The Incentive Effect of Abolishing China's Imperial Civil Examination

Home , Civil service, Jinshi, Juren, Test (assessment)

The Incentive Effect of Abolishing China’s Imperial Civil Examination on the Adoption

of Western Technology: A Regression Discontinuity Design

BAI, YING * Hong Kong University of Science and Technology

August 2013

Abstract Compared with the rigid political system of feudal Europe, Imperial China was a less oligarchic society that employed a civil examination to select scholar bureaucrats as ruling elites. This institution created a negative incentive to adopt West- ern technology and establish modern enterprises. Using prefectural-level panel data for the 1896-1910 period, this study compares the effects of the chance to pass the civil examination on the establishment of private firms adopting Western technology before and after the abolition of the examination system. Its findings show that prefectures with a higher quota of successful candidates tended to establish more such firms once the examination system was abolished. As higher quotas were assigned to prefectures whose agricultural tax in the Ming Dynasty exceeded 150,000 stones, we adopt a regression discontinuity design to resolve the potential endogeneity, find the results to remain robust.

Key Words: Incentive, Western Technology, Imperial Civil Examination JEL Code: N95, O10, O31

______

* Division of Social Science, Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong. Email: [email protected]

I. Introduction

Institutions and the incentives they create determine the economic outcomes of a society (North, 1990, 1994). The recent literature suggests that good institutions, including secure property rights (Acemoglu and Johnson, 2005; North and Thomas, 1973), efficient legal systems (La Porta, Lopez-de-Silanes, Shleifer and Vishny 1997, 1998), and fewer barriers to entry or the absence of oligarchies (Acemoglu, 2008; Acemoglu, Cantoni, Johnson and Robin- son, 2011; Olson, 1982), are correlated with better economic performance. Institutional roles have taken center stage in explaining why some regions, particularly Western Europe, experi- enced rapid technological progress and an unprecedented period of sustained growth from the

16th century onwards, which has shaped today’s global income distribution (Acemoglu, John- son and Robinson 2001, 2002, 2005; Galor, Moav and Vollrath 2009).

In the literature on new institutional economics, medieval Europe is regarded as an extremely oligarchic society is which political power rested in the hands of a small landed aristocracy and there were explicit entry barriers for other groups. Medieval England, for instance, was characterized by a rigid class system and a small landed nobility that dominated the sphere of politics. Under this rigid political system, the rising merchant class had no chance of becoming a ruling elite or sharing political power. Political reforms, such as those that took place in the wake of the 1688 Glorious Revolution in England and the French Revo- lution in the 19th century, removed many of the legal barriers protecting the ruling nobility

(Acemoglu, Cantoni, Johnson and Robinson 2010). These institutional changes also intro- duced the protection of property rights and equity before the law, which paved the way for the Industrial Revolution and sustained economic growth (Acemoglu, Johnson and Robinson

2005; Jones 1981; De Long and Shleifer 1993; North and Thomas 1973). This body of litera-

1 ture suggests that the rigidity of the former system compelled economic elites to enrich themselves and then demand and obtain favorable institutions.

In late Imperial China, in contrast, political power was much more diffused, and the society less oligarchic. A large part of the ruling class – scholar bureaucrats – did not have hereditary positions, but were selected on the basis of the imperial civil examination. There was no “effective legal barriers to status mobility,” according to Ho (1976: 54). Historical evidence shows that the land gentry and merchants translated “their economic and social power into cultural and educational advantages that enabled mainly the sons of gentry and merchants to pass the civil examinations” (Elman 2000: xix), thus allowing these elites to obtain political power through the political system (Elman 2000). Compared with the European feudal system, this institution provided the Chinese elites with great incentives to invest in a traditional education and take the civil examination. The knowledge required to pass the examination focused on memorization of the Confucian classics and other less productive do- mains (Jones 2008; Yuchtman 2009). Hence, the Chinese elites had few incentives to study other subjects, such as mathematics and science, which played key roles in launching the In- dustrial Revolution and a period of sustained economic growth in Europe (Clark and Feenstra,

2003; Huff 2003; Lin 1995).

This paper examines the negative incentive effect of the civil examination system on growth-promoting activities and the adoption of advanced Western technology by exploring the effect of the system’s abolition.1 The main empirical difficulty is that the abolition was universal, with no regional variation in policy implementation. To better understand the adoption of Western technology that resulted from the examination system’s abolition, we derive a

1 The decision to abolish the civil examination was announced on January 13, 1904, and it was officially abolished on September 2, 1905.

2 simple model incorporating two choices open to Chinese families: to adopt Western technology and establish a modern firm or to invest in preparation for the civil examination. In this model, it is assumed that families with a greater chance of passing the examination would be less likely to adopt Western technology and that these families would tend to establish more modern firms after its abolition. Accordingly, the regions with the greatest numbers passing the examination should be the most affected, which makes it possible to employ a difference- in-differences (DID) method to identify the causal effect of abolishing the civil examination on the adoption of Western technology.

We exploit the variation in the probability of passing the examination among prefectures – a level that falls between the provincial and county levels. To control the regional composition of successful candidates, the central government allocated a quota of successful candidates to each prefecture.2 In terms of the chances of individual participants, there were great inequalities among regions (Chang 1955). Using the number of private industrial firms above a designated size, with mechanization as a proxy for the adoption of Western technology, we compare the effects of these quotas on the establishment of private modern enterprises before and after the abolition of the examination system. Our empirical results show that the prefectures with a higher quota tended to establish more of these enterprises once the decision to abolish the examination system was taken on January 13, 1904.

However, it is possible that the quotas are correlated with other omitted variables, such as economic prosperity or Confucian culture, whose effects may also have changed after the system’s abolition. To correct such biases and resolve the potential endogeneity, we em-

2 China’s civil examination system included three stages of testing: a licensing examination in the prefectural capital, a qualifying examination in the provincial capital, and an academy examination in the national capital, with re-examination in the imperial palace. Only candidates who were successful in the lower-level examinations were eligible to sit for the higher levels. The central government allocated a quota of successful candidates in the licensing examination to each prefecture and county, whereas the quota for the higher level examination applied to the provinces.

3 ploy a regression discontinuity design, and find our results to remain robust. In the Ming

Dynasty (1368-1643), the prefectures whose agricultural tax exceeded 150,000 stones3 were classified as the most important. Higher quotas were allocated to the most important prefectures in the early Qing Dynasty (1644-1910). Mid-way through the period, a new rule for classifying regions by importance was adopted, but the quotas remained stable throughout the

Qing Dynasty.

This study makes three contributions to the literature. First, it enhances our under- standing of the role played by institutions in economic development. Extant studies find oligarchic societies to erect entry barriers, which creates inefficiency and induces decline in the long term (Acemoglu 2008). In this paper, we propose that in non-democratic societies, more oligarchy may lead to elites without political power exploiting other economic opportunities and attempting to change the political system through wealth. Less oligarchic societies, in contrast, provide incentives to join the political game, and thus other economic opportunities are forgone.

Second, this study sheds light on the reasons for the great divergence in technological expertise and per capita income seen across the globe in the past five centuries. A multitude of hypotheses, ranging from factor endowments and natural resources (Jones 1981; Diamond

1997; Sachs and Warner, 1995) to cultural and scientific traditions (Weber, 1930, 1964), and political and economic institutions (Acemoglu, Johnson and Robinson 2001, 2002, 2005;

Galor, Moav and Vollrath 2009), have been proposed. We suggest here that the rigidity of the

European political system in medieval times played an unexpected role in the bifurcation of the West and China. The rigid feudalism of medieval Europe stimulated landless elites to en-

3 “Stone” is the English word for a historical Chinese unit of mass. The Chinese character is pronounced Dan as a unit of measure.

4 rich themselves, thereby inducing political reform, whereas China’s less rigid system encour- aged elites to invest in a non-productive traditional education and gave them a negative incentive to adopt advanced technology.

Third, previous studies on the great divergence have focused primarily on the rise of

West Europe, with no systematic empirical effort to explain the decline of Imperial China.

Namely, why did China fail to undergo an industrial revolution or sustained growth, given its economic leadership in the 14th century and even in the 18th (Allen, 2004; Maddison 1998,

2001; Pomeranz 2000)? Landes (2006) describes two opportunities that China missed: 1) the opportunity to generate a self-sustaining process of technological advance based on its achievements and 2) the opportunity to learn from European technology once foreigners had entered the Chinese domain. This paper identifies the effects of the second missed opportunity empirically and supports the extant hypothesis that the civil examination constituted an institutional obstacle to the rise of modern science and industry (Huff 2003; Lin 1995).

The remainder of the paper is organized as follows. In Section II, we provide a brief review of the imperial civil examination in China, and, in Section III, introduce a simple economic model and our empirical specifications. We define the variables and introduce our data sources in Section IV. Section V presents our baseline results and the results of robustness checks, and we employ the regression discontinuity design in Section VI to resolve the potential endogeneity. Section VII summarizes our findings and concludes the paper.

II. Brief Introduction to the Chinese Imperial Civil Examination System

5 In AD 605, the Emperor Yang of the Sui dynasty (581 – 618) established the imperial examination system, which intellectuals were formally selected to be officials (Keay, 2008).4

Although the system was used on a small scale during this and the subsequent Tang dynasty

(618 – 907), it was expanded under the Song dynasty (960 – 1276) (Chaffee, 1995). After being interrupted during the Mongol Yuan dynasty (1276- 1368), the examination system became the primary channel for recruiting government officials during the Ming (1368 – 1644) and Qing (1644 – 1910) dynasties (Ebrey, 1999; Elman, 2000; Ho, 1962). The system was designed to select the best administrative officials for the state’s bureaucracy through a test of educational merit, and directly created a class of scholar bureaucrats independent of family status (Elman, 2000; Ho, 1962).5 The system helped the central government to capture and hold the loyalty of local-level elites (Man-Cheong, 2004), selected suitable and efficient managerial elites (Elman, 2000), and created cultural uniformity and consensus on basic values (Elman, 1990, 1991).

The structure and process of the civil examination system remained relatively stable especially in the late imperial period of China (the Ming and Qing dynasties). There were three stages of testing: the biennial licensing examination (Yuankao in Chinese), the lowest level, held in the prefectural capital after primary testing in the county seat;6 the triennial qualifying examination in the provincial capital (Xiangshi in Chinese); and the third level in the capital (academy examination, Huikao in Chinese) with re-examination in the imperial palace (palace examination, Dianshi in Chinese). Successful candidates received a new title

4 The beginnings of this practice can be found in the Han dynasty (Chaffee, 1995; Elman, 2000). 5 Ho (1962) discussed “the lack of effective legal barriers preventing the movement of individuals and families from one status to another” (Ho, 1962: 54) and shows the upward mobility was possible due to the civil examination system. However, Elman (2000) argued that the civil examination “was not a system designed for increased social mobility. Instead, it served as an institutionalized system of inclusion and exclusion that publicly legitimated the impartial selection of officials.” 6 New candidates had to be be chosen by magistrates and prefects for the county, department, and prefectural tests, before they qualified as “apprentice candidates”. After they were screened by the primary test, the successful candidates took a final licensing examination in the prefectural capital (Elman, 2000).

6 at each level and became automatically eligible to attend the next level of examination. Spe- cifically, the candidates who passed the licensing examination were termed Shengyuan or

Xiucai (literati), and those who passed the provincial, the academy, and the palace testing were Juren (recommended man, a provincial graduate), Gongshi (tribute personnel, a national degree), and Jinshi (presented scholar, a graduate of the palace examination), respectively.

The contents of all the examinations were dominated by the Confucian classics – the Four

Books and the Five Classics (Elman, 2000). The examinations were bound by many regula- tions and presented within strict frameworks. Form was even more important than content.

Under this examination system, candidates were required to grasp the Confucian classics and be able to write the eight-legged essays (Chang, 1955; Elman, 2000; Yuchtman, 2009).

Each of the degrees or titles carried different privileges, prestige, and income (Chang,

1955, 1962; Elman 1990, 2000; Glahn, 1996). Chang (1962) shows that the gentry received about 24 per cent of the national income, even though they constituted only about 2 per cent of the population. At the lowest level, the candidates who passed the district examination became members of the gentry (lower gentry), who were exempted from corporal punishment and corvee, and had the right to wear a scholar’s robes. This degree and title also provided the gentry with the opportunity to manage local affairs, become secretarial assistants to officials, and to teach – three important sources of income for Chinese gentry (Chang, 1962). The highest possible achievement was to become a government official, which brought great power and prestige. For instance, the district magistrate had great authority to carry out court orders, collect taxes, and implement the policies of the central government, all of which provided “the greatest opportunity for the rapid accumulation of wealth” (Chang, 1962: 7). Ni and Van (2006) estimate that circa 1873, the corrupt income of officials was14 to 22 times greater than their regular income, resulting in about 22% of the agricultural output being

7 owned by 0.4% of the population. Chang (1962) estimates that the average regular and extra income of a government official at that time was about 5,000 Tael silver per year. To qualify for appointment to government office, the candidates needed to pass at least the provincial level examination. If the candidates passed the higher-level academy or palace examinations, they were usually automatically appointed to a government position. In pursuit of the great power and prestige of an official, candidates began studying as young children and on average passed the district examination at 24. They were likely to continue to study for another ten years before passing the district examination.7

The dynasties distributed these opportunities to become rich according to provincial and prefectural quotas over the whole country “as an institutional means to confine and regu- late the power of elites” (Elman, 2000: 140). The quota system allowed the government to control the size of the gentry class, which consisted exclusively of candidates who had passed the licensing examinations. In the late empire, only 1.6 to 1.9 percent of the total population had gentry status (Chang, 1955; Elman, 2000).8 Furthermore, by using regional quotas, imperial officials could be recruited from the whole country. The control of the regional composition of the gentry was inherent in the initial stages of the licensing examination system. By restricting the prefecture or county quotas, the dynasty could control the number and location of candidates entering the official selection process. The quotas for the higher level examinations – Juren and Jinshi – were always allocated at the provincial level during the

Qing dynasty. There was no standard formula for the regional quota, but the size and importance of the administrative units were the key determinants (Chang, 1955). Thus, there

7 Chang (1955) estimates the average age at which examination candidates passed the district examination as 24, and suggests they passed the provincial and national examination at approximately 30 and 35 respectively. 8 Chang (1955) estimates the percentage of gentry in the total population was 1.2 percent in the early nineteenth century and increased to 1.9 percent in the late nineteenth century.

8 were great inequalities among different regions in terms of opportunities for individual participants.

After more than 1200 years of use, the introduction of Western knowledge systems led to a series of reforms after the First and Second Opium Wars and the decline of the imperial examination system. Wolfgan (1960) gives a detailed description of this phase of history.

After the defeat in the First Sino-Japanese War, some Chinese elites such as Kang, You-wei, and Liang, Qi-chao, recommended reforms to the examination system. In 1898, during the

Hundred Days’ Reform (11 June – 21 September), the reform party abolished the eight- legged essay from all examination levels, and modernized the traditional system by replacing

Confucian classics with mathematics and science. However, the Empress Dowager Cixi, the de facto ruler of the Qing Dynasty from 1861 to 1908, ended the reform and the imperial civil examination system resumed. After the Siege of the International Legations during the Boxer

Uprising in 1900, the Empress Dowager Cixi was forced to reform the traditional examination system. She abolished the eight-legged essay in 1901, but retained the three level examination structure. In late 1903 and early 1904, the Committee on Education submitted a memorandum urging the abolition of the examination system, and they received imperial approval on 13 January 1904. This led to the decisive abolition of the examination system. On 2 Sep- tember 1905, The Empress Dowager Cixi endorsed a memorandum ordering the discontinu- ance of the old examination system at all levels in the following year.

III. Economic Model and Empirical Specifications

III.A.1. Description

There are I infinitely-lived households with identical linear inter-temporally prefer- ences in the economy, each endowed with 1 unit of time per period. The household has a

9 fixed quantity of unskilled labor L and a fixed quantity of skilled labor L . The output of a

A household in period t is given by the aggregate output of the agricultural sector yt , and the

M manufacturing sector yt such that,

A M yytt y t. (1)

Production in the agricultural sector uses unskilled labor L and a type of intermediate good

xt . The output produced at period t is therefore

A  1 yxLtt . (2)

The intermediate goods are provided and monopolized by the government, including things such as infrastructure, irrigation facilities, public securities, and national defense. These intermediate goods are important for production and there will be no output if no intermediate

A goods are provided, so yLt (0, ) 0 . xt is produced using a specific group of labor ( St ) with a constant wage s . The production function can be written as

xtt S , (3)

where St can be regarded as troops, policemen, and other minor officials who are not selected through the civil examination process. Production in the manufacturing sector uses skilled

labor ( nt ) with adopted Western technology ( ), according to the formula

M yntt  . (4)

A household can allocate skilled labor, mt , to write the civil examination, join the government, and eventually monopolize the supply of intermediate goods. If the candidates pass the examination in period t , they will be able to produce the intermediate goods in peri-

od t 1. We assume the probability of passing the examination in period t 1 is tt1()m .

The probability function is increasing and concave, tt1()0m  and tt1()0m  , which

10 means that increased input will increase the likelihood of passing examination at a decreasing rate.

The objective of the government, the intermediate monopolist, is to maximize the profits over the current period. The inverse demand curve facing a monopolist charging the price is the marginal product of the intermediate goods, described as

A yxLtt(,)  11 ptt xL. (5) xt

Thus, the problem of the monopolist can be written as

1 tttttmax(p sx ) xL sx. (6) xt

From the first-order condition, xt and  t can be determined by

 2 1 xLx()1  t s . (7) s(1 ) 2 1  ()1 L  t  s

III.A.2. Skilled Labor Allocation

The problem that households face is determining the allocation of skilled labor between the manufacturing and the civil examination systems that will maximize expected income. This can be expressed in the following way:

 vt vt A M  maxEr ( yvvvvv px  )  Er [(1  ) y v  y v v ] vt vt , (8)  st. . nvv m L

where r is time preference, yv is the total output based on equation (1), pv is the price of

the intermediate goods based on equation (5), and tt is the benefits from the government based on the realized probability. Based on Equations (2), (5), and (7), the income of each

11 AM period can be written as (1 ) yyvv v . We describe this problem using the value function, such that

AM VyyrEVtttttttmax[(1 )1 ] nmtt, . (9)  st. . ntt m L

By this Bellman equation, the multi-period optimization is turned into a two-stage

problem. The household allocates mt to the civil examination system to equalize the margin-

M al profit of rEVtt1 with the marginal product of yt , which can be represented by the following first-order condition:

  rm tt1(). (10)

The left side of Equation (10) is the marginal product of the manufacturing sector and the right side is the marginal benefits of writing the civil service examination. Intuitively, we define the right side as the “marginal cost of adopting Western technology” into the manufacturing sector and the left side as the “marginal benefit of adopting Western technology”. As

tt1()0m  and tt1()0m  , the increase of labor flow to the civil service will decrease with the “marginal cost of adopting Western technology” and will eventually equalize the marginal benefit  . Equation (10) holds only if rL ( )  r  (0) . All skilled labor will be allocated to the manufacturing sector when  rL  ( ) and to the civil examination system when  r  (0) .

III.B. Empirical Specifications

The probability of passing the civil examinations is determined by the quota of successful candidates (Q ) and the number of candidates ( M ). Specifically, higher quotas will

12 increase the probability of passing the tests, and more candidates will decrease the probability.

Thus, the probability function tt1()m can be defined as

QQ  ()mmLn ( ). (11) tt1 MM t t

Thus, for the jth household the first-order condition can be rewritten as

rQ m  (1jt, ) . (12) M M

JJrQ m Because  (1jt, ) and HNM  , then jj11M M

r NH Q, (13)  where J and H represents the number of households and the total skilled labor in the economy. In this paper, there are no systematic labor data for the adoption of Western technology, but we can employ the number of modern enterprises ( Y ) as a proxy and assume that

N Y  , where  can be regarded as the average number of skilled laborers employed in a  firm, thus

Hr YQ . (14)  

The marginal effect of the quota on the number of firms is

YHr1   . (15) QQ 

H The first term describes the effect of quotas on the number of skilled laborers, and we Q define it as the “human capital” effect. This effect can be interpreted in several ways: the prefecture with more skilled labor might be allocated a higher quota, or higher quotas might attract more labor. The size of the quota may be positively related to the amount of skilled la-

13 r bor in a prefecture. The second term ,  , describes the negative incentive effect of a quo-  ta on adopting Western knowledge into the manufacturing industry. This paper will try to identify this effect. If there are no civil examinations, all skilled laborers will be allocated to the manufacturing sector, and the effect of the quota can be written as

YH1   . (16) QQ 

The main method used to identify the dis-incentive effect of the civil examinations is to compare the effects of quota size on the number of new modern enterprises before and after the abolition of the civil examination system, such that

YQZ00   00 iiii. (17) 1 1 1 1 YQZiiii  

The first equation (17a) estimates the effect of quota size on the number of firms established

1 Hr before the abolition of the civil examinations, where  0  as described in Equa-  Q  tion (15); the second equation (17b) estimates the effect of quotas after the abolition, and

1 H r  1  . The difference between  1 and  0 , namely 10    , is the negative  Q  incentive effect of the quota on the adoption of Western technology.

IV. Data

To conduct an analysis of the relationship between the abolition of the imperial civil examination system and entrepreneurship in Late Qing China, we construct a panel dataset that covers 263 prefectures in 18 provinces south of the Great Wall, between 1896 and 1910.

The locations are shown on the map of Qing China circa 1820 (Figure 1). We choose 1896 as our starting date because it marked the end of the First Sino-Japanese War (1 August 1894 –

14 17 April 1895), which signified the end of the Self-Strengthening Movement (1861 – 1895).

We choose 1910 as the end point of the analysis because the Xinhai Revolution of 1911 over- threw the Qing dynasty and attempted to establish a Republic in 1912. In the following section, we begin by defining the dependent variable, entrepreneurship, and then define the variables that are employed in the analysis.

Figure 1 about here

IV.A. the Adoption of Western Technology (Yit )

After the defeats in the First (1839-42) and Second (1856-60) Opium Wars, China began to adopt Western technologies and establish modern enterprises. Apart from the period of the Self-Strengthening Movement (1861-1895), initiated by Qing government officials to ac- quire military technology and armaments, this adoption was primarily carried out by private firms emerging in the mid-nineteenth century. We employ the number of mechanized industrial firms that are above a designated size as the proxy for measuring the adoption of West- ern technology. Chang (1989) compiled ten series of extant materials on Chinese private enterprises and listed all modern firms with their locations and establishment dates. All firms in this study meet the following five criteria: first, the firm is organized as a company; second, the capital is over 10 thousand dollars; third, mechanization is used; fourth, there are over 30 employees; fifth, the value of the output is over 50 thousand dollars. We plot the number of private enterprises by the year they were established in Figure 2.9 This figure clearly shows that the firms began in 1848, were negligible before 1878, and started to grow during the height of the Self-Strengthening Movement (1861-1895). After the defeat in the First Sino-

Japanese War (1894-1895) which marked the failure of the Self-Strengthening Movement, the number of newly opened firms suddenly decreased from 18 in 1894 to 3 in 1895 and 6 in

9 We only include the firms located in the 18 provinces covered in my empirical testing.

15 1896. After a period of slow growth (1895-1903), private enterprises sprouted at the end of

Qing Dynasty and the beginning of the Republic of China. For instance, 71 firms opened in

1916, the first year after the abolition of the imperial civil examinations, and 125 firms were established in 1912, the first year of the Republic of China.

Figure 2 about here

IV.B. The Quota for the Imperial Civil Examination (Q)i

In this paper, we use the quotas for the licensing examination - the lowest examination level, which was conducted in the prefectural capitals. This district examination was the threshold for gentry status. Candidates who passed the district examination became members of the gentry and earned the relevant privileges and income. The prefectural level of examination corresponds to our analysis of firms at the prefecture level. The quotas for successful candidates in these district examinations were assigned to each administrative unit, prefecture, and county. To compute this number, we first gathered the figures for the quotas in all administrative units and summed them. The Qing Hui Dian Shi Li (Kun, 1991), gives the quotas from the beginning to the end of the nineteenth century. According to Chang (1955), the district examination quotas were stable after 1724, with the biggest change happening during the

Taiping Rebellion, when the quota was increased as a result of quelling the revolt. Based on our data, the sum of the quotas in the 262 sample prefectures was 29,808 which was an increase from 24,698 before the Taiping Rebellion.10 Overall, during the period studied in this paper, the average prefectural quota, namely the number of successful candidates in licensing examination was 113.8 with a standard deviation of 75.8.

10 In Chang (1955), the sum of the quotas before the Taiping Rebellion was 25,089 and increased to 30,113 after the rebellion. The reason for the difference with our figures is we can not include the quotas for Eight Barner, Fengtian (Manchu) and the special quota for merchants.

16 IV.C. Control Variables ( Zi )

Geography is always a key factor in accounting for differences in economic prosperity.

To control for the differences in access to the coastline and navigable rivers, we include two dummy variables: coast – whether a prefecture is situated on the seaboard (coast); and riverside (Changjiang) – whether a prefecture is located along the Changjiang River, the most navigable river in China. Approximately 13.4 per cent of the sampled prefectures are located on the coast and 6.2 per cent are on the Changjiang River. Furthermore, we control for the longitude and latitude of the prefecture’s capital, as these might be correlated with conditions for agriculture and human health. In addition to geographical factors, we control for the size of a region because a larger region might have a higher quota of civil servants and more busi- nesses. To control for these effects, we include in the empirical estimates both the population density in 1880 and the size of the prefectures. In addition, provincial dummies and constant terms are also included. Details concerning the definition, data sources, and summary statistics of the pertinent variables are summarized in Table 1.

Table 1 about here

V. Empirical Testing

V.A. Baseline Results

To investigate the incentive effect of the abolition of the civil examination system, we first estimate the cross-sectional correlation between the quotas and the number of new private enterprises before (in 1903) and after the abolition (in 1904), and then compare the estimators, which can be represented by

YQZ00   00 iiiit, (18) 1 1 1 1 YQZiiiit  

17 in which  0 and  1 represent the effects of quota before and after the abolition, and

 10  is the incentive effect of the abolition of imperial civil examinations. Figure 3 shows that the effect of the quota was only 0.017 in 1903, but suddenly increased to 0.275 in

1904. The difference is 0.258 and is highly significant.

To examine the long-term effect that abolishing civil examinations had on entrepreneurship, we estimate the average effect of quotas during a period T , rather than only the first year before and after the abolition using the following equations:

YQZ00 T  000 it i i t it , (19) 1 1T 1 1 1 YQZit i i t it

0 1 0T where t and t represent the year dummies to control for year-specific effects and  and

 1T represent the effect of quotas before and after the abolition based on a period of T years.

The length of the period is variable. For instance, the length of the first period is two, when we compare the effects of quotas during 1904-1905 with the effect during 1902-1903, so that  21202  can be estimated. Overall, given different values of T (1,2,7T   ), a series of  TTT10  can be estimated. We graph the estimates in Figure 3, and find that

 0T and  1T are significantly different before and after the abolition of the imperial civil examinations. If a long period, such as seven years, is adopted the effect is only 0.065 before the abolition, and increases to 0.374 after the abolition. The difference is 0.310.

Figure 3 about here

V.B. Yearly Correlation between Quota and Entrepreneurship

The above analysis found that the effect of quotas significantly increased after the abolition. However, if the effect of quotas was gradually increasing before the abolition, our

18 estimator could be compounded with the pre-abolition trend. To examine this, we implement an alternative specification

1910  Yit  ( Q i Year dummy  t)  Z i t  it , (20)  1897

where Year dummy t equals 1 if   t and equals 0 otherwise. Because we have an interaction term of quota for each year, we drop the main effect of quotas. Then,   represents the yearly correlation between the quota and the number of new enterprises. Figure 4 (Panel A) plots the yearly correlation, along with the upper and lower bounds for 90% and 95% confidence intervals. There are two findings. First, the effect of quota on entrepreneurship does not show a clear trend in the seven years prior to the abolition. All seven coefficients are not significant and we cannot reject the null hypothesis that all pre-abolition effects are equal. Sec- ond, the effect immediately increases after abolition, from 0.033 in 1903 to 0.180 in 1904, and there is an upward trend after 1904, which is different from the constant trend prior to the abolition.

Figure 4 about here

It is possible that some omitted variables, which are correlated with economic growth and their effects, increase after 1904. If the apparent relationship between quotas and private firms is caused by these omitted variables, such variables might also affect other economic activities, such as the establishment of foreign firms. To rule out this possibility, we regress the number of new foreign enterprises on the quota and its interaction terms, with year dummies, such that

1910 Foreign F F F F Yit  ( Q i Year dummy  t)  Z i t it , (21)  1897

19 where  F represents the yearly correlation between the quota and the number of new foreign enterprises, and the right superscript F denotes the effect of the quota on the establishment of foreign firms. Figure 4 (Panel B) plots the yearly correlation and confidence intervals.

We find that the effect of the quota on the establishment of foreign firms does not show a clear pattern during the fourteen years included in this study, and furthermore, the trend of the effect of quotas on the number of new foreign firms is highly different from its effect on new private firms.

V.C. Dynamic Model

Because the number of firms established in a given year might be affected by the

number of firms established in the previous year (Yit1 ), we employ the following dynamic panel model with unobserved specific effects:

YY00 QZu  0000  t 1899,1900, 1903 it it1 i i i t it , (22) 11 1111 YYit it1 QZu i i  i  t it t 1904,1900, 1910 where  0 and 1 are the parameters associated with the lagged dependent variable. Equation

(22) can be estimated using the “system GMM” estimator (Arellano and Bover, 1995; Blun-

dell and Bond, 1998). yit 2 is employed as an instrument to correct for the endogeneity of

the lag of the dependent variable ( yit1 ) and because the standard error based on the “system

GMM” estimation is under-estimated, we will report the Windmeijer-corrected (2005) standard errors. Moreover, the validity of the GMM estimator is premised on the satisfaction of

two required conditions. The first is whether the error term vit has to be serially uncorrelated.11 A second condition required for the assumption of the validity of GMM estimates is that

11 Specifically, a serially uncorrelated error term means that the first-differenced error term vit should have significant first-order serial correlation but insignificant second-order serial correlation, which can be tested by the test-statistics m1 and m2 for the GMM estimator developed by Bond (2002).

20 the instruments are exogenous, and we thus report a Hansen test based on Hansen (1982).12

Estimation results of the dynamic panel models are presented in Table 2 (Panel A). AR(1) and AR(2) tests show that the null hypothesis of no serial correlations for the error terms cannot be rejected. The p-value of the Sargan/Hansen-statistics indicates that the over-identifying restrictions are not rejected. As for the estimation results, the coefficient of the quota increases significantly after the abolition of civil examinations, indicating that the inclusion of a lagged dependent variable does not change our baseline results. Moreover, we find that the effect of lagged numbers of private firms also increases, which implies that the adoption of

Western technology does exhibit strong time-dependent properties after the abolition. How- ever, the effect of quota only increases about 0.080 (Column (1), Panel A, Table 2), which is much smaller than the results in Figure 3.

Table 2 about here

Because the number of new private enterprises is a proxy for the adoption of Western technology, it is necessary to control for the effect of Western influence. Thus, we control for the number of foreign firms established in year t 1. To correct potential endogeneity, we

Foreign Foreign also use yit 2 as an instrument of Yit1 . The specification can be represented by the following equation:

YY00 YForeign   0 QZu  0000  t 1899, 1903 it it11 it i i i t it (23). 11Foreign 1 1111 YYit it11 Y it   QZu i i  i  t it t 1904, 1910

Consistent with the results in Panel A (Table 2), the effect of quota significantly increases after the abolition. The adoption of Western technology does exhibit strong time-dependent properties. Another finding is that foreign firms have a positive influence on the adoption of

1 12 Under the null hypothesis of joint validity, the vector of empirical moments ZEˆ is supposed to randomly N distribute around zero.

21 Western technology after the abolition, whereas they are negatively correlated with the adoption of Western technology before the abolition. This implies that the diffusion of Western technology depends on Chinese people having the incentive to learn.

Because the civil examinations were officially abolished on September 2, 1905, we address the problem of using 1904 (the decision to abolish the examinations was made on

January 13, 1904) as the year of policy change. To check the robustness of our results, we compare the effects post-1905 versus pre-1904 in the following manner:

YY00 QZ  010 t 1899, 1903 it it1 i i t it . (24) 11 111 YYit it1 QZ i i  t it t 1906, 1910

As before, the effect of quota significantly increases and the adoption of Western technology exhibits stronger time-dependent properties after the abolition (Column (1), Panel B in Table

2). Moreover, we also include the number of foreign firms established in the lagged year, and find that foreign firms had a positive influence after the abolition (Column (2), Panel B in

Table 2).

VI. Regression Discontinuity (RD) design

VI.A. Sources of Endogeneity

Our estimation might be biased by some unobserved time-invariant factor ( Fi ), which

might be related to the quota (Qi ). If the quota is positively related to the omitted variable,

for instance, the prefecture’s economic prosperity ( cov(QFii , ) 0), and economic prosperity

0 1 positively affects the establishment of firms ( cov(YFii , ) 0 ), the effects of quota (  and  )

22 13 might be over-estimated. We can rule out the effects of unobserved factors ( Fi ) that might

be related to the quota (Qi ) only when the effects of the omitted variables ( Fi ) does not change before and after the abolition of imperial civil examinations.

More seriously, other policy changes ( Pt ) during the period we are analyzing, such as the government introducing policies to encourage the diffusion of Western culture and sci-

ence and technology, might affect our results. Thus, the effects of omitted variables ( Fi ) might change over time. If so, our estimator might capture the effects of these omitted variables. To correct the potential bias, it is necessary to employ an exogenous and random varia-

ble i , which is correlated with the quota, but uncorrelated with other omitted variables, thus

cov(ii ,F ) 0 .

VI.B. Fuzzy RD design

Quotas were allocated to administrative units on the basis of the units’ size and importance (Chang, 1955).14 The schools of different administrative units were classified as large, medium, and small (first mentioned in 1647), and the admission quotas to Shengyuan were varied among these three types of schools. For instance, in an edict in 1658, the quota for a large school (big prefectures) was 20, a medium school (big county) 15, and a small school 4 (or 5). In 1670, the quota for medium schools increased to 12 and for small schools to 8.

13 For the same reason, if the quota is positively related to some omitted variables ( cov(QFii , ) 0 ), such as the prefecture’s Confucian ideology which is negatively related with business culture ( cov(YFii , ) 0 ), the effects of quota (  0 and  1 ) might be under-estimated. 14 “Quotas were allocated to administrative units. Each prefecture and district had its government school for Shengyuan, and each school had a specific quota of Shengyuan to be admitted in each examination. The size of these quotas varied according to the importance and size of the administrative unit.” (Chang, 1955: 77)

23 The importance of administrative units can be measured by the rank of their post designations. Every administrative unit was characterized according to the presence or absence of four attributes based on the four aspects of administration: economy, transportation, gov- ernance difficulty, and security. The four attributes were (1) “thoroughfare” (Chong), a center of communication; (2) “troublesome or abundant” (Fan), a great deal of official business; (3)

“fatiguing” (Pi), an indicator of the difficulty of collecting taxes; (4) “difficulty” (Nan), an indicator that a post had to cope with a crime-prone populace. Based on these four variables, there were sixteen possible post designations.15 This method was proposed in 1728, and in

1731 the Board of Personnel used this classification system to designate four post levels – most important, important, medium, and simple. This system continued until the end of the

Qing dynasty.

It is clear that the Qing government adopted the dichotomous approach of the Ming dynasty (Liu, 1993), in which a dummy variable was defined as “troublesome, abundant”

(Fan) or “simple” (Jian). 16 In the History of Ming (Zhang, 1987), the method is recorded as

“the administrative unit is classified as ‘troublesome, abundant’ if at least one of the following conditions is matched: a prefecture with agricultural taxes totaling more than or equal to

150 thousand stones, a Zhou (department) with agricultural taxes more than or equal to sev- enty thousand stones, or a county with agricultural taxes more than or equal to thirty thousand stones, or contains a prince’s mansion, or is the seat of a provincial government (a city with all three departments in it), or has troops stationed in it, or has vital postal routes, or provides

15 There is one designation consisting of four characters, four consisting of combinations of three characters, six consisting of two characters, four consisting of only one characters, and one without any characters. 16 Though the schools were divided into three types, the Qing government did not adopt the triple classification for administrative units. The Shunzhi Emperor ordered the Board of Personnel to classify the administrative units into the three types in 1655, but stopped the project in 1657.

24 for armed forces. If it meets none of these conditions it is classified as ‘simple’.”17 Thus, there was a clear demarcation between the important, or “troublesome, abundant” prefectures

(1Ii  ), and the unimportant or “simple” ones (Ii  0 ). If the agricultural tax ( x ) was more than or equal to 150 thousand stones, the prefectures were regarded as important. If the agricultural tax ( x ) was less than 150 thousand stones, but at least one of the other conditions

was met, the prefectures could still be ranked as “troublesome, abundant” (Ii  1). Overall, in the Ming dynasty, if the agricultural tax of a prefecture ( x ), an observed “assignment” variable, exceeds a known cutoff point of 150, it will significantly increase the likelihood of being

characterized as important or “troublesome, abundant” ( Ii  1 ).

We have illustrated the historical background in Figure 4. In the Ming dynasty, the

prefectures were classified into two groups: important (“troublesome, abundant”, Ii  1 ) and

unimportant (“simple”, Ii  0 ). One criterion was whether the agricultural tax was more or less than 150 thousand stones. If the agricultural tax exceeded 150 thousand stones ( x  150 ), the prefecture would be classified as important. In the subsequent Qing dynasty, the central government adopted this binary approach, and we hypothesize that the important prefectures were more likely to have large schools and therefore have larger quotas. The quotas remained stable from 1724 to 1850 when they were disrupted by the Taiping Rebellion. From the end of the Rebellion until the abolition of the civil examination system there was an increase of only about 10 per cent. In 1731, a new method for ranking administrative units was employed, but the examination quota stayed relatively stable and time dependent. In sum, if the agricultural tax exceeded 150 thousand in the Late Ming dynasty, the prefecture would have a larger

17 The classification standard is translated by the author from Chronicle 47, Vol. 70 in the History of Ming (Ming Shi) edited by Zhang (1987: 1722).

25 quota for officials. After 1737, a new classification system was adopted, and thus this tax cutoff point was no longer used to measure prefecture importance in the Late Qing dynasty.

Figure 4 about here

VI.C. Model Specifications

We use fuzzy RD to exploit the discontinuity in the expected value of treatment Ii ,

conditional on agricultural tax ( xi ). There is a jump in the expected value of Qi at c  150

(unit: a thousand stones), so that:

 f1(, xii Z ) if x i c , namely I i 1 EQ[iii | x , Z ] where f10 ( xii ) f ( x ) . (25)  f0 (, xii Z ) if x i c , namely I i 0

Replacing xiixc, the non-parametric version of fuzzy RD can be represented by:

* EQ[|0iii x ,] Z  EQ [| i  x ii 0,] Z  . (26)

The reduced-form conditional expectation is

* EY[|0iii x ,] Z  EY [| i  x ii 0,] Z  , (27)

Then,  can be estimated by dividing (7) by (6) and taking a limit as  tends to zero.

In Panel A of Figure 5, the fitted values come from nonparametric regressions using

Cleveland’s (1979) tricube weighting function and a bandwidth of 1, estimated separately for prefectures on either side of the cut value. Thus, they represent a moving average of quota across the cut value. The figure presents dramatic evidence that the important prefectures have a larger quota. An especially convincing feature is the evidence of a discontinuous increase when the taxes are just above the cutoff value of 150 thousand stones. When the cutoff point is exceeded, the quota increases about 0.36 (hundred). In Panel B, we perform the same analysis but replace the size of the quotas with the average number of private firms estab-

26 lished per year from 1897 to 1903, and find that there is only a 0.06 increase across the threshold. In Panel C, we graph the change of the average number of private firms established per year from 1904 to 1910, and the graph clearly shows that there is a jump at the cut off value, about 0.40. Though a convincing non-parametric estimation would require a large number of observations near the treatment threshold (Imbens and Lemieux, 2008), these figures still suggest that before the abolition, the size of the quota had a small effect on firm establishment (about 0.06/0.36=0.17), and this effect increased after the abolition (to

0.40/0.36=1.11), which supports the hypothesis that the prefectures with larger quotas tended to attract more firms after the abolition of the imperial civil examinations.

Figure 5 about here

Though non-parametric techniques have the advantage of not relying on the assumption of functional form, we do not have a large sample so a parametric method is employed.

The equation (25) can therefore be rewritten as

EQxZ[|,]iii f010 (,) xZ ii I i [(,) fxZ  ii f (,)] xZ ii, (28),

th where xiixc. Assuming that f1(,xZii ) and f0 (,xZii ) can be described by p order polynomials, we have

2 p EQ[|,]iii x Z00 01 x i   02 x i   0 pi x ** *2 *p IxxxZiiipii ( 01   2    )   (29) P **jj 0000IxxIZipijiii (   )  j1

The first stage can be written as

P **jj QIii0000()  jijiiii xxIZv  , (30) j1 and the second stage as

27 P jj* YQii ()00  0 jijiiii x   xIZ  . (31) j1

th In both stages, we control for p -order polynomials and their interaction with Ii , and then

we use Ii to instrument Qi , to correct the potential bias.

In the case of polynomial regressions, the key question is the choice of the order of the polynomial regressions. In this research, the Akaike information criterion ( AIC ) of model selection is used to decide the order of polynomial regression. In a regression discontinuity context, the AIC is given by nPln(ˆ 2 ) 2( 1) , where ˆ is the mean squared error of the regression (Lee and Lemieux, 2009), and we can choose P to minimize AIC . Another key issue is to choose the appropriate bandwidth ( ) around the regression cutoff point. General- ly, using a large bandwidth around the cutoff point is more precise because more observations are available, but the polynomial function can provide a closer approximation over a small bandwidth than over a large bandwidth. Thus, a number of bandwidths are chosen to illustrate the robustness of the results. The analysis starts with a small bandwidth   5 , namely the prefectures with 55x  are included in the regressions; and then the bandwidth increases step by step,   6,7,8,9,10. Figure 6 presents the geographic distribution of prefectures in different bandwidths.

Figure 6 about here

Table 3 examines the robustness of five specifications of the polynomial of x in the

given bandwidths. The first row reports results when only Ii is included, without any polynomial of x (the polynomial order P  0 ), and the second row reports the results with the inclusion of the 1st-order polynomial of x and its interaction (the polynomial order P  1).

The results with the quadratic polynomials ( P  2 ) and their interactions with Ii are reported

28 in the third row, the cubic polynomial ( P  3) and the quartic polynomial ( P  4 ) in the

fourth and fifth row respectively. The significance of Ii in all specifications and all bandwidths shows that the results are not sensitive to the order of the polynomial.

Table 3 about here

The AIC statistics for each specification are also reported, and the optimal polynomial order can be decided by minimizing AIC. For instance, for x [5,25] , the optimal polynomial

order is 1, which implies that only linear polynomials and their interaction with Ii are con- trolled. The optimal polynomial order is reported in the second last row, which shows that it is necessary to include the higher order of polynomial when the bandwidth is enlarged. Using

the optimal polynomial order, Table 2 shows that the effects of Ii range from 0.740 to 0.903, which means that if the agricultural taxes at the end of the Ming dynasty exceeded 150 thousand stones, about 74 to 90 additional spaces would be allocated to this prefecture’s quota.

VI.D. Instrumental Evidence

Because the validity of the instrument is based on the assumption that Ii first affected the quota in the early Qing dynasty and that the size of the quota had strong time dependence,

Ii should affect the quota in late Qing. If Ii had no effect on the quota in the later period of the Qing dynasty, then the validity of the instrument is uncertain. To evaluate the plausibility

of this assumption, the size of the quota (Wi ) before the Taiping Rebellion (1850-1864) is

regressed on Ii based on the specification of Equation (32):

P **jj WIii0000()  pijiiii x  xIZv . (32) j1

29 This equation has almost the same control variables as Equation (30) except that the population density in 1880 is replaced with that of 1776, because most of the dependent variables

(Wi ) are variables from before 1850. The robustness of the results show that our null hypoth-

esis, that Ii affects the quota throughout the Qing dynasty, is not rejected.

Table 4 about here

Another assumption of the instrument is that the classification system of administra-

tive units, namely post-designation, greatly changed in 1727. But if Ii still influenced the importance of prefectures in the late Qing, then our instrument might affect firm establishment via other channels. For instance, the Qing government could have paid more attention to the important prefectures and therefore more state enterprises were constructed in these locations.

To test the validity of this assumption, two variables, a binary variable, “troublesome, abundant” or not, and a category variable, “the rank of importance of prefecture – 1, simple, 2,

medium, 3, important, 4, the most important” – are regressed on Ii based on the specification

(32). The results are reported in the second and third rows respectively and there is no significant relationship.

The RD approach requires another identifying assumption: all outcomes except for quota size must vary smoothly at the cutoff point. Supposing W 1 and W 0 are potential out-

1 0 comes in the two groups with Ii  1 and Ii  0 . In this case EW(|,) xZ and EW(|,) xZ should be continuous at the threshold. To evaluate the plausibility of this assumption, we ex-

amine the following measurement of economic prosperity (Wi ): based on Equation (32), two dummies are used to signify the urbanization level in the mid-nineteenth century – whether the urban population was larger than 30,000, whether the urban population was larger than

30 50,000, and the number of bank branches open in the 1850s. We find that the effect of an

“abundant” ( Ii ) classification is insignificant. This suggests that Ii is not correlated with other variables that suggest economic prosperity – a key factor that might affect the establishment of enterprises.

To instrument the quota and estimate its effect on the establishment of new enterprises, the second stage of regression specified by equation (31) can be rewritten as:

P 00 00jj 0* 010 YQit i()00  0 j xxIZ i   j i i  i  t it j 1  , (33) P 1 1 1 1jj 1* 1 1 1 YQit i()00  0 j xxIZ i   j i i  i  t it j1

where we use Ii to instrument Qi while controlling for the polynomial of xi and their inter-

actions with Ii (Angrist and Pischke, 2009). The order of polynomials is determined by AIC statistics (Table 3). As in the baseline analysis, we graph the coefficients in Figure 7. The main results support our hypothesis that the effect of quota significantly increases after the abolition of the civil examinations. Compared with our baseline results, we find that the in- strumented effect of quota is much larger than the baseline results. For instance, after the abolition, the effect of quota increases by about 0.856 (for the first subsample with an agricultural tax greater than 50 thousand stones and less than 250 thousand, Panel A in Figure 7) compared with the seven years before the abolition. This effect is only 0.310 in the baseline results. Moreover, across different bandwidths, the results remain robust. For instance, from the comparison between the seven years pre- and post- abolition, the increase in the effects of quota ranges from 0.987 to 1.203 (Figure 7).18

Figure 7 about here

18 The increase of the quota’s effects is 0.987 in Panel A, 1.203 in Panel B, 1.098 in Panel C, 1.112 in Panel D, 1.082 in Panel E, and 1.041 in Panel F. Please refer to Figure 7 for details.

Moreover, we control for the number of firms established in the last year (Yit 1), and

employ Ii as the instrument of Qi such that:

P 00 000*0000jj YYititi1000 Q(  jijiii x xIZu )   itit  t 1899, 1903 j1 . (34) P 11 111*1111jj YYit it1000 Q i(  j x i j xIZu i i )  i  i  t it t 1904, 1910 j1

In Table 5, we find that the effect of quota significantly increases, which supports our hypothesis. For instance, in the first subsample (including all prefectures with agricultural taxes greater than 50 thousand stones and less than 250 thousand), the effects increase about 0.733 after the system’s abolition. These results remain robust across different bandwidths, in which the increases in the quotas’ effects range from 0.608 to 0.780.

Table 5 about here

VII. Conclusion

In this paper we account for the effects of initial institutions on the great divergence between Europe and China. The rigidity of the feudal system in Europe allowed economic elites to enrich themselves, and thus motivated possible political reform. However, in late imperial China, political power was much more diffuse and the political system was less oligarchic. The ruling elites were not hereditary, but selected through the imperial civil examination. This institution provided Chinese gentry and merchants with more incentives to invest in traditional Chinese education, and fewer incentives to seek other types of knowledge such as science and technology. Our empirical results support this hypothesis. The prefectures with higher quotas tended to establish more modern private enterprises once the examination system was abolished, which implies that the civil examination system provided negative incentives to adopt Western technology.

This paper helps us to understand the role of institutions in economic development. In strongly oligarchic societies, elites without political power might exploit other economic opportunities and attempt to change the political system with their increasing wealth; however, a less oligarchic society provides incentives for a wider variety of people to join the political game instead of pursuing other economic opportunities. This study also helps us to understand the effect of pre-existing institutions, especially the rigidity of a political system, on the great divergences in technology and per capita income between the West and China. Finally, the study identifies the effects of the civil examination system on the adoption of technology and supports the extant hypothesis that the civil examination system limited the rise of modern science and industry.

References Acemoglu, Daron (2008) “Oligarchic vs. Democratic Societies,” Journal of the European Economic Association 6: 1-44. Acemoglu, Daron, Davide Cantoni, Simon Johnson and James A. Robinson (2010) “From Ancien Regime to Capitalism: The French Revolution as a Natural Experiment,” in Jared Di- amond and James A. Robinson eds. Natural Experiments in History. Cambridge: Harvard University Press. Acemoglu, Daron, Davide Cantoni, Simon Johnson and James A. Robinson (2011) “The Consequences of Radical Reform: The French Revolution,” American Economic Review, forthcoming. Acemoglu, Daron and Simon Johnson (2005) “Unbundling Institutions,” Journal of Political Economy 113, 949-995. 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. Acemoglu, Daron, Simon Johnson and James A. Robinson (2002) “Reversal of Fortune: Ge- ography and Institutions in the Making of the Modern World Income Distribution,” Quarterly Journal of Economics 118: 1231-1294. Acemoglu, Daron, Simon Johnson and James A. Robinson (2005) “The Rise of Europe: At- lantic Trade, Institutional Change, and Economic Growth,” American Economic Review 95 (3): 546-579. Allen, Robert C. (2004) “Agricultural Productivity and Rural Incomes in England and the Yangtze Delta, c. 1620- c. 1820,” Nuffield College, mimeo.

33 Angrist, Jushua D. and Jorn-Steen Pischke (2009) Mostly Harmless Econometrics. Princeton University Pres Arellano, M. and Bover, O. (1995). “Another look at the instrumental variables estimation of error components models.” Journal of Econometrics, vol. 68, pp. 29–51. Blundell, R., and Bond, S. (1998). “Initial conditions and moment restrictions in dynamic panel data models.” Journal of Econometrics, vol. 87, pp.11–143. Bond, S. (2002). “Dynamic panel data models: a guide to micro data methods and practice.” Institute for Fiscal Studies Working Paper 09/02, Institute for Fiscal Studies, London. Chaffee, John W. (1995). The Thorny Gates of Learning in Sung China: A Social History of Examinations. Albany, NY: State University of New York Press. Chang, Chung-li (1955) The Chinese Gentry: Studies on Their Role in Nineteenth Century Chinese Society. Seattle: University of Washington Press. Chang, Chung-li (1962). The Income of the Chinese Gentry. Seattle, WA: University of Washington Press. Chang, Yufa. 1989. “Private Industries in the Late Ch'ing and the Early Republic of China, 1860-1916.” Bulletin of the Institute of Modern History, Academia Sinica (18): 315-561. "CHGIS, Version 4" Cambridge: Harvard Yenching Institute, January 2007 Clark, G. and Feenstra, R. (2003) “Technology in the Great Divergence” in Michael Bordo, Alan Taylor and Jeffrey Williamson eds. Globalization in Historical Perspective. Chicago: University of Chicago Press Cleveland, W. S. 1979. Robust locally weighted regression and smoothing scatterplots. Jour- nal of the American Statistical Association 74: 829-836. De Long, J. B. and A. Shleifer (1993) “Prices and Merchants: European City Growth before the Industrial Revolution,” Journal of Law and Economics 36(2): 671-702. Diamond, J. (1997) Guns, Germs, and Steel: The Fates of Human Societies. New York: Nor- ton. Ebrey, Patricia B. (1999). The Cambridge Illustrated History of China. Cambridge: Cam- bridge University Press. Elman, Benjamin A. (1990). Classicism, Politics, and Kinship: The Ch'ang-chou School of New Text Confucianism in Late Imperial China. Berkeley, CA: University of California Press. Elman, Benjamin A. (1991). “Political, Social, and Cultural Reproduction via Civil Service Examinations in Late Imperial China.” The Journal of Asian Studies 50(1): 7-28 Elman, Benjamin A. (2000). A Cultural History of Civil Examinations in Late Imperial China. Berkeley, CA: University of California Press. Ge, Jianxiong. 2000. China Population History (Zhongguo Renkou Shi). Shanghai: Fudan University Press. Glahn, Richard von (1996). Fountain of Fortune: Money and Monetary Policy in China, 1000-1700. Berkeley, CA: University of California Press. Galor, Oded, Omer Moav and Dietrich Vollrath (2009) “Inequality in Landownership, the Emergence of Human-Capital Promoting Institutions, and the Great Divergence,” Review of Economic Studies 76(1): 143-179.

34 Hansen, L. (1982). “Large sample properties of generalized method of moments estimators.” Econometrica, vol. 50 (3), pp. 1029–54. Ho, Ping-ti (1962) The Ladder of Success in Imperial China. New York: Columbia Universi- ty Press. Huff, Toby (2003). The Rise of Early Modern Science: Islam, China, and the West. Cambridge, UK: Cambridge University Press. Imbens, G., and T. Lemieux (2008) “Regression discontinuity Designs: A Guide to Practice.” Journal of Econometrics, 142: 615-635. Jones, E. L. (1981) The European Miracle: Environments, Economies, and Geopolitics in the History of Europe and Asia. Cambridge: Cambridge University Press. Keay, John (2009). China: A History. New York: Basic Books.

Kun, Gang. 1991. Qing Hui Dian Shi Li. Beijing : Zhonghua shu ju. La Porta, Rafeal, Lopez-de-Silanes, Florencio, Shleifer, Andrei and Vishny, Robert W. (1997) “Trust in Large Organizations,” American Economic Review 87 (2): 333-338. La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer, and Robert W. Vishny (1998) “Law and Finance," Journal of Political Economy 106: 1113-1155. Liang, Fang-zhong. 2008. Historical Statistics on Hukou, Land and Land Tax of China (Lidai hukou, tudi,tianfu tongji). Beijing: Zhonghua Book Company. Lin, Justin Yifu (1995). “The Needham Puzzle: Why the Industrial Revolution Did Not Orig- inate in China.” Economic Development and Cultural Change, 43(2), pp. 269-292. Liu, Cheng-yun (1993). “Chung Fan P’I Nan: A preliminary Study of Post Designations during the Ch’ing,” Bulletin of the Institute of Modern History, Academia Sinica (64): 175-204. Man-Cheong, Iona D. (2004). The Class of 1761: Examinations, State and Elite in Eighteenth Century China. Stanford， CA: Stanford University Press. Maddison, A. (1998) Chinese Economic Performance in the Long run. Paris: OECD Devel- opment Centre. Maddison, A. (2001) The World Economy: A Millennia Perspective. Paris: OECD Develop- ment Centre. Ni, S. and P. H. Van (2006). “High Corruption Income in Ming and Qing China.” Journal of Development Economics 81 (2): 316-336. North, D. C. (1990). Institutions, Institutional Change, and Economic Performance. New York: Cambridge University Press. North, D. C. (1994). Economic Performance through Time. American Economic Review 84(3): 359-368. Olson, Mancur (1982) The Rise and Decline of Nations: Economic Growth, Stagnation, and Social Rigidities. New Haven: Yale University Press. Pomeranz, Kenneth (2000). The Great Divergence: China, Europe and the Making of the Modern World. Princeton, NJ: Princeton University Press. Rozman, Gilbert. 1973. Urban Networks in Ch’ing China and Tokugawa Japan. Princeton, N.J.: Princeton University Press.

35 Sachs, J. and Warner, A. M. (1995) “Natural resource Abundance and Economic Growth,” NBER working paper No. 4398. Weber, M. (1930) The Protestant Ethic and the Spirit of Capitalism, translated by Talcott Parsons. London: Allen & Unwin. Weber, M. (1964) The Religion of China, translated and edited by H. H. Gerth. New York: Collier and Macmillan. Windmeijer, F. (2005). “A finite sample correction for the variance of linear efficient two- step GMM estimators.” Journal of Econometrics, vol. 126, pp. 25–51. Wolfgang Franke (1960). The Reform and Abolition of the Traditional China Examination System. Cambridge, Massachusetts: East Asian Research Center. Yuchtman, Noam (2009). “Teaching to the Tests: An Economic Analysis of Educational In- stitutions in Late Imperial and Republican China,” Harvard University, Working paper. Zhang Tingyu (1987) “The History of Ming” (Ming Shi). Shanghai: Shanghai Ancient Books Publishing House (Shanghai Guji Chuban She).

36 Figure 1: Geographic distribution of sampled prefectures

Source: "CHGIS, Version 4" Cambridge: Harvard Yenching Institute, January 2007.

37 Figure 2: The development of private modern industrial enterprises in China, 1840-1916

160

140

120

100

5 8 1 4 7 7 0 3 6 9 6 6 6 0 0 0 0 840 843 846 849 852 85 85 876 879 882 885 888 891 894 89 915 Year 1 1 1 1 1 1 1 18 18 18 1870 1873 1 1 1 1 1 1 1 1 19 19 19 19 1912 1

Data source: Chang, Yufa. 1989. “Private Industries in the Late Ch'ing and the Early Republic of China, 1860-1916.” Bulletin of the Institute of Modern History, Academia Sinica (18): 315-561.

38 Figure 3: The Effect of Quota on Firm Establishment, pre- and post- abolition of Imperial Civil Examination

Panel A: The Effect of Quota on the Establishment of Private Firms before and after the Abolition of Civil Examination, by the Length of Periods

0.450

0.400

0.350

0.300

0.250 Post 0.200 Pre

0.150

0.100

0.050

0.000 Year(s) 1234567

Panel B: The Difference in the Effects of Quota on the Establishment of Private Firms before and after the Abolition of Civil Examination, by the Length of Periods

0.450

0.400

0.350

0.300

0.250

0.200

0.150

0.100

0.050

0.000 Year(s) 1234567 Post - Pre 90% Confidence Interval 95% Confidence Interval

Data source: appendix

39 Figure 4: The Effect of Quota on Firm Establishment, by year

Panel A: The Effect of Quota on the Establishment of Private Firms

0.70

0.60

0.50

0.40

0.30

0.20

0.10

0.00 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 -0.10 Year

-0.20

The Effect of Quota 90% Confidence Interval 95% Confidence Interval

Panel B: The Effect of Quota on the Establishment of Foreign Firms

0.30

0.20

0.10

0.00 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910Year

-0.10

-0.20

The Effect of Quota 90% Confidence Interval 95% Confidence Interval

Data source: appendix

40 Figure 4: Historical Background of Regression Discontinuity Design

Ming Dynasty Qing Dynasty

From 1647 to 1724, the Quotas increased Quota allocation: government schools were during Taiping classified into three Rebellion levels and were allocated to specific quotas. Quotas remained stable until Taiping Rebellion

Discontinuous rule based on agricultural tax ( x ) at the cutoff The importance of point 150 administrative units: Ixi = 1, if ≥ 150 Importance Ixi = 0, if < 150

New method of post designations

1644 1731 1911 Year

41 Figure 5: Non-parametric estimation around the cutoff point Panel A: non-parametric estimation of RD on the number of quotas 1.5

-.5

-1 -15 -10 -5 0 5 10

Panel B: non-parametric estimation of RD on the number of private firms established per year before the abolition of civil examination system (1904 – 1910)

-.2 -15 -10 -5 0 5 10

Panel C: non-parametric estimation of RD on the number of private firms established per year after the abolition of civil examination system (1897-1903)

-1 -15 -10 -5 0 5 10

42 Figure 6: Geographical Distribution of Prefectures with Tax Data in Late Ming China

Source: "CHGIS, Version 4" Cambridge: Harvard Yenching Institute, January 2007.

43 Figure 7: Instrumental Evidence, by the Length of Periods with Different Bandwidths

Panel A: Tax - [5,25] Panel B: Tax - [4,26]

5.000 5.000 Post - Pre Post - Pre 90% Confidence Interval 4.500 90% Confidence Interval 4.000 95% Confidence Interval 4.000 95% Confidence Interval

3.500 3.000 3.000

2.500 2.000 2.000

1.000 1.500

1.000

0.000 0.500 1234567 Year(s) 0.000 Year(s) -1.000 1234567

Panel C: Tax - [3,27] Panel D: Tax - [2,28]

5.000 Post - Pre 5.000 Post - Pre 90% Confidence Interval 90% Confidence Interval 4.000 95% Confidence Interval 4.000 95% Confidence Interval

3.000 3.000

2.000 2.000

1.000 1.000

0.000 0.000 1234567 1234567 Year(s) Year(s) -1.000 -1.000

Panel E: Tax - [1,29] Panel F: Tax - [0,30]

5.000 Post - Pre 5.000 Post - Pre 90% Confidence Interval 90% Confidence Interval 4.000 95% Confidence Interval 4.000 95% Confidence Interval

3.000 3.000

2.000 2.000

1.000 1.000

0.000 0.000 1234567 1234567 Year(s) Year(s) -1.000 -1.000

Data source: appendix

44 Table 1: Definition of Variables and Data Sources

Variables Variables Definition Data Sources Observation Mean S.D.

Domestic firms Number of new erected firms, 1897-1910 A 3668 0.117 (0.699) Quota of imperial examination Quota of imperial civil examination (hundred) B 262 1.138 (0.756) Control Variables Changjiang River (=1, if riverside) C 262 0.062 (0.240) Coast (=1, if coastal) C 262 0.134 (0.341) Longitude C 262 111.561 (5.792) Latitude C 262 30.687 (4.953) Population density, 1880 (log-term) D 262 4.284 (1.095) Population density, 1776 (log-term) D 262 4.183 (1.111) Size (log-term) D 262 9.336 (0.770) Taxation in Late Ming dynasty Agricultural Tax (10,000 Stones) E 160 17.441 (33.352) Urbanization (1840s) Urban population larger than 30,000 F 262 0.359 (0.481) Urban population larger than 50,000 F 262 0.160 (0.368) Bank Branches Number of Bank Branches G 262 0.122 (0.496) Importance of prefecture Rank H 262 2.607 (0.972) Abundant H 262 0.760 (0.428)

A: Chang, Yufa. 1989. “Private Industries in the Late Ch'ing and the Early Republic of China, 1860-1916.” Bulletin of the Institute of Modern History, Academia Sinica (18): 315-561. B: Kun, Gang. 1991. Qing Hui Dian Shi Li. Beijing : Zhonghua shu ju. C: "CHGIS, Version 4" Cambridge: Harvard Yenching Institute, January 2007 D: Ge, Jianxiong. 2000. China Population History (Zhongguo Renkou Shi). Shanghai: Fudan University Press. E: Liang, Fang-zhong. 2008. Historical Statistics on Hukou, Land and Land Tax of China (Lidai hukou, tudi,tianfu tongji). Beijing: Zhonghua Book Company. F: Rozman, Gilbert. 1973. Urban Networks in Ch’ing China and Tokugawa Japan. Princeton, N.J.: Princeton University Press. G: H:

45 Table 2: Robustness check - Dynamic model Panel A: Post: 1904 – 1910 Vs. Pre: 1897 – 1903 (1) (2) Post- Pre- Post-Pre Post-Pre- Post-Pre Quota 0.103*** 0.023 0.080*** 0.098***0.019 0.079*** (0.034) (0.017) (0.030) (0.032) (0.016) (0.027) L. Foreign firms 0.783*** -0.447*** 1.230*** (0.149) (0.036) (0.144) L. Private firms 0.528*** 0.054 0.474*** 0.397*** 0.215*** 0.182*** (0.052) (0.088) (0.071) (0.072) (0.069) (0.021) AR(1) -2.55 -1.70 -2.30 -2.21 AR(2) 0.96 0.85 0.81 -1.02 Hansen test 0.190 0.385 0.637 0.514 Number of prefecture 1834 1834 1834 1834 Observations 262 262 262 262

Panel B: Post: 1906 – 1910 Vs. Pre: 1899 – 1903 (1) (2) Post- Pre- Post-Pre Post-Pre- Post-Pre Quota 0.158*** 0.016 0.142*** 0.107**0.025 0.082** (0.058) (0.017) (0.056) (0.043) (0.022) (0.037) Foreign firms 0.849*** -0.635*** 1.484*** (0.202) (0.083) (0.184) L. Private firms 0.403*** 0.103 0.299*** 0.367*** 0.128* 0.239*** (0.089) (0.079) (0.043) (0.099) (0.078) (0.062) AR(1) -2.48 -1.65 -2.10 -3.91 AR(2) 0.92 0.37 0.94 -0.98 Sargan/Hansen test 0.373 0.351 0.624 0.521 Number of prefecture 1310 1310 1310 1310 Observations 262 262 262 262 Notes: Standard errors in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1%; Control Variables ( Z ), year dummies and constant term are included in the regressions but not reported; AR(1) and AR(2) are test-statistics for the first-order and second-order serial correlation of the first-differenced residuals; Hansen test (P-value), which tests the validity that the instruments are exogenous, is reported.

46 Table 3: RD estimates of the effect of “Complicated” entitlement in Late Ming dynasty on the quota of licentiate in Late Qing dynasty

Bandwidth 1: [5, 25] 2: [4, 26] 3: [3, 27] 4: [2, 28] 5: [1, 29] 6: [0, 30] Polynomial of order A. Zero 0.514*** 0.447** 0.517*** 0.537*** 0.555*** 0.568*** (0.187) (0.174) (0.174) (0.175) (0.174) (0.169) R-squared 0.83 0.82 0.77 0.79 0.79 0.79 B. One (Linear polynomial) 0.903*** 0.756*** 0.611*** 0.541*** 0.442** 0.365** (0.198) (0.193) (0.196) (0.183) (0.178) (0.168) AIC: -150.88 -180.65 -176.93 -194.90 -205.98 -255.03 R-squared 0.80 0.81 0.74 0.73 0.73 0.76 C. Two (Quadratic polynomial) 0.836*** 0.729*** 0.740*** 0.794*** 0.840*** 0.875*** (0.227) (0.221) (0.213) (0.211) (0.208) (0.201) AIC: -149.97 -178.69 -177.30 -198.91 -216.75 -269.59 R-squared 0.81 0.81 0.75 0.75 0.77 0.79 D. Three (Cubic polynomial) 0.738*** 0.548** 0.552** 0.582*** 0.567*** 0.663*** (0.266) (0.220) (0.218) (0.205) (0.206) (0.218) AIC: -146.92 -178.26 -175.02 -197.67 -216.72 -268.43 R-squared 0.82 0.82 0.76 0.76 0.78 0.79 E. Four (Quartic polynomial) 0.846*** 0.579** 0.577** 0.562** 0.596*** 0.636*** (0.308) (0.258) (0.241) (0.223) (0.216) (0.214) AIC: -146.14 -175.75 -172.81 -196.10 -215.79 -267.26 R-squared 0.83 0.83 0.77 0.77 0.78 0.80 Optimal order of polynomial 1 1 2 2 2 2 Number of observations 71 82 90 101 108 134 Notes: Robust standard errors in parentheses. The optimal order of the polynomial is chosen using Akaike's criterion. * significant at 10%; ** significant at 5%; *** significant at 1%

47 Table 4: RD estimates of the effect of “Complicated” entitlement in Late Ming dynasty on other economic factors Bandwidth 1: [5, 25] 2: [4, 26] 3: [3, 27] 4: [2, 28] 5: [1, 29] 6: [0, 30]

On Quota, pre-1840 0.514*** 0.447** 0.517*** 0.537*** 0.555*** 0.568*** (0.187) (0.174) (0.174) (0.175) (0.174) (0.169) R-squared 0.83 0.82 0.77 0.79 0.79 0.78 On the “abundant” 0.082 -0.141 -0.173 -0.194 -0.168 -0.121 (0.213) (0.185) (0.228) (0.218) (0.207) (0.212) R-squared 0.36 0.36 0.30 0.28 0.25 0.25 On the importance -0.479 -0.498 -0.445 -0.553 -0.515 -0.426 (0.503) (0.436) (0.517) (0.500) (0.487) (0.470) R-squared 0.47 0.51 0.43 0.43 0.40 0.38 On Urbanization (>30,000), 1840s 0.258 0.360 0.347 0.264 0.320 0.388 (0.277) (0.251) (0.320) (0.305) (0.300) (0.281) R-squared 0.45 0.36 0.34 0.30 0.34 0.39 On Urbanization (>50,000), 1840s 0.184 0.343 0.153 0.189 0.218 0.267 (0.237) (0.247) (0.332) (0.318) (0.306) (0.286) R-squared 0.60 0.45 0.38 0.34 0.34 0.34 On Bank, 1850s 0.202 0.391 0.177 0.255 0.260 0.255 (0.345) (0.338) (0.422) (0.411) (0.401) (0.387) R-squared 0.37 0.30 0.26 0.25 0.24 0.24

Observations 71 82 90 101 108 133 Notes: Robust standard errors in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1%

48 Table 5: Robustness check - Dynamic model Instrumental Results Bandwidth [5, 25] [4, 26] [3, 27] [2, 28] [1, 29] [0, 30] Post- Quota 0.862*** 0.938*** 0.807** 0.781** 0.828*** 0.919** (0.291) (0.319) (0.326) (0.311) (0.321) (0.369) L.Y 0.368*** 0.347*** 0.342*** 0.503*** 0.497*** 0.496*** (0.119) (0.111) (0.106) (0.164) (0.164) (0.164) AR(1) -2.65 -2.84 -2.36 -2.85 -2.86 -2.86 AR(2) -1.23 -1.25 -0.19 -0.93 -0.93 -0.93 Sargan/Hansen test 0.663 0.540 0.406 0.869 0.868 0.871

Pre- Quota 0.129* 0.138* 0.114 0.173** 0.184** 0.205** (0.069) (0.073) (0.072) (0.079) (0.085) (0.099) L.Y -0.075 -0.078 -0.062 0.174 0.174 0.174 (0.219) (0.201) (0.091) (0.153) (0.153) (0.154) AR(1) -1.59 -1.67 -3.10 -2.23 -2.23 -2.23 AR(2) -0.24 -0.24 -1.30 0.58 0.58 0.59 Sargan/Hansen test 0.284 0.293 0.684 0.513 0.509 0.504

Post - Pre 0.733*** 0.780** 0.693** 0.608** 0.643** 0.714** (0.283) (0.311) (0.317) (0.300) (0.310) (0.356)

Number of prefecture 71 82 90 101 108 134 Observations 497 574 630 707 756 938 Notes: Standard errors in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1%; Control Variables ( Z ), year dummies and constant term are included in the regressions but not reported; AR(1) and AR(2) are test-statistics for the first-order and second-order serial correlation of the first-differenced residuals; Hansen test (P-value), which tests the validity that the instruments are exogenous, is reported.