Policy Research Working Paper 8837 Factions, Local Accountability, and Long-Term Development County-Level Evidence from a Chinese Province Hanming Fang Linke Hou Mingxing Liu Lixin Colin Xu Pengfei Zhang Development Economics Development Research Group May 2019 Policy Research Working Paper 8837 Abstract This paper investigates, both theoretically and empirically, on local grassroots support, which could be best secured the role of factional competition and local accountability if these leaders focused on local development. In addition, in explaining the enormous but puzzling county-level vari- the local guerrilla presence in the county further improved ations in development performance in Fujian province of the development performance either because it intensified China. When the Communist armies took over Fujian from local accountability of the county leader, or because it better the Nationalist control circa 1949, Communist cadres from facilitated the provision of local public goods beneficial to two different army factions were assigned as county leaders. development. The paper finds consistent and robust evi- For decades the Fujian Provincial Standing Committee of dence supporting these assumptions; being affiliated with the Communist Party had been dominated by members weak factions and having local accountability are both asso- from one particular faction, which we refer as the strong ciated with sizable long-term benefits in terms of growth, faction. Counties also differed in whether there was local education, private-sector development, and survival in the guerrilla presence prior to the Communist takeover. The Great Famine. The paper also finds that being affiliated model predicts that county leaders from the strong faction with the strong faction and adopting pro-local policies are were less likely to pursue policies friendly to local develop- associated with higher likelihood of political survival. The ment, because their political survival relied more on their empirical findings here suggest that factional competition loyalty to the provincial leader than on the grassroots sup- contributes to efficiency in non-democratic countries, and port from local residents. In contrast, the political survival that local accountability is a key ingredient for balanced of county leaders from the weak faction was based more development. This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at lxu1@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Factions, Local Accountability, and Long-Term Development: County-Level Evidence from a Chinese Province∗ Hanming Fang Linke Hou Mingxing Liu University of Pennsylvania & NBER Shandong University Peking University Lixin Colin Xu Pengfei Zhang World Bank Peking University Keywords: Local Accountability; Factional Politics; Political Survival; Development Performance; Famine. JEL Classification Numbers: O1, O43, H70, D72 ∗ We would like to thank Laura Baily, Loren Brandt, Shuo Chen, Angus Deaton, Melissa Dell, Yue Hou, Ruixue Jia, Asim Khwaja, Michael Kremer, Aaditya Mattoo, Wuyue You, Qi Zhang and seminar/conference participants in Princeton, the Second IMF/Atlanta Fed Conference on Chinese Economy (2017), Penn-UCSD Conference on “Chinese Institutions and Economic Performance” (2018), Fudan Conference on Chinese Political Economy (2018), and World Bank Policy Research Talk for helpful discussions and suggestions. We thank Shuo Chen and Bingjing Li for sharing their data. We gratefully acknowledge the financial support from Penn China Research Engagement Fund (CREF), and financial support from DFID through its World Bank Strategic Research Program. Part of the research was conducted when Fang was visiting the Department of Economics at Princeton University. We are responsible for all remaining errors. The views of this paper reflect the authors’ own and do not implicate the World Bank or the countries it represents. 1 Introduction There is a large literature on the causes of the cross-country differences in economic growth rates, emphasizing the roles of institutions (North, 2005), human capital and capital market imperfection (Lucas, 1990), culture (Barro, 1991; Barro and Lee 1994), among others. A recent and influential literature investigates the long-run impact of history on a nation’s economic performance, through its effect on the institutional development (La Porta, et al, 1998a, 1998b; Acemoglu, Johnson and Robinson, 2001, 2002). La Porta et al (1998a, 1998b) argue that history might impact current economic performance via the legal system. They show that, generally, common- law countries have the strongest, and French civil-law countries the weakest, legal protections of investors, with German- and Scandinavian-civil-law countries in the middle. For many developing countries, their colonial past affected the legal system they adopt, and casts a long shadow on its current economic performance. Acemoglu, Johnson and Robinson (2001, 2002) show that mortality rates among early European settlers in a colony strongly predict whether the colony ends up with “inclusive” institutions that protects property rights and shares prosperity, which in turn affects whether the country prospers today. 1, 2 Most of the existing literature, however, does not explain the vast differences in the economic performance across smaller geographical units within the same country or within the same province. 3 One exception is Banerjee and Iyer (2005). They focus on the effect of a specific historical institution in India — the system for collecting land revenue, and show that the present-day economic performance of different districts of India is very much related to the land revenue systems imposed by the British colonial rulers as a result of certain historical accidents. They show that, after controlling for a wide range of geographical differences, districts in India where the collection of land revenue from the cultivators was assigned to a class of landlords systematically under-perform, in term of agricultural investment, yields, and various measures of public investment and outcomes, relative to the districts where this type of intermediation was avoided. Another exception is Acemoglu, Reed and Robinson (2014), which examine the long-term consequences of political competition at the level of chiefdom in Sierra Leone. Here chiefs must come from exogenously given number of ruling families designated by the original British colonial authority. A large number of designated ruling families thus result in stronger political competition at the chiefdom level. They find that such competition leads to better indicators of development, but worse social capital. Dell et al (2018) examine the long-run impact of the historical state conditions in Vietnam, comparing Northern Vietnam (Dai Viet) where historically village was the fundamental administrative unit to Southern Vietnam where the 1 Engerman and Sokoloff (1997, 2000) argue that the fact that Europeans considered Brazil, but not the USA, to be suitable for growing sugar contributed to the much larger slave population in Brazil, and led to a much more hierarchical society in Brazil than in the USA. This caused a divergence in the types of institutions that evolved in the two countries, which in turn led to a divergence in the growth rates. 2 Acemoglu and Robinson’s (2012) book further develops this important thesis. 3 Gennaioli et al (2013) consider the determinants of regional development using a large dataset of sub-national regions from 110 countries, and focus on geographic, institutional, cultural, and human capital determinants of regional development. We focus on sub-national growth differences within one country, and we are able to examine long-term outcomes and potential mechanisms. 1 historical statecraft relied on more informal, personalized power relations but not on village intermediation. They use a regression discontinuity design to show that areas exposed to Dai Viet administrative institutions for a longer period prior to French colonization have experienced better economic outcomes over the past 150 years; and they argue that the mechanism for the better performance under Dai Viet is that institutionalized village governance led to more local cooperation. Our paper is in the same broad research agenda. We consider the within-provincial differences in a rich array of development outcomes in the Chinese province of Fujian after the Communist took power in 1949. We investigate, both theoretically and empirically, the role of political factions and local accountability in explaining the huge variations in the development performance across counties in Fujian province. The set of development outcomes we examine in this paper includes the economic growth rates between 1952-1998 and in the post-reform period (i.e., 1978-1998), the improvement in educational attainment between 1952-1990, the level of private sector development, as well as net birth rates and death rates during the Great Chinese Famine between 1959-1961, all at the county level. We also examine whether empirically political survival is part of our story. We provide a model that shed light on the underlying mechanisms that link the political faction, local accountability, and the development outcomes at the county level. The intuitions can be explained as follows. When the Communist armies took over Fujian province from the Nationalist control circa 1949, cadres mainly from two army factions–namely, the Third Field Army (FA3, henceforth) led by Ye Fei, and the Yangtze- River Detachment (YRD)– were assigned as county leaders. These cadres were commonly known as “South- Bound Cadres” because they came from northern provinces. The Fujian Provincial Standing Committee of the Communist Party, however, was always dominated by members from FA3, which we refer to as the strong faction. We argue that county leaders’ incentives regarding local development depended on whether they were from the strong faction, namely FA3, that dominated the provincial government. If county leaders belonged to FA3, then they were less likely to pursue policies that were friendly to local development, because as political incumbent, their political survival depended more on their connections to the provincial leader. On the other hand, if the local leader belonged to the weaker faction, his political survival depended more on local grassroot support, which could be best secured if he focused on local economic development. In addition, local guerrilla presence in the county further improved development performance either because it intensified local accountability of the county leader, or it better facilitated the provision of local public goods beneficial to development. We argue that the above mechanisms explain our finding that counties with leaders from the weaker faction have significantly better development outcomes in almost all the measures we examined, and the difference is stronger in counties that also had local guerrilla presence. Fujian province is a particularly suitable place to examine how faction affiliation and local accountability affect long-term development. Due to exogenous historical reasons, some counties had been governed by cadres affiliated with the strong faction in the province, others by the weak faction. Moreover, a significant share of counties had presence of Communist guerrilla before the 1949 takeover. As presumed by our model, weak-faction 2 counties and guerrilla-present counties likely had to depend more on grassroot support for political survival, and we should see systematical differences in long-term development. A key challenge for identifying the effects of weak faction affiliation (Weak Faction) and local accountability (Guerrilla) is that they may not be exogenous to long-term development. We do our best to ensure that the effects indeed likely reflect causal effects. First, we provide detailed institutional background of how different counties in Fujian ended up being in different counties, likely exogenously with respect to our indicators of long-term development. Indeed, we find that Weak Faction and Guerilla are not systematically related to initial income level and population, though they are related to the share of plain areas, which we include in subsequent empirical specifications. Second, in assessing the impact of Weak Faction and Guerrilla, we control for initial income, population, and local geography (i.e., the distance to the coast, and the fraction of mountainous areas in the county). The differences that we find are thus not a result of geography. Clearly, these counties shared the same legal system, religion and culture. Since we also controlled for the initial level of economic development in the county, the results cannot be attributed to conditional convergence. Third, we find that the estimates of the two key variables tend to be quite stable regardless of the amount of control—for instance, relative to strong-faction counties, weak-faction counties had a higher growth rate during 1952-1998 by 0.99 when no controls are added, and 0.91 when full controls are imposed. This is quite remarkable given the small sample we have. This robustness with respect to various control of correlates is consistent with exogenous assignment (Angrist and Pischke, 2009). Fourth, the key results remain robust quantitatively and qualitatively when we only using the neighboring-county pairs—for instance, when examining the effect of Weak Faction, the treatment is a county with Weak Faction being 1, and the control is a geographically neighboring county with Weak Faction being 0. This drastically reduces the sample but without changing our key conclusions, both qualitatively and quantitatively. We also rely on the model and derive the full implications of our causal story and test the channels. For instance, our model asserts that political survival would be higher with affiliation with the strong faction or with grassroot support, and we find supporting evidence using evidence of political survival during the Cultural Revolution. We also find Weak-Faction counties and Guerrilla counties tend to adopt stronger pro-local policies, as demonstrated by lower death rates during the Great Famine around 1960, greater private sector development, and higher changes in schooling attainment before and after the power change in 1949. When policies such as the implementation of One Child Policy or the Household Responsibility System were easily monitored by upper governments, the pro-local tendency associated with Weak Faction or Guerrilla disappear. Our empirical results yield coherent findings consistent with the implications of our model. First, based on our base specification with full controls, relative to strong-faction counties, weak-faction countries have an annual growth premium (in gross value of output) of 0.9 percentage points in 1952-1998, and this premium increases to 1.9 percentage points in 1978-1998, that is, the post-reform period. Similarly, relative to non-guerrilla counties, guerrilla-counties have a growth premium of 0.5 (1.5) percentage points in 1952-1998 (1978-1998). There is also robust evidence that the effects are especially pronounced when a county was both a weak-faction and a guerrilla 3 county, with the weak-faction-and-guerrilla counties growth premium (relative to strong-faction counties) often double that of weak-faction-without-guerrilla counties. Second, there is consistent evidence that weak-faction counties had significantly lower death rates during the Great Famine around 1960. Third, we offer evidence that the change in average schooling level for affected cohorts after the power change was positive and significantly larger—by around 37.5 percentage points —in weak-faction-and-guerrilla counties (relative to the strong-faction counties). Fourth, we find that weak-faction counties tend to be more conducive to private sector development. The SOE shares as measured by sales revenue and firm counts were about 22 and 15 percentage points lower, respectively, in weak-faction counties than in strong-faction counties. In addition, counties with local guerrilla presence also had lower SOE shares, though the effect is not statistically significant. Fifth, we provide evidence to directly support the political survival story of our model. In particular, counties affiliated with the strong faction or counties with lower death rates during the Great Famine tend to see their leaders that were purged during the initial years of the Cultural Revolution to be reinstalled earlier or more likely to be reinstalled, supporting the building block of the model, that is, both affiliation with strong factions and grassroot support are central for political survival in the autocratic regime. Finally, we offer further checks such as whether fiscal transfer explain the differences in development outcomes, and we could exclude this alternative interpretation. In general, our results tend to be quite robust to specifications with different controls, to the outlier concerns, and to using only the neighboring-county treatment-control pair. Our paper is related to several strands of literature in political economy. First, it is related to a recent literature on the effect of local accountability on development. Besley and Burgess (2002) develop a theoretical model to show that having a more informed and politically active electorate strengthens the incentives for governments to be responsive to citizens’ preferences. In their empirical setting of India, media presence plays the role of increasing local accountability. They show that state governments are more responsive to the falls in food production and crop flood damage by means of public food distribution and calamity relief expenditure in areas where newspaper circulation is higher (and thus electoral accountability is greater). Björkman and Svensson (2009) present a randomized field experiment on community-based monitoring of public primary health care providers in Uganda. They randomly choose communities in which residents are encouraged to be more involved in the state of the health service provision so as to strengthen their capacity to hold their local health providers accountable for their performance. They document large increases in utilization and improved health outcomes—reduced child mortality and increased child weight – in the treatment communities than in the control communities. 4 In our paper, local accountability is proxied by the local guerrilla presence prior to the Communist takeover. It is well known that guerrillas must rely on the support of the local residents for their survival; as such, guerrillas must foster a close and synergistic relationship with the local population. Therefore, leaders in counties with local guerrilla presence were likely more accountable to local residents, and could more easily mobilize the local 4 Mansuri and Rao (2013) provide a comprehensive overview of the theory and evidence for development strategies that are based on local community empowerment. 4 support. The results in our paper suggest that local accountability plays an important role in regulating the behavior of local politicians to be more favorable to local economic development in a non-democratic setting such as China under Communism. Furthermore, since sometimes economic growth is achieved at the expense of other development outcomes such as education, sectoral transformation, and sometimes even human sacrifices, our findings that local accountability is positively associated with many different aspects of development (i.e., growth, education, private sector development, and famine control) and at different stages of half a century of political transformation in China, suggest that local accountability are powerful, robust, and balanced mechanisms for ensuring long-term development. To the best of our knowledge, this is the first paper that considers the guerrilla presence as the historical origin of local accountability, and studies its comprehensive impact on local economic growth and other pro-local policies. 5 Second, our paper is related to the literature on the role of accountability for political survival in autocracies. The main idea is that concerns of political survival in autocracies drive the political leader to satisfy the demand of the people in order to avoid large-scale revolutions. For example, Acemoglu and Robinson (2001, 2006) argue that in nondemocratic societies, the elite will try to prevent revolution by making concessions to the poor (e.g., by redistributing income to them), because the poor poses a revolutionary threat even though they are excluded from political power. Smith (2008), and Bueno de Mesquita and Smith (2008, 2010) suggest that a leader can reduce the revolutionary threat by expanding public goods supply to potential revolutionaries. 6 Our paper focuses on the survival incentives of local leaders instead of national leaders. In our setup, local leaders can increase their chances of political survival either by strengthening their ties to the higher level officials, or by mobilizing the support from the grassroots within their jurisdiction. Moreover, we emphasize the interactions between the factional politics and local accountability in shaping the local leaders’ incentives when they choose local policies. We are able to provide direct evidence from power purge and recovery during the Cultural Revolution to offer support to our model. Third, our paper is related to the literature on the role of political competition in general and in communist China in particular. As summarized in Acemoglu, Reed, and Robinson (2014), existing studies on political competitions are few, and most focusing on the context of developed democratic countries such as United States (e.g., Ansolabehere and Snyder, 2006; Besley, Persson, and Sturn, 2010). Very few papers exploit exogenous variations in political competition, with notable exceptions in Besley and Preston (2007), which uses redistricting as the source of variations, and in Acemoglu, Reed and Robinson (2014), which uses the number of designated ruling family for potential chiefs in chiefdoms in Sierra Leone. We differ in capturing political competition 5 The literature on local accountability in non-democratic countries is relatively small, and if any, the focus is on upward accountability (i.e. accountability to the superiors) for the purpose of promotions. Few exceptions exist in the literature. Li (2014) shows that provincial leaders in China tend to implement policies more in favor of the citizens in response to intensified labor disputes, suggesting evidence for downward accountability. Distelhorst and Hou (2017) study the responsiveness of Chinese city government officials to appeals from putative citizens and find that their average responsiveness is comparable to that in democracies. 6See Bueno de Mesquita et al (2003) for a book-length treatment of the logic of political survival. 5 explained by local leaders’ factional affiliation, which was caused by the liberation pattern circa 1949. We show that stronger political competition faced by weak-faction leaders led to better long-term growth, social development (i.e., lower famine deaths, better education), and stronger private sector development. Our results are thus consistent with Becker (1958), Stigler (1972), Wittman (1989), and the new evidence of Acemoglu, Reed and Robinson (2014) that political competition contributes to efficiency by reducing distortions. Finally, our paper is related to the literature on factions in communist China. Most of the studies on factions in China focus on the post-reform period. Shih (2008) provides a systematic analysis of factions based on networks of leaders, and studies the impact of elite politics on monetary and banking policies. Shih, Adolph and Liu (2012) examine the role of factional ties with top leaders in the promotion to the Central Committee of the Chinese Communist Party. Few papers study the impact of factional ties based on military affiliations among the south-bound cadres. One exception is Shih, Zhang and Liu (2013), which examine a possible causal relationship between a region’s communist revolutionary legacy before 1949 and the variation in private sector development after 1949 for the case of Zhejiang province. Our paper offers a theoretical framework, and delves deeper and more systematically into the broad impact on a variety of development performance measures of the counties led by south-bound cadres from different military factions. The remainder of the paper is structured as follows. In Section 2 we present a simple model to highlight the role of political faction and local accountability on the development performance; in Section 3 we provide the institutional background of the factions and local guerrilla in the post-civil-war county and provincial governments of Fujian Province; in Section 4 we describe our data sources and present summary statistics for some of the key variables of interest; in Section 5 we present our empirical results, as well as robustness checks and alternative explanations; finally, in Section 6 we conclude. 2 Political Survival, Local Accountability, and Local Economic Performance: A Model 2.1 An Illustrative Example We first present a simple example to illustrate the main forces that local (i.e., county) leaders face in their decision making to ensure political survival. Consider two factions, w and s, representing the weak and the strong factions, respectively. The provincial government is controlled by the strong faction; indeed, the control of the provincial government is what makes the so-called “strong faction” strong. We focus on the incentives of the county-level leaders. A leader from faction f ∈ { w, s } faces an exogenous baseline probability ρ f ∈ (0, 1) of being stripped from power. It is reasonable to assume that ρs < ρw . However, whether the county leader will actually be stripped of power depends both on his faction and the policy actions he chooses. A leader’s faction is exogenously given and 6 can not be changed. He can choose from two possible policy actions: the first is pro-local economic development, which we denote by L ; and the second is anti-local (or, pro-upper) policy, which we denote by U . We can interpret L as extracting less local resources to send to the provincial government, and instead focusing on providing more local public goods; or L could represent following economic policies that are more suitable to the local conditions instead of following the policies preferred by the central leaders. Taking action L will endear the county leader to the local citizens, which helps protect him from being purged from power. Taking action U will curry favor from the provincial leaders, but whether it would translate into protection against being purged depends on whether the local leader belongs to the strong or the weak faction. Suppose that the actual probability that a county leader will be purged from his power is represented by the following matrix, where ρw > ρwU > ρwL, and ρs > ρsL > ρsU . In this simplest example, these assumptions say that for the local leaders belonging to the weak faction, it is more effective to protect himself from purge by adopting policy L , and hopefully to gain local support than to curry favor with the provincial leader. In contrast, for the local leaders in the strong faction, currying favors from the provincial leader is a more effective strategy for political survival. It is clear that the optimal action choice of a local leader whose goal is to maximize his probability of political survival will then depend on his faction: if he belongs to the weak faction, he will choose pro-local action L ; and if he belongs to the strong faction, he will choose the anti-local policy U . Faction Action L U w ρwL ρwU s ρsL ρsU Table 1: Probabilities of Political Purge for Local Leaders, as Functions of his Factions and Actions 2.2 A Model The above example highlights the key tradeoffs that local leaders of different factions may face in their choices of whether to adopt pro- or anti-local economic policies. In this sub-section, we enrich the model to show that the key forces identified in the illustrative example are robust under the assumption that local leaders from the strong faction has a comparative advantage over the local leaders from the weak faction in gaining the political support from the provincial leaders; we also incorporate the effect of local accountability in our model. As in the illustrative example, consider two factions: w, standing for the weak YRD faction; s, standing for the strong FA3 faction. We distinguish officials at two levels of the government, those at the provincial level and those at the county level. At the provincial level, as we will document empirically below (see Figure 1 in Section 3), the dominant faction, or the faction in power, has been those from FA3, s . We analyze the incentives of the officials at the county level. A county is led by cadres affiliated with faction f ∈ { w, s } ; and the county may or may not have guerrilla 7 presence, which we denote by g ∈ {0, 1} where g = 1 indicates that the county had guerrilla presence. We hypothesize that any county official faces possible shocks that may lead to their dismissal. The probability that a faction- f , f ∈ { w, s } , official in a county with guerrilla presence g ∈ {0, 1} will be dismissed is given by f exp −α f × κ g × Z − β f × T ∈ (0, 1) , ρ f , g ( Z, T ) = ρ0 (1) f ∈ (0, 1) denotes the baseline probability of a local official from faction f being purged from power; where ρ0 Z denotes the grassroot support that a faction- f county official enjoys from the citizens in the county, whose determinant will be specified below; T is the amount of “tax revenue” he collects from the citizens and sends to the higher-ups in the provincial government, or alternatively, the amount of local resources the county leader collects and spends on the projects dictated by the higher-ups in the provincial government that does not benefit the local citizens. We assume that α f and β f are both positive, in order to capture that a local official’s chance of political survival can be improved by either building strong local support or currying favors from the provincial leaders. Parameter κ g measures the effect of local accountability on the importance of grassroot support in political survival; without loss of generality, we normalize κ 0 = 1 and let κ 1 = κ > 1 to capture that grassroot support is more important in counties with strong local accountability as proxied by guerrilla presence. We make the following assumptions regarding the functions ρ f , g (·, ·) : Assumption 1 α f > 0, β f > 0 for both f ∈ { w, s } . To capture the difference between the weak faction and the strong faction, which lies in the fact that the strong faction has their allies controlling the provincial government, we assume that the transfers to the higher-ups are more effective in reducing the chances of being dismissed for county officials from the strong faction than those from the weak faction: Assumption 2 (Comparative Advantages of Strong vs. Weak Factions) α s / β s < α w / βw . Assumption 2 formalizes the idea that local leaders from the strong factions have comparative advantage relative to those from the weak faction to use upward transfers to reduce the probability of purges; likewise, local leaders in the weak factions have relative advantage to use local citizen support in reducing the probability of purges. Both Z and T are affected by the policy choices of the local leader. We proxy Z by the local economic outcomes. We hypothesize that the local citizens have an endowment (which could be interpreted either as income or labor) equal to E . Local citizens can choose to allocate their endowment E in two ways. They can spend it as investment (or labor supply for the market) I which produces f ( I ) > I . Or they can keep it under a mattress, which represents a storage technology (or working in their own backyard). We assume that f > 0 and f < 0, and satisfies Inada conditions. The output produced from investment I, however, can be taxed by the local government, while the storage technology is secret and not subject to taxation. 8 Local citizen’s wellbeing is measured by their after-tax income, which is given by: Z = (1 − τ ) f ( I ) + ( E − I ) . (2) where τ is the tax rate chosen by the county official. The “tax revenue” that the local leader can either send to the provincial leader or spend on projects dictated by the higher ups, T , is given by T = (1 − τ ) f ( I ) . (3) In a county where its local leader chooses tax rate τ, the citizens will choose I to maximize (2). The first order condition for citizens’ optimal investment I ∗ is given by (1 − τ ) f ( I ∗ ) = 1 . (4) (4) is also sufficient for optimality given our assumptions on f (·) . Since f > 0, we have 1 I ∗ (τ ) = f −1 . (5) 1−τ Moreover, f < 0 implies that 1 1 I ∗ (τ ) = < 0. (6) f ( I ∗ (τ )) (1 − τ )2 The local citizen’s after tax income evaluated at the optimal investment choice I ∗ (τ ) is given by Z ∗ (τ ) = (1 − τ ) f ( I ∗ (τ )) + [E − I ∗ (τ )] (7) It follows from the Envelope Theorem that Z ∗ (τ ) = − f ( I ∗ (τ )) < 0. (8) Together with (6), we have Z ∗ (τ ) = − f ( I ∗ (τ )) I ∗ (τ ) > 0. (9) For a given level of tax rate τ chosen by the local leader, the available tax revenue that the local leader can collect and transfer to the provincial level officials is then given by T ∗ (τ ) = τ f ( I ∗ (τ )) (10) (10) is the “Laffer Curve” that depicts the relationship between the total tax revenue and the tax rate. It is clear that T ∗ (0) = 0, and limτ →1 T ∗ (τ ) = 0. Thus T ∗ (τ ) is non-monotonic. Indeed simple algebra shows that T ∗ (τ ) = f ( I ∗ (τ )) + τ f ( I ∗ (τ )) I ∗ (τ ) , which does not have a definite sign. The sign of the second derivative of T ∗ (·) with respect to τ is more involved: T ∗ (τ ) = 2 f ( I ∗ (τ )) I ∗ (τ ) + τ f ( I ∗ (τ )) [ I ∗ (τ )]2 + τ f ( I ∗ (τ )) I ∗ (τ ) . 9 The first two terms are unambiguously negative, but the third term depends on the sign of I ∗ (τ ) . If I ∗ (τ ) < 0, then T ∗ (τ ) < 0. From (6), we have f ( I ∗ (τ )) I ∗ (τ ) (1 − τ )2 − 2 f ( I ∗ (τ )) (1 − τ ) I ∗ (τ ) = − 2 . (11) f ( I ∗ (τ )) (1 − τ )2 To convey our intuition as cleanly as possible, we make the following assumption: Assumption 3 The production function f (·) is such that T ∗ (·) is a globally concave function. Remark 1 A sufficient (but not necessary) condition for T ∗ (·) to be globally concave is f ≤ 0. This follows from the fact that f ≤ 0 is a sufficient (but not necessary) condition for I ∗ (τ ) < 0, which is in turn a sufficient condition for T ∗ (τ ) < 0. Now we can describe the choice problem of the county officials affiliated with faction f where the guerrilla presence is g . He/she will choose τ f , g to solve min ρ f ,g Z ∗ τ f ,g , T ∗ τ f ,g (12) τ f , g ∈[0, 1] where Z ∗ (·) and T ∗ (·) are respectively given by (7) and (10). The first order condition with respect to τ for county officials who belong to faction f with guerrilla presence g is f ,g + β f T α f κ g Z ∗ τ∗ ∗ f ,g = 0 τ∗ (13) f , g denotes the optimal solution for officials from faction- f county, f ∈ { s, w } , where the guerrilla presence Here τ ∗ is g ∈ {0, 1}. Proposition 1 Under Assumptions 1 and 2, we have the following predictions: 1. For a given g ∈ {0, 1} , s, g > τ w, g ; (a) local leaders from the strong faction will choose higher tax rates: τ ∗ ∗ (b) citizens in counties whose leader belongs to the strong faction will have lower after tax income: s, g < Z Z ∗ τ∗ ∗ τ∗ w, g ; (c) more taxes are collected from counties whose leaders belong to the strong faction: T ∗ τ ∗ s, g > w, g . T ∗ τ∗ 2. Similarly, for a given f ∈ { w, s } , f ,1 < τ f ,0; (a) local leaders in counties with guerrilla presence will choose lower tax rates: τ ∗ ∗ f ,1 > Z (b) citizens in counties with guerrilla presence will have higher after- tax income: Z ∗ τ ∗ ∗ τ∗ f ,0 ; 10 f ,1 < T (c) less taxes are collected from counties with guerrilla presence: T ∗ τ ∗ f ,0 . ∗ τ∗ 3. Local economic development as proxied by Z ∗ will be highest in counties with guerrilla presence and led by cadres affiliated with the weak faction, and worst in counties without guerrilla presence and led by cadres affiliated with the strong faction: Z ∗ τ∗ s, 0 s, 0 , Z = min Z ∗ τ ∗ ∗ τ∗ s, 1 , Z ∗ τ∗ w, 0 , Z ∗ τ∗ w, 1 , (14) Z ∗ τ∗ w, 1 s, 0 , Z = max Z ∗ τ ∗ ∗ τ∗ s, 1 , Z ∗ τ∗ w, 0 , Z ∗ τ∗ w, 1 . (15) Proof. To prove Part (1), consider g ∈ {0, 1} . The first order conditions for local leaders of strong and weak factions are respectively s, g + β s T αs κ g Z ∗ τ∗ ∗ τ∗ s, g = 0 (16) w, g + β w T αw κ g Z ∗ τ∗ ∗ τ∗ w, g = 0. (17) Taking the ratios, we have αs Z ∗ τ∗ s, g αw T ∗ τ∗ s, g = . (18) βs Z ∗ τ∗ w, g βw T ∗ τ ∗ w, g s, g ≤ τ w, g . As we showed in (9), Z Suppose τ ∗ ∗ ∗ (τ ) > 0. Together with Assumption 3, τ ∗ ≤ τ ∗ s, g w, g implies that Z ∗ τ∗ s, g T ∗ τ∗ s, g ≤ 1 and ≥ 1. (19) Z ∗ τ∗ w, g T ∗ τ∗ w, g The two inequalities in (19) and Assumption 2 then imply that αs Z ∗ τ∗ s, g αw T ∗ τ∗ s, g < , βs Z ∗ τ∗ w, g βw T ∗ τ ∗ w, g s, g > τ w, g as claimed in (a). Claim (b) follows from (8). For Claim (c), note that the contradicting (18). Hence, τ ∗ ∗ first order conditions (16) and (17) can be satisfied only if T ∗ τ ∗ s, g and T ∗ τ∗ w, g are both in the positive region because Z ∗ τ ∗ s, g and Z ∗ τ∗ w, g are both negative from (8). Hence τ s, g > τ w, g implies T ∗ ∗ s, g > T ∗ τ∗ ∗ τ∗ w, g . The proof of Part (2) is completely analogous to the proof of Part (1). To prove Part (3), simply note that Claim (b) in Part (1) yields: Z ∗ τ∗ s, 0 w, 0 , < Z ∗ τ∗ (20) Z ∗ τ∗ s, 1 < Z ∗ τ∗ w, 1 ; (21) and Claim (b) in Part (2) yields: Z ∗ τ∗ w, 0 w, 1 , < Z ∗ τ∗ (22) Z ∗ τ∗ s, 0 s, 1 . < Z ∗ τ∗ (23) Claims (14) and (15) follows from combining (20)-(23). 11 3 Institutional Background In this section, we present the historical and institutional background of Fujian province for the period under our consideration. Fujian province is a particularly suitable province to study how local accountability and factional politics may shape local development performances. First, during the early periods of the Chinese Civil War (July 1946- October 1949) between the Nationalist (Kuomintang, henceforth, KMT) and the Communist (People’s Liberation Army, henceforth, PLA) armies, there were local Communist guerilla presences in Fujian even though the province was under the formal rule of the Nationalist government; importantly, the presence of the guerilla forces varied significantly across the counties within Fujian province. This provides the cross- county variations in local accountability, which we proxy with local guerrilla presence. Second, there had been well-defined army-based factions among the county-level leaders in Fujian after the Communist takeover. The army-based factions resulted from two different forces within the PLA jointly taking over the administration of Fujian after the defeat of the KMT. Specifically, the liberation of Fujian from the KMT control experienced two phases. Between May to July 1949, PLA’s Second Field Army (FA2) led by Liu Bocheng and Deng Xiaoping entered Fujian province from the Southwest of neighboring Zhejiang province and the Northeast of neighboring Jiangxi province, and played an important role in liberating the ten counties in northern Fujian. However, FA2 was immediately mobilized to fight in Southwestern China. From June to October 1949, PLA’s Third Field Army (FA3) led by Marshall Chen Yi, replaced FA2 as the major military force in Fujian to attack the KMT forces that were then still controlling part of the Fujian Province. The 10th Battalion of FA3, led by Ye Fei, entered from the eastern part of Fujian province, eliminated the main KMT forces defending Fujian and took control of the major cities of Fuzhou, Zhangzhou and Xiamen. At the same time, the guerilla forces that were active along the borders of Fujian, Zhejiang, Jiangxi and Guangdong actively participated in the liberation of counties in central Fujian (including Pu Tian, Xian You) and western Fujian (including Shang Hang, Ming Nan and Ping He, among others). By May 1950, the communist took control of the Fujian province except for the outpost islands of Jing Men and Ma Zhu which have been under the control of the military force associated with Taiwan. As we will soon explain , the two different army forces that played a role in the liberation of Fujian became the basis of army factions we study in this paper. County Level Government. As in any power transition, the new Communist government needed to quickly install cadres at all levels of the governing bureaucracy. The Chinese bureaucracy under communism consists of two parallel but inter-related apparatus: the Communist Party organization (“the party”), and the People’s Government (“the state”). Hierarchically, below the central government, there are two hierarchical ladders at the local level: the provincial and the county leadership. At both levels, the Party is headed by a Party Secretary; and the People’s Government is headed by a chief (Governor at the provincial level and County Chief at the county level). Of course, there were hundreds of other positions at lower levels to fill. It was a huge task to build a 12 bureaucracy completely from scratch because the Communist government could not use any of the officials of the previous regime to staff these positions due to ideological differences and concerns for sabotage. The central government thus staffed these leadership positions by drawing from the military men and women from the armies that liberated Fujian. These newly installed cadres, commonly known as “South-bound Cadres” because they were mostly from the Chinese Communist’s power base in northern China, played an important role in the governance of the Mainland China after the Civil War. For Fujian provinces, even though both FA2 and FA3 played important roles in liberating Fujian from the KMT forces, FA2 was deployed to fight elsewhere before Fujian was fully controlled by PLA. In early 1949, members that were mainly drawn from FA2’s Communist bases in Hebei and Shanxi provinces, particularly in Tai Hang and Tai Yue Communist Revolutionary Bases, were assembled into a unit, known as Yangtze-River Detachment (YRD hereafter). YRD members were sent to southern provinces, including Fujian, to take up these leadership positions. As a result, the two major army factions from which the cadres in Fujian Province were drawn were FA3 and YRD. Specifically, FA3-affiliated cadres took over the top positions in the two major cities in Fujian – Fuzhou and Xiamen – as well as a total of 11 of the 59 counties in Fujian, including Long Yan, Yong An counties. YRD-affiliated cadres, organized into six units, were called to take over the remaining 48 counties in Fujian province. These counties include Jing Jiang, Jian Yang, Nan Ping, counties surrounding Fuzhou including Fu Qing, Ping Tan and Pu Tian, Zhang Zhou, and Fu An. 7 As mentioned earlier, another important variation for the liberation experience of counties was whether a local guerrilla force was present and participated in the liberation. When there was a local guerrilla force participating, the local government tended to have some members from that guerrilla force. We will interpret that local guerilla presence would lead to more local accountability. Provincial Level Government. At the provincial level, the Fujian Chinese Communist Party (CCP) Committee was the highest level decision-making body; and within the CCP Committee, the key members that played the pivotal role were the CCP Standing Committee, which controlled the rights to make decisions on political, personnel and economic matters. Historically, there were 13 members on the Standing Committee. As expected, the Standing Committee members were mostly either from FA3 or YRD. The cadres from these two factions were the major political forces in the post-liberation CCP Standing Committee in Fujian for decades. Figure 1 depicts the fraction of members from FA3 and YRD in Fujian CCP Standing Committee from 1950 to 1993, respectively. From the figure it is clear that the FA3 faction was the more dominant force in the Standing Committee than the YRD faction. Up to mid-1980s, FA3 members accounted for about 40-50%, and it was as high as over 60% in early 1950s, of the Standing Committee. The share by YRD members hovered around 20%. it should also be noted that, Ye Fei, the aforementioned commander of the 10th Battalion of FA3 that fought the KMT forces in Fujian, was appointed as the CCP Party Secretary of Fujian Province from October 1954 until May 1967. It is 7 Source: Jing De, Tie Min and Zhi Wan (2010). 13 also pertinent that cadres from local guerilla forces were marginalized in the Fujian Standing Committee. Only two Standing Committee members came from the local guerilla forces - Zeng Jingbing and Wei Jinshui – before the Cultural Revolution (1966-1976), and one of the two members (Zeng Jingbing) was removed in 1955. After the 1978 economic reform, only two people with experience in the local guerillas during the Civil War were on the Standing Committee. Thus cadres from the local guerilla forces were not a powerful element in the Standing Committee. While there were also members on the Standing Committee that did not belong to either FA3 or YRD factions, these members tended to have much shorter tenure on the Standing Committee, and their influence on Figure 1 Share of FA3 and YRD in Provincial Party Standing Committee in Fujian the political, personnel and economic decisions in Fujian was limited. Province during 1950-1993 1978 80 60 Percent (%) 40 20 0 1950 1960 1970 1980 1990 Year FA3 YRD Note: We implemented a two-step approach to calculate the shares of Share of FA3 and YRD in Provincial Party Standing Committee. In the first step, we read resumes Figure 1: Share of FA3 andof every Standing YRD Committee in Provincial member Party and identify Standing if they have working Committee experience in Fujian in Province During 1950-1993. the Third Field Army or they were members of YRD. In the second step, we divided Note: We calculate the shares ofthe FA3 and of number YRD in Provincial provincial Party Standing Standing Committee members Committee as follows. from FA3 and First, we read resumes of every YRD by the total number of provincial Standing Committee members year by year, respectively. Standing Committee member and identify if they had working experience in FA3 or if they were members of YRD. We then divide the number of Provincial Standing Committee members from FA3 and YRD, respectively, by the total number of provincial Standing Committee members year by year. Factional Conflicts. As is well known among students of the history of the Chinese Communist Party, the CCP has undergone numerous movements since it took power in 1949. At the national level, the best-known example to illustrate these political movements is the experience of Deng Xiaoping, who later became the architect of Chinese economic reform and the paramount leader of China after the death of Mao Zedong. Deng’s political career after the founding of the People’s Republic of China had several ups and downs. In July 1952, Deng assumed the 2 / 31 posts of Vice Premier and Deputy Chair of the Committee on Finance, and shortly afterwards, he took the posts of Minister of Finance and Director of the Office of Communications. Yet in 1954, he was removed from all of these positions, holding only the post of Deputy Premier. In 1956, he became the Head of the Communist Party’s Organization Department and a member of the powerful Central Military Commission. In Mao’s Anti-Rightist 14 Movement of 1957, Deng acted as Secretary General of the Secretariat and ran the country’s daily affairs with President Liu Shaoqi and Premier Zhou Enlai during the Great Leap Forward of 1957-1960. Yet, during the Cultural Revolution, Deng was twice purged from the central power apparatus. The first purge occurred in October 1969 when Deng was sent to Jiangxi province to work as a regular worker in a tractor factory. Only in 1974 when Premier Zhou Enlai fell ill with cancer was Deng brought back to politics as the First Vice-Premier. He was purged yet again in 1976 after the death of Premier Zhou Enlai when he was removed from all positions following the Tiananmen Incident of April 5, 1976. He re-emerged as the de facto leader of China following the death of Chairman Mao on September 9, 1976 and the purge of the Gang of Four in October 1976. 8 Within Fujian, there were serious power struggles between cadres from FA3 and local guerilla forces. As mentioned, Zeng Jingbing, one of the only two provincial CCP Standing Committee members with local guerilla background was removed in 1955. In 1957, during the Anti-Rightist Movement, many of the local leaders with guerilla background were purged and being stripped of power. In the 1959 Anti-Localism Movement, the Acting CCP Party Secretary of Fujian Province, Jiang Yizhen, was stripped of power and sent to work in a steel factory; Wei Jinshui, then Vice Governor of Fujian province, was also reprimanded. Both Jiang and Wei had local guerilla connections. 9 The political struggles among the provincial leadership had serious implications on the fortunes of many local leaders at the county level. In the presence of constant power struggles, local leaders faced serious risks of being purged. As shown below, local leaders from the relatively weak YRD tended to adopt economic policies that were more protective of local economic development in their areas of jurisdiction. Many of these decisions were driven to mobilize the grassroot support in order to increase their chances of political survival. Examples of Local Leaders’ Discretion in Economic Policies. We now provide some examples of how local leaders chose economic policies that were more or less friendly to local economic development. The leaders of Shanghang county belonged to the FA3 faction. In the majority of the post-liberalization period up to the dawn of the reform (i.e., 1978), FA3 cadres held the top leader position in the county. During the multitude of political movements in the Communist era, the local cadres from FA3 in general adopted leftist, collectivist economic policies. For example, during the Great Leap Forward Movement (1958-1962), the local cadres were strict in fulfilling the quotas of grain procurement, which led to severe famine in the county (see empirical evidence below). In addition, the local leaders also adopted more strictly the centralized economic policies, with the share of State-owned Enterprises (SOEs) in the local economy rising steadily from 1958 to as high as 70 to 88%. In 1985, SOEs still accounted for 56% of the total local industrial output. It was not until 1991 that individual and private ownership surpassed 30% for the first time. In contrast, Dong Shang County was governed largely by cadres from YRD and local cadres that they tutelaged. 8See Vogel (2011) for a detailed account of Deng’s life and his influence on China. 9Jiang Yizhen’s career was later rehabilitated in 1962 and became China’s Acting Minister of Agriculture in 1964. 15 These leaders overall were more reserved in the implementation of the leftist policies imposed from the central and provincial-level governments. During the Great Leap Forward Movement, the local cadres did not succumb to the calls for farmers to go all in to produce steel in the “backyard furnace,” as did in many leftist regions. Instead, they encouraged local residents to plant trees, and to finish several large-scale civil engineering projects, e.g., constructing ocean-crossing dikes; implementing the drinking water program, and claiming land from sea. These projects greatly facilitated local development, and won local support. Moreover, through self-financing and financing from overseas, the county established a series of small collective and state-owned firms. In 1988, the share of SOEs in Dongshan’s industrial output was only 24.3%, that of collective firms was 36.3%, and that of privately-owned firms was for 36.6%. A third county is Jing Jiang County which was led by cadres from the YRD faction, and it had strong local guerrilla presence before 1948. In fact, the local guerrilla leaders tended to hold important positions within the county administrative structure, such as County Chief (1949-1958), and party secretary (1972-1976). Before the Cultural Revolution, the cadres from YRD and the local guerrilla force allowed non-state economic activities to continue. Even in 1974, statistical report shows that village-level firms (in reality, most of these village-level firms were privately owned) accounted for 41.1% of the county-level industrial output; and by 1987, this ratio was 80%. 4 Data Sources and Descriptive Statistics 4.1 Data Sources We now describe our data sources. County/Provincial Leaders’ Factional Affiliations and County’s Guerrilla Presence. Here we rely on in- formation from two primary sources: (1). “History of the Communist Party in Fujian Province, 1926-1987”; (2). “Recollections on Yangtz-River Detachment”. We use these two primary sources to determine whether a county was assigned cadres affiliated with the FA3 or with the YRD. Moreover, we hand-collected the resume of every member of the Fujian Provincial Communist Party Standing Committee from 1950-1993. We identify if a member belongs to the FA3 faction or the YRD faction based on their working experiences listed on their resumes. To determine whether a local guerrilla force had strong presence in the counties during the pre-Communist liberation period, we hand check various county gazettes (as of May 1948). County-Level Development Performance from 1952-1998. We examine various measures of development performance at the county level from 1952 to 1998. First, measures related to economic growth and other economic outcomes are gathered from “Statistical Information on 50 Years of Fujian Province” and “Regional 16 Economies in Fujian,” which covers the period from 1952 to 1998 for all 59 counties in Fujian. We use this data set to construct the average annual real growth rate of gross value of output for agriculture and industries, separately for 1952-1998 (the whole sample period) and for 1978-1998 (the post-reform period). For outcome measures related to the Great China Famine (1959-1961), we construct two measures. The first measure, which we refer to as famine control, following Meng, Qian and Yared (2015), is defined as the ratio of the number of surviving births (per year) in the county during the period 1959-1961 relative to the number of surviving births (per year) in the same county during the period 1954-1957, as observed in the 1% sample of the 1990 China Population Census [see formula (24)]. The higher this ratio, the more successful the county was at controlling the famine. We also examine the county-level data rates (deaths per thousand) during the Great China Famine. The county-level number of deaths are mainly collected from the population statistical books published by the provincial Statistics Bureaus in the 1980s. 10 Scope of Study. Our analysis is restricted to the 59 counties or county-level cities in Fujian province. In the administrative system in China, there are also prefecture level cities (such as Fuzhou, Xiamen, and Zhangzhou). Prefecture-level cities are often treated differently in the statistical yearbooks, with many of the key economic indicators only collected at the prefecture level (instead of the district level within the prefecture, which would be more comparable to counties or county-level cities). We also restrict our analysis up to 1998 when the Chinese government initiated extensive redrawing of the county boundaries, especially because the housing reform drastically blurred the rural-urban boundaries. 4.2 Descriptive Statistics Among the 59 counties (shorthand for counties, or county-level cities) in our study, 11 counties were led by cadres affiliated with the FA3, among which 3 had local guerrilla presence; 48 counties were led by cadres affiliated with YRD, among which 22 had local guerrilla presence. In this subsection, we provide some descriptive statistics. To control for the differences in the initial conditions across counties, we construct the following variables: (1) The log of the average agricultural and industrial output per capita in 1952 (respectively, in 1978), denoted by Ln_GVOPC52 (respectively, Ln_GVOPC78); and (2) The log of the total population in 1952 (respectively, in 1978), denoted by Ln_Pop52 (respectively, Ln_Pop78). We construct two variables to measure the county’s geography that may be relevant to economic performance: 11 10The data is compiled by Kasahara and Li (2018), where they provide the details about the data construction and cross-validation with the other provincial level death stastistics.We are grateful to Bingjing Li for generously sharing the data with us. 11We have also tried the distance to Fuzhou – the provincial capital – but it is never statistically significant, and we thus opted not to include it to avoid multicollinearity in light of the few observations that we have. Since Fujian is a coastal province, we have also tried 17 The first is the share of plain (%) in the county’s total land areas where flat land is defined as land with less than 15 degrees of incline. This captures the amount of land that can be used as science/industrial parks or for agriculture production; it also captures the amount of land that is suitable for development. The second is The distance to key FDI source – Taiwan – as proxied by the distance to Xiamen, the city right across the Taiwan Strait from Taiwan. The distance is calculated based on the Global Position System (GPS) position of the center of each county to the center of Xiamen City. This variable may also capture the access to the commercial hubs of Fujian province, and investment from Taiwan, as well as other potential benefits such as better access to business personnel and technology. Table 2 provides summary statistics of the variables used in our analysis. It shows that there are huge variations in the annual growth rates across counties. During the whole study period of 1952-1998, the annual real growth rate ranged from -0.03% to 7.94%, with a mean of 2.91% and a standard deviation of 1.47%; and during the post-reform period of 1978-98, the annual growth rate ranged from 1.47% to 22.49%, with a mean of 7.11% and a standard deviation of 3.89%. Similarly, the Famine Control variable we created in formula (24) also shows large variations across counties, with a mean of 0.78 and a standard deviation of 0.14. That is, during the famine period of 1959-1961, the drop in live births averaged to an astounding 22%. Similarly, the death rates during the Great Chinese Famine ranged from 5.8‰ to 33.8‰ with a mean death rate of 13.4‰ and a standard deviation of 5.8‰. 4.3 Factions, Local Accountability and Annual Growth We first compare the annual growth rates of real agriculture and industrial output per capita (also referred to as Gross Value of Output per Capita, or GVO per capita, henceforth) in counties led by YRD-affiliated cadres and those led by FA3-affiliated cadres. We also compare the annual growth rates in counties with local guerrilla presence prior to 1948 and those without guerrilla presence. We use the GDP deflator to compute the real growth rates. We have comprehensive data for all counties covered by our study from 1952 to 1998. To take into account important structural changes after the reform starting in 1978, we present the above comparisons both for the whole study period of 1952-1998 and only the post-reform period of 1978-1998. Figure 2 plots the estimated kernel density distributions of the annual growth rates of real agriculture and industrial real output per capita by factions (FA3 vs. YRD, the top panel) or by local accountability as proxied by guerrilla presence (GuerrillaYes vs. GuerrillaNo, the bottom panel). The growth rates for the whole period of 1952-1998 were plotted in the left panel and those for the post-reform period of 1978-1998 were plotted in the right panel. The two graphs in the top panel show that the distribution of the growth rates among counties led by YRD-affiliated cadres is more heavily weighted on higher values than those in the counties led by FA3-affiliated controlling the length of seashore within the county. Again, it is never significant; thus we exclude it. 18 Variable Obs. Mean SD Min Max Annual Growth Rate 52-98 (%) 57 2.91 1.47 -0.03 7.94 Annual Growth Rate 78-98 (%) 57 7.11 3.89 1.47 22.49 Famine Control 58 0.78 0.14 0.46 1.11 Death Rate (Death Per 1000) 58 13.4 5.8 5.8 33.8 FA3 59 0.19 0.39 0 1 YRD 59 0.81 0.39 0 1 Guerrilla 59 0.42 0.5 0 1 YRD×GuerrillaNo 59 0.44 0.5 0 1 YRD×GuerrillaYes 59 0.37 0.49 0 1 Ln_GVOPC_52 58 7.95 0.45 5.99 8.66 Ln_GVOPC_78 58 7.9 0.48 5.99 9.23 Ln_Pop_52 59 2.67 0.69 1.34 4.33 Ln_Pop_78 58 3.36 0.66 1.99 4.96 Share of Plain (%) 59 10.96 9.7 1.53 41 Distance to Xiamen (Km) 59 184.2 90.01 21 342 Table 2: Summary Statistics of Main Variables. cadres, particularly so in the post-reform period. The two graphs in the bottom panel show that the growth rates are higher, probabilistically, in counties that had local guerrilla presence than those that did not have guerrilla presence; and the difference is also much more striking during the 1978-1998 period. In Table 3, we summarize the differences in the average annual real GVO growth rates between the counties led by cadres affiliated with FA3 and YRD, or between counties with or without guerrilla presence for the period of 1952-1998 and 1978-1998 separately. In the left side of Panel A, we show that, during 1952 to 1998, the average real GVO growth rates for FA3 counties was 2.10 percent, that for YRD counties was 3.08 percent, and the difference between YRD and FA3 counties was 0.99 percentage points, which is 0.67 standard deviations (SDs) of the mean growth rate of 2.91 percent. The difference in the growth rates between YRD and FA3 counties is statistically significant at the 10 percent level. However, if we focus on 1978-1998, the difference is much more striking: the average annual real GVO growth rate for FA3 counties was 4.37 percent, and that for YRD counties was 7.72 percent. The 3.17 percentage points per year YRD advantage in growth rate (about 0.82 SDs), is statistically significant at 1.4% level. Needless to say, the difference in economic growth rates between these two types of counties was huge, especially in light of the fact that they are from the same province. In the right side of Panel A in Table 3, we examine the differences in growth rates by whether a county had guerrilla presence as of May 1948. The differences in the counties with and without guerrilla presence are also striking. The guerrilla-present counties had an advantage in growth of 0.89 percentage points (or 0.6 SDs) in the 19 Panel A: FA3 vs. YRD, and Guerrilla vs. No Guerrilla County by Growth Rate (%) Sample County by Growth Rate (%) Sample Faction 1952-1998 1978-1998 Guerrilla Presence 1952-1998 1978-1998 FA3 2.10 4.37 11 No 2.51 5.52 25 (0.86) (1.6) (1.16) (2.44) YRD 3.08 7.72 48 Yes 3.43 9.26 34 (1.54) (4.03) (1.72) (4.52) YRD-FA3 0.99* 3.17** Yes-No 0.89** 3.72** (0.5) (1.25) (0.38) (0.93) 20 Panel B: Interactions of FA3, YRD with Guerrilla Presence County by Growth Rate 1952-1998 (%) Growth Rate 1978-1998 (%) Sample Faction\Guerrilla Yes No Yes-No Yes No Yes-No Yes No FA3 2.05 2.11 0.06 5.82 4.08 1.7 3 8 (0.69) (0.97) (0.63) (0.125) (1.69) (1.01) YRD 3.63 2.65 0.98** 9.75 6.01 3.74** 22 26 (1.74) (1.18) (0.43) (4.63) (2.44) (1.07) Table 3: Comparisons of Growth Rates across Counties in Fujian, by Faction and Guerrilla Presence (1952-1998 and 1978-1998). N : Standard errors in parenthesis. *, ** and *** denote significance at 10, 5 and 1 percent, respectively. Figure 2 Kernel Density Estimation of Annual Growth Rate of Agriculture and Industrial Real Output per Capita by Power .25 .5 YRD YRD FA3 FA3 .4 .2 .15 .3 Density Density .2 .1 .05 .1 0 0 0 2 4 6 8 0 5 10 15 20 25 growth rate of ag and ind real output PC, 1952-1998 growth rate of ag and ind real output PC, 1978-1998 guerrillaYes guerrillaYes .15 .3 guerrillaNo guerrillaNo .1 .2 Density Density .05 .1 0 0 0 2 4 6 8 0 5 10 15 20 25 growth rate of ag and ind real output PC, 1952-1998 growth rate of ag and ind real output PC, 1978-1998 Note: Panel top left is the annual growth rate of agriculture and industrial real output Figure 2: Kernel Density of Annual Growth Rate of Real Agriculture and Industrial Output per Capita: YRD vs. per capita, 1952-1998 (YRD vs. FA3); Panel top right is the annual growth rate of agriculture FA3 (Top) and Guerrilla vs. and Guerrillareal No industrial output per (Bottom), capita, and 1978-1998 for the (YRD vs. FA3); period 1952-1998 Panel (Left) and 1978-1998 (Right). bottom left is the annual growth rate of agriculture and industrial real output per capita, 1952-1998 (GuerrillaYes vs. GuerrillaNo); Panel bottom right is the annual growth rate of agriculture and industrial real output per capita, 1978-1998 and 3.72 vs. (GuerrillaYes period of 1952-1998, GuerrillaNo) percentage points in 1978-1998 (or 0.95 SDs). Both differences are statistically significant at 5% level. In Panel B, we examine the differences in average annual real GVO growth in guerrilla versus no-guerrilla counties within the groups of counties led by FA3- or YRD-affiliated cadres. During 1952-1998 and within the group of FA3 counties, there is no difference in growth rates with respect to the local guerrilla presence. In contrast, within the group of YRD-affiliated counties, those with guerrilla presence outgrew those without by 0.98 percentage points (or 0.7 SDs). Similarly, during 1978-1998 and within the group of FA3 counties, we do not find statistically significant differences in annual GVO growth rates between those with and without guerrilla presence, though the grow rates remain higher in counties 3 / 31 that had guerrilla presence. In contrast, within the group of YRD counties, those with guerrilla presence outgrew those without by 3.74 percentage points (or 0.95 SDs). It should be noted that, as shown in the last two columns in Panel B, the numbers of counties in the FA3 group with and without guerrilla presence are small, 3 and 8, respectively. 4.4 Factions, Local Accountability and Famine Severity During 1959-1961 Our second piece of descriptive evidence on the impact of factions and local accountability on development performance is how counties fared during the Chinese Great Famine (1959-1961). An estimated total of 16.5 21 million (Coale, 1981) to 45 million (Dikotter, 2010) individuals, mostly rural residents, died or failed to be born in the three year period. We examine two measures of famine severity at the county level. The first measure follows Meng, Qian and Yared (2015) and proxies the famine severity as proxied by the birth cohort size of survivors observed in 1990 census. The logic for this measure, as Meng, Qian and Yared (2015) argued, is that “famine increases infant and early childhood mortality rates and lowers fertility rates such that a more severe famine results in smaller cohort sizes for those born shortly before or during the famine.” Specifically, we define our variable “famine control,” the opposite of “famine severity,” in a county as the ratio of the number of surviving births (per year) in the county during the period 1959-1961 relative to the number of surviving births (per year) in the same county during the period 1954-1957, as observed in the 1% sample of the 1990 China Population Census: Surviving Births per Year from 1959-1961 in County C Famine ControlC = . (24) Surviving Births per Year from 1954-1957 in County C The higher the measure, the less severe the famine was in county C . Meng, Qian and Yared (2015) point out that birth cohort size measure of famine severity is not vulnerable to the misreporting because it is less influenced by the government’s desire to understate famine severity. However, it potentially suffers from the measurement error caused by differential death rates after the famine period across counties. The second measure is the newly compiled county-level death rates, number of deaths per thousand (see Section 4). Figure 3 plots the kernel density estimation of the distribution of the famine control proxied by birth cohorts (top panel) and death rates (bottom panel) among counties with FA3- and YRD-affiliated cadres (left), and counties with or without guerilla presence (right). The top panel of Figure 3 demonstrates that counties led by FA3-affiliated cadres had more severe famine as measured by birth cohort size than counties led by YRD-affiliated cadres; and it also shows that counties with guerrilla presence had lower famine severity. The bottom panel of Figure 3 shows that the death rates during the Great Chinese Famine is lower in counties led by YRD-affiliated cadres than in counties led by FA3-affiliated cadres; and similarly, they are lower in counties with guerrilla presence than those without. In Table 4, we compare the two famine severity measures, famine control and death rates in 1959-1961, between the counties led by FA3- and YRD-affiliated counties, and the counties with and without guerrilla presence (Panel A), as well as the interaction effects between FA3 v.s YRD factions and guerrilla presence on famine severity (Panel B). In the left side of Panel A, we find that YRD counties had significantly higher level of famine control by 0.11 (or 0.8 SDs), a large effect. We also find that counties led by YRD-affiliated cadres had a significantly lower death rate by 7.6 per thousand people, or 1.3 SDs, than counties led by FA3-affiliated cadres. In the right side of Panel A, we find that the counties with guerilla presence as of May 1948 also had a higher levels of famine control measure than counties without guerrilla presence, though the difference is not statistically significant. The death rates in counties with guerrillas is significantly lower by 2.65 per thousand people, or 0.46 22 Figure 3 Kernel Density Estimation of Famine Control during 1959-1960 by Power 3 4 YRD guerrillaYes FA3 guerrillaNo 3 2 Density Density 2 1 1 0 0 .4 .6 .8 1 1.2 .4 .6 .8 1 1.2 Famine Control Famine Control .15 YRD guerrillaYes FA3 guerrillaNo .1 .1 Density Density .05 .05 0 0 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 Death Rate Death Rate Note: Panel left is the famine control during 1959-1960 by Power(YRD vs. FA3); Figure 3: Panel right Kernel Density is the of the famine Famine control during Control 1959-1960 as Proxied (GuerrillaYes by Birth Cohorts vs. GuerrillaNo) (Top Panel) and Death Rates (Bottom Panel) at the County Level During the Great Chinese Famine of 1959-1961: YRD vs. FA3 (Left) and Guerrilla vs. No Guerrilla (Right). SDs, than those in counties without guerrilla presence. In Panel B of Table 4, we show the interaction effects of faction and local accountability. The left side of Panel B shows that, within FA3-affiliated counties, there is no statistically significant difference in both famine control and death rates between counties with and without guerrilla; however, within the group of YRD-affiliated counties, those with guerrilla presence had statistically significant higher famine control, and statistically significant lower death rates, than those without guerrilla presence. We interpret the better famine control and lower death rates during the Great Chinese Famine in counties led by YRD-affiliated counties and in counties with local guerrilla presence as evidence that the leaders in these counties undertook more pro-local policies during the Great Famine period. As has been pointed out in the 4 / 29 literature (Meng, Qian and Yared, 2015; and Fan, Xiong and Zhou, 2016), local officials’ incentives to inflate their grain production to meet the procurement quota from the central government played an important role in the magnitude of famine. The argument in our paper is that the local officials are driven by their incentives for political survival, which is in turn intimately related to their factional affiliations with the provincial level governments and their local accountability. County leaders whose political survival depends on grassroot support would tend to adopt policies that are more likely to blunt the devastating impact of the famine. Finally, we would like to point out that, while Figures 2 and 3 convey similar messages as Tables 3 and 4 respectively, the figures do highlight that the better development performance of YRD-affiliated counties over 23 Panel A: FA3 vs. YRD, and Guerrilla vs. No Guerrilla County Famine Control Death Rate County Famine Control Death Rate FA3 0.688 19.7 No Guerrilla 0.754 14.57 (0.158) (2.6) (0.153) (5.36) Obs. 11 10 25 25 YRD 0.798 12.1 Yes Guerrilla 0.807 11.92 (0.132) (0.6) (0.125) (6.08) Obs. 48 48 34 33 YRD-FA3 0.11** -7.6*** Yes-No 0.054 -2.65* [0.046] [1.75] [0.038] [1.51] Panel B: Interactions of FA3, YRD with Guerrilla Presence Famine Control Death Rates County Guerrilla Presence Guerrilla Presence Yes No Yes-No Yes No Yes-No FA3 0.56 0.73 -0.17 24.3 17.8 6.5 (0.09) (0.16) [0.097] (9.5) (7.5) [5.55] Obs. 3 8 3 7 YRD 0.84 0.76 0.08** 10.2 13.7 -3.47*** (0.09) (0.15) [0.037] (2.99) (4.4) [1.11] Obs. 22 26 22 26 Table 4: Comparisons of Famine Control and Death Rates (per 1000) Across Counties in Fujian, by Faction and Guerrilla Presence, 1959-1961. N : The numbers in parenthesis are standard deviations, and the number in square brackets are standard errors. *, ** and *** denote significance at 10, 5 and 1 percent, respectively. 24 FA3-affiliated counties, and counties with guerrilla presence over those without guerrilla presence, are not due to a few outlier; rather, the differences occur in the whole distributions. 5 Empirical Results The descriptive evidence in Section 4 is suggestive that the drastic differences in economic growth and the famine severity in 1959-1961 are related to the differences in the policy choices made by the local leaders from the weaker YRD faction and those from the stronger FA3 faction, or by the leaders in counties with and without guerrilla presence. In this section, we present more systematic empirical evidence of this connection in order to account for the possible differences in other factors, including geographical factors and the pre-l952 differences in economic conditions, that may have played a role in accounting for the results. 5.1 Initial Assignment: Testing for Randomness Before we move on to examine the effect of faction affiliation and local accountability , we examine how YRD counties and guerrilla counties differ from other counties in terms of basic characteristics. This would shed light on what variables we should control for, and whether selection issues are serious. In Table 5, we present simple linear probability model of the YRD and Guerrilla dummy to basic county characteristics such as initial income (measured at 1952), population (measured at 1952), and geography. 12 The idea is that if YRD and Guerrilla statuses are orthogonal to the key observables, they are also likely to be orthogonal to the unobservable residual (Altonji, Elder, and Taber, 2000). We also present multinomial Logit model to explain the three states of FA3, YRD without guerrilla, and YRD with guerrilla (with the default category being FA3). 13 The OLS regression results, reported in the first two columns in Table 5, suggest that YRD counties had similar initial income and population, but they tended to have somewhat higher share of plain areas. On average the distance to Xiamen (and Taiwan) does not differ much between YRD and FA3 counties. Counties with guerrilla presence tended to be more populous counties, but there is no systematic difference between counties with and without guerrilla presence in terms of initial income and geography in general. In the last two columns in Table 5, we report the multinomial Logit results where the default group is FA3. It again suggests that neither initial income nor geography matters for which of the three groups, YRD with Guerrilla, YRD without Guerrilla and FA3, would be assigned to a county in 1950. The only variable that is significant at the 10 percent level is initial county population. To summarize the results in Table 5, we think it is credible, at least as a first pass, to treat the county’s assignment to YRD- or FA3-affiliated cadres, or the assignment to having 12Our income and population variables are dated in 1952, the earliest number we can find. Ideally we would like to have values that precede 1949. 13We do not distinguish FA3 with and without guerrilla presence because we have very limited number of observations (i.e., 11) for FA3 counties. 25 OLS Multinomial Logit YRD Guerrilla YRD_GuerrillaNo YRD_GuerrillaYes Initial Conditions Ln_GVOPC_52 0.052 -0.048 0.586 0.317 (0.565) (0.651) (0.579) (0.776) Ln_Pop_52 0.053 0.303*** -0.122 1.374* (0.472) (0.000) (0.870) (0.100) Geography Share of Plain (%) 0.009* 0.001 0.062 0.084 (0.076) (0.891) (0.331) (0.183) Distance to Xiamen (KM) 0.001* -0.001 0.008 0.005 (0.067) (0.353) (0.123) (0.353) Constant -0.012 0.126 -5.534 -7.334 (0.987) (0.897) (0.558) (0.468) Obs. 58 58 58 R2 (Pseudo- R2 ) 0.07 0.253 0.159 Table 5: Correlations of the Factions of Local Leaders in 1949 and Guerrilla Presence with County Characteristics in 1952. N : FA3 is the reference group in multinomial logit regression. White Standard Errors are in parenthesis. ***, **, and * respectively indicate 1%, 5% and 10% statistical significance. guerilla presence or not, as close to random, especially after we control for the initial conditions and geography variables. In Section 5.4, we will further address this issue by focusing only on bordering counties. 5.2 Factions, Local Accountability and Economic Growth In this section, we report our regression results on the effect of factions (as captured by YRD vs. FA3), local accountability (as captured by Guerrilla or no Guerrilla) and their interactions on local annual economic growth rates. In Table 6, we examine the effect of YRD vs. FA3 factions on county level annual growth rates, for the whole period of 1952-1998 and for the post-reform period of 1978-1998 separately. We present the results for the full sample, as well as the results when we trim the 5% of the tail observations. In Columns (1) and (4), we do not control for any variables, and the estimates of YRD coefficients simply replicate the raw differences in annual growth rates between YRD counties and FA3 counties we reported in the left side of Panel A in Table 3: YRD-affiliated counties on average grew by 0.99 percentage points faster than FA3-affiliated counties during the whole sample period of 1952-1998, and by 3.17 percentage points faster in the post-reform period of 1978-1998. 26 Full Sample Trimming Tail 5% 1952-1998 1978-1998 1952-1998 1978-1998 (1) (2) (3) (4) (5) (6) (7) (8) Power Structure YRD 0.987*** 0.925*** 0.907*** 3.166*** 2.103*** 2.102*** 0.767*** 2.007*** (0.004) (0.004) (0.003) (0.000) (0.003) (0.002) (0.008) (0.000) Initial Conditions Ln_GVOPC_52 -1.876*** -2.127*** -1.985*** (0.000) (0.000) (0.000) Ln_GVOPC_78 -4.634*** -4.506*** -3.358*** (0.000) (0.000) (0.000) Ln_Pop_52 0.217 -0.417 -0.193 27 (0.464) (0.136) (0.349) Ln_Pop_78 0.839 -0.842 -0.864 (0.299) (0.282) (0.208) Geography Share of Plain (%) 0.055** 0.143*** 0.046* 0.081* (0.015) (0.007) (0.064) (0.06) Distance to Xiamen -0.004*** -0.010*** -0.004** -0.012*** (Km) -0.008 (0.005) (0.012) (0.001) Obs. 57 57 57 57 57 57 53 52 R2 0.049 0.403 0.58 0.089 0.49 0.645 0.432 0.528 Table 6: Effects of YRD vs. FA3 Factions on Annual Growth Rates. N : White Standard Errors are in parenthesis. Intercepts are not reported. *, **, *** respectively indicate 10%, 5% and 1% statistical significance. In Columns (2) [respectively, Column (5)], we also add controls for the log of the initial GVO per capita, and initial population size, in 1952 [respectively, in 1978]. In Columns (3) and (6) we further control for some measures of the county geography and the distance to Xiamen. Finally, in columns (7) and (8), we trim the tail 5% of the dependent variable to ensure that our results are not driven by outliers. The results are remarkably robust. Let us discuss Columns (3) and (6) where we include all the controls of initial economic condition, population and geography variables. Focusing on the 1952-1998 annual growth rate, we find that relative to FA3 counties, YRD counties exhibited an advantage of 0.91 percentage points per annum. In Column (7), we find that YRD advantage stays at 0.77 percentage points per annum even after we trim 5% of the tails. Focusing on the post-reform period of 1978-1998, the advantage of the YRD counties is estimated to be more than 2.1 percentage points per annum, and this advantage does not shrink much as we trim the 5% outliers. Table 6 also shows that initial income is associated with lower growth rate, which is consistent with conditional convergence hypothesis; that having a higher share of plain area is associated with higher growth and a closer proximity to Xiamen (and Taiwan) is associated with higher growth rate. 14 We now turn to local accountability, proxied by the presence or absence of guerrilla forces in the county as of May 1948. In counties of strong guerrilla presence, the local grassroots were more organized, and thus they were more able to act as a force to be reckoned with for the survival of the local leaders. As a result, and as modelled earlier, the local leaders in counties with guerrilla presence are likely to adopt more pro-local economic policies. Otherwise similar to in Table 7, in Table 7 we add the dummy of local Guerrilla presence and study the separate effects of YRD and guerrilla presence on local growth rates. We find that both YRD and Guerrilla dummies have statistically significant and positive effects on annual growth rates. Focusing on the results in columns (3) and (6), where we include the richest set of controls with the full sample, we find that YRD counties had a 0.86 percentage point higher average annual growth rate in 1952-1998 (or 0.6 SD), and 1.929 percentage points (or 0.5 SDs) in 1978-1998, than FA3 counties; and the effect of Guerrilla presence is also strong though slightly less pronounced: by 0.52 percentage point (or 0.35 SDs) in 1952-1998, and 1.52 percentage points (or 0.4 SDs) in 1978-1998. The results remain statistically significant and quantitatively similar when we trim 5% of tail counties. 14Reducing the distance to Xiamen by one SD (90KM) is associated with an increase in growth rate by 0.9 percentage point. 28 Full Sample Trimming Tail 5% 1952-1998 1978-1998 1952-1998 1978-1998 (1) (2) (3) (4) (5) (6) (7) (8) Power Structure YRD 0.867** 0.879*** 0.857*** 2.537*** 1.981*** 1.929*** 0.741** 1.890*** (0.015) (0.007) (0.004) (0.000) (0.004) (0.002) (0.012) (0.000) Guerrilla 0.818** 0.642** 0.524** 3.420*** 1.652** 1.517** 0.472* 1.488** (0.036) (0.035) (0.048) (0.000) (0.026) (0.018) (0.074) (0.010) Initial Conditions Ln_GVOPC_52 -1.860*** -2.096*** -1.975*** (0.000) (0.000) (0.000) Ln_GVOPC_78 -4.278*** -4.150*** -2.973*** (0.000) (0.000) (0.000) 29 Ln_pop_52 -0.001 -0.576** -0.337 (0.997) (0.045) (0.116) Ln_pop_78 0.358 -1.248 -1.247* (0.671) (0.112) (0.077) Geography Share of Plain (%) 0.055** 0.149*** 0.047** 0.091** (0.010) (0.003) (0.049) (0.016) Distance to Xiamen -0.004** -0.009*** -0.003** -0.010*** (Km) (0.014) (0.009) (0.019) (0.001) Obs. 57 57 57 57 57 57 53 52 R2 0.110 0.430 0.598 0.266 0.515 0.668 0.459 0.571 Table 7: Separate Effects of YRD and Guerrilla on Annual Growth Rates. N : White Standard Errors are in parenthesis. Intercepts are not reported. *, **, *** respectively indicate 10%, 5% and 1% statistical significance. Full Sample Trimming Tail 5% 1952-1998 1978-1998 1952-1998 1978-1998 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Power Structure FA3×GuerrillaYes 0.058 0.376 (0.453) (0.852) YRD×GuerrillaNo [b0 ] 0.551 0.620* 0.674** 0.690* 1.458** 1.352* 1.455** 1.547* 0.535* 1.372*** (0.118) (0.055) (0.029) (0.410) (0.034) (0.054) (0.032) (0.818) (0.062) (0.002) YRD×GuerrillaYes [b1 ] 1.528*** 1.424*** 1.288*** 1.306*** 5.198*** 3.422*** 3.180*** 3.289*** 1.160*** 3.086*** (0.001) (0.000) (0.000) (0.455) (0.000) (0.000) (0.000) (0.959) (0.001) (0.000) Initial Conditions Ln_GVOPC_52 -1.855*** -2.098*** -2.097*** -1.968*** (0.000) (0.000) (0.240) (0.000) Ln_GVOPC_78 -0.576** -4.325*** -4.243*** -3.089*** (0.294) (0.000) (0.000) (0.000) 30 Ln_pop_52 -0.028 -0.572** -4.210*** -0.345* (0.929) (0.042) (0.980) (0.096) Ln_pop_78 0.272 -1.245 -1.261 -1.252* (0.745) (0.108) (0.795) (0.075) Geography: Share of Plain (%) 0.052** 0.053** 0.140*** 0.142*** 0.044* 0.084** (0.016) (0.023) (0.005) (0.051) (0.063) (0.027) Distance to Xiamen -0.004** -0.004** -0.009*** -0.009*** -0.003** -0.010*** (Km) (0.010) (0.002) (0.007) (0.003) (0.016) (0.001) Obs. 57 57 57 57 57 57 57 57 53 52 R2 0.127 0.444 0.603 0.595 0.267 0.528 0.672 0.666 0.479 0.580 H 0 : b0 = b1 0.033 0.024 0.049 0.051 0.002 0.018 0.023 0.024 0.042 0.015 Table 8: The Interaction Effects of Factions and Guerrilla Presence on Annual Growth Rates. N : (1). White Standard Errors are in parenthesis. (2). Intercepts are not reported. (3). *, **, *** respectively indicate 10%, 5% and 1% statistical significance. (4) The last row reports the p-values of the hypothesis that b0 = b1 . In Table 8, we examine the interaction effects of factions and local accountability, and ask whether the YRD effect hinges on the presence of local guerrilla forces. We include YRD×GuerrillaYes and YRD×GuerrillaNo. 15 If there is an additional boost of being led by cadres from the weak YRD faction in counties with strong local accountability as proxied by local guerrilla presence, we expect to see the effect of YRD×GuerrillaYes to be significantly larger than YRD×GuerrillaNo. We find this to be true, and the results are robust. Focusing on Columns (3) and (7), counties in which the local leaders were affiliated with YRD faction but without guerrilla participation had 0.67 percentage points (or 0.45 SDs) higher growth rate in 1952-1998, and 1.46 percentage points (or roughly 0.4 SDs) higher in 1978-1998, than FA3 counties. In contrast, counties with YRD faction and guerrilla presence had 1.29 percentage points (or 0.9 SD) higher growth rate in 1952-1998, and 3.18 percentage points (or 0.8 SDs) higher growth rate in 1978-1998, than FA3 counties with no guerrilla presence. The Wald tests for the hypothesis that the coefficients for YRD×GuerrillaYes and YRD×GuerrillaNo terms are equal is rejected at 5.1% and 2.4% significance level respectively in Column (3) and (7). Thus, being led by cadres who lack political support in upper-level government could imply faster local economic growth, but when coupled with local accountability, its effects more than double. 5.3 Factions, Local Accountability and Famine Severity in 1959-1961 So far we have found a strong and robust association between faction of the local leaders and local account- ability and local economic growth, and offered evidence that counties with stronger local accountability and with leaders from weaker factions (in terms of weaker connection to the upper-level government) tended to have significantly higher growth rate in the post-reform period and in the five decades after the founding of the People’s Republic of China. However, the very nature of the empirical exercise suggests that it is almost impossible to establish impeccable causality. In Tables 6-8 above, we have controlled for geography and initial conditions; and in Table 5, we have shown that there does not seem to be significant correlations of the YRD or Guerrilla status of a county with the county’s observables in terms of initial economic conditions and geography. But the concern is that there may still be other omitted factors that drive our findings. In this section, we attempt to provide additional evidence to buttress the case for the causal impact of factions and local accountability on economic performance. In particular, we examine whether counties led by leaders affiliated with weak factions and with stronger local accountability led to better protection of local residents in the Great Chinese Famine (1959-1961). The Great Chinese Famine was an important event in modern Chinese history where roughly 30 million people died due to a combination of natural disaster and radical policies implemented by the central government (Coale, 1981; Dikotter, 2010). While one of the causes for the Great Chinese Famine was no doubt natural disaster, the 15In principle, we may also consider including FA3×GuerrillaYes. However, only 3 counties fit this category. Indeed, we have tried to include this term in the regression [see Columns (4) and (8) in Table 8], and found the estimate to be, not surprisingly, statistically insignificant. 31 Full Sample Trimming Tail 5% Bottom 25%∗ (1) (2) (3) (4) (5) Power Structure YRD 0.109** 0.116** 0.119** 0.087 -0.419** (0.050) (0.054) (0.056) (0.053) (0.182) Initial Conditions Ln_GVOPC_52 0.033 0.028 0.031 -0.081 (0.036) (0.039) (0.038) (0.101) Ln_pop_52 0.048** 0.041 0.032 -0.152* (0.022) (0.026) (0.024) (0.076) Geography Share of Plain 0.000 0.001 0.002 (%) (0.002) (0.002) (0.007) Distance to Xiamen -0.000 -0.000 0.001 (Km) (0.000) (0.000) (0.001) Obs. 58 57 57 53 58 R2 0.075 0.125 0.095 0.089 0.172 Table 9: Effect of Factions on Famine Control during the Great Chinese Famine 1959-1961 in Fujian. N : White standard errors are in parenthesis. Intercept not reported. In Column (5), the dependent variable is a dummy variable, which take the value of 1, if the county was in the bottom 25 percent of famine control (25 percent counties suffering the most severe famine during 1959-1961), and take the value of 0, otherwise. 32 literature suggests that the key determinant was likely the implementation of radical policies to procure grains for rapid industrialization that was favored by Mao Zedong, the top leader of China then. For instance, the incidence of famine was higher in grain-rich areas (Meng, Qian and Yared, 2015), which implies that it was grain procurement rather than production that caused famine. Indeed, Li and Yang (2005), Meng, Qian and Yared (2015) argue that the key factors behind the famine were the inelastic and regressive procurement policies. Yang (1996) and Kung and Lin (2003) offer evidence that radical policies were a key reason behind the widespread famine in China even though the average grain output was sufficient to support the whole population. 16 Kung and Chen (2011) provide evidence that provincial leaders with stronger career incentives (i.e., alternative members of the central committee versus full central committee members) were more radical in implementing the grain procurement policy despite the large scale of the natural disaster. Those alternative members had higher upside for their career movements and thus worked harder to try to please Mao by procuring grains aggressively. Fan, Xiong and Zhou (2016) argued that inflation of the production by local officials was partly responsible for the severity of the famine. This literature thus far has focused on either the role of provincial leaders such as their career incentives, or the policies at the central level (such as the grain procurement policies). We complement this literature by zeroing in on the incentives of the county level cadres, the role of local accountability, and the connection between the provincial and the county level cadres. Famine Control (Birth Rates). In Table 9 we relate Famine Control at the county level to whether a county’s leader was affiliated with the weaker faction YRD, with the same sequences of controls as in the case of annual growth rates reported in Table 5. The results on the effect of YRD are very robust, with the coefficient estimates ranging from 0.11 to 0.12 when outliers are not trimmed. When outliers (i.e., tail 5%) are trimmed, however, the YRD coefficient becomes statistically insignificant, though the magnitude is similar. These results thus suggests that counties with YRD-affiliated leaders reduced the severity of the famine as measured by famine control is likely to operate by reducing the incidence of extreme famine. 16Fan and Shi (2013) offer evidence that the Great Leap Forward industrialization movement and the details of the procurement policies contributed to the famine. 33 Separate Effects Interaction Effects Full Sample Trimming Tail 5% Bottom 25%∗ Full Sample Trimming Tail 5% Bottom 25%∗ (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Power Structure YRD 0.101* 0.114** 0.117** 0.077 -0.422** (0.055) (0.056) (0.058) (0.055) (0.182) Guerrilla 0.041 0.021 0.018 0.052 0.023 (0.036) (0.040) (0.042) (0.037) (0.111) YRD×GuerrillaNo [b0 ] 0.072 0.091 0.094 0.055 -0.369* (0.056) (0.057) (0.060) (0.055) (0.189) YRD×GuerrillaYes [b1 ] 0.152*** 0.154*** 0.157*** 0.134** -0.499** (0.050) (0.056) (0.058) (0.056) (0.189) Initial Conditions Ln_GVOPC_52 0.033 0.029 0.032 -0.079 0.034 0.031 0.034 -0.086 (0.037) (0.039) (0.038) (0.102) (0.038) (0.039) (0.038) (0.102) 34 Ln_pop_52 0.041 0.036 0.017 -0.159** 0.029 0.027 0.012 -0.120 (0.025) (0.029) (0.024) (0.076) (0.024) (0.028) (0.023) (0.078) Geography Share of Plain (%) 0.000 0.001 0.002 -0.000 0.001 0.003 (0.002) (0.002) (0.007) (0.002) (0.002) (0.007) Distance to Xiamen -0.000 -0.000 0.001 -0.000 -0.000 0.001 (Km) (0.000) (0.000) (0.001) (0.000) (0.000) (0.001) Obs. 58 57 57 53 58 58 57 57 53 58 Adjusted- R2 0.079 0.113 0.080 0.104 0.156 0.126 0.143 0.110 0.144 0.172 H0 : b0 = b1 0.031 0.136 0.142 0.035 0.238 Table 10: The Separate and Interaction Effects of Factions and Guerrilla Presence on Famine Control during the Great Chinese Famine in 1959-1961. N : (1). White Standard Errors are in parenthesis. (2). Intercepts are not reported. (3). *, **, *** respectively indicate 10%, 5% and 1% statistical significance. (4) The last row reports the p-values of the hypothesis that b0 = b1 . (5) In Columns (5) and (10), the dependent variable is a dummy variable, which take the value of 1, if the county was in the bottom 25 percent of famine control (25 percent counties suffering the most severe famine during 1959-1961), and take the value of 0, otherwise. To check this, we create an indicator variable which takes value 1 if a county’s famine control measure as constructed by formula (24) is in the bottom quartile among all counties. This indicator explicitly captures the possibility of extreme local famine. The results in Column (6) of Table 9 show that YRD counties had 42 percent lower chance to be in the bottom quartile of famine control among all counties, suggesting that being led by cadres from the weaker YRD faction was very effective in containing the famine disasters. In Table 10 we further allow the guerrilla presence in the county to play a role. In the left panel, we let YRD-affiliation and guerrilla presence to have separate effects; and in the right panel, we let YRD to interact with guerrilla presence. When Guerrilla presence and YRD factions are included separately, YRD matters much more than guerrilla presence, with the former being largely statistically significant, while the latter is never statistically significant. Since the procurement policy was a top-down mandate, it is not surprising that YRD affiliation would be more important in reducing famine since its cadres served as the link between the provincial government and the county governments, and thus could have a larger impact on how much grain was procured. Local guerrilla cadres did not have direct link to the provincial government, and thus could hardly directly push back the central mandate. Similar to Table 9, Columns (4) and (5) suggest that YRD-affiliated counties were able to largely reduce the extreme famine consequences. In Columns (6)-(10), we allow YRD’s effect to hinge on whether the county had guerrilla presence. Columns (6)-(8) show that counties with guerrilla presence led by cadres affiliated with the weak faction (i.e., belonging to the weaker YRD faction) achieved much better famine control with the coefficient around 0.15 to 0.17. In contrast, YRD counties without guerrilla presence did not have significantly better famine control, though the magnitude remain consistently positive. In Column (10) we focus on whether a county had a disastrous famine control (i.e., in bottom quartile). We find that it is 36.9% (49.9%, respectively) less likely for YRD counties without Guerrilla presence (with Guerrilla presence, respectively) than for FA3 counties to be in this extreme famine severity scenario. Death Rates. The second measure of famine severity we examine is the county-level average death rates (per thousand) during the period of 1959-1961. In Table 11, we present the estimates of the impact of faction on death rates. In the specification reported in Columns (4), which controls for the county’s initial conditions and geography variables, we find that counties led by YRD-affiliated cadres have 7.81‰ (about 1.4 SD) lower death rates, and it is statistically significant at 1% level. In Columns (5) and (6), we find that YRD-affiliation is associated with lower local death rates both at the extreme end of the distribution and the middle part of distribution: trimming the 5% of the tail observations, we find that having YRD-affiliated cadres is still associated with 4.2 ‰ lower death rates; and YRD-affiliation reduces the likelihood of being the most impacted county (top quartile in terms of death rates) by 39%. In Table 12, we add guerrilla presence in the county into our analysis of the death rates. Similar to Table 10, we include guerrilla presence both separately from YRD-affiliation [Columns (4)-(5)] and interacted with YRD 35 Full Sample Trimming Tail 5% Top 25%* (1) (3) (4) (5) (6) Power Structure YRD -7.611*** -7.925*** -7.807*** -4.228** -0.391** (2.576) (2.853) (2.883) (1.860) (0.195) Initial Conditions Ln_GVOPC_52 -2.095* -1.488 -0.366 -0.042 (1.199) (1.036) (0.736) (0.101) Ln_pop_52 -3.077*** -1.711* -1.423 -0.131 (0.765) (0.873) (0.887) (0.085) Geography Share of Plain (%) -0.105** -0.089** 0.002 (0.045) (0.038) (0.004) Distance to Xiamen 0.009* 0.010** 0.002** (Km) (0.005) (0.005) (0.001) Obs. 58 57 57 52 57 Adjusted R2 0.238 0.376 0.408 0.394 0.202 Table 11: The Effect of Factions on the Death Rates During the 1959-1961 Chinese Famine. N : (1). White Standard Errors are in parenthesis. (2). Intercepts are not reported. (3). *, **, *** respectively indicate 10%, 5% and 1% statistical significance. [Columns (6)-(10)]. When guerrilla presence is included in the regression separately, we find that its coefficients are not statistically significant, while the coefficient estimates on YRD remain negative and significant at around ˙ -7.8‰When we interact guerrilla presence with YRD, we find that both YRD counties with and without guerrilla presence have statistically significant lower death rates than FA3-affiliated counties. There is also strong evidence that YRD-affiliation joint with guerrilla presence reduce the probability of the county experiencing extreme famine severity (in terms of the county’s death rates being among the highest quartile among all counties) by about 40%. 36 Separate Effects Interaction Effects Full Sample Trimming Tail 5% Top 25%* Full Sample Trimming Tail 5% Top 25%* (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Power Structure YRD -7.298** -7.886*** -7.798** -4.214** -0.395** (2.769) (2.916) (2.936) (1.915) (0.196) Guerrilla -1.982 -0.433 -0.077 -0.108 0.032 (1.350) (1.330) (1.310) (0.985) (0.117) YRD×GuerrillaNo [b0 ] -6.021** -7.093** -7.173** -3.896** -0.380* (2.673) (2.861) (2.903) (1.881) (0.203) YRD×GuerrillaYes [b1 ] -9.491*** -9.045*** -8.672*** -4.729** -0.407* (2.606) (2.888) (2.937) (1.946) (0.206) Initial Conditions Ln_GVOPC_52 -2.101* -1.492 -0.372 -0.040 -2.131* -1.559 -0.417 -0.043 (1.204) (1.038) (0.754) (0.102) (1.191) (1.063) (0.753) (0.103) 37 Ln_pop_52 -2.925*** -1.688* -1.393 -0.141 -2.435*** -1.311 -1.218 -0.124 (0.830) (0.914) (0.920) (0.092) (0.851) (0.935) (0.930) (0.094) Geography Share of Plain (%) -0.105** -0.089** 0.002 -0.098** -0.085** 0.002 (0.046) (0.038) (0.004) (0.046) (0.039) (0.005) Distance to Xiamen 0.009* 0.010** 0.002** 0.009 0.010* 0.002** (Km) (0.005) (0.005) (0.001) (0.005) (0.005) (0.001) Obs. 58 57 57 52 57 58 57 57 52 57 R2 0.254 0.365 0.396 0.381 0.187 0.302 0.384 0.408 0.389 0.186 H0 : b0 = b1 0.002 0.077 0.145 .359 0.819 Table 12: The Separate and Interaction Effects of Factions and Guerrilla Presence on on Death Rates during the Great Chinese Famine in 1959-1961. N : (1). White Standard Errors are in parenthesis. (2). Intercepts are not reported. (3). *, **, *** respectively indicate 10%, 5% and 1% statistical significance. (4) The last row reports the p-values of the hypothesis that b0 = b1 . (5) In Columns (5) and (10), the dependent variable is a dummy variable, which take the value of 1, if the county was in the bottom 25 percent of famine control (25 percent counties suffering the most severe famine during 1959-1961), and take the value of 0, otherwise. 1952-1998 1978-1998 (1) (2) (3) (4) (5) (6) YRD 0.972*** 0.684** 1.961** 1.545** (0.351) (0.309) (0.782) (0.635) Guerrilla 0.468* 1.374** (0.277) (0.670) YRD×GuerrillaNo [b0 ] 0.465 0.980 (0.309) (0.657) YRD×GuerrillaYes [b1 ] 1.058*** 2.645*** (0.364) (0.820) Other Control Variables Yes Yes Yes Yes Yes Yes Obs. 15 51 51 15 51 51 Adjusted- R2 0.233 0.589 0.598 0.424 0.662 0.669 H0 : b0 = b1 0.071 0.037 Table 13: Effects of Power Structure on Growth Rates in Fujian, Border-Sharing Counties Only. N : White standard errors are in parenthesis. Intercept not reported. Columns (1) and (4) use subsample of FA3 counties and their borders sharing counties. Columns (2) and (5) use subsample of guerrilla counties and their borders sharing counties. Other control variables include Ln_GVOPC_52 (Ln_GVOPC_78), Ln_pop_52 (Ln_pop_78), Share of Plain (%) and Distance to Xiamen. 5.4 Robustness Checks: Border Counties Some may argue that, given the uneven distribution of YRD and FA3 counties over the province, the difference in development outcomes we have so far documented may reflect difference in other unobserved differences in geography or aspects not captured by our limited controls such as the share of plain in the county and the distance to Xiamen (and therefore Taiwan). That is, the concern is that such controls are unlikely to be sufficient. A useful robustness check would thus be to hold these unobserved geography or other aspects as constant as possible. To this end, when we estimate the effect of YRD, we only keep FA3 and YRD counties that are neighbors. Since counties are geographically small—there are almost 3000 counties in China now—neighboring counties tend to be similar in geography (and culture). This stringent control results in a much smaller sample of 15 counties (versus 57 counties in the full sample). When we estimate the effect of guerrilla (and its interactive effect with YRD), we only keep counties that had guerrilla and their neighboring counties. The restricted sample has 51 counties. Since our results are quite robust across various sets of controls, we present the specification with the comprehensive controls. The results on annual growth rates when restricted to the border counties are presented in Table 13, and those on famine control and death rates are in Table 14. In Columns (1) and (4) in Table 13, we examine the effect of factions (measured by the YRD dummy) on border-sharing counties with observations from only 15 border-sharing counties. We find that the results, both 38 Famine Control Death Rates (1) (2) (3) (4) (5) (6) YRD 0.176** 0.171*** -3.549 -10.545*** (0.062) (0.060) (4.362) (3.201) Guerrilla 0.018 -0.335 (0.040) (1.273) YRD×GuerrillaNo [b0 ] 0.151** -10.050*** (0.063) (3.221) YRD×GuerrillaYes [b1 ] 0.200*** -11.154*** (0.060) (3.173) Other Control Variable Yes Yes Yes Yes Yes Yes Obs. 15 51 51 14 51 51 Adjusted- R2 0.485 0.171 0.191 0.501 0.566 0.571 H0 : b0 = b1 0.244 0.271 Table 14: The Effects of Power Structure on Famine Control and Death Rates during the 1959-1961 Chinese Famine in Fujian, Border-Sharing Counties Only. N : White standard errors are in parenthesis. Intercept not reported. Columns (1) and (4) use subsample of FA3 counties and their borders sharing counties. Columns (2) and (5) use subsample of guerrilla counties and their borders sharing counties. Other control variables include Ln_GVOPC_52 (Ln_GVOPC_78), Ln_pop_52 (Ln_pop_78), Share of Plain (%) and Distance to Xiamen. qualitatively and quantitatively, are very similar to our results from the full sample. For instance, the YRD effect on annual growth rates in the period 1978-1998 is now 1.96 percentage point (vs 2.10 percentage point in Column 6 of Table 6), and on the growth rates in the whole period of 1952-1998 is 0.97 percentage point (vs. 0.91 percentage point in Column 3 of Table 6). This is quite remarkable, since the estimation sample is now only a quarter of the previous, already quite small, sample. In Columns (2) and (5), we use the 51 neighboring counties with and without guerrilla in the estimation. When we examine the effect of both YRD-affiliation and guerrilla presence, again we find the results qualitatively and quantitatively similar for the growth rates in both periods. For instance, the effect of YRD on the growth rate in 1978-1998 is 1.55 percentage point (vs. 1.93 percentage point in Column 6 of Table 7), and that of guerrilla is 1.37 percentage point (vs. 1.52 percentage point in Column 3 of Table 7). In Columns (3) and (5) where we allow for the interactive effect of YRD and guerrilla presence, there are some differences in results. Relative to FA3 counties, YRD counties without guerrilla presence no longer has statistically significant advantage in growth rates, though the advantage remains and the magnitude remains sizable — being 0.98 (vs. 1.46 in Column 7 of Table 8) for 19878-98. However, relative to FA3 counties, YRD counties with guerrilla presence still exhibits sizable and significant advantage in growth, 1.06 (vs. 1.29 in Column 3 of Table 8) for 1952-98, and 2.65 (vs. 3.18 in Column 7 of Table 8) for 1978-1998. In Table 14, we report similar results for both famine control and death rates in 1959-1961 using border-sharing 39 counties only. In Column (1), we find that, when using only border-sharing counties, the effect of YRD-affiliation on famine control becomes even more pronounced; the estimated coefficient of YRD is 0.17 (vs. 0.12 in Column 3 of Table 9). When we include both YRD and guerrilla presence effects, the estimated YRD effect is now 0.17 (vs. 0.12 in Column 3 of Table 10); the effect of guerrilla presence remains statistically insignificant. In Column (3), we look at the interaction effect of YRD and guerrilla presence, the coefficient estimate of YRD×GuerrillaYes is now 0.20 (vs. 0.16 in Column 8 of Table 10); that of YRD×GuerrillaNo is now 0.15 and statistically significant (vs. 0.09 and statistically insignificant in Column 8 of Table 10). In Columns (4)-(6), we report the results for death rates when only border-sharing counties are included in the analysis. In Column (4), we find that the estimated coefficient of YRD affiliation on death rates during the Great Chinese Famine is still negative at -3.55‰ , but it is no longer statistically significant. When we include both the YRD and guerrilla presence in Column (5), however, the YRD effect is -10.5‰ and statistically significant at 1% level, and the effect of guerrilla presence is negative but statistically insignificant. In Column (6), the interaction effects of YRD×GuerrillaYes and YRD×GuerrillaNo are both negative and statistically significant; the magnitude of these estimates is larger than the corresponding numbers in Table 11. The fact that the results in Tables 13 and 14 using a smaller set of neighboring counties are similar to those with the full set of the counties thus renders strong support to the positive role of weak factions and local accountability in facilitating local development. 5.5 Were the Local Leaders from the Stronger Faction Starved of Resources from the Higher Level Government? Why does counties led by cadres affiliated with the stronger FA3 faction perform worse than the counties led by cadres affiliated with the weaker YRD faction? Is it possible that counties led by FA3 factions did not receive as much resource as the counties led by YRD factions? To assess this hypothesis, we plot in Figure 4 the kernel density distribution of the average fiscal expenditure/fiscal revenue ratio for 1950 and 1957, the only two years for which these statistics are available at the county level. A ratio larger than 1 indicates that the county received net transfers from the higher level government (as local government debt was prohibited then), and a ratio lower than 1 indicates that the county transferred net resources to higher level governments. Figure 4 shows that the density plot for YRD counties is to the left of the density plot for the FA3 counties, indicating relatively lower transfer to the YRD counties. This is confirmed in Table 15 where we regress the average expenditure-revenue ratio for 1950 and 1957 on the YRD dummy. We find that the ratio is about 26.8 percentage point lower (Column 3) for YRD counties than for FA3 counties when we include all the controls. However, trimming the 5% tail outliers renders the YRD dummy statistically insignificant even though the point estimate is still negative, suggesting that YRD counties are more prone to extreme resource extraction by the higher-level governments. Despite the proneness of resource extraction by the higher level governments, the counties led by YRD-affiliated cadres were still able to achieve 40 Figure 6. Kernel Density Estimation of Average Expenditure-Revenue Ratio, 1950 and 1957 YRD vs. FA3 1 YRD FA3 .8 .6 Density .4 .2 0 0 .5 1 1.5 2 Expenditure-Revenue Ratio Note: We construct the variable of Average Expenditure-Revenue Ratio to proxy fiscal favoritism by the provincial authority. Expenditure-Revenue Ratio=(Exp50+Exp57*Def50-57)/ (Rev50+Rev57*Def50-57) Figure 4: Kernel Density Estimation of Average Expenditure-Revenue Ratio, 1950 and 1957. Where Exp50 is the county's fiscal budgetary expenditure in 1950, Inc50 is the county's fiscal budgetary revenue. Def50-57 is the GDP deflator to convert 1957 Renminbi (RMB) into 1950 RMB. a better development performance for their counties. This suggests that the mechanisms must be more efficient resource allocations in the YRD-affliated counties. 5.6 Faction, Local Accountability or Something Else? Another concern is that the main results may be driven by the fact that cadres affiliated with FA3 and YRD factions might be different in terms of their skills in managing economic affairs. After all, FA3 cadres specialized in fighting wars because they spent most of their careers in the formal troop led by the Chinese Communist Party. In contrast, YRD cadres may be more experienced in working with locals because they originated from the revolutionary bases in Hebei and Shanxi province. If this was the case, the difference in the growth rate for two groups of counties can presumably be driven by skill difference of the local leaders, not by political survival incentives due to different levels of connections with provincial leaders or by local accountability. To address this concern, we argue that, if either skills and/or knowledge in economic development are important driving forces behind the differences in the development performances of FA3- and YRD-affiliated counties, then we would expect the growth gaps between these two groups of counties to decrease (and eventually disappear) over time, as FA3-affiliated cadres obtained more experience on their positions. In Table 16, we present the regression results, where we replicate the regressions in Tables 6-7, but replace the dependent variable with the annual real growth rate between 1984 and 1998. Comparing the results in Table 15 with those in Tables 6-7, we find that for every specification, the coefficient estimate for YRD is larger in the 1984-1998 period than that in the overall 1978-1998 period. Thus, skills and/or knowledge on the job does not seem to be the main factor in explaining the growth gaps between FA3 and YRD counties; if anything, the gap increases rather than decreases over time. 17 17If FA3 cadres, however, prefer to promoting their former colleagues, this kind of official selection practice will prolong their disadvantage in managing economic affairs and might cause persistent gaps in growth rate between these two kinds of counties. As we do not observe cadres’ skills or knowledge directly, it is impossible to definitively reject the mechanism of social heritage. 41 Full Sample Trimming Tail 5% (1) (2) (3) (4) Power Structure YRD -0.357** -0.266* -0.268* -0.186 (0.151) (0.140) (0.156) (0.149) Initial Conditions Ln_GVOPC_52 -0.317* -0.319* -0.254 (0.160) (0.175) (0.168) Ln_pop_52 -0.206** -0.212** -0.147 (0.083) (0.101) (0.090) Geography Share of Plain (%) 0.001 -0.001 (0.008) (0.008) Distance to Xiamen -0.000 0.000 (Km) (0.001) (0.001) Obs. 53 52 52 48 R2 0.062 0.147 0.110 0.037 Table 15: The Effect of Factions on Average Expenditure-Revenue Ratios between 1950 and 1957. Full Sample Trimming Tail 5% (1) (2) (3) (4) Power Structure YRD 4.297*** 2.864*** 3.072*** 2.844*** (1.027) (1.024) (0.932) (0.723) Initial Conditions Ln_GVOPC_78 -5.515*** -5.425*** -3.474*** (1.752) (1.478) (0.900) Ln_pop_78 1.466 -1.138 -0.001 (1.238) (1.297) (0.907) Geography Share of Plain (%) 0.193*** 0.150*** (0.066) (0.056) Distance to Xiamen -0.018*** -0.019*** (Km) (0.005) (0.005) Obs. 58 57 57 53 Adjusted- R2 0.083 0.405 0.597 0.623 Table 16: The Effects of Factions on Growth Rates in Counties in Fujian, 1984-1998. 42 5.7 Evidence of Pro-Local Policies 5.7.1 Improvement in Local Educational Achievement We now present evidence that the counties led by YRD-affiliated cadres indeed adopted pro-local policies that facilitated local development. Education, a key component of human capital, can be an important source of local growth. Table 17 regresses the log changes in the average years of schooling for individuals born in 1950-1972 from those born prior to 1930, as observed in the 1990 Census, on the county’s YRD affiliation, Guerrilla presence and their interactions. As before, we control for the initial conditions and geography. Note that those born prior to 1930 would have been 22 years old in 1952, and would mostly have finished their formal schooling by then. Similarly, cohorts born between 1950 and 1972 would be at least 18 by 1990 and would have completed their formal schooling by then. We use the log differences in the average schooling of the 1950-1972 cohorts and the cohorts prior to 1930 as the proxy for the improvement in local educational improvement. In Columns (1) and (2) of Table 17 where all the counties (with non-missing variables) are included in the analysis, we find that counties with YRD affiliations and with Guerrilla presence are associated with larger percentage increases in educational achievement, though the differences are not statistically significant. In Column (3) where we interact the YRD and the Guerrilla dummies, we find that counties with both YRD affiliation and Guerrilla presence are associated with 37.5 percentage points higher increase in average education, and the effect is statistically significant at 10% level. In the right panel of Table 17, we trim the tail 5% of the counties from the analysis. In Column (4), we find that, relative to counties led by FD3 cadres, those led by YRD cadres had a 30.1 percentage points higher increase in average schooling levels; in Column (5), we include both YRD and Guerrilla dummies in the regression, and the results indicate that the YRD premium in education increase is 28.7 percentage points, statistically significant at 10% level. Guerrilla presence is also associated with a premium in education increase of 13 percentage points, though not statistically significant. In Column (6), we include the YRD and Guerrilla interactions. We find that most of the YRD effect on educational improvement result from the counties led by YRD-affiliated cadres that also had guerrilla presence, and the premium is 38 percentage points (relative to FA3 counties). In counties led by YRD-affiliated cadres that had no guerrilla presence, the educational achievement growth is also faster than that in FA3 counties (by 22 percentage points), but the coefficient estimate is not statistically significant. The results in Table 17 confirm that counties led by cadres from weaker factions that in addition had stronger local accountability did pursue policies that led to faster improvement in local educational achievement. We believe that these are likely levers through which factional competition and local accountability impact the local development. 43 Full Sample Trimming Tail 5% (1) (2) (3) (4) (5) (6) Power Structure YRD 0.273 0.262 0.301** 0.287** (0.216) (0.218) (0.135) (0.137) Guerrilla 0.162 0.130 (0.128) (0.113) YRD×GuerrillaNo [b0 ] 0.209 0.220 (0.248) (0.142) YRD×GuerrillaYes [b1 ] 0.375* 0.380** (0.201) (0.152) Initial Conditions Ln_GVOPC_52 -0.025 -0.014 -0.014 -0.055 -0.045 -0.041 (0.128) (0.125) (0.124) (0.116) (0.113) (0.111) Ln_pop_52 -0.254 -0.301 -0.293 -0.073 -0.113 -0.119 (0.201) (0.203) (0.200) (0.118) (0.123) (0.122) Geography Share of Plain (%) 0.011 0.011 0.011 -0.000 0.000 -0.001 (0.011) (0.011) (0.011) (0.008) (0.008) (0.008) Distance to Xiamen 0.001 0.001 0.001 -0.000 -0.000 -0.000 (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Obs. 55 55 55 51 51 51 R2 0.087 0.095 0.095 0.084 0.105 0.110 H0 : b0 = b1 0.309 0.221 Table 17: The Effect of Factions and Guerrilla Presence on the Changes in the Average Years of Schooling for Individuals Born in 1952-1972 Relative to Those Born Prior to 1930, as Observed in the 1990 Census. Note: The dependent variable is the difference in log of the average education level of the cohort born between 1952 and 1970 and the log of the average education of the cohort born before 1930, as observed in the 1990 Census. In the right panel, the 5% tail observations are trimmed from the analysis. The last row reports the p-values of the hypothesis that b0 = b1 . White standard errors are in parenthesis. 44 5.7.2 State-Owned Enterprise (SOE) Shares in the Local Economy in 1998 Another pro-local policy that county leaders might have undertaken was to encourage the development of the private sector. In this subsection, we examine whether counties led by YRD-affiliated cadres and counties with guerrilla presence are more conducive to private sector development. Since China started its large-scale SOE reform in 1998 (see Hsieh and Song, 2015), and the top-down nature of that SOE reform likely would hide the footprint of the locally-initiated private sector development after 1998 (Huang et al., 2017), we focus on the share of the SOE in the county economy as of 1998. We use two measures of SOE share in the economy: The first is based on SOE share of the county’s total sales revenues; and the second is based on SOE share in terms of counts of firms. In Panel A of Table 18, we regress the SOE sales-revenue share in 1998 on the county’s YRD affiliation, Guerrilla presence and their interactions, after controlling for the initial conditions and geography. In Columns (1)-(3), we include the full sample; in Columns (4)-(6), we trim the tail 5% of the counties. In Column (1), we find that, relative to counties with FA3 affiliation, counties with YRD affiliations are associated with 22 percentage points lower in the SOE’s sales-revenue share in the county’s economy in 1998. In Column (2) when we include both YRD and Guerrilla dummies, we find that the YRD coefficient barely barged, while the guerrilla presence is not significantly associated with the SOE share. In Column (3), where we interact YRD and Guerrilla dummies, we find that counties with both YRD affiliation and Guerrilla presence are associated with 25.3 percentage points lower SOE sales-revenue share, and counties with YRD affiliation but without Guerrilla presence are associated with 21.3 percentage points lower SOE sales-revenue share, and both effects are statistically significant at 1% level. These effects are qualitatively and quantitatively similar when we trim the tail 5% of the counties. In Panel B of Table 18, we regress the SOE-count share in 1998 on the county’s YRD affiliation, Guerrilla presence and their interactions, after controlling for the initial conditions and geography. Overall, the qualitative results are similar to those in Panel A. An exception is that the guerrilla presence is more strongly related to the reduction of SOE importance (in firm counts). For instance, the weak-faction-and-Guerrilla counties have a drop in the SOE-count share by 22.8 percentage points (and statistically significant), but the weak-faction- without-Guerrilla counties have a drop that is half in size, around 11.7 percentage points (and not statistically significant). The results in Table 18 confirm that counties led by cadres from the weak faction and with stronger local accountability pursued policies more conducive to private sector development. Since private sector tends to be more productive than the SOEs (Shleifer 1998; Megginson and Netter, 2001), a lower SOE share in the local economy tends to be associated with faster local economic growth. We believe that this is likely another lever through which factional competition and local accountability impact the local development. 45 Full Sample Trimming Tail 5% (1) (2) (3) (4) (5) (6) Panel A: SOE sales-revenue share in the County in 1998 Power Structure YRD -22.866*** -22.610*** -20.370*** -20.353*** (6.494) (6.585) (6.304) (6.407) Guerrilla -2.547 -0.282 (6.196) (6.322) YRD×GuerrillaNo [b0 ] -21.343*** -19.706*** (7.418) (7.271) YRD×GuerrillaYes [b1 ] -25.320*** -21.454*** (7.441) (7.404) H0 : b0 = b1 0.597 0.818 Panel B: SOE Count Share in the County in 1998 Power Structure YRD -15.965** -15.157** -13.466** -13.020* (6.697) (6.784) (6.608) (6.806) Guerrilla -8.046 -6.805 (5.295) (5.427) YRD×GuerrillaNo [b0 ] -11.735 -9.559 (7.214) (7.180) YRD×GuerrillaYes [b1 ] -22.782*** -19.747*** (6.947) (6.844) H0 : b0 = b1 0.071 0.131 Obs. 56 56 56 52 52 52 Controls for Initial Conditions Yes Yes Yes Yes Yes Yes Controls for Geography Yes Yes Yes Yes Yes Yes Table 18: The Effect of Factions and Guerrilla Presence on the Share of SOE in the County Economy in 1998. Note: The dependent variable in Panel A is the SOE share of sales revenue in the county in 1998; the dependent variable in Panel B is the SOE share in terms of the number of firms (count share) in 1998. In the right panel, the 5% tail observations are trimmed from the analysis. All regressions control for the initial conditions and geography variables as in Table 17.. The last row of each panel reports the p-values of the hypothesis that b0 = b1 . White standard errors are in parenthesis. 46 5.8 Limits to Pro-Local Policies One interesting question is how much discretion would county level cadres have in choosing pro-local policies in an authoritarian regime? It is reasonable to hypothesize that there are limits to what local leaders can do to promote local growth; and that the pro-local policies that local leaders may implement are limited to those not easily observed by upper-level leaders, especially if they only have loose tie to upper-level leaders. As in our theory, the local leaders’ behavior relies on two factors: connection with upper-level authority (the factional connection) and connections with the grassroot (local accountability). Since the higher level government holds the power of promotion/demotion or even political survival, the local leaders dare not overtly disobey the center. Therefore, we implicitly assume that YRD cadres are likely to take policies favorable to local residents only when doing so does not openly defy their superiors. In other words, two types of cadres will perform in the same way when they implement tasks that are easily monitored by the upper-level authority. Figure 7. Kernel Density Estimation of Population Growth Rate during 1984-1998, YRD vs. FA3 1.5 YRD FA3 1 Density .5 0 0 .5 1 1.5 2 Population Growth Rate Figure 5: Kernel Density Estimation of Population Growth Rates Among Counties Led by YRD- and FA3- affiliated Cadres, 1984-1998. 5.8.1 One Child Policy We use the population growth after One-Child policy to test the above hypothesis. First, the One-Child Policy was launched as a national policy by the central government in 1979; it is a highly visible policy. Second, at all levels of the government hierarchy, Population and Family Planning Commissions were established after this policy, which made enforcement of the policy easily observable. Due to these reasons, even though taking a loose birth policy would increase the local cadres’ popularity among the local population, we expect that YRD-affiliated cadres would not openly defy the central government’s national policy. Figure 5 shows that the kernel density plot of the population growth rates among counties with YRD- affiliated leaders and those with FA3-affiliated leaders between 1984-1998. The two density plots do not exhibit 47 Full Sample Trimming Tail 5% (1) (2) (3) (4) Power Structure YRD -0.034 -0.072 -0.053 -0.094 (0.111) (0.104) (0.109) (0.109) Initial Conditions Ln_GVOPC_78 -0.154 -0.165 -0.277*** (0.208) (0.199) (0.092) Ln_pop_78 0.023 0.046 -0.008 (0.099) (0.114) (0.090) Geography Share of Plain (%) -0.004 0.002 (0.009) (0.006) Distance to Xiamen -0.000 0.000 (Km) (0.001) (0.001) Obs. 57 56 56 51 R2 0.001 0.046 0.054 0.144 Table 19: The Effect of Factions on the Population Growth Rates in the Counties in Fujian, 1984-1998. 48 Number of Months between the Start of HRS in the County from December 1978 (1) (2) Power Structure YRD -1.647 (3.074) YRD×GuerrillaNo 1.375 (3.615) YRD×GuerrillaYes -5.389 (3.428) Initial Conditions Ln_GVOPC_52 -6.783** -7.467*** (2.561) (2.291) Geography Share of Plain (%) 0.354 0.517 (0.340) (0.358) Distance to Xiamen 0.030* 0.025 (0.017) (0.016) Ln(Elevation) 17.834 23.913 (19.374) (21.418) Constant 22.934 16.822 (49.828) (50.463) Obs. 58 58 R2 0.169 0.238 Table 20: The Effect of Power Structure on the Initiation of the Household Responsibility System (HRS) in Counties in Fujian Province. obvious differences in mean population rates, though there is more dispersion among YRD-affiliated counties. This pattern is confirmed in the regression analysis in Table 19: the coefficients for power structure are never statistically significant in explaining population growth rates in all specifications. 5.8.2 Household Responsibility System (HRS) At the beginning of the rural reform in early 1980s, Chinese central government launched the household responsibility system (HRS), which allowed households to contract land from collective organizations. Households could make operating decisions independently within the limits set by the contract agreement, and freely dispose of surplus production over and above national and collective quotas. HRS represents one of the most significant institutional reforms in the Chinese agricultural sector. It gave the farmers the residual claim rights for production 49 from the land after meeting the procurement quota. When it was initiated in secret in Xiaogang Village of Fengyang County in Anhui Province in December 1978, it was prohibited by the central government. It was clearly a pro-local policy. By 1982, it was recognized by the central government as a form of institutional innovation, and in 1983 it spread throughout China and most of the village collectives adopted HRS. In Table 20 we test whether factions and local accountability measures might have affected the timing of the initiation of HRS in the county. We collect the time (year and month) that a county government unofficially or officially allowed for HRS from the county gazettes. We use the number of months between the county’s initiation of the HRS and December 1978 as the dependent variable, a larger value of which indicates a later adoption of the HRS . We find that YRD-affiliated counties allowed or adopted HRS about 1.6 months earlier than FA3-affiliated counties, but the difference is statistically insignificant. In Column (2) we find that the difference is mainly coming from YRD-affiliated counties with guerrilla presence, where HRS was permitted or adopted about 5.4 months earlier than FA3 counties. But the difference is again not statistically significant. The two case studies reported in this subsection on One-Child Policy and HRS adoption suggest that there is a limit on how much local leaders are willing to go against the central government in their desire to adopt pro-local polices. In counties with weaker factions, or those feature local accountability pressure, their leaders would choose more pro-local policy, but their political instinct also limits that they would not ostensibly defy central government if the policy in question is clearly observable to the higher level governments. Doing so would have implied higher probability of being politically purged. 5.9 Factions, Grassroot Support and Political Survival: Some Direct Evidence We have advanced the idea that local politicians’ decisions about what types of local development policies to pursue are shaped by their incentives of political survival, and that their chances of political survival depend on the factional support from higher-level government as well as the grassroot support from local citizens. As we illustrate in our model presented in Section 2, local cadres from the strong faction may find it advantageous in terms of political survival to cater to the policies desired by the higher level officials instead of pro-local policies; in contrast, local cadres from the weak faction may find pro-local development policies to enhance their political survival. To provide direct evidence for this mechanism is not easy, as the career paths of the county-level cadres were impossible to track. In this section, we exploit a unique historical event of the Cultural Revolution (1966-1976) to provide direct evidence about the mechanisms that relate faction affiliation, grassroot support and political survival. After the initiation of the Cultural Revolution in 1966, county-level Communist Party Committees gradually lost power amid the chaos. To sustain political order, the central authority launched the so-called “Three Support, Two Military” (shorthand for “Support the Left, Support the Peasants, Support the Workers; Military Training, Military Control”) Movement in 1967, which facilitated military cadres to organize the Core Leading Group of the County Revolutionary Committee to be in charge of local administrations. In essence, local cadres were 50 (1) (2) FA3 -3.238** -6.042 (1.254) (6.308) Famine Control -6.224** -7.078** (3.070) (3.337) FA3×Famine Control 4.019 (8.863) Constant 9.609*** 10.292*** (2.553) (2.782) Obs. 53 53 R2 0.116 0.103 Table 21: The Effect of FA3 and Famile Control (1959-1961) on Retaking the Power During 1971 and 1979. purged from power by the military in the process. The county Core Leading Group stayed in power until late in 1970, when the county Party Committees were reestablished and military cadres gradually retreated from the county leadership. That is, the county Party Committees attempted to regain their leadership around the end of 1970. Consistent with our model, we expect that those counties that either kept a close connection with the provincial leaders or had strong grassroot support were likely to regain their power from the military-led Core Leading Group faster after 1971. In Table 21, we attempt to directly test the effect of the connection to higher level government and the grassroot support on the ability of a faction to regain power from the military Core Leading Group. The dependent variables in Table 21 are measures of speed to regain power by the original faction in the county between 1971 and 1979. For simplicity, we use the OLS specifications, and the dependent variable is “years to regain power during the period.” We use FA3 and Famine Control during the Great Chinese Famine (1959-1961) to respectively proxy the connection to the provincial leaders and grassroot support. In Columns (1)-(2), the negative OLS coefficient estimates of FA3 and Famine Control mean that FA3-affiliated counties and counties with strong grassroot support regain power faster. Interestingly, in Column (2) we find that the interaction term FA3×Famine Control is positive, suggesting that counties that already had strong connections to the higher-level government benefits less from pro-local policies in their speed of regaining power; however, the coefficient estimate is not statistically significant. The results from Table 21 provide direct evidence that connections to strong factions and grassroot support both expedited the local cadres’ regaining of power after the purge; in addition, grassroot support is not as effective for the strong FA3 faction as for the weak YRD faction in expediting their regaining power. Both are consistent with the mechanisms of political survival described in our model presented in Section 2. 51 6 Conclusion What explains persistent regional difference in development performance? Do factional competition of the local leader and local accountability matter? In this paper, we investigate, both theoretically and empirically, the role of factional competition and local accountability in explaining the enormous local variations in development performance. Our evidence comes from the county-level variations in Fujian province in China. When the Communist armies took over Fujian from the Nationalist control circa 1949, cadres from two different army factions, namely the Third Field Army (FA3) and the Yangtze-River Detachment (YRD), were assigned as county leaders. Counties also differed in whether there was local guerrilla presence prior to the Communist takeover. The Fujian Provincial Standing Committee of the Communist Party, however, was dominated by members from FA3, which we refer to as the strong faction. We argue that local leaders’ incentives regarding development policies depended on whether they were from the strong faction in the provincial government. County leaders from the strong faction were less likely to pursue policies friendly to local development, because their political survival depended more on their loyalty to the provincial leader than on the grassroot support from local residents. In contrast, similar to politicians facing stronger electoral competition in democratic countries, the political survival of county leaders from the weak faction was more based on the local grassroot support, which could be best secured if they focused on local development. In addition, local guerrilla presence in the county further improved development performance either because it intensified local accountability of the county leader, or it better facilitated the provision of local public goods beneficial to development. We offer evidence that the historical assignment of local cadres’ factions and whether the county had guerrilla presence had little to do with the initial economic conditions, though it was somewhat related to geography, as proxied by the share of plains in the county and the distance of the county from Xiamen – the provincial commercial hub near the Taiwan Strait. We provide robust evidence that counties led by cadres affiliated with the weak faction and counties with local guerrilla presence tend to have less severe famine – measured both by the degree of birth cohort shrinkage and the death rates — during the Great Chinese Famine (1959-1961), and had significantly faster real annual economic growth rates in the whole period of 1952-1998, but particularly in the post-reform period of 1978-1998. We also find positive interaction effect between weak factions and local accountability: counties with guerrilla presence and led by cadres affiliated with the weak YRD factions tend to grow the fastest and had the least severe famine. The magnitudes of these effects also imply first-order importance. For instance, counties led by weak YRD factions with guerrilla presence had 3.3 percentage points higher annual growth rates in 1978-1998 (or 0.85 SD of county growth rates). We further provide evidence that two potential mechanisms for such long-term effects could be improvement in local education, and the development of private sector. Our findings underscore the important roles of political survival in shaping local leaders’ policy choices that may have drastic implications about economic growth, as well as lives and deaths, literally and figuratively. We believe that the insights we uncovered in this paper from Fujian province are robust and general in settings 52 of one-party-ruled countries, where political purges and turnovers are important concerns for politicians. De facto one-party rule has been quite common in the world. Indeed, at the country level over the past four decades, about 45 percent of country-years (among a sample of 179 countries) can be characterized as having de facto one-party rule. 18 In future work, we will continue to examine the connections between factional competition, local accountability, political survival, and local development in other provinces in China, and around other parts of the world. References [1] Acemoglu, Daron, Simon Johnson and James A. Robinson (2001). “The Colonial Origins of Comparative Development: An Empirical Investigation.” American Economic Review, 91 (5), 1369-1401. [2] Acemoglu, Daron, Simon Johnson and James A. Robinson (2001). “Reversal of Fortune: Geography and Institutions in the Making of the Modern World Income Distribution.” Quarterly Journal of Economics, 117 (4), 1231-1294. [3] Acemoglu, Daron and James Robinson, James (2001). “A Theory of Political Transitions.” American Economic Review, 91, 938–963. [4] Acemoglu, Daron and James Robinson (2006). Economic Origins of Dictatorship and Democracy. Cam- bridge, MA: Cambridge University Press. [5] Acemoglu, Daron and James R. Robinson (2012). Why Nations Fail: the Origins of Power, Prosperity and Poverty. New York: Crown Publishing. [6] Acemoglu, Daron, Tristan Reed, and James A. Robinson (2014). “Chiefs: Economic Development and Elite Control of Civil Society in Sierra Leone.” Journal of Political Economy, 122 (2), 319-368. [7] Altonji, Joseph, Todd Elder and Christopher R. Taber (2005). “Selection on Observed and Unobserved Variables: Assessing the Effectiveness of Catholic Schools.” Journal of Political Economy, 113 (1), 151- 184. [8] Ansolabehere, Steven and James M. Snyder. (2006). “Party Control of State Government and the Distribution of Public Expenditure.” Scandinavian Journal of Economics, 108 (4), 547-69. [9] Banerjee, Abhijit and Lakshmi Iyer (2005). “History, Institutions, and Economic Performance: The Legacy of Colonial Land Tenure Systems in India.” American Economic Review, 95(4): 1190-1213. 18This is based on calculation using the Database of Political Institutions 2015 (Cruz, Keefer, and Scartascini, 2016). 53 [10] Barro, Robert J. (1991). “Economic Growth in a Cross Section of Countries.” Quarterly Journal of Eco- nomics, Vol. 106, No. 2, 407-443. [11] Barro, Robert J. and J.W. Lee (1994). “Sources of Economic Growth.” Carnegie-Rochester Conference Series on Public Policy, Volume 40, 1-46. [12] Becker, Gary S. (1958). “Competition and Democracy.” Journal of Law and Econ., Vol. 1, 105-109. [13] Belton F. and Jian C. (1997). “The Coast-Noncoast Income Gap, Productivity and Regional Economic Policy in China.” Journal of Comparative Economics, Vol. 25, No. 2, 220-236. [14] Besley, Timothy, and Robin Burgess (2002). “The Political Economy of Government Responsiveness: Theory and Evidence from India,” Quarterly Journal of Economics, 177, 1415–1451. [15] Besley, Timothy, and Ian Preston (2007). “Electoral Bias and Policy Choice: Theory and Evidence,” Quarterly Journal of Economics, 122(4), 1474–1510. [16] Besley, Timothy, Thorsten Persson, and Daniel Sturm (2010). “Political Competition and Economic Perfor- mance: Evidenced from the United States,” Review of Economic Studies, 77(3), 1329-52. [17] Björkman, Martina and Jakob Svensson (2009). “Power to the People: Evidence from a Randomized Field Experiment on Community-Based Monitoring in Uganda.” Quarterly Journal of Economics, 124 (2): 735-769. [18] Bueno de Mesquita, Bruce and Alastair Smith (2008). “Political Survival and Endogenous Institutional Change.” Comparative Political Studies, 42, 167–197. [19] Bueno de Mesquita, B. and Smith, A. (2010). “Leader Survival, Revolutions, and the Nature of Government Finance.” American Journal of Political Science, 54(4), 936–950. [20] Bueno de Mesquita, Bruce, Alastair Smith, Randolph M. Siverson, and James D. Morrow (2003). The Logic of Political Survival. Cambridge: The MIT Press. [21] Coale, A. J. (1981). “Population Trends, Population Policy and Population Studies in China.” Population and Development Review, 7, 85-97. [22] Cruz, Cesi, Philip Keefer, Carlos Scartascini (2016). “Database of Political Institutions, 2015 Update (DPI2015).” Updated version of Thorsten Beck, George Clarke, Alberto Groff, Philip Keefer, and Patrich Walsh, 2001. "New Tools in Comparative Political Economy: The Database of Political Institutions." World Bank Economic Review, 15, 165-176. [23] Dell, Melissa, Nathan Lane, and Pablo Querubin (2018). “The Historical State, Local Collective Action, and Economic Development in Vietnam.” Econometrica, 86(6), 2083-2121. 54 [24] Dikotter, F. (2010). Mao’s Great Famine: The History of China’s Most Devastating Catastrophe, 1958-1962. Bloomsbury Publishing USA, Walker. [25] Distelhorst, Gregory and Yue Hou (2017). “Constituency Service Under Nondemocratic Rule: Evidence from China.” Journal of Politics, 79(3): 1024–1040. [26] Engerman, Stanley L., and Kenneth L. Sokoloff (1997). “Factor Endowments, Institutions, and Differential Paths of Growth Among New World Economies: A View from Economic Historians of the United States.” in “How Latin America Fell Behind: Essays on the Economic Histories of Brazil and Mexico 1800-1914.” Stanford University press, 260-304. [27] Fan, Ziying, Wei Xiong, and Li-An Zhou (2016). “Information Distortion in Hierarchical Organizations: A Study of China’s Great Famine.” Working Paper, Princeton University. [28] Gennaioli, Nicola, Rafael LaPorta, Florencio Lopez-de-Silanes, and Andrei Shleifer (2013). “Human Capital and Regional Development.” Quarterly Journal of Economics, 128(1), 105-164. [29] Hsieh, Chang-Tai and Michael Zheng Song (2015). “Grasp the Large, Let Go of the Small: The Transfor- mation of the State Sector in China.” Brookings Papers on Economic Activity, Spring, 295-346. [30] Huang, Zhangkai, Lixing Li, Guangrong Ma, and Lixin Colin Xu (2017). “Hayek, Local Information, and Commanding Heights: Decentralizing State-Owned Enterprises.” American Economic Review, 107(8), 2455-2478. [31] Jia, Ruixue, Masa Kudamatsu and David Seim (2015). “Political Selection in China: Complementary Roles of Connections and Performance.” Journal of the European Economic Association, Vol. 13, No. 4, 631-668. [32] Kasahara, Hiroyuki and Bingjing Li (2018). “Grain Exports and China’s Great Famine, 1959-1961: County- Level Evidence.” Working Paper, National University of Singapore. [33] Kung, James and Justin Y. Lin (2003). “The Causes of China’s Great Leap Famine, 1959–1961.” Economic Development and Cultural Change, 52, 51–73. [34] Kung, James and Chen Shuo (2011). “The Tragedy of the Nomenklatura: Career Incentives and Political Radicalism during China’s Great Leap Famine.” American Political Science Review, 105(1), 27-45. [35] La Porta, Rafael, Florencio Lopez de Silanes, Andrei Shleifer, and Robert Vishny (1998a). “Law and Finance.” Journal of Political Economy, 106 (6), 1113-1155. [36] La Porta, Rafael, Florencio Lopez de Silanes, Andrei Shleifer, and Robert Vishny (1998b). “The Quality of Government.” Journal of Law, Economics and Organization , 15 (1), 222-279. 55 [37] Li, Wei and Dannis Tao Yang (2005). “The Great Leap Forward: Anatomy of a Central Planning Disaster.” Journal of Political Economy, 113(4), 840-877. [38] Li, Hongbin and Li-An Zhou (2005). “Political Turnover and Economic Performance: The Incentive Role of Personnel Control in China.” Journal of Public economics, 89(9), 1743-1762. [39] Li, Yuan (2014). “Downward Accountability in Response to Collective Actions: The Political Economy of Public Goods Provision in China.” The Economics of Transition, 22(1), 69-103. [40] Lucas, Robert E. (1990). “Why Doesn’t Capital Flow from Rich to Poor Countries?” American Economic Review Papers and Proceedings, 80(2), 92-96. [41] North, Douglas (2005). Institutions, Institutional Change, and Economic Performance. Cambridge: Cam- bridge University Press. [42] Mansuri, Ghazala and Vijayendra Rao (2013). Localizing Development: Does Participation Work? Inter- national Bank for Reconstruction and Development/The World Bank. [43] Maskin, Eric, Yingyi Qian, Chenggang Xu (2000). “Incentives, Information, and Organizational Form.” Review of Economic Studies, 67(2), 359-378. [44] Megginson, William, Jefferey Netter (2000). “From State to Market: A Survey of Empirical Studies on Privatization.” Journal of Economic Literature, 39, 321-389. [45] Meng, Xin, Nancy Qian and Pierre Yared (2015). “The Institutional Causes of China’s Great Famine, 1959–1961.” Review of Economic Studies, 82 (4): 1568-1611. [46] Miquel, Gerard Padró I. (2007). “The Control of Politicians in Divided Societies: The Politics of Fear.” Review of Economic Studies, 74(4), 1259-1274. [47] Olsen, Mancur (1993). “Dictatorship, Democracy, and Development.” American Political Science Review, 87(3), 567-576. [48] Persson, Petra and Ekaterina Zhuravskaya (2016). “The Limits of Career Concerns in Federalism: Evidence from China.” Journal of the European Economics Association, 14(2), 338-374. [49] Shih, Victor (2008). Factions and Finance in China: Elite Conflict and Inflation. Cambridge University Press. [50] Shih, Victor, Christoper Adolph and Mingxing Liu (2012). “Getting Ahead in the Communist Party: Explaining the Advancement of Central Committee Members in China”, American Political Science Review, 106, 166-187. 56 [51] Shih, Victor, Qi Zhang and Mingxing Liu (2013). “Guerilla Capitalism: Revolutionary Legacy, Political Cleavage, and the Preservation of the Private Economy in Zhejiang.” Journal of East Asian Studies, 13: 379-407. [52] Shleifer, Andrei (1998). “State versus Private Ownership.” The Journal of Economic Perspectives, 12(4), 133-150. [53] Smith, Alastair (2008). “The Perils of Unearned Income.” The Journal of Politics, 70(3), 780-793. [54] Stigler, George J. (1972). “Economic Competition and Political Competition.” Public Choice, 13, 91-106. [55] Vogel, Ezra F. (2011). Deng Xiaoping and the Transformation of China. Cambridge, Mass.: Belknap Press of Harvard University Press. [56] Wittman, Donald. (1989). “Why Democracies Produce Efficient Results.” Journal of Political Economy, 97 (6), 1395-1424. [57] Xi, Tianyang, Yang Yao and Muyang Zhang (2018). “Competence versus Incentive: Evidence from City Officials in China.” Journal of Comparative Economics, DOI: 10.1016/j.jce.2018.04.001. 57