Buddha s Grace Illuminates All: Temple Destruction, School

Transcription

1 Buddha s Grace Illuminates All: Temple Destruction, School Construction, and Modernization in the 20 th Century China Shaoda Wang Boxiao Zhang October 2015 Preliminary Version, Please Do Not Cite or Circulate Abstract This study documents a novel natural experiment in the early 20 th century China called the Temple Destruction Movement (TDM), where local governments were required to take over the assets of Buddha and Taoist temples to support local modern schooling. We show that before the TDM, the prior stock of average temple assets was not correlated with various measures of human capital and economic development; while after the TDM started, regions with higher prior stock of average temple assets constructed more schools, enrolled more students, evolved more modern elites, and also experienced faster urbanization. Moreover, we find that the impacts of the TDM persist in the 21 st century: regions with higher prior stock of temple assets before the TDM have higher human capital and better economic performance in Keywords: Temple Destruction Movement; Human Capital; Economic Development JEL: O15; O18; N35 Department of Agricultural and Resource Economics, U.C. Berkeley. Guanghua School of Management, Peking University. We are indebted to Noam Yuchtman and Se Yan for their generous support. We would also like to thank Alain de Janvry, Elisabeth Sadoulet, Debin Ma, Yu Hao, Paul Katz, Yang Xie, Xi Lu, Chicheng Ma, Ye Wang, Xuenong Zhou, and seminar participants at the third International Symposium on Quantitative History for their helpful comments. Financial support from the Berkeley Economic History Lab (BEHL) is gratefully acknowledged. Any remaining errors are solely ours. 1

2 1 Introduction For over 1,300 years, especially in the late imperial period ( ), China s civil society in general, and its traditional education system in particular, were fundamentally shaped by the civil service exams (Elman, 2000). At various levels, tutors, schools, and academies trained students with Confucian classics, which were the official materials of the exams. To prepare for these exams, students had to devote themselves mainly into writing poems and eight-legged essays, which partly led to the academic tradition of valuing humanities over science and technology (Lin, 1995; Huff, 2003). In the late 19 th and early 20 th centuries, China experienced consecutive military defeats against the western powers and Japan, which led to the abolishment of the civil service exams, and the explorations of ways to generate modern human capital and new elites. Among the numerous attempts made by the government, not surprisingly, constructing modern schools was the most important and persistent one. Since the central government faced severe debts and deficits, the provision of education funding was almost entirely undertaken by lower-level governments and local elites. To help the local elites raise adequate amount of funding for modern schooling, the central government implemented a novel policy aimed to close most of China s total 200,000 Buddha\Taoist temples and take over most of their assets to finance modern education, which was known as the Temple Destruction Movement (TDM). The TDM was first proposed and approved in 1898, but due to political and ideological reasons never actually implemented until In this paper, making use of a unique prefecture-level panel dataset on modern education in the late Imperial and early Republican eras of China, we first show that before the movement started, stock of average temple assets (measured by the average number of temples in 1820) was orthogonal to various kinds of indicators for human capital and economic prosperity, and was not correlated with either the levels or the trends of average number of schools and students. However, after the movement started, places with higher stock of temple assets ex ante constructed more schools, enrolled more students, and later 2

3 evolved more elite human capital (measured by average number of top university alumni and oversea students), and also experienced faster economic growth (measured by urbanization rate). Furthermore, we provide evidence that the increase of human capital and its positive effects on economic growth persists until the 21 st century: places with higher prior stock of average temple assets in 1820 have higher levels of human capital and economic performance in Our study speaks to several strands of literature. First, we add to the increasing literature on the formation and persistence of human capital through time. Economic historians have long argued that formation of modern human capital is essential for a society s transition towards a modern economy (Yuchtman, 2014), but there is few empirical evidence documenting how exactly does modern human capital evolve, our paper fills in this blank. Also, there have been papers showing that mass human capital investment in history could have persistent effects on today s educational levels and investments (Huillery, 2009; Wantchekon et al., 2015; Nunn, 2011). Different from these works, we also study how elite human capital could evolve from investment in mass human capital, and also estimate the effects of human capital on economic development in the mid-run and in the long-run. Second, our paper contributes to the literature on the relationship between human capital and economic development. There exists a huge literature on whether human capital could boost economic development, some using cross-country comparison (Hall and Jones, 1999; Glaeser et al., 2004; Acemoglu et al., 2014), others using cross-regional analysis (Acemoglu and Dell, 2009; Gennaioli et al., 2013). But as pointed out by many researchers, investment in education is an endogenous decision affected by many factors, such as institutions (North and Thomas, 1973, Acemoglu et al., 2002, Acemoglu et al., 2014, Engerman et al., 2012, Galor et al., 2009), political structure (Lindert, 2004), fiscal capacity (Bardhan and Mookherjee, 2006), ethnicity and culture (Easterly and Levine, 2012, Fernandez, 2011, Peri, 2012), which will also affect economic outcomes. Therefore, estimating the causal effects of human capital on development remains challenging. Some of the more recent works adopt historical natural experiments to identify this causal relationship (Ferraz et al., 2015; Hornung, 2014; Becker et al., 2011; Squicciarini and Voigtlaender, 2015), basically showing that human capital has positive impacts on 3

4 development, and such effects could persist in the long run. Our paper reassures these findings in the context of 20 th century China using a comprehensive historical dataset. We also differentiate from these recent papers by studying and documenting the formation of elite human capital in addition to only mass human capital, which is believed to be crucial to the transition and upgrade of an economy (Cantoni and Yuchtman, 2014). Third, our paper is also related to the large literature on the effects of religion on economic growth. Since Max Weber (2002), whether Protestantism could actually encourage hard work and in turn contribute to economic prosperity has led to heated debates and mixed evidence in the empirical literature. Cantoni (2013) finds no relationship between Protestantism and growth, many papers find evidence supporting Weber s proposed channel (Stulz and Williamson, 2003; Barro and McCleary, 2003), but others argue that it is not religion itself, but the human capital and other by-products of it, that contribute to growth (Glaeser and Sacerdote, 2008; Becker and Woessmann, 2009). In the context of China, Bai and Kung (2015) find that Protestantism promoted growth in China through building hospitals and schools, Chen et al (2015) find similar results, but also suggest that this human capital channel is not exclusive. To our best knowledge, despite the large literature on Protestantism, there has been few studies on the social and economic effects of Buddhism and Taoism, which are highly important religions in the context of China, and many other Asian countries as well. Our results suggest that these two religions contributed to modernization significantly, but likely only through the channel of their assets being appropriated to support modern schooling, and not through shaping social preferences or any other channels. The remainder of this paper is organized as follows. Section 2 will document the historical background of the formation of modern human capital in China and the Temple Destruction Movement. Section 3 discusses the data and presents descriptive statistics. Section 4 will present the effects of the TDM on both mass and elite human capital formation, and the effects on mid-run economic development. Section 5 presents evidence of the persistence of human capital and its effects on economic growth in the 21 st century. Section 6 discusses the robustness of our results. Section 7 concludes. 4

5 2 Historical Background Fundamentally shaped by the civil service exam, China s traditional education system were constituted by tutors, traditional schools and academies. At the junior level, there were several kinds of public or private traditional schools, including community schools (shexue), charitable schools (yixue) and clan schools (zuxue). According to imperial edict, those schools objective was jiaohua, can be translated as civilization or enculturation. They trained students to learn basic reading and writing skills, and to recite Confucian classics, including the Four Books and Five Classics (Sishu Wujing), which were the standard textbooks of the civil service exam. For exam participants, they usually needed to hire tutors or attend academies to obtain senior education afterwards, including writing poems and eight-legged essays (Baguwen). In the late 19th and early 20th centuries, after consecutive military defeats against the western powers and Japan, Chinese government and elites realized the challenges they faced in technology, as well as in culture and education. In 1902, China issued the first comprehensive plan of building its own modern education system, and began to construct modern primary schools, secondary schools and universities. Instead of Confucian classics, students were required to learn modern knowledge about science, technology and humanities in those schools. During , Japan succeeded in the Russo-Japanese War, which was the first major military victory in the modern era of an Asian power over a European nation. Japan's success was attributed to the Meiji Restoration and Modernization, which set an example for the Qing Dynasty and facilitated a new round of education reform, including the abolishment of the civil exam system as well as the traditional education system, and the further construction of modern schools (Franke, 1960). According the experiences of western countries, public provision of modern education requires consistent financial support. However, in late Qing Dynasty, both central and local governments faced severe debts and deficits, and were unable to build and sustain a modern education system. In 1895, after the Sino-Japanese war, the deficit 5

6 of Qing government reached 15 million tael of silver. 1 In 1910, the number increased to 80 million (Zhang, 1996). To raise adequate funding to support modern schooling, a novel policy was prompted, which aims to activate traditional temple assets to finance local education, and is often referred to as the Temple Destruction Movement (TDM). Early in 1898, one of the most important Chinese politicians, Zhang Zhidong, wrote a famous proposal (Quanxue Pian) to the Emperor Guangxu and the Empress Dowager Cixi, suggesting that the government should take over the assets of those Buddha temples, and use the money to build primary schools in the local communities, so that China will cultivate more literate youngsters and generate modern human capital to compete against the western countries. As part of the tremendous 1898 reform plan (Wuxu Bianfa), the government accepted this advice, and planned to change 70% of the country s temples into primary schools, and take over 70% of these temples assets to use as education budget. However, due to political and ideological reasons, the 1898 reform plan only lasted for 104 days, before most of the innovative policies were terminated by the Empress Dowager Cixi, and 7 of the major designers of the reform were sentenced to death, even the emperor himself was imprisoned in the Forbidden City (Xu 2008). As a result, the TDM was also banned by the government. However, under the pressure of raising enough fund for local education, some local governments implemented this policy secretly at smaller scales. Realizing such cases, in 1905, the central government re-emphasized that the TDM was illegal. In 1912, the Qing government was overthrown, and the Republican of China was founded. In the same year, President Yuan Shikai issued the 31 rules of temple administration, which formally urged the local governments to take over all (rather than 70%) of the temple assets to construct modern schools. Soon after that, the TDM thrived across the whole country, and persisted with local authorities even though the centralized state power disappeared again after Yuan Shikai died in Nearly twenty years later, in the 1930s, the Kuomintang reestablished the centralized state power and formally 1 Tael is a Chinese traditional unit of weight, 1 tael 37.8 g. 6

7 facilitated the movement once again. Therefore, the TDM persisted for more than 40 years, until the establishment of the PR China (Katz 2012, 2013). Buddha and Taoist temples had accumulated large amount of assets throughout the history, including buildings, land, and cash. While there is no official survey data, some studies estimate temple assets using historical documents. Xu (2010) summarizes that, in late Qing Dynasty, there were about 2 million Buddha and Tau temples, and they totally owned about 16 million houses, 13,000 square kilometers land and millions tael of silver. As a matter of fact, the imperial China actually had a long-standing historical tradition that governments solve their fiscal crisis by seizing temple assets. Following that tradition, it was natural that the Qing and Republican governments regarded temple assets as viable sources of funding for modern education. During the TDM, local governments and elites took over tremendous amount of wealth from temples, and used it to support the construction of a modern education system. With adequate funding collected from temple assets, modern education thrived quickly across the country. As shown in figures 1 and 2, during the first half of the 20th century, the development of modern education has been boosted twice 2. In 1910s and 1930s, both the amount of schools and students had increased greatly, which are consistent with the periods that the TDM was conducted and facilitated. [Insert Figure 1 and 2 about here] Historical studies also find that the Temple Destruction Movement had huge destructive effects on Buddhism and Taoism in China. Before the Temple Destruction Movement, in Chinese history, there were four similar movements of seizing temple assets, summarized by the Buddhists as four greatest disasters (Sanwu Yizong Fanan). However, none of them can be compared to the TDM. The most common form of destruction was to use temples houses as schools. From existing historical evidences, more than 70% of schools were transformed from temples. In addition, some local governments also regulated temples land as part of schools assets, or imposed high religion tax on temples (Xu, 2010). 2 The data we use in the figures will be introduced in section 3. 7

8 3 Data and Key Variables The data for our analysis is collected from assorted sources, and constructed as several datasets. First, we have a prefectural-level panel dataset, with information on modern education, temple and other control variables in the early 20 th century ( ), which will be used to investigate the effects of the TDM on school construction. Second, we obtain personal information of top university alumni and oversea students before the establishment of the PR China, and use the average number of students as the measure of elite human capital from the 1890s to the 1940s. Third, we use a county-level crosssectional dataset, with information on urbanization rate and other control variables in 1920 and 1964, to identify the effects of school construction on economic development in the mid-run. Finally, we collect county-level average income and other indicators of human capital in 2000, and try to investigate the TDM s positive effects on long-term economic growth. 3.1 Prefectural-level Panel Data of Modern Education and Temple Assets Modern Education Data The original education data basically includes three historical periods, from the late Qing Dynasty to the Republic of China. It records detailed information on the number of schools, number of students, types of schools and education budget at the county-level. In the first period, this database includes three years education data in the late Qing Dynasty, which is the year of 1907, 1908 and In the second period, this database includes two years education data in the beginning of the Republic of China, which is the year of 1915 and The above five years survey data was respectively complied as the first, second, third, fourth and fifth time of the Statistical Chart of Education (Jiaoyu Tongji Tubiao). In 1905, during the political reform in the late Qing Dynasty, the Ministry of Education was established for modern education reform. Between the year of 1907 and 1909, the Minister of Education was Zhang Zhidong, who was well-known for his progress of sponsoring modern education in Hubei. Since the modern education reform had been 8

9 conducted in China for nearly ten years at that time, Zhang Zhidong carried out those surveys to investigate and evaluate the reform in nationwide. However, after the third survey in 1909, the nationwide survey of modern education was interrupted by the 1911 Revolution. Till the year of 1915 and 1916, in the early Republic of China, the Ministry of Education successfully conducted the fourth and fifth surveys again under the comparatively stabilized atmosphere. Afterwards, the national survey of modern education was once again interrupted by conflicts and wars. 3 The third period is much longer than the first two periods, which is between the year of 1919 and From 1920s to 1930s, the national surveys of modern education were missing because of the unstable political atmosphere. Therefore, the county-level education data comes from several provincial surveys conducted by some local education authorities. This dataset includes five provinces county-level education data in 1920s, and nine provinces in 1930s. The above surveys provide by far the best and the most complete national county-level education data, which has been cited as credible historical data in the history literature (Shang, 2001; Su, 2007). The specific sources of education data can be specifically traced and is summarized in table A1 in Appendix. Based on the original county-level education data with multiple years, we then do two important adjustments to reorganize it into a prefectural-level panel data with five years. In the first period, we observed that the data of 1907, 1908 and 1909 is basically complete with little missing values. Therefore, we keep all the 1907, 1908 and 1909 s data in the first period. However, in the second period, we observed that the data of 1915 has a complete coverage, while the data of 1916 has many missing values. We then only use the 1915 s data to represent the second period and drop the data of As for the third period, we set 1930 as the representative year, and for each available province, we choose data whose survey year is closest to In summary, we now have the adjusted data with five representative years, for each period. A tricky problem is how to append them into a panel data. From the late Qing Dynasty to the Republic of China, the changes of county-level administrative divisions were frequent and complicated. Furthermore, we are lack of any 3 The detailed process of the five times nationwide education survey has been fully studied by several historians, see Guan (1999), Ma and Lu (2002), and Wang (2010). 9

10 GIS map to handle the changes. Luckily, most of the changes are within prefectures, and we can avoid the tricky problem by summarizing the data from county-level to prefecturelevel. Based on 1820 s map from CHGIS, 4 we are finally able to construct a prefecturallevel education panel data with years of 1907, 1908, 1909, 1915 and The modern education data has 264 prefectures in 18 provinces in 1907, 1908, 1909 and 1915, and 135 prefectures in 10 provinces in We use the number of primary schools and students as the main indicator of modern education development. Temple and Traditional School Data The data for temple assets is documented in the Imperial Encyclopedia of the Qing Empire (Daqing Yitong Zhi), which includes prefecture-level information on the important local Buddha and Taoist temples. This book was compiled by Qing government for more than 150 years, and was final-edited in For each prefecture, we obtain the total number of important temples as the indicator of temple assets before the TDM. Our data includes 264 prefectures in 18 provinces, which can be perfectly matched to 1820 s map from CHGIS. Furthermore, we also collect the number of traditional schools in 1820, and include it in our regression as a control variable. Population, Area and Jinshi Data We also include other important socioeconomic information in our study. We have collected prefectural-level population data for six years, including 1776, 1820, 1850, 1880, 1910 and We have also collected area of each prefecture. In Qing dynasty, there are more than 26,000 successful candidates of the civil exam (Jinshi). From the Official Records of Jinshi in Qing Dynasty (Qingdai Jinshi Timing Lu), we identify each exam successful candidate s prefecture of origin and construct a prefectural-level data of the total number of civil exam graduates. 4 The China Historical Geographic Information System provide a prefectural-level GIS map in

11 We match the aforementioned datasets to the panel dataset of education. Except for the number of primary schools and students, all the other variables are time-invariant. 3.2 Prefecture-level Panel Data of Top University Alumni and Oversea Students From assorted historical archives and alumni books, we obtain personal information of top university alumni and oversea students. We identify 16,439 students who graduated before 1950, from the top 2 universities in China, Peking University and Tsinghua University, both located in Beijing. We divide the era into eight 5-year periods, and for each prefecture, we summarize the number of students enrolled in the top 2 universities in each period. For oversea students, we identify 9,001 individuals who were born between 1890s and 1940s. Similarly, we divide the birth era into 10-year periods, and for each prefecture, we count the number of future oversea students born in each period. By above method, we construct two prefecture-level panel data of the number of top university alumni and oversea students. Furthermore, we match the datasets to the panel datasets of education, population and area, especially the variable of average temples. 3.3 County-level Data of Urban Population and Urbanization Rate The data of urban population and urbanization rate in 1920 are obtained from The Christian Occupation of China (Stauffer, 1922), which contains population of all counties and urban population of cities with more than 25,000 residents. Therefore, we compute the proportion of population living in urban areas as a measure of urbanization in We also collect population data in 1964 from the The Complied Statistics of the Second National Population Census (National Bureau of Statistics, 1982). The data includes total population, male population and the number of households. In the Census data, we also know in each county how many people are registered as urban residents, and how many as rural residents. The residents were largely classified by their types of occupation: urban residents typically participate in non-agricultural production, and vice versa. Since in the 1960s migrations across regions were banned in most cases, the ratio of urban residents in a region is an accurate measure of its level of urbanization. Therefore, we can obtain urbanization rate and average family size in

12 We merge the above two datasets together, and construct a county-level dataset of urbanization rate. Furthermore, we identify the prefecture each county belongs to, and combine the data with prefectural-level variables. In the end, we have information for 1,700 counties from 18 provinces in our data. 3.4 County-level Contemporary Data of Economic Growth and Human Capital To study the long-term impacts of the TDM, we assemble contemporary socioeconomic indicators of Chinese counties. We obtained county-level data on economic growth and human capital in 2000, including aggregate GDP, government subsidies on education and culture, from the Statistical Materials of Public Finance of Cities and Counties; and data on population, literacy rates and years of schooling from the Fifth National Population Census (2000). One of the challenges faced by our study is that because of the long historical span, political regimes changed and territorial borders of many counties also shifted. We address this challenge by first comparing the GIS data for both 1820 and 2000 and then converting prefecture-level variables in the early 20 th century to those in 2000 using overlapping area as weights. To achieve those ends, we obtain historical GIS data from the China Historical Geographic Information System (CHGIS). Then we construct a data set that combines historical education and temple data, and contemporary socioeconomic data. 3.5 Key Variables All variables used in this study are summarized in table 1, containing their definition, sources and statistics. In the following, we discuss several key variables. Average Temple Assets Ideally, to construct a proxy for temple assets, we would like to have detailed information on the number of temples in each prefecture, along with measurements of scale and wealth for each temple. However, as discussed in section 3.1, the Imperial Encyclopedia of the Qing Empire only lists those important temples, and doesn t include the small ones. 12

13 Therefore, we could only proxy for the stock of temple assets using the number of important temples in each prefecture. For this to be a good proxy, we need to assume that the (unobservable) overall temple asset is positively correlated with the number of important temples, which seems to be reasonable. Also, since whether a temple is listed as important or not is a subjective judgement of the authors of the Imperial Encyclopedia of the Qing Empire, the standards may be inconsistent across regions, and lead to measurement errors. We need to assume that the measurement error is not correlated with local education conditions, which means that the authors do not change their criterion for temples according to the local education levels, which is also likely to be satisfied. We use 1820 population as denominator to calculate the average number of important temples, and obtain the proxy variable of average temple assets. Average Schools\Students We use the education data and population data described in section 3.1 to calculate the average schools and average students. 5 For modern schools\students in 1907, 1908, 1909 and 1915, we use the population of 1910 as denominator. For schools\students in 1930, since we only have population data for 1910 and 1953, we use 1910 population as denominator, but also try to use 1953 and the simple average of 1910 and 1953, for all the three cases, the results are largely the same. Average Top University Alumni and Oversea Students With the student data in section 3.2, combined with population data, we calculate the average number of enrolled top university students and born oversea students in every period. Similarly, for periods after 1920s, we use 1910 population as denominator, but also try to use 1953 and the simple average of 1910 and 1953, for all the three cases, the results are largely the same. 5 Average Schools= Schools/Population, Average Students= Students/Population. 13

14 Urbanization Rate and Schools Differences In the county-level data, we have urbanization rate in 1920 and We define the Urban as the difference between two rates. 6 We also have average schools in 1909 and 1915, and can define School and ln(school) in the same way. 7 Contemporary Economic Development and Human Capital In our contemporary county-level data, we have average income (GDP per capita) and urbanization rate as the indicator of economic development level in To investigate the mechanisms, we use average years of schooling, literacy rate and culture and education expenditure per capita to measure the level of human capital. [Insert Table 1 about here] 4 Empirical Results: Short- and Mid-Terms 4.1 TDM and Mass Education As argued above, prefectures with more temple assets ex ante, thanks to the TDM, had easier access to education funding during the Republican era, and were thus more likely to invest in modern education, which would lead to better economic development later on. To formalize this logic, we first adopt a DID model, and test whether the places with more temple assets ex ante indeed constructed more schools during the movement. We define the prefectures with temples more than the median value as treated, the others as control. The information of schools is available in years 1907, 1908, 1909, 1915, and As discussed above, the Temple Destruction Movement really started to peak after 1912, so we can regard the changes between 1907 and 1909 as pre-trends, and the changes between 1909 and 1930 as post-trends. [Insert Figure 3 about here] Figure 3 shows the trends in the average number of schools for treated and control groups. In 1907, both groups had essentially the same amount of schools; before the TDM 6 Urban= urbanization rate in urbanization rate in School= average primary schools in average primary schools in ln(school)= ln(average primary schools in 1915) ln(average primary schools in 1909). 14

15 was introduced, the school construction for both groups followed very similar trends, and their difference in levels were also almost negligible. Differences between the prior trends is statistically insignificant, so the identical trends assumption holds, which makes our Difference-in-Differences approach likely to be a valid one. Moreover, before the treatment, difference in levels in the two groups is also statistically insignificant, which provides further credibility to our identification strategy. During the post-treatment period ( ), the number of schools in the treated group (prefectures with temples above the median value) increased much faster than that in the control group. Moreover, the differences between trends enlarged between 1915 and 1930, a naïve but intuitive interpretation for this fact is that the control group started to face increasing marginal costs of taking over temple assets, while the treated group, having more temples, did not face such problem. In figure 4, we show the trends in the number of students for treated and control groups in years 1907, 1908, 1909, 1915 and We see that both the pre-trends and the post-trends of average primary students are very similar to that of the schools. [Insert Figure 4 about here] We quantify the observations from the graphical analysis using a DID method: School iτ = τ { } α τ Temple i Year τ + X iτ β + ρ τ + μ i + ε iτ (1) where School iτ is defined as the number of primary schools per 10,000 people in prefecture i, Temple i is defined as the number of temples per 10,000 people in prefecture i in 1820, Year τ is the time dummy. X iτ is a set of control variables that vary both across units and time, ρ τ is a time effect common to all prefectures in period τ, μ i is a time-invariant effect unique to prefecture i, and ε iτ is a prefecture time-varying error distributed independently of μ i and ρ τ. The year of 1907 is left as a comparison. [Insert Table 2 about here] As shown in table 2, the interactions of temples and 1908 and 1909 are not statistically significant, which is consistent with the parallel trends assumption. The interactions temple*1915 and temple*1930 are both statistically significant at 1% level, which indicates that the prefectures with more temples have a faster speed of school construction 15

16 after the movement started. We try to control for different variables, including the timevariant impacts of ln(population), ln(area), Average Traditional Schools and Average Jinshi Number, in column 2 and 3, and the prior results are highly robust. Since nearly half of prefectures have missing values in 1930 s data, in column 4, we only include the data of 1907, 1908, 1909 and 1915, and the prior results remain the same. In column 5, we exclude prefecture fixed-effects, and include the interaction term temple*1907, to test whether regions with different number of temples have different number of schools in As we can see, the coefficient is statistically insignificant, indicating that temples are unlikely to be correlated with number of schools in method: We also estimate the effect of TDM on primary school students using the same DID Student iτ = τ { } α τ Temple i Year τ + X iτ β + ρ τ + μ i + ε iτ (2) where the dependent variable has been changed as Student iτ, which is defined as the number of primary schools per 10,000 people in prefecture i. As shown in table 3, patterns the results are very similar to table 2. Before the TDM, parallel trends are satisfied, but after the TDM, places with higher stock of average temples enrolled significantly more students. Results are highly robust to the inclusion of various control variables, column 5 indicates that the initial level of average students is also uncorrelated with average stock of temples. [Insert Table 3 about here] 4.2 TDM and Elite Education As shown in section 4.1, the TDM led to construction of more schools and enrollment of more students, indicating that the TDM has very likely contributed to massive education. However, did the TDM affected only mass human capital, or did it also contribute to the formation of elite human capital? In this section, we measure elite human capital using average number of alumni from top universities and average number of oversea students, and provide evidence that before the TDM, the formation of elite human capital was uncorrelated with the stock 16

17 of average temples; but after the TDM, regions with more temples cultivated significantly more modern elites. We provide suggestive evidence that higher education expenditure per student due to the TDM might contributed to elite human capital. Alumni of Top Universities For top 2 university alumni, we aggregate the individual level data to the prefecture level and match it with the stock of average temples and other prefecture level variables. We also aggregate the yearly data to 8 five-year periods, so that the change of alumni for each prefecture becomes smooth. Our results are not sensitive to different methods of aggregation. As shown in section 4, the TDM started to peak after Considering the regular time of primary and middle schools education, we would expect that the stock of average temples had and only had effects on top university enrollment after the 1920s. Therefore, we define a dummy variable Post τ, which equals one if period τ is after 1920, and zero otherwise. We then estimate the following equation: Alumni iτ = α Temple i Post τ + X iτ β + ρ τ + μ i + ε iτ (3) where Temple i is the stock of average temples in prefecture i in 1820, X iτ is a vector of prefecture level control variables that varies with time. ρ τ is time fixed effect, μ i is prefecture fixed effect, ε iτ is the error term that varies both with time and prefecture. As shown in the first three columns of table 4, a 10% increase in the stock of average temples would result in about 2% increase in average top university alumni after the movement. [Insert Table 4 about here] In order to present the dynamic effects and test for pre-trends, we also explore a slightly different specification where the stock of average temples is interacted with each period, and the coefficients and confidence intervals are plotted in figure 5. As we can see, there is no differences in trends for the pre-periods, and significantly different trends in the post-periods, suggesting that our diff-in-diff approach is likely to be valid. [Insert Figure 5 about here] 17

18 Oversea Students We do almost exactly the same exercise with oversea students as outcome variable. We aggregate the individual data to prefecture level, and aggregate the yearly data into 6 tenyear periods. The record of oversea students document their birth year rather than year of going abroad, so we define periods according to year of birth. Since the TDM started in 1912, we would expect that only individuals born after 1900 could benefit from the stock of average temples. Therefore, in this specification, Post τ equals one for periods after 1900, and zero otherwise. We estimate the following equation: Oversea iτ = α Temple i Post τ + X iτ β + ρ τ + μ i + ε iτ (4) where Oversea iτ is defined as the average number of oversea students in prefecture i in period τ, other variables are defined same as before. As shown in table 4, previous stock of temple assets would strongly affect the ratio of oversea students after the TDM started. A 10% increase in average temple assets would lead to 1.2% increase of average oversea students. Same as for top university students, we run another specification that interact average temples with each period, and plot the coefficients and confidence intervals. As shown in figure 6, temples were not correlated with change in oversea students before the movement, while places with more temples had significantly more oversea students after the movement started. [Insert Figure 6 about here] 4.3 Mid-Term Impacts on Economic Development As discussed in sections 4.1 and 4.2, because of the TDM, regions with more temples accumulated higher human capital, at both mass and elite levels. A natural question would then be, whether the advantage in human capital later transformed in to a margin in economic development? Before the Temple Destruction Movement As shown before, human capital was uncorrelated with temples in both levels and trends. Ideally, we would like to conduct same tests for our economic outcomes. However, there 18

19 are very few economic indicators that have multiple periods of data available before the TDM started. So instead, we do the following two tests. First, we collect cross-sectional data on average tax revenue before the TDM, and run an OLS regression of tax on temples. Since average tax revenue is a good indicator for local economic prosperity, and our hypothesis is that temples are orthogonal to economic development before the TDM, we should expect to see that the stock of temples have no effects on average tax revenue. As we can see from columns 1-3 in table 5, this is exactly the case. Second, while we do not have multiple periods for most economic indicators, we do have a cross-sectional dataset on the distribution of modern factories across the country in We first argue that since the TDM only started in 1912, and education should need longer time to influence industrialization, the establishment of factories before 1916 should not be affected by the TDM. Furthermore, since there were very few modern factories before the self-strengthen movement (1861), we could actually interpret the level of factories in 1916 as the trends of factory establishment between 1861 and This way, by running a regression of the 1916 factories on 1820 temples, we are actually also testing whether the stock of temple assets affected the trend of industrialization in the late-qing era (pre-tdm). As shown in columns 4-6 in table 5, temples are uncorrelated with factories, indicating that the parallel trends assumption is also likely to hold for economic development. [Insert Table 5 about here] After the Temple Destruction Movement Since information for China s regional economic outcomes before the 1980s is largely missing, for economic outcomes, we follow the literature and specify the degree of urbanization as a proxy for economic development, given that this indicator is believed to be highly and positively correlated with per capita income (Acemoglu et al 2005; Bai and Kung, 2014). We estimate the following equation: 19

20 urban i = α Temple i + X i β + μ prov + ε i (5) where urban i is the speed of urbanization during , Temple i stands for the average number of temples in 1820 in the prefecture that county i belongs to, X i is a vector of county-level control variables, μ prov is province fixed effect, ε i is the error term. As shown in columns 1 and 2 of table 6, counties with higher stock of average temples in 1820 experienced faster urbanization during While exclusion restriction could not be directly tested, we provide suggestive evidence by exploring the channels through which temples might affect growth: if temples affect economic outcomes only through human capital, once including measures of human capital as control variables in the regressions, temples should no longer have significant impacts on economic outcomes. As shown in columns 3 and 4 of table 6, once control for mass education, the magnitude and significance of the coefficient for temples drop sharply; when we further control for elite education in in columns 5 and 6, the magnitude further drops and the significance totally disappears. These results suggest that human capital accumulation is likely to be the source of economic development we observed in the data. [Insert Table 6 about here] 4.4 Mechanisms Mechanisms for Elite Human Capital Formation Since temples and elite human capital are uncorrelated for periods before the TDM, it is likely that the parallel prior trends assumption is satisfied: places with more temples and places with fewer temples did not have significant differences in human capital accumulation. To further test this hypothesis, we regress the average number of Jinshi on the average number of temples. As shown in table 7, there is no statistically significant relationship between Jinshi and temples, this reassures that the TDM is an appropriate natural experiment to study the effects of human capital accumulation. [Insert Table 7 about here] Why might the TDM improve elite human capital formation? One straight-forward channel is that since more schools were constructed, more students were enrolled in 20

21 modern schools, and thus more elites were selected out of a larger pool. In addition to this explanation, since temple assets were used to not only construct schools, but also finance students, it could be possible that the TDM also cultivated elites by improving the quality of education. We test this hypothesis by regressing the average education fund per student on the average amount of temples using a fixed effect model. As shown in table 8, places with more temples had higher education fund per student, which suggests that the quality of education was also better. Considering the large heterogeneity of education quality in the Republican era, it is reasonable that places with more sufficient funding could provide much better education, and their students will be more likely to stand out and become elites. [Insert Table 8 about here] Mechanisms for Economic Development As shown in section 4.3, by exploiting the TDM as a natural experiment, we find significant positive effects of education investment on economic development. However, remaining concerns include the following two: 1. It might be the case that the TDM added to the budget of local governments where there were more temples, and not 100% of the added budget was on school construction, or by adding new education budgets, they could reallocate the original government education budget for other purposes. If that is the case, then the observed effects on urbanization could also be explained by resource reallocation and general government spending, rather than just increased investment in education. 2. During the TDM, thousands of temples were shut down, as a result, many monks decided to return to laity (Katz, 2013). It might be the case that these monks returned to laity added to the local labor supply, and directly contributed to local development. In this section, we discuss these two concerns respectively. Qualitative evidence and placebo tests will be provided to address the concerns. Resource Reallocation Both government documents and anecdotal evidence suggest that during the late Imperial 21

22 and early Republican eras of China, massive primary education was supported exclusively by the local gentries, and the governments at different levels were really hands-off on such issues (Shang, 2001; Su, 2007). Given that the local governments did not have any budget prepared for massive education before the TDM, we may not need to worry too much about the possibility that by using temple assets to build schools, they can then reallocate the original education budget in other ways, and boost the economy. However, even though the TDM requires that the temple assets should be used only for massive education, we may still be concerned that some governments would simply use the temple assets in other ways, for instance, invest in basic infrastructures, and thus contribute to the local economy. If that is the case, temples may affect the economy through channels other than education, then we could no longer interpret findings of section 4.3 as a causal relationship between human capital and economic development. However, as we discussed in section 4.3, once human capital is controlled for, temples no longer affect growth. This indicates that it is unlikely that temples boosted the economy in ways other than education, which is inconsistent with this hypothesis. In addition, qualitative evidence suggests that local governments hardly had any budget prepared for modern education other than raising from local elites and temples, and education budget was almost the only legitimate reason to ask for their donations. As a result, using these funds to serve other purposes is difficult for the local governments, and could easily cause severe social unrest and even local conflicts (Xu, 2010; Katz, 2013). Therefore, both the data and the historical evidence seem to be inconsistent with this alternative hypothesis. Increased Labor Supply As shown in section 4.3, temples have no effects on growth after controlling for human capital, so it is unlikely that the temples, in addition to boosting investment in education, had also contributed to the economy by adding the monks returned to laity to local labor supply. To further address this concern, we carry out a falsification test. We have shown that the treatment group (prefectures with temples above the median value) and the control group (prefectures with temples below the median value) had largely comparable 22

23 socioeconomic conditions before the TDM, and we have also shown that the TDM was happening very rapidly between 1909 and So if it was the case that many monks returned to laity joined the labor force due to the TDM, given that the monks did not have their own land and were thus unlikely to participate in agricultural production, it is likely that we would observe the treatment group had a higher worker/population ratio in the 1930s. From the Chinese Industry Report, we collect total numbers of industry workers in 135 counties in However, as shown in table 9, the average temple assets has little impact on the ratio of workers in 1934, meaning that the monks returning to laity concern may not be a severe one. [Insert Table 9 about here] 5 Empirical Results: Long-Term In section 4, we have shown the effects of the TDM on mass and elite human capital, and also the effects of such human capital growth on economic development. In this section, we would like to further explore the persistence of these effects. As explained in section 3, we matched a county level socio-economic dataset with our historical prefecture level panel dataset. Using this merged dataset, we could estimate the following equation: Y i = α Temple i + X i β + μ prov + ε i (6) where Y i is the outcome of interest for county i in year 2000, Temple i stands for the average number of temples in 1820 in the prefecture that county i belongs to, X i is a vector of county-level control variables, μ prov is province fixed effect, ε i is the error term. Persistence of Human Capital We first test whether the positive effects of the stock of average temples persisted into the 21 st century. As shown in table 10, counties with higher stock of temples in 1820 have significantly higher average years of schooling, literacy rate, and per capita culture and education expenditure in On average, a one standard deviation increase of average temples in 23

24 that prefecture would lead to the average years of schooling in that county to increase by 0.1 years, the illiteracy rate to drop by 6%, and the per capita culture and education expenditure to increase by about 4%. These results are highly robust to various types of control variables. Specifically, if one concerns that the TDM has direct impact on religious practices in different regions, which would in turn affect the outcomes, we show that including other measurements of religious practices (average vicars, average churches) does not change the result at all, which could arguably reduce the concern. [Insert Table 10 about here] Persistent Impacts on Economic Development We now examine whether the persistence of human capital would contribute to economic development in the long run. As shown in table 11, counties with higher stock of temples in 1820 have significantly higher per capita GDP and urbanization rate in On average, a one standard deviation increase of average temples in that prefecture would lead to per capita GDP to increase by about 300 Yuan (about 10% of the mean), and urbanization rate to increase by about 6%. To justify the causal effects of human capital on these economic outcomes, we explore the potential channels through which temples could affect the economy: if temples affect economic outcomes through ways other than human capital, by including measures of human capital as control variables in the regressions, temples should still have significant impacts on economic outcomes. As we can see from columns 3 and 6 of table 11, after including three measures of human capital (average years of schooling, literacy rate, average culture and education expenditure) as control variables, the positive effects of average temples totally disappear. Therefore, it is unlikely that temples affect the economy through channels other than human capital, and thus the results of table 11 could likely be interpreted as causal evidence that human capital accumulation contributed to economic growth in the long run. [Insert Table 11 about here] 24