Quick viewing(Text Mode)

War, Bureaucracy, and State Capacity: Evidence from Imperial China

War, Bureaucracy, and State Capacity: Evidence from Imperial China

War, , and State Capacity: Evidence from Imperial

Peng Peng

November 12, 2018

Abstract

Warfare has been at the center of the analysis of state building. This paper explains why provisional financial strategy, driven by fiscal needs of warfare, impairs state capacity through undermining bureaucratic recruitment process. I exploit an original dataset between 1644 and 1911 in the Imperial China to provide empirical evidence that officials who entered the bureaucracy by purchase were less intrinsically motivated and less capable of mobilizing resources when managing crisis. This focus on bureaucracy as a pillar of state capacity provides an alternative pathway to effective state building, adding to the bellicist theory derived from the history of Western Europe.

1 Introduction

Tilly(1992) famously argues that “states made wars, war made states.” However, China did not experience state development as occurred in Western Europe, despite extensive exposure to conflicts. This is even more puzzling, considering China’s head start on building an effective state. In 221 BCE, China’s first unitary state, the , abolished national feudal titles; appointed national officials to local jurisdictions; standardized the currency, measures and weights; established a uniform written language; and centralized the tax system. By contrast, l’Acad´emiefran¸caisewas only created in 1635 and early censuses were conducted by the Church and not the state in Western Europe.

In this paper, I introduce and provide evidence for a mechanism through which fiscal needs driven by warfare directly weaken state capacity: the recruitment process for

1 bureaucracy. State capacity is defined as a state’s ability to implement its goals (Mann, 2012). Local administrators are the on-the-ground implementers of state policy, so their motivation and ability have impacts on the actual implementation of policy. If rulers compromise merit-based bureaucratic recruitment for provisional financial relief, a deterioration of state capacity follows because these administrators are less intrinsically motivated and less capable. In the bellicist theory, an effective bureaucratic apparatus is often assumed. A few works that relate to bureaucracy use size as an indicator of its strength (Brewer, 2002; Garfias, 2018). This is problematic because a bulky government is likely to be inefficient, weak, and characterized by patronage. In sum, bureaucracy may not enhance state capacity, conditional on the officials who constitute it. The current literature on state capacity focus mainly on the political side of the state, for example, representative institutions and electoral systems (Mares and Queralt, 2015; North and Weingast, 1989). Fewer work consider bureaucratic characteristics as determinants of state capacity. My work intends to fill the gap.

This argument is particularly relevant to developing countries in which broadening tax bases entails prohibitively high political and economic costs and government is inflicted with financial difficulty. Imperial China provides a suitable case to test this relationship because of two characteristics. First, Imperial China was an agri- cultural economy. By understanding the institutional consequences of its rulers’ financial strategies, we can gain a deeper understanding of the political and eco- nomic constraints of state building under which underdeveloped economies operate. Second, its bureaucratic selection was highly routinized and merit-based, allowing me to test the effects of recruitment procedures on state capacity. I analyze data from a unique dataset I create for Imperial China between 1644 and 1911, and show that officials who entered the administration by purchasing their offices weakened state capacity. I also show two key mechanisms by which this occurred, namely, they were more likely to over-extract and less able to mobilize resources from local elites during crisis management.

This paper makes three contributions. First, I contribute to the discussion of bureau- cracy and state capacity. The existing literature mostly focuses on fiscal capacity (Beramendi et al., 2018; Queralt, 2018; Dincecco et al., 2011). Bureaucracy, espe- cially local administration, as I show, is an important pillar of state capacity, but lacks systematic discussion. Second, my work sheds light on the linkage between irregular public revenues and state capacity. While it is politically expedient for rulers in developing countries to obtain revenues through sale of offices, this strat- egy undermines state building in the long run. Scholars studying export-oriented economies and rentier states have pointed out that over-reliance on abundant natural

2 resources, in particular, petroleum, makes rulers lose the inventive to invest in both extractive capacity and bureaucratic capacity because revenues are too easily ob- tained with the help of a small, specialized bureaucracy (Chaudhry, 1989). Foreign aid and external borrowing also elicit moral hazard: states spend the easy money without broadening the tax base and strengthening the bureaucracy (Br¨autigam and Knack, 2004; Queralt, 2018). My argument is similar in logic to these scenar- ios: a windfall of money may help rulers avoid politically difficult bargaining, but it arrests state capacity. I differ from this literature by proposing and providing em- pirical support for how irregular public revenues impair the recruitment procedures for local administration directly. Third, this paper contributes to the discussion of Great Divergence (Pomeranz, 2009; Rosenthal and Wong, 2011) and state develop- ment in China. We know relatively more about state development or lack thereof in Africa and Latin America (Herbst, 2014; Boone, 2003; Centeno, 2002; Bates, 2014, 2015). By contrast, less work has been done on the development and atrophy of state organizations in China, which has a long history of meritocratic bureaucracy.

2 Argument

In this section, I first review the bellicist theory of state building and its limited explanatory power in Africa and Latin America. I then emphasize a key feature of the theory, namely, the linkage between revenue strategy and domestic institu- tions. Finally, I argue that when rulers compromise a meritocratic bureaucratic recruitment for provisional fiscal relief, state capacity weakens. Non merit-recruited bureaucrats are inherently driven to maximize their own private interests at the ex- pense of public good; they are also less skilled at mobilizing resources to implement policies designed by the political center to improve state capacity.

2.1 The Bellicist Theory of State Building and Its Limited Explanatory Power

The bellicist theory has many variants but its main content can be summarized as follows. In a context of continual threats of interstate warfare, the state needs to raise sufficient material resources to compete militarily. This revenue imperative compels rulers to invest heavily in building state institutions and, in particular, tax institutions. Tax institutions are instrumental to rulers because they can create reliable records of taxpayers, their economic activities, and their assets (Tilly, 1992;

3 Dincecco, 2011) and thereby provide rulers with a more reliable source of revenue. In exchange for tax revenues, rulers agree to establish representative institutions that cede powers to taxpayers (Stasavage, 2016, 2010; Dincecco, 2011). An effective state is thereby established, a state that is able to secure a broad taxation base and large-scale borrowing in private markets (Dincecco, 2011; North and Weingast, 1989). This story of state making obtained its inspiration from Western European history, especially the history of England and the Netherlands where pre-existing representative institutions provided the organizational infrastructure for tax bar- gaining.

However, the bellicist theory has been found of limited use and explanatory power in other regions. Geography and the character of warfare have been singled out as the primary reasons for why other regions did not experience similar state de- velopment. Despite frequent conflicts, political disorder still pervades Africa and Latin America (Herbst, 2014; Centeno, 2002). States are unable to monopolize the legitimate use of violence, the basic function of the state as defined by . Herbst(2014) explains the prevalence of weak states in Africa by pointing to its geography. Africa’s barren and arid soil lowered the potential gains of fighting terri- torial wars; consequently, African rulers fought wars to plunder labor and animals, but not to establish effective territorial control with permanent state organizations. In the case of Latin America, Centeno(2002) emphasizes the character of warfare, arguing that only external wars amount to severe threats to rulers, and internal wars in Latin America merely divided elites and slowed down the development of political authority.

Unlike Africa and Latin America, state institutions developed very early in China. As the first Chinese state, the Qin laid the groundwork for a centralized tax and bureaucratic system (Hui, 2005). Figure1 shows the territory of the Qin state and the comparison with present-day China.1 Its control stretched all the way to Mongolia in the north and Northern Vietnam in the south. These state institutions, however, did not give China a head start. As in Africa and Latin America, the Chinese state did not become effective, taxing on a broad base and regulating social and economic activities. Dincecco and Wang(2017) point to the differences in political geography between China and Europe. They argue that China’s sheer size made external warfare less frequent, whereas the high level of fragmentation of Western Europe after the fall of the Carolingian Empire made it more prone to external warfare. This view echos the research of Stasavage(2010), who argues that the small size of European polities lowered transaction costs of establishing and

1The maps come from https://www.travelchinaguide.com/intro/history/qin/qin-dynasty- map.htm

4 maintaining representative institutions.

[Figure 1 about here.]

Certainly, geography and the character of warfare are important external factors in shaping state institutions. My theoretical approach differs from the above arguments by focusing on domestic institutions as a determinant of state capacity. Figure2 shows that wars occurred on a frequent basis in Imperial China from 1650 to 1900.2

[Figure 2 about here.]

The key and often overlooked feature of the bellicist theory is the linkage between fis- cal strategy and domestic institutions. Establishing a professionalized tax agency is not globally optimal. Customs, excise, land tax, external borrowing, domestic credit, export tax, sales of office, and even plundering are all possible revenue strategies, producing different institutional consequences. State building in Western Europe was preceded by the growth of Atlantic trade (Acemoglu et al., 2005) and a domes- tic credit market (North and Weingast, 1989), providing necessary conditions for collecting commercial taxes (?). The institutional consequence was power-sharing institutions that ceded power to tax payers (North and Weingast, 1989; Dincecco, 2011). Queralt(2018) points out that external borrowing weakens state capacity because the possibility of renegotiating or repudiating loans weakens incentives to invest in state capacity. Alternatively, a ruler may reap benefits from oil production or foreign aid (Chaudhry, 1989; Br¨autigamand Knack, 2004). Chaudhry(1989) finds that in Saudi Arabia, oil revenues transformed the function of bureaucracy from collecting broad-based taxes to redistributing wealth gained from exporting oil (Chaudhry, 1989). Br¨autigamand Knack(2004) make a case for how foreign aid produces moral hazard. Political elites have weak incentives to change a situation in which large amounts of aid provide exceptional resources for patronage and fringe benefits. In its extreme version, allow corruption in tax departments or an ineffective internal revenue service to continue. Government annual accounts look deceptively healthy but state capacity is weakened.

When the main source of government income is agricultural taxes, as in Imperial China and the French ancien r´egime (Kuhn, 1970; Morrisson and Snyder, 2000),

2I collect war data from Catalog of Historical Wars in China published by the Nanjing Military Academy (2003). This catalog contains dates, location, and leaders of conflicts that took place in China from approximately 1000 BCE to 1820 BCE. The Catalog derives this information from Chinese official historical books, known as the Twenty-Four Histories.

5 a cost-effective way to raise revenue is to simply sell offices. Increasing taxes on the agricultural population does not yield high profits for rulers because harvests in preindustrial societies were highly volatile and highly dependent on natural factors like rainfall and sunlight. At the same time, public offices were the most important source of social prestige. This is my point of departure: Chinese emperors used sale of offices as their financial instrument when pressed by conflicts, which is com- pletely different from broad-based taxes as in Western Europe. The institutional consequence was a corrupt and inefficient bureaucracy. The relationship between sale of offices, state capacity, and conflict is captured well by Feng Guifen, a Qing scholar-official:

In the past ten years [1850s], sales of rank have been frequent, and civil government has therefore been weakened. When civil government is weakened, social ferment becomes critical; when social ferment becomes critical, the public revenues are strained; when the public revenues are strained, there is increased sale of rank. This is the way in which one rebellion leads to another. In discussion of present-day governance, I consider the abolition of sale of rank to be the first principle.3

2.2 Bureaucratic Recruitment and Its Impacts on State Ca- pacity

In this section, I present an argument that the process to select bureaucrats, as driven by the financial needs of the ruler, has significant impact on the ability of government to meet challenges.

The argument centers on two actors in an agricultural economy with a centralized bureaucracy: a ruler and local officials. The ruler’s power emerges from military dominance. The ruler strives to maximize his chances of survival, which is achieved by investing in agricultural production. This set-up makes sense in an agricultural economy as the ruler relies primarily on land tax to finance government expenditures. At the same time, his chances of holding onto power are inverse to the likelihood of peasant rebellions, which are usually fueled by heavy taxes. In such societies, policies that increase state capacity are those that promote the state’s ability to steer and promote agricultural production. Local officials are the other political player. These administrators seek to maximize rents extracted from being in office,

3Feng Guifen, Jiaofenlu Kangyi (Protests from the Hut of Revision), 1897 edition (reprint, Taipei: Wen-hai Press) 1.17b-19a. Translated in Mary Wright, The Last Stand of Chinese Con- servatism, p. 85

6 chances of getting promoted, and fulfillment of intrinsic values. On the one hand, they differ in their ability to implement policies made by the ruler, at the political center. On the other hand, they engage in rent-seeking activities to varying degrees, which is particularly prevalent if they are paid poorly. When material benefits and intrinsic values are in conflict, we expect to see officials trade off self-interest and public good, characterized by policies that promote the social welfare of the local community. Officials who value their own economic interests would subvert policies that are designed to increase state capacity because they are unwilling to let go of the opportunity of extraction. A “virtuous” official, by contrast, is willing to sacrifice the opportunity of extraction to national policies that are designed to increase state capacity. When national policy and self-interest are not in conflict, local administrators act to increase state capacity. But we should still be able to observe variation in the extent to which the center’s policy is implemented as a result of varying levels of competence of local officials. Facing the same social and political constraints, a competent official is able to countervail these limiting factors on his performance and push forward for greater state capacity. These limiting factors are especially crucial when crisis occurs, imposing restraints on the available resources an official can use. The major resource providers are local elites and the ruler at the political center. Competence is thus reflected in one’s ability to get resources from local elites and the political center.

This argument is closely related to selecting “good types” for both democratic and authoritarian legislatures (Fearon, 1999; Chattopadhyay and Duflo, 2004; Manion, 2017). Bureaucracy is also about selecting good types, and only a bureaucracy composed of “good types” can increase state capacity. As Besley(2005) argues, the nature of the workings of government depends ultimately on the officials who constitute it, among which honest and competent ones are the most desirable. Lo- cal administrators are the actual implementers of national policy and, as such, a critical determinant of state capacity. Yet, the internal workings of the state and the individuals who provide basic public services have received less attention in the study of state capacity.

In short, bureaucracy may or may not enhance state capacity, depending on its composition. Local administration undercuts the ruler’s power when the selection procedure is not effective in selecting self-motivated and competent officials. When bureaucratic recruitment is merit-based, state capacity is strengthened because vir- tuous and capable local administrators act in ways consistent with the ruler’s ob- jective.

This characterization of local helps explain why external in-

7 ducements like wars undermine state capacity in countries with low income but a relatively routinized bureaucracy. Rulers can sell offices to raise revenue without disturbing agricultural production. The institutional consequence of this irregular public revenue strategy is that the bureaucratic selection fails to select “good types.” Incompetent and less social-minded officials are selected into public offices, crippling state organizations and bureaucratic performance. We observe this phenomenon in the Ottoman Empire, the Chinese Empire, and pre-colonial Vietnam. The fates of rulers in these regimes share a common feature: when local administration is merit- based, their rule is stable; when the composition of local administration is disrupted by external shocks like wars, their power is endangered.

Imperial China provides a highly suitable case to test this argument because of its agricultural economy and political institutions. The had a highly centralized fiscal system featuring inter-provincial revenue sharing. The central state was able to steer flows of revenues to provinces in fiscal need (Kaske, 2011). The size of the Qing’s local administration changed hardly at all, as one can see from Table1. Its composition changed, however. In 1764, 73 percent of officials had academic credentials; 22 percent had become local officials by purchase. In 1871, officials by purchase had risen to 51 percent of local officialdom. That is, half of the local administration was staffed with officials who had not taken the competitive exam. These changing patterns indicate that officials were more likely to be selected based on their willingness and ability to pay, and less so on their academic credentials or merit.

[Table 1 about here.]

Historically, we observe that sale of office as a revenue strategy does not necessarily contribute to worsened state capacity. Under the French ancien r´egime, the sale of offices increased social mobility because many of the buyers were from the middle class, not aristocrats. Purchase of office may merely reflect career motivation. This is probably the case when existing selection into public office is not based on merit and purchasing office provides an alternative path toward political power. Guardado (2018) looks at sale of offices in the Spanish Empire and its effects on long-term eco- nomic development. She uses social status to proxy internal motivation and argues that low-status officials were more likely to be involved in extraction. For my case, I am able to separate motivation and competence, leveraging the institutional details of the Imperial Chinese Civil Service Exam. Table2 summarizes my argument and the embedded mechanisms. I test empirically the relationship between bureaucratic recruitment and state capacity in Sections 6 and 7 and the possible mechanisms in

8 Section 8.

[Table 2 about here.]

3 Institutional Background

In this section, I review the relevant institutional details of the official recruitment and appointment systems in the Qing dynasty.

The Chinese Imperial Civil Service Exam took shape and matured between 750 and 1250, interrupted by the Mongol conquest in the thirteenth century. It then continued into late Imperial China until its abolition by the Qing emperor in 1905. It was the major path to office. Its evaluation was almost exclusively based on merit, measured by mastery of written classical literacy. Success in the civil service exam was the sine qua non of social prestige in Imperial China because success rates were extremely low (Zhang, 1955). Other channels of administrative recruitment also existed, such as purchase and inheritance, but these paths were deemed ”irregular” and, therefore, less prestigious (Zhang, 1955; Elman, 2000).

Figure3 shows how competitive these exams were. The numbers in the brackets indicate the success rates of advancing to the next higher level of exam. According to Zhang , a prominent Chinese historian, during the Qing dynasty approxi- mately 2 million candidates sat for the entry-level licensing exams and 1.5% of them (roughly 30,000) passed. About 1500 people passed the provincial exams, and about 300 people passed the metropolitan exams (Zhang, 1955).

[Figure 3 about here.]

The Civil Service exam was structured hierarchically. There were three levels of exams altogether: prefectural, provincial, and metropolitan. Sometimes the em- peror would supervise a palace test to select the best from those who had passed the metropolitan exam, adding a fourth tier to the hierarchy of examinations. The prefectural level was the licensing exam, in which county and prefectural schools chose eligible candidates for the provincial examinations. These exams were super- vised by prefectural officials and provincial educational commissioners. Students were required to write two essays, one based on a passage from the The Four Books and the other from the The Five Classics, both Confucian Classics containing the

9 core value and belief systems of .4 Students’ exam papers were judged on calligraphy, writing style, and content (Elman, 2000). Students who passed pre- fectural exams were granted the title of gentry, and provided with state-sponsored stipends paid in rice. Their family members were exempt from the labor corvee and land tax. The average age of these licentiates was between 17 and 37 years old (Zhang, 1955).

The contents of both provincial and metropolitan exams usually consisted of three sections that tested the knowledge of candidates beyond the scope of moral teachings of Confucianism. Table3 shows an example of the exam content. Table4 demon- strates the wide range of topics covered in policy questions. In the policy section, students were tested on their knowledge of selection of officials, cultivating talent, the economy, history, local governance, agriculture, and defense.(Elman, 2000). We can thus infer from the recruitment process that candidates who passed the civil service exams were not only relatively very immersed in Confucian moral teachings but also relatively competent because they had higher levels of literacy, numeracy, and understanding of local administration, the economy, and the legal system.

[Table 3 about here.]

[Table 4 about here.]

Ample historical evidence verifies that the selection process was fair and relatively free of corruption. Exam cheaters were expelled and punished by a ban on taking subsequent exams. The center hired a large number of copyists to recopy all the exam papers and assigned them secret codes in case the exam graders recognized the writing. If a student bribed a grader with more than 80 ounces of silver, both faced the death penalty (Elman, 2000).

The local government consisted of three levels of administration: province, prefecture and county. The prefect was the head at the prefectural level and my data were collected at this level. A prefect was responsible for tax collection, maintaining social order, adjudicating legal cases, and providing public goods, such as famine

4The Four Books were selected by Zhu in the dynasty to serve as a general introduction to Confucian thought. Five Classics are the the Book of Poetry, the Book of Documents, the Book of Change, and the Book of Annals. The Book of Poetry is a collection of 305 poems. The Book of Documents is a collection of documents and speeches of rulers and officials. The Book of Rites describes ancient rites and court ceremonies. The Book of Change contains a Chinese divination system of I-Ching. The Book of Annals is a historical record of the Confucius’s state during 722-481 BC.

10 relief and granaries (Qu, 1969). Overall, there were 267 prefectures in total. This is also the level of administration that I analyze.

The term “office selling” is actually a simplification (Kaske, 2011) because the Im- perial Chinese state rarely sold offices blatantly. Instead it sold honorific titles and academic degrees as qualifications to become officials. Both purchasers and exam takers were organized into cohorts based on the years they registered and were appointed to fill vacancies of official posts in local government by the Ministry of Personnel (, 1974).5 In the early Qing dynasty, the state had a lottery system of appointment to avoid corruption. Prospective officials drew lottery every month when a local post was vacant. In 1723, Emperor Yongzheng permitted provincial governors to appoint capable prefects and magistrates to regions with high crime rates and delayed tax payments, as a response to varying local conditions. The majority of the posts were still appointed by the lottery system (Zhang, 2011). The Qing government also stipulated a rule of avoidance that forbade officials to hold office in their native or neighboring provinces and banned their clan members and maternal relatives from serving in the same province (Qu, 1969).

To summarize, the bureaucratic recruitment was meritocratic and competitive and the appointment system was centralized and well-organized. The avoidance law suggested that officials could not use his family and lineage to influence resource allocation during crisis management. The lottery appointment system also gave me some leverage when addressing the strategic appointment of local officials.

4 Measuring State Capacity

Measuring state capacity is considered a thorny issue in political science research (Hendrix, 2010). Scholars have used various measures, for example, aggregate tax revenue, tax revenue/GDP ratio, income tax/GDP ratio, size of bureaucracy, and size of military budget (Queralt, 2018; Beramendi et al., 2018; Garfias, 2018). Ana- lytically, these measures cannot separate a state’s policy preferences from its ability to implement policy. For example, a high tax revenue/GDP ratio not only reflects a state’s ability to extract from society but also captures its preferences for welfare spending or outright extraction. Moreover, these measures are highly correlated with regime type, which makes it impossible to tease out the mechanisms linking state capacity and other variables of interest. Instead, I use the historical data on disas-

5The Qing central government consisted of six ministries: the Ministry of Personnel, the Min- istry of Rites, the Ministry of War, the Ministry of Revenue, the Ministry of Justice, the Ministry of Public Works

11 ter relief in Imperial China to measure state capacity, which is an output measure capturing the state’s ability to meet challenges during disasters. This is a minimal- ist approach to capture the core function of state, or infrastructural power (Mann, 2012). The data come from Collections on Economic History from the Veritable Records of the Qing.

This measure has several advantages. First, it is analytically separate from how policy priorities are chosen and thus allows me to inquire into causal mechanisms. Unlike tax/GDP ratio that captures both the willingness and capacity of the state to implement its goals, relieving the famine-stricken population from famine was consistently of the highest priority to the Chinese emperors. On the one hand, the empire was primarily reliant on agricultural taxation for public revenues. On the other, famine often led to social unrest and rebellion (Kuhn, 1970; Li, 2007). Disasters like droughts and floods could easily displace millions of people, creating a fertile ground for social unrest as population movement became rampant and loss of homes and property became widespread. In 1813, for example, an extremely serious drought hit Zhili (today’s Beijing and Hebei province) and quickly extended to Henan and Shandong provinces. Henan and Shandong provinces were unable to provide relief grain, leading to a peasant rebellion later that year (Li, 2007). This phenomenon is not distinct to China: rulers of agricultural countries everywhere consider famine relief as their highest priority because of its impact on tax revenues and social order. The persistence of chronic hunger in Sub-Saharan Africa is an indication of low state capacity (Baro and Deubel, 2006).

Second, the measure accurately captures the theoretical construct of interest, state capacity. The Imperial Chinese state stipulated a comprehensive famine relief plan at the national level, but the extent to which this plan was carried out was decided by local administration, which is exactly the concept of infrastructural power (Mann, 2012). The goal of the famine relief plan was to restore agricultural production as quickly as possible, by sending displaced farmers back to their homes. The plan consists of the four following steps. First, the official investigates to ascertain the severity of the disaster and classifies it on a 10-point scale, with 10 the most severe and disasters rated below 5 in no need of relief efforts. Second, the official classifies local households by degree of need and distributes extra cash and grain to those in need. Third, the official sets up distribution centers to supply food to refugees, with the grain provided from inter-regional transport by the central state, personal donations of officials, and contributions of local gentry. Finally, the official permits sale of grain at below-market prices (Li, 2007). Local administrative leadership was clearly critical to this plan because all four steps require coordination of personnel and resources. In reality, an official might cut steps either because of unwillingness to

12 spend energy on disaster relief or because of an inability to mobilize needed resources from the central state or local gentry. For example, an official could underrate the severity of disasters so as to avoid the obligation to organize famine relief. He could also bypass the household investigation and only set up grain distribution centers for refugees. All the famine relief steps from distributing extra grain and cash to allowing sale of grain at low prices are documented in the data source Collections on Economic History from the Veritable Records of the Qing.

5 Data on Bureaucratic Recruitment

To measure my independent variable, bureaucratic recruitment, I construct an orig- inal dataset of bureaucracy in the Qing dynasty (1644-1911). Using all the data, I code how the prefect entered the bureaucracy (i.., by exam or by purchase), his surname, his ethnicity (-Chinese or Manchu), and his place of birth. Qing China comprised 267 prefectures in total. I analyze a simple random sample of 50 prefec- tures. The explanatory variable of key theoretical interest is an indicator variable reflecting whether or not a prefect became an official by purchase, which takes the value of 1 if the official entered bureaucracy by purchase and 0 otherwise.

The data on bureaucrats come from Chinese local gazetteers that have been digi- tized, which are all in classical Chinese. Local gazetteers of Imperial China were compiled by members of the local elite and were produced under the sponsorship of local officials. They contain copious materials on local administration, the economy, and society, handed down from generation to generation. Figure4 shows an example of the original gazetteers.

[Figure 4 about here.]

I first check provincial gazetteers, which normally contain detailed administration information for their respective prefectures from the to the Qing dy- nasty. I use the information on Qing administration. In cases of missing information on an individual official’s background, I turn to the relevant prefectural gazetteers6. Table 11 in Appendix A presents summary statistics of my data.

6To improve data quality, I check gazetteers of the prefecture where the bureaucrat served his office and the prefecture where he was born. In Imperial China, holding public office and passing Civil Service Exam were highly respected and local gazetteers documented students who passed Imperial Civil Service Exams.

13 6 Empirical Strategy and Results

My primary objective in this section is to investigate whether bureaucratic recruit- ment affects state capacity. I have conjectured that we should expect officials who enter public office by purchase to be associated with lower state capacity.

Before turning to estimation results, I present some descriptive statistics on the relationship between bureaucratic recruitment and famine relief in Figures5 to7. Figure5 reports the total number of famine relief efforts conducted by officials by purchase and officials by exam from 1662 to 1820. I also calculate the simple moving averages (SMA) of total number of famine relief efforts conducted by both types in Figure6 and the per person number of famine relief efforts in Figure7. 7 All three figures demonstrate a consistent pattern: officials by purchase delivered fewer famine relief efforts. Obviously, the descriptive statistics do not control for other variables. nor are the statistics broken down by locality.

[Figure 5 about here.]

[Figure 6 about here.]

[Figure 7 about here.]

I now turn to the estimation strategy. I estimate the relationship between bureau- cratic quality and state capacity as follows:

Reliefit = α + βpurchaseit + γdisasterit + θpurchaseit ∗ disasterit + λXit + it.

In this equation, Reliefit, the number of famine relief efforts, is estimated as a linear function of the following variables taken in prefecture i and year t. P urchaseit is equal to 1 if the prefect entered public administration by purchase. Disasterit is 8 equal to 1 if either a drought or a flood happened. Xit is a vector of individual- level and prefectural controls, which include an official’s ethnicity and a dummy that 7The moving averages allow me to look at smoothed data rather than focusing on the volatile fluctuations that are inherent in all disaster data. 8The data come from The Yearly Charts of Dryness/Wetness in China for the Last 500 Years published by the China Meteorological Society. To be sure, this measure is not ideal: the models should take into account duration and severity of disasters. Yet, in a premodern agricultural society, disasters usually lasted for months, so the official document only records whether or not a disaster occurred in a particular year, including both flood and drought.

14 registers whether he was newly appointed. The estimator θ captures the difference between officials by exam and officials by purchase in exerting famine relief efforts, conditional on the disaster occurrence. In all models below, I use province and year fixed effects to account for time-varying trends and underlying characteristics of provinces. Indeed, historical documents suggest that officials in the late Qing dynasty were more corrupt and less capable than those who served in the earlier period (Wakeman, 1977). Famine relief efforts were also likely to be coordinated at the province level (Li, 2007). I use robust standard errors in all models to account for heteroskedasticity. In Appendix B.4, I reestimate the models using clustered standard errors with prefecture fixed effects.

As a next step, Table5 reports OLS estimates of famine relief efforts with different model specifications. The coefficient on the interaction term is consistently negative and statistically significant. In Column 1, an official by purchase is associated with 0.278 fewer famine relief efforts when disaster occurs. Specifications 2 to 4 take individual characteristics into consideration. Manchu is an indicator variable that is equal to 1 if the official belonged to the Manchu ethnicity and 0 otherwise. The rulers of the Qing dynasty were Manchus, a minority ethnic group that came from Manchuria, today’s Northeastern China. In 1644 and 1645, they defeated the Ming Dynasty and came to rule China. Han Chinese officials were more adept at administration than were their Manchu counterparts. However, in the early decades, the Imperial Civil Service Exam allocated higher per capita quotas to Manchus although they were less educated.9 Omitting this variable may cause an over-estimation of the coefficient on purchase in magnitude because worse state capacity could be a result of incompetence associated with ethnic favoritism rather than bureaucratic selection. The coefficient on Manchu is negative, indicating fewer famine efforts conducted by Manchu prefects. The second individual control is newly appointed, which takes the value of 1 if the official was newly appointed to the prefecture and 0 otherwise. Political scientists find that locally-embedded officials provide more public goods because they are either more informed or held accountable informally by the local community (Tsai, 2007; Bhavnani and Lee, 2018; Manion, 2006). By this logic, a newly appointed official may conduct fewer famine relief efforts because he lacks local information and resources. The coefficient on newly appointed is not statistically significant. The time frame of this dataset is between 1644 and 1911, and I have less missing information on officials between 1662 and

9One caveat to the coding of Manchu is that I coded anyone who belonged to Eight Banners as Manchu. Eight Banners were administrative and military divisions under the Qing dynasty into which all Manchu households were placed. But Manchu rulers rewarded Bannermen identity to Han Chinese before the Manchus conquered China and later Eight Banners also absorbed Mongolians. So a Manchu person can be ethnically Mongolian or Han Chinese, but they have long acculturated to Manchu culture and enjoyed political privileges (Elman, 2000).

15 1820, as shown in Figure 11 in Appendix A. I rerun the regression model including a full set of individual-level controls using observations between 1662 and 1820 in Column 4. In this restricted model, an official by purchase is associated with 0.303 fewer famine relief efforts when an disaster occurs.

[Table 5 about here.]

In summary, in this section I have presented two types of evidence to suggest that officials by exam conduct more famine relief efforts than do officials by purchase. First, simple descriptive statistics in Figures5 and6 are suggestive of this pattern. Second, in Table5, statistical models that include controls for year-specific shocks and province-specific characteristics show evidence of the negative impact of low administrative quality on state capacity.

7 Alternative Explanations

The estimation results reported in Section 6 control for unobserved time trends and unobserved heterogeneity at the province level. Nonetheless, there remain several reasons why we might still be cautious about the interpretation. In this section, I consider some alternative explanations for variations in state capacity: (1) trans- portation and commercialization, (2) Western colonial influence, (3) information costs, and (4) selection bias.10

7.1 Income, Transportation, and Geography

To begin with, only areas with a certain level of productivity in agriculture can sustain large urban centers and a dense population. Higher economic productivity could lead to higher state capacity, and population is usually used as a proxy for wealth in preindustrial society (Acemoglu et al., 2002). More populous areas may have a stronger state presence and broader state functions. In Column 1 of Table6, I add population and size to the regression model. Both have positive impacts on famine relief efforts.

[Table 6 about here.]

10I address potential measurement errors of my dependent variable including whether it reflects extraction and whether early state investment lowers the probability of disaster occurring in Ap- pendix B.

16 Second, proximity to a transportation network reduces the transaction costs of the state’s presence in society (Acemoglu et al., 2015). In my case, this is particularly important given that the Qing state relied on the Grand Canal to transport tribute grain to supply the capital with staple food and ship it to provinces in need. The canal linked Yantze river and other major rivers with Beijing (see Figure8)(Li, 2007; Pierre-Etienne, 1990). Prefectures with better access to the coast and rivers may thus have more famine relief efforts. I therefore include two variables coastal city and access to main river. Both of them are indicators that take the value of 1 if the prefecture was either a coastal city or had access to a major river in China. Transportation clearly has a positive effect on the state’s ability to mobilize and coordinate resources during natural disasters. Having access to the sea or major rivers is associated with approximately 0.3 times more famine relief efforts.

[Figure 8 about here.]

Third, geography may still bias my estimates. Suppose that the emperor prioritizes rice-growing provinces over others when he distributes limited resources because rice is one of the most important tribute grains to the center. At the same time, thanks to the suitable weather and climatic conditions, the grain storage in public granaries should be higher, giving local officials more grain to distribute during disasters. I include a variable rice suitability to proxy for agricultural productivity. The variable takes a value from 1 to 5, with 5 indicating the highest agricultural suitability for growing rice. The data come from Food and Agriculture Organization of the United Nations. The coefficient on rice suitability is statistically significant and positive in Columns 5 to 8, implying that agriculturally suitable prefectures witnessed more famine relief efforts. I also include a variable terrain ruggedness to indicate the terrain ruggedness of a prefecture, which is highly correlated with famine relief efforts, as shown in Columns 6 to 8. The coefficient on purchase*disaster remains significant.

7.2 Western Influence

State building outside Western Europe is very likely to be influenced by colonial powers. Colonial settlers open up trade ports and invest in legal and physical in- frastructure that yields profits (Boone, 2003; Ali et al., 2015; Dell et al., 2017). This is particularly relevant in my case as Western presence in some Chinese cities was strong. 1840 was a turning point for the Qing Empire because in this year, the

17 British engaged the Qing in the First Opium War and thus opened China to West- ern influence (Wakeman, 1977). In 1842, the Qing was forced to sign the Treaty of Nanking, which granted indemnity and extraterritoriality to Britain, opened five treaty ports to foreign merchants, and ceded Hong Kong to the British Empire. Other Western countries followed the British in establishing treaty ports in China. Figure9 maps the locations of treaty ports in China between 1840 and 1920. Most of them were concentrated on the coast and alongside River but some were in Tibet and Yunnan, bordering British India and French Indochina. In treaty ports, Westerners were allowed to establish hospitals and churches and even have their own police forces. Western organizations may have become de facto providers of public goods, a substitute for local government. Foreign workers were documented to dis- tribute relief and raise contributions (Li, 2007). I include an indicator variable treaty post that takes the value of 1 if the prefecture was a treaty port and 0 otherwise. The coefficient on treaty port is not statistically significant in Columns 7 and 8 of Table6, suggesting that the Qing state was still the major provider of famine relief during natural disasters. The coefficient on purchase*disaster remains robust.

[Figure 9 about here.]

7.3 Information Costs

Monitoring bureaucrats in democracies is costly (Przeworski et al., 1999). It is even more so in authoritarian regimes (Manion, 2015). Stasavage(2010) argues that distance determines the transaction cost of establishing representative institutions. In a similar vein, one may argue that the number of disaster relief efforts may be lower in places far away from the capital in Imperial China because the Emperor had higher costs of monitoring. As a result, local administrators might shirk their responsibilities. To account for this explanation, I first retrieve the latitude and longitude of each prefecture and then calculate the distance between a prefecture and the capital, Beijing, measured in kilometers. As shown in Table7, distance has a negative impact on local state capacity and the coefficient is statistically significant throughout all columns. A one kilometer increase in distance from a prefecture to Beijing is associated with 0.001 fewer famine relief efforts, and the coefficient purchase*disaster remains robust in Table7.

[Table 7 about here.]

18 7.4 Selection Bias

The strategic thinking of rulers in the appointment system may also impose a threat to the validity of inference. If the ruler’s goal is to maximize his chances of political survival, then he is very likely to appoint officials with competence to regions that are strategically important. Table 12 in Appendix B shows the t-test of the group means on a range of variables that may enter into the ruler’s strategic appointment. An official by purchase was not more likely to be appointed to prefectures that were far away from the capital or to prefectures that were more likely to experience peasant revolts. But an official by purchase was more likely to be appointed to a prefecture with higher rice suitability. This might be a function of my sub-sample. As this is the case, then my estimates are underestimated because the ruler should be more willing to support agriculturally suitable prefectures during disasters. Table 13 shows that the previous disaster records cannot predict whether a prefecture had an official by purchase, implying that emperors were not more likely to appoint officials by purchase to prefectures that were disaster-prone.

8 Mechanisms

Having established the link between sale of office and state capacity, I proceed to understand the underlying mechanisms. The Civil Service Exam was able to select officials who were more skilled because of the institutional design. Candidates were rigorously tested on their knowledge of the economy, geography, legal codes, and politics. Based on the main players of disaster relief, namely, the central state, local elites, and local administration, I lay out four possible mechanisms. The first is the role of central-local connections: officials by exam were more active and effective in “lobbying” for policies in favor of their jurisdiction, leading to better local state capacity. The second mechanism is social mobilization: the social prestige associated with passing the Imperial Civil Service Exam reduced the transaction costs of bargaining with local elites when officials demanded local resources during disasters. The third mechanism is intrinsic motivation: officials who gained office by exam harbored a stronger sense of public virtue, and therefore exerted more effort in famine relief. Finally, the emperor may have perceived officials by exam as more loyal and therefore diverted more resources to them during disasters.

19 8.1 Mobilization of Central Resources

Mobilizing resources is a key component of competence. Resource mobilization is a process of obtaining resources from resource providers, to implement pre-determined goals. I expect a competent local official to be better able to obtain financial and personnel support from the central state in a timely manner. The standard proce- dure was that a local official delivered written requests to his superiors and ministries asking for financial help. This kind of request had to be substantiated with rigorous computation and household investigation. An official had to be precise about the severity of disaster and how much grain relief was needed. Only when a request was considered as legitimate would interregional transport of grain be granted. During the process of conducting comprehensive surveys, investigating household needs, computing the needed financial help, and writing requests, high levels of classical literacy, numeracy, and communication skills were essential. These skills were criti- cal given that the Qing’s fiscal system was highly centralized and local government was strained by a tight budget all the time (Zelin, 1992). Local government had no choice but to rely on an inter-regional transport system that required massive di- version of grain supplies from tribute grains, and circulation of grain was facilitated through canals and public granaries, both sanctioned by the state so a poor harvest in one region could be offset by a good one from another.

I create an interaction term official*purchase to test officials’ differential ability to mobilize resources from the central state when disaster occurs, conditional on the prefecture’s connection with the political center in Table 8.A. Here, official is the number of officials born in prefecture i who achieved higher than or equal to the level of a vice-provincial governor in the administration. The data come from ?. Ninety percent of these 4,200 high-level officials passed the metropolitan exams. The co- efficient on the interaction term is not statistically significant, providing suggestive evidence that the fiscal transfer system was still centralized and “decentralized lob- bying” for inter-regional transport grain may not have played a key role in famine relief in the Qing dynasty. But it does not rule out the role of other forms of politi- cal connections, for example local favoritism and factions, in determining local state capacity.

[Table 8 about here.]

20 8.2 Social Mobilization

As explained in Section 2, an official’s competence is also embodied in his ability to mobilize local financial and personnel resources. Because of the social prestige associated with the Civil Service Exam, officials by exam were more respected by the local community. During normal times, they demonstrated a greater amount of knowledge and a higher level of effort regarding governance to the benefit of the local community. This clear sense of commitment was the source of effective resource mobilization during crisis. Local elites may assume that both they and officials were working hard for the success of the common enterprise, i.e., combating disaster and recovering agricultural production, therefore they may be more willing to invest in local state capacity. By contrast, localities governed by officials by purchase may be trapped in a sub-optimal equilibrium. Not trusting officials by purchase, local elites may be less willing to cooperate with local administration during disasters to guard against corrupt behavior. In the context of Imperial China, local elites and officials by exam were both academic degree holders, sharing social values and political prestige. For these two reasons, I expect the two types of bureaucrats to have differential abilities to utilize resources from local elites during disasters.

To test this mechanism, I use a variable quota to proxy for the strength of local elites in a prefecture. These data come from ?. In the Chinese Civil Service exam, the numbers of successful candidates at each exam level were controlled by a quota system, with quotas for the prefectural examination assigned at the prefecture level. As explained above, only candidates who passed the prefecture-level exam were granted the title of gentry and treated as local elites. Thus, the quota system served as an institutional means of regulating the power of local elites by dictating its size (Elman, 2000). A prefecture with higher quotas thus had a larger number of local elites. The empirical results of testing this mechanism are shown in Table 8.B, in which I create an interaction term quota*purchase to capture the difference in mobilizing resources from local community by officials. The interaction term is consistently negative and statistically significant in Columns 1 to 3 of Table 8.B, suggesting that officials by exam were more effective in mobilizing resources from local elites. Figure 10 presents the marginal effects of being an official by purchase on relief efforts in a prefecture conditional on quotas. The differential impact of an official’s ability in utilizing resources was stronger when local elites were more powerful. This empirical result is different from that of Xu(2017), who shows that political connections harmed bureaucratic performance in the British Empire. One possible reason is that political elites were homogeneous in Imperial China in terms of educational background and income source.

21 The quantitative results are consistent with qualitative historical evidence. Histori- ans document that officials persuaded local elites to contribute to relief funds or set up distribution centers for grain. In the late Qing period, local elites also funded charitable institutions to shelter widows and orphans, but these institutions were initiated and sponsored by local governments (Li, 2007).

[Figure 10 about here.]

8.3 Public Virtue

In this section, I provide evidence that officials by exam were less likely to extract from local community. These “honest” officials were more likely to adopt policies that increase state capacity in the long run. Moreover, I present qualitative evidence suggesting that public virtue can be an outcome of socialization.

The political economy literature on political selection has focused on the use of re- election as a reward for exercising restraint. But not all politicians are narrowly self- interested, some harbor what I call an intrinsic motivation of “public virtue” when they are in office. For these intrinsically motivated officials, the greatest satisfaction of office does not derive from seeking rents but come from the realization of an inner desire to realize the public good. When self-interest and intrinsic values are in conflict, they are likely to sacrifice the opportunity of extracting rents for intrinsic values. To be sure, an official’s motivation is unobservable; we can only infer it from behavior. But using policy to infer motivation has certain advantages. First, it is more “truthful” than reported motivation. Second, it has better external validity and policy implications than experimental studies that rarely touch on real trade-offs between personal income and intrinsic values.

In my case, I expect officials by exam are less likely to engage in over-extraction. The most important policy stimulant in an agricultural society after a disaster occurs is tax reduction, which is designed to stimulate productivity and encourages refugees to relocate to their hometowns. More importantly, reducing taxes significantly reduces local officials’ illegal income because they are no longer “eligible” to collect surplus tax. Zelin(1992) writes that a surplus tax was levied on the tax grain in order to cover eventual losses during transport. It was originally a transport-loss surcharge allowing the tax collectors to deliver exactly the amount of tax grain requested according to the tax quota. But in reality, this transport-loss surcharge became a cash cow for local administrators as they often charged as much as 20 or 30 per cent, even up to 50 per cent. Implementing tax reduction policies means that the local

22 officials lost this source of illegal income. Furthermore, granting tax remissions had to be proposed by local officials after they had investigated the severity of disasters and evaluated the loss of tax revenues due to disasters. They then submitted requests to the central state and asked for tax remissions for their locality. This process was physically arduous because the local administration had to complete a thorough investigation and computation. In 1728, the Qing emperor admonished a local official for refusing to implement tax remission under his jurisdiction (Pierre-Etienne, 1990). We can safely conclude that tax remissions make officials less likely to be involved in extraction and that requesting such policies from the central state implies an official valuing his intrinsic values more than personal material interests.

Table9 shows the empirical test of this relationship. I regress the number of tax re- ductions in prefecture i and year t against whether an official entered officialdom by purchase when disaster occurred. The interaction term disaster*purchase is consis- tently negative and statistically significant in Columns 1 to 3, suggesting that facing disasters, officials by purchase were less likely to request tax reduction policies from the central state.

Besley(2005) argues that some politicians have qualities that make them trust- worthy and these qualities are hard-wired into preferences rather being dependent on external reinforcement. Certainly, these preferences are hard-wired, but I use some qualitative evidence to suggest that these preferences can be an outcome of socialization, more accurately, exam socialization. The curriculum of the Imperial Chinese Civil Service exam was conducive to a virtuous bureaucracy. Confucian classical texts were a main component of the exam content, as reported above in Table1. Students had to spend years memorizing and internalizing these moral teachings before they finally took the exam. Benevolence (ren) was considered the highest virtue in Confucian teachings:

The Master said, wealth and honor: these are what everyone desires, but if he cannot obtain them without violating his principles, he will not pursue them. Poverty and lowliness: these are what everyone hates, but if he has to abandon his principles to get rid of them, he will not avoid them. If the gentleman avoids ren, how shall he make his name? A gentleman does not sacrifice ren for a meal. Under good circumstances, he upholds it; in deepest distress he upholds it (The Book of Analects, 4:5 (1998)).

More than ethical teaching, ren is also the basis of Confucian political theory. Ac- cording to Confucius, an inhumane ruler loses his Mandate of Heaven or the right to

23 rule, and his subjects are entitled to rebel against him. A ruler should be empathetic to the sufferings of his subjects and try his best to alleviate their suffering. In policy terms, this usually means low tax rates so that the people can keep most of their harvest to themselves:

Ai Gong asked You : “It is a year of bad harvest, and my basic needs cannot be met; what is to be done?” You Ruo replied: “Why not tax people’s income at 10 percent?” Ai Gong said: “I still don’t have enough revenues if I tax them at 20 percent. If the people have enough income, rulers will also have enough. But if the people starve, how can a ruler expect to obtain revenues?”(Mencius, 12:9 )

These Confucian Classics all share one feature: they try to make aspiring officials internalize policy preferences consistent with the well-being of the people, which is understood as a low tax burden. In fact, qingyao (light corvee and low tax rates) was the state policy for almost all dynasties (Zelin, 1992). These historical materials provide strong evidence that exam preparation is itself a process of socialization that attempt to make officials internalize social values.

[Table 9 about here.]

8.4 Loyalty as A Competing Mechanism

A competing hypothesis to explain the empirical regularity between bureaucratic quality and state capacity is that officials by exam were perceived as more loyal than were officials by purchase. Officials by purchase could be seen as opportunistic and had no vested interest in maintaining regime stability by the political center. The central state, as a result, might transport fewer needed resources to prefectures governed by officials by purchase out of concern for misuse of intergovernmental aid. To test this competing mechanism, I analyze the relationship between how an official entered officialdom and local state capacity, conditional on his ethnicity. I split the data into two groups: Manchu officials and Han-Chinese officials. In the Qing dynasty, Manchus were deemed more loyal because they shared the same ethnic identity with the rulers and the high-level officials were disproportionately Manchus. By contrast, Han officials were more capable but less trustworthy (Wakeman, 1977). Despite being less educated, the Manchus received generous quotas in the Civil Service Exam from the rulers. Moreover, they were allowed to take the exams in

24 their own language rather than classical Chinese, and the exam questions were much simpler, without policy questions (Elman, 2000). For Manchus, the Civil Service Exam was unable to select competent officials. In Columns 1 and 2 of Table 10, whether a Manchu official got his office via purchase has no statistically significant effect on local state capacity, suggesting that loyalty cannot explain the relationship between whether an official took the Civil Service Exam and local state capacity.

[Table 10 about here.]

9 Conclusion

The literature on state development has focused on military campaigns as a financial inducement for rulers to invest in state capacity. But this theory is not sufficient to explain state development or lack thereof beyond Western Europe. Despite extensive warfare, China’s early state institutions did not develop into an effective state that was able to regulate the economy and tax on a broad base. I argue that Chinese rulers’ financial strategy to cope with military expenditures, namely, sale of offices, undermined merit-based bureaucratic recruitment, which in turn impaired state capacity, measured by the state’s ability to handle crisis.

To test this theoretical argument, I use an original dataset on the Imperial Chinese Civil Service Exam between 1644 and 1911. My main empirical analysis provides evidence for the positive linkage between merit-based bureaucratic recruitment and state capacity. Making entry to the bureaucracy conditional on passing the Im- perial Chinese Civil Service Exam produced a capable pool of bureaucrats. Facing disasters, officials by exam were more likely to exert famine relief efforts compared to officials by purchase. In particular, merit-based recruitment promoted esprit de corp among political elites who were linked by the Civil Service Exam. Their common sense of commitment to upholding public interest curbed corruption, and lowered the transaction cost of utilizing local resources for the purpose of crisis manage- ment. But the emperors’ myopic choices of irregular public revenues undercut the institutional foundation of building effective state. Officials who were selected into bureaucracy via purchase were not motivated to restrain from corruption or other behaviors that weakened government performance.

The discussion in this paper puts emphasis on bureaucracy as a pillar of state ca- pacity, which tends to be overlooked by the existing literature on state capacity. This argument has implications for finding alternative approaches to building ef-

25 fective states in developing countries because a selective, merit-based bureaucracy provides a cost-effective way to increase government performance. My argument pre- sented above shares a core insight with Dahlstr¨omand Lapuente(2017) in the sense that we do not use the principal-agent models but instead discuss why motivation takes precedence over monitoring. Besides performance-based pay and tournaments among public employees for high-level positions, merit-based recruitment provides another possible mechanism to improve government performance. It also has impli- cations for the role of bureaucracy in long-term economic performance. Much of the discussion of economic growth has focused on regime type (Przeworski et al., 1999); public administration receives less attention. In reality, public administration consti- tutes the day-to-day implementers of national strategies. A competent bureaucracy is associated with the economic take-off in East Asia (Evans, 2012). Analysis of bureaucratic characteristics can shed light on the causal mechanisms through which economic growth is achieved by improving the quality of public administration.

I conclude with two potential directions for future research. My paper examines the deleterious effects on state capacity of compromised merit-based bureaucratic recruitment but it does not discuss the scope conditions. Future research should analyze the conditions under which the trade-off between provisional fiscal relief and sale of offices may take place from a broader cross-national perspective and its long-term consequences on economic performance. Second, future research should explore under what conditions political leaders are able to establish a meritocratic bureaucracy and other state institutions out of low-capacity equilibria. Countries such as China, Japan and South Korea built effective states without meeting the necessary conditions of the bellicist model. This line of inquiry is closely related to the discussion of state-led industrialization and will provide a different approach to state making from the bellicist theory.

26 References

Acemoglu, D., Garcia-Jimeno, C., and Robinson, J. A. (2015). State capacity and economic development: A network approach. American Economic Review, 105(8):2364–2409.

Acemoglu, D., Johnson, S., and Robinson, J. (2005). The rise of europe: Atlantic trade, institutional change, and economic growth. American economic review, 95(3):546–579.

Acemoglu, D., Johnson, S., and Robinson, J. A. (2002). Reversal of fortune: Geog- raphy and institutions in the making of the modern world income distribution. The Quarterly Journal of Economics, 117(4):1231–1294.

Ali, M., Fjeldstad, O.-H., Jiang, B., and Shifa, A. B. (2015). Colonial legacy, state- building and the salience of ethnicity in sub-saharan africa. The Economic Journal.

Baro, M. and Deubel, T. F. (2006). Persistent hunger: Perspectives on vulnerability, famine, and food security in sub-saharan africa. Annual Review of Anthropology, 35:521–538.

Bates, R. (2014). Markets and States in Tropical Africa: the Political Basis of Agricultural Policies. University of California Press.

Bates, R. (2015). When Things Fell Apart. Cambridge University Press.

Beramendi, P., Dincecco, M., and Rogers, M. (2018). Intra-elite competition and long-run fiscal development. Journal of Politics.

Besley, T. (2005). Political selection. Journal of Economic Perspectives, 19(3):43–60.

Bhavnani, R. R. and Lee, A. (2018). Local embeddedness and bureaucratic perfor- mance: Evidence from india. The Journal of Politics, 80(1):71–87.

Boone, C. (2003). Political Topographies of the African state: Territorial authority and Institutional Choice. Cambridge University Press.

Br¨autigam,D. A. and Knack, S. (2004). Foreign aid, institutions, and governance in sub-saharan africa. Economic Development and Cultural Change, 52(2):255– 285.

Brewer, J. (2002). The Sinews of Power: War, Money and the English State 1688- 1783. Routledge.

27 Centeno, M. A. (2002). Blood and Debt: War and the Nation-state in Latin America. Penn State Press.

Chattopadhyay, R. and Duflo, E. (2004). Women as policy makers: Evidence from a randomized policy experiment in india. Econometrica, 72(5):1409–1443.

Chaudhry, K. A. (1989). The price of wealth: Business and state in labor remittance and oil economies. International Organization, 43(1):101–145.

Dahlstr¨om,C. and Lapuente, V. (2017). Organizing Leviathan: Politicians, Bureau- crats, and the Making of Good Government. Cambridge University Press.

Dell, M., Lane, N., and Querubin, P. (2017). The historical state, local collective action, and economic development in vietnam.

Dincecco, M. (2011). Political transformations and public finances: Europe, 1650– 1913. Cambridge University Press.

Dincecco, M., Federico, G., and Vindigni, A. (2011). Warfare, taxation, and political change: Evidence from the italian risorgimento. The Journal of Economic History, 71(4):887–914.

Dincecco, M. and Wang, Y. (2017). Violent conflict and political development over the long run: China versus europe. Annual Review of Political Science.

Elman, B. (2000). A Cultural History of Civil Examinations in Late Imperial China. University of California Press.

Evans, P. (2012). Embedded Autonomy: States and Industrial Rransformation. Princeton University Press.

Fearon, J. (1999). Electoral accountability and the control of politicians: Selecting good types versus sanctioning poor performance. Democracy, Accountability, and Representation, 55:61.

Garfias, F. (2018). Elite competition and state capacity development: Theory and evidence from post-revolutionary mexico. American Political Science Review, 112(2):339–357.

Guardado, J. (2018). Office-selling, corruption and long-term development in peru. American Political Science Review, 112:971–995.

Hendrix, C. (2010). Measuring state capacity: Theoretical and empirical implica- tions for the study of civil conflict. Journal of Peace Research, 47(3):273–285.

28 Herbst, J. (2014). States and Power in Africa: Comparative Lessons in Authority and Control. Princeton University Press.

Hui, V. T.-b. (2005). War and State Formation in Ancient China and Early Modern Europe. Cambridge University Press.

Kaske, E. (2011). Fund-raising wars: Office selling and interprovincial finance in nineteenth-century china. Harvard Journal of Asiatic Studies, pages 69–141.

Kuhn, P. (1970). Rebellion and Its Enemies in Late Imperial China: Militarization and Social Structure, 1796-1864, volume 49. Harvard University Press.

Li, L. M. (2007). Fighting famine in north china: State, market, and environmental decline, 1690s-1990s.

Manion, M. (2006). Democracy, community, trust: The impact of elections in rural china. Comparative Political Studies, 39(3):301–324.

Manion, M. (2015). Information for autocrats: Representation in Chinese local congresses. Cambridge University Press.

Manion, M. (2017). “good types” in authoritarian elections: The selectoral connec- tion in chinese local congresses. Comparative Political Studies, 50(3):362–394.

Mann, M. (2012). The Sources of Social Power: Global Empires and Revolution, 1890-1945, volume 3. Cambridge University Press.

Mares, I. and Queralt, D. (2015). The non-democratic origins of income taxation. Comparative Political Studies, 48(14):1974–2009.

Morrisson, C. and Snyder, W. (2000). The income inequality of france in historical perspective. European Review of Economic History, 4(1):59–83.

North, D. C. and Weingast, B. R. (1989). Constitutions and commitment: the evolution of institutions governing in seventeenth-century england. The Journal of Economic History, 49(4):803–832.

Pierre-Etienne, W. (1990). Bureaucratie et Famine en Chine au Dix-Huiti`emeSi`ecle. Ecole des hautes etudes en sciences sociales.

Pomeranz, K. (2009). The Great divergence: China, Europe, and the Making of the Modern World Economy. Princeton University Press.

Przeworski, A., Stokes, S., and Manin, B. (1999). Democracy, Accountability, and Representation. Cambridge University Press.

29 Qu, T. (1969). Local Government in China Under the Ching, volume 95. Stanford University Press.

Queralt, D. (2018). The legacy of war on fiscal capacity.

Rosenthal, J.-L. and Wong, R. B. (2011). Before and Beyond Divergence. Harvard University Press.

Stasavage, D. (2010). When distance mattered: Geographic scale and the develop- ment of european representative assemblies. American Political Science Review, 104(4):625–643.

Stasavage, D. (2016). Representation and consent: Why they arose in europe and not elsewhere. Annual Review of Political Science, 19:145–162.

Tilly, C. (1992). Coercion, Capital, and European States, AD 990-1990. Oxford Blackwell.

Tsai, L. (2007). Solidary groups, informal accountability, and local public goods provision in rural china. American Political Science Review, 101(2):355–372.

Wakeman, F. (1977). Fall of Imperial China. Simon and Schuster.

Xu, D. (1974). Qingdai juanna zhidu(Contributions in Qing Dynasty). Wen hai chuban she.

Xu, G. (2017). The costs of patronage: Evidence from the british empire. American Economic Review.

Zelin, M. (1992). The Magistrate’s Tael: Rationalizing Fiscal Reform in Eighteenth- century Ching China. University of California Press.

Zhang, Z. (1955). The Chinese Gentry. University of Washington Press.

Zhang, Z. (2011). Qingdai fu ting zhou dengji zhidu de queding(the ranking system of prefecture and county in the qing dynasty). Mingqing luncong (History of Ming and Qing), 11.

30 Figure 1: The Qin Dynasty Map and Its Comparison with Present-day Chinabu0

[][]

[][]

31 Figure 2: Government Expenditure and Wars

32 Figure 3: The Hierarchy of Imperial Chinese Civil Service Exam

Palace Tests (Hanlin Academy)

Metropolitan Exams( 20%)

Provincial Exams( 5%)

Prefectural Exams( 1.5%)

33 Figure 4: Example of Local Gazetteers

34 Figure 5: Total Famine Relief Efforts

35 Figure 6: Total Famine Relief Efforts (20 Years Moving Average)

36 Figure 7: Average Famine Relief Efforts (20 Years Moving Average)

37 Figure 8: Grand Canal in Imperial China

38 Figure 9: Treaty Ports in China

39 Figure 10: Marginal Effect of Being An Official by Purchase on Mobilizing Resources From Local Elites

40 Table 1: Percentage of Officials Becoming Local Officials Through Civil Ex- aminations, Inheritance, Purchase during the Qing Dynasty

Year No.of Officials Exam Inheritance Purchase Other 1764 2,071 72.5% 1.1% 22.4% 4% 1840 1,949 65.7% 1.0% 29.3% 4% 1871 1,790 43.8% 0.8% 51.2% 4.2% 1895 1,975 47.9% 1.2% 49.4% 1.5% Source: Ping-ti Ho, The Ladder of Success in Imperial China New York: Columbia University Press 1962), p. 49, table 2. Li Tie, Zhongguo Wenguan Zhidu (The Chinese Civil Service System)(Peking: Chinese Political University, 1989), p. 171.

41 Table 2: Bureaucracy as Determinant of State Capacity

Hypothesis Causal mechanisms Observable indicators Intrinsic motivation: Merit-based bureaucratic recruitment The process of socialization of intrinsic values of tax reduction increases state capacity. holding offices will curb over-extraction. Mobilizing resources: Officials who are selected into the bureaucracy connections with resource holders. through competitive exam are better able than others to mobilize resources. Loyalty: The political center prioritizes officials selected ethnic identity by competitive exam over others because they are more loyal.

42 Table 3: Format of Provincial and Metropolitan Civil Service Examinations During the Qing Dynasty, 1787-1792

Session No. No. of Questions ONE 1. Four Books 3 quotations 2. Poetry question 1 poetic model TWO 1. Change 4 quotations 2. Documents 4 quotations 3. Poetry 4 quotations 4. Annals 4 quotations 5. Rites 4 quotations 6. Discourse 1 quotation (dropped in 1793) THREE 1. Policy questions 5 essays Source: Elman, Benjamin (2000), A Cultural History of Civil Examinations in Late Imperial China. Berkeley: Uni- versity of California Press., p. 564.

43 Table 4: Qing Dynasty Policy Questions Classified by Topic: Zhejiang Province, 1646-1859 (460 Questions, Top 15 Ranks Only)

Rank Topic % of Total 1 Classical Studies 14.1 2 Learning/Political Selection 10.7 3 Economy 9.6 4 World Ordering 7.8 5 History 7.4 6 Tao-hsueh 6.1 7 Literature/Poetry 5.1 7 Local Governance 5.1 9 Philology 4.2 10 National Defense 3.8 11 Law 3.1 13 Agriculture 2.7 13 Military Matters 2.7 15 People’s Livelihood 2.2 Source: Elman, Benjamin (2000), A Cultural History of Civil Examinations in Late Imperial China. Berkeley: University of California Press., p. 720. The original data are from Pen-ch’ao Che- san-ch’ang ch’uan-t’i pei-k’ao (Complete listing of all questions from the three sessions of the Zhejiang provincial civil examinations during the Qing Dynasty). Compiled ca. 1860. World Ordering mainly refers to statecraft.

44 Table 5: Baseline Results

Dependent Variable: Famine Relief (1) (2) (3) (4) OLS OLS OLS OLS purchase -0.019 -0.008 -0.008 0.019 (0.050) (0.050) (0.050) (0.058) disaster 0.441 0.440 0.440 0.492 (0.059) (0.059) (0.059) (0.081) disaster*purchase -0.278 -0.282 -0.282 -0.303 (0.093) (0.093) (0.093) (0.121) Manchu -0.093 -0.093 -0.064 (0.052) (0.052) (0.068) newly appointed -0.001 -0.034 (0.051) (0.066)

Observations 8,491 8,491 8,491 5,921 R-squared 0.143 0.143 0.143 0.142 Year FE Y Y Y Y Province FE Y Y Y Y Time 1644-1911 1644-1911 1644-1911 1662-1820 Note: Bolded coefficients indicate statistical significance, with p<.05 adopted as level of significance.

45 Table 6: Income, Geography and Western Influence 0.000 0.133 -0.105 -0.044 (0.113) (0.133) 0.461 0.458 0.508 (0.130) (0.129) (0.158) 0.282 0.273 0.279 0.266 (0.053) (0.052) (0.053) (0.067) .05 adopted as level of significance. < Dependent Variable: Famine Relief 0.247 0.247 0.207 0.179 0.183 (0.060) (0.060) (0.058) (0.058) (0.059) (0.070) 0.268 0.264 0.264 0.353 0.396 0.410 0.456 (0.106) (0.105) (0.105) (0.111) (0.117) (0.121) (0.151) (1) (2) (3) (4) (5) (6) (7) (8) OLS OLS OLS OLS OLS OLS OLS OLS 0.418 0.4200.000 0.410 0.000 0.410 0.000 0.410 0.000 0.409 0.000 0.408 0.000 0.458 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 -0.279 -0.279 -0.262 -0.262 -0.267 -0.258 -0.254 -0.286 (0.050) (0.050)(0.058) (0.050) (0.058)(0.093) (0.050) (0.058) (0.093) (0.050) (0.058) (0.092) (0.050) (0.058) (0.092) (0.050) (0.058) (0.092) (0.058) (0.058) (0.092) (0.079) (0.092) (0.118) (0.000) (0.000)(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Note: Bolded coefficients indicate statistical significance, with p purchasedisaster disaster*purchase 0.014population 0.015 0.005access to main river 0.005rice suitability 0.007terrain ruggedness treaty 0.006 post 0.008ObservationsR-squaredYear 0.023 FEProvince FEIndividual 8,491 ControlsTime 0.154 8,491 Y Y Y 0.155 8,491 Y 1644-1911 0.156 8,491 1644-1911 Y Y 1644-1911 1644-1911 0.156 8,491 Y 1644-1911 1644-1911 Y 1644-1911 Y 0.158 8,491 1662-1820 Y 0.159 8,491 Y Y 0.159 5,986 Y Y Y 0.155 Y Y Y Y Y Y Y Y Y size coastal city

46 Table 7: Information Cost

Dependent Variable: Famine Relief (1) (2) (3) (4) OLS OLS OLS OLS purchase -0.014 -0.005 0.037 0.036 (0.050) (0.050) (0.069) (0.085) distance -0.001 -0.001 -0.001 -0.001 (0.000) (0.000) (0.000) (0.000) disaster*purchase -0.264 -0.268 -0.266 -0.283 (0.093) (0.093) (0.106) (0.138)

Observations 8,491 8,491 6,967 4,662 R-squared 0.145 0.145 0.182 0.180 Year FE Y Y Y Y Province FE Y Y Y Y Individual Controls N Y Y Y Prefecture Controls N N Y Y Time 1644-1911 1644-1911 1644-1911 1662-1820 Note: Bolded coefficients indicate statistical significance, with p<.05 adopted as level of significance.

47 Table 8.A Mobilizing Resources from Central Government Dependent Variable: Famine Relief (1) (2) (3) OLS OLS OLS purchase -0.121 -0.120 -0.170 (0.095) (0.095) (0.094) official*purchase -0.003 -0.003 -0.000 (0.002) (0.002) (0.002)

Observations 4,563 4,563 4,563 R-squared 0.205 0.206 0.212 Year FE Y Y Y Province FE Y Y Y Individual Controls N Y Y Prefecture Controls N N Y Time 1644-1911 1644-1911 1644-1911 Table 8.B Mobilizing Resources from Local Elites Dependent Variable: Famine Relief (1) (2) (3) OLS OLS OLS purchase 0.153 0.141 0.110 (0.162) (0.163) (0.157) quota*purchase -0.002 -0.002 -0.002 (0.001) (0.001) (0.001)

Observations 4,563 4,563 4,563 R-squared 0.212 0.212 0.219 Year FE Y Y Y Province FE Y Y Y Individual Controls N Y Y Prefecture Controls N N Y Time 1644-1911 1644-1911 1644-1911 Note: Bolded coefficients indicate statistical significance, with p<.05 adopted as level of significance.

48 Table 9: Intrinsic Motivation

Dependent Variable: Tax Reduction (1) (2) (3) OLS OLS OLS purchase 0.039 0.037 0.126 (0.134) (0.135) (0.134) disaster 0.612 0.613 0.501 (0.130) (0.130) (0.126) disaster*purchase -0.537 -0.548 -0.457 (0.216) (0.216) (0.213)

Observations 8,492 8,492 8,492 R-squared 0.243 0.244 0.278 Year FE Y Y Y Province FE Y Y Y Individual Controls N Y Y Prefecture Controls N N Y Time 1644-1911 1644-1911 1644-1911 Note: Bolded coefficients indicate statistical significance, with p<.05 adopted as level of significance.

49 Table 10: Competing Mechanism: Loyalty

Dependent Variable: Famine Relief (1) (2) OLS OLS purchase 0.132 0.071 (0.084) (0.118) disaster*purchase -0.283 -0.130 (0.176) (0.185)

Observations 2,345 1,933 R-squared 0.235 0.293 Year FE Y Y Province FE Y Y Individual Controls Y Y Prefecture Controls N Y Time 1644-1911 1644-1911 Note: Bolded coefficients indicate statistical significance, with p<.05 adopted as level of significance.

50 Appendix A Descriptive

A.1 Data Quality

Figure 11 presents the available observations by each year. As we can see, the data quality is better between the time period from 1660 to 1820. I use this restricted time period in my empirical analyses. In the future, I will complete data collection.

Figure 11: Number of Available Observations

A.2 Summary Statistics

Table 11 presents the summary statistics of my dataset.

51 Table 11: Summary Statistics

Variables Obs Mean SD Min Max purchase 8492 0.274 0.446 0 1 relief 13399 0.566 2.488 0 54 tax reduction 13400 2.256 5.637 0 128 tax revolt 5106 0.028 0.193 0 3 turnover 8492 0.364 0.518 0 4 Manchu 8492 0.276 0.447 0 1 official 13400 37.2 53.756 1 235 quota 13400 172.92 91.170 35 423 population 13400 2120000 1560000 170664 6845663 size 13400 14970.22 8635.051 2511 44260 coast 13400 0.2 0.4 0 1 per capita tax in 1820 13400 0.113 0.104 0.009 0.641 difficulty to tax() 13400 0.32 0.466 0 1 main river 13400 0.72 0.449 0 1 rice 13400 1.923 0.876 0 3.556 distance 13400 972.712 536.525 39.786 2080.143 treaty 13400 0.1 0.3 0 1 disaster 13400 0.147 0.354 0 1

52 Appendix B Additional Empirical Analyses

B.1 Selection Bias

Table 12: Balance Table

Variable exam purchase difference t-value rice suitability 1.949 2.038 -.089 -4.348 (0.011) (0.017) (0.020) main river 0.692 0.704 -0.012 -1.070 (0.006) (0.009) (0.011) terrain ruggedness index 0.175 0.172 0.002 0.352 (0.004) (0.007) (0.008) coast 0.209 0.203 0.005 0.550 (0.005) (0.008) (0.010) distance 1005.621 1014.178 -8.556 0.619 (7.259) (11.623) (13.814) tax revolt 0.022 0.014 0.021 0.951 (0.004) (0.007) (0.003) disaster 0.538 0.535 0.004 0.294 (0.006) (0.010) (0.012)

Table 13: Selection Bias

Dependent Variable: Purchase (1) (2) (3) (4) OLS OLS OLS OLS

disaster 0.001 -0.004 -0.004 -0.005 (0.018) (0.017) (0.017) (0.017) disaster t-1 -0.007 -0.007 (0.018) (0.018) disaster t-2 -0.013 (0.012)

Observations 8,492 6,968 6,967 6,967 R-squared 0.069 0.083 0.083 0.084 Year FE Y Y Y Y Individual Controls Y Y Y Y Prefecture Controls N Y Y Y Time 1644-1911 1644-1911 1644-1911 1644-1820 Province FE Y Y Y Y Note: Bolded coefficients indicate statistical significance, with p<.05 adopted as level of significance.

53 B.2 Early State Investment

Although I have controlled for potential endogeneity by including fixed effects, a different interpretation of my findings is that they reflect the cumulative effects of early state capacity, namely, investment in local infrastructure had a negative cumulative effect on the probability of disaster occurring and thus, led to fewer famine relief efforts. In this case, more famine relief efforts actually imply lower state capacity. If early state investment in local infrastructure was the main driver for fewer occurrence of disasters, we would expect to see a statistically significant negative relationship between them. I measure early state investment in local in- frastructure in two ways: (1) the per capita tax in 1820 at prefectural level, (2) the designations by the Qing state indicating characteristics of prefectures pi (difficult to gather taxes)11. Table 14 shows that early state investment in infrastructure has no significant impact on the probability of disaster occurring, implying that famine relief is not an inverse measure of state capacity.

Table 14: Early State Capacity

Dependent Variable: Disaster (1) (2) OLS OLS

taxpc1820 -0.088 (0.123) difficulty of taxation(pi) -0.004 0.033 (0.023) (0.042)

Observations 5,270 1,940 R-squared 0.189 0.177 Year FE Y Y Province FE Y Y Individual Controls Y Y Prefecture Controls Y Y Time 1723-1911 1820-1911 Note: Bolded coefficients indicate statistical significance, with p<.05 adopted as level of significance.

To further illustrate this relationship, I regress the variable disaster against purchase. As Table 15 shows, the variable purchase at both time t and t-1 has no statistically

11The data on land tax at prefectural level were rare. I restrict the sample to the time period from 1820 to 1911 when I use the per capita tax in 1820. In 1723, the Qing central state started ranking prefectures based on the difficulty of governance. Pi, the difficulty to tax, is a proxy for fiscal capacity at prefectural level. It is an indicator variable that takes the value of 1 if it had been difficult to collect taxes and 0 otherwise. I restrict the sample to the time period from 1723 to 1911 when I use pi

54 significant impact on whether a prefecture experienced disaster, further confirming that officials exhibited different abilities and harbored different motivations when implementing policies designed by the political center. Table 15: Early State Capacity (II)

Dependent Variable: Disaster (1) (2) (3) (4) OLS OLS OLS OLS

purchase 0.001 -0.003 0.003 -0.001 (0.008) (0.010) (0.010) (0.012) purchase t-1 -0.000 -0.001 (0.010) (0.012)

Observations 8,492 6,968 5,789 4,692 R-squared 0.125 0.133 0.143 0.152 Year FE Y Y Y Y Province FE Y Y Y Y Individual Controls Y Y Y Y Prefecture Controls N Y N Y Time 1644-1911 1644-1911 1644-1911 1644-1911 Note: Bolded coefficients indicate statistical significance, with p<.05 adopted as level of significance.

B.3 Extraction

A competing explanation for the empirical results is that the variable famine relief actually measures officials’ preferences for extraction as they were able to divert resources from the political center to their jurisdiction. To alleviate this concern, I use the data on tax revolt from 1800 to 1911 to test this relationship. The variable tax revolt is the number of tax revolt occurred in a prefecture i and year t. It is reasonable to assume that the number of tax revolts was higher in a prefecture with a corrupt official because he was more likely to be involved in collecting surcharges. If famine relief captures extraction rather than local state capacity, we would expect to see that famine reliefs predict tax revolts. As Table 16 shows, from Columns 1 to 3, past and current famine relief efforts cannot predict tax revolts.

B.4 Spatial Correlation of Errors

Given the nature of famine relief, spatial clustering may affect the validity of the results. It is very likely that famine reliefs are not independent from each other

55 Table 16: Extraction and Famine Relief

Dependent Variable: Tax Revolt (1) (2) (3) OLS OLS OLS

purchase -0.003 -0.003 -0.003 (0.008) (0.008) (0.008) relief 0.001 0.001 0.001 (0.002) (0.002) (0.002) relieft−1 -0.001 -0.001 (0.001) (0.001) relieft−2 -0.000 (0.002)

Observations 2,512 2,512 2,512 R-squared 0.146 0.146 0.146 Year FE Y Y Y Province FE Y Y Y Individual Controls Y Y Y Prefecture Controls Y Y Y Time 1800-1911 1800-1911 1800-1911 Note: Bolded coefficients indicate statistical significance, with p<.05 adopted as level of significance. in the same prefectures. I re-estimate the models using prefecture fixed effects and clustered standard errors at the prefectural level in Table 17. The variable disaster*purchase remains statistically significant and negative in Columns 1 to 4.

56 Table 17: Spatial Clustering

Dependent Variable: Famine Relief (1) (2) (3) (4) OLS OLS OLS OLS purchase 0.079 0.066 0.079 0.096 (0.074) (0.075) (0.074) (0.077) disaster 0.408 0.409 0.408 0.446 (0.087) (0.088) (0.087) (0.099) disaster*purchase -0.255 -0.250 -0.255 -0.265 (0.114) (0.114) (0.114) (0.124) new appointment -0.018 -0.018 -0.058 (0.075) (0.075) (0.089) Manchu -0.110 -0.110 -0.088 (0.071) (0.071) (0.090) population 0.000 0.000 0.000 (0.000) (0.000) (0.000) size 0.000 0.000 0.000 (0.000) (0.000) (0.000) coastal city 0.291 0.239 0.489 (0.070) (0.076) (0.083) access to main river 0.069 0.135 (0.022) (0.020) rice suitability 0.223 0.398 (0.053) (0.063) terrain ruggedness index -0.564 -0.629 (0.100) (0.082) treaty post -0.076 -0.166 (0.035) (0.048)

Observations 8,491 8,491 8,491 5,986 R-squared 0.178 0.177 0.178 0.177 Year FE Y Y Y Y Province FE Y Y Y Y Prefecture FE Y Y Y Y Individual Controls Y N Y Y Prefecture Controls N Y Y Y Time 1644-1911 1644-1911 1644-1911 1662-1820 Note: Bolded coefficients indicate statistical significance, with p<.05 adopted as level of significance.

57