Essays on Economic Development and Social Capital by Hyunjoo Yang B.A., Incheon National University, 2008 M.Sc., London School of Economics and Political Science, 2010 M.A., Brown University, 2011 Dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Department of Economics at Brown University Providence, Rhode Island May 2016 c Copyright 2016 by Hyunjoo Yang This dissertation by Hyunjoo Yang is accepted in its present form by the Department of Economics as satisfying the dissertation requirement for the degree of Doctor of Philosophy. Date David N. Weil, Advisor Recommended to the Graduate Council Date Stylianos Michalopoulos, Reader Date Nathaniel Baum-Snow, Reader Approved by the Graduate Council Date Peter M. Weber, Dean of the Graduate School iii Vita Hyunjoo Yang was born in South Korea, on December 17, 1980. He received B.A. from Incheon National University in 2008, and M.Sc. from London School of Economics in 2010. He entered Brown University in 2010, and received M.A. in 2011 and Ph.D. in 2016. iv Acknowledgements I am deeply grateful to my advisors, David Weil, Nathaniel Baum-Snow, and Ste- lios Michalopoulos for their support. I also thank Kenneth Chay, Andrew Foster, Raphael Franck, Oded Galor, J. Vernon Henderson, Sriniketh Nagavarapu and Louis Putterman for valuable advice. I thank Kanghyock Koh for being the best colleague, coauthor, and friend I have ever had at Brown. I have benefited from extraordinary research assistance provided by Dahae Yang. I also appreciate Taeyoung Ryu for his guidance on data collection. I am grateful to professors at the Incheon National Uni- versity, especially Do-Suk Han, Taeho Kim, and Myungheon Lee, for encouraging me and helping me to pursue graduate study in economics. Finally, I thank my family for their love and support. Particularly, I am extremely fortunate to have my wife, Min Gyung Kim, who helped me and supported me in so many ways. This dissertation is dedicated to her. v Contents Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi Introduction 1 1 Family Clans and Public Goods: Evidence from the New Village Beautification Project in South Korea 3 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.2.1 Family Clans . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.2.2 New Village Beautification Project . . . . . . . . . . . . . . . 11 1.2.3 Family Clans and Social Capital . . . . . . . . . . . . . . . . . 14 1.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 1.3.1 Family Clans and Village Heterogeneity . . . . . . . . . . . . . 16 1.3.2 Economic Outcomes . . . . . . . . . . . . . . . . . . . . . . . 18 1.3.3 Geographic and Economic Characteristics of Villages . . . . . 22 1.4 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 1.5 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 1.5.1 Effect of Clan Heterogeneity on Cement Grades . . . . . . . . 26 1.5.2 Effect of Clan Heterogeneity on Land Donations . . . . . . . . 31 vi 1.5.3 Effect of Clan Heterogeneity on Agricultural Mechanization . 33 1.5.4 The Concave Relationship between Clan Heterogeneity and the Production of Public Goods . . . . . . . . . . . . . . . . . 35 1.6 Case Study: Effects of Lineage Homogeneity on Cooperation and Par- ticipation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 1.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 2 The Effect of War on Local Collective Action: Evidence from the Korean War 68 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 2.2 Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 2.2.1 The Korean War (1950-1953) . . . . . . . . . . . . . . . . . . 72 2.2.2 War Damages in South Jeolla Province . . . . . . . . . . . . 73 2.2.3 Anti-communist Purges . . . . . . . . . . . . . . . . . . . . . . 74 2.3 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 2.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 2.4.1 Population Censuses . . . . . . . . . . . . . . . . . . . . . . . 77 2.4.2 Severity of Conflict . . . . . . . . . . . . . . . . . . . . . . . . 77 2.4.3 The NVCS Data . . . . . . . . . . . . . . . . . . . . . . . . . 78 2.4.4 The Family Clan Data . . . . . . . . . . . . . . . . . . . . . . 79 2.5 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 2.5.1 Effects of War on Cooperation . . . . . . . . . . . . . . . . . . 81 2.5.2 Alternative Explanations . . . . . . . . . . . . . . . . . . . . 82 2.5.3 Effects of War on Population Size . . . . . . . . . . . . . . . . 84 2.5.4 Comparison of the Effects of Conflict Through Social Division and Conventional Battles . . . . . . . . . . . . . . . . . . . . . 85 2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 vii 3 The Effects of High Speed Trains on Local Economies: Evidence from the Korea Train Express 101 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 3.2 The Korea Train Express . . . . . . . . . . . . . . . . . . . . . . . . . 104 3.3 Data and Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . 105 3.3.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 3.3.2 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . 106 3.4 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 3.4.1 Impact of KTX on Local Economic Activity . . . . . . . . . . 108 3.4.2 Falsification Checks . . . . . . . . . . . . . . . . . . . . . . . . 110 3.4.3 Case Study of Daegu and Gimcheon City . . . . . . . . . . . . 111 3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 Bibliography 127 viii List of Tables 1.1 Summary statistics for village level analysis . . . . . . . . . . . . . . 40 1.2 Summary statistics for village level analysis (continued) . . . . . . . 41 1.3 Correlation coefficients between heterogeneity measures . . . . . . . . 41 1.4 Summary statistics for land donation analysis . . . . . . . . . . . . . 42 1.5 Summary statistics from power tiller analysis . . . . . . . . . . . . . 43 1.6 Effects of clan heterogeneity on public good production . . . . . . . . 44 1.7 Effects of TOPSHARE on public good production . . . . . . . . . . . 45 1.8 Effects of clan heterogeneity on land donation . . . . . . . . . . . . . 46 1.9 Effects of clan heterogeneity on power tiller ownership . . . . . . . . 47 1.10 Timeline of the New Village Beautification Project . . . . . . . . . . 58 1.11 Top priority village projects identified from a government survey . . 58 1.12 Effects of clan heterogeneity on public good production for non-split villages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 1.13 Effects of clan heterogeneity on cultivated land . . . . . . . . . . . . . 60 1.14 Effects on power tiller ownership with a cement grade control . . . . 61 1.15 Mean comparisons of village characteristics by clan concentration . . 62 1.16 Mean comparisons of village characteristics of split v.s. non-split vil- lages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 2.1 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 2.2 OLS and FE estimates of the effects of the conflict on cooperation . 89 ix 2.3 Falsification checks using ∆pop in pre- & post-war periods . . . . . . 90 2.4 Effect of war on population trends . . . . . . . . . . . . . . . . . . . 91 3.1 Diff-in-Diffs estimates of the impact of the KTX on light intensity . . 120 3.2 Estimates of the slope changes of distance to the KTX . . . . . . . . 121 3.3 Falsification checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 x List of Figures 1.1 Comparison of ethnic fractionalization measures across countries . . 48 1.2 Comparison of linguistic fractionalization measures across countries . 48 1.3 An example of family clan data from Family Names in Chosun . . . . 49 1.4 Histogram of TOPSHARE . . . . . . . . . . . . . . . . . . . . . . . 50 1.5 Histogram of HERFINDAHL . . . . . . . . . . . . . . . . . . . . . . 50 1.6 Histogram of POLARIZATION . . . . . . . . . . . . . . . . . . . . . 50 1.7 Spatial distribution of the Herfindahl Index for family clans . . . . . 51 1.8 Spatial distribution of the polarization index for family clans . . . . 52 1.9 An example of village data from New Village Comprehensive Survey . 53 1.10 Spatial distribution of cement project grades . . . . . . . . . . . . . 54 1.11 An example of a land donation list from Glorious Footsteps . . . . . 55 1.12 Standardized mean differences of village characteristics . . . . . . . . 56 1.13 Effect of clan heterogeneity on public good production . . . . . . . . 56 1.14 Effect of clan heterogeneity on land donation . . . . . . . . . . . . . . 57 1.15 Effect of clan heterogeneity on power tiller ownership . . . . . . . . . 57 1.16 A map of South Korea with provincial boundaries . . . . . . . . . . . 64 1.17 The average number of agricultural machines per agricultural house- hold . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 1.18 Land tilling and transportation . . . . . . . . . . . . . . . . . . . . . 65 1.19 Location of Moonsung village in the case study . . . . . . . . . . . . 66 xi 1.20 Location of villages that did not experience geographical split between 1930-1970 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 2.1 Location of South Jeolla . . . . . . . . . . . . . . . . . . . . . . . . . 92 2.2 The Korean War and UN Forces . . . . . . . . . . . . . . . . . . . . . 93 2.3 Major battle sites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 2.4 Severity of conflicts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 2.5 Effect of conflict on public use in South Jeolla . . . . . . . . . . . . . 96 2.6 Effect of conflict on public use in North Kyungsang . . . . . . . . . . 96 2.7 Effect of conflict on population trend . . . . . . . . . . . . . . . . . . 97 2.8 The Pusan Perimeter . . . . . . . . . . . . . . . . . . . . . . . . . . 98 2.9 Map of North Kyungsang province . . . . . . . . . . . . . . . . . . . 99 2.10 Effect of bombing on population trends . . . . . . . . . . . . . . . . . 100 3.1 The KTX network . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 3.2 Comparison of light intensities, 1994 and 2013 . . . . . . . . . . . . . 115 3.3 Pixel-level comparison of light intensity changes . . . . . . . . . . . . 116 3.4 Townships boundaries in 2016 . . . . . . . . . . . . . . . . . . . . . . 117 3.5 Differential light trends . . . . . . . . . . . . . . . . . . . . . . . . . . 118 3.6 Differential slopes of the KTX on local economies . . . . . . . . . . . 119 3.7 Light intensity, Seoul Metropolitan Area, 1994 . . . . . . . . . . . . . 123 3.8 Comparison of light intensities, 1994 and 2013 . . . . . . . . . . . . . 124 3.9 Light growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 3.10 Difference in the slopes of KTX distance . . . . . . . . . . . . . . . . 126 xii Introduction In the first two chapters of this dissertation, I examine the determinants of social capital, family clan heterogeneity and ideological conflicts. The third chapter assesses the effects of high speed rail on rural economic activity. In Chapter 1, I introduce heterogeneity in family clan membership as an important determinant of social capital. Ethnic and linguistic heterogeneity are widely studied as determinants of social capital, conflict, and institutional quality. In many cultures, another important dimension of heterogeneity is family clan membership. I study the relationship between family clan diversity in South Korean villages and the voluntary production of public goods and contributions of private resource for village projects. Under the 1970-1971 New Village Beautification Project, the government distributed resources to each village for the production of village public goods. Subsequently, the government systematically evaluated how well these resources were applied. I combine these data with information on village family clan structures collected by the Japanese Colonial Government, as well as records of land donations for village projects between 1970 and 1980. I find an inverted-U-shaped effect of group heterogeneity on the improvement of public goods and on the average amount of donated land per household. I suggest that the concave relationship reflects the trade-off between better coordination among clan members and less accountability of clan leaders as village clan homogeneity increases. 1 In Chapter 2, I investigate whether war has important long-term economic conse- quences. Existing literature suggests a lack of long-term effects related to the short- term destruction of physical capital and population reduction. Increased ideological and social division as a result of war, on the other hand, may produce persistent eco- nomic and social outcomes. I investigate the effect of the 1950-1953 Korean War on cooperation within rural communities in South Korea. Combining census data and unique data on village level collective action, I find that residents of townships that experienced more intense conflicts due to the prolonged presence of the North Korean Army and communist influences during the war were less likely to cooperate 20 years after the war ended. Further, I provide evidence that the reductions in township populations due to the conflict persisted over 40 years. The empirical results suggest that the impacts of the war persisted in the form of increased ideological and social division. Chapter 3 is a joint work with Kanghyock Koh. We empirically test the economic impacts of the introduction of the Korean high speed rail system, the Korea Train Express (KTX), in 2004. For analysis, we use nighttime light data from satellite observations as a proxy for local economic activity and construct a 20-year panel data set at the township level. This novel data set enable us to investigate local economic impacts on a small spatial scale. Using the difference in differences approach, we find that there was a 10% increase in local economic activity in rural townships adjacent to KTX stations 10 years after the launch of the KTX. 2 Chapter 1 Family Clans and Public Goods: Evidence from the New Village Beautification Project in South Korea 1.1 Introduction Heterogeneous communities tend to have worse economic outcomes than homogenous ones. Researchers have found that heterogeneous communities have more conflict, and have fewer public goods, and that members exhibit less trust and interact less often (Knack and Keefer, 1997; Collier and Hoeffler, 1998; Alesina and La Ferrara, 2000; Miguel and Gugerty, 2005).1 Many scholars have investigated the roles of ethnic, linguistic and religious het- erogeneity in communities (Mauro, 1995; Alesina et al., 2003; Garc´ıa Montalvo and Reynal-Querol, 2005). In this paper, I introduce family clans as a different, but com- 1 See Costa and Kahn (2003) for an in-depth review. 3 plementary dimension of heterogeneity in a society. In many societies, family clans play significant economic and political roles. A family clan is broadly defined as a group of people who share the same paternal lineage. In Iraq, for example, individ- uals are more strongly bounded by clan and tribal ties than by ethnic or religious background (Hassan, 2007).2 Although Somalia is the most ethnically homogenous country in Africa, clan warfare devastated the country in the 1990s (Arnold, 2001). In Syria, armed tribes and clans have actively participated in its current civil war (International Crisis Group, 2015). Historically, cleavages between Scottish clans heightened during civil wars, such as the 1689-1745 Jacobite Risings (Barthorp and Embleton, 1982). Likewise, a single clan in Uzbekistan wielded considerable political influence by occupying many important ministry seats (Collins, 2002). I use family clan organization in South Korean villages to study the effect of village clan heterogeneity on the production of public goods. Studying family clans in Korea offers a number of advantages. First, South Korea has little variation in other important dimensions of heterogeneity explored in the literature such as ethnicity, race, language, and landholding, thereby enabling the investigation of one dimension of group heterogeneity (i.e., family clans).3 Second, family clans are located in villages with typically comprised of about 100 households on average. Cooperation, social sanctions, and information transmission, some key ingredients of social capital, could be prevalent in such small communities.4 Third, as I show in this paper, there exist excellent data on both clan composition and public good outcomes. To measure family clan heterogeneity, I use family clan data from Family Clans in Chosun, part of the 1930 census conducted by the Japanese Colonial Government in 2 Immediately after Iraqi independence, tribes were estimated to have 100,000 rifles in their pos- session, whereas the government had 15,000 (Marr, 2011). 3 As an additional advantage is that, unlike ethnicity or race, clan membership is easily identifiable. In general, a Korean individual can easily identify his or her clan by the family name and the ancestor’s place of origin. 4 Due to data limitations, U.S. metropolitan statistical areas (MSAs), which have millions of people, are often used in the social capital literature to study community-level phenomena. 4 Korea. These village-level cross-sectional data include the clan names and the total number of households belonging to every clan in each village in South Korea, as long as the share of clan households exceeded 10% of total village households. Using the share of clan households in each village, I construct heterogeneity measures based on the percentage of the households belonging to the most dominant clan in a village, the Herfindahl Index of clan concentration, and a polarization index suggested by Garc´ıa Montalvo and Reynal-Querol (2005) to capture clan polarization in a village. The identification strategy of this paper relies on the historically determined set- tlement of family clans. Family clans settled in Korea hundreds of years ago. By controlling for family clan identities and township fixed effects, I eliminate alterna- tive explanations that the results are driven by the presence of specific family clans or unobserved characteristics that vary across townships. Furthermore, the estimates are robust to various controls for potential determinants of village clan structure, such as distance to town centers, rivers, and roads, terrain ruggedness, altitude and soil type. In my first empirical analysis, I use the voluntary production of public goods as an outcome variable. This measure originates from government evaluations of how well government resources were used to produce village public goods. Under the New Village Beautification Project (henceforth, NVBP) in 1970, the government distributed 335 bags of cement to every village in South Korea, irrespective of village characteristics. These bags of cement were dropped at the entrance to each village for the production of public goods. In consultation with village elites who were members of village development committees, residents decided how to use cement, and produced public goods. As only cement was provided, village residents contributed voluntary labor and private resources. Unbeknownst to the village residents when they received the bags of cement, the government decided to evaluate how much public goods had been improved with a grade (A, B or C) the following year. I 5 digitized the village evaluations and other village characteristics from the New Village Comprehensive Survey (NVCS), which was published by the government in 1972. I find that there was a robust relationship between family clan heterogeneity and the production of village public goods. The relationship, however, was nonlinear. Using the household share of the largest clan in a village as a heterogeneity measure, I show that there was a concave relationship between the family clan heterogeneity and the probability of getting an A grade. The concave curve peaked when the share of the largest clan in a village is around 40%. Using the Herfindahl Index yields a sim- ilar inverse-U-shaped relationship. Additionally, the polarization index is positively associated with improvement of public goods and the coefficient is statistically signif- icant. A change from no polarization to the maximum level increases the probability of getting an A grade by 4 percentage points. In the second set of empirical analyses, I use land donation as a measure of private contributions for village projects. Village residents donated private land to enable the production of public goods, such as village roads. I digitize Glorious Footsteps, a government publication documenting village public projects in detail between 1970 and 1980. It contains a donation list with the name of each donor, and the amount and type of land donated. I show that the amount of land donated per village household during the 1970s also has a concave relationship with the family clan heterogeneity. When the household share of the largest clan in a village is used as a heterogeneity measure, the concave curve peaks around 0.6.5 I also find a positive association between the polarization index and land donation. My last findings relate to agricultural mechanization as a consequence of village improvements, particularly village roads. Many anthropologists and sociologists have found that village residents widened village roads in the 1970s to facilitate the use of 5 As an alternative outcome measure, I use the total amount of donated land as well as the amount of donated land as a percentage of the total amount of cultivated land in a village. The results are qualitatively the same. 6 wheeled agricultural machines in order to saves labor costs.6 If villages successfully improved village roads through village projects that relied on private contributions such as land donations, it is likely that they would have utilized more wheeled agri- cultural machines. Based on township-level agricultural census data from 1970 and 1980, I show that, 10 years after the NVBP, the change in the number of two-wheeled power tillers per agricultural household between 1970 and 1980 has a positive and statistically significant relationship with clan heterogeneity. The polarization index is positively related with changes in the number of power tillers. The persistent concave relationship between family clan heterogeneity and the pro- duction of and contributions toward public goods may reflect the trade-off between coordination and accountability. Scholars have found that in more homogeneous so- cieties, members coordinate better and contribute more to public goods (Miguel and Gugerty, 2005). Yet other studies have shown that highly homogenous communities lack competition, enabling elites to waste resources in the absence of checks and bal- ances (Platteau and Gaspart, 2003; Acemoglu et al., 2013). In traditional societies without strong political institutions, highly homogenous communities may be likely to have both a high level of coordination and highly autocratic elites. In Korea, clan elders have considerable power in clan and village matters (Lee, 1997; Bang, 2004; Kim, 2009), and there has been, and still is, no formal government or political organi- zation in villages.7 Therefore, as clan homogeneity increases, the provision of public goods may improve since such communities have better coordination. However, as communities become highly homogeneous, the provision of public goods may worsen because clan elites could be more prone to despotism with no institutional counter- vailing force. This reasoning is also consistent with the positive relationship between the polarization and public goods found in my data. 6 It has been widely documented that demand for wider village roads due to increasing wages and the introduction of the power tiller, the most popular agricultural machine in Korea. In a survey, farmers listed road improvement projects as the highest priority at that time (Park, 1998). 7 Public goods such as schools or medical clinics are all located in township districts. 7 My study is most closely related to the literature on the effects of group hetero- geneity on social capital. Ample evidence shows that group heterogeneity is detri- mental to economic and social outcomes such as investment, corruption, the chace of civil war, provision of public goods, group participation and trust (Mauro, 1995; Collier and Hoeffler, 1998; Alesina et al., 1999; Glaeser et al., 2000; Garc´ıa Montalvo and Reynal-Querol, 2005). To my knowledge, this is the first paper in which family clans are used as a measure of group heterogeneity. The results of this paper also complement economic literature on the effects of kinship groups on economic outcomes such as insurance, information sharing, re- source pooling, and credit access (Rosenzweig, 1988; Besley et al., 1993; Munshi, 2003; Fafchamps and Gubert, 2007).8 Additionally, this paper contributes to the role of group heterogeneity and social capital on agricultural modernization. Most related is the work of Isham (2002), who showed that the adoption of improved fer- tilizer is positively associated with village level ethnic homogeneity and household participation in village organizations. The rest of the paper is organized as follows. In the next section, I describe the context of the study; specifically, I provide detailed background information on family clans and the NVBP, a rural intervention program aimed at the improvement of public goods in South Korean villages. I also provide a conceptual framework to explain the relationship between clans and social capital. In section 3, I explain my data collection. In section 4, I discuss empirical strategies before presenting and discussing the regression results in section 5. In section 6, I present a case study before offering some concluding remarks in section 7. 8 See Cox and Fafchamps (2007) for a detailed overview of economic literature on kinship networks. Sociologists also have studied the effects of family and relatives on: out-migration (Palloni et al., 2001); protecting property rights (Peng, 2004); technology adoption (Warriner and Moul, 1992); and coping with long-term personal emergencies (Litwak and Szelenyi, 1969). 8 1.2 Background 1.2.1 Family Clans Clans have fulfilled important economic or political functions in many regions of the world. When governments are weak, clans often assume government functions such as settlement of disputes and protection of property and members from outsiders. In South Korea, clan membership is not only fundamental in traditional social in- teraction, it also offers many benefits such as mutual support in farming and in emergencies. The common ancestor is worshipped collectively. Clan membership also increases emotional security (Song, 1982). A family clan in Korea is defined as a group of people who share the same paternal lineage. Clan membership can be identified by a family name and the ancestor’s place of origin. Every native born Korean belongs to a clan. Unlike ethnicity or religion in other countries, however, there are no politically dominant clans in national politics or violent inter-clan conflicts. Further, Korean villages have extremely low ethnic or linguistic fractionalizations. According to measures by Alesina et al. (2003), the values of ethnic and linguistic fractionalization are both 0.002, one of the lowest values among the countries across the world (see Figure 1.1). Family clans in a village are often highly concentrated in rural villages. Often the name of a village reflects the dominant clan (e.g., Kim’s Village) (Yu, 1986). Villages are often classified by the presence of clans: “the most fascinating persistent example of Korean kin organization is the consanguineous village” (Jacobs, 1985, p. 212). Jacobs (1985) classified villages into four types: (a) all residents are from a single lineage; (b) a single lineage dominates; (c) the residents from strongly competitive lineages; and (d) the residents are from a number of weak lineages. In my data, out of 1,298 villages, there are 23 villages (2%) in which all residents were from a single clan, 143 villages (11%) with one dominant clan (i.e., >50% of members), 163 villages 9 (13%) with large concentrations of two or more clans (i.e., no dominant clan), and 969 villages (75%) with weak lineages.9 A survey performed by the Japanese Colonial Government in the 1930s indicates that out of 1,227 sampled villages with large concentrations of dominant clans, the histories of 17% could be traced back 500 years, 53% between 300 and 500 years, 28% between 100 and 300 years, and just 2% less than 100 years (Korean Studies Advancement Center, 2014).10 These kinship ties remained strong in the early 1970s, even when non-rural areas were experiencing rapid political and economic changes (Kim, 1985). In 1930, there were more than 660 different family clans.11 One explanation for the high concentrations of clans in villages is the change in the inheritance law in the 17th century (Korean Studies Advancement Center, 2014). Prior to 17th century, a father could leave his land equally to both sons and daughters. This enabled daughters to remain in the villages where they were born and their husbands from different clans to move in. Since Korea is a patrilineal society, daughters, once married, were considered to be members of their husbands clans. Hence, different clans could reside in the same village. Once the changed inheritance law excluded daughters from being inheritors, they tended to leave their native villages to live with their husbands.12 This prevented the inherited land from 9 For calculations, I use villages that had not split between 1930 and 1970. I use the polarization index to identify villages with large concentrations of two or more large clans and no dominant clan. To obtain the number 163, I count villages in which the largest clan share does not exceed 50% and the polarization index is above 0.3, roughly the 75th percentile of the sample. 10 The long settlement histories of family clans may show that the initial characteristics that attracted founding ancestors 300 years ago may not have been relevant in the early 1970s, the study period of this paper. Banerjee and Somanathan (2007) used similar argument to justify using historical caste compositions as a regressor over contemporary ones. However, the concern of differential geographical endowment remains. Thus I control extensively for geography related variables. 11 My own calculation based on the family clan data-set used in this paper. 12 A family with no sons often adopted a son from relatives in order to bestow land, and more importantly, continue the family paternal line and ancestor worshipping duties. 10 being owned by outsiders and therefore reduced the inflow of people from different clans.13 Major factors influencing clan settlements were historical. While few systematic studies have been performed, the concentrations of clans in certain locations can be attributed to: (a) relocation due to wars, (b) settlement on land gifted by kings, (c) settlement of loyal families near the tombs of ancestors, and (d) settlement of families of retired senior government officials (Jacobs 1985; Korean Studies Advancement Center 2014). The settlement of elite members raises the concern that the higher quality clans may be concentrated in villages that are conducive to better provision of public goods. In the analysis, I control for clan identities to rule out the alternative explanation that differential characteristics of clans are driving my results. 1.2.2 New Village Beautification Project In this section, I describe the unique Korean rural development policy that led to the production of village public goods in the early 1970s. The South Korean government distributed 335 bags of cement to every village between 1970 and 1971 as part of the New Village Beautification Project (NVBP). The purpose of this project was to encourage village residents to produce public goods. As each bag of cement weigh about 40 kilograms, each village thus received a total of 13.4 tons of cement (Ministry of Home Affairs, 1983, p. 22). Cement was distributed between October 1970 and June 1971, so that villages could take advantage of labor availability during the agricultural off-season (Hwang, 1980). An important aspect of this project was that each village could decide how to use the cement, as long as it was used to produce public goods for the village. Cement was to be used for “village projects meeting villagers’ common needs based upon their general consensus” (Moore, 1984, p. 587). The government suggested several 13 In future work, I plan to exploit the timing of the change in the inheritance law to investigate the possibility of exogenous variation in village clan compositions. 11 potential uses for the cement. For example, villages could improve village roads, re- pair river embankments, build compost/manure collection points, repair public wells or construct common laundry facilities (Kyunghyang, 1970). To decide how the ce- ment would be used, local government officials encouraged villages to create village development committees with five to 10 members each. Since public projects often required substantial land and labor contributions from village members, decisions on village projects, such as the widening of roads, were made in democratic ways, such as voting by a show of hands during the village meetings. Typically, the eldest male from each household attended these meetings (Park, 1998). In the year following the distribution of the cement under the NVBP, the govern- ment systematically evaluated each village on how well the cement had been used to improve village public goods (see Table 1.10 for the timeline of the NVBP). Each vil- lage was given either an A, B, or C grade.14 Township or county government officials visited villages and assigned grades. Using cement for mostly private projects (e.g., paving kitchen floors, building stone fences around houses) resulted in getting a C grade. An A or B grade was given to villages that used the cement to produce public goods. For example, some villages widened and straightened village roads. Others established new village roads. Some villages fixed sewage and drainage pipes. Some created common laundry facilities or village wells. A large number of villages chose to improve transportation infrastructure, partic- ularly roads. A government survey of villages shows that improving transportation was considered to be a top priority by village residents. Table 1.11 shows that the three most desired projects identified by village households related to improving roads and fixing bridges. Farmers wanted to use more agricultural products for farming and transportation. However, the roads were not wide or straight enough to use wheeled machines. Furthermore, villagers wanted to improve roads so that they could access 14 The original classifications were independent village, self-help village, and basic village. I have replaced these labels with grades A, B, and C, respectively. 12 modern modes of transportation such as trucks, buses, taxies or cars (Ministry of Home Affairs, 1978). Villagers provided private resources to improve village road in- frastructure. In 1973, the largest fraction of labor days spent on village projects were dedicated to road improvement. According to national statistics, 29 million out of a total of 36 million labor days (81%) were spent on village roads (Ministry of Home Affairs, 1973, p. 116). The considerable effort and contributions put toward improving village roads in the 1970s was partly due to increasing agricultural wages and the introduction of a new labor-saving agricultural machine, the two-wheeled power tiller. Prior to the introduction of power tillers–the most popular agricultural machine in the history of Korean agriculture–there was little incentive to build wider, straight roads because traditional technology did not require them. Figure 1.17 shows the trend of the number of wheeled agricultural machines per agricultural household. Prior to 1970, there were few households that owned these machines. Starting 1970, there was a rapid adoption of power tillers. Power tillers was highly popular: in 2000, there was 0.7 power tillers per every agricultural household in the country. Before 1970, farmers tilled land using animal power, and residents mostly carried goods on their backs (see Figure 1.18). While I do not have information on the initial quality of village roads, given the nature of traditional agricultural technology in Korean villages, initial road quality, particularly the width of roads, did not seem to differ based on clan structure and other village characteristics. Based on personal interviews with several village elders who participated in the cement projects, most village residents did not seem to be aware that their projects would be evaluated and that they could earn rewards based on how well they used the cement to produce public goods. In the year following the cement distribution, additional resources were given to villages that had received A or B grades. However, this decision had not been planned in advance. After seeing the grade distribution 13 for the cement projects, President Park Chung-hee, who had initiated the NVBP, suggested providing additional resources only to villages that had received A or B grades (Kim, 2006). 1.2.3 Family Clans and Social Capital Why does family clan heterogeneity matter for the production of village public goods and hence for village social capital? I argue that the benefits of participating in village projects and the costs of free riding are both higher if there are more members of the same clan in a village. First, the incentive of a village resident to participate in village projects may have depended on the increased benefit, either in utility or profit, to the participating individual and on the increased benefit to relatives. Therefore, if more residents of a village were related by blood, each individual receives more marginal utility, ceteris paribus. This may have been especially true in the Korean rural context, because clan members exhibited strong social integration, and had a sense of mutual solidarity (Brandt, 1972). In field interviews, Yi (1981) described the positive role of clans in village projects: Of particular importance, according to people interviewed, was the ques- tion of clan. The villages dominated by one clan-group evidently had an easier time getting their residents to cooperate in various...[village] projects; where two or more clans were present, this task was more diffi- cult. (pp. 448-449) Additionally, improving durable public goods by building new roads or bridges benefits not only current clan members, but also future generations, which may have been an incentive to contribute to infrastructure projects. An example from a single clan-dominated village illustrates this point. Among the households comprising Ho- am village in Kyungsang North Province, 95% belonged to the Milyang Park Clan in 14 the 1970s. When land owners were reluctant to donate the lands required to widen village roads, a clan member Kyusam Hong said, “even if we are poor now, let us make our younger generations praise us for giving them better roads” (Ministry of Home Affairs, 1978, p. 552).15 Clans also provide better coordination for labor- intensive agriculture and resolution of conflict (Seo, 1997), which often are mediated by clan elders before they get out of hand (Yesa Moonhan, 2001). Second, the cost of free riding is higher when there are more clan members in a community. Social punishment can be strong when members of the same kinship group have semi-permanent relationships and repeated face-to-face interactions in a village, which is the case in clans in South Korea. Collective ancestor worship cere- monies are performed multiple times each year. Additionally, socially unacceptable behavior may damage not only the reputation of the wrongdoer, but also the repu- tations of immediate family members, such as parents.16 Even if individuals did not care about the benefits to current and future clan members when making participa- tion decisions, it is still reasonable to believe that the social costs may still have had a strong influence in rural Korean societies. Social sanctions have been shown to affect various economic outcomes in rural communities in other countries as well. In rural Ghana, La Ferrara (2003) provide evidence that social sanctions influence the loan default rate. Likewise, Miguel and Gugerty (2005) show that ethnically diverse communities in Kenya fail to impose sanctions on parents who do not contribute to school funding. Lineage groups in other countries also exhibit strong solidarity. For example, in Chinese villages, “lineage groups inculcate a sense of obligation to the group...based 15 If being dominated by a single clan is advantageous to public goods improvement, one may expect that by 1970, these villages would already have had better public goods. However, villages had received few government resources prior to the NVBP. Historically the government had provided resources for the production of public goods located primarily in town centers, not in villages. Even irrigation, an important village public good, mostly relied on proximity to rivers and rainfall, instead of planned irrigation systems. 16 Posner (1980) described this kind of behavior as collective responsibility. 15 on concepts of family and shared patrilineal descent” (Tsai, 2007, p. 359). Mu and Giles (2014) argue that “mutual trust between villagers built through common family lineage may lead to less conflict” (p. 21). In both the U.S. and Hungary, Litwak and Szelenyi (1969) find that relatives are the most helpful group when an individual is dealing with a long-term emergency such as a broken leg that takes 3 months to heal. Among American survey respondents, 73% indicated that relatives would help “very much” in the broken leg scenario, whereas only about 30% indicated that neighbors and friends would help “very much.” In contrast, villages with heterogenous clan membership could face cooperation challenges. Anecdotal evidence suggests that villages with multiple clans may be less cooperative. For example, one village had “forty different family names and it was difficult for them to cooperate” (Ministry of Home Affairs, 1978, p. 625). Likewise, members of another village were described as “selfish because there are numerous families with different clans” (Ministry of Home Affairs, 1978, p. 402). 1.3 Data 1.3.1 Family Clans and Village Heterogeneity For clan membership data, I digitized a publication called Family Names in Korea which was part of the 1930 population census by the Japanese Colonial Government. The publication lists family clans in villages if the number of households associated with a single family clan comprised more than 10% of total households in the village. In some villages, more than one clan met the 10% threshold. For each village in South Korea, data include the number of households belonging to a specific clan and the name of each clan, including ancestral place of origin and family name. Figure 1.3 shows the original version of the cross-sectional data by the Japanese Colonial Government. Another survey on clan membership was conducted in 1985 16 as a part of a population census. However, clan data in 1985 are available only at the township level, and not at the village level. Townships comprise the lowest administrative division in South Korea, and each village belongs to a township. Each village may also include multiple hamlets, which are concentrated pockets of dwellings within a village. According to the data, out of 3,124 villages: no single clan comprised at least 10% of the households in 1,708 villages (55%); at least one clan comprised at least 10% of the households in 1,048 villages (34%); and two or more clans each comprised at least 10% of the households in 368 villages (11%). The mean of the share of households belonging to the largest clan in the village is 16%, with a standard deviation of 0.23. The median share of the largest clan is 0, and the 75th percentile is 0.28. I used multiple measures to capture family clan heterogeneity in a village. First, I used the share of households belonging to the largest clan among the total households in a village (TOPSHARE). Figure 1.4 shows the distribution of TOPSHARE. Second, I constructed the Herfindahl Index (HERF) to measure the family clan concentration within a village. The index is calculated by the formula n X HERFi = share2i , (1.1) i=1 where sharei is the share of clan i in a village, and n is the number of clans exceeding the 10% threshold. A higher index value implies less diversity in clan and higher concentration of a single clan. Figure 1.7 shows the spatial distribution of the index. To capture the effect of the presence of multiple clans within a village, I used a polarization index (POLAR) following Garc´ıa Montalvo and Reynal-Querol (2002). The index is computed using the following formula: n n X 0.5 − sharei o2 P OLARi = 1 − sharei . (1.2) i=1 0.5 17 This index captures the strength of polarization between clans. For example, a village with two clans that each comprised 30% of households would have a higher index value than a village with a single clan that comprised 60% of households. An index value of 1 implies maximum polarization–two clans each comprising 50% of households in a village. An index value of 0 means there was a single clan comprising 100% of households, or there were multiple clans that each comprised a minimal percentage of households in a village. Figure 1.8 shows the spatial distribution of the index. Table 1.3 shows correlations between homogeneity measures. All three measures are highly correlated. TOPSHARE and HERF have a correlation coefficient of 0.92. TOPSHARE and POLAR have a correlation coefficient of 0.83. HERF and POLAR have the least correlation, at 0.64. 1.3.2 Economic Outcomes Public Goods Data Village characteristics and information on the production of public goods come from the New Village Comprehensive Survey (NVCS) published by the Department of Home Affairs in 1972. This government publication includes cross-sectional data from 1971, including names, demographic information, and other characteristics of entire villages in South Korea. Important for this study, the NVCS includes the cement project grades for all villages in South Korea. A total of 16,301 villages received A or B grades, closely matching the official figure of 16,600 villages according to news outlets at that time.17 Figure 1.9 shows the original format of the data. Each page of the NVCS has a list of the villages in each township. A map of each township is also included, which shows the village boundaries. Each map also shows the locations of important landmarks such as electrification, rivers, roads, railroads and highways. I digitized these variables from each township map. Village characteristics used in the 17 See, for example Maeil Kyungjae (1971). 18 empirical analysis of this paper are all from the NVCS. Table 1.1 provides summary statistics. While data are available for the universe of villages in South Korea, I restricted the sample to Kyungsang North Province for the analysis (see Figure 1.16). Kyungsang North Province is the largest province in South Korea with an area of 19,028 square kilometers (7,347 square miles) and a population of 2.6 million in 2010.18 Kyungsang North Province has the most concentrated settlement of clans and hence the most variations in clan structures in villages (Kim, 2012). I used the grades assigned to each village under the NVBP as a measure of the production of public goods by constructing a dummy variable that equals 1 if the grade is an A, and 0 otherwise. I also constructed an alternate dummy variable that equals 1 if the grade is either an A or B, and 0 otherwise. Of the 5,539 villages included in the NVCS that meet my sampling criteria (see Section 3.3), 357 villages received an A grade (6%), 2,417 villages received a B grade (44%), and 2,765 villages received a C grade (50%). Figure 1.10 shows the spatial distribution of the cement project grades. I used the probability of getting an A cement project grade as a measure of road improvement. There are several reasons why this may be sensible. As discussed in detail in the background section, the most desired village project was road improve- ment and more than 81% of total labor hours contributed to village projects were related to road improvement (Ministry of Home Affairs, 1973; Park, 1998). Since road improvement required coordinated donation of land among village residents, road related project could be the most difficult village project. Additionally, when more systematic evaluation of the performance on village projects was introduced later, road improvement was one of top five major criteria (Ministry of Home Affairs, 1983). 18 Kyungsang North Province is slightly smaller than the state of New Jersey in the U.S. 19 Since clan data were collected in the 1930s and cement project grades were recorded in the 1970s, I reconstructed the village geographies based on 1930 data. Some village boundaries changed between 1930 and 1970; most frequently, villages split into separate villages. When a village had split into multiple villages between 1930 and 1970, I merged geography data for the villages and used the weighted mean of characteristics and cement project grades based on the number of household. To overcome data limitations, I imputed total village households in 1930 by multiplying the total number of township households in 1930 by a village’s share of total township households in 1970. Data on Land Donation I compiled data on the amount of land donated for village projects by village resi- dents from Glorious Footsteps, a government publication by Ministry of Home Affairs (1978). This publication has detailed information on village projects for about 350 villages across the country between 1970 and 1980. Importantly for this paper, it contains a list of land donors for village projects for each village (see Figure 1.11 for the original format). For each donor, the list contains information on the type and the amount of land donated. The type of land include land for housing, rice paddies, regular fields, and forest land. For the empirical analysis of the paper, I summed up the total amount of land donated for each village by land type. The main analysis used the total amount of cultivated land donated, which is the sum of rice paddies and regular fields. In robust test, I also tried different combination of summation of each type. The results of the paper were robust to these combinations. Glorious Footsteps also contain the full distribution of family names of each village. Unlike the family clans data from the Japanese Colonial Government, there is no truncation of missing data at the 10% household share. In the analysis using land 20 donation as an outcome, I use the share of each family names as a measure of lineage heterogeneity. Additionally, Glorious Footsteps contains detailed information on land of a village, such as amount of total land, cultivated land, regular dry field, rice paddies, and forest. Because irrigation in rural area was mostly preformed in rice paddies, I computed irrigation rate using the he proportion of rice paddies out of total cultivated land. These characteristics are used as control variables in the analysis on land donation. Table 1.4 provides summary statistics. While Glorious Footsteps data have detailed information for each village, the main shortcoming of the data is that there is no explicit information on how villages were selected. While the introduction of the publication indicated that the villages were chosen to be representative, there is possibility that villages that performed relatively better in village projects. Therefore, the results using these data may indicate the relationship between group heterogeneity and the amount of land donated among the villages that were relatively more active in village projects. Data on Agricultural Machines I used agricultural census in 1960, 1970, and 1980 for the data on agricultural ma- chines owned by village residents. Agricultural machines and tools in the data include power tillers, ox carts, hand carts, water pumps, sprayers and combines. The most relevant information for the analysis is the number of power tillers. Unlike other agricultural machines, I was able to track the changes in the number of power tillers owned by villagers from 1970 to 1980. Due to changes in survey questions of agricul- tural census, data on other agricultural machines are not available for both 1970 and 1980. Since the agricultural census are available at the township level, my empirical analysis on power tillers are also at the township level. Township characteristics are 21 directly from agricultural census. Clan heterogeneity were aggregated from the village level clan heterogeneity measures. I took the weighted average of clan heterogeneity in each village belonging to the same township. The weight was the number of households in each village. Agricultural census also has various township characteristics. These include the number of villages, literacy, occupations, landholdings, and irrigation rate. I com- puted a measure of poverty rate from the share of agricultural households which sold harvested crops to the market. This variable captures subsistence farming. I com- puted the rate of the change in agricultural household between 1960 and 1970 to capture the migration rate. Table 1.5 provides summary statistics. 1.3.3 Geographic and Economic Characteristics of Villages I used various control variables that could be potential determinants of both clan settlements and production of public goods. Michalopoulos (2012) showed that geo- graphic characteristics such as elevation and land quality are important determinants of ethnolinguistic diversity. Similarly, clan diversity could be also explained by ge- ography. Hence the failure to control for these may result in biased estimates as production of public goods using cement may also depend on geographic endowment. For example, to use cement for construction, cement needs mixing with pebbles and water. Therefore, the proximity to river may also an important factor. Additionally, proximity to river may increase productivity of crop field which could attract clans for settlement. I compiled spatial geography data and created village-level geographical character- istics using ArcGIS software. Data on village boundaries in 2014 are from Geoservice Korea 2014. Terrain Rugged Index was obtained from Nunn and Puga (2012). I also used soil types information from the 2007 Digital Soil Map of the World by the Food and Agriculture Organization. A map of river networks in 2014 is from Water 22 Resource Management Information System, a web portal created by the Ministry of Land, Infrastructure, and Transport in South Korea. Road network data in 2014 comes from the National Transport Information Center in South Korea. Elevation data are from the 2000 Shuttle Radar Topography Mission by the U.S. National Aero- nautics and Space Administration. I also used maps showing major battles during the 1950-1953 Korean War, obtained from the U.S. West Point Military Academy website. Village characteristics in 1970 were also obtained from the digitized data from the NVCS. These include village characteristic of demographics, occupations, distance to township centers, proximity to national transportation infrastructure, and some local geographic characteristics. 1.4 Empirical Strategy In this paper, I empirically investigate three questions. First, I use village cement project grades to analyze the relationship between the improvement of public goods and group heterogeneity. Second, I use land donation data to investigate the rela- tionship between the amount of land donations and group heterogeneity. Finally, I evaluate the relationship between agricultural mechanization and group heterogene- ity. While I use separate data for each analysis, they are conceptually related. Since road infrastructure projects were most popular during the 1970s, as described in sec- tion 2, private land was needed to be donated in order to widen, straighten existing roads, and build new ones. The objective (and consequence) of road improvement was the utilization of labor saving agricultural machines. Therefore, I present the relation- ship between clan diversity and cement project grade as evidence of improved road infrastructure. Land donation for village projects is a measure of private contribution 23 for village public goods. The increase in ownership of two-wheeled power tillers 10 years after the NVBP can be interpreted as a consequence of road improvement. Equation 1.3 shows the the main empirical specification used for all outcomes: Yv = α + β heterogeneityv + Xv0 γ + v , (1.3) where heterogeneityv is the measure of clan heterogeneity for village v; Yv is the outcome of interest, such as the improvement of public goods, land donation, and agricultural mechanization; and Xv is a vector of village characteristics used as con- trol variables. The main explanatory variable, heterogeneitym,v , measures the het- erogeneity of village clan compositions, TOPSHARE, HERF, and POLAR. β is the coefficient of interest which estimates the relationship between clan homogeneity and the outcome variable of interest. The identification of the empirical analysis relies on the historically determined patterns of the settlement of clans. The identifying assumption is that, once control- ling for the potential determinants of clan settlement, the error terms are uncorrelated with heterogeneityv . It is reasonable to assume that the settlement was largely deter- mined by geographic and spatial characteristics such as soil quality, access to water sources, terrain ruggedness, and distance from towns, I extensively control for these relevant variables. The main empirical analysis of this paper is on the relationship between family clan heterogeneity and cement project grades at the village level. I include township fixed effects, δm , to account for across township differences in various unobserved char- acteristics which also influences the production of village public goods. Additionally, I control for clan identities, τc , to rule out that the results are driven by specific clans which may be better at producing public goods. Therefore, the identifying 24 assumption is, E(t,v | heterogeneityv , Xv , δm , τc ) = 0, (1.4) that is, clan heterogeneity is exogenous conditional on potential determinants of clan settlement, township fixed effects and clan identities. By controlling for township fixed effects, I compare villages within the same town- ship which is the lowest administrative unit in the country. The average area of a township in my sample is small. It is roughly similar to the land area of Syracuse in the state of New York (25 square miles). Therefore township fixed effects could account for unobserved variables that vary even in a small spatial scale. Addition- ally, I control for village-level geographic characteristics such as terrain ruggedness, distance from rivers and public road network. Using village data, I checked whether mean values of village characteristics differed significantly between two groups of villages: group 1 had TOPSHARE values below the median, and group 0 had TOPSHARE values above the median. Following Kline (2013), I calculated standardized mean differences for village characteristics using the formula (µ1 − µ0 )/σ0 , where µ1 and µ0 are the means of a variable for groups 1 and 0, respectively, and σ0 is the standard deviation of group 0. Figure 1.12 shows standardized mean differences for various village characteristics. The values are mean differences relative to the standard deviation of each variable. Most variables have standardized mean differences that are less than 0.1 of the stan- dard deviation in absolute terms. The average differences in absolute terms are 0.05 of the standard deviations for all village variables shown in the figure. Village character- istics with the largest mean differences are village altitude (0.13) and distance from the town center (0.11), which are not drastic. Potentially important geographical determinants of clan settlements, such as ruggedness, distance from rivers, distance from major roads, and soil type, all show differences less than 0.1. Distance from battle sites during the Korean War, which may be a proxy for differences in initial 25 village public goods due to war time destruction, also show little differentiation (0.01). Other evidence of the exogeneity of the heterogeneity variable is that the estimated coefficients of heterogeneity do not differ whether control variables are included or not. Based on the context of the study, reverse causality can also be ruled out. Vil- lage residents could sort into places with preferred types or quantities of public goods. However, Korean clans–and the ancestors of typical village residents in general–settled in places hundreds of years ago. As farmers often were reluctant to sell farm lands bestowed by ancestors, mobility between villages was limited. Additionally, the gov- ernment did not provide resources for the production of public goods in villages prior to 1970. Hence, the lack of public goods overall may have weakened villagers’ incen- tives to move based on differences in public goods. 1.5 Empirical Results Empirical results show that clan heterogeneity was systematically related to the im- provement of public goods, land donation, and agricultural mechanization. In all three sets of empirical results, I find a concave relationship between outcomes and clan diversity. I suggest a possible explanation for the concavity results at the end of this section. 1.5.1 Effect of Clan Heterogeneity on Cement Grades Clan heterogeneity is systematically related to the improvement of village public goods, measured by NVBP cement project grades. There is a concave and statistically significant relationship between the group heterogeneity measure and the probability of getting an A grade. 26 I used the following specification to test the main hypothesis: 0 cement gradem,v = α + β heterogeneitym,v + Xm,v γ + δm + τc + i , (1.5) where cement gradem,v is a dummy variable with a value of 1 if village v in township m received an A grade from the government under the NVBP, and 0 otherwise. This is the measure of the improvement of village public goods using provided government resources. δm denotes township fixed effects and τc is clan identities.19 Figure 1.13 presents the relationship between TOPSHARE, the percentage of households belonging to the largest clan in a village, and the probability of getting an A grade (sub-figure A). Since the outcome variable is a dummy variable, I divided the sample into 40 bins using the percentiles of TOPSHARE and took the average value of the outcome variable for each bin. Each bin has about 80 villages. The graph shows that the relationship was non-linear and concave. At the low level of TOPSHARE, the increase in TOPSHARE resulted in the increase in the probability of getting an A grade. However, at around 0.5, the relationship becomes negative. A further increase in TOPSHARE results in a reduction of the outcome variable. The relationship between HERF, the Herfindahl Index, and the outcome variable is qualitatively similar to TOPSHARE results (sub-figure B). Sub-figure C shows the positive relationship between POLAR, the polarization index, and the outcome variable. Table 1.6 reports the estimation results. The dependent variable is the dummy variable indicating whether a village received an A grade or not. Panel A shows estimates without any control variables. Panel B includes full controls and town- ship fixed effects. Specifications in each column are based on different heterogeneity measures. Column 1 in panels A and B shows a linear relationship between cement 19 In an alternate specification, the dummy variable equals 1 if a village received an A or B grade and 0 otherwise. 27 project grade and TOPSHARE. Column 2 includes a squared term of TOPSHARE, capturing nonlinearity of the TOPSHARE measure. In column 3, HERF is used as the heterogeneity measure. Column 4 includes the squared term of HERF. In column 5 POLAR is used. Panel B includes various control variables. These include the total number of households, percentage of households engaged in agricultural occupations, average cultivated area per agricultural household, percentage of the population under age 14, distance from the township center, whether the local administrative office was in the village, the number of sub-villages (or hamlets), and the age of the village leader. Additionally, panel B includes dummy variables indicating whether a village had electricity access, and was located next to the sea, a river or stream, county roads, regional roads, national roads, highways, and railroads. It also includes geo- graphic classifications of each village based on the NVCS. I created dummy variables for each of the eight different classifications: agricultural villages, villages near cities, villages near highways, semi-urban villages, villages in mountainous regions, fishing villages, villages near rivers, and villages near the sea.20 Following Michalopoulos and Papaioannou (2013, 2014), I included variables to present geographic character- istics that were potential determinants of village clan structures including the Terrain Ruggedness Index (Nunn and Puga, 2012), altitude and soil type based on classifica- tions from the Food and Agricultural Organization as proxies for crop suitability. I also included distance from major battle fields during the 1950-1953 Korean war, to account for differences in initial levels of public goods due to war time destruction. TOPSHARE does not predict the probability of getting an A grade, with or with- out control variables. However, as expected from Figure 1.13 showing a concave rela- tionship, if the squared term of TOPSHARE (TOPSHARE2 in the table) is included, 20 Classifications are mutually exclusive. For example, if a village can be classified as a fishing village, then this village is not classified as a village near the sea. Specific criteria (e.g., percentage of the population engaged in fishing as an occupation) were used such as the fraction of fishing population to designate each village as either a fishing village or a village near the sea. 28 both TOPSHARE and TOPSHARE2 are statistically significant. As shown in Table 1.7, the estimates are robust to inclusion of control variables. Since TOPSHARE2 has a negative value, the relationship between TOPSHARE and the probability of getting an A grade is concave, and peaks at 0.41. The interpretation is that at low levels of TOPSHARE, an increase in homogeneity is associated with greater improvement of public goods. However, once TOPSHARE reaches 0.41, the relationship becomes negative, and an increase in TOPSHARE is associated with less improvement of pub- lic goods. Similarly, HERF has a concave relationship with the outcome variable that peaks at 0.36. Polarization is positively associated with the outcome variable and the estimates are statistically significant. An increase in polarization from 0 to 1 is associated with a 5 percentage point increase in the probability of receiving an A grade, which is about one-fourth of the standard deviation of the outcome variable, or about 70% of the mean of the outcome variable. In the Appendix, I also show that the concave relationship persists when the sample is restricted to villages that did not split between 1930 and 1970. Because the available clan data were collected in 1930, I used 1930 geographic data in my analysis. If a village split between 1930 and 1970, I merged the split villages back into one. This introduces measurement errors in group heterogeneity; restricting the analysis to non-split villages might reduce the measurement errors. The results show that among non-split villages, concavity still exists and both TOPSHARE and TOPSHARE2 are statistically significant (see Table 1.12). It is possible that the results are mainly being driven by other characteristics that are correlated with the heterogeneity measures. For example, villages with a high per- centage of households belonging to the largest clan already may have had substantial levels of public goods, thus their improvement of public goods might have been low. However, road improvement only gained importance around the time when the ce- ment was distributed. The demand for better roads coincided with the introduction 29 of the power tiller (a two-wheeled, labor-saving agricultural machine), and increases in agricultural wages in the late 1960s. Therefore, it is not likely that some villages already had wider and straighter roads prior to the NVBP, because there was lit- tle incentive to create better road infrastructure. Traditional agricultural technology simply did not require wider roads.21 Furthermore, I controlled for distance from the major battles during 1950-1953 Korean War, to account for initial differences in the levels of village public goods due to destruction. While the cement projects may have complemented to existing village public goods, major public goods such as schools and medical clinics were located at the centers of townships, not in villages. A higher irrigation rate may result in higher agricultural yield, which in turn may lead to higher demand for village public goods such as roads. During the study period, the majority of irrigation was based on natural sources such as rivers and rainfall, and less than 20% of irrigation was performed using a formal irrigation system. While I do not have data on irrigation for the empirical results shown in this section, I extensively control for geographic variables that could predict irrigation such as river network and terrain ruggedness. In the following two outcome variables, land donation and the adoption of power tillers, I explicitly controlled for irrigation rate in the two other empirical analyses. The results are robust to its inclusion.22 Another alternative explanation is that clan diversity may be correlated with geographic characteristics that happen to be conducive to cement use. Since cement must be mixed with pebbles and water, the proximity of these materials, not clan diversity itself, may have influenced the improvement of public goods. I extensively controlled for geographic variables such as proximity to rivers, terrain ruggedness and altitude to account for these factors, and the results are robust to their inclusion. 21 The most widely used method of transportation was A-frame backpack carriers. For tilling, animal power was most often used. These two technologies did not require wider roads. 22 I collected historical high quality maps produced by the Japanese Colonial Government in 1918. The maps contain information on land types, including irrigated rice paddies and regular dry farm fields, and the locations of villages. I plan to digitize maps and include the irrigation rate variable in my analysis. 30 1.5.2 Effect of Clan Heterogeneity on Land Donations I turn to the relationship between the amount of donated land per household and clan heterogeneity. Similar to the cement project results in the previous section, I find a concave relationship. I used the following specification for the analysis: 0 land donationp,v = α + β heterogeneityp,v + Xp,v γ + θp + i , (1.6) where land donationp,v is the amount of donated land per household in a village v and province p, homogeneityv,p denotes heterogeneity measures in the village, and the vector Xv,p contains village level controls. Unlike the previous empirical analysis, here I used province fixed effects denoted by θp , since the donation data are from all provinces in South Korea23 . Data are available for less than 30 villages in each province so there is a limited number of villages from each township. Figure 1.14 shows the relationship between TOPSHARE and the amount of cul- tivated land donated. Similar to cement project grade results, the relationship was non-linear and concave. The relationship between HERF and land donation was sim- ilarly concave. There was a positive relationship between POLAR and land donation. Table 1.8 shows the results. Panel A presents bivariate relationships between het- erogeneity measures and the average amount of donated land per household. Panel B includes village controls and province fixed effects. Control variables include the number of village households, the percentage of agricultural households, average cul- tivated area per household, total amount of cultivated land, and total amount of land. As expected from the relationship shown by the figures, TOPSHARE has a con- cave and statistically significant relationship with land donation. The concave curve peaks around 0.6. HERF also has a concave relationship with land donation that 23 A province consists of townships. 31 peaks at 0.5. There is also a positive relationship between POLAR and the outcome variable. However, HERF and POLAR estimates are not statistically significant when controls are included. These results are related to the previous analysis on cement project grades and group heterogeneity. The production of public goods requires contributions of private resources such as voluntary labor and land. The results in this section indicates that either low or high clan diversity in a village is associated with a relatively smaller amount of land donated. If land donation has a concave relationship with clan diver- sity, it is possible that the production of public goods also has a concave relationship with clan diversity. It is possible that the concavity pattern of the amount of donated land is due to the concavity of the land endowment with respect to clan diversity. That is, villages with a medium level of clan diversity had more cultivated land per household so they tended to donate more. However, I showed in the Appendix that the total cultivated land per household does not have a concave relationship with clan diversity. While the estimates are not statistically significant, it had a convex relationship with clan heterogeneity given the positive sign of the coefficient of quadratic term when control variables are included (Table 1.13). Another concern is that the clan diversity measure in this analysis is based on the family name distribution instead of the actual clan distribution. For example, the family name Kim is associated with many different clans. Since I constructed clan diversity based on the share of each family name in the village, clan diversity has measurement errors. Another concern is that the donation list does not include compensated land contributions. It is possible that some villages had residents with higher incomes who were able to collectively purchase the land needed for village projects. Although the land donation data do not include income or wealth level, I controlled for the 32 rate of irrigation as a proxy for income since a typical village relied on rice crops as a primary source of income. The results shown here are robust to the inclusion of irrigation rate. 1.5.3 Effect of Clan Heterogeneity on Agricultural Mecha- nization One major reason to improve public goods, particularly village roads, was to adopt labor-saving agricultural technology in the 1970s. As discussed in Section 2, once agricultural wages began to increase and power tillers had been introduced, village residents increasingly demanded improvements to village road infrastructure. As shown in previous results, if clan heterogeneity had effects on the improvement of public goods through contributions of private resources such as land and labor, one would expect that power tiller adoption may also be systematically associated with clan heterogeneity. In this section, I investigates this relationship. I find that the change in the number of tillers owned per household in a township is positively associated with clan homogeneity, and positively correlated with the polarization index. The econometric specification is: 0 ∆tiller1970−1980,m = α + β heterogeneitym + Xm γ + i , (1.7) where ∆tiller1970−1980,m is the change in the number of power tillers owned per house- hold in township m, heterogeneitym is the heterogeneity measure, and X is a vector containing the control variables. Figure 1.15 shows the relationship between TOPSHARE and the changes in the number of power tillers owned per household between 1970 and 1980. While there is some indication that the relationship is concave, it is much less pronounced compared 33 to the previous results on cement project grades and land donation. Further, the most of the values of TOPSHARE is from 0 and 0.4. This is due to using township-level information. By aggregating at township level, the variability of the values of village- level TOPSHARE was much reduced. The relationship between HERF and land donation was also concave. Additionally, there was a positive relationship between POLAR and the outcome variable. Table 1.9 shows the regression results. There is a positive and statistically sig- nificant linear relationship between TOPSHARE and the changes in the number of power tillers per household between 1970 and 1980. TOPSHARE and TOPSHARE2 are both statistically significant, suggesting a concave relationship. Since R2 of the specification using both the level and quadratic terms of TOPSHARE (Column 2) are higher than R2 of the specification with level terms only (Column 1), the Col- umn 2 specification provides a better fit with the data. Polarization is also positively associated with the outcome variable and the estimates are statistically significant. These results, however, cannot be interpreted directly as consequences of the pro- duction of public goods between 1970 and 1980. If the production of public goods is the main driver of the increase in power tillers, one would expect that heterogeneity measures should not predict the outcome when controlling for cement project grade. In the Appendix, I show that even after including the probability of receiving an A cement project grade as a control, clan diversity still predicts the outcome (Table 1.14). This implies that other factors related to clan heterogeneity were associated with the increase in the number of tillers. The estimated coefficient of the effect of cement project grade on power tillers has economically large and statistically signifi- cant, which suggests that the improvement of roads was also an important driver of the increase in power tillers. Another concern is that since the analysis is at the township level, the clan het- erogeneity measure is also aggregated at the township level, which could lead to 34 measurement error. Since the projects were mostly implemented at the village level, examining township level data may result in some form of aggregation bias. This is a data limitation, since data on agricultural machines from the agricultural census are at the township level. Additionally, the magnitude of the coefficients drops significantly as more controls are added when estimates from panels A and B are compared. It is possible that some omitted variables could completely explain away the effect of clan heterogeneity on power tiller ownership. 1.5.4 The Concave Relationship between Clan Heterogene- ity and the Production of Public Goods Empirical findings in the previous sections show a robust concave relationship between family clan heterogeneity and the production of public goods and resource contribu- tions. In the existing literature, scholars have shown that group heterogeneity has a linear relationship with the provision of public goods. The concave relationship found in this study may imply that homogeneity yields not only benefits, as suggested in previous literature, but also some possible costs. Finding an empirical answer to this question using data is beyond the scope of this paper. However, a possible trade-off between coordination and accountability could explain the concave relationship. The social capital literature provides evidence that more homogeneous societies have better coordination among members and make more contributions to public goods. Yet, another strand of literature has shown that the absence of checks and balances provides incentives for elites to waste resources.24 For example, Acemoglu et al. (2013) shows that, in Sierra Leone, the number of families 24 Several authors have emphasized the complementarity between social capital and the quality of leaders. Durlauf and Fafchamps (2004) argued that the delivery of public goods depends on local trust and leadership. Krishna (2001) shows that social capital is associated with better social outcomes only in the presence of strong organizational leadership. 35 in a chiefdom was negatively related to comparative development outcomes, and positively related with social capital, suggesting the control of family organizations by the elite. It is possible that in traditional societies without strong political institutions, highly homogenous communities are likely to have a high level of coordination as well as highly autocratic elites. This could be a reasonable characterization of Korean rural villages in the 1970s. The strong influence of the Confucian doctrine of obedience to family elders may have resulted in elders having powerful influence in clan and village matters. This may have been especially true in villages, since they lack formal governmental or political organization. Therefore, the potential tradeoff between coordination and accountability could imply that, as clan homogeneity increases, village residents coordinate better. However, clan elites could more easily control the clan organization and use resources for themselves in the absence of checks and balances. At a higher level of clan homogeneity, public goods provision could be negatively affected. 1.6 Case Study: Effects of Lineage Homogeneity on Cooperation and Participation In this section, I provide qualitative evidence on the benefit of a village’s homogenous lineage in terms of increased contributions to public goods through better coordina- tion and cooperation. This case study links the three sets of empirical results pre- sented in this paper: clan heterogeneity is systematically related to the improvement of public goods, land donations, and agricultural mechanization. Moonsung village in Kyungsang North Province provides a concrete example of the willingness of dominant clan members to donate land (Lim and Lim, 2013). The village is located in the southeastern part of the province (see Figure 1.19). There 36 were 68 households in the village in 1970, and the village was one of the poorest in the region. Approximately 67% of the village was mountainous. The education level was low: just 5% of the population had completed junior high school, and the majority had completed elementary school at most. In October 1970, the village decided to widen 1,800 meters of village access road using the cement from the NVBP. The problem was that 0.25 hectares of land had to be donated. Since agricultural land was the most important asset and the source of income, it was difficult to persuade landowners to donate their precious land. The Namyang Hong clan was the dominant clan in the village, and more than 50% of total households were members. The cohesiveness of the Namyang Hong clan was considered to be very strong. At village meetings, Soon-rak Hong and Seon- pyo Hong volunteered to donate 0.06 hectares and 0.05 hectares, respectively. These two were probably from the Namyang Hong Clan since the family name Hong is not common. Another 17 land owners soon donated their land as well. Since the majority of households was from the same clan, perhaps the two Hongs were more willing to donate their land. If the lack of willingness to donate land was the most binding constraint in improving roads, then the homogeneity in lineage groups in a village could be beneficial for road improvements. The road improvements created benefits. Better roads enabled the villagers to transport goods using hand carts on wider roads instead of carrying them on their backs. Further, the improved roads facilitated agricultural mechanization in Munsung village in the late 1970s. 1.7 Conclusion In South Korean rural villages, village clan heterogeneity had a systematic relation- ship with the production of village public goods, land donation for public projects, 37 and changes in the number of power tillers owned per household. In contrast to the existing literature on the effects of ethnic diversity on the provision of public goods, I found that the relationship between clan diversity and public goods is not linear, but concave, due to the trade off between coordination and leader accountability. If this trade-off is a salient feature of the provision of public goods, then strength- ening political institutions to prevent clan leaders from dominating communities could be an important policy implication. By providing checks and balances, improved po- litical institutions could prevent elites from controlling community members and the high level of social capital could be better utilized for more economically productive uses. The New Village Beautification Project, which this paper is based on, was a pilot project for a following full-blown rural development program called the New Village Movement or Saemaul Undong (SU). SU has recently gained international attention from agencies such as the United Nations Development Programme (UNDP) and the Asian Development Bank as a model for rural development. For example, the UNDP, in partnership with the Korean government, plans to apply lessons from SU to development projects in six developing countries: Uganda, Rwanda, Vietnam, Bolivia, Lao People’s Democratic Republic, and Myanmar. However, while there is an abundance of case studies and qualitative research on SU, few empirical research studies have been conducted on the factors that influenced the success of SU. To my knowledge, this is the first empirical paper in which an economist has systematically evaluated the effect of group heterogeneity on the successful uses of public resources provided under the NVBP using large-scale village-level data. The paper has several limitations. In the absence of clear natural experiments on group heterogeneity, identification concerns remain. While I extensively controlled for potential determinants of clan settlement and production of village public goods, it is possible that some omitted variables may have biased the results. Additionally, 38 the outcome measure of production of public goods in this paper was based on eval- uation grades given by the government. More detailed data on actual improvements to different types of public goods would provide further insights. Finally, by collect- ing relevant data, researchers could directly analyze the relationship between group heterogeneity and leader accountability in future studies. 39 Figures and Tables Table 1.1: Summary statistics for village level analysis mean sd min max Dependent Variables Village with an A grade (dummy) 0.07 0.21 0 1 Heterogeneity Measures TOPSHARE 0.16 0.23 0 1 HERF 0.08 0.17 0 1 POLAR 0.14 0.21 0 1 Village Characteristics Leader age 39.7 6.07 19 73 Average cultivated area per household (ha) 0.94 0.66 0 11.5 Total number of households 180 201 11 4,527 Fraction of agricultural households 0.86 0.17 0.05 1 Fraction of population with age below 14 0.39 0.04 0.12 0.66 Number of hamlets 2.09 1.05 1 9 Distance from town center (km) 4.61 3.55 0 30 Local admin office in village (dummy) 0.05 0.17 0 1 Electricity access in village (dummy) 0.30 0.44 0 1 Sea next to village (dummy) 0.03 0.16 0 1 River passes in village (dummy) 0.37 0.44 0 1 National road passes village (dummy) 0.15 0.33 0 1 Regional road passes village (dummy) 0.19 0.36 0 1 County road passes village (dummy) 0.16 0.34 0 1 Highway passes village (dummy) 0.02 0.12 0 1 Railroad passes village (dummy) 0.06 0.23 0 1 Fraction of improved roofs 0.40 0.25 0 1 Observations 3,124 40 Table 1.2: Summary statistics for village level analysis (continued) mean sd min max Village Classifications by NVCS Agricultural village (dummy) 0.78 0.40 0 1 Village near urban center (dummy) 0.07 0.25 0 1 Village near highway (dummy) 0.05 0.20 0 1 Village belong to town center district (dummy) 0.05 0.19 0 1 Village near mountainous area (dummy) 0.02 0.15 0 1 Village with more than 50% of people with fishing occupation (dummy) 0.01 0.11 0 1 Village located less than four kilometers from coast (dummy) 0.02 0.12 0 1 Geographical and Spatial Variables Area (km2 ) 6.4 6.2 0 94.2 Terrain Ruggedness Index 169 95 2 549 Distance from major roads (km) 3.7 9.4 0 219.5 Distance from major river (km) 1.8 9.2 0 219.6 Distance from major battles during Korean 27.8 18.4 0.1 248.7 War (km) Altitude (m) 212 144 2 955 FAO Soil Type 3964 (dummy) 0.25 0.44 0 1 FAO Soil Type 4295 (dummy) 0.26 0.44 0 1 FAO Soil Type 4352 (dummy) 0.27 0.45 0 1 FAO Soil Type 4290 (dummy) 0.09 0.29 0 1 FAO Soil Type Other (dummy) 0.12 0.32 0 1 Observations 3,124 Table 1.3: Correlation coefficients between heterogeneity measures TOPSHARE HERF POLAR TOPSHARE 1 HERF 0.92 1 POLAR 0.83 0.67 1 41 Table 1.4: Summary statistics for land donation analysis mean sd min max Dependent Variable Amount of cultivated land donated per household (hectare/100) 0.36 0.49 0 3.4 Explanatory Variables TOPSHARE 0.35 0.19 0.09 1 HERF 0.22 0.17 0.02 1 POLAR 0.48 0.17 0 0.89 Village Characteristics Number of village households 96.5 55.6 20 360 Fraction of agricultural households 0.79 0.23 0.03 1 Irrigation rate 0.60 0.49 0 1 Cultivated area per household (hectare / household) 1.09 0.94 0.07 11.2 Total amount of land (hectares) 195.3 228.2 7.1 1,906 Total amount of cultivated land (hectare) 66.1 32.1 3 193 Observations 207 42 Table 1.5: Summary statistics from power tiller analysis mean sd min max Dependent Variable Number of power tillers per agricultural house- 0.19 0.09 0 0.43 holds in 1980 Explanatory Variables TOPSHARE 0.17 0.12 0 0.72 HERF 0.07 0.06 0 0.25 POLAR 0.12 0.1 0 0.57 Township Characteristics Number of villages 12.46 4.54 3 33 Illiteracy rate 0.14 0.04 0.02 0.3 Number of agricultural households 1659 595 398 4,433 Total cultivated area 1442 517 331 3,476 Irrigation rate 0.54 0.15 0.03 0.82 Average cultivated area per agricultural house- 0.89 0.15 0.48 1.3 hold Fraction of harvested agricultural household 0.76 0.17 0.1 1 which sold harvest to market Number of tillers per agricultural household 0.005 0.01 0 0.13 Number of ox carts per agricultural household 0.1 0.09 0 0.44 Number of hand carts per agricultural house- 0.23 0.16 0 0.7 hold Rate of change of agricultural household be- 0.05 0.08 -0.33 0.4 tween 1960 and 1970 Observations 239 Note: All control variables are based in 1970. 43 Table 1.6: Effects of clan heterogeneity on public good production Dependent variable: (Prob. getting an A grade) (1) (2) (3) (4) (5) PANEL A: No controls TOPSHARE 0.02 0.13*** (0.02) (0.04) TOPSHARE2 -0.16*** (0.05) HERF -0.00 0.15** (0.02) (0.06) HERF2 -0.21*** (0.06) POLAR 0.04** (0.02) Peak of concavity 0.41 0.36 Observations 3,124 3,124 3,124 3,124 3,124 R-squared 0.00 0.00 0.00 0.00 0.00 PANEL B: Full controls TOPSHARE 0.03 0.14*** (0.02) (0.05) TOPSHARE2 -0.17*** (0.06) HERF 0.01 0.16** (0.02) (0.07) HERF2 -0.22*** (0.08) POLAR 0.05** (0.02) Peak of concavity 0.41 0.36 Mean of Y 0.07 0.07 0.07 0.07 0.07 Observations 2,668 2,668 2,668 2,668 2,668 R-squared 0.16 0.16 0.16 0.16 0.16 # of townships 220 220 220 220 220 Note: Robust standard errors in parentheses. In Panel B, standard errors are cultured at township level. Dependent variable is a dummy variable which equals one if a village received an A cement grade, and zero otherwise. This measure captures the production of village public goods. An A cement grade is an indication of more production of public goods compared to a B or C grade. TOPSHARE is the household share of the largest clan in the village. TOPSHARE2 is the squared term of TOPSHARE. HERF is the Herfindahl Index based on village clan structure. POLAR is the polarization index by Garc´ıa Montalvo and Reynal-Querol (2002). Panel A does not have control variables. Panel B includes various village characteristics including village size, demographic, prox- imity to urban infrastructure and geographical characteristics. Additionally, it includes township fixed effects and family clan identities. * p < 0.1, ** p < 0.05, *** p < 0.01. 44 Table 1.7: Effects of TOPSHARE on public good production Dependent variable: (Prob. getting an A grade) (1) (2) (3) (4) (5) TOPSHARE 0.02 0.13*** 0.12*** 0.13*** 0.14*** (0.02) (0.04) (0.05) (0.05) (0.05) TOPSHARE2 -0.16*** -0.16*** -0.16*** -0.17*** (0.05) (0.06) (0.06) (0.06) Basic control N N Y Y Y Urban proximity N N N Y Y Geography N N N N Y Township FE N N Y Y Y Mean of Y 0.07 0.07 0.07 0.07 0.07 # of townships 245 245 245 220 Observations 3,124 3,124 3,124 3,124 2,668 Note: Robust standard errors in parentheses. When township fixed effects are included, standard errors are cultured at the township leve. Dependent variable is a dummy variable which equals one if a village received an A cement grade, and zero otherwise. This measure captures production of village public goods. An A cement grade is an indication of more production of public goods com- pared to a B or C grade. TOPSHARE is the household share of the largest clan in the village. TOPSHARE2 is the squared term of TOPSHARE. * p < 0.1, ** p < 0.05, *** p < 0.01. 45 Table 1.8: Effects of clan heterogeneity on land donation Dependent variable: Size of cultivated land donation per HH (1) (2) (3) (4) (5) PANEL A: No controls TOPSHARE 7.80* 37.83*** (4.06) (11.41) TOPSHARE2 - 33.24*** (11.54) HERF 3.73 26.82** (4.52) (13.16) HERF2 -29.87** (13.40) Polarization 7.56 (4.58) Peak of concavity 0.57 0.45 Observations 207 207 207 207 207 R-squared 0.02 0.04 0.00 0.02 0.02 PANEL B: Full controls TOPSHARE 4.66 24.59* (3.93) (11.44) TOPSHARE2 -22.09* (10.60) HERF 1.10 7.05 (4.19) (13.81) HERF2 -7.68 (14.17) Polarization 2.07 (5.03) Peak of concavity 0.56 0.46 Observations 205 205 205 205 205 R-squared 0.17 0.19 0.17 0.17 0.17 # of provinces 8 8 8 8 8 Note: Robust standard errors in parentheses. In Panel B, standard errors are cultured at the province level. Dependent variable is the size of donation of cultivated land per household for village projects in a village. This measure captures private contribution to public goods by village residents. TOPSHARE is the household share of the largest family name in the village. TOPSHARE2 is the squared term of TOPSHARE. HERF is the Herfindahl Index based on family name distribution. POLAR is the polarization index by Garc´ıa Montalvo and Reynal-Querol (2002). Panel A does not have control variables. Panel B includes various village characteristics including village size, share of agricultural households in the village, irrigation rate, cultivated area per household, total area of the village. It also include province fixed effects. Top one percent of the outcome variable is dropped as outliers. * p < 0.1, ** p < 0.05, *** p < 0.01. 46 Table 1.9: Effects of clan heterogeneity on power tiller ownership Dependent variable: ∆ Number of power tillers per HH 1970-1980 (1) (2) (3) (4) (5) PANEL A: No controls TOPSHARE 0.25*** 0.61*** (0.06) (0.18) TOPSHARE2 -1.04** (0.50) HERF 0.36*** 1.11*** (0.11) (0.30) HERF2 -3.93*** (1.51) POLAR 0.21*** (0.06) Peak of concavity 0.29 0.14 Mean of Y 0.18 0.18 0.18 0.18 0.18 Observations 245 245 245 245 245 R-squared 0.07 0.09 0.05 0.08 0.05 PANEL B: Full controls TOPSHARE 0.11** 0.33** (0.04) (0.13) TOPSHARE2 -0.63* (0.36) HERF 0.11 0.47** (0.07) (0.21) HERF2 -1.84* (0.97) POLAR 0.09** (0.04) Peak of concavity 0.26 0.13 Mean of Y 0.18 0.18 0.18 0.18 0.18 Observations 245 245 245 245 245 R-squared 0.60 0.61 0.60 0.60 0.60 Note: Robust standard errors in parentheses. Dependent variable is the changes in the number of household owned power tillers per household between 1970 and 1980. TOPSHARE is the weighted average of household share of the largest family clan in the village, aggregated at the township level with weight being the number of village households. TOPSHARE2 is the squared term of TOP- SHARE. HERF is the Herfindahl Index based on family clan distribution. POLAR is the polariza- tion index by Garc´ıa Montalvo and Reynal-Querol (2002). Panel A does not have control variables. Panel B include various township level characteristics including township population size, size of cultivated land, literacy, irrigation rate, ownership of agricultural machines in 1970. * p < 0.1, ** p < 0.05, *** p < 0.01. 47 Figure 1.1: Comparison of ethnic fractionalization measures across countries Ethnic Fractionalization Measures (Alesina et al, 2002) South Korea Hong Kong China Argentina Turkey Singapore Sri Lanka India United States Switzerland Brazil Kazakhstan Iran Malawi Indonesia Afghanistan Kenya Uganda 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Source: Alesina et al. (2003) Figure 1.2: Comparison of linguistic fractionalization measures across countries Linguistic Fractionalization Measures (Alesina et al, 2002) South Korea Brazil Argentina China Hong Kong Turkey United States Singapore Sri Lanka Switzerland Malawi Afghanistan Kazakhstan Iran Indonesia India Kenya Uganda 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Source: Alesina et al. (2003) 48 Figure 1.3: An example of family clan data from Family Names in Chosun “A-po” Township “Yin” Village “Park”: Family Name “Mil-Yang”: Ancestor’s Place of Origin “58”: Number of Households 49 Figure 1.4: Histogram of TOPSHARE 6 4 2 0 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 TOPSHARE Figure 1.5: Histogram of HERFINDAHL 8 6 4 2 0 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 HERF Figure 1.6: Histogram of POLARIZATION 6 4 2 0 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 POLAR 50 Figure 1.7: Spatial distribution of the Herfindahl Index for family clans 51 Figure 1.8: Spatial distribution of the polarization index for family clans 52 Figure 1.9: An example of village data from New Village Comprehensive Survey 53 Figure 1.10: Spatial distribution of cement project grades 54 Figure 1.11: An example of a land donation list from Glorious Footsteps 55 Figure 1.12: Standardized mean differences of village characteristics Leader age Average cultivated area per household Total number of households Fraction of agricultural households Fraction of population with age below 14 Number of hamlets Distance from town center Local admin office in village (dummy) Electricity access in village (dummy) Sea next to village (dummy) River passes in village (dummy) National road passes village (dummy) Regional road passes village (dummy) County road passes village (dummy) Highway passes village (dummy) Railroad passes village (dummy) Fraction of improved roofs Agricultural village (dummy) Village near urban center (dummy) Village near highway (dummy) Village belong to town center district (dummy) Village near mountainous area (dummy) Village with fishing occupation (dummy) Village near coast Area Terrain Ruggedness Index Distance from major roads Distance from major river Distance from Korean War battles Sea Level FAO Soil Type 3964 (dummy) FAO Soil Type 4295 (dummy) FAO Soil Type 4352 (dummy) -.5 -.25 0 .25 .5 Note: The figure plots standardized mean differences of village characteristics between villages with TOPSHARE above (group 1) and below the median (group 0). The dots represent the standardized mean differences of each variable. The lines represent the 95% confidence intervals. The median value of TOPSHARE is zero. The values are calculated by (µ1 − µ0 )/σ0 , following Kline (2013), where µ1 and µ0 are the means of a variable for group 1 and group 0, respectively, and σ0 is the standard deviation of group 0. Figure 1.13: Effect of clan heterogeneity on public good production .15 .15 .15 Prob. of Getting A Grade Prob. of Getting A Grade Prob. of Getting A Grade .1 .1 .1 .05 .05 .05 0 0 0 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 TOPSHARE HERF POLAR (a) TOPSHARE (b) HERFINDAHL (c) POLARIZATION Note: Scatter plot showing relationship between probability of getting an A grade (y axis) and group heterogeneity measures (each sub-figure) in the raw data. Each point in the graphs represents the average value of the probability of getting an A grade in a bin with approximately 80 villages in each bin. There are 40 bins. The lines represent the quadratic fit of the data for (a) and (b), and the linear fit for (c). 56 Figure 1.14: Effect of clan heterogeneity on land donation 25 25 25 20 20 20 Donated land per household Donated land per household Donated land per household 15 15 15 10 10 10 5 5 5 0 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 TOPSHARE HERF POLAR (a) TOPSHARE (b) HERFINDAHL (c) POLARIZATION Note: Scatter plot showing relationship between the amount of cultivated land donated per household (y axis) and group heterogeneity measures (each sub-figure) in the raw data. Each point in the graphs represents the average value of the probability of getting an A grade in a bin with approximately 4 villages in each bin. There are 60 bins. The lines represent the quadratic fit of the data for (a) and (b), and the linear fit for (c). Figure 1.15: Effect of clan heterogeneity on power tiller ownership Changes in power tiller ownership per HH 1970-1980 Changes in power tiller ownership per HH 1970-1980 Changes in power tiller ownership per HH 1970-1980 .3 .3 .3 .25 .2 .2 .2 .1 .1 .1 .15 0 0 0 .1 .2 .3 .4 0 .05 .1 .15 .2 .25 0 .1 .2 .3 .4 TOPSHARE HERF POLAR (a) TOPSHARE (b) HERFINDAHL (c) POLARIZATION Note: Scatter plot showing relationship between the change in number of power tillers owned per household between 1970 and 1980 (y axis) and group heterogeneity measures (each sub-figure) in the raw data. Each point in the graphs represents the average value of the probability of getting an A grade in a bin with approximately 5 villages in each bin. There are 50 bins. The lines represent the quadratic fit of the data for (a) and (b), and the linear fit for (c). 57 Appendix Tables Table 1.10: Timeline of the New Village Beautification Project Year Event 1969 Excess supply of cement by cement industry 1970 Distribution of cement bags to villages 1970-1971 Production of village public goods 1971 Evaluation by the government (A,B,C grades) 1971 Decision for rewarding villages Table 1.11: Top priority village projects identified from a government survey Order Description of Project 1 Village access roads to be straightened and widened 2 Old bridges over streams to be reconstructed 3 Village roads to be widened and straightened 4 Sewage systems in village area to be improved 5 Thatched roofs to be replaced by cement made tiles 6 Old fences of farm houses to be repaired 7 Traditional wells for drinking water must be improved 8 Village hall to be constructed 9 River banks to be repaired 10 Feeder roads to fields to be developed 11 Rural electrification to be speeded up 12 Village owned telephones to be installed 13 Village owned bathhouse to be built 14 Children playground to be constructed 15 Laundry place in riverside to be improved 16 Trees and flowers to be planted for beautification Source: Park (1998) 58 Table 1.12: Effects of clan heterogeneity on public good production for non-split villages Dependent variable: (Prob. getting an A grade) (1) (2) (3) (4) (5) PANEL A: No controls TOPSHARE 0.03 0.23*** (0.02) (0.07) TOPSHARE2 -0.28*** (0.08) HERF 0.01 0.25*** (0.03) (0.09) HERF2 -0.32*** (0.10) POLAR 0.09** (0.04) Observations 1,298 1,298 1,298 1,298 1,298 R-squared 0.00 0.01 0.00 0.01 0.01 PANEL B: Full controls TOPSHARE 0.04 0.18** (0.04) (0.09) TOPSHARE2 -0.20** (0.09) HERF 0.02 0.14 (0.04) (0.13) HERF2 -0.16 (0.13) POLAR 0.06 (0.05) Observations 1,087 1,087 1,087 1,087 1,087 R-squared 0.17 0.17 0.17 0.17 0.17 # of townships 202 202 202 202 202 Note: Robust standard errors in parentheses. Samples are restricted to villages that did not split between 1930 and 1970. In Panel B, standard errors are cultured at the township level. Dependent variable is a dummy variable which equals one if a village received an A cement grade, and zero other- wise. This measure captures production of village public goods. An A cement grade is an indication of more production of public goods compared to a B or C grade. TOPSHARE is the household share of the largest clan in the village. TOPSHARE2 is the squared term of TOPSHARE. HERF is the Herfindahl Index based on village clan structure. POLAR is the polarization index by Garc´ıa Mon- talvo and Reynal-Querol (2002). Panel A does not have control variables. Panel B includes various village characteristics including village size, demographic, proximity to urban infrastructure and ge- ographical characteristics. It includes township fixed effects and clan identity dummies. * p < 0.1, ** p < 0.05, *** p < 0.01. 59 Table 1.13: Effects of clan heterogeneity on cultivated land Dependent variable: Amount of cultivated land per HH (1) (2) (3) (4) (5) PANEL A: No controls TOPSHARE -0.12 1.07* (0.23) (0.61) TOPSHARE2 -1.31** (0.61) HERF -0.26 0.45 (0.26) (0.85) HERF2 -0.93 (0.90) POLAR 0.09 (0.31) Observations 205 205 205 205 205 R-squared 0.00 0.01 0.00 0.00 0.00 PANEL B: Full controls TOPSHARE -0.10 -0.22 (0.10) (0.27) TOPSHARE2 0.13 (0.25) HERF -0.12 -0.40 (0.12) (0.38) HERF2 0.36 (0.37) POLAR -0.15 (0.13) Observations 205 205 205 205 205 R-squared 0.89 0.89 0.89 0.89 0.89 # of provinces 8 8 8 8 8 Note: Robust standard errors in parentheses. In Panel B, standard errors are cultured by province. Dependent variable is the amount of cultivated land per household in a village. The unit of the de- pendent variable is hectares. TOPSHARE is the household share of the largest family name in the village. TOPSHARE2 is the squared term of TOPSHARE. HERF is the Herfindahl Index based on family name distribution. POLAR is the polarization index by Garc´ıa Montalvo and Reynal-Querol (2002). Panel A does not have control variables. Panel B includes various village characteristics including village size, share of agricultural households in the village, irrigation rate, cultivated area per household, total area of the village. It includes province fixed effects. Top one percent of the outcome variable is dropped as outliers. * p < 0.1, ** p < 0.05, *** p < 0.01. 60 Table 1.14: Effects on power tiller ownership with a cement grade control Dependent variable: ∆ number of power tillers per HH 1970-1980 (1) (2) (3) (4) (5) PANEL A: Full controls TOPSHARE 0.11** 0.33** (0.04) (0.13) TOPSHARE2 -0.63* (0.36) HERF 0.11 0.47** (0.07) (0.21) HERF2 -1.84* (0.97) POLAR 0.09** (0.04) Mean of Y 0.18 0.18 0.18 0.18 0.18 Observations 245 245 245 245 245 R-squared 0.60 0.61 0.60 0.60 0.60 PANEL B: Full controls with the average share of A cement villages Share of A grade villages 0.08*** 0.08** 0.09*** 0.08*** 0.09*** (0.03) (0.03) (0.03) (0.03) (0.03) TOPSHARE 0.11** 0.31** (0.04) (0.12) TOPSHARE2 -0.59* (0.35) HERF -0.12 2.96* (0.07) (1.70) HERF2 -1.70* (0.94) POLAR 0.09** (0.04) Observations 245 245 245 245 245 R-squared 0.61 0.62 0.61 0.61 0.61 Note: Robust standard errors in parentheses. Dependent variable is the changes in the number of household owned power tillers per household between 1970 and 1980. TOPSHARE is the weighted average of household share of the largest family clan in the village, aggregated at the township level with weight being the number of village households. TOPSHARE2 is the squared term of TOP- SHARE. HERF is the Herfindahl Index based on family clan distribution. POLAR is the polariza- tion index by Garc´ıa Montalvo and Reynal-Querol (2002). Panel A includes various township level characteristics including township population size, size of cultivated land, literacy, irrigation rate, ownership of agricultural machines in 1970. Panel B adds another control which is the average share of A cement villages in a township. * p < 0.1, ** p < 0.05, *** p < 0.01. 61 Table 1.15: Mean comparisons of village characteristics by clan concentration TOPSHARE above below diff. t-stat median median Cement grade (=1 if Excellent/Good) 0.60 0.53 0.07∗∗ 2.55 HERF 0.26 0.00 0.26∗∗∗ 23.16 POLAR 0.37 0.00 0.37∗∗∗ 39.47 Village Characteristics Leader age 39.8 39.7 -0.07 0.18 Average cultivated area per household 0.97 0.97 -0.00 -0.13 Total number of households 122 111 11.27∗∗∗ 2.81 Fraction of agricultural households 0.88 0.88 0.01 0.91 Fraction of population with age<14 0.39 0.39 -0.00 -1.38 Number of of sub-villages 2.36 2.40 -0.04 -0.55 Distance from town center 4.74 5.41 -0.66∗∗∗ -3.19 Local admin office in village (dummy) 0.03 0.03 -0.00 -0.02 Electricity access in village (dummy) 0.31 0.29 0.02 0.89 Sea next to village (dummy) 0.01 0.02 -0.01 -1.18 River pass in village (dummy) 0.44 0.38 0.05∗ 1.9 National road passes village (dummy) 0.13 0.16 -0.03 -1.39 Regional road pases village (dummy) 0.17 0.16 0.01 0.39 County road passes village (dummy) 0.17 0.15 0.01 0.63 Highway passes village (dummy) 0.02 0.02 -0.00 -0.14 Railroad passes village (dummy) 0.07 0.05 0.01 0.86 Fraction of improved roofs 0.38 0.37 0.02 1.05 Agricultural village (dummy) 0.81 0.77 0.04 1.53 Near urban village (dummy) 0.06 0.08 -0.02 -1.17 Near highway village (dummy) 0.06 0.04 0.01 1.10 Village belong to town district (dummy) 0.03 0.03 -0.01 -0.78 Village near mountainous area (dummy) 0.03 0.04 -0.01 -0.71 Village with more than 50% of people 0.01 0.01 -0.00 -0.26 with fishing occupation (dummy) Village located less than four kilometers 0.01 0.02 -0.02∗∗ -2.54 from coast (dummy) Observations 485 813 Note: * p < 0.1, ** p < 0.05, *** p < 0.01. 62 Table 1.16: Mean comparisons of village characteristics of split v.s. non-split villages non- split diff. t-stat split Cement grade (=1 if Excellent/Good) 0.55 0.51 0.05∗∗ 3.06 TOPSHARE 0.18 0.19 -0.01 -1.28 HERF 0.10 0.07 0.02∗∗∗ 3.69 POLAR 0.14 0.15 -0.01 -1.15 Village Characteristics Leader age 39.7 39.6 0.12 0.52 Average cultivated area per households 0.97 0.92 0.05∗∗ 2.14 Total number of households 115 226 -111∗∗∗ -18.30 Fraction of agricultural households 0.88 0.84 0.04∗∗∗ 6.40 Fraction of population with age<14 0.39 0.39 0.00 0.10 Number of of sub-villages 2.39 1.88 0.51∗∗∗ 12.96 Distance from town center (km) 5.16 4.22 0.94∗∗∗ 7.18 Local admin office in village (dummy) 0.03 0.05 -0.02∗∗∗ -3.32 Electricity access in village (dummy) 0.30 0.31 -0.01 -0.73 Sea next to village (dummy) 0.02 0.04 -0.02∗∗∗ -2.78 River pass in village (dummy) 0.40 0.35 0.05∗∗∗ 3.28 National road passes village (dummy) 0.15 0.15 0.00 0.16 Regional road pases village (dummy) 0.16 0.21 -0.05∗∗∗ -3.54 County road passes village (dummy) 0.16 0.16 -0.00 -0.34 Highway passes village (dummy) 0.02 0.02 0.00 0.23 Railroad passes village (dummy) 0.06 0.07 -0.01 -1.19 Fraction of improved roofs 0.37 0.42 -0.05∗∗∗ -5.68 Agricultural village (dummy) 0.79 0.78 0.01 0.61 Near urban village (dummy) 0.07 0.07 0.00 0.47 Near highway village (dummy) 0.05 0.04 0.00 0.42 Village belong to town district (dummy) 0.03 0.06 -0.02∗∗∗ -3.46 Village near mountainous area (dummy) 0.04 0.02 0.02∗∗∗ 3.68 Village with more than 50% of people 0.01 0.02 -0.01∗∗∗ -3.39 with fishing occupation (dummy) Village with more than 50% of people 0.00 0.00 -0.00∗ -1.66 with aquaculture occupation (dummy) Village located less than four kilometers 0.02 0.02 -0.00 -0.21 from coast (dummy) Observations 1298 1826 Note: * p < 0.1, ** p < 0.05, *** p < 0.01. 63 Appendix Figures Figure 1.16: A map of South Korea with provincial boundaries Note: Kyungsang North Province is highlighted. 64 Figure 1.17: The average number of agricultural machines per agricultural household Source: Agricultural census Figure 1.18: Land tilling and transportation Note: Power tillers (left), traditional land tilling (middle), and A frame (right) Source: http://cfile6.uf.tistory.com/image/160FB14B4F9D6EED296D56, http://www.zipul. com/coding/sub4/sub5.asp?bseq=9&mode=view&cat=-1&aseq=448&page=6&sk=&sv=, http:// blog.joins.com/media/folderlistslide.asp?uid=silhouette7&folder=12&list_id=6922086 65 Figure 1.19: Location of Moonsung village in the case study 66 Figure 1.20: Location of villages that did not experience geographical split between 1930-1970 67 Chapter 2 The Effect of War on Local Collective Action: Evidence from the Korean War 2.1 Introduction War can cause immense damage and lead to countless deaths of both military per- sonnel and civilians. As a result of the destruction of physical capital and the loss of human capital, production and income both decrease in the short term. Uncertainty remains, however, as to whether there are persistent economic consequences of war. If a war causes short term disturbances of physical capital accumulation or reduces the population level, the neoclassical growth model predicts no changes in the growth path (i.e., the economy will quickly converge back to its pre-war state). On the other hand, if a war changes fundamental aspects of an economy, such as institutions and social norm, the long-run growth path of the economy can be permanently altered. The 1950–1953 Korean War provides compelling historical evidence of community- level social division. During the war, members of the North Korean People’s Army 68 (NKPA) stationed in South Korea executed a significant number of civilians labeled as anti-communists. At the time, typical farmers were functionally illiterate and lacked knowledge of communism. Yet, they had to side with either anti- or pro- communist groups, often involuntarily. This unprecedented social division severely damaged community-level social cohesion. In this paper, I investigate whether the ideological conflict inflicted during the Korean War is associated with lasting damage to the social fabric of affected commu- nities using census data and a novel data on collective action.1 As a measure of the severity of conflict, I use the changes in the civilian population that occurred during the period from 1949 (just before the war) to 1954 (immediately after the war) fol- lowing Davis and Weinstein (2002). As a measure of community cooperation, I use the Korean government’s evaluations of the use of public resources distributed to each village under the 1970-1971 New Village Beautification Project. Each village received bags of cement intended for the production of village public goods. A year later, the government systematically evaluated each village’s cement usage and assigned one of three grades: A, B, or C. A village received an A grade if it produced relatively more public goods than a village with a B grade. A village received a C grade if it produced few public goods. Since the production of public goods requires voluntary labor and private contributions, I use the probability of receiving either an A or B cement project grade as a proxy for community cooperation. I demonstrate that the severity of conflict has an impact on community coopera- tion 20 years after the war ended. A 10% reduction in a township’s civilian population was associated with a 2 percentage point reduction in the probability of using cement for the production of public goods. A township is an administrative unit comprised of 10 to 20 villages. The effect is statistically significant and the magnitude is eco- 1 Investigating the role of social division on conflict is important, given mounting evidence of the effect of contemporary social divisions on economic outcomes such as income, investment, corruption, institutional efficiency and public goods provision (Knack and Keefer, 1997; Alesina et al., 1999; Alesina and La Ferrara, 2000; Banerjee et al., 2005; Miguel and Gugerty, 2005; Khwaja, 2009). 69 nomically meaningful. A one standard deviation decrease in the civilian population is associated with a decrease of one-fifth of the standard deviation in the cement measure. I then analyze whether the reduction in population level during war was short term and whether the population converged back to the pre-war trend. For the analysis, I divide townships into two groups depending on whether a township experienced a decrease in the civilian population or not during the war. Using the population trend of the group without population reductions during the war as a counterfactual population trend, I show that the population reduction from the war persisted for more than 40 years. I then turn to investigate whether the social division is a channel through which the conflict affected community cooperation. The unique Korean context allows me to compare the effects of war through social divisions and through destruction of physical capital from conventional battles within the same national boundary. South Jeolla province did not experience conventional war battles, but political purges were frequent. On the other hand, North Kyungsang province suffered by military battles between NKPA and the UN forces, but it experienced little purges. Consistent with the hypothesis that the social division has lasting influences on cooperation, I find that the severity of conflict was associated with community cooperation only in South Jeolla. In North Kyungsang, I find little association between conflict and cooperation. This work contributes to the literature on the impact of violence on social capital in two ways. First, I introduce ideological conflict within community as a novel explanation for the association between violence and social capital. In the existing literature, the evidence of the effects of conflicts on social capital is mixed, suggesting the existence of various channels through which conflicts could influence social capital. Some scholars find the positive effects of violence on social capital, typically measured by trust from survey data (Bozzoli et al., 2011; Cassar et al., 2011; Becchetti et al., 70 2014). Other scholars find positive effects of civil conflicts on social capital, such as political participation and measurement from experiments (Bellows and Miguel, 2006, 2009; Blattman, 2009; Gilligan et al., 2014). Second, I provide empirical evidence of the persistence of the effect of conflict on social capital. While there is a lack of extensive research on the topic of persistent damages in social capital, my empirical results contrasts with De Luca and Verpoorten (2015) who find that armed conflict in Uganda decreased trust and associational membership only temporarily. They document that the negative effect lasted only a few years. One possible reason of the different degree of persistence could be related to whether perpetrators of violence were just following orders (returning soldiers in Uganda) or whether they actively destroyed community social fabric (political purges by community residents in South Korea). Additionally, this work is broadly related to literature on the effects of war on economic outcomes today. In the existing literature, scholars find little evidence of long-run effects of war associated with the destruction of physical capital (Davis and Weinstein, 2002; Brakman et al., 2004; Miguel and Roland, 2011). However, my empirical results resonate with the literature on the effect of civil conflicts in African countries that document the existence of long run effects from conflict (Blattman and Annan, 2010; Voors et al., 2012; Besley and Reynal-Querol, 2014). My work also contributes to literature on political purges in general as well as purges perpetuated by communists (Getty, 1987; Chandler, 1999; Strauss, 2002; Ace- moglu et al., 2011). To my knowledge, this is one of the first empirical papers on the effect of political purges on social capital, as well as on the effect of the Korean War on economic and social outcomes. The rest of the paper is organized as follows. In the next section, I describe the context of the study on the Korean War and anti-communist purges. Then I explain 71 my empirical strategy in section 3. In section 4, I describe data before proceeding to empirical results in section 5. I provide some concluding remarks in section 6. 2.2 Context 2.2.1 The Korean War (1950-1953) After obtaining independence from the Japanese colonial government in 1945, the Korean peninsula was divided into two governments, one in the north backed by So- viet Union and the other in the south supported by the United States. North Korea invaded South Korea on June 25th, 1950, with support from the Soviet Union and communist China. The South Korean army was ill prepared. On the other hand, the NKPA possessed Russian T-34 tanks and had support from heavy artillery. Hast- ings (1987) observed that “communists...[were] checked more by terrain and natural obstacles than by the [South Korean] forces as they forged through the gaps in the hills.” Just two months after the war began, when the Joint United Nations Forces intervened to counter the North Korean attacks, most parts of South Korea were already occupied by the NKPA. The battles ended with the armistice in 1953. Damages from the war were severe. The total value of property losses in South Korea was estimated to be approximately similar to the entire gross national product of South Korea in 1949. It is estimated that 3 million people were either killed, wounded or missing during the war. Furthermore, approximately 5 million refugees fled war-torn areas (Oberdorfer, 1997). The number of deaths and casualties from the Korean War was significant compared to other major wars. While the number of battle deaths during the Korean War was smaller than battle deaths during the Vietnam War or during the World War I, non-battle deaths totaled 21,000, almost 72 twice the number from the Vietnam War (Edwards, 1998).2 Moreover, the number of North Korean and Chinese casualties exceeded 500,000 (Edwards, 2003). 2.2.2 War Damages in South Jeolla Province South Jeolla province, which is the focus of this study, is located in the southwestern corner of the country (see Figure 2.1). The province was mostly poor and agrarian throughout Korean history (Wickham, 1999). During the war, the province experi- enced a disproportionately large number of civilian deaths compared to the rest of the nation. According to one estimate, more than 70% of total civilian deaths occurred in this region (Park, 2005). When the UN forces launched their counterattack, they landed in the west near Seoul, the capital city of South Korea, and Pusan in the southeast, which is the second largest city.3 As a result, some members of the North Korean army were trapped in South Jeolla Province because their escape routes were cut off (see Figure 2.2). The NKPA was essentially trapped in this region whereas North Korean soldiers in other regions were able to retreat back to the North more easily because they had easy access to escape routes back through mountains and the east coast (Gibney, 1992). It was reported that 15,000 NKPA soldiers and local communist supporters re- mained in the South (Korea Institute of Military History, 2001). Even by the middle of May in 1951, 11 months after the war began, the guerrilla forces were not com- pletely eliminated (Korea Institute of Military History, 2001). The NKPA was able to linger in the mountains because the strategic priority of UN Forces was not to eliminate trapped NKPA soldiers, but to recapture the capital city of South Korea 2 The total battle deaths during the Korean War was estimated to be 34,000. During the Vietnam War, the battle deaths were 47,000. During the World War I, the battle deaths were 54,000 (Edwards, 1998). American casualties during the Korean War were 50,000 dead and 291,000 wounded (Edwards, 2003). 3 See Hastings (2010) for details on General MacAurthur’s Inchon landing on September 15, 1950. 73 and force the NKPA to retreat back to the north.4 As a result, while civilian deaths did occur, no major battles between the NKPA and UN Forces took place in South Jeolla during the war (see Figure 2.3). 2.2.3 Anti-communist Purges The prolonged presence of North Koreans in South Jeolla severely damaged social cohesion and increased tensions and hostility within communities. Oberdorfer (1997, p. 10) notes: One of the most important consequences of the war was the hardening of ideological ... lines. The antipathy ... was deepened into a blood feud among family members, extending from political leaders to the bulk of the ordinary people ... The thirteen-hundred-year-old unity of the Korean people was shattered. The NKPA set up ad hoc courts called people’s courts to purge anti-communists in villages. Accusations were typically made by village members, and people who were labeled as anti-communists were executed onsite, often by their accusers (Park, 2005). Due to the presence of the NKPA, one had to take sides with either the pro-communist or anti-communist group, often involuntarily. This ideological divide severely damaged social cohesion. For example, once the NKPA came to town, an elementary school teacher who was a communist sympathizer killed his own pupils whose parents were thought to be anti-communists (Kim, 2003). Some historians have documented that existing conflicts within communities were amplified as some rival groups exploited the people’s courts and accused other groups of being anti- communists (Park, 2005; Park, 2010). 4 UN Forces did not take part in eliminating NKPA troops hiding in the mountains. The U.S. Joint Chiefs of Staff issued a directive to the Chief of the UN Command that “guerrilla activities should be dealt with primarily by the forces of the Republic of Korea, with minimum participation by United Nations contingents (Schnabel, 1972, p. 183).” 74 Park’s memoirs provide a vivid story related to the people’s courts (Park, 1999, p. 59). He was accused of being an anti-communist by another village member who had a personal grudge toward him. Park wrote: “This is of of the vilest enemies!” he yelled, grabbing my hair and shaking my head mercilessly. ... [he] was raving happily with this opportunity for revenge. I was moved to the second cell and there I found the principal of Songlim Girls’ Middle School. He was imprisoned on the accusation that he had been a leading figure in anti-communist education. In the areas like Naju and Muan, the communists held a people’s court several days earlier. When prisoners were dragged out and presented before the people, the leftists and families holding grudges gathered and called out, ”Yes, yes. Kill that one, too!” They shouted out together influenced by the mass psychology. It was rare for one or two prisoners to survive out of several hundred. Those unfortunate people ... were falsely accused as a result of personal animosity or intrigue by their own neighbors. 2.3 Empirical Strategy To identify the effect of conflict on community cooperation and population trends, I employ two different specifications. First, to estimate the effect of war on the propen- sity for cooperation within community, I use the following cross-sectional empirical specification: cooperationi = α + β conf licti + Xi γ + θc + ei , (2.1) 75 where cooperationi is the measure of cooperation in township i, conf licti is the mea- sure of conflict severity in township i during the war, Xi is a vector of controls, and θc is the county fixed effects. To identify the effects of conflict, one needs to ensure that selection into conflicts are based on unobserved but fixed county-level characteristics. If this assumption holds, Equation 2.1 provides a consistent estimate of β.5 Second, to investigate whether population which experienced a short-term reduc- tion during the war converged back to pre-war trend, I divide the sample into two groups. First group is the treatment group that had more conflicts. The control group had relatively less conflicts. The precise definition of the measure of severity of conflicts will be discussed in section 4. The following specification is used: log(popit ) = α + β treatmenti + γt yeart + δt yeart · treatmenti + Xit ν + θc + it(2.2) where the outcome is log population at town i and year t. Xit is a vector of controls. θc is the county fixed effects. The coefficient of interest is δt which shows differences in the level of population between the treatment and control group. δt 6= 0 implies that the mean of the population of the treatment group in year t is different from the mean of the population of the control group in the same year. If δt 6= 0 after the end of war, it implies that the population level of the treatment group does not converge back to pre-war level at year t. 5 While it is possible that the NKPA chose hiding places based on town characteristics, such as overall degree of politically left-leaning tendencies. However, during the war, it might be difficult to acquire accurate information on political preference of residents. Moreover, the urgency of finding hiding place during the war may resulted in more random choice of hiding locations. Perhaps the most important determinants of the choices of hiding places would be ruggedness and altitudes which may prevented easy access from UN forces. I plan to include extensive geographic controls. 76 2.4 Data Primary data sources for the analysis are population censuses in various years and the New Village Comprehensive Survey (NVCS). 2.4.1 Population Censuses I use population censuses from year 1925 to 1990 to measure the changes in the number of civilian populations at the township level. Population census was collected every five years on average. The data contain various township characteristics including the number of the population, the number of illiterate population, the number of people with agriculture-related occupations, the number of Japanese population, the number of single and married people, and the number of people with different age groups, for example, between 0 and 14, and between 15-24. It would be ideal to have detailed breakdowns of population changes such as by age, gender, education level and migration destinations to assess detailed effects of war on population movement. However, the census data do not have more detailed population breakdowns. Therefore, I use overall population trend for the analysis. 2.4.2 Severity of Conflict The explanatory variable of interest is the severity of conflicts due to NKPA. To capture the severity of conflict within a township, the ideal data would be the the number of people’s courts held in a township. Unfortunately, these data are not available. Instead. I use census data and calculate the changes in civilian population right before the war (1949) and right after the war (1955) as a measure of severity of conflict, following Davis and Weinstein (2002). The severity measure, ∆pop49,55 is 77 defined as ∆pop49,55 = log(pop1955 ) − log(pop1949 ). (2.3) The changes in population reflect both war casualties and the reduction in population who migrated out to avoid conflict. Additionally, the population change also reflects other migration flows as well as births and other deaths. During the war, however the most prominent factor of population changes could be war-related migration and deaths. In my data, almost half of the townships experienced a reduction in pop- ulation during the war. Before the breakout of the war, however, there were few townships that experienced a reduction in population.6 Figure 2.4 shows the spatial variation in ∆pop49,55 . It shows that there are multiple pockets of regions where there is a large concentration of a relatively large reduction in the civilian population. While there is no centrally concentrated regions with a large decline in the population within the province, the existence of concentrations requires me to employ empirical strategy of including county fixed effects to eliminate across-county variations driving empirical results. 2.4.3 The NVCS Data To construct a measure of community cooperation, I use a government publication, the New Village Comprehensive Survey (NVCS) in 1972 which recorded the government assessment on the the production of public goods under the New Village Beautification Project, a rural intervention program in 1970. I digitized the data into an electronic format for analysis. 6 If the reduction in population during the war captures the severity of conflict, I expect that ∆pop49,55 is relatively uncorrelated with ∆popt , the population changes in periods before and after the war period. I calculate corr(∆popt−1 , ∆popt ) for every population census year data from 1925 to 1990. I find that corr(∆pop44,49 , ∆pop49,55 ) is not only approximately zero, but also it has lowest value among all corr(∆popt−1 , ∆popt ) from other census periods. 78 Under the New Village Beautification Project, each village was given the same amount of bags of cement bags by the government to produce village-level public goods. Since only cement was provided by the government, other resources such as land, labor, and equipments were voluntarily supplied by village members. Further, the usage of cement was collectively decided by village members.7 The government systematically evaluated each village the following year and clas- sified villages depending on the actual usages of cement. Some villages used cement for production of public goods such as improving village roads and building common laundry facilities. These villages received an A or B grade. Other villages used cement privately, such as kitchen floor improvements, and received a C grade. Using data on the government classification of cement projects, I construct the public use variable which takes the value one if a village received an A or B grade and zero otherwise. Since the unit of analysis in this paper is township, I compute the weighted average of village-level public use dummy at the township level. The weight is the number of households of each village. This measure could be a reasonable proxy for cooperation among village members for a couple of reasons. First, without any agreement among village members to use cement for public goods, it would be difficult to produce public goods. Second, even conditional on agreement to produce public goods, village members still have to voluntarily provide land and labor.8 2.4.4 The Family Clan Data I use Family Names in Chosun, a part of population census in 1930 by the Japanese Colonial Government to construct a lineage diversity measure at the village-level. 7 Village council members decided how to use cement then decided the usage of cement through votes from the head of each village household. 8 It was particularly difficult to donate private agricultural field for road improvement, such as widening village road because the average cultivated area was already quite low. Korea had a successful land reforms in the late 1940s. Each farmer could own land only up to 3 hectares. 79 The family clan data contain the number of households belong to each family clan in a village as long as the clan household share exceeds 10% of the total number of households. Using the household share of each clan in a village, I construct the family clan Herfindahl Index for measuring clan concentration and include it as a control variable in the analysis. As the unit of study is township, I take the weighted average of the Herfindahl Index of villages in the same township. The weight is the household share of each village in a township. The study region is South Jeolla province which experienced the most severe conflicts during the war because of the extended period of presence of NKPA. South Jeolla province has population of 1.7 million in 2010 and the size is roughly similar to the state of Connecticut in the U.S. The analysis is at the township level because the population census is the main data set which provides information at town level which is the lowest administrative unit. Urban regions in the province are excluded from the sample because the outcome variable, public use, is only available in rural townships. Table 2.1 presents summary statistics of townships. According to the 1949 popu- lation census, a township had population of roughly 10,000 on average. Agricultural occupation consisted of 85% of all occupations of township residents. The population was relatively immobile with 74% of population were born in the same township they resided when census was conducted. This is not surprising because farmers often in- herited land from ancestors, and they were reluctant to sell ancestors’ land and move elsewhere. The illiteracy rates were high, almost approaching 80%. The mean and the median of the main explanatory variable, ∆pop49,55 was approximately zero and the standard deviation was 0.08. 80 2.5 Empirical Results This section presents two sets of estimation results. I first show estimates of the effect of the conflicts during the war on cooperation within community. I then show whether the reduction of the civilian population during the war persisted. 2.5.1 Effects of War on Cooperation To test whether ideological conflicts had adverse effects on community cooperation, I examine the relationship between ∆pop49,55 and public use. Figure 2.5 plots ∆pop49,55 and public use. It suggests that there is a positive relationship between these two variables. The figure shows that a township that experienced more severe conflict (lower value of ∆pop49,55 ) was less likely for its pop- ulation to cooperate 20 years after the end of the war (lower value of public use). To confirm the patterns shown in the figure, I estimate the empirical specification in Equation 2.1. Table 2.2 shows results. I correct for heteroskedasticity in standard errors. When county fixed effects are used, I cluster standard errors at the county level. For the analysis, I use the population census data and the NVCS data. Column 1 does not include any control variable. The coefficient of ∆pop49,55 indicates that one percentage point decrease in the population during the war – more severe conflict – is associated with a decrease of the probability of using government-provide cement for public use by 29 percentage points. The estimate is highly statistically significant at 1 percent level. Column 2 adds the pre-war population level right before the war as a control. The coefficient changes only slightly. Column 3 adds pre-war township controls, and Column 4 include county fixed effects. The estimated coefficient is 0.16 and it is statistically significant at 10 percent level. While the magnitude of the coefficient decreased as more controls were added, the coefficient of ∆pop49,55 remains practically large given the standard deviation of the outcome variable is 0.1. These 81 results suggest that internal social division is associated with community cooperation and its consequences could be harmful and long-lasting.9 2.5.2 Alternative Explanations In this section, I evaluate alternative explanations on the relationship between the reduction of civilian population during the war and community cooperation. These include location specific amenities, migration to avoid conflict, and more generally, selection on unobserved variables. Location Specific Amenities It is possible that the regressor ∆pop49,55 predicts cooperation because of the existence of a third factor, such as time-invariant location specific amenities. While county fixed effects take differences in amenities at the county-level into account, there is still a possibility that township-level differences may still exist. These amenities could draw people into a township and also make town residents more likely to cooperate. This could drive spurious results. I carry out a placebo test that uses ∆pop in pre-war periods in South Jeolla province. If a presence of a third factor drives the results, I expect to see that placebo ∆pop in other periods will be also positively predict the outcome measure, that is β > 0. Table 2.3 shows that data do not support evidence that a time-invariant third fac- tor drives my results. I substitute ∆pop49,55 with ∆pop with different time periods in my preferred econometric specification, Column 4 of Table 2.2. Each row of Table 2.3 represents the estimated coefficients βˆ for each separate ∆pop. The results show that ∆pop in pre-war periods do not predict community cooperation. Except ∆pop49,55 , the coefficients of ∆pop of pre-war periods are statistically insignificant and the sign 9 The results in this paper contrast with existing literature on limited long-run impacts through destructions in physical capital (Davis and Weinstein, 2002; Brakman et al., 2004; Miguel and Roland, 2011). 82 of the coefficients are mostly the opposite of the results I find in the main results in Table 2.2. Migration to Avoid Conflict I also evaluate whether the relationship between ∆pop49,55 and public use shown in the previous section is due to migration of township residents to avoid conflict during the war, i.e., ∆pop49,55 captures migrations to avoid actual conflict instead of civilian casualties due to the war. Because the NKPA advanced to this region, it is reasonable to assume that the capitalists or the anti-communists were more likely to leave townships to avoid purges. However, these selective out-migration of people with right-leaning ideology would re- sult in less ideological diversity of the remaining township residents, i.e., a negative relationship between ∆pop49,55 and public use. This contrasts with the positive asso- ciation that I find in the data. Selection on Unobserved Variables While I employ county fixed effects to eliminate across-county differences driving results, the concern of potential biases still remains from the selection on unobserved variables.10 The effect of conflicts on outcome could be driven by selection because the magnitude of main coefficients of Table 2.2 does change as more controls are included. I use a statistical test suggest by Altonji et al. (2005) to check whether unob- served characteristics could dominate the main coefficient of ∆pop49,55 . Table 2.2 includes selection test statistics. The tests indicate that it seems unlikely that esti- mated coefficient is mostly due to selection. Conditional on county fixed effects being included, the explanatory power of unobserved characteristics should be at least five 10 Instrument variable strategy will alleviate this concern in more systematic way. Work on IV strategy is on progress. 83 times greater than the explanatory power of control variables used in my study to claim that the estimate is entirely due to selection. 2.5.3 Effects of War on Population Size Existing literature on the effect of war typically show that the population reduction during the war is temporary and it converges back to pre-war trend level quickly. For example, Davis and Weinstein (2002) document a rapid recovery of population in Japanese cities from bombing. Nagasaki took less than 15 years for the population recovery. Similar results of the convergence of population were shown in the case of bombing in rural regions in Vietnam (Miguel and Roland, 2011). Unlike the effects of bombing and destruction of physical capital, an increase in social divisions during the war may have lasting effects on the population and prevent the convergence of population. Residents may not wanted to live socially divided villages, or potential residents could be more reluctant to move into villages with uncooperative residents. To test the convergence of population after the war is ended, I compare the population trend of a group of townships which experienced the decline in civilian population during the war (treatment group) and another group without the population reduction (control group), which serves as a counterfactual population trend. That is, treatement group is consists of townships with ∆popi,4955 < 0. Control group has ∆popi,4955 ≥ 0. Figure 2.7 plots the difference in the average population size of the two groups by year. I calculate the differences by estimating Equation 2.2 using the census data. The estimated δt captures the differences in the population sizes between the two groups for year t. I plot the δt in the figure. Prior to the beginning of the war in 1950, there was no differences in the average population size. Between 1950 and 1960, there was a 15% drop in population in the treatment group. This initial drop is expected because the treatment and control group are defined based on whether a township faced a 84 reduction in population during the war. The reduction in the population, however, were sustained 40 years after the war up to 1990. The differences in the population size reached 20% by 1980 and the differences are statistically significant. Table 2.4 presents quantitative evidence that the township population size does not converge to pre-war population trend. The table shows the estimates of δt from various specifications. Column 1 has no control variables. Column 2 and 3 adds controls and county fixed effects. The results shows that there was little difference in the population size before the war. The estimates are mostly statistically insignificant. After 1955, however, the population gap persisted up 1990. 2.5.4 Comparison of the Effects of Conflict Through Social Division and Conventional Battles The main hypothesis of this paper is that heightened social divisions during the war are associated with community cooperation. On the other hand, if the civilian casualties are from conventional war battles, rather than through social division, then there could be a lack of such effect on cooperation. South Korea provides an unique context to compare the social division effect and the war battle effect within the same country. North Kyungsang province experienced military battles between NKPA and UN forces during the war. It contained Pusan Perimeter, a heavily fought battle lines (see Figure 2.8). Unlike Jeolla South, political purges were rare because North Korean soldiers could easily retreat back to north through the east coast and through mountains when UN forces successfully fought back (see Figure 2.2). Since purges were mostly absent in North Kyungsang but were frequent in South Jeolla, I expect that the relationship between the conflict and cooperation only holds in South Jeolla. This is because North Kyungsang did not expect much social divisions due to few political purges. 85 To test this intuition, I run the same regression, using Equation 2.1, in North Kyungsang. The results are consistent with the idea that the war effect through so- cial division lowers cooperation but not through civilian casualties due to battles: the reduction in the civilian population in North Kyungsang did not predict community cooperation. The estimates using North Kyunsang province show a negative associa- tion of ∆pop49,55 and public use, and estimates are not statistically significant. Figure 2.5 and 2.6 compare the bivariate relationship between ∆pop49,55 and public use in South Jeolla and North Kyungsang. While South Jeolla shows a relatively strong positive relationship, North Kyungsang shows weakly negative or no relationship. Additionally, through this exercise of the comparison of different provinces with various channels of war damages, I am able to reject an alternative hypothesis that there is a common omitted variable drives spurious correlation between ∆pop and cooperation across provinces. Otherwise, the sign of, and possibly the magnitude of, ˆ would also have been similar across provinces. the estimated coefficients, β, 2.6 Conclusion In this paper, I find evidence of persistent economic and social consequences of war. Specifically, I find a robust association between the severity of conflict a community experienced during the Korean War and cooperation within that community 20 years after the war ended. Further, the reduction in the population during the war did not converge back to the pre-war trend, even 40 years after the war ended. Evidence of the consequences of war through social division, however, is far from complete in this paper. I examine only two outcomes: community cooperation and the population trends. Assessing other important economic, political and social outcomes would provide further understandings of the effects of war. 86 In this paper, I examine the effects of political purges by communists during the Korean War. The general message—that political purges could have negative long- run consequences—could apply to other contexts such as political purges instigated by Mao in communist China, or by Stalin in the Soviet Union. The results in this paper also have the potential to provide policy guidance. Iden- tifying the major channel of war damages can help policymakers use recovery funds more efficiently. For regions with heavy reductions in physical capital, it would be advisable to boost capital accumulation and rebuilding efforts. However, for regions with damaged social cohesion, it might be necessary to develop policies for rebuilding trust and confidence among community members. Spending resources in rebuilding physical capital alone in these regions may not produce the intended results of revi- talizing the regional economy. As this paper has shown, people may not migrate to regions where residents do not trust each other, fail to cooperate, or exhibit lower levels of social capital. 87 Figures and Tables Table 2.1: Descriptive statistics obs mean SD min max Independent Variable public use 216 0.477 0.109 0.139 0.921 Explanatory Variable ∆pop4955 220 0.001 0.080 -0.381 0.213 Control Variables population 1949 227 11613 5783 4046 60251 illiteracy rate 227 0.799 0.047 0.639 0.896 % ag occupation 227 0.853 0.097 0.400 0.960 family clan Herfindahl Index 227 0.008 0.012 0.000 0.126 % Japanese pop 227 0.010 0.016 0.000 0.111 % native born 227 0.736 0.088 0.339 0.955 % single individuals 227 0.479 0.017 0.437 0.521 % individuals w/ age 0-14 227 0.403 0.014 0.360 0.438 % individuals w/ age 15-24 227 0.171 0.008 0.143 0.202 88 Table 2.2: OLS and FE estimates of the effects of the conflict on cooperation Dependent variable: public use (mean 0.48, s.d. 0.11) (1) (2) (3) (4) ∆pop49,55 0.29∗∗∗ 0.30*** 0.20** 0.16* (0.11) (0.11) (0.08) (0.09) log(population 1949) -0.03 -0.08*** -0.29** Illiteracy rate 0.01 0.24 % ag occupation -0.19 -0.09 Family clan Herfindahl Index -0.13 0.29 % Japanese pop 1.47 1.59 % native born 0.44*** 0.19 % single individuals -0.19 -0.71 % individuals w/ age 0-14 0.34 0.73 % individuals w/ age 15-24 1.40 0.75 County FE N N N Y Observations 216 216 216 216 R-squared 0.04 0.05 0.13 0.47 R Selection ratio: 1.27 (βOLS R , βFF E ), 1.99 (βOLS F , βOLS ), 5.06 (βFRE , βFF E ) Notes: Robust standard errors in parentheses. Standard errors in Column 4 are clustered at the county level. Data are from population censuses and the New Village Comprehensive Survey. The public use variable equals one if a village used cement from the government to produce village level public goods and zero if cement was used for private usage. As the analysis is at the township level, village level public use is averaged at the township level with the number of village household as the weight. ∆pop49,55 is changes in town popula- tion between 1949 and 1955. This measure is a proxy for the severity of conflict within a township. Selection ratios are based on (Altonji et al., 2005). * p < 0.1, ** p < 0.05, *** p < 0.01 89 Table 2.3: Falsification checks using ∆pop in pre- & post-war periods Dep. var.: public use Explanatory variable βˆ1 s.e. ∆pop 1925-1930 0.04 0.13 ∆pop 1930-1935 -0.01 0.09 ∆pop 1935-1944 -0.10 0.06 ∆pop 1944-1949 -0.11 0.11 ∆pop49,55 (original regressor) 0.16* 0.09 ∆pop 1955-1960 -0.23 0.14 ∆pop 1960-1966 -0.18 0.12 Notes: Robust standard errors in parentheses. Standard errors are clustered at the county level. Data are from population censuses and the New Village Comprehensive Survey. I run regressions with the specification identical to Table 2 Column 4 except the regressor. Each row shows the changes in population in various years as regressors. For example, ∆pop 1925-1930 is a measure of population changes between year 1925 and 1930. * p < 0.1, ** p < 0.05, *** p < 0.01 90 Table 2.4: Effect of war on population trends Dep. var.: log(population) (1) (2) (3) Before the War treatment*year1930 -0.02 -0.02 -0.02 treatment*year1935 -0.05 -0.05 -0.05∗∗∗ treatment*year1944 -0.02 -0.02 -0.02 treatment*year1949 -0.03 -0.03 -0.03 After the War treatment*year1955 -0.15∗∗ -0.14∗∗ -0.14∗∗∗ treatment*year1960 -0.14∗∗ -0.13∗∗ -0.13∗∗∗ treatment*year1966 -0.13∗ -0.13∗∗ -0.12∗∗∗ treatment*year1970 -0.15∗∗ -0.15∗∗ -0.14∗∗∗ treatment*year1975 -0.14∗∗ -0.14∗∗ -0.14∗∗∗ treatment*year1980 -0.19∗∗ -0.19∗∗∗ -0.19∗∗∗ treatment*year1985 -0.19∗∗ -0.21∗∗∗ -0.21∗∗∗ treatment*year1990 -0.16∗ -0.18∗∗ -0.17∗∗∗ Controls N Y Y County FE N N Y Observations 2808 2808 2808 R-squared 0.21 0.48 0.52 Notes: Robust standard errors in parentheses. Standard errors in Column 3 are clustered at the county level. Data are from population censuses and the New Village Comprehen- sive Survey. I construct a panel data set in which a township has multiple observations for each years. The treatment dummy equals one if a township experienced negative growth between just before and just after the war (∆pop49,55 ) and zero otherwise. The treatment group experienced relatively more conflict compared to the control group. The variables year19XX indicates a dummy for year 19XX. The table shows interaction terms between treatment dummy and year dummies to indicate the average population differences in two groups. Controls variables are identical to Table 2.2. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01 91 Figure 2.1: Location of South Jeolla South Jeolla province is located in southwestern part of South Korea. The province is highlighted in gray color in the map. The map shows township boundaries. 92 Figure 2.2: The Korean War and UN Forces Note: UN forces landed near Seoul (west) and also counter attacked the NKPA from the southeast toward northwestern direction (direction of arrows). As a result, some of the NKPA units were trapped in the southwest area. Source: Department of History, US Military Academy 93 Figure 2.3: Major battle sites Note: The numbers in the map show the location of major battles between the NKPA and UN Forces during the Korean War. South Jeolla province (southwestern region) escaped major battles. Source: The 8th Army Staff Historian’s Office (1972) 94 Figure 2.4: Severity of conflicts The figure shows the spatial variation of ∆pop49,55 , changes in township population right before and the right after the Korean War. Red colors indicate townships that experienced reductions in population (more severe conflict). Green colors shows townships with a positive increase in popula- tion. Yellow colors show minimal changes in population. A darker red implies a more reduction in population. A darker green implies a more increase in population. 95 Figure 2.5: Effect of conflict on public use in South Jeolla 1 .8 public_use .6 .4 .2 -.4 -.2 0 .2 Δpop4955 The figure shows bivariate relationship between ∆pop1949,1955 and public use in South Jeolla province. Figure 2.6: Effect of conflict on public use in North Kyungsang 1 .8 public_use .6 .4 .2 -.4 -.2 0 .2 .4 .6 Δpop4955 β_hat = -0.036, s.e. = 0.146, R2 = 0.0007 The figure shows bivariate relationship between ∆pop1949,1955 and public use in North Kyungsang province. 96 Figure 2.7: Effect of conflict on population trend 0 coeff. of interaction terms -.2 -.3 -.1 1930 1940 1950 1960 1970 1980 1990 year The graph shows that the reduction in the civilian population caused by the war did not con- verge back to pre-war trend after 40 years. The solid line shows the differences in population between treated group and control group by plotting the estimated coefficients of the interaction term, treatment·year dummy. The dotted lines indicate the 95% confidence interval. Treated group is defined as townships that experienced a negative population growth during the war. Townships in control group did not experience population reduction. The differences in population between treatment and control group before the war (1950-1953) were little and were mostly statistically insignificant. On the other hand, the reduction in population during the war persisted 40 years after the war. 97 Figure 2.8: The Pusan Perimeter Note: The Pusan Perimeter shown in the map is the battle lines between the NKPA and UN forces. The highlighted region in the southeastern part of the map indicates the only part of South Korea which was not occupied by the NKPA. Source: Stueck (2002) 98 Figure 2.9: Map of North Kyungsang province North Kyunsang province is located in southeastern part of South Korea. The province is highlighted in red color in the map. The map shows township boundaries. 99 Figure 2.10: Effect of bombing on population trends Note: Davis and Weinstein (2002) provide evidence of a rapid recovery of population from bombing. Nagasaki took less than 15 years for its population trend to reach the pre-war trend. Hiroshima took 30 years. Source: Davis and Weinstein (2002) 100 Chapter 3 The Effects of High Speed Trains on Local Economies: Evidence from the Korea Train Express 3.1 Introduction Accurate assessments of economic costs and benefits of large scale infrastructure in- vestments, such as dams, airports, and express highway systems, are important for policymakers. Bad cost-benefit analysis could result in significant amounts of re- sources being directed to economically inefficient investments with long-term bud- getary consequences. Investments in high speed rail (HSR) systems have attracted attention worldwide as a way to improve transportation infrastructure. As of 2006, there were more than 20,000 kilometers of high speed rails in operation in 15 countries, such as France, Germany, Spain, China and Japan (Campos and De Rus, 2009). Despite significant investment in HSR systems, there has been a lack of systematic evaluations on their economic impacts. In this paper, we examine economic benefits 101 and costs of HSR systems on local economic activity.1 Specifically, we investigate how the Korean HSR system, the Korea Train Express (KTX), generated economic impacts on local economies. The KTX began operating in 2004 between Seoul and Busan, with a total rail length of more than 400 kilometers connecting these two biggest cities and other major cities in South Korea. The main innovation of this paper is the use of a 20-year panel data set to mea- sure local economic activity on a small spatial scale. We constructed a measure of township-level economic activity annually from 1994 to 2013 using nighttime light data from satellite observations as a proxy.2 Since the KTX began operating in 2004, the data set enabled us to follow economic activities of townships about 10 years before and after its launch. Furthermore, the panel nature of the data set enabled us to control for time-invariant and region-invariant unobserved characteristics. This is particularly important since our dependent variable is local economic activity, which may have many confounding factors. Moreover, there is limited concern for sam- ple selection because our panel data set is perfectly balanced. Often, administrative boundaries change over time, complicating the construction of balanced panel data sets on a small spatial scale. We overcame this difficulty by applying fixed adminis- trative boundaries from 2016 to all nighttime light satellite maps across years. 1 Existing literature provides mixed evidence on the economic impact of HSR systems. Some researchers have suggested that improved transportation networks foster economic activity. Sands (1993), for example, showed that the introduction of the HRS system in Japan was associated with higher population and employment growth in connected cities. Banister and Berechman (2003) provided similar evidence from the French HSR system. Kwon (2014) documented more growth in the retail sector and increased land prices in cities connected to the KTX. On the other hand, other researchers have demonstrated negative effects on local economies. Economic activity in locales with relatively poor economic conditions could be shifted to larger cities once they are connected via HSR networks (Whitelegg and Holzapfel, 1993; Thompson, 1995; Vickerman, 1997). For example, cancer patients in smaller cities began switching to hospitals located in metropolitan areas once their cities were connected to the KTX (Kim et al., 2010). Additionally, Chung and Lee (2011) showed that, 5 years after the launch of the KTX, there was large population shift toward Seoul and Chungcheong provinces from southern provinces where the KTX was in operation. Ultimately, it is an empirical question as to whether positive or negative effects dominate. 2 Henderson et al. (2012) showed that nighttime activity measured by nighttime light sensors from satellites can be used as a reliable proxy for GDP. 102 In this paper, we provide robust empirical evidence that the KTX has had a positive impact on local economic activity. Using the difference in differences (DD) approach, we show that rural townships adjacent to KTX stations grew about 10% more after the introduction of the KTX. To use the DD approach, we divided the sample into two groups: townships located near KTX stations and townships located farther away from KTX stations. We define a township as near a KTX station if the distance to the nearest KTX station is less than 34 kilometers (the median distance in the data), and far from a KTX station if the distance to the nearest KTX station is more than 34 kilometers. Prior to the introduction of the KTX in 2004, these two groups of townships showed parallel trends in economic activity, providing the crucial justification for using the DD approach as the research design. We then generalized key variables used in our DD approach in two ways. First, instead of dividing townships into two groups and using a group indicator variable, we used the distance in kilometers between each township and the nearest KTX station as the main regressor. Second, rather than using an indicator variable for the period post-2004 (i.e., the period of KTX operations), we included year dummies for each year and interacted these with the main regressor described above. Consistent with results in the DD estimation above, we provide robust evidence that the relationship between the distance to the nearest KTX station and economic activity strengthened progressively after 2004, but not before. This paper contributes to the growing economics literature on the empirical analy- sis of the relationship between economic activity and various modes of transportation, such as highways (Baum-Snow, 2007; Michaels et al., 2012); railroads (Fogel, 1994; Donaldson, 2010); and subways (Billings, 2011; Gonzalez-Navarro and Turner, 2016). Two economics papers are closely related to ours: one is by Ahlfeldt and Feddersen (2015), who provided evidence of a positive impact on GDP in municipalities near high speed rail stations in a number of states in Germany; the other is by Zheng and 103 Kahn (2013), who showed that land prices rose in cities adjacent to high speed train stations in China. Our paper is unique in that it is based on data for the universe of rural townships that cover most of a country. This paper proceeds as follows. In Section 2, we provide institutional details on the KTX. In Section 3, we describe the data and empirical strategies we employed. In Section 4, we discuss the main empirical results. We provide concluding remarks in Section 5. 3.2 The Korea Train Express The KTX began operating in April 2004 between Seoul and Busan over a 410- kilometer high speed rail line. The trains were capable of reaching 300 kilometers per hour. The introduction of the KTX cut the travel time between Seoul and Busan in half. In 2004, it took 2 hours and 40 minutes to travel between the two cities via the KTX, while it took 5 hours and 20 minutes via an express bus, and 4 hours and 30 minutes via the existing fast train (Suh et al., 2005). Figure 3.1 shows KTX stations and major HSR lines. The major Seoul-Busan line connects Seoul in the northwest and Busan in the southeast. Additionally, the Seoul-Gwangju line branches out from the Seoul-Busan line near Daejeon station. As shown on the map, Gangwon province does not have any HSR service in operation. In the empirical analysis, we exploit this fact and use rural townships in Gangwon province to perform a falsification test of the effect of the KTX. The Korean HSR service successfully competed with other conventional modes of transportation. A few years after the KTX began operating, the transit share for automobiles decreased by 15 percentage points from 70% to 55% along the Seoul- Daejeon route, while the transit share for the KTX increased by 28 percentage points. Along the Seoul-Daegu route, the transit share for air travel was reduced to 0%, 104 forcing closure of the air route, while the transit share for the KTX increased to 59% (Kim and Lee, 2011). 3.3 Data and Empirical Strategy 3.3.1 Data As a measure of local economic activity, we used nighttime light satellite data from the U.S. National Oceanic and Atmospheric Administration (NOAA). The NOAA data consist of annual maps of nighttime lights of the world, collected by visible and infrared light sensors of satellites owned by the U.S. Department of Defense. We used the Version 4 Defense Meteorological Satellite Program Optical Linescan System Nighttime Lights Time Series Data from 1994 to 2013, which are publicly available on the NOAA website.3 A nighttime light satellite map is composed of pixels, each of which contains information on the intensity of nighttime light with integer values ranging from 0 to 63. A higher number indicates more intense light. The area of each pixel is roughly 1 square kilometer. Figure 3.2 shows nighttime light maps of South Korea in 1994 and 2013. Figure 3.9 depicts the increase in the amount of light from 1994 to 2004 and from 2004 to 2013, along with all of the KTX stations in South Korea. To create township-level panel data using nighttime light data, we used spatial data on township boundaries in 2016. These data are from the National Spatial Information Clearinghouse.4 A major advantage of empirical analysis at the township level is the potential for large variations in treatment status. The 20-year panel data 3 The raw data are available at http://ngdc.noaa.gov/eog/dmsp.html. 4 The data can be downloaded at https://www.nsic.go.kr/ndsi/. 105 set we use in this study has more than 25,000 observations at the township level. Figure 3.4 depicts township boundaries of South Korea in 2016.5 We exclusively used rural townships in our analysis and excluded urban districts. The light sensors of the satellites are saturated by the strong lights emitted from urban areas. The majority of dongs, the township-equivalent administrative unit in urban areas, were already top-coded in 1994. Moreover, dongs are often too small to cover even one pixel on the nighttime light maps, generating missing values of light intensity. To identify treatment status, we used the distance of a township from the nearest KTX station. To compute the distance measure, we first geocoded the locations of KTX stations using their GPS coordinates. Then, we recorded the center point of each township by locating the centroid of each township boundary. In our context, a centroid is the center of mass of the two-dimensional space within the boundary of each township. Our definition of the distance of a township from the nearest KTX station is the computed Euclidean distance from the GPS coordinates of the KTX station to the location of the centroid of the township, expressed in kilometers. 3.3.2 Empirical Strategy To identify the effects of the introduction of the KTX, we compared trends of night- time light intensity between two groups of townships—those located near KTX sta- tions and those located far from KTX stations—10 years before and after the launch of the KTX. To implement this strategy, we used the DD approach, exploiting the timing of the introduction of the KTX in 2004. A panel data structure enabled us to control for unobserved variables that are time invariant and region invariant and to compare economic trends. 5 Figure 3.4 omits Jeju province and other islands that are far away from any KTX station and hence excluded from our analysis. 106 We used the following specification for the first part of our empirical analysis: LIGHTi,t = α0 + α1 T REATi + α2 P OSTt + α3 T REATi · P OSTt (3.1) 0 +Xi,t Ψ + τi + i,t , where T REATi is the treatment status dummy, indicating whether the township i is located within 34 kilometers of the nearest KTX station or not, LIGHTi,t is the average light intensity of township i in year t, P OSTt is an indicator of the period after the launch of the KTX in 2004, Xi,t is a vector of control variables, and τi is county fixed effects. The coefficient of interest is α3 which measures the impact of the KTX on local economies, based on changes in light intensity. If the positive impact of improvement in the transportation infrastructure dominates, then we expect α3 > 0. If negative effects dominate, then α3 < 0. If these two opposing effects cancel each other, or there are no effects, then α3 = 0. While some KTX lines, such as the Seoul-Gwangju line, began service in 2010, we define the treatment period as 2004 and later. Except Gangwon province, which was completely cut off from the influence of the KTX, other provinces received partial treatment because some parts of travel itineraries from those regions include the KTX line. For example, one could use the KTX to travel from Seoul to Daejeon and then use the regular train service to travel from Daejeon to Gwangju. This definition of the treatment would result in measurement errors. However, the direction of the bias is toward zero because townships were considered treated in 2004, whereas the actual treatment did not occur until 2010, resulting in dilution of treatment effects. Therefore, the estimates serve as a lower bound of the true effects. We also used a more generalized version of the specification above by replacing the treatment group dummy with the distance to the nearest KTX station and by replacing a post-policy year dummy with year dummies for each year. We used the 107 following specification: P P LIGHTi,t = δ0 + δ1 DISTi + t γt Y EARt + t λt DISTi · Y EARt (3.2) 0 +Xi,t Θ + σi + ξi,t , where DISTi is the distance of township i to the nearest KTX station, LIGHTi,t is the average light intensity of township i in year t, Y EARt is a year dummy indicating whether the observation year is t or not, Xi,t is a vector of control variables, and σi is county fixed effects. The coefficients of interest are λt which show the differential slope of the KTX distance and light intensities compared to the slope in 1994, since we omit the year dummy for that year. If there are economic impacts due to the launch of the KTX, we expect that λt = 0 when t < 2004 and λt 6= 0 when t ≥ 2004. 3.4 Empirical Results 3.4.1 Impact of KTX on Local Economic Activity Using the DD approach following Equation 3.1, we found empirical evidence of a positive impact of the KTX on local economic activity. Figure 3.5 shows trends of the average light intensities over the 20-year period for both the treatment and control groups. Prior to 2004, the trends are parallel, which satisfies the crucial identification assumption for using the DD approach. After 2004, the gap between the two groups becomes wider. Figure 3.5 also shows the gap of the average light intensities of the treatment and control groups. Before 2004, the gap between the two groups is relatively stable. After 2004, the gap progressively increases, suggesting that the effect of the KTX could be a trend shift, not just a shift in the level of local economic activity. 108 Table 3.1 confirms the results from the figures. In Column 1, the estimated impact of the KTX is 2.3 and the coefficient is statistically significant at the .01 level. Using trends from townships located far from KTX stations as counterfactual trends, the light intensity of townships near KTX stations would have been about 21.7 in the absence of the KTX. Hence, the estimated increase in light intensity of 2.3 implies that there was about a 10% increase in economic activity due to the introduction of the KTX. In Column 2, we added control variables which do not change the estimated value of the coefficient. In Column 3, we added county fixed effects to control for unobserved characteristics that are time-invariant at the county level. In Column 4, we included county fixed effects as well as control variables. In all of these cases, the estimates are stable and highly statistically significant. We then generalized key variables used in the DD estimation from the previous subsection in two ways. First, we used the distance from the nearest KTX station as the main regressor, replacing the treatment group dummy. Second, we replaced the dummy variable for the post-KTX launch year with year dummies (Equation 3.2). Based on the results, we are able to draw a similar conclusion that there was a positive impact of the KTX on local economic activity, as shown in the previous subsection. Figure 3.10 plots estimated λt and the 95% confidence intervals. λt are the coefficients of the interaction terms of year dummies and the KTX distance variable. We observe that the slope of the KTX distance against light intensity rarely changed before the introduction of the KTX in 2004. The coefficients are mostly near zero and the confidence interval contains zero as well. However, the slope becomes progressively steeper. As λt have negative values and the slope in 1994 is decreasing, we can infer that the slope becomes steeper post 2004. Moreover, the confidence interval does not include zero post 2004, which suggests that the changes in the slope are statistically significant. 109 Table 3.2 provides the values plotted in Figure 3.10. We used two specifications. The first two columns show the estimates of λt and standard errors from a regression without covariates and fixed effects. The next two columns include both covariates and county fixed effects. In both specifications, the changes in the slope of KTX distance relative to the slope in 1994 are mostly around zero. After 2004, the mag- nitude of the changes in slope increases progressively and the estimates are highly statistically significant. Since the estimates of the slope in 1994 are about -0.17, the changes in the slope of -0.9 after 2010 are substantial. 3.4.2 Falsification Checks In this subsection, we present falsification checks. If the increase in economic activity is due to the introduction of the KTX, rail stations that are not connected to the KTX should have a negligible impact on the trend of local economic activity after 2004, ceteris paribus. We used data from Gangwon province and Jeju Island to perform the falsification checks. Gangwon did not receive KTX connections during the study period, and Jeju Island does not have railway connections to the mainland. In Figure 3.6, we plot estimated λt using the same regression specification (2). We then used the distances to the rail stations in Wonju, Chooncheon and Gangreung to define each townships treatment status. There was little differential change in economic activity in Gangwon Province over time. Moreover, these estimates are mostly not statistically significant (Table 3.3). Another possible explanation is that connections to other major cities, such as Seoul, became more important after 2004 and economic activity shifted toward loca- tions with better transportation connectivity in general. While the falsification checks above provide little support for this argument, we also checked whether townships close to airports had more economic activity after 2004. We found little evidence to support this explanation. The estimated λt for townships on Jeju Island shows 110 few changes before or after 2004 (Figure 3.6). The last column of Table 3.3 provides estimates showing the lack of statistical significance. 3.4.3 Case Study of Daegu and Gimcheon City “After being connected to the KTX, the speed of development in Gim- cheon is as fast as the speed of a KTX train.” – Mayor of Gimcheon Complementing our main empirical analysis, we end this section by presenting a case study that concretely shows how the introduction of the KTX has contributed to a local economy. Daegu is a large metropolitan city that reached its economic peak in the 1970s and 1980s thanks to its large textile industry. Once the Korean economy transitioned to electronics and other heavy industries, the Daegu economy lost momentum. After its existing Dong Daegu rail station was connected to the KTX in 2004, the travel time to Seoul was cut in half, from more than 3 hours to 1.5 hours. The demand for housing has since increased sharply. In 2014, the average price for a condominium was the third-highest in the country, after Seoul and the Kyunggi region. Condominium prices even surpassed prices in Incheon, a large city next to Seoul without a KTX connection. Daegu’s price had lagged Incheon’s since 2000. The 7.7% growth rate of condominium prices in 2014 was the highest among all 17 provinces and cities (Ha, 2015). The impact of the KTX on Gimcheon, a small city northeast of Daegu, was also strong.6 Since the KTX station became operational, Gimcheon has experienced an 6 In 2003, the launch of the Gimcheon KTX station was publicly announced. Construction began in 2008 with total budget of 148 billion Korean Won. The KTX station was built in a rural township, Nam Myeon. The KTX line was connected in 2010 with 40 KTX train stops in the station daily in both directions to and from Seoul. After being connected, it took 90 minutes to travel to Seoul and 70 minutes to travel to Busan (Park, 2013). 111 economic boom. The city has attracted 12 public institutions and 5,000 workers to the region. By 2014, seven public institutions had relocated there (Ha, 2015). Moreover, the demand for condominiums has skyrocketed in Nongso Myeon and Nam Meyong, townships that are adjacent to the station. A new industrial complex was built near Oemo Myeon, the first new complex in the Gimcheon region since 1990s. Officials in Gimcheon city expected a population inflow of 26,000 and 10,500 new jobs by the end of 2015 (Park, 2013). Figure 3.8, which is based on pixel-level nighttime light data, graphically shows how economic activity in the Daegu and Gimcheon regions grew rapidly after 2004. Most of the surrounding area had negative growth between 1993 and 2004 (Panel a). Specifically, the regions between Gimcheon and Daegu had negative growth. Once the KTX was introduced, these regions between two the KTX stations grew rapidly (Panel b). 3.5 Conclusion Transportation improvements could be an important driver of economic growth. In this paper, we estimated the effect of a new HSR system on local economic activity using a novel panel data set on a small spatial scale. Taken together, our DD estimates imply that the introduction of the KTX led to a 10% increase in local economic activity 10 years after the launch of the HSR service in rural townships adjacent to KTX stations. This paper has some limitations. Our results do not fully explain the mechanism through which HSR systems may cause increases in economic activity. An increase in economic activity could be driven by either a shift of activity from elsewhere or the generation of new activity at a given location. We did not analyze the population trend, another important outcome that reflects economic activity. We were not able 112 to analyze sectoral changes due to the launch of the KTX. Was the growth driven by new construction in residential areas or further development of the service sector? These questions are left for future research. 113 Figures and Tables Figure 3.1: The KTX network Source: https://en.wikipedia.org/wiki/Korea_Train_Express 114 Figure 3.2: Comparison of light intensities, 1994 and 2013 (a) 1994 (b) 2013 Source: U.S. National Oceanic and Atmospheric Administration night time light data, 1994, 2013 115 Figure 3.3: Pixel-level comparison of light intensity changes (a) 1994-2004 (b) 2004-2013 Source: U.S. National Oceanic and Atmospheric Administration night time light data, 1994, 2004, 2013 116 Figure 3.4: Townships boundaries in 2016 Source: Korea National Spatial Information Clearinghouse, 2016 117 Figure 3.5: Differential light trends 30 14 13 20 Light Intensity Differences 12 10 Light Intensity 11 0 10 -10 9 -20 8 -30 7 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 Year Near KTX Away from KTX Near - Away Source: U.S. National Oceanic and Atmospheric Administration night time light data, 1994-2013 Note: The graph plots average light intensity of townships near and far way from a KTX station. The average light intensity for each township is computed using the 2016 township boundaries. We use the median distance from the nearest KTX station, 34 kilometers, to divide these two groups. The light intensity difference line is constructed by the vertical gap between light intensity of townships near and far away from KTX stations. Urban districts, Gangwon Province, and islands are excluded in the sample. 118 Figure 3.6: Differential slopes of the KTX on local economies -.1 -.08 -.06 -.04 -.02 0 .02 .04 .2 .4 Diff. in Slope for Rail Stations Diff. in Slope for Jeju Airport -.8 -.6 -.4 -.2 0 -1 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 Year KTX (main regressor) Chooncheon Seo Wonju Gangreung Jeju Airport Source: U.S. National Oceanic and Atmospheric Administration night time light data, 1994-2013 Note: This figure shows estimates of λt in Equation 3.2 in connected lines using various regressors: the distance from: the nearest KTX stations (main regressor); Chooncheon rail station; Seo Wonju rail station; Gangreung rail station; and Jeju Airport. The specification include covariate and county fixed effects. Each line tracks the yearly changes in the estimated slope of the distance of a rural township from the different regressors described above on the average light density of rural townships. Since the year dummy for 1994 is excluded, the values in the graph represent the changes in the slope relative to the slope in 1994. The sample includes all rural townships within 100km from each distance measure. For Jeju airport results, the sample includes rural townships in the Jeju Island only. 119 Table 3.1: Diff-in-Diffs estimates of the impact of the KTX on light intensity (1) (2) (3) (4) treatment × post 2.3*** 2.3*** 2.3*** 2.3*** (0.28) (0.27) (0.55) (0.55) Observations 25,300 25,300 25,300 25,300 R-squared 0.15 0.21 0.59 0.63 County FE N N Y Y Covariate N Y N Y Source: U.S. National Oceanic and Atmospheric Administration night time light data, 1994-2013 Note: Standard errors in parentesis. Standard errors are corrected for heteroskedasticity. They are clustered at county-level when county fixed effects are used. Treatment dummy equals one if ob- servations with distance from the nearest KTX station below median value of the distance from the sample (34km) and zero otherwise. The covariate includes the size of the township. Data ex- clude urban districts and rural townships from islands and from Gangwon Province. *** p<0.01, ** p<0.05, * p<0.1. 120 Table 3.2: Estimates of the slope changes of distance to the KTX Year (1) Coeff. (1) S.E. (2) Coeff. (2) S.E. 1995 0.00 0.02 0.00 0.01 1996 -0.01 0.02 -0.01* 0.01 1997 0.01 0.01 0.01 0.01 1998 0.01 0.01 0.01** 0.01 1999 0.01 0.02 0.01 0.01 2000 -0.02 0.02 -0.02*** 0.01 2001 -0.04** 0.02 -0.04*** 0.01 2002 -0.04** 0.02 -0.04*** 0.01 2003 -0.01 0.01 -0.01 0.01 2004 -0.03* 0.02 -0.03** 0.01 2005 -0.01 0.02 -0.01 0.01 2006 -0.03* 0.02 -0.03** 0.01 2007 -0.05*** 0.02 -0.05*** 0.01 2008 -0.05*** 0.02 -0.05*** 0.01 2009 -0.06*** 0.02 -0.06*** 0.01 2010 -0.09*** 0.02 -0.09*** 0.02 2011 -0.09*** 0.02 -0.09*** 0.01 2012 -0.07*** 0.02 -0.07*** 0.02 2013 -0.08*** 0.02 -0.08*** 0.02 County FE N Y Covariate N Y Observations 25,300 25,300 R-squared 0.21 0.70 Source: U.S. National Oceanic and Atmospheric Administration night time light data, 1994-2013 Note: Standard errors in parentesis. Standard errors are corrected for heteroskedasticity. They are clustered at county-level when county fixed effects are used. The table shows estimates of λt in Equation 3.2 along with the standard errors. Since the year dummy for 1994 is excluded, the esti- mates represent the changes in the slope relative to the slope in 1994. The covariate includes the size of the township. Data exclude urban districts and rural townships from islands and from Gangwon Province. *** p<0.01, ** p<0.05, * p<0.1. 121 Table 3.3: Falsification checks KTX Chooncheon Seo Wonju Gangreung Jeju Airport DIST -0.20*** -0.04 0.05 -0.02 -0.32 1995 0.00 0.00 0.00 0.01 0.13* 1996 -0.01 0.02 -0.01 0.01 0.06 1997 0.01 -0.02 -0.01 0.01 -0.01 1998 0.01 -0.02 -0.01 0.00 0.04 1999 0.01 -0.01 -0.01 -0.01 0.14 2000 -0.02 -0.01 0.00 -0.01 0.02 2001 -0.04** 0.02 -0.01 0.00 0.01 2002 -0.04** 0.01 0.00 -0.01 0.05 2003 -0.01 -0.01 0.00 -0.02 0.05 2004 -0.03* 0.01 0.02 -0.03* 0.01 2005 -0.01 0.00 0.02 -0.04*** 0.03 2006 -0.03* 0.00 0.03 -0.03* 0.08 2007 -0.05*** 0.01 0.03 -0.03* -0.02 2008 -0.05*** 0.01 0.04 -0.04* -0.04 2009 -0.06*** 0.00 0.03 -0.01 -0.08 2010 -0.09*** 0.02 0.02 -0.02 -0.03** 2011 -0.09*** 0.00 0.03 -0.02 -0.06 2012 -0.07*** 0.02 0.03 -0.01 0.02 2013 -0.08*** 0.02 0.03 -0.01 0.06* County FE Y Y Y Y Y Covariate Y Y Y Y Y Observations 25,300 4,400 6,740 2,260 240 R-squared 0.70 0.77 0.81 0.52 0.66 Source: U.S. National Oceanic and Atmospheric Administration night time light data, 1994-2013 Note: Standard errors are corrected for heteroskedasticity and are clustered at county-level. The table shows estimates of λt in Equation 3.2. Since the year dummy for 1994 is excluded, the esti- mates represent the changes in the slope relative to the slope in 1994. The covariate includes the size of the township. Data exclude urban districts and within 100 kilometers from the nearest KTX station, except Jeju Island which include all rural townships within the Jeju Island. *** p<0.01, ** p<0.05, * p<0.1. 122 Appendix Figure 3.7: Light intensity, Seoul Metropolitan Area, 1994 Source: U.S. National Oceanic and Atmospheric Administration night time light data, 1994 123 Figure 3.8: Comparison of light intensities, 1994 and 2013 (a) 1994 (b) 2013 Source: U.S. National Oceanic and Atmospheric Administration night time light data, 1994, 2013 124 Figure 3.9: Light growth (a) 1994-2004 (b) 2004-2013 Source: U.S. National Oceanic and Atmospheric Administration night time light data, 1994-2013 Note: The graph plots the growth of light intensity from the 1994-2004 period (a) and the 2004-2013 period (b). The average light intensity in year 1994, 2004, and 2013 is computed at the county- level, using 2016 county boundaries. The growth of light is then computed by log differences in corresponding years. Darker regions have higher growth rates. 125 Figure 3.10: Difference in the slopes of KTX distance .05 Diff. in Slope of Distance from KTX Station -.15 -.1 -.05 0 1994 1999 2004 2009 2014 Year 95% CI 95% CI Coefficients Source: U.S. National Oceanic and Atmospheric Administration night time light data, 1994-2013 Note: This figure shows estimates of λt in Equation 3.2 along with 95% confidence interval of the estimates. The specification include covariate and county fixed effects. The solid line tracks the yearly changes in the estimated slope of the distance of a rural township from the nearest KTX station on the average light density. Since the year dummy for 1994 is excluded, the values in the graph represent the changes in the slope relative to the slope in 1994. Urban districts, Gangwon Province, and islands are excluded in the sample. 126 Bibliography Acemoglu, Daron, Tarek A Hassan, and James A Robinson (2011) “Social structure and development: A legacy of the Holocaust in Russia,” Quarterly Journal of Economics, Vol. 126, pp. 895–946. Acemoglu, Daron, Tristan Reed, and James A Robinson (2013) “Chiefs: Elite control of civil society and economic development in Sierra Leone,” Working Paper 18691, National Bureau of Economic Research. Ahlfeldt, Gabriel M and Arne Feddersen (2015) “From periphery to core: Measur- ing agglomeration effects using high-speed rail,” working paper, London School of Economics. Alesina, Alberto, Reza Baqir, and William Easterly (1999) “Public goods and ethnic divisions,” Quarterly Journal of Economics, Vol. 114, pp. 1243–1284. Alesina, Alberto, Arnaud Devleeschauwer, William Easterly, Sergio Kurlat, and Ro- main Wacziarg (2003) “Fractionalization,” Journal of Economic Growth, Vol. 8, pp. 155–194. Alesina, Alberto and Eliana La Ferrara (2000) “Participation in heterogeneous com- munities,” Quarterly Journal of Economics, Vol. 115, pp. 847–904. Altonji, Joseph G, Todd E Elder, and Christopher R Taber (2005) “Selection on observed and unobserved variables: Assessing the effectiveness of Catholic schools,” Journal of Political Economy, Vol. 113, p. 151. Arnold, Guy (2001) Guide to African political and economic development: Routledge. Banerjee, Abhijit, Lakshmi Iyer, and Rohini Somanathan (2005) “History, social di- visions, and public goods in rural India,” Journal of the European Economic Asso- ciation, Vol. 3, pp. 639–647. Bang, Ki-joong (2004) Fascist governing policy of Japanese colonial power and peasant lives: Hye-an. Banister, David and Joseph Berechman (2003) Transport investment and economic development: Routledge. 127 Barthorp, Michael and Gerry Embleton (1982) The Jacobite Rebellions 1689-1745: Osprey Publishing. Baum-Snow, Nathaniel (2007) “Did highways cause suburbanization?” Quarterly Journal of Economics, pp. 775–805. Becchetti, Leonardo, Pierluigi Conzo, and Alessandro Romeo (2014) “Violence, trust, and trustworthiness: Evidence from a Nairobi slum,” Oxford Economic Papers, Vol. 66, pp. 283–305. Bellows, John and Edward Miguel (2006) “War and institutions: New evidence from Sierra Leone,” American Economic Review, Vol. 96, pp. 394–399. (2009) “War and local collective action in Sierra Leone,” Journal of Public Economics, Vol. 93, pp. 1144–1157. Besley, Timothy, Stephen Coate, and Glenn Loury (1993) “The economics of rotating savings and credit associations,” American Economic Review, Vol. 83, pp. 792–810. Besley, Timothy and Marta Reynal-Querol (2014) “The legacy of historical conflict: Evidence from Africa,” American Political Science Review, Vol. 108, pp. 319–336. Billings, Stephen B (2011) “Estimating the value of a new transit option,” Regional Science and Urban Economics, Vol. 41, pp. 525–536. Blattman, Christopher (2009) “From violence to voting: War and political participa- tion in Uganda,” American Political Science Review, Vol. 103, pp. 231–247. Blattman, Christopher and Jeannie Annan (2010) “The consequences of child soldier- ing,” Review of Economics and Statistics, Vol. 92, pp. 882–898. Bozzoli, Carlos, Tilman Br¨ uck, and Tony Muhumuza (2011) “Conflict experiences and household expectations on recovery: Survey evidence from northern Uganda,” Working Paper 1785436, SSRN. Brakman, Steven, Harry Garretsen, and Marc Schramm (2004) “The strategic bomb- ing of German cities during World War II and its impact on city growth,” Journal of Economic Geography, Vol. 4, pp. 201–218. Brandt, Vincent SR (1972) A Korean village: Between farm and sea: Harvard Uni- versity Press. Campos, Javier and Gines De Rus (2009) “Some stylized facts about high-speed rail: A review of HSR experiences around the world,” Transport Policy, Vol. 16, pp. 19–28. Cassar, Alessandra, Pauline Grosjean, and Sam Whitt (2011) “Civil war, social cap- ital and market development: Experimental and survey evidence on the negative consequences of violence,” Working Paper 1917111, SSRN. 128 Chandler, David Porter (1999) Voices from S-21: Terror and history in Pol Pot’s secret prison: University of California Press. Chung, Il Hong and Seong Woo Lee (2011) “The effects of KTX on population dis- tribution between 2004 and 2009,” Journal of the Korean of Regional Science As- sociation, Vol. 27, pp. 121–138. Collier, Paul and Anke Hoeffler (1998) “On economic causes of civil war,” Oxford Economic Papers, Vol. 50, pp. 563–573. Collins, Kathleen (2002) “Clans, pacts, and politics in Central Asia,” Journal of Democracy, Vol. 13, pp. 137–152. Costa, Dora L and Matthew E Kahn (2003) “Civic engagement and community het- erogeneity: An economist’s perspective,” Perspective on Politics, Vol. 1, pp. 103– 111. Cox, Donald and Marcel Fafchamps (2007) “Extended family and kinship networks: Economic insights and evolutionary directions,” Handbook of Development Eco- nomics, Vol. 4, pp. 3711–3784. Davis, Donald R and David E Weinstein (2002) “Bones, bombs, and break points: The geography of economic activity,” American Economic Review, Vol. 92, pp. 1269–1289. De Luca, Giacomo and Marijke Verpoorten (2015) “Civil war, social capital and resilience in Uganda,” Oxford Economic Papers, Vol. 67, pp. 661–686. Donaldson, Dave (2010) “Railroads of the Raj: Estimating the impact of transporta- tion infrastructure,” Working Paper 16487, National Bureau of Economic Research. Durlauf, Steven N and Marcel Fafchamps (2004) “Social capital,” Working Paper 10485, National Bureau of Economic Research. Edwards, Paul (2003) The Korean War: A historical dictionary (Historical dictio- naries of war, revolution, and civil unrest, No. 23): Scarecrow Press. Edwards, Paul M (1998) The Korean War: An annotated bibliography, No. 10: Green- wood Publishing Group. Fafchamps, Marcel and Flore Gubert (2007) “The formation of risk sharing networks,” Journal of Development Economics, Vol. 83, pp. 326–350. Fogel, Robert William (1994) Railroads and American economic growth: Cambridge Univ Press. Garc´ıa Montalvo, Jos´e and Marta Reynal-Querol (2002) “Why ethnic fractionaliza- tion? Polarization, ethnic conflict and growth,” Polarization, Ethnic Conflict and Growth (September 2002). UPF Economics and Business Working Paper. 129 (2005) “Ethnic polarization, potential conflict, and civil wars,” American Economic Review, Vol. 95, pp. 796–816. Getty, John Arch (1987) Origins of the great purges: The Soviet Communist Party reconsidered, 1933-1938, Vol. 43: Cambridge University Press. Gibney, Frank (1992) Korea’s quiet revolution: From garrison state to democracy: Walker. Gilligan, Michael J, Benjamin J Pasquale, and Cyrus Samii (2014) “Civil war and so- cial cohesion: Lab-in-the-field evidence from Nepal,” American Journal of Political Science, Vol. 58, pp. 604–619. Glaeser, Edward L, David I Laibson, Jose A Scheinkman, and Christine L Soutter (2000) “Measuring trust,” Quarterly Journal of Economics, Vol. 115, pp. 811–846. Gonzalez-Navarro, Marco and Matthew A Turner (2016) “Subways and urban growth: Evidence from Earth,” working paper, Brown University. Ha, Inshik (2015) “People and money flock into Daegu and Gimcheon,” Hankyung Newspaper, May 21, 2015. Hassan, Hussein D (2007) “Iraq: Tribal structure, social, and political activities,” Report for Congress 22626, Congressional Research Service. Hastings, Max (1987) The Korean War: Simon and Schuster. Henderson, J Vernon, Adam Storeygard, and David N Weil (2012) “Measuring eco- nomic growth from outer space,” American Economic Review, Vol. 102, pp. 994– 1028. Hwang, In-jeong (1980) Comprehensive agricultural development in South Korea: As- sessment and outlook: Korea Rural Economic Institute. International Crisis Group (2015) “Chiefs: elite control of civil society and economic development in Sierra Leone,” Middle East Report 163, International Crisis Group. Isham, Jonathan (2002) “The effect of social capital on fertiliser adoption: Evidence from rural Tanzania,” Journal of African Economies, Vol. 11, pp. 39–60. Jacobs, Norman (1985) The Korean road to modernization and development: Univer- sity of Illinois Press. Khwaja, Asim Ijaz (2009) “Can good projects succeed in bad communities?” Journal of Public Economics, Vol. 93, pp. 899–916. Kim, Cheongryum (2006) From a poor country to an advanced country: Random House Joongang. 130 Kim, Eunjeong and Euijong Lee (2011) “Facilitation of KTX Economic Zone and improvement of the quality of life,” Urban Affair, Vol. 46, pp. 8–13. Kim, Jin-Hyun, Jae-Hee Lee, and Jin-Hee Lee (2010) “Changes in healthcare utiliza- tions of cancer patients since the launch of KTX,” Journal of the Korean Society for Railway, Vol. 13, pp. 236–243. Kim, Kyong-dong (1985) Man and society in Korea’s economic growth: Sociological studies: Seoul National University Press. Kim, Sangkyu (2012) “Study in the mass migration of lineage villages and ethos,” Cultural Study of the Northeast Asia, Vol. 32, pp. 5–17. Kim, Young-mee (2009) Their New Village Movement: Blue History Press. Kline, Rex B (2013) Beyond significance testing: Statistics reform in the behavioral sciences: American Psychological Association. Knack, Stephen and Philip Keefer (1997) “Does social capital have an economic payoff? A cross-country investigation,” Quarterly Journal of Economics, pp. 1251– 1288. Korean Studies Advancement Center (2014) “Culture of aristocracy in Andong region,” URL: http://www.ugyo.net/cf/frm/tuFrm.jsp?CODE1=02&CODE2=03& CLSS=6&sBookNmbr=F003&sMok_Nmbr=4, (Retrieved on November 10, 2014). Krishna, Anirudh (2001) “Moving from the stock of social capital to the flow of benefits: The role of agency,” World Development, Vol. 29, pp. 925–943. Kwon, Youngjong (2014) “Construction of KTX and regional development,” Urban Affair, Vol. 49, pp. 27–31. Kyunghyang (1970) “Saemaul Undong in every village,” Kyunghyang Newspaper, p.1 Oct. 2 1970. La Ferrara, Eliana (2003) “Kin groups and reciprocity: A model of credit transactions in Ghana,” American Economic Review, Vol. 93, pp. 1730–1751. Lee, Kwang-kyu (1997) Psychological anthropology of Korean culture: Jibmoondang. Lim, Han-Sung and Jae-Kang Lim (2013) “A case study on success factors of Saemaeul Undong,” Korean Journal of Local Government Studies, Vol. 17, pp. 153–176. Litwak, Eugene and Ivan Szelenyi (1969) “Primary group structures and their func- tions: Kin, neighbors, and friends,” American Sociological Review, Vol. 34, pp. 465–481. Maeil Kyungjae (1971) “Three billion Won investment in Saemaul Undong,” Maeil Kyungjae Newspaper, September. 17 1971. 131 Marr, Phebe (2011) The modern history of Iraq: Westview Press. Mauro, Paolo (1995) “Corruption and growth,” Quarterly Journal of Economics, Vol. 110, pp. 681–712. Michaels, Guy, Ferdinand Rauch, and Stephen J Redding (2012) “Urbanization and structural transformation,” Quarterly Journal of Economics, Vol. 127, pp. 535–586. Michalopoulos, Stelios (2012) “The origins of ethnolinguistic diversity,” American Economic Review, Vol. 102, p. 1508. Michalopoulos, Stelios and Elias Papaioannou (2013) “Pre-colonial ethnic institutions and contemporary African development,” Econometrica, Vol. 81, pp. 113–152. (2014) “National institutions and subnational development in Africa,” Quar- terly Journal of Economics, Vol. 129, pp. 151–213. Miguel, E. and M.K. Gugerty (2005) “Ethnic diversity, social sanctions, and public goods in Kenya,” Journal of Public Economics, Vol. 89, pp. 2325–2368. Miguel, Edward and Gerard Roland (2011) “The long-run impact of bombing Viet- nam,” Journal of Development Economics, Vol. 96, pp. 1–15. Ministry of Home Affairs (1973) The Saemaul Movement: Ministry of Home Affairs. (1978) Glorious Footsteps: Ministry of Home Affairs. (1983) Saemaul Undong in Korea: Ministry of Home Affairs. Moore, M. (1984) “Mobilization and disillusion in rural Korea: The Saemaul move- ment in retrospect,” Pacific Affairs, Vol. 57, pp. 577–598. Mu, Ren and John Giles (2014) “Village political economy, land tenure insecurity, and the rural to urban migration decision: Evidence from China,” World Bank Policy Research Working Paper. Munshi, Kaivan (2003) “Networks in the modern economy: Mexican migrants in the US labor market,” Quarterly Journal of Economics, Vol. 118, pp. 549–599. Nunn, Nathan and Diego Puga (2012) “Ruggedness: The blessing of bad geography in Africa,” Review of Economics and Statistics, Vol. 94, pp. 20–36. Oberdorfer, Don (1997) The two Koreas: A contemporary history: Basic Books. Palloni, Alberto, Douglas S Massey, Miguel Ceballos, Kristin Espinosa, and Michael Spittel (2001) “Social capital and international migration: A test using information on family networks,” American Journal of Sociology, Vol. 106, pp. 1262–1298. Park, Hongshik (2013) “Gimcheon-Gumi KTX station as the engine of growth for Gimcheon (Korean),” Newsis Newspaper, November 17, 2013. 132 Park, Jin-Hwan (1998) The Saemaul Movement: Korea’s approach to rural modern- ization in 1970s: Korea Rural Economic Institute. Park, Yongjae (1999) The memoirs of Hun Pong: Homa and Sekey Books. Peng, Yusheng (2004) “Kinship networks and entrepreneurs in China’s transitional economy,” American Journal of Sociology, Vol. 109, pp. 1045–1074. Platteau, Jean-Philippe and Frederic Gaspart (2003) “The risk of resource misap- propriation in community-driven development,” World Development, Vol. 31, pp. 1687–1703. Posner, Richard A (1980) “A theory of primitive society, with special reference to law,” Journal of Law and Economics, Vol. 23, pp. 1–53. Rosenzweig, Mark R (1988) “Risk, implicit contracts and the family in rural areas of low-income countries,” Economic Journal, Vol. 98, pp. 1148–1170. Sands, B (1993) “The development effects of high-speed rail stations and implications for Japan,” Built Environment, Vol. 19, pp. 257–284. Schnabel, James F (1972) Policy and direction: The first year: Office of the Chief of Military History, United States Army. Seo, Bongdo (1997) Family clans: Suwon Sisa. Song, Sonhui (1982) Kinship and lineage in Korean village society: Indiana University. Strauss, Julia C (2002) “Paternalist terror: The campaign to suppress counterrevolu- tionaries and regime consolidation in the People’s Republic of China, 1950-1953,” Comparative Studies in Society and History, Vol. 44, pp. 80–105. Stueck, William (2002) Rethinking the Korean War: A new diplomatic and strategic history: Princeton University Press. Suh, Sunduk, Keun-Yul Yang, Jae-Hoon Lee, Byung-Min Ahn, and Jeong Hyun Kim (2005) “Effects Of Korean Train Express (KTX) operation on the national trans- port system,” Proceedings of the Eastern Asia Society for Transportation Studies, Vol. 5, pp. 175–189. The 8th Army Staff Historian’s Office (1972) Key Korean War battles fought in the Republic of Korea: The 8th Army Staff Historian’s Office. Thompson, Ian B (1995) “High-speed transport hubs and Eurocity status: The case of Lyon,” Journal of Transport Geography, Vol. 3, pp. 29–37. Tsai, Lily L (2007) “Solidary groups, informal accountability, and local public goods provision in rural China,” American Political Science Review, Vol. 101, pp. 355– 372. 133 Vickerman, Roger (1997) “High-speed rail in Europe: Experience and issues for future development,” Annals of Regional Science, Vol. 31, pp. 21–38. Voors, Maarten J, Eleonora EM Nillesen, Philip Verwimp, Erwin H Bulte, Robert Lensink, and Daan P Van Soest (2012) “Violent conflict and behavior: A field experiment in Burundi,” American Economic Review, Vol. 102, pp. 941–964. Warriner, G Keith and Trudy M Moul (1992) “Kinship and personal communication network influences on the adoption of agriculture conservation technology,” Journal of Rural Studies, Vol. 8, pp. 279–291. Whitelegg, John and Helrnut Holzapfel (1993) “The conquest of distance by the destruction of time,” High Speed Trains: Fast Tracks to the Future, pp. 203–212. Wickham, John Adams (1999) Korea on the brink: from the 12/12 Incident to the Kwangju Uprising, 1979-1980: National Defense University. Yesa Moonhan (2001) “Twenty clan villages,” URL: http://news679.ndsoftnews. com/news/articleView.html?idxno=14833, (Retrieved on July 2, 2014). Yi, Man-gap (1981) Toward a new community: Institute of Saemaul Undong Studies, Seoul National University. Yu, Tae-yeung (1986) The Patterns of rural development in Korea, 1970s: Institute of Saemaul, Kon-Kuk University. Zheng, Siqi and Matthew E Kahn (2013) “China’s bullet trains facilitate market integration and mitigate the cost of megacity growth,” Proceedings of the National Academy of Sciences, Vol. 110, pp. E1248–E1253. 134