities on the development of western-style primary schools, we regress, at the county level, the number of students per 1,000 people in Christian primary schools on the density of Christian converts,13 the latter of which was instrumented by frequencies of historical droughts and floods. We report the results in column (1) of Table 5. Given that our dependent variable is censored at zero, we estimate a Tobit model instead of OLS.14 The IV result in panel A indicates a statistically and economically significant effect of Christian activities on the enrollment rate of the Christian primary school: transforming the coefficient into marginal effect (evaluated at the sample mean) suggests that one standard deviation in convert per 1,000 people increase the enrollment rate of Christian primary school by 0.92 standard deviations. [Table 5 is about here] Christian missionaries were not only building more schools, but more critically, different schools. These schools were completely different from the old Confucius schools. They introduced new curriculums which put a big emphasis on both natural sciences, such as mathematics, physics and chemistry, and social sciences, such as law and economics. Practical subjects such as foreign language studies and engineering were also included. English, for example, was offered to 58% of the students in the 4th or 5th grade (Stauffer, 1922, p. 1075). Class lectures were combined with practices and laboratory experiments. Another big advantage of Christian schools was stable and sufficient financial support. Stauffer (1922) wrote a special report on the status of Christian schools in China based on a survey conducted by the Chinese Education Commission in 1920. According to this report, over half of the funding of Christian schools came from

12 The rests are special institutions, including orphanages and schools for the blind or deaf. 13 If one assumes that portion of school-aged children is constant across counties, then this result can be viewed as clear evidence for the casual effects of Christian activities on the enrollment rate of the school-aged children in Christian elementary schools. 14 The OLS results are not significantly different from the Tobit ones. foreign donations (Stauffer, 1922, p. 1094).15 Continuous and sufficient funding prevented Christian schools from being disrupted by frequent political turbulences in the early twentieth century. By contrast, public schools were often plagued by insufficient financial support from the government. For example, in 1911, the education budget accounted for merely 1.5% of the total government spending (Stauffer, 1922, p. 1068).16 In the early 20th century, teachers and administrators in the public schools, and even staffs of the Ministry of Education frequently stroke and protest to enforce the payment of arrears in salary (China Educational Commission, 1922, p. 22). Better schools educate better students. Take student promotion rate as an example. In 1920, 21% of the Christian junior elementary school students entered senior elementary schools (as opposed to 10% for the public school students), and 10% of the Christian senior elementary school students entered middle schools (as opposed to 3.4% for the public school students) (Stauffer, 1922, p. 404). Moreover, Christian schools educated a large amount of professionals who were urgently needed by the society. Stauffer (1922, p. 409) reports a surveyed on 5,500 high school graduates in 1918: 30% of them continued their study in college; 30% of them worked for churches; 20% of them went to teach in other schools; and the rest of 20% went for business or other professions such as doctor, nurse and lawyer. By contrast, statistics showed that 70% of the graduates from public middle schools had difficulty in finding positions (China Educational Commission, 1922, p. 19). In 1920, there were 6,700 Christian schools, while the total number of public primary schools already reached 125,000. But one should also consider the following fact before being able to appreciate the impacts of Christian education: there were 0.8 converts for every 1,000 residence, yet 2.2 Christian schools students for every 1,000 school-aged children. One of the reasons a small group of Christian missionaries and converts in China can yield profound influence on the long-run development is that they set a model for many Chinese to follow. In fact, this was one intention for which Protestants promoted modern education in China. As being said in one of the reports from a missionary education association of China, “by perfecting and strengthening this arm of the service (Christian schools), we increase the probability that the future governmental educational system of China will be largely influenced and molded by such superior examples.” (Silby, 1902, p 621) Wang (1997) documents three spillover effects of Christian schools to the modernization of Chinese education system. First, combining western origins with Chinese reality, most textbooks written or edited by missionaries were well customized to suit the needs of Chinese students. Therefore, they were widely adopted by domestic schools (Wang, 1983, p112). Second, as previously mentioned, a large amount of graduates from Christian schools chose to become teachers in domestic schools, where they taught foreign language and science that are in urgent need of

15 31% came from tuition fees, and the rest of 18% came from domestic donations. 16 60% of the government budget was diverted to military purpose and war indemnity. those schools.17 Third, aside from textbooks and curriculums, both the education system and management model in Christian schools were adopted by Chinese education reformers.18 Empirically, we corroborate the spillover effects of Christian schools in China with two pieces of statistical evidence. First, we investigate, within a county, whether more intensive Christian activities induced higher enrollment rate for public elementary schools. The results in column (2) of Table 5 support this conjecture. In particular, our IV Tobit coefficient implies that one standard deviation in convert per 1,000 people increases the enrollment rate of public elementary school by 0.70 standard deviation. In other words, 1 more convert per 1,000 people increases the enrollment rate of public elementary schools by 0.58% (the mean value is 1.19%). Next, we examine, across counties, whether the enrollment rate of public elementary schools in one place was influenced by the Christian activities in the nearby counties. Specifically, consider the following specification:

ln(Student/pop1920i)= α +β1 ln(Christianityi)

' +β2 ln(Christianity_Contiguousi)+Xi γ+ εi , (2) where the enrollment rate of elementary schools in county i (Student/pop1920i)is regressed on both the Christian activities in its own location (Christianityi) and the

Christian activities in the neighboring locations (Christianity_Contiguousi). The latter is constructed by averaging the convert densities in all counties contiguous to county i, weighted by the distance of each of these counties to county i. The instrument for this variable, which is the average frequency of droughts and floods in the nearby counties, can be constructed in a similar fashion.

Column (3) in Panel B reports the OLS results of β1 and β2. They are both significantly positive, and the effects of Christian activities in the own counties are larger than those in nearby counties (β1>β2). However, once we estimate the model with instruments (column (3) in Panel A), the neighboring effects disappear. We further investigate this issue by dividing the overall sample into two groups: counties with positive number of converts and counties with no converts. Their results are reported in column (4) and (5) of Table 5, respectively. Notice that in column (5), only

β2 can be estimated by the construction of the subsample (Christianityi=0). Clearly, the across-county spillover effects of Christian on elementary school enrollment concentrated on counties without any Christian activities. Specifically, for counties without any convert, its elementary school enrollment rate will increase by 0.97 standard deviation if the convert population density in its contiguous counties increases by 1 standard deviation. That is to say, increasing 1 convert per 1,000 people

17 For example, graduates from Tengchow College, a famous school established by the American Missionary Calvin Mateer, showed high competiveness in their job markets among Chinese schools (Wang, 1983). 18 For example, Zhang Zhidong, the Huguang Governor since 1896 who was considering to establishing new-type schools in Hubei province, had sent many of his officials to Boone Memorial School, a famous Christian academy established by the United States Episcopal Church in 1871, to study and learn from its education system and management model (Wang, 1983, p113). leads to the increase in the enrollment rate by 1.61% (the mean value is 1.06%).

6. Health Care

6.1 Hospitals and Clinics Herbal therapy was the main form of medical treatment in China until western medical science was introduced in China in the nineteenth century. Christian missionaries played a critical role in China’s transition from traditional medicine to western medicine. Peter Parker, an American missionary, built up the first modern (ophthalmic) hospital in China in 1837. More than 2,000 patients received treatment in the first year of its operation (Bush, 1879). Modern medical system gradually emerged in China since then. By 1889, 61 Christian hospitals and 44 clinics were operating in China.19 This number more than doubled in the first two decades of the twentieth century, reaching 326 and 244 respectively (Stauffer, 1922, p96). These Christian hospitals and clinics operated in all the provinces in China proper. The annual number of inpatients soared to nearly 150 thousands and that of outpatients over one million (Stauffer, 1922, p623). John K. Fairbank, a renowned historian, commented on this, “Modern Western medicine in China was to an important degree a consequence of missionary demonstration and instruction” (Fairbank, 1983). Figure 5 presents the correlation between Christian activities and the numbers of hospitals at provincial level.20 The positive correlation is very strong: 1% increase in the number of converts is associated with 1% more in modern hospital. [Figure 5 about here] With strong support from Christian organizations, these Christian hospitals were usually equipped with advanced facilities and were well financed. Dr. Harold Balme, the dean of the School of Medicine in Shandong Christian University, surveyed the situation of 165 Christian hospitals in 1919. The study found that most of the surveyed hospitals owned medical laboratories, 75 hospitals had the capability to conduct laparotomy, and 24 had been equipped with X-ray machines (Balme and Stauffer, 1920). 6.2 Public Health What the missionaries brought to China was not only western hospital system, but more importantly the idea of western public health. Whenever large epidemics broke out in disaster-struck areas, missionaries were usually foremost in organizing effective remedial and preventive measures. For example, in 1872, cholera struck Tianjin. Missionaries provided medical aid to people who were infected. In 1911-1912, a highly deadly pneumonic plague broke out in northeast China. A group of

19 Records of the General Conference of the Protestant Missionaries of China, held at Shanghai, May 7-20, 1890. (pp. 733), published by General Books LLC. 20 The main reason we cannot conduct this analysis at the finer geographical level is that the number of hospital in 1920 is only available at the provincial level. missionaries set up the Anti-Plague Bureau to fight with the plague. In frequent disaster reliefs, Christian missionaries found that lack of public health knowledge and consequent outbreaks of epidemics were main reasons of large death toll in the disasters. Disseminating knowledge of disease control became a priority in the disaster relief. The prevention measures included sterilization of medical equipment and tools, disinfection of food and drinking water, control of flies and mosquitoes, etc. Missionaries enforced quarantine in disaster struck areas and effectively controlled expansion of contagious diseases. According to a study by Dr. Balme in early twentieth century, 69 Christian hospitals (42% of total) were equipped with quarantine facilities (Balme and Stauffer, 1920). A good example is control of leprosy. Local governments and missionaries collaborated to enforce quarantine lepers, and the number of leprosy cases reduced drastically in the early twentieth century (Stauffer, 1922, p437-438). Health conditions of women and children were critical in public health. Christian missionaries worked hard to help improve health conditions of women and children. Female missionaries played a critical role in disseminating knowledge of obstetrics and gynecology among Chinese women. They made regular visits to pregnant women and provided medical services in case of necessity.21 Because of enormous efforts of female missionaries, obstetricians, midwives and nurses began to practice in many places in China, resulting in significant decline of women and infant mortality in the early twentieth century. 6.3 Medical Education Christian missionaries were also pioneers in promoting modern medical education in China. They believed that “scientific medicine in China must not continue indefinitely to be a ‘foreign doctrine’” and “the medical profession of China must become national if it is to be universally accepted” (Mac Alister, 1921). As a result, Christian organizations had established 116 medical education institutions by 1920. Among these institutions, 10 were medical colleges, which accounted for one third of all the medical colleges in China (Stauffer, 1922, p.425). Most of these colleges continued to operate today and trained a large amount of excellent doctors. The other 106 institutions were nurse schools (China Educational Commission, 1922, p. 416). Usually these schools were larger and equipped with better facilities than public schools. In fact, these Christian medical schools provided an advanced model of medical education that public schools were trying to follow. Setting up schools was not the only way that missionaries promoted medical education. They translated a large number of western medical books to Chinese. A famous missionary, John Glasgow Kerr, published the first medicine journal in China

21 For example, as early as in 1888, seven Canadian missionaries conducted a comprensive survey of health conditions of women and children in Zhangde Prefecture, Henan Province. Meanwhile, with the fund raised in the home country, they set up gyneacological clinics and provided free medical services to local women and children. They also provided monthly training class to women on the knowledge of obstetrics and gynaecology. See Song (2008, p. 31) for details. in 1868.22 A few years after that, the Medical Missionary Association of China, set up in 1886, published its own journal in 1888 with a focus on transmitting western medical knowledge to the Chinese. Missionaries published the first Chinese medicine dictionaries in 1908, which standardized the Chinese medical terms (China Mission Year Book. 1912, pp. 267-268). The legacy of these efforts lasts to today.

7. Persistency

The status of Christianity was overturned when the People’s Republic of China (PRC) was established in 1949. The government then was highly concerned that Christian missionaries were representatives of “western imperialism”.23 Consequently, the central government established the official Christian church system and severed their connections with foreign Christian organizations. By 1952, all foreign missionaries left China. The government took over all of the Christian schools and hospitals, and regulated all Christian activities since then. As the revolution progressed, Christianity, along with other religions in China, was smashed to the bottom. All religious activities, including the government-regulated churches, were banned during the Cultural Revolution (1966-1976). Christianity revived in China since the Reform and Open-up in 1978. Government-regulated Christian churches began to active again, and expanded rapidly in China. Christianity is now among the five biggest religions in China. Despite rapid growth, Christianity is still under tight regulation by the Chinese government. Religious services can only be practiced in the government-sanctioned churches, and these churches are not subsidiaries of any foreign church. Family fellowships are considered illegal, and can only practice services underground. The disruption of Christian mission in China raised a very critical issue in our analysis: how could the effects of Christian activities around 1920 persist to today, after all the political turmoils of the wars and the revolution? We test the channels of persistency in Table 6. In Columns 1-3 in Table 6, we use three variables to capture China’s education development: educational expenditure of local government normalized by population (Column 1), average years of schooling (Column 2) and literacy rate (Column 3) among the local residence. Similar to the previous section, we use frequencies of droughts and floods from 1900 to 1920 as instruments for log(convert/population) in 1920. The first stage results are not reported here because they are the same as that of the Column 5 in Table 3. [TABLE 6 ABOUT HERE]

22 John Glasgow Kerr (1824 – 1901) was a Presbyterian medical missionary. He came to China with the American Presbyterian Mission in 1854, and soon became the head of the Ophthalmic Hospital in Canton and later the Guangzhou Boji Hospital (The Canton Hospital). He worked there for 47 years and treated about 1 million patients. Besides his outstanding contribution in medical care, he was also famous for his student, Dr. Sun Yat-sen. 23 This point was stressed many times by the Premier Zhou Enlai in the meetings with Chinese Christian leaders. Panel A reports the second stage of the IV estimates. For the same reason discussed in Section 4.2, 2SLS delivers a higher magnitude than OLS does (reported in Panel B). The results in Columns 1-3 confirm that the past Christian activities have positive and significant effects on a county’s long-run education achievements. Take average years of schooling as an example, evaluating at the mean, 1 standard deviation increase in converts per 1,000 people in 1920 leads to 0.92 standard deviation increase in average years of schooling in 2000. Christian organizations help local communities build hospitals, introduce modern medical technologies, and improve public sanitation. Such endeavors changed people’s consciousness of public health and generated long-lasting effect in the improvement of public health conditions. We examine this effect by regressing two indictors of health conditions in 2000 on the Christian activities in the 1920s: child mortality rate in 2000 (Column 4), and public health expenditure during 2005 (Column 5). The results show strong positive effects of historical Christian activities on today’s health conditions. 1 standard deviation increase in converts per thousand people decreases the children mortality rate by 0.86 standard deviations. When evaluating at the sample mean, this amounts to 37% decrease in children mortality rate. Our empirical findings in Table 6 are consistent with what we see in reality. In fact, the complete disruption of Christianity only last for no more than twenty years (1950s to 1970s). During these two decades, people’s religious beliefs were suppressed, but not exterminated. The religious activities revived very rapidly since it became legal again in the late 1970s and 1980s. Many people who benefited from the Christian missionaries in the 1920s survived the wars and the revolutions. As a result, the effects of Christian missionaries eighty years ago persisted all the way to today.

8. Other Channels

Becker and Woosman (2009) argue that Protestantism affects long-run economic performance mainly through the improvement in literacy rate. Therefore, if one controls for measures of current education in the GDP regression, the effect of the religious activities should go away. In Table 7, we implement a similar exercise in our context. Particularly, we want to see if the influence of historical Christian activities on current economic performance works mainly through improvements in educational achievements or public health. [INSERT TABLE 7 HERE] We control for education variables in Columns 1-3 and health variable in Columns 4-6. All of these channel controls have expected signs and are highly significant. Compared with our baseline 2SLS results (0.16), adding in education controls reduces coefficients on Christian activities by about 40 percent; adding in health controls reduces coefficients on Christian activities by about 27 percent. In the last column of Table 6, we add in the years of schooling and mortality rate simultaneously. The coefficient on Christian activities is reduced to 0.09, and still remains significant at 10% statistical level. A magnitude of 0.09 still has significant economic meaning. The reduction in statistical significance is mainly due to the relatively larger standard errors, which remains stable around 0.05 across all specifications. We hereby conclude that, there are many other profound social and ideological impacts that Christian activities may have on Chinese communities, such as openness to new ideas, acceptance to modern technologies, or entrepreneurship. These are too complicated to be completely captured by the observable variables such as education and public health. These other potential channels deserve attention for future researches. Becker and Woosman (2009) argue that Protestantism affects long-run economic performance mainly through improvements in literacy rate. Therefore, if one controls for measures of current education in the GDP regression, the effect of the religious activities vanishes. Here, we implement a similar exercise in our context. In particular, we want to see if the influence of historical Christian activities on current economic performance works through channels other than improvements in educational achievements or public health. Ideally, we want to estimate the following model:

ln(pcGDP )     ln(Christian )   Edu   Health  X   2000 2 2 1920 2 2000 2 2000 2 2 (1) where we control for both the education and public health status of county i in year 2000 along with the Christian activities in 1920 on the right-hand side. The OLS results for this model are reported in Table 7, where we experiment different combinations of alternative measures in education and public health. As expected, both of these controls have large and significant effects on the county level economic performance. For example, column 2 of Table 7 reveals that 1 percent increase in years of schooling is associated with 0.78 percent increase in GDP per capita. This result is comparable to existing studies using across country variations.24 As to the effects of Christian activity, comparing with OLS results in the last column of Table 2, its coefficient decreases by 20-40% but remaining significant at 1% level. The potential problem of such a model is that not only the Christian activity in 1920 but also the education and health controls are endogenous. Unobservable county-level characteristics affecting education and health status can also affect economic outcomes. Even though we use two variables of droughts and floods frequencies to provide exogenous variations for historical Christian activities, no independent instrument is available in our context for education and health status. We pursue a similar strategy proposed in Becker and Woosman (2009) to bound

24 A similar study in the literature using similar specification to us is Mankiw et al (1992). According to their across-country analysis, 1 percent increase in years of schooling is associated with 0.66 percent increase in GDP per capita. the effect of Christian activities on GDP per capita net of education and health effects. Specifically, consider the following model:

ln(pcGDP2000 )  Edu2000 Health2000  2  2 ln(Christian1920)  X 2  2 (2)

where  is calculated based on the estimates of  2 from equation (1) while adjusting for potential biases found in other, well identified studies;  is calculated from  2 in a similar manner. Based on an extensive review on the return to education literature, Card (1999) concludes that the upward bias due to ignorance of ability account for about 10 percent of the OLS estimates. Moreover, studies using institutional changes to create exogenous variation in schooling actually receive estimates 20-40 percent higher than those of OLS. The main reason is that institutional improvements usually affect people with low education outcomes, who tend to have higher marginal return to schooling. Similar downward biases in OLS may apply to our context as well. Relative to the return to education literature, studies trying to evaluate effects health improvement on economic performance based on regional variations are much less conclusive (Jack and Lewis, 2009). They always suffer from omitted variable bias and reverse causality problems (Mankiw, 1995, p303-304). There are very few studies trying to construct instrumental mental variables for health measures in order to solve the endogeneity problem. Gallup and Sachs (2001), and Bloom et al (2004) are two examples. Even for these studies, their identification strategies are always questioned and debated, and their 2SLS results are not significantly different from the OLS. Guided by the above literature, we put an upper bound and lower bound on  and  in order to bound the estimate in equation (2). In practice, we experiment value of  ranging from 0.6 to 1.4 of  2 estimated from equation (1). We conservatively choose a much wider range for  , starting from 0.8 to 3.6 of  2 , given that the biased direction and magnitude are not clear based on the return to health literature.

Table 8 reports estimates of  2 where Christian activity in 1920 is instrumented by historical extreme weather frequency. Each cell is a separate regression. We choose years of schooling as the measure of education developments and children mortality rate as the measure of health conditions. The results using other measures deliver similar conclusions. The titles of each column and row indicate how we adjust OLS estimates of return to schooling and health in equation (2) when we estimate  2 .In particular, in cell [1,1], when both  and  equal to zero, we obtain the point estimates of the total Christian effect in Table 3. [INSERT TABLE 8 HERE] Column (1) reports results when we just take out the contribution of education in the GDP per capita regression while ignore the effects of health improvements. Coefficients on Christian activities range from 0.09 to 0.12, and remain both statistically and economically significant, suggesting that educational improvement accounts for about 25 to 43 percent of the total effect of Christianity. Row (1) reports results when only the contribution of health improvement is taken out from GDP per capita regression while the effect of education is ignored. For the wide range of  considered by us, the coefficient on Christianity remains at vicinity of 0.13, suggesting that long-run health improvements represent about 19 percent of the total effect of Christianity.25 For most of the reasonable combinations, effect of Christianity remains significant, ranging from 0.07 to 0.1. At extreme, in row (7) when the true effect of return to education is 40 percent higher than that of the OLS estimates, coefficient on Christianity starts to lose significance when the true effect of return to health is more than 120 percent higher than that of the OLS estimates (Row 7 and column 4 to 7).

Even for these cases, magnitude of the β2 still has significant economic meaning. We hereby conclude that, even though improvement in education and health status contribute to about 50% of the long-run effects of Christianity on economic performance, there are many other potential channels not captured by our studies. These other channels include openness to new ideas, acceptance to modern technologies, or entrepreneurship. These are all profound social and ideological impacts that on Chinese communities, but are too complicated to be measured or captured by other observable attributes.

9. Conclusion

In this paper, we study whether the spread of Christianity in the early twentieth century China, a large but closed and backward society, generates any long-run social economic influence. We construct a dataset mapping the historical Christian activities and current economic indicators at the county level. Utilizing this within-country variation, we find that Christian activities have significant positive effects in promoting long-run regional economic prosperity. One major identification challenge is that, if counties with better geographical, social, or economic conditions were more attractive to missionaries in the past, and these counties continue to have better economic performance in the present, we might obtain a positive correlation between past Christian activities and present economic outcomes, even though the former do not have any causal effects on the latter. We pursue several strategies to establish causality. First, through historical facts, anecdotal evidence, and statistical illustrations, we argue that Christian missionaries were actually oriented towards the less developed area to preach the word. Therefore, if there is any selection effect, it should make our OLS regression under-estimate rather than over-estimate the true effects of Christian activities on economic development. Second, taking advantage of the fact that Christian churches actively involved in “disaster relief” as a way to gain mutual trust with the local communities, we use historical frequencies of extreme weather as instruments for Christian

25 In fact, it is possible to calculate the cutoff value of χ and δ beyond which β starts to become insignificant. The χ value for β to lose significance at 5 percent level is 1.6 times χ2, and the corresponding δ value is 4 times δ2. Such large degrees of under estimation for OLS are very implausible. activities. Our 2SLS estimates confirm and further enhance the OLS results on the positive effects of Christian activities. Third, our findings are robust controlling for detailed geographical and contemporary climate characteristics. They are also robust among hinterland and rural counties. We then investigate the alternative channels through which Christian activities might generate such positive effects. As an important part of their charity work, missionaries help local people to build up schools and hospitals, and introduce modern science and technologies into the vast hinterland of China. For Chinese people, such interactions with the Western culture encourage them to challenge traditional norms, and to open their mind to new ideas. Empirically, we find the improvement in human capital accumulation, including education and health status, account for a major part of the positive effects of the Christian activities on economic performance. However, there are still some other profound and unobservable channels that these two indicators cannot capture, which we will leave for future researches. References

Acemoglu, Daron, Simon Johnson, and James Robinson, Yunyong Thaicharoen, “Institutional Causes, Macroeconomic Symptoms: Volatility, Crises and Growth”, NBER Working Paper No. 9124, Issued in August 2002. Acemoglu, Daron, Simon Johnson, and James Robinson, “Institutions as the Fundamental Cause of Long-Run Growth”, NBER Working Paper No. 10481, Issued in May 2004. Arruñada, Benito, “Protestants and Catholics: Similar Work Ethic, Different Social Ethic”, The Economic Journal, Volume 120, Issue 547, September 2010, pp. 890–918. Bai, Ying, and James Kung, "Diffusing Knowledge While Spreading God's Message: Protestantism and Economic Prosperity in China, 1840-1920." Manuscript,April 2011. Balme, H. & Stauffer, M.T. (1920). An Enquiry into the Scientific Efficiency of Mission Hospitals in China (pp. 430) Barnett, S. W., and J. K. Fairbank, eds, 1985, Christianity in China: early Protestant missionary writings, Harvard studies in American-East Asian relations, vol 9. Barro, Robert J, and Rachel M. McCleary, “Religion and Economic Growth across Countries”, American Sociological Review, Volume 68, No 5, October 2003, pp. 760-781. Becker, Sascha O, and Ludger Woessmann, “Was Weber Wrong? A Human Capital Theory of Protestant Economic History”, The Quarterly Journal of Economics, Volume 124, Issue 2, 2009, pp.531-596. Becker, Sascha O, Peter Egger, and Maximilian von Ehrlich, "Absorptive Capacity and the Growth Effects of Regional Transfers: A Regression Discontinuity Design with Heterogeneous Treatment Effects," CEPR Discussion Papers 8474, C.E.P.R. Discussion Papers, 2011. Becker, Sascha O, and Ludger Woessmann, "Luther and the Girls: Religious Denomination and the Female Education Gap in 19th Century Prussia," Stirling Economics Discussion Papers 2008-20, University of Stirling, Division of Economics. Becker, Sascha O, and Ludger Woessmann, "Knocking on Heaven’s Door? Protestantism and Suicide," The Warwick Economics Research Paper Series (TWERPS) 966, University of Warwick, Department of Economics, 2011. Becker, Sascha O, and Ludger Woessmann, “The effect of Protestantism on education before the industrialization: Evidence from 1816 Prussia,” Economics Letters, Volume 107, Issue 2, 2010, pp. 224-228. Cantoni, Davide, “The economic effects of the Protestant Reformation: testing the Weber hypothesis in the German Lands”, Working Papers, 2010. Cheng, Leonard, and Yum Kwan, “What are the determinants of the location of foreign direct investment? The Chinese experience”, Journal of International Economics, 51, 2000, 379-400 China Continuation Committee, Special Committee on Survey and Occupation, Milton Theobald Stauffer, “The Christian occupation of China: A general survey of the numerical strength and geographical distribution [sic] of the Christian forces in China, made by the Special Committee on Survey and Occupation, China Continuation Committee, 1918-1921”, Chinese Materials Center, 1922. China Continuity Committee, 1914, The Christian Yearbook of China,p.70(in Chinese) China Educational Commission, 1922, Christian education in China; a study made by an educational commission representing the mission boards and societies conducting work in China. New York City, Committee of Reference and Counsel of the Foreign Missions Conference of North America Cohen, Paul A., “Christian missions and their impact to 1900”, in the Cambridge History of China, Vol. 10, Late Ch’ing, 1800-1911, Part 1,editedbyDenis Twitchett, and John K. Fairbank,. Cambridge University Press, 1978 Delacroix, Jacques, and Francois Nielsen, “The Beloved Myth: Protestantism and the Rise of Industrial Capitalism in Nineteenth-Century Europe,” Social Forces, Volume 80, 2001, pp.509–553. Educational Association of China, 1896, “Records of the Second Triennial Meeting of the Education Association of China,” Held at Shanghai, May 6-9, 1896, p. 41 Ekelund, Robert B., Jr. Robert F.Hébert, and Robert D. Tollison, “An Economic Analysis of the Protestant Reformation”, Journal of Political Economy, Volume 110, No. 3, June 2002, pp. 646-671. Elman, Benjamin A, 2006, A Cultural History of Modern Science in China. Cambridge, MA: Harvard University Press. Fairbank, J. K., 1974, The Missionary Enterprise in China and America, Harvard Studies in American-East Asian Relations, Cambridge, MA: Harvard University Press. P13. Gallego, Francesco, and Robert D Woodberry, “Christian Missionaries and Education in Former African Colonies: How Competition Mattered” Journal of African Economies, Volume 19, Issue 3, March 2010, pp. 294-329. Gu, Weimin, 2010. Christianity and Modern Chinese Society, Shanghai: Shanghai People’s Press. Guiso, Luigi, Paola Sapienza, and Luigi Zingales, “Does Culture Affect Economic Outcomes?” Journal of Economic Perspectives, Volume 20, Issue 2, 2006, pp. 23-48. Huff, Toby. 2003. The Rise of Early Modern Science: Islam, China, and the West. Cambridge. UK: Cambridge University Press. Lee, Timothy, (1889),“New rules for making Shanxi rich”. Globe Magazine. In Chinese. Lin, Justin Yifu. 1995. “The Needham Puzzle: Why the Industrial Revolution Did Not Originate in China,” Economic Development and Cultural Change, 43(2): 269-292. MacAlister, Donald, 1921, Preface of “China and Modern Medicine: A Study in Medical Missionary Development”, by Haroll Balme, London, United Council for Missionary Education. McCleary, Rachel M. and Robert J. Barro, “Religion and Economy,” Journal of Economic Perspective, Volume 20, Issue 2, 2006, pp.49-72. McCleary, Rachel M. and Robert J Barro, “Religion and Political Economy in an International Panel,” Journal for the Scientific Study of Religion, Volume 45, Issue 2, 2006, pp. 149-75. Nunn, Nathan, "The Long Term Effects of Africa's Slave Trades," Quarterly Journal of Economics, Volume 123, Issue 1, February 2008, pp. 139-176. Nunn, Nathan, "Religious Conversion in Colonial Africa," American Economic Review Papers and Proceedings, Volume 100, No. 2, May 2010, pp. 147-152. Records of the General Conference of the Protestant Missionaries of China, held at Shanghai, May 7-20, 1890. (pp. 733). General Books LLC. Shepard D.,1968, A two-dimensional interpolation function for irregularly-spaced data, Proceedings of the 1968 23rd ACM national conference, ACM, pp.517-524 Silby, Rev. J. A., 1902, “Appeal to Foreign Mission Boards for Trained Educators for China,” Chinese Recorder, Vol 33： 619-621. Sinnott, R.W. 1984. Virtues of the Haversine.Sky and Telescope, 68 (2). 158. Stauffer, Milton T. 1922. The Christian occupation of China: a general survey of the numerical strength and geographical distribution of the Christian forces in China. Shanghai: China Continuation Committee. Stock, James H. and Motohiro Yogo. 2005. "Testing for Weak Instruments in IV Regression," in Identification and Inference for Econometric Models: A Festschrift in Honor of Thomas Rothenberg. Donald W. K. Andrews and James H. Stock, eds. Cambridge University Press, pp.80–108. Wang, Shenyin. (1983). A brief introduction of the Cheeloo University’s history. In Committee of Cultural and Historical Data of the CPPCC (Eds.), Selected Historical Data (Volume 91, pp. 112-113). Beijing: Chinese Culture and History Press. In Chinese. Wang, Yousan., 1991, History of Religion in China, Page 985-986, Jinan,Shandong: Qilu Press. Woodberry, Robert D. “Missions and Economics,” Encyclopedia of Missions and Missionaries. Jonathan J. Bonk (ed.). New York: Routledge, 2007. Woodberry, Robert D. “The Social Impact of Missionary Higher Education.” pp. 99-120 in Christian Responses to Asian Challenges: A Globalization View on Christian Higher Education in East Asia. Philip Yuen Sang Leung and Peter Tze Ming Ng (eds.). Hong Kong: Centre for the Study of Religion and Chinese Society, Chung Chi College, Chinese University of Hong Kong, 2007. Woodberry, Robert D. “Religious Roots of Economic and Political Institutions,” Forthcoming in Oxford Handbook of the Economics of Religion. Rachel McCleary (ed.). Oxford: Oxford University Press, 2010. Wylie, Alexander and William Gamble (1867). Memorials of Protestant Missionaries to the Chinese. Shanghai: American Presbyterian Mission Press. Yang, Senfu. 1968. China History of Christianity (Zhongguo Jidujiao Shi). Taibei: Taiwan Shang Wu Press. Figure 1.Converts per 1,000 people in 1920

Convert per 1k people (by quantile) .00 - .01 .50 - .80 .01 - .10 .80 - 1.00 .10 - .20 1.00 - 1.50 .20 - .30 1.50 - 2.50 .30 - .50 2.50 - 25.00

Figure 2. GDP per capita in 2000

GDP per capita in 2000 (by quantile, in 1k yuan)

.00 - 1.80 4.20 - 5.00 1.80 - 2.50 5.00 - 5.80 2.50 - 3.00 5.80 - 7.20 3.00 - 3.60 7.20 -10.00 3.60 - 4.20 10.00 - 35.00 Figure 3 Correlation between convert per 1,000 people in 1920 and GDP per capita in 2000 10 9 8 7 GDP capita per (log), 2000 6 5

-4 -2 0 2 4 6 Converts per capita (log), 1920

Figure 4. Correlation between initial economic conditions and Christian activities

(church density on the left; vicar density on the right) in 1920

(a) (b) 8 8 6 6 4 4 2 2 Population Density (log), 1920 Population Density (log), 1920 0 0 -2 -2 -8 -6 -4 -2 0 -6 -4 -2 0 Churches per 1000 people (log), 1920 Vicars per 1000 people (log), 1920 coef = -.21, se = .029 coef = -.117, se = .029

(c) (d) -2 -2 -4 -4 -6 -6 Tax per capita (log), 1820 Tax per capita (log), 1820 -8 -8 -10 -10 -8 -6 -4 -2 0 -6 -4 -2 0 Churches per 1000 people (log), 1920 Vicars per 1000 people (log), 1920 coef = -.211, se = .029 coef = -.118, se = .029 Figure 5. Correlation between Western Medical Development and Christian activities in 1920

4 Shangdong

FujianGuangdong 3.5

ShangdongJiangsu Zhili

Hubei 3 3 Hunan Zhejiang Jiangxi Hunan Henan 2.5 Anhui JiangsuGuangdong 2 Jiangxi Henan Yunnan ZhejiangFujian Hubei 2 Guangxi Guizhou Zhili ln(# of Hospitals), 1920 of Hospitals), ln(# 1 ln(# of Phamacies), 1920 Phamacies), of ln(#

YunnanGuizhou 1.5 Anhui

Guangxi 0 1

8.5 9 9.5 10 10.5 11 8.5 9 9.5 10 10.5 11 ln(# of Converts), 1920 ln(# of Converts), 1920 beta coef = 1.01, t-stat = 5.05, N = 14, R2 = 0.68. beta coef = .39, t-stat = 2.05, N = 14, R2 = 0.27. Table 1. Summary Statistics NMeanStd.MinMax

Panel A. County Data I. Historical Religious Activities Converts / 1,000 person 1743 0.76 1.62 0.00 27.71 Church / 1,000 person 1743 0.02 0.04 0.00 1.45 Vicar / 1,000 person 1743 0.02 0.04 0.00 0.46 Conditional on being positive Converts / 1,000 person 1473 0.90 1.72 0.00 27.71 Church / 1,000 person 1225 0.02 0.05 0.00 1.45 Vicar / 1,000 person 1346 0.03 0.04 0.00 0.46

II. Economic Performance GDP per capita (yuan) 1743 5419.20 3972.15 109.02 31064.16

III. Education Edu Exp. per capita (yuan) 1743 94.23 45.55 7.26 541.38 Year of Schooling 1743 7.22 1.00 1.85 11.06 Literacy (%) 1743 89.21 8.02 24.35 99.45 Mortality Rate (%) 1743 6.34 1.23 1.48 12.21 Infant Mortality Rate (%) 1743 2.32 1.89 0.00 18.65

IV. Foreign Firms Asset Share (%) 1743 11.87 17.67 0.00 95.66 Revenue Share (%) 1743 11.89 17.87 0.00 96.83 Employee Share (%) 1743 11.34 16.33 0.00 92.93

V. Control Variables Coast Province 1743 0.40 0.49 0.00 1.00 Distance to Capital City (km) 1743 187.41 108.07 0.58 611.09 Temperature (°C) 1743 13.52 3.68 0.16 23.72 Rain (cm) 1743 9494.68 3537.73 89.70 22525.05 Hist. Flood Freq (1900-1920) 1743 0.38 0.10 0.00 1.00 Hist. Draught Freq (1900-1920) 1743 0.28 0.10 0.00 0.80 Hist. Flood Freq (1978-2000) 1743 0.29 0.06 0.01 0.56 Hist. Draught Freq (1978-2000) 1743 0.46 0.08 0.18 0.82

Panel B Provincial Data (1960-2008) Converts / 1,000 person 1108 0.86 0.61 0.10 2.57 GDP growth rate 1108 8.70 9.31 -57.10 45.20 GDP per capita growth rate 1108 7.20 9.22 -57.20 45.30 Table 2. Relationship between Historical Church Activities and Current Economic Performance Dependent Variable is Log GDP per Capita in 2000 (1) (2) (3) (4) (5) log (Convert/Pop1920) 0.12 0.08 0.05 0.05 0.05 [0.01]*** [0.01]*** [0.01]*** [0.01]*** [0.01]*** Longitude 0.04 0.04 0.04 [0.01]*** [0.01]*** [0.01]*** Latitude -0.00 -0.01 -0.01 [0.01] [0.01] [0.01]** Dist to Prov. Capital -0.13 -0.13 -0.13 (in logs) [0.02]*** [0.03]*** [0.02]*** Altitude -0.10 -0.10 -0.10 (in logs) [0.02]*** [0.02]*** [0.02]*** Temperature -0.29 -0.38 (in logs) [0.11]** [0.13]*** Rain 0.09 0.21 (in logs) [0.12] [0.14] Current Flood Freq. -1.47 (1978-2000) [0.46]*** Current Draught Freq. -0.82 (1978-2000) [0.34]** Reg. FE NO YES YES YES YES R-sq 0.11 0.23 0.30 0.31 0.31 N 1743 1743 1743 1743 1743 Notes: The Christian Activities variable log (Convert/Pop1920) is the logarithm of disciples of each county in 1920 normalized by population. Coefficients are reported with standard errors in brackets. ***, **, and * indicate significance at 1%, 5% and 10% levels. Table 3. Relationship between Historical Church Activities and Current Economic Performance (1) (2) (3) (4) (5) Panel A. Second Stage. Dependent Variable is Log GDP per Capita in 2000 log (Convert/Pop1920) 0.16 0.17 0.17 0.17 0.15 [0.03]*** [0.05]*** [0.04]*** [0.04]*** [0.04]*** Location Controls NO NO YES YES YES Climate Controls NO NO NO YES YES Current Natural Disaster NO NO NO NO YES Reg. FE NO YES YES YES YES N 1743 1743 1743 1743 1743 Panel B. First Stage. Dependent Variable: log (Convert/Pop1920) 1900-1920 Flood Freq. 1.58 1.88 2.08 2.31 2.24 [0.51]*** [0.50]*** [0.49]*** [0.49]*** [0.52]*** 1900-1920 Draught Freq. 6.21 4.08 4.90 4.78 5.08 [0.56]*** [0.58]*** [0.55]*** [0.56]*** [0.59]*** Location Controls NO NO YES YES YES Climate Controls NO NO NO YES YES Current Natural Disaster NO NO NO NO YES Reg. FE NO YES YES YES YES F-stat on IV 73.87 24.75 40.82 37.84 39.23 Over Identification 0.26 0.32 0.09 0.26 0.46 (p-value) Notes: The Christian Activities variable log (Convert/Pop1920) is the logarithm of disciples of each county in 1920 normalized by population. Coefficients are reported with standard errors in brackets. ***, **, and * indicate significance at 1%, 5% and 10% levels. The Hansen's J statistic is used for the test of over identifying restrictions in the presence of heteroskedasticity. The joint null hypothesis is that the instruments are valid instruments, i.e., uncorrelated with the error term, and that the excluded instruments are correctly excluded from the estimated equation. Table 4. Robustness Check

Omitting No. of Hinterland Rural Catholic Influential Land Tax Vicars Churches Sample Sample Mission Observation Station (1) (2) (3) (4) (5) (6) (7) Panel A. Second Stage, Dependent Variable is Log GDP per Capita in 2000 log (Convert/Pop1920) 0.16 0.34 0.15 0.16 0.12 [0.05]*** [0.10]*** [0.04]*** [0.04]*** [0.05]*** log (Vicar/Pop1920) 0.43 [0.16]*** log (Church/Pop1920) 0.49 [0.16]*** Land Tax 0.13 [0.09] No. of Catholic Stations 0.02 [0.00]*** Control Variables YES YES YES YES YES YES YES Number of obs. 1660 1048 1613 1743 1743 1743 1743 Panel B. First Stage. 1900-1920 Flood Freq. 2.40 1.68 2.32 2.21 1.86 0.24 0.69 [0.45]*** [0.64]*** [0.56]*** [0.52]*** [0.53]*** [0.32] [0.30]** 1900-1920 Draught 4.41 3.81 5.28 5.05 4.72 1.46 1.59 Freq. [0.51]*** [0.89]*** [0.64]*** [0.59]*** [0.60]*** [0.33]*** [0.33]*** F-stat on IV 37.22 9.46 34.99 38.80 33.95 12.09 12.98 Over Identification 0.88 0.05 0.97 0.45 0.88 0.52 0.08 (p-value)

Panel C. OLS Estimates. Dependent Variable is Log GDP per Capita in 2000 log (Convert/Pop1920) 0.05 0.03 0.05 0.05 0.04 [0.01]*** [0.01]*** [0.01]*** [0.01]*** [0.01]*** log (Vicar/Pop1920) 0.06 [0.01]*** log (Church/Pop1920) 0.05 [0.01]*** Notes: 1. The Christian Activities variable log (Convert/Pop1920) is the logarithm of converts of each county in 1920 normalized by population. Coefficients are reported with standard errors in brackets. ***, **, and * indicate significance at 1%, 5% and 10% levels. The Hansen's J statistic is used for the test of over identifying restrictions in the presence of heteroskedasticity. The joint null hypothesis is that the instruments are valid instruments, i.e., uncorrelated with the error term, and that the excluded instruments are correctly excluded from the estimated equation. 2. Control Variables include location controls like longitude, latitude, (log) distance to the capital city and (log) altitude, climate variables like temperature and precipitation, current natural disaster rate and regional fixed effect. Table 5 Relationship between Historical Church Activities and Education

Church Primary Public Primary Primary Schools in Total (IV) Enrollment Rate Schools Schools Convert/Pop Convert/Pop (in logs) Overall Sample (IV Tobit) (IV Tobit) 1920>0 1920=0

(1) (2) (3) (4) (5) Panel A. Second Stage Estimates log (Convert/Pop1920) 1.29 0.37 0.48 0.53 [0.15]*** [0.07]*** [0.24]** [0.22]** log (Avg -0.10 -0.15 0.46 Convert/Pop1920 in the neighborhood) [0.17] [0.15] [0.25]* Controls YES YES YES YES YES N 1743 1743 1743 1452 249 F-stat on IV set 1 36.31 36.31 19.62 21.34 3.80 F-stat on IV set 2 39.76 30.93 Over Identification 0.00 0.31 0.32 0.93 0.83 (p-value) Panel B. Estimates without Instruments log (Convert/Pop1920) 1.22 0.12 0.244 0.247 [0.03]*** [0.01]*** [0.01]*** [0.02]*** log (Weighted Avg. 0.106 0.066 0.159 Convert/Pop1920) [0.02]*** [0.02]** [0.04]** Notes: 1. The Christian Activities variable log (Convert/Pop1920) is the logarithm of converts of each county in 1920 normalized by population. In order to study the spillover effects of Christian Activities on education level, we measure the neighboring Christian activity intensity by the weighted average of Convert/Population in 1920 in its contingent neighbor counties (take logs). We use the weighted average of current Flood/Draught frequency in its neighbor counties as IVs for the neighboring Christian activity intensity. 2. Coefficients are reported with standard errors in brackets. ***, **, and * indicate significance at 1%, 5% and 10% levels. The Hansen's J statistic is used for the test of over identifying restrictions in the presence of heteroskedasticity. The joint null hypothesis is that the instruments are valid instruments, i.e., uncorrelated with the error term, and that the excluded instruments are correctly excluded from the estimated equation. 3. Control Variables include location controls like longitude, latitude, (log) distance to the capital city and (log) altitude, climate variables like temperature and precipitation, current natural disaster rate and regional fixed effect. Table 6. Effects of Church Activities on Other Social Economic Outcomes

Human Captial

(1) (2) (3) (4) (5) Panel A. Second Stage Dependent Variable (in Year Childhood Health Edu Exp. Literacy logs) Schooling Mortality Exp. ln(Convert/Pop1920) 0.05 0.06 0.04 -0.33 0.04 [0.03]* [0.01]*** [0.01]*** [0.05]*** [0.05] Location Controls Yes Yes Yes Yes Yes Climate Controls Yes Yes Yes Yes Yes Current Natural Disaster Yes Yes Yes Yes Yes Reg. FE Yes Yes Yes Yes Yes Number of obs. 1743 1743 1743 1743 1743 Panel B. First Stage. Dependent Variable: ln(Converts/Pop1920) 1900-1920 Flood Freq. 2.24 2.24 2.24 2.24 2.24 [0.52]*** [0.52]*** [0.52]*** [0.52]*** [0.52]*** 1900-1920 Draught Freq. 5.08 5.08 5.08 5.08 5.08 [0.59]*** [0.59]*** [0.59]*** [0.59]*** [0.59]*** F-stat on IV 38.79 38.79 38.79 38.79 38.79 Over Identification 0.00 0.04 0.01 0.00 0.00 (p-value) Panel C. OLS Estimates. Dependent Variable is Log GDP per Capita in 2000 ln(Convert/Pop1920) 0.02 0.01 0.00 -0.04 0.03 [0.01]*** [0.00]*** [0.00]** [0.01]*** [0.01]*** Notes: The Chrisitian Activities variable ln(Convert/Pop1920) is the logarithm of converts of each county in 1920 normalized by population. Coefficients are reported with standard errors in brackets. ***, **, and * indicate significance at 1%, 5% and 10% levels. The Hansen's J statistic is used for the test of overidentifying restrictions in the presence of heteroskedasticity. The joint null hypothesis is that the instruments are valid instruments, i.e., uncorrelated with the error term, and that the excluded instruments are correctly excluded from the estimated equation. Table 7. Alternative Channels

(1) (2) (3) (4) (5) (6) Dep. Var.: Log GDP per Capita in 2000 ln(Convert/Pop1920) 0.03 0.04 0.04 0.03 0.03 0.03 [0.01]*** [0.01]*** [0.01]*** [0.01]*** [0.01]*** [0.01]*** Education Expenditure 0.64 0.56 [0.03]*** [0.04]*** Year Schooling 0.79 0.69 [0.11]*** [0.13]*** Literacy 0.86 1.05 [0.13]*** [0.13]*** Childhood Mortality Rate -0.07 -0.06 -0.07 [0.02]*** [0.02]** [0.02]*** Health Expenditure 0.09 0.29 0.3 [0.02]*** [0.02]*** [0.02]*** Location Controls Yes Yes Yes Yes Yes Yes Climate Controls Yes Yes Yes Yes Yes Yes Current Natural Disaster Yes Yes Yes Yes Yes Yes Reg. FE Yes Yes Yes Yes Yes Yes N 1743 1743 1743 1743 1743 1743 Notes:

The Chrisitian Activities variable ln(Convert/Pop1920) is the logarithm of converts of each county in 1920 normalized by population. Coefficients are reported with standard errors in brackets. ***, **, and * indicate significance at 1%, 5% and 10% levels. The Hansen's J statistic is used for the test of overidentifying restrictions in the presence of heteroskedasticity. The joint null hypothesis is that the instruments are valid instruments, i.e., uncorrelated with the error term, and that the excluded instruments are correctly excluded from the estimated equation. Table 8.

ln(ChildrenDeathRate) (1) (2) (3) (4) (5) (6) (7) 0 80% δ 100% δ 120% δ 150% δ 200% δ 250% δ (1) 0 0.16 0.14 0.14 0.14 0.13 0.12 0.11 [0.04]*** [0.04]*** [0.04]*** [0.04]*** [0.04]*** [0.04]*** [0.04]**

(2) 80% χ2 0.12 0.1 0.1 0.1 0.09 0.08 0.07 [0.04]*** [0.04]** [0.04]** [0.04]** [0.04]** [0.04]* [0.04]* (3) 90% χ2 0.12 0.1 0.09 0.09 0.08 0.07 0.06 [0.04]*** [0.04]** [0.04]** [0.04]** [0.04]** [0.04]* [0.04] (4) 100% χ2 0.11 0.09 0.09 0.09 0.08 0.07 0.06 [0.04]*** [0.04]** [0.04]** [0.04]** [0.04]* [0.04]* [0.04] (5) 110% χ2 0.1 0.09 0.08 0.08 0.07 0.06 0.05 ln(Years of schooling) [0.04]** [0.04]** [0.04]** [0.04]* [0.04]* [0.04] [0.04] (6) 120% χ2 0.1 0.08 0.08 0.08 0.07 0.06 0.05 [0.04]** [0.04]** [0.04]* [0.04]* [0.04]* [0.04] [0.04] (7) 140% χ2 0.09 0.07 0.07 0.07 0.06 0.05 0.04 [0.04]** [0.04]* [0.04]* [0.04] [0.04] [0.04] [0.04] Table A1

Dep. Var. Tax per capita 1820 GDP per capita 2000 (1) (2) (3) (4) Flood Intensity -3.25 -3.28 1800-1820 [0.38]*** [0.39]*** Drought Intensity -2.48 -2.91 1800-1820 [0.40]*** [0.40]*** Flood Intensity 0.22 -0.39 1978-2000 [0.44] [0.44] Drought Intensity 0.27 -0.22 1978-2000 [0.37] [0.36] Geographic Controls No Yes No Yes Regional FE Yes Yes Yes Yes R-sq 0.4 0.41 0.23 0.27 N 1498 1498 1453 1453 Data Appendix

1. Mapping historical information with current administrative units

Refer to Figure A1 as an illustrative example, where squares represent 1999’s county boundaries and circles represent 1920’s county boundaries. For the latter we 1920 have measures of Christian activities (Xi , i=1,…,5), including number of converts, vicars, and churches. The corresponding 2000 county level measures (X1999), shown as the grey area, is calculated by

1999 xxkk  kAE ~ where

1999 1920 sA xxA 1 s1 ... s xx1999 1920 E E 5 s 5 Figure A1. An illustrative map The implicit assumption is that the church activities in 1920 are equally distributed with a county.

2. Convert station level information into county level

Given the climate measures at the station level (Yj, j=1,…,J), the county level measures (Yc) are constructed by

YYwcjij YYwcjcj   jJ1, jJ1,..., where wij is the weight, which is constructed using Shephard’s method (Shepard, 1968):

2 distcj wcj  2  distck kJ1,..,